Guide to Hydrological Practices, 6th edition, Volume II - Hydrology.nl

Guide to Hydrological PracticesVolume IIManagement of Water Resources andApplication of Hydrological PracticesWMO-No. 168

Guide toHydrological PracticesVolume IIManagement of Water Resources andApplication of Hydrological PracticesWMO-No. 168Sixth edition2009

WMO-No. 168© World Meteorological Organization, 2009The right of publication in print, electronic and any other form and in any language is reserved byWMO. Short extracts from WMO publications may be reproduced without authorization, providedthat the complete source is clearly indicated. Editorial correspondence and requests to publish, reproduceor translate this publication in part or in whole should be addressed to:Chairperson, Publications BoardWorld Meteorological Organization (WMO)7 bis, avenue de la Paix Tel.: +41 (0) 22 730 84 03P.O. Box 2300 Fax: +41 (0) 22 730 80 40CH-1211 Geneva 2, SwitzerlandE-mail: publications@wmo.intISBN 978-92-63-10168-6NOTEThe designations employed in WMO publications and the presentation of material in this publication do notimply the expression of any opinion whatsoever on the part of the Secretariat of WMO concerning the legalstatus of any country, territory, city or area, or of its authorities, or concerning the delimitation of its frontiersor boundaries.Opinions expressed in WMO publications are those of the authors and do not necessarily reflect those of WMO.The mention of specific companies or products does not imply that they are endorsed or recommended by WMOin preference to others of a similar nature which are not mentioned or advertised.

VOLUME IITABLE FOR NOTING SUPPLEMENTS RECEIVEDSupplementNo.DatedbyInserteddate123456789101112131415161718192021222324252627282930

CONTENTSPREFACE ..............................................................................................................................................ACKNOWLEDGEMENTS ......................................................................................................................PagexixiiiCHAPTER 1 – INTRODUCTION ............................................................................................................1.1 Background .............................................................................................................................. II.1-11.2 Scope ....................................................................................................................................... II.1-11.3 Contents of the Guide .............................................................................................................. II.1-21.4 The Hydrological Operational Multipurpose System ................................................................. II.1-4References and further reading ............................................................................................................. II.1-4CHAPTER 2 – HYDROLOGICAL SERVICES ...........................................................................................2.1 Introduction ............................................................................................................................. II.2-12.2 Responsibilities and functions of Hydrological Services ............................................................. II.2-12.2.1 Nature of the products and services of a Hydrological Service .................................. II.2-12.2.2 Clients of hydrological products and services ........................................................... II.2-22.2.3 Managing relationships with clients ......................................................................... II.2-32.2.4 Hydrological products and services .......................................................................... II.2-32.2.5 Functions and activities of a Hydrological Service ..................................................... II.2-32.2.6 Evaluation of products and services, and quality management ................................. II.2-42.2.7 Legal basis for operations and organizational arrangements ..................................... II.2-52.2.8 Managing relationships with other institutions ......................................................... II.2-62.2.9 Data exchange ........................................................................................................ II.2-72.3 Planning and strategy .............................................................................................................. II.2-82.4 Human resources management and capacity-building ............................................................. II.2-92.4.1 Management ........................................................................................................... II.2-92.4.2 Training and continuing education .......................................................................... II.2-102.5 Financial and asset management .............................................................................................. II.2-112.5.1 Sources of revenue .................................................................................................. II.2-112.5.2 Budgeting and monitoring financial performance .................................................... II.2-122.5.3 Asset management .................................................................................................. II.2-122.5.4 Database security .................................................................................................... II.2-12References and further reading ............................................................................................................. II.2-13CHAPTER 3 – INTEGRATED WATER RESOURCES MANAGEMENT ......................................................3.1 Introduction ............................................................................................................................. II.3-13.1.1 Sustainable water development ............................................................................... II.3-13.1.2 The changing nature of the resource ....................................................................... II.3-13.1.3 Changing attitudes to management ........................................................................ II.3-23.2 Integrated water resources management .................................................................................. II.3-33.3 Rationale for integrated water resources management ............................................................. II.3-43.3.1 Water quantity and quality ...................................................................................... II.3-43.3.2 Surface water and groundwater ............................................................................... II.3-43.3.3 Upstream and downstream considerations .............................................................. II.3-43.3.4 Water, land and other resource systems ................................................................... II.3-53.3.5 Environment, the economy and society ................................................................... II.3-53.3.6 Vertical and horizontal fragmentation: systems and silos .......................................... II.3-53.3.7 Collaboration, coordination and coherence ............................................................. II.3-6II.1-1II.2-1II.3-1

viGUIDE TO HYDROLOGICAL PRACTICES3.4 Evolution of integrated water resources management .............................................................. II.3-73.4.1 United States of America: Ohio conservancy districts, Tennessee Valley Authority ..... II.3-73.4.2 Canada: Conservation authorities ............................................................................ II.3-73.4.3 United States and Canada: the great lakes ............................................................... II.3-73.4.4 Australia: total catchment management .................................................................. II.3-83.4.5 New Zealand: Resource Management Act ................................................................ II.3-83.4.6 South Africa ............................................................................................................. II.3-83.5 Perspectives on integrated water resources management ......................................................... II.3-93.5.1 Dublin Conference: Earth Summit, 1992 ................................................................. II.3-93.5.2 World Water Council and the World Water Fora ....................................................... II.3-103.5.3 Global Water Partnership ......................................................................................... II.3-103.5.4 World Summit on Sustainable Development, Johannesburg, 2002 ........................... II.3-113.6 Elements of best practice for integrated water resources management ..................................... II.3-113.6.1 Alternative interpretations: comprehensive versus integrated approaches ................ II.3-113.6.2 Vision for a desirable future ...................................................................................... II.3-123.6.3 Spatial scale: watershed, subwatershed, tributary and site ...................................... II.3-123.6.4 Partnerships and alliances ........................................................................................ II.3-133.6.5 Links to regional planning and impact assessment ................................................... II.3-133.6.6 Designing institutional arrangements ....................................................................... II.3-143.6.7 Monitoring and evaluating ...................................................................................... II.3-153.7 Cautions regarding integrated water resources management ................................................... II.3-153.7.1 When to apply integrated water resources management ......................................... II.3-153.7.2 Implementation gap ................................................................................................ II.3-15References and further reading ............................................................................................................. II.3-15CHAPTER 4 – APPLICATIONS TO WATER MANAGEMENT ..................................................................4.1 Water resource assessment and water projects ......................................................................... II.4-14.1.1 The need for water resource assessment .................................................................. II.4-14.1.2 Water resource assessment programme components ............................................... II.4-14.1.3 Evaluation of water resource assessment activities .................................................... II.4-34.1.4 Water projects ......................................................................................................... II.4-34.1.5 Purposes served by a water management project .................................................... II.4-34.1.6 Multi-purpose projects ............................................................................................. II.4-44.1.7 Project cycle ............................................................................................................ II.4-44.1.8 Preliminary investigation of water management projects ......................................... II.4-54.2 Estimating reservoir capacity and yield ..................................................................................... II.4-74.2.1 General ................................................................................................................... II.4-74.2.2 Concepts of yield ..................................................................................................... II.4-84.2.3 Estimation of storage–yield relationships .................................................................. II.4-114.2.4 Classifications of yield .............................................................................................. II.4-124.2.5 Probabilistic approach ............................................................................................. II.4-134.2.6 Multi-purpose reservoirs and operating rules ........................................................... II.4-164.2.7 Multi-reservoir water resource systems ..................................................................... II.4-174.2.8 Incidental effects of reservoirs .................................................................................. II.4-184.2.9 Remote-sensing estimates of reservoir capacity ........................................................ II.4-194.2.10 Climate change ....................................................................................................... II.4-204.3 Flood management .................................................................................................................. II.4-204.3.1 General ................................................................................................................... II.4-204.3.2 Flood management strategies .................................................................................. II.4-204.3.3 Integrated flood management ................................................................................. II.4-214.3.4 Structural measures .................................................................................................. II.4-224.3.5 Non-structural measures .......................................................................................... II.4-274.3.6 Flood emergency management ............................................................................... II.4-294.4 Irrigation and drainage ............................................................................................................. II.4-304.4.1 Irrigation ................................................................................................................. II.4-30PageII.4-1

CONTENTS4.4.2 Agricultural drainage ............................................................................................... II.4-394.4.3 Use of remote-sensing and general information systems in irrigation and drainage .. II.4-424.5 Hydropower and energy-related projects .................................................................................. II.4-424.5.1 General ................................................................................................................... II.4-424.5.2 Hydropower ............................................................................................................ II.4-434.5.3 Operation of a hydroelectric system ........................................................................ II.4-514.5.4 Other projects related to energy production ............................................................ II.4-524.6 Navigation and river training .................................................................................................... II.4-544.6.1 Application of hydrology to navigation .................................................................... II.4-544.6.2 Classification of river training ................................................................................... II.4-594.6.3 Erosive forces due to channel flow ........................................................................... II.4-604.6.4 Erosive forces caused by waves and craft ................................................................. II.4-624.6.5 Evolution and characterization of river bends ........................................................... II.4-624.6.6 Determination of design discharges and stages ....................................................... II.4-634.7 Urban water resources management ........................................................................................ II.4-644.7.1 General ................................................................................................................... II.4-644.7.2 Urban development impacts .................................................................................... II.4-654.7.3 Urban storm drainage design .................................................................................. II.4-664.7.4 Urban drainage control measures ............................................................................ II.4-714.7.5 Urban drainage management .................................................................................. II.4-714.7.6 Remote-sensing estimates for land use ..................................................................... II.4-734.8 Sediment transport and river channel morphology ................................................................... II.4-734.8.1 General .................................................................................................................... II.4-734.8.2 Catchment erosion .................................................................................................. II.4-734.8.3 Channel erosion ...................................................................................................... II.4-734.8.4 River systems ........................................................................................................... II.4-734.8.5 Flow regimes and bed forms .................................................................................... II.4-754.8.6 Transportation of sediments in channels .................................................................. II.4-774.8.7 Sedimentation .......................................................................................................... II.4-804.8.8 Sediment control measures ...................................................................................... II.4-814.9 Water quality and the conservation of aquatic ecosystems ........................................................ II.4-814.9.1 General ................................................................................................................... II.4-814.9.2 Relationships between water quantity and water quality .......................................... II.4-824.9.3 Effects of water resources projects on water quality in streams and rivers ................. II.4-834.9.4 Effects of water resources projects on water quality in large lakes and reservoirs ...... II.4-844.9.5 Water quality changes caused by pollution .............................................................. II.4-844.9.6 Measures to reduce effects of pollution on water quality .......................................... II.4-864.10 Hydroecology ........................................................................................................................... II.4-864.10.1 Introduction ............................................................................................................ II.4-864.10.2 Environmental management of rivers ...................................................................... II.4-874.10.3 Basic notions of river morphology and ecology ........................................................ II.4-894.10.4 Ecological impacts of water resources projects ......................................................... II.4-934.10.5 Mitigation of ecological impacts .............................................................................. II.4-95References and further reading ............................................................................................................. II.4-96viiPageCHAPTER 5 – EXTREME VALUE ANALYSIS ..........................................................................................II.5-15.1 Introduction ............................................................................................................................. II.5-15.2 Statistical series and return periods ........................................................................................... II.5-15.3 Probability distributions used in hydrology ............................................................................... II.5-35.3.1 Normal family: N, LN and LN3 ................................................................................ II.5-45.3.2 Extreme value distributions: Gumbel, generalized extreme value and Weibull .......... II.5-45.3.3 Pearson type III family .............................................................................................. II.5-65.3.4 Halphen family: types A, B and type B –1 ................................................................... II.5-65.3.5 Generalized logistic distribution ............................................................................... II.5-7

viiiGUIDE TO HYDROLOGICAL PRACTICES5.3.6 Generalized Pareto distribution ................................................................................ II.5-75.3.7 Non-parametric density estimation method ............................................................. II.5-75.4 Hypothesis testing .................................................................................................................... II.5-75.4.1 Wald–Wolfowitz test for independence and stationarity ........................................... II.5-85.4.2 Mann–Kendall test for trend detection ..................................................................... II.5-95.4.3 Mann–Whitney test for homogeneity and stationarity (jumps) ................................. II.5-95.4.4 Sample size and length of record ............................................................................. II.5-105.4.5 Grubbs and Beck test for detection of outliers .......................................................... II.5-105.4.6 Bayesian procedures ................................................................................................ II.5-105.5 Population statistics and parameter estimation ....................................................................... II.5-115.5.1 Parameter calculation methods ................................................................................ II.5-115.5.2 Use of logarithmic transformations .......................................................................... II.5-125.5.3 Historical Information .............................................................................................. II.5-125.5.4 Record augmentation .............................................................................................. II.5-125.5.5 Analysis of mixed populations .................................................................................. II.5-135.5.6 Frequency analysis and zeroes ................................................................................. II.5-135.6 Probability plots and goodness-of-fit tests ................................................................................ II.5-145.6.1 Plotting positions and probability plot ..................................................................... II.5-145.6.2 Goodness-of-fit tests ................................................................................................ II.5-145.6.3 Information criteria .................................................................................................. II.5-145.7 Rainfall frequency analysis ......................................................................................................... II.5-155.7.1 Assessment of rainfall data for frequency analysis ..................................................... II.5-155.7.2 At-site frequency analysis of rainfall .......................................................................... II.5-165.7.3 Regional rainfall frequency analysis .......................................................................... II.5-185.7.4 Frequency analysis of area-averaged rainfall ............................................................. II.5-195.7.5 Storm rainfall analysis for hydrological design applications ...................................... II.5-205.8 Low-flow analyses .................................................................................................................... II.5-315.8.1 General ................................................................................................................... II.5-315.8.2 At-site low-flow frequency analysis ........................................................................... II.5-325.8.3 Low-flow frequency estimation at partial-record sites using base-flow measurements.. II.5-345.8.4 Regionalization of low-flow frequency statistics ........................................................ II.5-355.8.5 Flow-duration curves ............................................................................................... II.5-365.9 Frequency analysis of flood flows .............................................................................................. II.5-385.9.1 Regionalization of flood flows .................................................................................. II.5-385.9.2 Homogeneous region delineation ............................................................................ II.5-395.9.3 Regional flood estimation methods .......................................................................... II.5-405.9.4 At-site and regional flow–duration–frequency approach .......................................... II.5-425.9.5 Combination of single-site and regional data ........................................................... II.5-425.9.6 Flood frequency analysis and climate variability ....................................................... II.5-435.10 Estimating design floods ........................................................................................................... II.5-445.10.1 General ................................................................................................................... II.5-445.10.2 Design floods ........................................................................................................... II.5-455.10.3 Data preparation ..................................................................................................... II.5-475.10.4 Design flood computation techniques ..................................................................... II.5-485.10.5 Flood hydrograph conceptual models ...................................................................... II.5-515.10.6 Snowmelt contribution to flood ............................................................................... II.5-525.10.7 Calculating discharges from urban drainage systems ............................................... II.5-525.10.8 Risk .......................................................................................................................... II.5-52References and further reading ............................................................................................................. II.5-53CHAPTER 6 – MODELLING OF HYDROLOGICAL SYSTEMS ................................................................6.1 Mathematical deterministic models .......................................................................................... II.6-16.1.1 Black box models ..................................................................................................... II.6-26.1.2 Artificial neural networks .......................................................................................... II.6-3PageII.6-1

CONTENTS6.1.3 Conceptual models .................................................................................................. II.6-46.1.4 Distributed models .................................................................................................. II.6-86.1.5 Parameter evaluation ............................................................................................... II.6-106.1.6 Selection of models ................................................................................................. II.6-116.2 Time series and spatial analysis ................................................................................................. II.6-126.2.1 Stochastic simulation of hydrological time series ...................................................... II.6-136.2.2 Change detection in hydrological records ................................................................ II.6-146.2.3 Spatial analysis in hydrology .................................................................................... II.6-176.3 Modelling hydrological systems and processes .......................................................................... II.6-186.3.1 Introduction ............................................................................................................ II.6-186.3.2 Rainfall–runoff relationships ..................................................................................... II.6-196.3.3 Groundwater modelling .......................................................................................... II.6-256.3.4 Snowmelt models .................................................................................................... II.6-306.3.5 Streamflow routing ................................................................................................... II.6-396.3.6 Modelling other processes ....................................................................................... II.6-426.4 Modelling challenges ............................................................................................................... II.6-486.4.1 Accuracy and availability of input data ..................................................................... II.6-486.4.2 Ungauged basins ..................................................................................................... II.6-486.4.3 Coupling of models ................................................................................................. II.6-49References and further reading ............................................................................................................. II.6-50ixPageCHAPTER 7 – HYDROLOGICAL FORECASTING ...................................................................................II.7-17.1 Introduction to hydrological forecasting .................................................................................... II-7-17.1.1 Scope ...................................................................................................................... II-7-17.1.2 Hydrological forecast operations .............................................................................. II-7-17.1.3 End-to-end hydrological forecasting systems ........................................................... II-7-27.1.4 Uncertainty and probabilistic forecasts ..................................................................... II-7-37.1.5 Dissemination of forecasts and warnings .................................................................. II-7-47.1.6 Decision support ..................................................................................................... II-7-67.1.7 Cooperation with the National Meteorological Service ............................................ II-7-77.2 Data requirements for hydrological forecasts ............................................................................ II-7-77.2.1 General .................................................................................................................. II-7-77.2.2 Data required to establish a forecasting system ........................................................ II-7-87.2.3 Data required for operational purposes .................................................................... II-7-97.3 Forecasting techniques .............................................................................................................. II-7-107.3.1 Requirements for flood forecasting models .............................................................. II-7-107.3.2 Flood forecasting methods ...................................................................................... II-7-117.3.3 Model updating techniques ..................................................................................... II-7-147.3.4 Forecast verification ................................................................................................. II-7-157.4 Forecasting flash floods ............................................................................................................ II-7-167.4.1 National flash flood programmes ............................................................................. II-7-167.4.2 Local flash flood systems .......................................................................................... II-7-177.4.3 Wide-area flash flood forecasts ................................................................................. II-7-197.4.4 Flash flood guidance ................................................................................................ II-7-197.4.5 Dam-break flash flood forecasting ............................................................................ II-7-217.4.6 Storm surges in rivers .............................................................................................. II-7-217.4.7 Urban flooding ........................................................................................................ II-7-227.4.8 Flooding from local drainage ................................................................................... II-7-227.5 Long-term forecasting .............................................................................................................. II-7-237.5.1 Water supply forecasting ......................................................................................... II-7-237.5.2 Flow recession forecasting ....................................................................................... II-7-247.5.3 Time-series analysis .................................................................................................. II-7-257.6 Snowmelt forecasts .................................................................................................................. II-7-257.6.1 General ................................................................................................................... II-7-25

xGUIDE TO HYDROLOGICAL PRACTICES7.6.2 Snowmelt runoff processes in lowland and mountain rivers ..................................... II-7-257.6.3 Short- and medium-term snowmelt runoff forecasts ................................................ II-7-267.6.4 Long-term snowmelt forecasts ................................................................................. II-7-267.7 Forecasts of ice formation and break-up ................................................................................... II-7-277.7.1 General ................................................................................................................... II-7-277.7.2 Long-term ice forecasts ............................................................................................ II-7-287.7.3 Ice jams and methods of forecasting high water levels ............................................. II-7-28References and further reading ............................................................................................................. II-7-31Page

PREFACEIn September 2000, world leaders agreed to theUnited Nations Millennium Declaration, from whichwas soon derived a set of eight time-bound andmeasurable goals and targets for combating poverty,hunger, disease, illiteracy, environmental degradationand gender inequality. These eight objectivesare known as the United Nations MillenniumDevelopment Goals (MDGs). The attainment of eachof these Goals depends, to a great extent, on theavailability of appropriate freshwater and on theprotection of the population from the ravages offlooding. This, in turn, places a major responsibilityon the National Hydrological andHydrometeorological Services to support the necessaryactions at the national level, in the face ofever-increasing demands on the limited freshwaterresources available to WMO Members. In transboundarybasins, in particular, where concerns areoften driven by the need for equitable distribution ofthese limited resources, appropriate operationalmechanisms to share them may have to be establishedand maintained among the relevant ripariancountries.One of the purposes of the World MeteorologicalOrganization (WMO) is to promote the standardizationof meteorological and hydrological observationsand to ensure uniform publication of observationsand statistics. With this objective, the WorldMeteorological Congress has traditionally adoptedTechnical Regulations laying down the meteorologicaland hydrological practices and procedures to befollowed by Members of the Organization. TheseTechnical Regulations (WMO-No. 49) are supplementedby a number of manuals and guidesdescribing in more detail the practices and proceduresthat Members are requested or invited tofollow, respectively, in monitoring and assessingtheir respective water resources. It is therefore hopedthat improved uniformity and standardization inhydrological practices and procedures will alsocontribute to enhanced collaboration among WMOMembers and further facilitate regional and internationalcooperation.The aim of the Guide to Hydrological Practices is toprovide the relevant information on current practices,procedures and instrumentation to all thoseengaged in the field of hydrology, thereby enablingthem to carry out their work more successfully.Complete descriptions of the theoretical bases andthe range of applications of hydrological methodsand techniques are beyond the scope of this guide,although references to such documentation areprovided wherever applicable. Detailed proceduresfor monitoring hydrological parameters are dealtwith in the specific WMO manuals.It is hoped that this guide will be of use, not only toMembers’ National Services, but also to various otherstakeholders and agencies involved in water resourcesmanagement in general, and in water resourcesmonitoring and assessment in particular. The WMOCommission for Hydrology (CHy) has thereforedecided to make this guide a “living” document,which will be updated periodically and posted onthe Internet. This Guide will also represent one ofthe building blocks of the WMO Quality ManagementFramework – Hydrology, which is currently beingdeveloped in order to support Members and theirNational Services by ensuring that the activities theyundertake, such as hydrological data acquisition ordelivery of services and products, are indeedperformed efficiently and effectively. Users of theGuide are therefore invited to continue providingtheir comments and suggestions for its furtherimprovement.The Guide to Hydrological Practices is published inEnglish, French, Russian and Spanish. However, aswith previous versions, several Members haveannounced their intention to translate this Guideinto their national languages.It is a pleasure to express my gratitude to the WMOCommission for Hydrology for taking the initiativeto oversee the revision of the Guide to HydrologicalPractices.(M. Jarraud)Secretary-General

CHAPTER 1INTRODUCTION1.1 BACKGROUNDHydrology is the science that deals with the occurrenceand distribution of the waters of the Earth intime and space, both above and below the landsurface, including their chemical, biological andphysical properties, and their interaction with thephysical environment (WMO/UNESCO, 1992). Itprovides an understanding of various phases ofwater as it passes from the atmosphere to the Earthand returns to the atmosphere. As such, it forms thebasis for water resources assessment and managementand the solution of practical problems relatingto floods and droughts, erosion and sediment transportand water pollution. Increasing stress on theavailable water resources in the search for improvedeconomic well-being and concerns for the pollutionof surface water and groundwater havehighlighted the central role of hydrology in allwater and environment initiatives.To provide guidance in monitoring this vitalresource, which is central to the developmentand well-being of humankind, the WorldMeteorological Organization (WMO) Commissionfor Hydrology, at its first session (WashingtonDC, 1961), recognized the urgent need for thepreparation of a guide to the relevant operationalpractices. As a result, the first edition waspublished in 1965 as the Guide to HydrometeorologicalPractices.The second and third editions of the Guide werepublished in 1970 and 1974, respectively. The thirdedition was entitled Guide to Hydrological Practices inrecognition of the broader scope of its contents.Subsequently, during its fifth session (Ottawa,1976), the Commission approved the revision ofand substantial additions to the Guide to produce afourth edition, which was issued in two volumes.Volume I dealt with data acquisition and processingand Volume II with analysis, forecasting and otherapplications. Volumes I and II of the fourth editionwere published in 1981 and 1983, respectively.With the evolution of technology and the Hydrologyand Water Resources activities within WMO, thefifth edition of the Guide was published in 1994 asone consolidated volume. It was also published ona CD-ROM for easy outreach to a wider watermanagement community beyond the traditionalWMO constituency.In 1999, the World Meteorological Congressadopted “Weather, Climate and Water” as the officialsubtitle of the Organization. The followingyear, the Commission for Hydrology, at its eleventhsession in Abuja, Nigeria, recommended that thesixth edition of the Guide be published as a livedocument to be uploaded to the Internet andupdated more frequently, as and when required.1.2 SCOPEThe accepted principles of integrated water resourcesmanagement dictate that, in order to achieveenvironmental sustainability and economicproductivity, rivers must be managed at the basinlevel. Today, when water is perceived to be a matterof universal concern, various stakeholders, at thenational as well as at international level, participateand play important roles in the process. Manyinstitutions and agencies within a country areengaged in the collection of hydrological data andinformation. These data may be collected by variousagencies using different measurement procedures.The resulting lack of homogeneity in theobservations gives rise to a lack of confidence. It isimperative, therefore, that all these partners bemade aware of the manner in which the hydrologicaldata are collected, the limitations and the reliabilityof the data, and how they are to be managed by theresponsible organizations in the basin. Transparencyin data collection, storage and sharing is an essentialelement for cooperation among various users. Aquality management framework for hydrometryand hydrological information is fundamental inusing hydrological information from diversesources.The growing demand for freshwater resources hasincreasingly focused the attention of governmentsand civil society on the importance of cooperativemanagement. Sharing the benefits of cooperationand even conflict prevention stem from a broadunderstanding of the principles and mechanismsthrough which these results can be achieved.Transboundary rivers have the potential to bringcountries together both economically and politicallyor, conversely, they can cause economic andpolitical tensions. The risk factor in decisionmakingin water resources management is a

II.1-2GUIDE TO HYDROLOGICAL PRACTICESfunction of hydrological variability. The risks canbe mitigated through cooperative management oftransboundary rivers. Cooperation in transboundaryriver management is fundamentally a politicalactivity. Allocation of the resources or distributionof the benefits is essentially dependent on theknowledge of water availability and the relatedhydrological variability. A shared and acceptedknowledge of the resources, their projected availabilityand the confidence in their accuracy greatlyhelp in assessing the feasibility and fairness ofalternative management and investmentscenarios.A lack of homogeneity in the data on the land phaseof the hydrological cycle limits the scientific capacityto monitor changes relevant to climate and todetermine the causes of variability and change inthe hydrological regime. River discharge has a rolein driving the climate system, as the freshwaterflows into the oceans may influence thermohalinecirculation. For easy and reliable use, the quality ofsuch data and the procedures for its acquisition,storage and exchange should in general followcertain specified standards and protocols.All of these factors increased the need for ensuringthe quality of hydrological data. WMO, with avision to provide expertise in international cooperationin weather, climate, hydrology and waterresources, issues international guidance materialand standards, and it is hoped that this Guide willform an important link in the quality managementframework for hydrological practices. To meet suchrequirements, continuing efforts have been madeto expand and improve the Guide, now in its sixthedition. It is expected that this Guide will be usefulto agencies – not only to National HydrologicalServices, but also to other stakeholders.This Guide addresses all aspects of the land phase ofthe hydrological cycle, especially its phases uponand under the surface of the land. In conjunctionwith the manuals published by WMO, it providesdetailed information on those areas that fall withinthe scope of the hydrology and water resourcesactivities of the Organization designed to supportNational Hydrological Services and services with asimilar mission.The Guide forms part of an overall framework ofrecommended practices and procedures providedby Technical Regulations (WMO-No. 49) Volume III –Hydrology, as approved by WMO. Members areinvited to implement these recommended practicesand procedures in developing their HydrologicalServices and activities.1.3 CONTENTS OF THE GUIDEIt is difficult to set a clear dividing line between thescience of hydrology and the practice of waterresources planning and management. Nevertheless,for practical reasons, it was necessary to split theGuide into two volumes as shown in Figure II.1.1.Volume I, entitled Hydrology – From Measurementto Hydrological Information, deals with networks,instruments, methods of observation and primarydata processing and storage. It contains ten chapters,beginning with an introduction and an outlineof the contents in Chapter 1.Chapter 2, entitled Methods of observation,deals with the design and evaluation of hydrologicalnetworks and provides an overview ofinstruments and methods of observation forvarious hydrological elements that are describedin detail in the subsequent chapters. Precipitationmeasurement in Chapter 3 is covered in all itsaspects, ranging from the location of raingaugesto the observation of precipitation by remotesensing.The chapter covers liquid and solidprecipitation, including their quality. Chapter 4,Evaporation, evapotranspiration and soil moisture,addresses both direct and indirect methodsand also briefly reviews methods for evaporationreduction.Chapter 5, Surface water quantity and sedimentmeasurement, is pivotal and deals with measurementof flow in rivers and the capacity of lakes andreservoirs. It is also concerned with the measurementof sediment discharge. This subject matter isdiscussed in greater detail in the Manual on StreamGauging (WMO-No. 519) and the Manual onOperational Methods for the Measurement of SedimentTransport (WMO-No. 686), to which the reader isinvited to refer for more information.Chapter 6, which is entitled Groundwater, isconcerned with measurements from wells and thehydraulic properties of aquifers. It also looks insome detail at various remote-sensing techniquesfor groundwater observation.The development of water resources is not onlyconstrained by their quantity but also by theirquality. Accordingly, Chapter 7, Water quality andaquatic ecosystems, addresses subjects ranging fromsampling methods to remote-sensing. Chapter 8,Safety considerations in hydrometry, discussestopics ranging from the safety of personnelperforming the measurements to safeguardingrecording stations and the samples collected.

CHAPTER 1. INTRODUCTIONII.1-3Network designInstrumentsMethods ofobservationNetwork operationData transmissionFeedbackHistorical andreal-time dataHydrological forecastingPrimary and secondary dataprocessing and storageSecondary data processingData analysisUsersWater resourcesplanningHistorical dataDesign dataderivationReal-timedataWater managementsystems operationHistorical and real-time dataOther usesNavigationAgriculturePowerproductionIndustryEcologyFigure II.1.1. Hydrological systemLastly, Chapters 9 and 10, Data processing andquality control and Data storage, access anddissemination, respectively, include the disseminationof data for use by the wider watercommunity.Volume II deals with the application of theinformation referred to above in hydrologicalforecasting and the planning and design of variouswater projects. Accordingly, the volume is entitledManagement of Water Resources and Applicationof Hydrological Practices. It consists of sevenchapters beginning with an introduction andoutline of the contents in Chapter 1.Chapter 2 provides guidance on the managementof hydrological services, including human resourcesaspects and financial and asset management.Chapter 3 introduces integrated water resourcesmanagement and emphasizes the vital role of qualityhydrological data in addressing various complexwater management issues. Chapter 4 highlights theuse of hydrological information in applications towater management, namely estimating reservoir

II.1-4GUIDE TO HYDROLOGICAL PRACTICEScapacity and yield, flood management, irrigationand drainage, hydropower and energy-relatedprojects, navigation and river training, urban waterresources management, sediment transport andriver channel morphology and environmentalissues. Chapter 5 deals with extreme value analysis,and Chapters 6 and 7 address the modelling ofhydrological systems and hydrological forecasting,respectively, as two of the key functions ofHydrological Services in water management.While a measure of standardization is desirable andcan be achieved with respect to instruments, methodsof observation and publication practices, this israrely the case with respect to hydrological analysisand applications. Therefore, the emphasis inVolume II is on presenting alternative approachesto the solution of selected problems, which havebeen demonstrated through experience to be bothpractical and effective. Rather than recommendingone approach or technique in preference to another,attention is drawn to the principal features andadvantages of each approach. The final choice willdepend on a multitude of factors, including therelevant hydrological and climatic regimes, availabledata and information and the purposes to beserved, and can only be made in the light of a fullunderstanding of a specific situation. During thepast few years, the increasing availability of microcomputershas permitted the introduction of moresophisticated analytical methods and techniques.Some of these have now been widely adopted andhave therefore been introduced into this Guide.The space limitations of this Guide restrict the amountof material that can be presented. For more detailedinformation on the subjects discussed, the readershould consult the following publications: fordischarge measurement, the Manual on Stream Gauging(WMO-No. 519, Volumes I and II) and on sampling,the GEMS/Water Operational Guide (UNEP, 2005). Thereader is also referred to international standards dealingwith methods for liquid flow measurements inopen channels prepared by member countries of theInternational Organization for Standardization (ISO).ISO has developed more than 50 standards for varioustypes and methods of measurement. Valuable referencescan also be found in the proceedings of theinternational symposiums, seminars and workshopson hydrometry organized by the InternationalAssociation of Hydrological Sciences (IAHS), WMOand the United Nations Educational, Scientific andCultural Organization (UNESCO).A full description of the theoretical base for therecommended practices and detailed discussion oftheir methods of application are beyond the scopeof this Guide. For such details, the reader is referredto appropriate WMO manuals and technical reports,as well as to other textbooks, handbooks and technicalmanuals of national agencies. In particular,further detailed guidance on instruments and methodsof observation is given in the Guide toMeteorological Instruments and Methods of Observation(WMO-No. 8) and the Guide to ClimatologicalPractices (WMO-No. 100).References and suggestions for further readingappear at the end of each chapter.1.4 THE HYDROLOGICAL OPERATIONALMULTIPURPOSE SYSTEMIn recent decades, hydrological science and technologyhave made substantial progress and significantcontributions have been made by field hydrologiststo the development and management of waterresources. So as to facilitate the sharing of hydrologicalpractices among the National HydrologicalServices, a technology transfer system known as theHydrological Operational Multipurpose System(HOMS) was developed by WMO and has been inoperation since 1981. It offers a simple but effectivemeans of disseminating information on a wide rangeof proven techniques for the use of hydrologists.HOMS transfers hydrological technology in the formof separate components. These components can takeany form, such as a set of drawings for the constructionof hydrological equipment, reports describing awide variety of hydrological procedures and computerprograms covering the processing and storage ofhydrological data, as well as modelling and analysisof the processed data. To date, over 180 componentshave been made available, each operationally usedby their originators, thus ensuring that every componentserves its purpose and has been proved inpractice. These descriptions appear in the HOMSReference Manual (HRM) which is available online athttp://www.wmo.int/pages/prog/hwrp/homs/homs_en.html in English, French, Russian andSpanish. The present Guide is further enrichedthrough cross-references to the relevant HOMScomponents, which are included at the beginning ofthe relevant sections of this Guide.References and further readingUnited Nations Environment Programme GlobalEnvironment Monitoring System (GEMS)/WaterProgramme, 2005: Global Environment MonitoringSystem (GEMS)/Water Operational Guide. Fourth

CHAPTER 1. INTRODUCTIONII.1-5edition, Inland Waters Directorate, Burlington,Ontario.World Meteorological Organization, 1980: Manualon Stream Gauging. Volumes I and II, OperationalHydrology Report No. 13, WMO-No. 519, Geneva.———, 1983: Guide to Climatological Practices. Secondedition, WMO-No. 100, Geneva.———, 1988: Technical Regulations. Volume III,Hydrology, WMO-No. 49, Geneva.———, 1989: Manual on Operational Methods for theMeasurement of Sediment Transport. OperationalHydrology Report No. 29, WMO-No. 686, Geneva.———, 1994: Guide to Hydrological Practices. Fifth edition,WMO-No. 168, Geneva.———, 1996: Guide to Meteorological Instruments andMethods of Observation. Sixth edition, WMO-No. 8,Geneva.———, 2000: Hydrological Operational MultipurposeSystem (HOMS) Reference Manual. Second edition,Geneva.———/United Nations Educational, Scientific andCultural Organization, 1992: InternationalGlossary of Hydrology. WMO-No. 385,Geneva.

CHAPTER 2HYDROLOGICAL SERVICES2.1 INTRODUCTIONMost Hydrological Services operate in the publicsector and are therefore influenced by trends ingovernmental policy and practice. What is more,they work in a rapidly evolving environment characterizedby the following factors:(a) Heightened global commitment to sustainablemanagement of natural resources and the environment,combined with efforts to improve theliving conditions of the poor, who generally arethe most dependent on natural resources;(b) An expanding emphasis on the need for integratedwater resources management, as pressureon the world’s water and other natural resourcescreates a general awareness that resources mustbe developed and managed in a sustainablemanner;(c) A seemingly inexorable increase in the impactwrought by natural disasters, particularly floodsand droughts. At the same time, risk managementis becoming more widely adopted;(d) Increased investments by Hydrological Servicesin capital assets and staff retraining, which isthe cost of offering new or improved products;(e) A mounting expectation that public servicesshould be accountable not only to electedrepresentatives, but also to the public at large.Public services should conduct their affairswith efficiency, effectiveness and economy.This expectation can culminate in the threatof litigation when the general public feels letdown by government agencies;(f) Keener competition for resources in the publicsector, as governments seek to reduce taxationwhile meeting rising public expectations;(g) The growing impact of globalization, whichaffects Hydrological Services both directly andindirectly;(h) The impact of socio-economic trends on theday-to-day operations of Hydrological Services,such as the increasing involvement of womenin professional activities and the ever-growinguse of the Internet or the Web-based delivery ofhydrological data and products.Effective resource management requires accurateinformation as a basis for planning, implementingand monitoring resources. However, to be fullyimplemented, integrated water resources managementdemands a wide range of hydrological andrelated information, which may not be easily available.In order to obtain such information, aNational Hydrological Service (NHS) requiresproper institutional development to meet thesenew challenges and must develop appropriatecapabilities and/or establish partnerships or strategicalliances with complementary agencies.The general public constantly seeks higher-qualityhydrological information and products. Such highqualityproducts and processes can provide acompetitive edge for Hydrological Services that arerequired to compete in the marketplace.Maintenance of verifiable standards are also necessaryand thus, as a general rule, a HydrologicalService should make quality management a focalpoint of its business, and its director should takeultimate responsibility for the quality of hydrologicalproducts.Public agencies are continually required to do morewith less, and often are obliged to recover someoperating costs and seek commercially profitableprojects in order to reduce the burden on thetaxpayer.As a result, the director of a Hydrological Servicemust constantly monitor changes in its environmentso as to provide appropriate responses.2.2 RESPONSIBILITIES AND FUNCTIONSOF HYDROLOGICAL SERVICES[HOMS A00]2.2.1 Nature of the products and servicesof a Hydrological ServiceHydrological data and information are, by andlarge, excludable public goods because the marginalcost of supplying the data to an additional client ispractically zero, and yet access to the informationcan be controlled. A Hydrological Service maytherefore choose to allow open access to some information,for example, making it accessible on theInternet, which places it irrevocably in the publicdomain, and to restrict access to other information,for example, only releasing it to selected clients,under strict contractual conditions regarding subsequentrelease. In this sense, services such as the

II.2-2GUIDE TO HYDROLOGICAL PRACTICESissuing of warnings to the public are pure publicgoods.A National Hydrological Service can provide hydrologicaldata and information at the national level ina cost-effective way. The implications of this are asfollows:(a) Products or public services, such as publicwarnings, can only be provided through publicfunding, because the Hydrological Serviceconcerned cannot easily recover its costs fromthe beneficiaries;(b) To obtain or sustain funding from governmentsources for the provision of both pureand excludable public goods, it is necessary todemonstrate their value or merit to society;(c) Products or services that are excludable publicgoods can be provided on the basis of profitabilityor recovery of costs from the beneficiaries.Authority may be required for such operations.Proper accounting practices, financial transparencyand fair charges will be expected;(d) Boundaries between the management of purepublic goods and excludable public goods mayshift as a result of, for example, evolving technology,contractual arrangements and publicinformation. A Hydrological Service may beable to influence such boundaries, if it is in thenational interest;(e) When lobbying for sources of funding, managersshould review their products and services toensure that they are in line with the mandateand pricing structure of the NHS.2.2.2 Clients of hydrologicalproducts and servicesWho are the clients of a Hydrological Service? Inprinciple, for a National Hydrological Service, theultimate client is the general public, represented byelected officials at the national, state/provincial andlocal levels. This includes the general public of thefuture, who will be the beneficiaries of hydrologicalwork that is being done today.Governmental policies and national developmentgoals, and the information required to supportthem, influence an NHS fundamentally. For example,in many developing countries the beginningof the twenty-first century has seen a growingnational emphasis on poverty alleviation. Themanagement of a Hydrological Service shouldmonitor governmental policies and analyse theimplications of these policies for the individualService. What products and services will theHydrological Service need to provide in order tosupport national policies and goals? Do the Service’scurrent products and services contribute to this? Inother words, the management should ensure thatthe Service’s products and services have the greatestpossible value. This is best done objectively bymeans of cost-benefit, cost-effectiveness andpoverty analyses, among others. The public’s interestmay be varied, so that a Hydrological Servicemay have a variety of clients, in addition to thetraditional ones. A Hydrological Service may alsooffer private services to clients who are prepared topay for them. The range of such clients will varyfrom country to country, depending on the natureof the national economy.The management of a particular HydrologicalService must implement frequent surveys to identifypotential clients. The management shouldcontinually monitor trends in demand for water,national and provincial policies and developmentgoals, political manifestos, trends and events invarious economic sectors, as well as internationalagreements and agreements with donor agenciesand other development partners.Client expectations of all businesses, includingthose in the public sector, are rising continuallyand businesses must continually seek to meet orexceed those expectations. Hydrological Servicesare no exception. To ensure the future of the Service,managers must encourage a client focus amongtheir staff. The single most important client is theperson to whom the director reports, for instance,the Minister for the Environment. The future of aHydrological Service depends on how successfullythe director markets the Service to that person anddemonstrates how the Service can be useful to theclient.A marketing strategy should meet the followingaims:(a) Identify the current and potential clients of theHydrological Service, and maintain and updatea client database;(b) Identify the products and services required bythe clients which the Service can provide;(c) Identify the most suitable mode or place ofdelivery of the product or service to the client,for example, use of the Internet to provideaccess to real-time data, fax warnings andconventional written reports with data annexeson CD-ROM;(d) Determine a pricing policy for different productsand services, and for different clients;(e) Specify the types of people involved in deliveringthe product or service;(f) Characterize the processes of product or servicedelivery according to the needs of the clients;

CHAPTER 2. HYDROLOGICAL SERVICESII.2-3(g) Promote the Hydrological Services wherepotential clients can be clearly identified andcontacted directly.2.2.3 Managing relationships with clientsA crucial element of running any business is maintainingeffective relationships with clients. Goodcommunication with clients is essential in ensuringthat they be made aware of the HydrologicalService’s capabilities. Good public relations andhonest feedback are necessary for client satisfaction.As members of the general public constitutethe bulk of the clientele of a public-sectorHydrological Service, it is essential that the publicbe kept informed of activities and outputs, and beprovided with opportunities for feedback. A HydrologicalService should have a high profile, that is, bevisible, and ensure that the public be made aware ofits work. World Water Day, held on 22 March ofeach year, provides such an opportunity.advice on an appropriate response to a contaminantspill on a major river, or on how torespond to an evolving drought.The management of a Hydrological Service shouldseek to develop added-value products and services,and to move out of the “data trap”, in which theService merely provides data from which otherpeople extract value. Capacity-building, in terms ofstaff skills, information management technology,quality assurance and marketing will be necessary.Other changes may have to be made to institutionalarrangements as well, for example, permitting aService to retain the revenue that it generates. Theproducts and services offered by a HydrologicalService have value and therefore are economicgoods. Hydrological Service staff should learn topackage their products to meet the needs of clients.They should also realize that client needs changewith the evolving climatic and economicsituation.2.2.4 Hydrological products and servicesThe basic products of a Hydrological Service arewater-related data and information. Data andinformation are of value for decision-making.Hence, a Hydrological Service might be seen asproviding increased confidence or reduced risk toits clients, as they make water-related decisions. Infact, one measure of the value of these data andinformation is their impact on the decisions thatare made.There is a continuum in the products that a Servicemight provide:(a) Water-related data and observations obtainedfrom an observing network. Hydrologicaldatabase management systems provide basicstatistics such as daily, monthly, seasonal andannual means or maxima, which are useful toclients;(b) Water-related information, such as a comprehensiveassessment of national water resources,the statistics of flood events or maps of spatial/temporal trends in groundwater quality;(c) A monitoring service, designed to provide veryspecific data or information at a particularlocation for a particular client, for example, toindicate when the dissolved oxygen concentrationdownstream from an outfall falls below aspecified minimum value;(d) Knowledge and understanding of water-relatedphenomena and water resources;(e) Advice on decision-making, where informationis developed into recommendations forresponses to certain conditions, for instance,2.2.5 Functions and activities of aHydrological ServiceThe functions of a Hydrological Service shouldreflect the products and services required by theclient. The Technical Regulations (WMO-No. 49) setout the core functions of a Hydrological Service inVolume III – Hydrology, D.1.1, 8.3. These includethe following activities: developing standards andquality assurance programmes; designing andoperating observation networks; collecting,processing and preserving data; assessing userrequirements for water-related data and information;and providing such data and information,for example, hydrological forecasts and waterresources assessments.Hydrologists today need a much broader view ofhydrology, including ecological, biological andhuman-use aspects of the aquatic system.Accordingly, the activities of many HydrologicalServices are becoming increasingly diverse as theydeal with different types of data and information.Hydrological Services should continually monitorchanging demands for water-related data, informationand advice so that they can allocate resourcesappropriately. An early start is always desirable,because baseline measurements and trend informationwill be required for many purposes in thefuture.Special national circumstances might require additionalbasic activities, such as monitoring riverchannel erosion and migration, or reservoirsedimentation.

II.2-4GUIDE TO HYDROLOGICAL PRACTICESThe functions and activities of a HydrologicalService are not fixed in time, but change inresponse to the evolving needs and expectationsof society and according to technological developments.The management of a Hydrological Serviceshould continually monitor changes in its functionalenvironment and assess their implicationsfor the Service. For example, in recent years, theactivities of some Hydrological Services havechanged significantly in response to the followingdevelopments:(a) Recognition of the hydrological significanceof climate change, bringing about a newemphasis on drought monitoring,forecasting of extreme events and time-seriesanalysis;(b) The near-universal adoption of computerdatabase management systems, leading to thepublication of hydrological yearbooks anddissemination of hydrological data and productsin electronic form;(c) Cooperative agreements among the riparianStates in transboundary river basins;(d) Adoption of regional political agreements,with consequent changes and adaptations ofstandards, regulations and directives to whichparticipating countries must abide, such as theEuropean Union Water Framework Directive2000/60/EC (EC, 2000), which has broughtconsiderable change to Hydrological Servicesin both the European Union and potentialmember States.In the past, the fundamental activity of aHydrological Service was to design and operate abasic network of observing stations. This enabled anational assessment to be made of the country’swater resources, thus providing a basic set of data tomeet future needs for data at all locations and for awide range of purposes. This also called for thetechnical capacity to provide estimates at locationsfor which there were no field data at all.It is becoming difficult to sustain the concept of abasic national network in many countries. Insome countries, promotion of integrated waterresources management on a river basin basis,excellent though this concept may be, has led tomonitoring efforts being focused on particularapplications of data at the expense of nationalcoverage. Therefore, it is important to demonstratethe benefits of a comprehensive, integratedHydrological Service, in which broadly based datacollection is more economical, both in the operationof monitoring networks and datamanagement, as well as in the assessment of waterresources.A Hydrological Service may also engage in theprovision of private services. Examples include thefollowing:(a) Compulsory monitoring of incoming waterquality below a wastewater outfall for a factory;(b) Monitoring of reservoir inflows and powerstation outflows for a hydropower company;(c) Water-related data required for an environmentalimpact assessment for private use;(d) Provision of information for a private irrigationcompany;(e) Groundwater bore monitoring for a watersupply authority.These may require the establishment of specialpurpose networks or project networks (or individualstations) to meet specific client needs. The clientwould cover the cost and own the products, whichcould only be archived or disseminated upon theclient’s request. The management of a HydrologicalService should seek such opportunities. Specialprojects have many benefits for a HydrologicalService, including increased revenue, spreading ofoverhead costs across a wider range of clients,opportunity to develop new skills, heightenedprofile and support, and increased innovation. Thedata produced by these activities could also be integratedinto information generated from the basicnational network.2.2.6 Evaluation of products and services,and quality managementThe final stage in marketing is to obtain feedbackfrom clients on the products and services. There aremany ways of obtaining feedback. Perhaps thesimplest is a friendly telephone call a few days afterthe product has been delivered, to enquire whetherit has met expectations. More formally, clients maybe asked to complete a simple questionnaire.Possibly the ultimate evaluation tool is to arrange aclient satisfaction survey. Essentially, the aim ofsuch a survey is to determine the original expectationsof the client and measure the extent to whichsuch expectations have been met. One way ofensuring customer satisfaction is to establish a qualitymanagement framework guaranteeing thedevelopment of products according to precise,replicable and agreed procedures and standards.Defined standards are an essential basis for qualityassurance of a Hydrological Service’s products andservices. Increasingly, clients require knowledge ofthe standards that are being achieved by the Serviceto satisfy their clients. In general, standards may bespecified for the procedures that are used by theService, and for the attributes of the products that

CHAPTER 2. HYDROLOGICAL SERVICESII.2-5the Service produces. It is important to rememberthat standards are needed, not only for technicalactivities related to hydrometric data collection andprovision of metadata, but also for all other activitiesundertaken within the Service, such as finances,staff performance and long-term planning. A reviewof International Organization for Standardization(ISO) standards for metadata from a WMO perspectivecan be found at http://www.wmo.int/pages/prog/www/WDM/reports/ET-IDM-2001.html.Quality management should be carried out in asystematic way. In other words, a HydrologicalService should have a quality management systemin place that assures clients that its products andservices meet the standards of quality that havebeen defined for them. A Service may find thatoperating a well-documented quality managementsystem can also be of great assistance should itbecome involved in legal proceedings related to itsdata and information products.A comprehensive quality management system isoften perceived as being expensive to implement.In practice, however, a quality management systemshould be no different from the data/productmanagement procedures that the HydrologicalService uses to make measurements, convey themto the office, process and archive them, and transmitthem to clients. Carrying out these proceduresefficiently requires the following:(a) Documented procedures for each step of thedata and information flow;(b) Defined standards for measurement and processingprocedures, the measurements (data)themselves and derived products;(c) Staff training and overview;(d) Assigned responsibilities;(e) Clearly documented data.These elements of data management are alsocomponents of quality management. A comprehensivequality management system might includeadditional components such as the following:(a) Verification that standard procedures are beingfollowed, for example, by independent checkson flow-rating curves or field work;(b) Validation that archived data meet definedstandards, for example, by cross-comparisonbetween neighbouring stations;(c) Documented evidence that all aspects of thesystem are being consistently monitored,for example, a training record for each staffmember.Although the cost of quality, implicitly high quality,is commonly perceived to be high, the cost ofpoor quality may well be higher. A Service maydiscover that observations it had made over severalyears were worthless because of a hitherto unrecognizedfault in an instrument, or that it mustcompletely reprocess a flow record because a weirwas incorrectly rated. Such remedial measures incura much higher cost than would have been involvedin initially checking the instrument or the rating.2.2.7 Legal basis for operations andorganizational arrangementsAlmost all countries have Hydrological Services thathave been explicitly established by some form oflegal instrument or that carry out functions that areprovided for or are enabled by legislation. However,in some cases, such legislation does not establish aspecific agency or even identify a governmentagency that is required to collect hydrological informationin the process of discharging its otherresponsibilities. In such cases, authority may comefrom an annual appropriation of funds, rather thanthe establishment of a law to include hydrologicalactivities.A great variety of legal or quasi-legal instrumentsgiving varying degrees of authority is possible, forexample, a national water policy, statute or law,water code, decree, order or inter-ministerial agreement,depending on the system of government.WMO (1994) provides a number of case studies thatshow the diverse arrangements that are possible. Inmany countries, water resources are now managedunder the authority of a water law, a law establishinga water sector head office such as the NationalWater Resources Council, a law on environmentalprotection, a law on natural resources management,or similar statute. In these cases, the emphasis ofthe law is on aspects of resource management, suchas allocation, resource pricing, or administration ofpermits. Hydrology may receive only passing reference,perhaps by being granted the authority tocollect appropriate information.There is a marked trend towards establishing riverbasin agencies that have comprehensiveresponsibilities for water management, includingthe provision of water-related information. Suchagencies can now be found on every continent. Inmany cases, these river basins and their agencies aretransnational, such as the Zambezi River Authorityin Southern Africa. In some countries, there iscomplete coverage by river basin agencies, whereasin others only the principal rivers are covered.Management of water resources by river basinagencies or by subnational civil administrationsintroduces a need to harmonize standards,

II.2-6GUIDE TO HYDROLOGICAL PRACTICEScoordinate data exchange, avoid duplication andassure national interests, for example. These tasksmay be assigned to a national head office andcarried out by its secretariat, which may be providedby the NHS or be completely separate. Inter-agencycoordination and liaison are absolutely necessaryin such cases.The increasing complexity of decision-making inthe water sector with a variety of stakeholders andactors, often with conflicting interests, necessitatesa clear definition of the roles and responsibilities ofeach player. Furthering the main aim of integratedwater resources management requires that data andinformation be available to all participants.Therefore, an appropriate legal framework andsome form of legal instrument are desirable toprovide a basis for a Hydrological Service’s operations.In particular, such a legal framework may beneeded to provide authority for activities or functionssuch as traversing private property as part ofmaintaining a monitoring station; charging fees forthe delivery of products or services; requiringother organizations, including those ofthe private sector, to provide copies of their data foraddition to the national archives; ortransnational activities or liaison.When laws related to water are being revised,managers should seek to participate in the draftingprocess. In particular, they should try to implementmeasures that have been successful in other countriesand attempt to introduce these into theirnational legislation. Contacts with other organizations,such as WMO and its regional associationworking groups, will provide useful ideas. WMOpublications also provide practical guidance (seeWMO, 1994, 2001a).The functions of a National Hydrological Servicemay be undertaken by a National HydrometeorologicalService, by one main sectoralHydrological Service or by federal HydrologicalService overseeing various state or regionalHydrological Services.In a survey of 67 countries carried out in 1991(WMO, 2001a), four principal models were foundfor organizing Hydrological Services at the nationallevel. Some 51 per cent of those surveyed hadnational hydrological or hydrometeorologicalagencies ; 1 per cent, regional (subnational) hydrologicalor hydrometeorological agencies; 42 percent, both national and regional hydrological orhydrometeorological agencies; 6 per cent hadneither national nor regional hydrological orhydrometeorological agencies.The organizational arrangements of HydrologicalService are very diverse. Much depends on the legalsystem, governmental structure and stage ofeconomic development, and successful examplessuggest that effective operational hydrology can beconducted under a variety of circumstances.Although managers of a Hydrological Service mayhave limited influence on organizational arrangementsat the national level, they should take everyopportunity to participate in organizational restructuring.They should draw on the experience ofHydrological Service managers in other countries inorder to propose changes that will help improve theperformance of their Service.At the level of an individual Hydrological Service,the organizational structure will largely depend onthe Service’s functions, products and activities. Asthese are always evolving, so too, should the structureevolve.Managers of a Hydrological Service should draw onthe experience of other Services when consideringappropriate organizational structures. Extensiveinformation on the relative merits of differentorganizational models, such as pyramidal structuresor flat structures, is always useful.In a country with several Hydrological Services,defined standards are of particular importance inensuring comparability of hydrological data andproducts. A key role of the NHS or lead HydrologicalService is to establish national standards. The samecould be said of an international river basin inwhich there are several NHSs. In this case, a key rolefor a river basin organization would be to establishstandards for the whole basin and assist NHSs inachieving them.The Technical Regulations, Volume III – Hydrology(WMO-No. 49) provides a set of long-establishedtechnical standards, as does the ISO Handbook 16 –Measurement of Liquid Flow in Open Channels(ISO, 1983).The technical standards adopted by a HydrologicalService provide an objective basis for performancemonitoring and appraisal. These standards shouldbe incorporated into a Service’s objectives.2.2.8 Managing relationships withother institutionsWater is vital to many sectors of the economy, andmany governmental as well as non-governmentalorganizations are likely to have an interest in water.Indeed, most countries have several organizations

II.2-8GUIDE TO HYDROLOGICAL PRACTICESregime in which only the transfer costs are levied,and this is essentially the principle expressed inResolution 25 (Cg-XIII). A number of HydrologicalServices have experimented with financial arrangementsfor data transfer in recent years, and thegeneral consensus seems to be that the approachadvocated by Resolution 25 (Cg-XIII) is preferred.In practice, the situation is more difficult in transboundaryriver basins, where issues of nationalsovereignty and national development outweighall others. In these circumstances, NHSs can onlyinsist that Resolution 25 (Cg-XIII) be followed.Many Hydrological Services consider it useful toprovide data to educational institutions and internationalscientific projects at no charge. On theother hand, if the data are to be used for consultingwork, there is no reason why a Hydrological Servicecannot require payment of a charge based on thecost of obtaining, verifying, storing and transferringthe data concerned.2.3 PLANNING AND STRATEGY[HOMS A00]Perhaps a director’s most important responsibilityis to implement the Hydrological Service’s planningand strategy development. To successfully respondto changing conditions and demands, a Serviceneeds a director with vision and the ability to implementactions. Planning and strategy developmentimply change. Few people like change, especiallywhen it is imposed on them, and the managers of aHydrological Service need skill in managing change.In particular, the organizational culture of manyServices may need to shift from one that has a technicalfocus to one that focuses first and foremost onclients.Managers of a Hydrological Service need plans andstrategies that ensure the Service allocates itsresources to achieve its most important goals.Diverse plans of different duration should beformulated to match identifiable goals. A strategicplan will provide a view of the overall direction ofthe Service, for example, for a period of five years.In times of rapid change, it is difficult to lookahead even five years; therefore the plan wouldneed to be updated regularly. An annual plan setsout the specific intentions and desired results to beachieved during a single year of operation; it isusually associated with a budget. A developmentplan focuses on the process of building a Service’scapacity to carry on its business, and may considera time period of 10 years or more for this purpose.In addition, there may be plans that focus onparticular aspects of the Service’s operations, suchas a staff training plan.A comprehensive plan is likely to include some ormost of the following elements:(a) Vision – how we want our world to be;(b) Mission – the reason for which the HydrologicalService exists;(c) Principles or values – the fundamental andunchanging beliefs that relate to the work ofthe Service;(d) Review of achievements during the last planningperiod;(e) Analysis of strengths, weaknesses, opportunitiesand threats (SWOT);(f) Goals and desired outcomes – broad statementsof what is to be achieved;(g) Objectives and desired outputs – specific targets:measurable results and standards, together witha time frame;(h) Actions – specific actions that will be used toachieve objectives and outputs;(i) Financial budget;(j) Performance criteria and indicators – measuresthat will be used to check progress.A strategic or long-term plan would not specifyactions and might include only an indicativebudget. An annual plan, however, might brieflysummarize many sections from an existing strategicplan and place more emphasis on defining theproposed actions and associated budget.The above-mentioned list commences with thehigh-level vision and mission statement, continueswith an appraisal of how the Service has performed,an honest evaluation of its condition (the strengthsand weaknesses of the Service) and of its businessenvironment (opportunities for new business andthreats from competitors or adverse changes in theenvironment), and then moves on to the specificationof actions and the means of measuring whetherthese have been successful. It is easy to obtain plansfrom other agencies to develop ideas on appropriateapproaches and formats. The WMO StrategicPlan (WMO-No. 1028) should, for example, beavailable to a Hydrological Service director and maybe obtained from the WMO Secretariat. Other servicesin the WMO group are an obvious source ofguidance; for instance, the Australian Bureau ofMeteorology (1995, 2005, 2006) has plans on arange of time scales, which might provide usefulexamples for other services.A Hydrological Service that is a component of aparent organization may have a strictly defined

CHAPTER 2. HYDROLOGICAL SERVICESII.2-9planning and budgeting format and process towhich managers would adhere. However, wherethere is more freedom, managers should take planningseriously. At times, where resources are lackingand it appears as though the Service receives norecognition or encouragement, planning may seemto be a pointless exercise. However, it is perhapsunder these conditions that planning is most necessaryto set a positive course for the future andprovide an impetus for change.A plan is not solely an internal document, but iscommonly used to promote the Service and as thebasis for a performance agreement or contractbetween the director and the senior official towhom the director reports. In this case, the planwill be negotiated with the senior official, as well asthe Service’s staff.Planning procedures are an essential component ofmanagement and are dealt with in many textbooksand all tertiary-level business managementprogrammes. Hydrological Service managers shouldmake planning a basic part of their business managementstudies.Planning need not be technical or time-consuming,although techniques such as discounted cashflow analysis to select the most promising ofseveral alternative courses of action can be used. Itis, perhaps, most important to effectively involvestakeholders in the process, that is, not only seniormanagement, but all the Service’s staff, clients andpotential collaborators. A mix of a top-down andbottom-up generation of ideas is desirable, facilitatedby consultation with clients and otherstakeholders. The director and senior managementshould set the overall direction of the Service, onthe basis of their understanding of the wider businessand political environment. Other staff mayhave a more hands-on perspective on strengthsand weaknesses, and personal links with clientsand collaborators. Commonly, individual departmentswill make proposals for components of theplan, which will be incorporated, modified oromitted according to the chosen selectionprocedure.A useful starting point for appraising the presentcondition of a Hydrological Service is the WaterResources Assessment: Handbook for Review of NationalCapabilities (WMO/UNESCO, 1997).Before a Hydrological Service’s strategic, annual orother plan is implemented, managers must developa clear link between the Service’s plan and theresponsibilities and duties of its staff. It is essentialthat managers focus the attention of their staff onthe results that they are expected to achieve, andnot simply on the tasks that they are to carry out.An essential aspect of planning is appraisal of pastperformance. In many countries, government agenciesare required to provide an annual report to thenational assembly of elected representatives, andthis provides the ultimate in performance appraisal.Even where they are not required to do so, directorsof Hydrological Services should review at leastannually their Services’ activities, achievementsand changing environment. The findings might bepresented in different ways and degrees of detail forvarious audiences: for elected representatives andclients, brief, focusing on contributions to nationallife; for staff, detailed, highlighting technical andproduct/service matters; and for management andplanning staff, comprehensive, including an analysisof deficiencies and adverse changes in theenvironment.Appraisal of the performance of the entireHydrological Service provides the basis for identifyingits strengths and weaknesses, and fordeveloping a plan that builds on these strengthsand eliminates any weaknesses. Managers shouldappraise the Service’s performance in terms of theperformance criteria and indicators previouslydefined by the plan, and should consider howsuccessfully the Service is achieving its vision andmission and governmental policies and goals.Feedback from public- and private-sector clientsalike is an invaluable element of performanceappraisal. A Service that delivers products that aretechnically first class but contribute little towardsachieving governmental goals or meeting commercialclients’ needs is unlikely to receive consistentsupport or funding for future planning periods.2.4 HUMAN RESOURCES MANAGEMENTAND CAPACITY-BUILDING[HOMS A00, Y00]2.4.1 ManagementMost organizations consider their most importantresource to be their staff. This is true andmanagers of successful Hydrological Servicesknow this well. As the role and functions of aHydrological Service evolve, the staffing requirementsand management style of the Service mayneed to change as well. Hence, a Service that ismodernizing or developing value-added products,for instance, is likely to require more staff skilled

II.2-10GUIDE TO HYDROLOGICAL PRACTICESin information technology. Such staff will performtheir tasks differently from staff with traditionalfield skills and will require different levels andstyles of supervision.The long-term success and health of a HydrologicalService rests in the hands of the director and managers.To discharge their responsibilities effectively,they require skills in a number of areas. The directorshould ensure that the entire management teamhas the following assets:(a) Diplomatic and administrative skills to functionsuccessfully in the public service environmentor as a State-owned company;(b) The ability to monitor and understand thebusiness environment and translate it intoplanning the Service’s programmes;(c) Skills in all areas of business management– human resources, finances, capitalassets, product quality, information technology– as appropriate to the Service;(d) Leadership skills and motivation;(e) Marketing and communication skills that areneeded to develop effective relationships withclients, the public and elected representatives,investors/donor agencies and the “owner”;(f) Technical and scientific knowledge required toensure that the Service has the technology itneeds;(g) The ability to represent the Service and thenational interest at an international level.The director should place as much emphasis onmanagement training as on technical capacitybuilding.If indeed staff are the most important resource,managers should select staff with great care. Theyshould appoint or reassign staff to meet the demandsof the Service’s strategic and annual plans so thatwork groups have the human resources needed toachieve their assigned objectives. A director shouldtake staff succession seriously, that is, identificationand preparation of junior staff to advance to moreresponsible levels as senior staff retire. A combinationof experience and training will be needed tosuitably prepare such staff.A contract between the Hydrological Service and anemployee is an essential basis for effective and fairhuman resources management. Legal requirementswith regard to employment contracts vary fromcountry to country, and managers of HydrologicalServices should be familiar with the employmentrelatedlegislation under which they operate. For aHydrological Service that is a parastatal or statebody, the form of contractual arrangements foremployees is normally specified by national civilservice regulations.The Director of a Hydrological Service that lacksclosely specified arrangements for employeecontracts should seriously consider developingthem. The main benefit of a contractual agreement,for both employer and employee, is that the relationshipis specific and transparent, allowing anyshortcomings on either side to be addressed in anobjective manner.A job description for every staff member is anessential management tool. It provides both aclear statement of what the Service expects of theindividual and a basis for setting personal objectives,implementing performance appraisals andidentifying training and personal developmentopportunities.Job descriptions and objectives provide the basis forthe appraisal of staff performance, which is on a parwith planning and strategy development in termsof importance to the managers and staff of a HydrologicalService. In many organizations, performanceappraisal is linked to preparation of staff developmentplans for individual staff members, for workgroups – for example, a work group that is expectedto take on new responsibilities – or for the entireService. Staff development plans will be used forfuture performance appraisals, in part to ensurethat proposals have been implemented and to evaluatetheir success as tools for enhancingperformance.When human resources management tools such asjob descriptions, setting of objectives and performanceappraisal are introduced, staff members can beresistant and sceptical. However, a manager willfind that they will be more likely to cooperate iftools to enhance their prospects are used sensibly,constructively and persistently over a period of oneor two years. It cannot be overemphasized thatmanagement tools must be used with understanding;if not, they are likely to be of little value, andeven counter-productive. This implies that thedirector and managers of a Hydrological Serviceshould ensure that their own performance meetsthe Service’s needs.2.4.2 Training and continuing educationTraining and continuing education are of criticalimportance to both management and staff, thecollective goal being to enable staff members makethe greatest possible contribution to achieving theService’s mission. Training and education should

CHAPTER 2. HYDROLOGICAL SERVICESII.2-11be managed in a structured way, possibly bypreparing a training plan for the Service or forindividual staff members. It should correspond totraining needs analyses that are part of the performanceappraisal process. These analyses may also becarried out independently, for example, whenmanagers are considering new procedures, productsor services, organizational restructuring orsome other response to the changing businessenvironment, and when there is a need to matchcurrent competence with that of the past.2.5 FINANCIAL AND ASSETMANAGEMENT [HOMS A00]Financial management has become a basic aspect ofa director’s work, as governments worldwide imposemore stringent financial disciplines. Normally,financial management procedures are defined by aHydrological Service’s parent organization, and thedirector and/or selected management staff receiveappropriate training in these procedures. Nevertheless,a director should make every effort todevelop a much more sophisticated grasp of financialmanagement than the basic minimum.Accounting procedures in the public sector are generallyprescribed by government and a HydrologicalService that is part of a government department orState-owned enterprise will have to follow thesescrupulously. This is to ensure transparency andaccountability, that is, to make sure that the Service’sfinancial accounts are clear and comprehensible,resources are spent for the designated purpose,responsibilities for financial transactions can beidentified and funds are not siphoned off throughcorruption – unfortunately a fact of life in developedas well as developing countries.2.5.1 Sources of revenueA major concern of managers of a HydrologicalService – indeed, of any organization – is the sourceof income or revenue required to maintain theService’s operations and assets. In most countries,the government has been and will continue to bethe predominant source.The recent trend worldwide is for governments torequire or enable public sector agencies to findsources of commercial revenue in addition to allocationsfrom the national budget. Some Serviceshave made significant progress in identifying nongovernmentalclients or willing clients within thepublic sector. Value-added products and services arethe most profitable. A Service should focus itsenergy on seeking new sources of revenue only inareas that are consistent with its primary mandateand where a good, that is, profitable, business casecan be made.Commercial work requires a legal mandate, and themanagers of Hydrological Services that engage incommercial activities should be familiar withnational laws and regulations relating to commerce.In most countries, there are relatively few nongovernmentalclients for value-added products thatare potential sources of considerable commercialrevenue. This is especially the case in developingcountries where the pressure on a HydrologicalService to find supplementary sources of revenue isalso likely to be greatest. Most of the products andservices of a Hydrological Service, and the databasesand other assets that are needed to provide thoseproducts and services, are, however, public goodsfor which the government is the logical purchaser.A Hydrological Service may nevertheless be requiredto recover some of the costs of its public servicesand products.Economic theory indicates that the appropriateapproach to cost recovery is to charge for the directand associated overhead costs of providing theproduct or service, including the administrativecost of recovering costs, as well as depreciation ofthe assets used. Where the product or service uses ahydrological database or other asset provided atpublic expense, it is economically inefficient toattempt to charge part of the cost of providing thatasset. Potential clients strongly object to suchcharges and refuse to use the service at all. Thisresults in underutilization of the public asset, theuse of inferior alternatives such as guesswork andhence economic inefficiency. The experience ofHydrological Services that have attempted to chargefor data generally confirms this. Increasingly, it isregarded as preferable to provide unrestricted publicaccess to data via the Internet at no charge. Thisreduces the cost of meeting data requests and canhelp enhance the Service’s reputation.As a means of imposing financial discipline andachieving maximum transparency, a governmentmay choose to administer funds in ways other thanmaking an allocation in the national budget. Theseinclude:(a) Providing funds through a non-governmentalorganization such as a National ResearchCouncil, which allocates funds on a competitivebasis and/or in terms of defined nationalneeds for information;

II.2-12GUIDE TO HYDROLOGICAL PRACTICES(b) Establishing the Hydrological Service as a Stateownedenterprise and administering publicfunds on the basis of a contract for definedoutputs and services. In the extreme, thecontract could be awarded on a competitivebasis, potentially to another provider;(c) Introducing a government contract betweenthe appropriate Minister and/or the Ministerfor Finance and the director of the HydrologicalService, to provide defined outputs and services.The director of a Hydrological Service is unlikelyto have much influence on such a decision, whichwill reflect overall government policy. However,the director should seek guidance from other directorsin similar circumstances, either in otherorganizations in the same country or HydrologicalServices in other countries, and attempt to negotiatecontractual arrangements that provide themost favourable conditions for future work.Lastly, it is worth recalling that a way of increasingeffective revenue is to reduce costs, for example, bymoving from paper-based to Internet-based disseminationof information. Of course, the Serviceshould ensure that the quality of the product orservice does not suffer, but is preferably enhancedfrom the user’s perspective.2.5.2 Budgeting and monitoringfinancial performanceBudgeting should be an integral part of annualplanning. As the Hydrological Service defines itsproposed programme of objectives and activities, itwill need to define the associated costs and, throughan iterative process, revise the proposed programmeso that its cost is consistent with likely revenue. Justas it is desirable for operational staff to be involvedin annual planning, so too should they be involvedin setting the budget. They will, after all, have towork with and within it.As a rule, Hydrological Services that are parastatalor state bodies, budgeting procedures and timetablesare strictly defined. The annual planningprocess must therefore be timed accordingly.The Hydrological Service is likely to be requiredto submit its budget in a defined format to itsparent organization, in terms of specified lineitems in a chart of accounts. The managers ofthe Service should ensure that the internal processof preparing a budget provides an end resultthat can readily be converted into the requiredformat, but may prefer to use a format that ismore appropriate to the Service’s business orsimpler to use.The completed budget should be a key componentof the annual plan and a means of monitoringperformance against the plan.2.5.3 Asset managementIn simple terms, the purpose of asset managementis to ensure that the value of the organization’sassets is maintained, and therefore that the organizationcontinues to be a going concern that has theresources to do business. Therefore, it is of considerableimportance to all managers and staff of aHydrological Service. Asset management basicallyinvolves acquisition, replacement, mainten ance,protection and disposal of assets.2.5.4 Database securityA Hydrological Service’s single most important assetis its database. Means of protecting this asset willdepend on the data storage media that are used, butthere is no doubt that the director of a Service mustensure that it is protected. In a number of countries,data rescue projects have been necessary togather together all data – original records, usuallyon paper – place them in a secure location andconvert them into an electronic format that is moremanageable on a long-term basis. Such projects,admirable though they may be, should becomenecessary only under circumstances completelybeyond the control of the Hydrological Service.There can be few excuses for a director who allowshis Service’s basic asset to be degraded.Paper media, for example, observers’ notebooks,recorder charts and machine-punched tapes, areinvaluable because they usually provide the originalrecord that must be consulted if questions ariseabout data validity, or if data reprocessing isrequired for some reason. They should be stored insuch a way that they will be subject to the smallestpossible degree of damage by insects, water, rot,sunlight, fire, earthquake or simple loss. Originalrecords should be the responsibility of a singleoffice; if such an arrangement is not possible,however, the same person in each office in whichthe records are stored should be responsible forthem. Whatever the case, the location of originaldocuments should be carefully tracked, for example,if they are released for reprocessing. If aHydrological Service does not have the facilities orexpertise to permanently archive its paper media,the national archive, museum or library may beable to assist.As paper media are subject to deterioration, copiesshould be made. Commonly, microfilm or

CHAPTER 2. HYDROLOGICAL SERVICESII.2-13microfiche copies are made, but the obsolescenceof this technology presents difficulties for thefuture. Electronic storage of scanned images isnow an economical alternative, using CD-ROMsor other even higher density media. In this case,technological obsolescence is perhaps an evengreater concern than microfilm; thus theHydrological Service will need a procedure forregular migration of electronic archives ontosuccessive generations of storage media.The secure long-term storage of original andprocessed records in electronic form, for example,incoming telemetered data, or an entire computerdatabase, requires procedures that are not so muchsophisticated as disciplined. It is essential to makeregular, frequent backups of data, followingrigorously defined procedures so that data are notlost before they reach the archive, and to makeliberal comments on archived data so that subsequentusers can understand any changes that have beenmade.References and further readingAustralian Bureau of Meteorology, 1995: Long-term Plan1995–2010. Commonwealth of Australia, Bureau ofMeteorology, Melbourne.———, 2005: Strategic Plan 2005–2010. Commonwealthof Australia, Bureau of Meteorology, Melbourne(http://www.bom.gov.au/info/leaflets/strategic-plan-2005-10.pdf)———, 2006: Operational Plan 2005–06. Commonwealthof Australia, Bureau of Meteorology, Melbourne.European Commission, 2000: Directive 2000/60/EC ofthe European Parliament and of the Council of23 October 2000 establishing a framework forCommunity action in the field of water policy. WaterFramework Directive, European Union, Brussels(http://www.wmo.ch/pages/prog/www/ois/Operational_Information/coolC14Sky/VolumeD/Amendments/AdditionalDataProducts/02_Resolution%2040.pdf).International Organization for Standardization, 1983:Measurement of Liquid Flow in Open Channels. ISOStandards Handbook 16, Geneva.World Meteorological Organization, 1994: The LegalBasis and Role of Hydrological Services (M.P. Mosley).(WMO/TD-No. 602), Geneva.———, 1995: Abridged Final Report with Resolutions of theTwelfth World Meteorological Congress (WMO-No. 827), Geneva.———, 1999: Abridged Final Report with Resolutions of theThirteenth World Meteorological Congress (WMO-No. 902), Geneva.———, 2001a: The Role and Operation of NationalHydrological Services (P. Mosley), Technical Reports inHydrology and Water Resources No. 72 (WMO/TD-No. 1056), Geneva.———, 2001b: Exchange of Hydrological Data and Products(P. Mosley), Technical Reports in Hydrology andWater Resources No. 74 (WMO/TD-No. 1097),Geneva.———, 2006: Technical Regulations. Volume III –Hydrology (WMO-No. 49), Geneva.———, 2007a: Basic Documents. No. 1 (WMO-No.15),Geneva.———, 2007b: WMO Strategic Plan (WMO-No. 1028),Geneva.———/United Nations Educational, Scientific andCultural Organization, 1997: Water ResourcesAssessment: Handbook for Review of NationalCapabilities. Geneva and Paris.

CHAPTER 3INTEGRATED WATER RESOURCES MANAGEMENT3.1 INTRODUCTION3.1.1 Sustainable water developmentSince the 1970s, there has been a growing awarenessthat natural resources are limited and thatfuture development must come to terms with thisfact. The concept of sustainability has on the wholegained wide acceptance, although its meaning mayvary, depending on the person. Sustainable developmentas defined by the International Union forConservation of Nature (IUCN), the United NationsEnvironment Programme (UNEP) and the WorldWildlife Fund (WWF) is adopted in this Guide:“improving the quality of human life while livingwithin the carrying capacity of supporting ecosystems”.(IUCN/UNEP/WWF, 1991).Is there any way to measure the sustainability ofdevelopment? If account can be taken of naturalvariability and trends in water resources availability,it is arguable that the effects of developmentwill be reflected in changes in the resource base.The so-called ecological footprint is a tool that isused to measure the amount of land and water areasrequired to produce the resources it consumes andabsorb its wastes (see, for example, http://www.footprintnetwork.org/). It has been estimated thattoday’s global population has an ecological footprintthat is 20 per cent larger than the biocapacityof the Earth. Monitoring of the quantity and qualityof water in natural systems – streams, lakes,underground, snow and ice – thus becomes aprerequisite for tracking the extent to which developmentcan be sustained.The building of adequate databases through themonitoring of hydrological systems is a fundamentalprerequisite of water resources assessment andmanagement. This chapter reviews the adequacy ofcurrent monitoring networks and techniques in thelight of a changing resource base and evolving watermanagement philosophies related to sustainabledevelopment.3.1.2 The changing nature of the resource3.1.2.1 Natural changesThe hydrological system, driven by meteorologicalconditions, is constantly changing. Over longperiods of time – decades to millenniums – variationsin the receipt of energy from the Sun, actingthrough the atmospheric system, cause importantchanges in hydrological regimes. For example,changes in the distribution and extent of ice massesand vegetation cover usually reflect hydrologicalchanges.Recently, there has been increasing awareness thatinteractions between the air and the sea haveextremely important effects on climate. El NiñoSouthern Oscillation events, for example, with teleconnectionsover wide areas, may have far-reachinghydrological ramifications, which are particularlyrelevant when associated with droughts and floods.Longer- term atmospheric phenomena such as thePacific Decadal Oscillation and their teleconnectionsmay also affect hydrological systems.Natural events of a completely different type, suchas major volcanic eruptions with massive emissionsof dust and gases into the atmosphere, can alsoimpact the hydrological system significantly.3.1.2.2 Human-induced changesHuman activities increasingly affect hydrologicalsystems. Some of the more important activities arelisted below:(a) The construction of dams and diversions hasa major impact on flow regimes and sedimenttransport in many of the world’s rivers, as wellas on ecological systems in donor and recipientbasins;(b) Changes in land use often produce major effectson hydrological regimes as follows:(i) Deforestation often leading to morepronounced flood peaks and increasedsoil erosion;(ii) Draining of wetlands, often bringingabout changes in the runoff regime;(iii) Frontier road and railway constructioncausing erosion, changes in human settlementand land-use change;(iv) Farming practices, resulting in varyinginfiltration rates and groundwaterrecharge;(v) Urbanization, prompting characteristicallyflashy runoff;(c) The quality of water in many places has beenadversely affected by industrial and municipal

II.3-2GUIDE TO HYDROLOGICAL PRACTICESwaste and agricultural practices such as the useof fertilizers and pesticides;(d) The emission of greenhouse gases leading toclimate change and related changes to hydrologicalsystems. According to the Fourth AssessmentReport of the Intergovernmental Panelon Climate Change (IPCC, 2007), the duration,location and frequency of extreme weather andclimate events are likely to change, and wouldresult in mostly adverse impacts on biophysicalsystems;(e) Long-range transport of air pollutants can causeecological damage far from the emission sites.Monitoring systems should take into account thesemany changes in order to better understand thehydrological system, predict water availability andmanage resources effectively. In particular, anthropogenicclimate change driven by significantincreases in atmospheric greenhouse gases over thepast few centuries and the resulting effects ofclimate change on hydrological systems pose enormouschallenges for water managers. Given theuncertainty in regional climate scenarios and thelikelihood that younger water resources managerswill witness events not previously recorded inhuman history, there is an even greater need forquality assured hydrological datasets and robustphysically based models than ever before.3.1.3 Changing attitudes tomanagementThere have been significant socio-economic changesin many parts of the world. Rapid populationgrowth, particularly in developing countries and inburgeoning urban centres, combined with industrializationand rising living standards, have increasedthe demand for water. Pollution in many regionshas reduced the quantities of safe drinking water.Groundwater levels have declined in many regions.Growing demand, outstripping supply, will becomemore common. Thus, more efficient and effectivewater management is imperative.The past few decades have witnessed dramaticchanges in water management. There have beentwo important underlying themes. First, there is agrowing awareness that water is a fundamentalelement in the natural environment. The presenceand movement of water through all biologicalsystems is the basis of life. Water, land and biologicalsystems must be viewed as interlinked, andmonitoring of the various components of theecosystem should be harmonized. Secondly, wateris absolutely essential to all forms of economicactivity, for example, agriculture and foodWater consumption by region (per cent) Global water consumption (km 3 /year)6 0005 0004 0003 0002 0001 00002520151050(a)1900 1920 1940 1960 1980 2000(b)Total consumptionAgricultureIndustryMunicipal economyReservoirsAsiaEuropeWorldNorth AmericaAfricaAustralia and OceaniaSouth America1900 1920 1940 1960 1980 2000Figure II.3.1. Trends in world water consumptionby (a) activity and (b) regionproduction, for much of industrial production andfor the generation of energy. Water is also a criticalfactor in human health. Too much water, in theform of floods, or too little, such as drought, canlead to human and environmental disasters.Figure II.3.1(a) shows trends in world waterconsumption from 1900 to 2000. Globally, consumptionduring that period increased tenfold; by theyear 2000, almost half of the available water supplieswas in use. Agriculture, particularly irrigation,remained the primary consumer despite a continuingdecline in water use from 90.5 per cent in 1900to 62.6 per cent in 2000. During the same period,industry’s share of water consumption rose from 6.4per cent to 24.7 per cent; that of cities with the samegrowth rate climbed from 2.8 per cent in 1900 to 8.5per cent in 2000 (United Nations, 1997).How has water consumption compared with theavailable water resources in each of the world’smajor regions during the twentieth century?Figure II.3.1(b) answers this question in terms ofpercentages calculated on the basis of theoreticalresources, that is, the amount of water flow in rivers.According to these calculations, Europe and Asiaclearly consumed much greater shares of their waterresources than North America, Africa and, particularly,Australia and Oceania, and South America . Itis also clear that Europe and Asia had the highestgrowth in consumption, except for South America,where the increase was offset by plentiful reservesof water.Growing awareness of the pervasive nature of water,in addition to its importance in the natural

CHAPTER 3. INTEGRATED WATER RESOURCES MANAGEMENTII.3-3environment and in human activity, has led to therecognition that a holistic approach to itsmanagement is necessary. Development of theresource for human use may have a detrimentalenvironmental impact while, conversely, changesin the natural resource base may limit or otherwiseaffect human activities. These changes have led tothe holisitic approach known as integrated waterresources management.3.1.3.1 Watershed managementThere is general recognition that the naturalmanagement unit is the river basin. It makes senseto manage the water resources within a river basinand in a coordinated manner, as the water is oftenused several times as it moves from the headwatersto the river mouth. It also makes sense to manageall natural resources – vegetation, soils and the like– within the basin unit. Water demands for humanactivities should also be managed within the basinin an integrated manner.Unfortunately, political boundaries do not normallycoincide with basin boundaries. Rivers often crossinternational frontiers and traverse states or provinceswithin countries. Globally, about half of allthe land surface falls within international basinsand more than 200 significant basins are internationalin character.3.1.3.2 Management fragmentationIt is common that several agencies or institutionswithin a State have authority over different aspectsof water resources management. Departments orministries of environment, agriculture, energy,industry, and health often have conflictingmandates.All too often, the monitoring networks within aState are also fragmented politically and institutionally.Even within single agencies, theresponsibilities for water quantity and waterquality monitoring often are not coordinated.Vague institutional responsibilities and mandateswithin countries, and conflicting demands onwater use between countries (within internationalbasins) or inter-state disputes withinfederal States, pose real problems for the establishmentand maintenance of effectivemonitoring networks.It is against this complex background of rapidlychanging water management philosophies, politicaland socio-economic realities and the resourcebase itself that several actions must be taken.These include the design and operation of monitoringsystems, the storage and dissemination ofdata, followed by the use of those data as the basison which sound decisions can be taken to plan,design and operate water projects and issue warningsand forecasts of hydrologically significantevents.3.2 INTEGRATED WATER RESOURCESMANAGEMENT [HOMS A00]The term integrated may be defined as having allparts combined into one harmonious whole, coordinatingdiverse elements.Integrated water resources management can beinterpreted at three different levels. First, it involvesthe systematic consideration of various dimensionsof water: surface and groundwater, and quantityand quality. The key is that water represents anecological system, containing interrelated parts.Each part can influence, and be influenced by, otherparts, and therefore needs to be planned for andmanaged with regard to those interrelationships. Atthis level, attention normally is given to how tointegrate considerations related to water securityand water quality.At the second level, managers recognize that whilewater is an ecological system, it also interacts withother resource systems, ranging from terrestrial toother environmental systems. This second level isbroader than the first, and turns attention to matterssuch as flood-plain management, drought mitigation,erosion control, irrigation, drainage, non-pointsources of pollution, protection of wetlands andfish or wildlife habitat and recreational use. At thislevel, integration is needed because many waterproblems are triggered by land use or other developmentdecisions involving major implications foraquatic systems.The third level is broader still, and directs themanager toward interrelationships among theeconomy, society and the environment – of whichwater is but one component. Here, the concern isthe extent to which water can facilitate or hindereconomic development, reduce poverty, enhancehealth and well-being and protect heritage.All three levels highlight the fact that planners andmanagers deal with a mix of systems, which ofteninvolves hierarchical relationships. As a result, akey feature of integrated water resources managementis the application of a systems or ecosystem

II.3-4GUIDE TO HYDROLOGICAL PRACTICESapproach. Another key feature is the need to befocused and results oriented, as there is always adanger of defining systems or issues so broadly thatthey become impractical from a managementperspective.To quote the Inter-American Development Bank(1998), integrated water resources managementinvolves decision-making on development andmanagement of water resources for various uses,taking into account the needs and desires of differentusers and stakeholders.In sum, the keys to effective integrated waterresources management are a systems perspective, afocused and results-based approach, and partnershipsand stakeholders. In the present chapter,attention is given to the rationale behind theseaspects, how they have been applied in practice,what general lessons have been learned and whatcautions should be borne in mind.3.3.2 Surface water and groundwaterIn many regions of the world, groundwater is themajor source of the flow in surface streams duringthe dry season. In addition, certain land-basedactivities, such as those causing leakage from undergroundstorage tanks, can lead to pollution ofaquifers. Other land-based activities, for instance,withdrawal, which is implemented to meet urbanor agricultural needs that exceed rates of recharge,can also bring about the depletion of groundwaterreserves.Given the interconnections identified above, inorder to achieve effective management of aquaticsystems, it is necessary to study and manage surfacewater and groundwater as connected systems,particularly to ensure secure water supplies ofacceptable quality. An integrated approach encourages– indeed, requires – the joint management ofsurface water and groundwater systems.3.3 RATIONALE FOR INTEGRATED WATERRESOURCES MANAGEMENT3.3.1 Water quantity and qualityThe responsibility for managing the quantity, orsupply, and quality of water is often assigned toseparate agencies. This can be attributed to historicaladministrative reasons that are unrelated to thesubject at hand, but also to the rationale that sucha division nurtures efficiency because it enablesprofessionals to focus on a specific aspect of watermanagement. This practice has generally resultedin two groups or cultures of water professionals –managers of clean water and managers of dirtywater – who operate separately.A major disadvantage of this separation of authorityfor water quantity and quality is that the causesof, and therefore, solutions to, quantity and qualityproblems are frequently interdependent. Forexample, if flow in a river system drops because ofnatural variability, there may not be enough waterto meet water-use needs or sufficient capacity toassimilate wastes deposited in the river. As a result,dams and reservoirs may be constructed in orderto enhance storage to meet user needs and toprovide augmented flow in the dry season in orderto meet water quality standards. To achieve theoptimum design of such dams and reservoirs,water quantity and quality needs should either beconsidered jointly or integrated in managementpractices.3.3.3 Upstream and downstreamconsiderationsDecisions or action taken in the upstream part of ariver basin or catchment have implications fordownstream areas. For example, point and nonpointpollution entering a river in the upper part ofa basin may produce negative health or otherimpacts on downstream users, whether human orother species. Conversely, if officials in downstreamurban areas determine that they can reduce theirvulnerability to flooding by building storage damsand reservoirs in the upper part of their basin, thenthe upstream residents may suffer. This happensthrough inundation of urban and agricultural landcaused by reservoir backwater, leading to loss ofhousing and livelihood for some farmers, and sometimesdamage to, or loss of, heritage or areas such asburial grounds or historical sites.The interconnections between areas of a river basinor catchment are often cited as a compelling reasonfor using the basin or catchment as the spatial unitfor integrated water resources management. Such arationale is logical. However, it must be understoodthat the relevant basin or catchment for surfacewater may not coincide with the spatial extent ofan aquifer. It should never be assumed that surfaceand groundwater systems have the same spatialextent. The possibility of such a disconnection ofthe spatial boundaries of surface and groundwatersystems poses a challenge to water managers, forwhich there is no obvious answer. Another challengearises when interbasin transfers occur,requiring a perspective extending beyond the

CHAPTER 3. INTEGRATED WATER RESOURCES MANAGEMENTII.3-5upstream and downstream needs in one basin, so asto consider the interconnections between two ormore river basins.Another challenge for defining the spatial boundariesof a management system based on ecosystemcharacteristics is the presence of various administrativeand political boundaries. Rivers, andsometimes lakes, have been used to delineateboundaries between municipalities, provinces,states and countries and are shared by several countriesor subnational administrative units. As a result,management of such rivers and lakes requires theinvolvement and collaboration of various partners.The most flagrant example is the Danube river,whose basin is shared by 19 countries. Ensuringthat upstream and downstream interests andconcerns are addressed in situations involvingdifferent countries poses a significant challenge forimplementing an integrated approach.3.3.4 Water, land and otherresource systemsMany water problems originate on land. To achieveflood damage reduction, for example, it is generallynot sufficient to manage or control the variabilityof water levels in rivers and lakes through dam,dykes and levees. Land-use activity related to urbandevelopment and agriculture can result in theremoval or shrinking of wetlands, forest systemsand grasslands, which in turn exacerbates erosionand flooding problems. Indeed, it is claimed thatmuch flood damage along the Ganges river in Indiaand along the Indus river in Pakistan can be attributedto the removal of forests in the Himalayas.Furthermore, initiatives to enhance water qualitymust often start with attention to activity associatedwith other resource systems. Thus, the use ofpesticides, herbicides and fertilizers to improveagricultural productivity is often a major contributorto non-point sources of pollution, requiringattention to land-based activities to tackle the pollutionof water systems.Another challenge is the long-range transport ofairborne pollutants. Even if an integrated approachis taken by key managers within the basin, theyusually do not have the authority to deal withsources of pollution originating outside the basin,at times hundreds of kilometres away.3.3.5 Environment, the economy andsocietyHistorically, water management has been dominatedin developed and developing nations bythree professions: engineering, agriculture andpublic health. As a result, engineers began focusingon structural solutions for issues ranging fromwater security – whether for urban, industrial oragricultural use – to water quality and flooddamage. In addition, health professionals startedturning their attention to the treatment anddisposal of sewage and other wastes detrimental tohealth.The domination of water management by engineeringand health professionals led to an emphasis ontechnical and economic perspectives. During the1960s, there was a growing awareness that environmentalaspects should receive greater attention,followed by the recognition that social or culturalissues also required special consideration. Suchrecognition led to a gradual acceptance of the desirabilityof teams – multidisciplinary, at least, orinterdisciplinary, at best – to gather and integrate arange of professional and disciplinary views indeveloping management approaches. While eachdiscipline represented in a multidisciplinary teamyields discipline-specific results and leaves the integrationof the various contributions to a third party,building an effective team requires overcomingmany obstacles and challenges. This is necessary todevelop and apply new knowledge, with teammembers working together as equal stakeholders toaddress a common challenge. However, such teamsare essential if the intent is for strategies to addressenvironmental, economic and social aspects in anintegrated manner.3.3.6 Vertical and horizontalfragmentation: systems and silosNotwithstanding the compelling reasons for usingintegrated water resources management, there arepragmatic reasons for public agencies beingstructured to focus on one or a subset of resourcesystems. Hence, it is common to find separatedepartments or ministries of agriculture, forestry,wildlife and natural resources. The separation offunctions into different agencies is known ashorizontal fragmentation when, for a given levelof government – national, state or local –responsibility for a particular resource is assignedto various agencies. Such arrangements require arange of technical expertise represented on a teamthat can concentrate on issues and opportunitiesrelated to that resource and, where appropriate,develop working relationships with users ofthe resource. Along with such organizationalstructures, inter-departmental committees or taskforces may be used to coordinate different interests,mandates and perspectives.

II.3-6GUIDE TO HYDROLOGICAL PRACTICESWithout coordination and collaboration, there is areal danger of losing effectiveness and efficiency.For example, as a ministry of agriculture carries outits mandate to increase lower-cost food production,it may seek to drain wetlands to put more land intofarming, or it may encourage the use of fertilizers orother chemicals to boost crop production. Incontrast, a ministry of natural resources may introduceprogrammes aimed at protecting or expandingwetlands in order to enhance wildlife habitat andcapacity, and delay runoff during storms, therebyreducing downstream flooding. Such programmescan also serve to discourage the use of agrochemicalsin order to reduce pollution of waterways usedby fish, birds and other species. The activities of theaforementioned ministries might not result in a netchange in the amount or type of wetlands in a jurisdiction,while expending significant funds to drainwetlands in some areas and enlarge them inothers.As mentioned above, horizontal fragmentationrefers to the division of responsibility within onelevel or layer of government. Vertical fragmentationoccurs when agencies at different levels ofgovernment – national, state or local – share aninterest in or responsibility for a resource, such aswater. For example, a State agency might design,build and operate a dam and reservoir, one purposeof which is to provide water for nearby communities.At the same time, a local government agencymight be responsible for distributing the waterfrom a reservoir to households, industry and farmirrigation systems. When this occurs, there is a realneed for mechanisms or processes to coordinatemandates and activities among the levels ofgovernment. Vertical and horizontal fragmentationcan create obstacles to integrated waterresources management; therefore, integration isessential.One way to overcome such fragmentation has beento do away with specific resource-oriented agenciesand create more broadly based ones, such as ministriesof the environment or of sustainabledevelopment. This was done by the CanadianFederal Government, which eliminated its InlandsWater Directorate and re-distributed the professionalstaff among various divisions withinEnvironment Canada. The aim was to ensure thatwater was considered along with other resourceissues in order to achieve sustainable development.The theory was logical. However, it soon becameapparent that other federal staff or clients had greatdifficulty finding the water specialists with whomto raise their concerns. Furthermore, it became clearthat most farmers or other water users thought ofthemselves as having a water or waste problem,rather than a sustainable development problem.The concepts of vertical and horizontal fragmentationillustrate that water and other resourcemanagers face their greatest difficulties in handlingso-called edge or boundary problems – thosebetween the mandates and responsibilities of twoor more agencies, and thus must be dealt with in ashared or partnership manner. Agencies usually doa very competent job dealing with problems ortasks that are clearly within their mandates andauthority. In contrast, those with edge or boundarycharacteristics offer serious challenges, and thuscall for an integrated approach, despite the practicaladministrative problems to be overcome.3.3.7 Collaboration, coordination andcoherenceWhich criteria should be used to evaluate thesuccess of a given management approach? It iscommon for the following criteria to be applied:effectiveness, in terms of achieving desired outputsor outcomes; efficiency, with regard to producingthe desired effects without wasting time and energy;and equity, ensuring a fair distribution of benefitsand costs of the desired results. The discussion inthe previous subsections indicates that manyaspects can hinder effectiveness, efficiency andequity.Integrated water resources management is a toolthat can help managers meet such criteria. Thefollowing factors are essential to achieve integration:collaboration, or the act of working together;coordination, that is, harmonious adjustments orworking together, or, arranging in proper order; andrelationships and coherence, that is, logical connectionor consistency, harmonious connection of theparts of a whole.Integration is a means to an end, not an end initself. As a result, the use of integration in watermanagement should be preceded by a shared visionabout a desired future condition or state. Withoutsuch direction, it is difficult to determine whichparts need to made into a whole, who should beworking together to establish a proper order andrelationships and what logical connections need tobe made.The rationale for integration as one tool to help inachieving a vision is to allow a desired future conditionto be achieved effectively, efficiently andequitably. Integration is usually advocated becauseof its potential contribution to all three criteria. It

CHAPTER 3. INTEGRATED WATER RESOURCES MANAGEMENTII.3-7contributes to effectiveness by helping ensure thatdifferent needs and opportunities are consideredand incorporated into plans and activities; to efficiencyby helping ensure that actions of one agencyor organization do not undo the actions of anotheragency; and to equity, by forcing consideration ofdifferent values and interests of variousstakeholders.3.4 EVOLUTION OF INTEGRATED WATERRESOURCES MANAGEMENTIntegrated water resources management is not anew concept. It has existed in one form or anotherfor well over half a century in the voices and writingsof eminent water experts such as Gilbert White.In 1977, the United Nations Water Conference inMar del Plata adopted a resolution promoting theconcept. Later, integrated water resources managementwas highlighted as a guiding principlecontained in the 1992 Dublin Statement on Waterand Sustainable Development (ICWE, 1992). Morerecently, the Global Water Partnership programmehas been based on this concept. (Tortajada, 2003).In various countries and regions, initiatives havebeen taken to manage water using integrated waterresources management as a basis, even if the termwas not being used. This section reviews experiencewith selected approaches to integrated waterresources management.3.4.1 United States of America: Ohioconservancy districts, TennesseeValley AuthorityIn 1933, the Tennessee Valley Authority was establishedso that initiatives related to hydropowerdevelopment, navigation and flood control in theTennessee river basin could be pursued in a coordinatedand integrated manner. Without theAuthority, different agencies responsible for powersupply, navigation and flood control would mostlikely operate independently, and thus miss anopportunity to design and operate activities tocomplement one another. In addition, the TennesseeValley Authority became involved in other initiatives,such as rural planning, housing, health care,libraries and recreation, because no agenciesprovided such services or facilities.3.4.2 Canada: conservation authoritiesLegislation was passed in 1946 in Ontario to createconservation authorities, catchment-based organizationsformed through a partnership ofmunicipalities with the provincial government(Richardson, 1974; Mitchell and Shrubsole, 1992).The trigger was the realization that individualmunicipalities did not have the resources or authorityto take initiatives such as the construction andoperation of upstream dams and reservoirs for flooddamage protection, which would benefit an individualmunicipality, as well as other downstreamcommunities. In 2005, there were 36 conservationauthorities in Ontario, covering areas in which over90 percent of the population lived.Conservation authorities were founded on thefollowing principles, which have had enduringvalue:(a) The best management unit was the watershed:Many of the economic staples of the province,such as agriculture and timber, depended onwater and terrestrial resources, highlighting theneed for an integrated approach;(b) Local initiative was essential: A conservationauthority would only be established when twoor more municipalities in a watershed agreed tocollaborate with each other and the provincialgovernment;(c) Provincial-municipal partnership was key:Although the provincial government wouldnot impose a conservation authority, it wouldparticipate as a partner. However, this featurealso meant that areas with few people or amodest tax base would not be able to form aconservation authority, because there would beno local capacity to raise the required funds;(d) A comprehensive perspective was required:Many land-based problems were caused bytoo much or too little water, and water-basedproblems often were influenced by land-basedactivities. Thus, a comprehensive approachwas promoted, meaning that water and associatedland-based resources would be consideredtogether;(e) Coordination and cooperation were important:Any new conservation authority was requiredto create links with provincial and municipalagencies responsible for other natural resources,the environment and planning.3.4.3 United States and Canada: the GreatLakesThe Great Lakes basin, shared by the United Statesand Canada, covers a surface area up to the outletof Lake Ontario of 765 990 km 2 , 521 830 km 2 ofwhich is land and 244 160 km 2 , water. It containsabout 20 per cent of the world’s surface freshwatersupply, and is home to over 40 million people –14 per cent of the total United States population

II.3-8GUIDE TO HYDROLOGICAL PRACTICESand 30 per cent of the total Canadian population.Its governance involves two national, eight state,two provincial and hundreds of municipalgovernments.The two national governments signed theBoundary Waters Treaty in 1909, and through thattreaty created the International Joint Commission.It has six commissioners, three from the UnitedStates and three from Canada. It is a permanentbi-national body and forum for internationalcooperation and conflict resolution regarding airpollution, water quality, regulation of water levelsand water flows between Canada and the UnitedStates along their common border. The Commissionhas quasi-judicial, investigative and surveillanceroles and two operational arms: the Great LakesWater Quality Board and the Great Lakes ScienceAdvisory Board. The Commission drafted twoimportant agreements, the 1972 and 1978 GreatLakes Water Quality Agreements, with amendmentsenacted in 1987. The 1978 agreement wasthe catalyst for the application of an ecosystemapproach.3.4.4 Australia: total catchmentmanagementBurton (1986) may be credited with the developmentof total catchment management in Australia.In 1947 the State Government of New South Walescreated the Conservation Ministry to coordinatethe management of the water, soil and forestresources in the state. In 1950, legislation waspassed to create the Hunter Valley ConservationTrust, with responsibility for the coordinatedmanagement of water and land resources in thatvalley, inland from Newcastle on the coast of NewSouth Wales. At a state-wide level, the State Premierin 1984 approved the creation of the Inter-Departmental Committee on Total CatchmentManagement and subsequently announced that atotal catchment management plan would be developedfor each of the major river valleys in NewSouth Wales.3.4.5 New Zealand: Resource ManagementActNew Zealand’s experience with integrated waterresources management goes back to the 1940s(Memon, 2000). Later, starting in the 1960s,catchment control plans for soil conservationand river control were initiated, and these werefollowed in the 1970s with basin-wide resourceinventories and informal water allocationplans.The Resource Management Act of 1991 wasmajor milestone, distinguished by its provisionof “a statutory basis for a relatively integratedapproach to environmental planning” (Memon,2000). Furthermore, the Act replaced a largenumber of separate and sometimes inconsistentand overlapping acts related to the use of land,water, air and geothermal resources. Under thislaw, duties are divided among three levels. Thecentral government focuses on policy and monitoring.At the subnational level, water and otherresource and environmental management tasksare undertaken within a two-tier system involvingdirectly elected multiple-purpose regionalcouncils and territorial local authorities: cityand district councils. The 12 regional councilsare set up according to major river basincatchments.3.4.6 South AfricaThe Water Act of 1956 was introduced to achievea fair allocation of water between competing agriculturaland industrial needs. A key aspect of thelegislation was giving water rights to riparianproperty owners, leaving them free to retain waterthrough dams and other means. By the mid1990s, however, it was recognized that the 1956statute had some serious limitations. First, waterquality concerns were not being systematicallyincorporated into management decisions whichnormally emphasized water quantity allocation.One result was growing organic pollution andeutrophication. Second, water requirements forthe environment were not being adequatelyrecognized. Third, at least in many rural areas,access to water was viewed as inequitable. Lastly,several analyses during the 1990s had noted theneed for a more integrated approach to watermanagement (Department of Water Affairs andForestry and Water Resources Commission, 1996;Lazarus, 1997; Gorgens and others, 1998). Theoutcome was the White Paper on a NationalWater Policy for South Africa, which had beenstarted in 1995 through consultations whichextended over two years (Department of WaterAffairs and Forestry, 1997).South Africa introduced the new Water Services Actin 1997 and the National Water Act in 1998 to alterthe way in which water was managed. A key aspectof the new approach was the incorporation of integratedwater resources management. The NationalWater Act recognized that water was a scarce andunevenly distributed national resource and also aresource that belonged to all people, not just riparianlandowners. Sustainability and equity were

CHAPTER 3. INTEGRATED WATER RESOURCES MANAGEMENTII.3-9identified as fundamental principles, and meetingthe basic needs of present and future generationswas a key objective, along with protecting the environmentand meeting international obligations forshared water resources. Social and economic developmentwas also to be promoted through theallocation and use of water.The National Water Act emphasizes decentralization.Key new institutions include catchmentmanagement agencies, through which responsibilityfor water management is delegated to theregional or catchment level and which involvelocal communities. Each catchment managementagency is responsible for a catchment managementstrategy, and through it the agency haspowers to manage, monitor, conserve and protectwater resources; make rules to regulate water use;require the establishment of managementsystems; and temporarily control, limit or prohibitthe use of water during periods of water shortage.Water user associations also are established underthe legislation, with the main purpose of helpingand coordinating individual water users.Based on the experience in South Africa, Van Zyl(1999) concluded that the successful use of integratedwater resources management requiresrecognition that integrated water resources managementhas the following characteristics:(a) It is a team business. A team requires commonunderstanding and individual and team fitnessand skills, rules and regulations for the gameand proper organization. This also requirescoaching, coordination, policies, integratedstrategies and planning;(b) It is about winning and achieving goals. Thiscalls for commitment and passion for the game,individual and team motivation, team spirit,and mutual trust and respect for the game, theteam and supporters;(c) It is about superior strategies. This requiresunderstanding of the real business, involvementof the right players and champions,addressing value systems, tactical organization,entrepreneurship, boundarylessness, innovationand the creation of a winning culture;(d) It is about champions, people with vision, initiativeand passion and outstanding leadership;(e) It is an exercise in public administration andpolitical science. There must be support for theprogramme or it will fail.The above points are relevant beyond South Africa,and deserve attention when designing, establishingor implementing integrated water resourcesmanagement strategies.3.5 PERSPECTIVES ON INTEGRATEDWATER RESOURCES MANAGEMENT3.5.1 Dublin Conference: Earth Summit,1992Prior to the United Nations Conference onEnvironment and Development, commonlyknown as the Earth Summit, in Rio de Janeiro inJune 1992, the International Conference on Waterand the Environment was held in January 1992 inDublin, Ireland. It was convened by WMO onbehalf of all nations with an interest in freshwater.This was the most all-embracing event focused onglobal water issues since the United Nations WaterConference in Mar del Plata in Argentina in March1977. The purpose of the Dublin Conference wasto identify priority issues related to freshwater,and to recommend actions to address them (ICWE,1992). The ideas and proposals from Dublin weretaken to the Earth Summit, and many of therecommendations were subsequently included inAgenda 21, the strategy for sustainable developmentin the twenty-first century (Young andothers, 1994).The Dublin Statement on Water and SustainableDevelopment, the main output from the conference,emerged from deliberations by more than 500people from 114 countries, 28 United Nations agenciesand organizations and 58 non-governmentaland intergovernmental organizations. The preambleof the Dublin Statement asserts that concertedaction is needed to reverse trends of over-consumption,pollution, and rising threats from both floodsand droughts. Action needs to come from local,national and international levels, and four principlesguide future initiatives. The first principle hasbeen interpreted as a call for integrated watermanagement:Fresh water is a finite and vulnerable resource,essential to sustain life, development and the environment.Since water sustains life, effectivemanagement of water resources demands a holisticapproach, linking social and economic developmentwith protection of natural ecosystems.Effective management links land and water usesacross the whole of a catchment area or groundwateraquifer.The other principles emphasized the followingneeds:(a) A participatory approach, involving users,planners and policymakers at all levels, withdecisions to be taken at the lowest appropriatelevel;

II.3-10GUIDE TO HYDROLOGICAL PRACTICES(b) An enhanced role for women in the provision,management and safeguarding of water;(c) The recognition that water has economic valuein all its competing uses and thus should also beconsidered an economic good. Managing wateras an economic good was viewed as an importantway to achieve efficient and equitable use,and to encourage conservation and protectionof water resources. In addition, all human beingshave the basic right to access to clean water andsanitation at an affordable price.The first principle of the Dublin Statement, a callfor a holistic approach which, since the EarthSummit has usually been referred to an integratedapproach, is the principle which has received themost attention. It emphasized that water problemscannot be treated in isolation, and indeed shouldbe considered in relation to land-based and landuseplanning issues. This was not a revolutionaryprinciple, as the Organisation for EconomicCooperation and Development (1989) had previouslypublished guidelines for integration relativeto water management. Subsequently, researcherssuch as MacKenzie (1996) urged adoption of anecosystem [holistic] approach, noting that it “canbe seen as both comprehensive (in scope) and integrated(in content)”.Agenda 21 was the key outcome of the Earth Summit(United Nations, 1992). Chapter 18 of the Agendadeals with freshwater resources and provides acompelling rationale for integrated water resourcesmanagement:...the holistic management of freshwater as afinite and vulnerable resource, and the integrationof sectoral water plans and programmeswithin the framework of national economic andsocial policy, are of paramount importance foraction in the 1990s and beyond. The fragmentationof responsibilities for water resourcesdevelopment among sectoral agencies is proving,however, to be an even greater impediment topromoting integrated water management thanhad been anticipated.In this assessment, Agenda 21 highlights the challengeof edge or boundary problems, noted earlier,as well as the significance of vertical and horizontalfragmentation.3.5.2 World Water Council and the WorldWater ForaThe World Water Council was established in 1996to provide an open platform for the discussion ofwater issues. Detailed information on the Councilmay be found on its Website, http://www.worldwatercouncil.org/.The Council is an initiative of water specialists, theacademic community and international agencies. Itregularly convenes World Water fora to discusswater issues, develop proposals for action and highlightthe importance of water. The First World WaterForum was held in Marrakech, Morocco, in March1997, the second, in The Hague, Netherlands, inMarch 2000, the third, in Tokyo, Japan, in March2003 and the fourth, in Mexico City, Mexico, inMarch 2006. The Fifth World Water Forum isplanned to be held in Istanbul, Turkey, inMarch 2009.The World Water Fora have been consistent inendorsing integrated water resources management.For example, the Ministerial Declaration of TheHague on Water Security in the 21st Century statedas follows:The actions advocated here are based on integratedwater resources management, that includes theplanning of management of water resources, bothconventional and non-conventional, and land.This takes account of social, economic and environmentalfactors and integrates surface water,groundwater and the ecosystems through whichthey flow. It recognizes the importance of waterquality issues.The Ministerial Declaration emerging from theThird World Water Forum in Japan also endorsedintegrated water resources management:Whilst efforts being taken so far on waterresources development and management shouldbe continued and strengthened, we recognizethat good governance, capacity-building andfinancing are of the utmost importance tosucceed in our efforts. In this context, we willpromote integrated water resourcesmanagement.3.5.3 Global Water PartnershipThe Global Water Partnership was set up in 1996,the same year as the World Water Council. Detailedinformation can be found at http://www.worldwatercouncil.org/.The Partnership is an international network withsupport from a number of countries and internationalfunding agencies. Its mandate is to supportintegrated approaches to sustainable water

CHAPTER 3. INTEGRATED WATER RESOURCES MANAGEMENTII.3-11management consistent with the Dublin and Rioprinciples by encouraging stakeholders at all levelsto work together in more effective, efficient andcollaborative ways. Its primary function is toencourage the exchange of information and, quiteexplicitly, to promote integrated water resourcesmanagement. As an international network, it isopen to all bodies involved in water management– governments of developed and developing countries,United Nations agencies, multilateral banks,professional associations, research institutes, theprivate sector and non-governmentalorganizations.The Global Water Partnership network includes asecretariat in Stockholm and nine technical advisorycommittees for each the following regions:Southern Africa, West Africa, the Mediterranean,Central and Eastern Europe, Central America,South America, South Asia, South-East Asia andChina.3.5.4 World Summit on SustainableDevelopment, Johannesburg, 2002The World Summit on Sustainable Developmentwas held ten years after the Earth Summit in Rio deJaneiro. A plan of implementation was prepared tobuild on and extend the actions proposed inAgenda 21. Section IV of the Plan addresses mattersrelated to protecting and managing the naturalresource base of economic and social development,and the first topic covered was an integratedapproach to their management.With respect to an integrated approach, the Planstipulated as follows:Human activities are having an increasing impacton the integrity of ecosystems that provide essentialresources and services for human well-beingand economic activities. Managing the naturalresources base in a sustainable and integratedmanner is essential for sustainable development.In this regard, to reverse the current trend innatural resource degradation as soon as possible,it is necessary to implement strategies whichshould include targets adopted at the nationaland, where appropriate, regional levels to protectecosystems and to achieve integrated managementof land, water and living resources, whilestrengthening regional, national and localcapacities.With reference to freshwater, the Plan stated thatthe objective should be to develop integratedwater resources management and efficiency plansby 2005, with support to developing countries,through actions at all levels to “develop andimplement national/regional strategies,plans and programmes with regard to integratedriver basin, watershed and groundwatermanagement”.3.6 ELEMENTS OF BEST PRACTICE FORINTEGRATED WATER RESOURCESMANAGEMENT3.6.1 Alternative interpretations:comprehensive versus integratedapproachesAt the beginning of this chapter, the term integratedwas defined as having all parts combinedinto a harmonious whole or coordinating diverseelements. This definition has led to integrated waterresources management being characterized as asystems, ecosystem, holistic or comprehensiveapproach. However, emphasis on having all partscombined into a harmonious whole has alsoprovided integrated water resources managementwith its greatest challenge.At the strategic planning level, it is appropriate tointerpret integrated water resources management asa comprehensive approach that seeks to identifyand consider the broadest number of variables thatare significant for the coordinated management ofwater and associated land and environmentalresources. However, if such an interpretation iscontinued at the operational level, experience hasshown that this contributes to inordinately lengthyperiods of time needed for planning, and also resultsin plans which are usually insufficiently focused tobe of value to managers.Given the above challenges, a comprehensiveapproach should be used at the strategic planninglevel to ensure that the widest possible perspectiveis maintained, in order to avoid overlooking anykey external or internal variable or relationship.However, at the operational level, more focus isneeded. In that regard, an integrated approach,while maintaining interest in systems, variablesand their interrelationships, is more selective andfocused, concentrating on the subset of variablesand relationship judged to be the most importantand amenable to being influenced by managementactions. If such a distinction is made betweencomprehensive and integrative interpretations of asystems, ecosystem or holistic approach, it shouldbe possible to complete planning exercises in a

II.3-12GUIDE TO HYDROLOGICAL PRACTICESmore reasonable length of time, identify the mostimportant priorities for action, and thereby meetthe needs of managers and users (Mitchell, 1990).3.6.2 Vision for a desirable futureIntegrated water resources management is a meansto an end, not an end to itself. As a result, before itsimplemention, or as an initial step in such a process,it is important to have a well-established visionor direction about a desired future condition for anarea or catchment. Integrated water resourcesmanagement will be one instrument to assist in itsachievement.A vision articulates the destination towardswhich a group or society agrees to aim. Thevision represents a future which in significantways is better or more desirable than the present.Without such direction, it is difficult to determinewhich parts need to be brought togetherinto a whole, and who should be workingtogether to arrange a proper order and establishrelationships.Developing a shared vision can be a major challenge,since at any given time a range of values,interests and needs will exist among different stakeholdergroups in a river basin or catchment.However, if there is no sense of direction, or clearlydefined ends, integrated water resources managementwill not be able to create one. Thus, plannersand managers must understand that without avision it is unlikely that integrated water resourcesmanagement will be an effective tool. Even worse,it may be discredited because it did not deliver avision, something it was never intended to do.When thinking about a vision for the future, it ishelpful to distinguish among what is most probable,desirable and feasible. Planners and managersoften focus first on identifying most probablefutures, and insight on this is very valuable.However, too often, they stop there, or then movedirectly to considering what would be feasiblefutures, in light of what is deemed as most probable.An important point to remember is that themost probable future may not be the most desirablefuture; that is precisely why planners and managersseek to create a vision – to determine the desiredfuture condition.3.6.3 Spatial scale: watershed, subwatershed,tributary and siteIt is important to make a distinction among differentsituations when applying integrated waterresources management. The need to adjust theamount of detail included as spatial scale changes isespecially significant. In a report focusing on lessonslearned and best practices related to watershedmanagement, three Ontario conservation authorities(2002) shared some interesting insights.In Ontario, watershed planning, equivalent tointegrated water resources management, isconducted on four different scales, “with the levelof detail increasing as the size of planning areadecreases”. In that context, the most logical andefficient way to conduct integrated water resourcesmanagement is to start with a catchment or riverbasin plan, then develop sub-catchment or subwatershedplans on a priority basis, andsubsequently follow those with tributary plans,and finally with environmental site plans, asappropriate. Key lessons indicate that what is doneat each stage provides direction and informationfor the next lower level and also helps avoid orminimize the potential for duplication.However, financial constraints often result in subbasinor sub-watershed plans being prepared first,and integrated later into an overall basin or catchmentplan for integrated water resourcesmanagement. In a similar way, tributary plans maybe completed before the sub-catchment plans. Thethree Canadian conservation authorities distinguishamong the four levels of integrated waterresources management in the following ways:(a) Basin or catchment plans: Such plans cover largeareas. These plans include goals, objectives andtargets for the entire basin and document bothenvironmental resources and environmentalproblems. They also provide catchment-widepolicy and direction for protecting surface andgroundwater, natural features, fisheries, openspace systems, terrestrial and aquatic habitats,and other important features;(b) Sub-basin or sub-catchment plans: These plansinvolve a smaller area compared with the basinor catchment level plan. On this spatial scale,enhanced detail is provided to allow localenvironmental issues to be addressed. Goals,objectives and targets for management of thesub-catchment are identified. Sub-basin or subcatchmentplans dealing with integrated waterresources management are custom designedto reflect local conditions and concerns.Recommendations may be included subsequentlyin official plans, secondary plans,growth management plans or other municipalplanning instruments;(c) Tributary plans: Plans on this scale are usuallyprepared to guide proposals for significant land

CHAPTER 3. INTEGRATED WATER RESOURCES MANAGEMENTII.3-13use changes, such as proposals for sub-divisions,large-scale water taking, gravel extraction,intensive agriculture and industrial estates.These are prepared for a portion of a sub-catchmentand generally cover an area ranging from2 to 10 km 2 . Ideally, the boundaries of a tributaryplan should align with the drainage basinof a tributary, but this is not always possiblein practice. Recommendations emerging fromtributary plans generally appear in secondaryland-use plans, official land-use plan amendments,conditions for draft plan approval or forsite plan approval;(d) Environmental site plans: Such plans areusually developed to meet conditions set out ina draft plan. They provide details on proposedenvironmental and stormwater measures, andare usually submitted in parallel with plansfor grading, erosion or sediment control andsite servicing. Recommendations from environmentalsite plans normally appear inengineering design drawings for draft plans fora subdivision or industrial estate.The four scales or levels identified above deserveattention from all planners and managers involvedin integrated water resources management. Bybeing aware that various levels of detail are appropriateon different spatial scales, planners andmanagers can increase the likelihood that issuesand problems will be addressed at a suitable level ofdetail, overlap or duplication of work will beavoided, the time needed to complete integratedwater resources management plans will be reducedand capacity for implementation will be boosted. Ifall of these are accomplished, the credibility andvalue of integrated water resources managementwill be enhanced.3.6.4 Partnerships and alliancesIntegrated water resources management wasdesigned to ensure a holistic or ecosystem approach,and to facilitate the coordination of initiatives bydifferent stakeholders. With regard to the latter, astrong motivation is required to break down what isoften referred to as the silo effect, or the tendencyof agencies to take decisions with regard only totheir own mandates and authority, without referenceto those of other organizations. In this manner,there is a reasonable expectation that integratedwater resources management will be more effectiveand efficient compared with a non-integratedapproach. However, in promoting a holisticapproach, integrated water resources managementcan experience tension with arrangementsfor including participatory mechanisms. Manyindividuals, communities or stakeholder groups donot always give attention to the entire system, butrather only to that part or aspect related to theirown needs and interests. Thus, individuals oftenfocus on the impacts of catchment management ontheir own property, while municipal governmentsfrequently worry about the area under their responsibility.As a result, if integrated water resourcesmanagement and participatory methods are to beused together, care must be taken to understand thestrengths and limitations of both.Collaboration allows stakeholders to join forces insharing their views on different aspects of a problem,and then together explore differences andsearch constructively for solutions going beyondany one stakeholder’s capacities and limitations. Inthis way, they can share resources, enhance eachother’s capacity for mutual benefit and achieve acommon purpose by sharing risks, responsibilitiesand rewards (Gray, 1989; Himmelman, 1996).In addition to the above features, Gunton and Day(2003) point out that it is essential to determine ifa collaborative approach is appropriate in anyspecific situation. In their view, a collaborativeapproach “may not work in all circumstances” . Tohelp determine when participatory approaches areappropriate, they identify five pre-conditions forsuccess:(a) Commitment of decision-making agencies to aparticipatory approach;(b) Commitment of all stakeholders;(c) Urgency for resolution of an issue or issues;(d) Absence of fundamental value differences;(e) Existence of feasible solutions. In their view,the challenge is not whether all pre-conditionsare met perfectly, but whether they aremet adequately enough to allow a participatoryprocess to begin.3.6.5 Links to regional planning andimpact assessmentIntegrated water resources management plans orstrategies often lack a legislative or statutory basis.This can have several negative consequences. First,agencies receiving recommendations from an integratedwater resources management plan maysimply ignore them, believing that they fall outsidetheir legislated mandate or mission. Second, if agenciesdo strive to implement recommendations fromsuch a plan, they have to determine what prioritythese recommendations should have relative toother responsibilities. For either of these reasons,there is a high probability that little action will betaken.

II.3-14GUIDE TO HYDROLOGICAL PRACTICESOne way to overcome this problem is to link recommendationsto instruments – such as officialregional or municipal land-use plans, or environmentalimpact assessments – which have a statutorybasis. It was for that reason that in 3.5.3 the discussionhighlights how recommendations fromintegrated water resources management catchment,sub-catchment, tributary or environmental siteplans in Ontario were incorporated into officialplans, secondary plans or environmental impactassessment processes.Water planners and managers should thereforefamiliarize themselves with the opportunities toconnect the recommendations from integratedwater resources management plans to regional orlocal land-use official plans or to environmentalimpact assessment processes, when these have astatutory basis. Another alternative is to strive for astatutory basis for integrated water resourcesmanagement but, at the moment, such arrangementsare the exception, not the rule.3.6.6 Designing institutionalarrangementsOnce a vision is established, it is important toconsider the institutional arrangements – the formaland informal mix of values, rules, organizationalstructures and cultures, mechanisms and processes– available for implementing integrated waterresources management.Experience suggests that governments often lookfirst to make changes to organizational structures,such as when ministries of the environment werecreated in the 1970s or ministries of sustainabledevelopment, in the 1990s. However, this can beeffective only where edge or boundary problems areidentified, as highlighted in 3.2.6, and are thereforerarely the best place to start. As a result, when introducingor modifying institutional arrangements forintegrated water resources management once avision has been established, planners and managersshould do the following:(a) Determine what actions can be taken to givecredibility or legitimacy to integrated waterresources management. This is usually done byhaving some combination of a legislative base,administrative policy commitment and ongoingfinancial support;(b) Decide which management functions are tobe integrated. Given their utility-like characteristics,some functions, such as water supply,sewage treatment and waste disposal, could beallocated to the private sector, while others,such as flood-plain management or wetlandprotection, should be allocated to the publicsector on the basis of their common propertycharacteristics;(c) Determine appropriate organizational structures,on the principle that structures shouldfollow, not lead, functions. A continuumof structures exists, ranging from one large,centralized, multiple-purpose organization tomany, small, decentralized, single- or limitedfunctionorganizations. Each arrangement hasstrengths and weaknesses and all encounteredge or boundary problems;(d) As structures will never align perfectly withfunctions, next consider what mix of processes– for example, public participation and impactassessment – and mechanisms, such as interdepartmentaltask forces or committees, will bemost effective to ensure coordination, collaborationand coherence among different agenciesor groups. The following observation has beenmade (Grindle and Hilderbrand, 1995):Capacity builders need to create active mechanismsfor interaction and coordination. Formalmeans of communication and coordinationcan be created, such as high-level and technical-levelcoordination committees, interlockingboards of directors or advisors, joint workshopsand seminars, and relocating offices or improvingtechnology so that communication isphysically easier. Informal communicationscan be stimulated to supplement and supportthese formal interactions.The value of such initiatives has been reiteratedby noting that:interventions which allow professionals towork alongside one another as equals areincreasingly important. Such interventionsinclude networking and twinning arrangements,as well as workshops, seminars andplatforms for cooperation which facilitate thesharing of knowledge (Franks, 1999);(e) Most importantly perhaps, managers and plannersshould establish organizational culturesand staff attitudes to foster collaborationand cooperation, rather than competition.According to Grindle and Hilderbrand (1995),“Without exception, the organizations thatperformed well were able to inculcate a senseof mission and commitment to organizationalgoals among staff” and “one of the most importantsets of findings is the evidence that relatesorganizational performance to the strength andorientation of its organizational culture”. Such a

CHAPTER 3. INTEGRATED WATER RESOURCES MANAGEMENTII.3-15supportive culture can be created and nurturedthrough training and education programmesfocusing on the nature of and need for collaborativeprocesses, and conflict resolution.The above concepts together provide a frameworkto assist planners and managers as alternatives areconsidered regarding appropriate institutionalarrangements to support integrated water resourcesmanagement.3.6.7 Monitoring and evaluatingAs noted in 3.5.2, it is important to have a desirablefuture for which to aim and then use integratedwater resources management as a means to achieveit. It is equally important to include provision formonitoring and evaluation so that the journeytoward a desirable future can be tracked and, ifnecessary, adjusted.To ensure a results-based focus, it is normal to monitorfor effectiveness: are objectives being achieved?In addition, attention should be given to efficiency(are objectives being attained in the most cost-effectivemanner?) and equity (are the benefits and costsbeing distributed fairly?). Another dimensionincreasingly receiving attention is transparency oraccountability: is it possible to see how decisionsare taken and resources allocated?Comparing the above product- and process-orienteddimensions provides a systematic basis against whichto assess progress, or lack of progress, related to therole of integrated water resources management inhelping achieve a vision. Without systematic monitoringfollowed by evaluation, the opportunity tolearn from experience is reduced, as well as theopportunity to make adjustments in the light of newinformation, knowledge and experience.3.7 CAUTIONS REGARDING INTEGRATEDWATER RESOURCES MANAGEMENT3.7.1 When to apply integrated waterresources managementIt is all too often assumed that integrated waterresources management is good or desirable.However, because integration does not occur withoutcosts, care should be taken when decidingwhether or not integrated water resources managementis appropriate. Staff time and other resourcesare required to accomplish integration; thoseresources are then not available for other needs ortasks. Often overlooked is the need to establish thatserious resource scarcity and/or environmentaldegradation problems are the result of many interconnectedcausal factors whose resolution requiresan integrated approach. In contrast, many situationsare characterized by relatively straightforwardproblems that can be handled effectively by oneagency or organization. If such a situation exists,integrated water resources management is unlikelyto be needed. However, if there are multiple causes,or the actions of numerous agencies or participantsmight work at cross-purposes or could be designedto complement each other, then integrated waterresources management will be appropriate (Hooperand others, 1999).3.7.2 Implementation gapOnce it has been decided that integrated waterresources management is appropriate, it is importantto ensure that capacity exists to move from conceptto action. As indicated in 3.5, many challenges canbe encountered when striving to implement integratedwater resources management: too broad aninterpretation which leads to difficulties in completinganalyses and plans in a timely manner, lack of avision to be achieved through use of integrated waterresources management, lack of recognition of theneed to change the detail sought as the spatial scalechanges, confusion over the role of partners or stakeholders,lack of credibility or legitimacy of aintegrated water resources management plan, inadequateinstitutional arrangements and low monitoringand evaluation capacity. Any one or a combinationthereof can hinder integrated water resourcesmanagement. Most of these aspects are not uniqueto the approach, but are generic challenges for planningand management. Nevertheless, if theseshortcomings are not recognized and addressed, theywill most likely contribute to integrated waterresources management being ineffective, and thus toits being discredited.References and further readingBurton, J.R., 1986: The Total Catchment Managementconcept and its application in New South Wales.Proceedings of the Hydrology and Water ResourcesSymposium, 1986. Brisbane, Queensland, GriffithUniversity.Department of Water Affairs and Forestry, 1997: WhitePaper on a National Water Policy. Department ofWater Affairs and Forestry, Pretoria.Department of Water Affairs and Forestry, and WaterResearch Commission, 1996: The Philosophy andPractice of Integrated Catchment Management:

II.3-16GUIDE TO HYDROLOGICAL PRACTICESImplications for Water Resource Management inSouth Africa. Discussion Document, WRCReport No. TT 81/96, Pretoria, Water ResearchCommission.Görgens, A., G. Pegram, M. Uys, A. Grobick, L. Loots, A.Tanner and R. Bengu, 1998: Guidelines for CatchmentManagement to Achieve Integrated Water ResourceManagement in South Africa. WRC Report No. KV108/98, Pretoria, Water Research Commission.Gray, B., 1989: Collaborating: Finding Common Groundfor Multiparty Problems. San Francisco, Jossey-Bass.Abstract: http://www.colorado.edu/conflict/peace/example/gray7278.html.Grindle, M.S. and M.E. Hilderbrand, 1995: Buildingsustainable capacity in the public sector: what canbe done? Public Administration and Development,15:441–463. Abstract: at http://www.grc-exchange.org/info_data/record.cfm?ld=35.Gunton, T.I. and J.C. Day, 2003: The theory andpractice of collaborative planning in resource andenvironmental management. Environments, 31(2):5–19. Abstract: http://goliath.ecnext.com/coms2/summary_0199-1243923_ITM&referid=2090.Himmelman, A.T., 1996: On the theory and practice oftransformational collaboration: from social serviceto social justice. In: Creating Collaborative Advantage(E. Huxham, ed.). Thousand Oaks, California, SAGEPublications.Hooper, B.P., G.T. McDonald and B. Mitchell, 1999:Facilitating integrated resource and environmentalmanagement: Australian and Canadian perspectives.Journal of Environmental Planning and Management,42:747–766.Inter-American Development Bank, 1998: Strategy forIntegrated Water Resources Management, PublicationNo. ENV-125. Washington, DC, Inter-AmericanDevelopment Bank.Intergovernmental Panel on Climate Change (IPCC),2007: Fourth Assessment Report – Climate Change2007. Geneva.International Conference on Water and theEnvironment, 1992: International Conference onWater and the Environment: Development Issuesfor the 21st Century, Keynote Papers. Geneva, ICWESecretariat, World Meteorological Organization.International Union for Conservation of Nature/UnitedNations Environment Programme/World WildlifeFund, 1991: Caring for the Earth: a Strategy forSustainable Living. Gland.Jenkins, H.A., 1976: A Valley Renewed: The History of theMuskingum Watershed Conservancy District. Kent,Ohio, Kent State University Press.Lazarus, P. 1997: Towards a Regulatory Framework forthe Management of Groundwater in South Africa.WRC Report No. 789/1/98. DWAF Report Geo2.2(389). Pretoria, Water Research Commission andDepartment of Water Affairs and Forestry.MacKenzie, S.H., 1996: Integrated Resource Planning andManagement: The Ecosystem Approach in the GreatLakes Basin. Washington, DC, Island Press.Memon, P.A., 2000: Freshwater management policiesin New Zealand. In: EnvironmentalPlanning and Management in New Zealand,(P.A. Memon and H. Perkins, eds). NorthPalmerston, New Zealand, Dunmore Press.Abstract: at http://www3.interscience.wiley.com/cgi-bin/abstract.14322/ABSTRACT.Mitchell, B. and D. Shrubsole, 1992: Ontario ConservationAuthorities: Myth and Reality, Department ofGeography Publication Series No. 35. Waterloo,Ontario, University of Waterloo.Mitchell, B. (ed.), 1990: Integrated Water Management:International Experiences and Perspectives. London,Belhaven Press.Organisation for Economic Cooperation andDevelopment, 1989: Water Resource Management:Integrated Policies. Paris, Organisation for EconomicCooperation and Development.Richardson, A.H., 1974: Conservation by the People: TheHistory of the Conservation Movement in Ontario to1970. Toronto, University of Toronto Press.Tortajada, C., 2003: Workshop on Integrated WaterResources Management for South and South-EastAsia, Bangkok, Thailand, 2–4 December 2002. WaterInternational, 28(1)130–131.United Nations, 1977: Report of the United NationsWater Conference, Mar del Plata, 14–25 March 1977,E/CONF.70/29. New York, United Nations.———, 1992: Report of the United Nations Conferenceon Environment and Development, Rio de Janeiro,3–14 June 1992. United Nations publicationE.93.I.8, Three Volumes. New York, UnitedNations.———, 1997: Comprehensive Assessment of the FreshwaterResources of the World. New York, United Nations.Van Zyl, F., 1999: Above all else what do we have to domake a sustainable impact on diffuse source pollution?Proceedings of the International Conferenceon Diffuse Pollution. C. Barber, B. Humphries andJ. Dixon (eds). Wembley, Western Australia, CSIROLand and Water.World Summit on Sustainable Development,2002: Plan of Implementation, at http://www.johannesburgsummit.org.html/documents/summit_docs/2309_planfinal.html.Young, G.J., J.C.I. Dooge and J.C. Rodda, 1994: GlobalWater Resource Issues. Cambridge, CambridgeUniversity Press.

CHAPTER 4APPLICATIONS TO WATER MANAGEMENT4.1 WATER RESOURCE ASSESSMENT ANDWATER PROJECTS [HOMS A00]4.1.1 The need for water resourceassessmentWater resource assessment is the determination ofthe sources, extent, dependability and quality ofwater resources, which is the basis for evaluatingthe possibilities of their utilization and control(UNESCO/WMO, 1997). Water resource assessmentis of critical importance to sound and sustainablemanagement of the world’s water resources. Severalreasons for this may be cited (WMO/UNESCO,1991):(a) The world’s expanding population is placingincreasing demands on water for drinking, foodproduction, sanitation and other basic socialand economic needs, but the world’s waterresources are finite. The rising demand hasreached its limit in some areas and will reachthe limit in many other areas within the nexttwo decades. Should present trends continue,the world’s water resources will be fully utilizedbefore the end of the next century;(b) Human activities are becoming increasinglyintensive and diverse, producing a definite,ever-growing impact on natural resourcesthrough depletion and pollution. This is particularlythe case for water, whose quality formany purposes can be severely degraded byphysical changes and by pollution caused by awide range of chemicals, microorganisms, radioactivematerials and sediments;(c) Water-related natural hazards, such as floods,droughts, and tropical cyclones, causeimmeasurable destruction of human life andproperty, and have so during the course ofhistory. Deforestation and urbanization, inparticular, have exacerbated flood hazards byincreasing the magnitude and frequency offloods;(d) There is a growing realization that the world’sclimate is not constant, and indeed may wellbe changing in response to human activity.While the relationship between increasedglobal temperatures and greenhouse-gasinducedwarming has been widely publicized,more attention should be paid to the effects ofclimate on the distribution of rainfall, runoff,and groundwater recharge, which are likely tobe significant. It cannot be assumed that thepatterns of these hydrological phenomena willnot change.Effective water management can be achievedthrough sound decision-making based on reliabledata and information on the status and trends ofwater resources, including quantity, quality, statisticson events such as floods, for example, and usefor human purposes.Water resource assessment is generally a prerequisitefor water resources development andmanagement, as recognized as early as 1977 by theUnited Nations Water Conference held in Mar delPlata (Resolution 1 and Recommendation A of theMar del Plata Action Plan). The Conference stressedthe need for greater knowledge about the quantityand quality of surface-water and groundwaterresources, and for comprehensive monitoring toguide the management of these resources.Furthermore, the International Conference onWater and the Environment, held in Dublin from26 to 31 January 1992, recommended a number ofactions in support of national water resource assessment(United Nations, 1992).4.1.2 Water resource assessmentprogramme componentsIn order to permit a preliminary assessment of availablewater resources on which to base national orregional long-term plans for overall water resourcesdevelopment, a basic water resource assessmentprogramme involves the collection and processingof existing hydrological and hydrogeological data,plus the auxiliary data required for their arealinterpolation.These plans should be based on or geared to presentand future water needs. The components of a waterresource assessment programme are shown, inFigure II.4.1, and are mainly as follows (UNESCO/WMO, 1997):(a) The collection of hydrological data – historicaldata on water cycle components at a number ofpoints distributed over the assessment area;(b) The collection of physiographic data – data onthe natural characteristics of the terrain thatdetermine the areal and time variations of thewater cycle components, such as topography,

II.4-2GUIDE TO HYDROLOGICAL PRACTICESCollection of hydrologicaldata – water cyclecomponents, includingquantity and quality ofsurface and groundwaterCollection of physiographicdata – topographic,soils, geologyBasic andappliedresearchEducationandtrainingTechniques of arealassessment of waterresources – regionalizationtechniquesWater-resourcesinformation – data banks,mapsUsers – planning, designand operation of waterresources facilitiesFigure II.4.1. Components of a basic water resource assessment programmesoils, surface and bed rock geology, land useand land cover;(c) The techniques used for the areal assessmentof water resources – techniques for convertingdata into information and for relating thehydrological data to the physiographic data toobtain information on water resource characteristicsat any point of the assessment area.A basic water resource assessment programme isconsidered adequate if these three components areavailable and if, by relating them, they aresufficiently accurate to supply the water resourcesinformation required for planning purposes at anypoint of the assessment area. The countryconcerned will need to define the type ofinformation required for planning, the manner inwhich this information is produced and transmittedto users and the effects of a lack of or inaccurateinformation on the decision-making process at theplanning stage.All basic water resource assessment activitiesrequire skilled personnel, sound equipment andtechniques for field surveys, network design andoperation, and development of reliable areal interpolationtechniques. This, in turn, may requiretraining and education of the required personnel,and basic and applied research to develop therequired technology. An analysis of these activitiescan provide indications of their adequacy for thepurpose of basic water resource assessment or, ifinadequate, the additional means to be devoted tothem to provide the required base for future developmentof an adequate water resource assessmentprogramme.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-34.1.3 Evaluation of water resourceassessment activitiesWater resource assessment is a national responsibility,and any evaluation of the extent to which it isbeing undertaken adequately in a country is alsothe responsibility of the country concerned. TheWMO/UNESCO Water Resources Assessment:Handbook for Review of National Capabilities(UNESCO/WMO, 1997) was prepared with the aimof increasing the capabilities of countries to evaluatetheir achievements in water resource assessmentand to provide a general framework for determiningtheir needs and the actions necessary to achieveminimum requirements. The methodologyproposed in the Handbook comprises the full rangeof topics and activities that are included in a waterresource assessment programme. The current levelsof basic water resource assessment are comparedwith minimum acceptable requirements in terms ofinstallations, equipment, skilled personnel, education,training and research. It contains detailedchecklists for each component (see Figure II.4.1)and offers advice as to how each activity might beevaluated, in most cases in quantifiable terms.The results of the evaluation will be different foreach country, depending on the characteristics ofthe corresponding basic water resource assessmentprogramme and the country’s characteristics andneeds. Nevertheless, a minimum set of results isexpected in practically each case. This set includesthe following items:(a) An analysis of the existing institutional frameworkfor carrying out a basic water resourceassessment programme with its resulting advantages,disadvantages and related constraints;(b) A comparative evaluation of the measurementnetworks and indications of network elementsthat require improvement with respect tostation density, equipment, operational andsupervisory staff, and other factors;(c) A review of the available surveys andprogrammes for collecting and processingphysiographic data pertinent to basic waterresource assessment;(d) An evaluation of the application of varioustechniques for areal extension of basic waterresource assessment and related data- and information-transfertechniques;(e) An analysis of the hydrological informationrequirements for long-term planning, of theproduction and flow of this information tothe user, and of the results of the use of suchinformation in the planning process, whichdemonstrates the basic water resource assessmentprogramme’s adequacy or inadequacy;(f) An estimation of the personnel and skillsrequired for an adequate basic water resourceassessment programme and appraisal of existingeducation and training programmes comparedwith current and future requirements;(g) A review of basic and applied research activitiesin the country and region, their adequaciesor inadequacies for water resource assessmentcompared with current and future needs,including needs for regional and internationalscientific and technological cooperation;(h) Definition of the major gaps in the programmewith regard to institutional framework, financialresources, instrumentation, techniques andothers;(i) Recommendations for eliminating inadequaciesof the basic water resource assessmentprogramme through national or regional cooperationand/or international aid.4.1.4 Water projectsWater is needed in all aspects of life. The overallobjective of water resource management is to makecertain that adequate supplies of good-quality waterare available for the population and various socioeconomicdevelopments of the society, whilepreserving the hydrological and biochemical functionsof the ecosystem. There is a growing awarenessthat development, including water resources development,must be sustainable. This implies that theworld’s natural resources must be managed andconserved in such a way as to meet the needs ofpresent and future generations.This chapter provides guidance on the applicationof the hydrological analysis methods described inChapters 5, 6 and 7 for the design and operationof water management projects in order to meet theabove-mentioned objective. In addition to theanalysis to be undertaken as described in thischapter, a number of social, economic and environmentalconsiderations should also be takeninto account; however, these are beyond the scopeof this Guide.4.1.5 Purposes served by a watermanagement projectAs explained in Chapter 3, an integrated approachto river basin planning and management is suitablefor handling the cross-sectoral activities. Theholistic management of freshwater as a finite andvulnerable resource, and integration of the sectoralwater plans and programmes within the nationaleconomic and social policy, are of paramountimportance. Consideration of equitable and

II.4-4GUIDE TO HYDROLOGICAL PRACTICESresponsible use of water is central to addressing theUnited Nations Millennium Development Goalsand eradicating poverty.The natural water cycle is spatially and temporallycomplex, and yet fulfilling human needs requires astable water supply. It is therefore essential to implementwater resources development and managementstrategies which generally involve some form ofengineering intervention. Pressures on watersystems due to growth in population and economicdevelopment have made it imperative that theengineering analyses needed for water developmentprojects be more impartial and scientific based thanin the past.A water management project may serve one or moreof the following objectives:(a) Municipal water supply;(b) Irrigation;(c) Industrial water supply;(d) Groundwater management;(e) Power generation;(f) Flood management;(g) Navigation;(h) Recreation, aesthetics and tradition;(i) Salinity and sediment control;(j) Pollution abatement;(k) Fish and wildlife conservation;(l) Other environmental considerations.4.1.6 Multi-purpose projectsWith the increasing level of development and use ofwater resources throughout the world, it is becomingever more important to plan projects that canserve a number of purposes simultaneously. Forexample, a planned storage reservoir may provideboth water supply and flood control downstream.Hydrological data required for the design of a multipurposeproject are basically an aggregate of thedata required for the various single purposesinvolved. The methods of analysis, although similarto those applied in design of single-purpose projects,are more complex. A series of plans involving combinationsof project sizes and methods of operationmust be made to determine the optimum plan.Conflicts can arise when attempting to managewater resources for a number of needs. The challengeof designing and operating systems to servemultiple functions is discussed in 4.2.4.1.7 Project cycleThe project cycle is illustrated in Figure II.4.2. Thecycle starts with the identification process, in whichthe following questions should be answered:(a) Is the project technically feasible?(b) Will total benefits exceed costs?(c) Who benefits?Section planningIdentificationEvaluationPreparationOperationAppraisalImplementationFigure II.4.2. Project cycle

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-5(d) Are there better alternatives?(e) Are social and environmental costs accepted?Project preparation calls for a clear definition ofproject goals and objectives. Once they are defined,it is possible to identify relevant problems and feasiblesolutions. In addition, it is necessary to analysepertinent data and information.4.1.8 Preliminary investigation of watermanagement projectsBefore appreciable expenditure of time andmoney can be justified for the planning of a watermanagement project, a preliminary investigationmust be made of its feasibility, desirability, scopeand its possible effect on those hydrologicalfactors that influence the environment and theefficiency of other projects. Although the investigationhas to be based on whatever material maybe available, for example fragmentary hydrologicalrecords, old maps and reports, it must becarried out with great care because it is at thisstage that conceptual planning decisions areoften made and that important aspects and consequencesof the project may become apparent. Ifthe preliminary investigation indicates that theproject potential is favourable, then more detailedstudies would be initiated.The types of hydrological data required for watermanagement are given in Table II.4.1 below.Table II.4.1. Data required for water managementPurpose Features Concern Required dataReconnaissanceHydrologyDrainage networkWatershedsSpringsDistinction of perennial from intermittent andephemeral streamsPhysiographyGeologyTopography and morphologySoil cover and typesUrbanizationMeteorology10, 11Temperature distributionWind distributionSnowpack distributionStreamflow1, 2, 3, 4, 7, 8, 9 – at selected sitesFloods4, 5, 6Groundwater12, 13Flood controlStructuresWater levelDepth–discharge relationship for important pointsHydraulic–topographic relationships in the floodplain4, 5, 6, 8Flood-plain occupancyRainfallStatistics of heavy rainfall in the general area underconsiderationPairs of floods and their causing precipitationFlood warningForecastTravel times of floodsTime lag between precipitation and runoffFlood synchronization at different tributariesPredictionTime series of floodsTime series of heavy precipitationFlood zoning andinsuranceFlood extentArea–duration–frequency of floodsScour and sedimentation by floods

II.4-6GUIDE TO HYDROLOGICAL PRACTICESPurpose Features Concern Required dataIrrigationDemandPrecipitation 10Evapotranspiration 11TranspirationSoil moisture Soil typeGroundwater levelSupplyStreamflow1, 2, 3, 4, 7, 8, 9Groundwater12,13Reservoir1, 2, 3, 4, 5, 6, 8, 9, 10, 11Groundwatermanagement,including rechargeAquifersReservoirs and pondsGroundwaterStreamflow12,131, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11Bank infiltrationStreamflow3, 4, 6, 7, 8, 9WellsStreamflow1, 2, 3, 4, 5, 6, 8, 9, 10, 11Power generationHigh-head damsStreamflow1, 2, 3, 4, 5, 6, 8, 10, 11Low-head damsStreamflow2, 3, 4, 6, 7, 8Tailwater depth–discharge relationshipNavigation Channels Water depthFlood flowsDepth–discharge relationship for important points2, 3, 7, 84, 6Rates of high water riseTime lag between rises at different points along thestreamsTime lag from heavy precipitation to high waterSnowmelt distributionMunicipal supplyRiversStreamflow andspringflow1, 2, 3, 4, 7, 9ReservoirsStreamflow1, 2, 3, 4, 5, 6, 8, 9, 10, 11Groundwater12,13Industrial useRiversStreamflow1, 2, 3, 4, 7, 8, 9ReservoirsStreamflow1, 2, 3, 4, 5, 6, 8, 9, 10, 11AquifersGroundwater12,13

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-7Purpose Features Concern Required dataRecreation,aesthetics andtraditionLakes and reservoirsPhysiographyClimateStorage–elevation relationshipShoreline propertiesWave possibilities97, 10, 11Air temperature distributionWind distributionRiversStreamflow1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11PhysiographyChannel geometryDepth–discharge velocity relationshipsBank soil and coverStreamflow2, 3, 4, 6, 7, 8, 9Reservoir release variationsSalinity andsediment controlDilutionCleaningStreamflowFloods2, 3, 4, 6, 7, 8, 94, 6, 8, 9Reservoirs1, 2, 3, 4, 5, 6, 8, 9, 10, 11Pollution abatementDilutionStreamflow1, 2, 3, 4, 7, 8, 9CleaningFloods4, 6, 9Reservoirs1, 2, 3, 4, 5, 6, 8, 9, 10, 11Fish and wildlifeconservationRiversStreamflowLakes and reservoirs2, 3, 4, 6, 7, 8, 9Water level fluctuation distribution9StructuresResulting changes in water depth, velocity,temperature, sediment load and bankcharacteristics, upstream and downstreamNote: Numbers refer to common hydrological data listed below.1 – Series of monthly and annual volume of streamflow2 – Mean daily discharge series3 – Low-flow frequency distribution4 – Frequency distribution of high discharges5 – Frequency distribution of large-volume floods6 – Shapes of typical flood hydrographs7 – Ice cover information8 – Erosion, sediment transport and deposition9 – Water quality10 – Precipitation distribution in space and time11 – Evaporation distribution in space and time12 – Aquifer extent and characteristics13 – Series of water levels of relevant aquifersVarious applications for water managementare discussed in the following sections of this chapter:4.2 provides information about estimating yieldand fixing reservoir capacity; 4.3 is devoted to floodmanagement; 4.4 to irrigation and drainage; 4.5 tohydropower and energy-related projects; 4.6 tonavigation and river training; 4.7 to urban waterresources management; 4.8 to sediment transportand river channel morphology; 4.9 and 4.10 aredevoted to environmental issues.4.2 ESTIMATING RESERVOIR CAPACITYAND YIELD [HOMS K75]4.2.1 GeneralThis section addresses the yields achievable andthe storages required to maintain certain levels ofyield, with respect to water resource systems. Thefocus is on surface water, although the waterresources practitioner should always be aware of

II.4-8GUIDE TO HYDROLOGICAL PRACTICESthe hydrological interdependencies betweensurface water and groundwater which simplyconstitute different occurrences of the sameresource in the hydrological cycle. Most principlesare explained with respect to single river and singlereservoir systems, although approaches for dealingwith complex water resource systems comprisingmultiple reservoirs in different basins are alsoaddressed.The yield from a water resource system is thevolume of water that can be abstracted at a certainrate over a specific period of time, generallyexpressed as an annual volume, such as million m 3per year. The rate at which water needs to beabstracted may vary throughout the year, dependingon the intended use. For domestic, industrialand mining uses, water is required at a relativelyconstant rate throughout the year, whereas strongseasonality occurs with respect to irrigation.Natural streamflow, in contrast, is much more variable.Rivers typically display strong seasonality intheir natural runoff, compounded by withinseasonfluctuations in flow as well as largevariations in total annual runoff.If a constant abstraction rate is considered, thehighest yield that can be abstracted from an unregulatedriver is equal to the lowest flow in the riveras demonstrated in Figure II.4.3. By regulatingstreamflow by means of dams, water can be storedduring periods of high flow for release during periodsof low flow, as shown by the dotted lines on thediagram. This increases the rate at which water canbe abstracted on a constant basis and, consequently,the yield. The greater the storage, the greater theyield that can be abstracted within the constraintsof the physical characteristics of the system. Largerannual yields can also be obtained where seasonaldemand patterns show good correlation with thestreamflow characteristics. For ease of description, aconstant abstraction rate is used in the remainderof the chapter, unless otherwise specified.In areas where the average annual streamflow, ormean annual runoff, is well in excess of waterrequirements, but where the minimum streamflowmay drop below the required abstraction rate, thefocus is typically on determining the reservoircapacity required for bridging the period of lowflow in order to maintain the desired yield. Asstreamflow varies from year to year, low flows (similarto floods) are not always of the same severityand duration. The amount of water that can beabstracted without failure, the yield, consequentlyalso varies from year to year. Consideration musttherefore also be given to the economics of whethersufficient storage should be provided to maintainthe yield even under the most severe low flow(drought) conditions, or whether it would be moreeconomical to provide less storage and therebyaccept some degree of failure from time to time,with respect to supplying the full yield. Thus thechallenge is to weigh the expected benefits againstthe risk, the associated costs of failure and the costof storage.Where streamflow is limited or where a waterresource is already highly utilized, the focus shiftsto the optimal utilization of water and the yieldachievable from different storage capacities, ratherthan determining the storage required to maintaina desired yield. In such cases resource optimizationadds an additional dimension to the initial problemof risk versus cost optimization.With the high degree of water resources developmentand utilization already existing in most partsof the world, resource optimization is becomingincreasingly important. Due attention is thereforegiven to aspects of resource optimization in thedescriptions that follow.HighFlow rate or yieldZeroStorage of high flowsYield withstorageTypical streamflowsequenceYield withoutstorageOnly one year shown.Storage and release cyclestypically extend over severalyears.Release fromstorageJan Feb Mar April May June July Aug Sept Oct Nov DecFigure II.4.3. Streamflow and storage4.2.2 Concepts of yieldThis section presents the main parameters and keyconcepts related to the determination of yield, asreferred to in this chapter.4.2.2.1 SequenceA sequence is defined as a chronological time seriesof data such as streamflows, precipitation and evaporationfor a particular location. Monthly data aremost commonly used with respect to the determinationof yields for large water resource systems,although longer or shorter intervals may beselected.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-9As can be interpreted from Figure II.4.3, the configurationof the streamflow sequence has adeterminant influence on the yield that can beobtained from a river. Should the streamflow beconstant, for example, a yield equal to the averageflow can theoretically be abstracted without anyflow regulation. The larger the variability of streamflow,the greater the requirements for regulationthrough storage.4.2.2.2 Natural influencesLosses from rivers, such as those due to evaporationand infiltration, are part of the hydrological cycleand are reflected in observed streamflow recordsused as a basis for the determination of yield. Thecreation of large water bodies as a means of flowregulation changes the pre-project natural equilibriumconditions, normally resulting in additionalevaporation and infiltration.4.2.2.2.1 EvaporationThe depth of water that evaporates annually from areservoir surface may vary from about 400 mm incool, humid climates to more than 2 500 mm inhot, arid regions. Therefore, evaporation is animportant consideration in many projects anddeserves careful attention.Methods for estimating reservoir evaporation frompan observations and meteorological data aredescribed in Volume I, 4.2, of this Guide. In theabsence of pan evaporation or other appropriatemeteorological observations at or near the reservoirsite, regional estimates of these quantities are usedto assess reservoir evaporation.It is important that the net evaporation be takeninto account in the water balance calculations (seeVolume I, 4.2.3), thus allowing for precipitation onthe reservoir surface. As periods of high and lowstreamflow normally correspond to wet and dryclimatological conditions, the most realistic representationof the actual conditions is obtained wherethe precipitation or evaporation and streamflowsequences are fully synchronized. Should correspondingevaporation data not be available, as oftenoccurs in practice, average monthly data could beused, at least to reflect seasonal variations.Evaporation from reservoir surfaces not only representsa loss in potential yield, but also a loss of waterfrom the surface water resource system, and thus anet loss of resource. Specific attention should thereforebe given to the minimization of evaporation.Wherever practicable, the minimum storage surfacearea per unit volume of storage should be sought inthe selection of dam sites. Extensive research hasbeen conducted into evaporation suppression bythe spreading of monomolecular films on watersurfaces (see Volume I, 4.4.1), but practical problemsin the application of these techniques to largestorages still remain unsolved. Thermal stratificationin reservoirs and the temperature differencebetween inflow and outflow can have a significantimpact on reservoir evaporation. These influencesare difficult to reliably quantify from theory andcan best be judged from comparisons to existingstorages.4.2.2.2.2 InfiltrationLosses from reservoirs due to infiltration or seepageare highly dependent on local hydrogeologicalconditions.Dam sites are mostly selected where foundationconditions are good and the geological formationsunderlying the reservoir basin are relatively impermeable.In such cases, groundwater levels in thevicinity of a new reservoir generally stabilize arounda new equilibrium some time after the new reservoirhas been brought into operation, with relativelysmall and even insignificant losses due toinfiltration.Where a reservoir basin or a part thereof is underlaidby permeable strata such as sands, dolomiteand karst formations, the losses can be significant.Measures to control such losses can be technicallydifficult and expensive to implement, and canrender a project unfeasible.Estimates of expected infiltration and seepagelosses may be derived from geological investigationsof the reservoir basin and dam site, andfrom comparisons with existing reservoirs insimilar conditions. Unlike losses caused by evaporation,those caused by infiltration and seepagedo not necessarily constitute a net loss of resourceas they may contribute to groundwater rechargeor to discharge downstream from a controlstructure.4.2.2.2.3 SedimentationSediment deposition in reservoirs reduces the availablestorage over time and therefore can have animpact on the long-term yield from a reservoir andthe feasibility of a reservoir design project.Any reservoir design must therefore account for thevolume of sediment expected to accumulate over

II.4-10GUIDE TO HYDROLOGICAL PRACTICESthe economic life of a dam and to provide for thisthrough an equivalent volume of additional storagein the design of the reservoir or through an assumeddecrease in the storage, and therefore the yield,over time. If it has deposited sediment in a reservoir,the river is likely to erode its downstreamchannel more than previously, and this should betaken into account during early planning stages.It is essential to develop a bathymetric surveymonitoring programme to ensure that expectedsediment accumulations are consistent with thosethat occur in reality. This is especially true inregions where sediment transport is episodic andlinked to unpredictable and extreme hydrologicalevents, such as in semi-arid and arid areas, andalso where catchment land-use changes haveincreased sediment production. Bathymetricsurveys can be undertaken using standard waterdepth sounding methods, or through appropriateremote-sensing approaches. Depending on thesediment load and grain size distribution, as wellas the streamflow and reservoir basin characteristics,much of the sediment deposition may occurin the upper reaches of a reservoir basin. Thismakes it expensive or difficult to remove the sediment.However, it is possible to design outlet works(scour gates) that can be used periodically to scoursome of the accumulated sediment from a reservoirbasin. More detail on sediment discharge,transport characteristics and the possible scouringof sediment can be obtained from Volume I,Chapter 5, and Volume II, 4.8.4.2.2.3 Human influencesWater resources developments and some land-useactivities upstream of a project site alter the naturalstreamflow characteristics at the project and canhave significant impacts on its yield. Water resourcesdevelopments can include regulation structures,diversions, abstractions, return flows and transfersfrom other catchments. Land-use activities with thegreatest impact on water resources and sedimentloads are as follows: urbanization, afforestation,deforestation, cultivation of certain crops such asrice and sugar, denudation of land and some formsof rainwater harvesting. The subsequent sections ofthis chapter relate to such activities and provideadvice that is of use in assessing the likely impact ofhuman influences.It is important that human influences be properlyaccounted for in determining the yield from a waterresource system. In particular, any trends should benoted, and due consideration be given to possiblefuture developments.4.2.2.4 Observed streamflowObserved or actual streamflow sequences refer tothe streamflow data as recorded in the field (seeVolume I, Chapter 5). Therefore, such records inherentlyreflect the impacts of human influences and,with the exception of natural, or virgin, catchments,show some variations over time. In general,observed sequences require some processing for theinfilling of missing data (see Volume I, 9.7.2) and toaccount for the impacts of development.4.2.2.5 Naturalized streamflow sequencesNaturalized streamflow sequences are representativeof the streamflow conditions prior to influencesby humankind. In totally undeveloped catchments,the observed streamflows reflect the natural conditionsperfectly. For catchments where developmenthas occurred, realistic estimates can be made ofwhat the streamflows would have been under naturalconditions by calculating the impacts of thevarious influencing factors and adjusting theobserved streamflow sequences accordingly.4.2.2.6 Synthetic streamflow sequencesA synthetic streamflow sequence is one that is artificiallyproduced by using a computer model. Twokinds of synthetic sequences are used with respectto streamflow: deterministic sequences and stochasticsequences.Deterministic sequences are mainly used to fill inand extend incomplete streamflow sequences. Thisis achieved through the use of hydrological models,as described in Chapter 6.A stochastic sequence is one that randomly variesin time, possibly with some dependence structure,and purports to offer alternatives to the observedsequence as a means of assessing what might plausiblybe experienced in future (Box and Jenkins,1970), (Pegram and McKenzie, 1991) and (Hipeland others, 1977). The statistical properties ofstochastically generated sequences are such thatthey are considered to originate from the samepopulation and to be generated from the samenatural processes that characterize the natural ornaturalized sequences on which they are based. Thestochastic sequences referred to in this chapterprimarily relate to streamflow and are used to studythe probabilistic behaviour of yield from reservoirs.However, the same principles for selection andprocessing can be applied to sequences of rainfalland other hydrological variables of importance tothe investigation of water resource systems.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-114.2.2.7 Target draftTarget draft is the volume of water that one aims todraw from a reservoir or water resource system tosupply requirements over a specified period, generallyexpressed as an annual total.4.2.3 Estimation of storage–yieldrelationshipsMany computer models have been developed forthe calculation of storage–yield relationships andare relatively easily accessible. This section brieflydescribes the basic principles underlying thesemodels; further details are provided in 4.2.4 and4.2.5.4.2.3.1 Numerical procedureIn its most basic form, yield analysis is a simplesequential mass balance exercise between waterentering a reservoir (streamflow, precipitation) andwater released or lost from the reservoir (abstraction,evaporation, spillage). The equation to besolved is the following:S i= S i–1+ I i+ P i– E i– D i– O i= S i–1+ ΔS i(4.1)where S irepresents the storage at the end of timeinterval i, S i–1represents the storage at the beginningof time interval i, I iis the inflow during intervali, P iis the precipitation during interval i, E iis theevaporation during interval i, D iis the draft orabstraction during interval i, O iis the outflow orspillage during interval i and ΔS iis the change instorage during interval i.Where a time step of a week or longer is used, theaverage surface area of the reservoir between timeintervals t i–1and t iis used to calculate the volumesof precipitation and evaporation.Where the storage needs to be determined to maintaina certain draft, the equation is solved fordifferent assumed maximum storage capacities (S)to find, in an iterative way, the capacity where thereservoir is drawn down to barely touching empty,or the minimum operating level. Where a damalready exists or the storage is fixed, the abstractionrate which can be maintained can be determinedby substituting storage with draft as the variable inthe equation. The sequence of levels of reservoirstorage which result from solving the equation isreferred to as the storage trajectory.The trajectory will generally be bounded by thefull and minimum operating level states. Ingeneral, the trajectory for a given inflow sequenceand abstraction rate will be a function of the startingstorage level and will differ from starting levelto starting level. However, once corresponding fullor minimum operating level states have beenreached for the range of starting storages, thetrajectories from that point onwards will be indistinguishablefor a given inflow sequence andabstraction rate.The period of maximum drawdown of a reservoir,that is, from a full state of storage down to the minimumoperating level and recovering until it reachesthe full level again, is referred to as the criticalperiod. To reach stability in the analyses, it is importantthat the critical period be clearly defined bythe reservoir trajectory.Careful inspection of the trajectory with respect tothe occurrence of low flow periods remains important,however. A potentially more severe low flowperiod than defined by the critical period may occurat the beginning or end of the inflow sequence, butwhere the first or last part of such a potentiallymore severe low flow period may be truncated bythe record length of the inflow sequence available.Should this be suspected, an adjustment may judiciouslybe made to the abstraction rate by accountingfor the net change of storage over the period of thesequence analysed.In the simplified case described above, the yield ofthe system was assumed to be equal to the abstractionrate. One may, however, aim to abstract moreor less water from a resource rather than the yield ofthe reservoir or water resource system. The relevanceof target draft to the yield characteristics of awater resource system is described in more detail in4.2.4.4.2.3.2 Graphical approachThe graphical approach offers a simple alternativefor visually presenting the results of sequentiallysolving equation 4.1.In a reservoir subject to an inflow I and draft D, thestorage S at time t is mathematically defined as:ttS t = S 0 + ∫ ( I − D )dτ = S 0 + ∫ Idτ00t(4.2)* *− ∫ Ddτ = S 0 + I t − Dt0(For ease of demonstration, the influences of evaporationand precipitation are not included above,

II.4-12GUIDE TO HYDROLOGICAL PRACTICESand the draft is representative of all outflow. Spillagewould occur where the inflow mass curve exceedsthe draft mass curve as shown in Figure II.4.4.)Plots of the cumulative sums I* and D* representthe inflow and draft mass curves, respectively, withS 0being the initial reservoir storage. Figure II.4.4illustrates how the required storage capacity S isdetermined for a constant draft D with the constraintthat no failure is allowed during the sequenceanalysed. The procedure employs the concept of asemi-infinite (bottomless) reservoir. The constantdraft corresponds to a constant slope of the draftmass curve D*. A line, parallel to D*, is drawnthrough each peak on the inflow mass curve I*. Thedesign storage capacity S is the maximum verticaldistance between any point on I* and any of thelines that are parallel to D*.The graphical approach was widely used in the past.However, computing power has increased enormouslyover the years, facilitating the solution toequation 4.1. In addition, the digital approachoffers great flexibility in analysing various scenarios;as a result, the graphical approach is nowseldom used, if ever.4.2.3.3 Influence of record lengthAlthough there are no formal guidelines for theminimum period of record, reasonable stabilitywith respect to yield analyses is generally reachedwith a record length of 10 to 20 times the criticalperiod. Where little variability in streamflow occursand where the need is mainly for seasonal storage(less than one year), a minimum record period of10 to 20 years may be acceptable. However, inAccumulated inflow and outflow in 10 9 m 39876543210Inflow mass curve l*max d = SOutflow mass curve D*Radial scale for discharge in m 3 s –15040d t 30D*20l* t1001 2 3 4 5 6 7 8 9 10 11 12Time t in yearsFigure II.4.4. Mass-curve approach for determiningreservoir storage capacitysemi-arid to arid areas, over-year storage is generallyrequired, as critical periods of 5 to 10 years andlonger are common. A record length of 50 to100 years should preferably be used in such cases.Even where reasonably long streamflow recordsexist, worse floods and worse droughts than thosehistorically observed are bound to occur in future. Itis also virtually certain that the exact configurationof a streamflow sequence, as recorded in the past,will never be exactly repeated in future. It is evident,however, that the longer the period of record onwhich the inflow sequence is based, the more reliablethe estimation of the yield is likely to be. Whilehistorical records are the only factual informationavailable, improved perspective on possible futureextreme events can be gained through the stochasticgeneration of streamflow, as described in 4.2.2.6.4.2.4 Classifications of yieldThe yield characteristics of water resource systemsare more complex than can be described by a singleformula such as equation 4.1 and requires a morecomplete description than that already beenprovided.Concepts were developed for the classification ofyield from a reservoir or water resource system asbase yield, firm yield, secondary yield, non-firmyield and average yield (Basson and others, 1988).These facilitate a graphic representation of thebehaviour of a reservoir or water resource system asshown in simplified form in Figure II.4.5. The valuesfor defining the diagram are obtained from solvingequation 4.1 for various target drafts and by recordingthe relevant results.Such diagrams enhance further understanding ofthe behaviour of a system under various operationalconditions. They are particularly useful where waterresources are highly utilized, where high variabilityof streamflow occurs and where yield determinationand management of complex water resourcesystems is necessary.Base yield is defined as the minimum yield over aspecified number of consecutive time intervals thatcan be abstracted from a river or reservoir systemfed by a given inflow sequence while attempting tosatisfy a given target draft associated with a specifieddemand pattern for water and a specifiedoperating policy. The base yield initially increaseswith increased target draft until a stage is reachedwhen the reservoir is unable to yield continuouslyat the target draft, resulting in base yields lowerthan the target draft.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-13Firm yield is defined as the maximum base yield.For analyses based on historically observed streamflows,reference is made to the historic firm yieldto distinguish it from the firm yields derived byprobabilistic methods as described in 4.2.5. Thehistoric firm yield associated with a particularreservoir capacity may vary with length of inflowsequence. In fact, it is likely to be smaller for alonger inflow sequence because there is anincreased probability of a more severe low-flowsub-sequence occurring.The yields obtained according to the methodologiesdescribed in 4.2.3.1 and 4.2.3.2 are analogousto the historic firm yield.Secondary yield is the yield that can be abstractedin excess of target draft. As defined, it is withdrawnfrom a reservoir only while the reservoir is at itsfull supply level. The assessment of secondaryyield for various maximum abstraction capacitiescan therefore provide a valuable measure of thepotential for further development of a waterresource system. Secondary yield is often used forthe generation of additional (secondary) hydropoweror other interim beneficial uses, whereappropriate facilities exist.Non-firm yield is the average yield that can beabstracted from a water resource system in excess ofYield →Secondary zoneNon-firm zoneFirm zoneTD 1TD 2A 1Draft →Installed abstraction capacityTarget draftFigure II.4.5. Simplified draft–yield responsediagrambase yield, but not exceeding the target draft. Thisis not a continuous yield and cannot, therefore, berelied on at a specific assurance of supply.Average yield is the sum of the base yield and nonfirmyield averaged over the period analysed. Itprovides a measure of the yield which can, onaverage, be abstracted from a system, but wherepart of the yield cannot be continuouslysupported.Total yield is simply the sum of the base, secondaryand non-firm yields.Further interpretation of the draft–yield responsediagram is given by Basson (Basson and others,1994).4.2.5 Probabilistic approach4.2.5.1 Storage–flow dependabilityrelationshipThe design and operation of a storage reservoir is animportant component of most water resourcesdevelopment projects. While the design storagecapacity will depend on the demand for water to bemet from the reservoir, the major factor affectingthis decision will be the available flow in the riverat the location of the planned reservoir. It fluctuatesfrom year to year, however, and this variabilitymust be taken into account.A correct estimation of reservoir capacity is veryimportant. If it is not sufficient, the project will notserve the community to its full sustainable potentialand may lead to wasting scarce water resources.However, over-estimation of the storage capacitymay result in the reservoir rarely filling despite itshigh construction cost, thus rendering the projectuneconomical. Therefore, the criteria for choosingreservoir size should include not only the overalldemand, but also the reliability with which thatdemand should be met. For example, 75 per centdependable yield means that the quantity of waterrequired for irrigation will be availability for at leastthree out of each four years. A 100 per cent dependableyield means that the required supply of waterwill be available every year – a 100 per cent successrate – but this can only be achieved if the supplyrate is less than that for 75 per cent dependableyield. This in turn would only satisfy a far lowerdemand.Different countries have different criteria for planningwater resources projects. The concept of apercentage dependability, where a certain level of

II.4-14GUIDE TO HYDROLOGICAL PRACTICESfailure is acceptable, is frequently adopted in developingcountries because they give paramountimportance to the economic feasibility of projects.On the other hand, in developed countries – theUnited States of America, for example – the principlecriterion is to meet the requirements for aparticular purpose with nearly 100 per centcertainty. The percentage also varies according tothe type of services to be provided by the reservoir.As a rule, it may be set at 75 per cent for irrigation,90 per cent for hydroelectric power generation and100 per cent for domestic water supply projects.The following methodology can be employed fordetermining flow of a certain reliability at a particularpoint in a river:(a) Annual gross yield, or the natural flow volume,is also known as the virgin, or historical, flow. Itis defined as the flow that would have occurredat that point of the river had there been neitherany abstractions from nor additions to the flowupstream from sources outside the river system.In this, both natural seepage and evaporationare ignored. The natural flow can be determinedby adding the observed flow, upstreamwater used for irrigation, domestic and industrialuses both from surface and groundwatersources, increases in water volumes held bythe reservoirs (both surface and subsurface)and evaporation losses from the reservoirs, anddeducting return flows from different uses fromsurface and groundwater sources. This is representedby the following equation:R n= R o+ R ir+ R d+ R gw– R ri– R rd– R rg+ S + E (4.3)where R nis the natural flow, R ois the observedflow, R iris the withdrawal for irrigation, R dis thewithdrawal for domestic, industrial and otherrequirements, R gwis the groundwater withdrawal,R riis the return flow from irrigated areas,R rdis the return flow from domestic, industrialand other withdrawals, R rgis the return flowfrom groundwater withdrawal, S is the increasein storage of the reservoirs in the basin and E isthe net evaporation from the reservoirs.If inter-basin transfers are involved – whetherinto or from the river basin – the amountsthereof will have to be respectively deductedfrom or added to R n;(b) To ascertain the percentage dependability ofthe flow at a given point on the stream where acontinuous record of natural flows for a numberof N years is available, the annual values ofnatural flows are arranged in a descendingorder. Each year’s flow so arranged is assignedthe serial number from top to bottom and if Mis the serial number of the flow in any year, thepercentage dependability for the flow of thatyear (D) is calculated by applying the formula100M/N. Some authorities prefer the formulato be expressed as 100M/(N+1);(c) The year that would represent a particularlydesired percentage of dependable flow can bedirectly ascertained by rearranging the relationshipto M = DN/100 or D(N+1)/100 and theamount of flow of that dependability read outfrom the natural flow series. In cases where thederivative of M is not a whole number, a smalladjustment may be required in the values offlows of the two years between which M falls soas to achieve the closest dependable flow correspondingto the exact percentage of dependability;(d) The same results are obtained by creating anascending series of natural flows rather than adescending series.The natural flows worked out by equation 4.3 canalso be used to apportion the flow in river amongvarious potential users such as riparian States.4.2.5.2 Risk of failure and reliability of supplyMany definitions of failure of a water resourcesystem can be formulated. The definition favouredin this chapter is where failure of a water resourcesystem is defined as the inability of the system tosupply the base yield associated with a specifictarget draft. Risk of failure of a water resource systemcan be defined as the probability of not being ableto supply the base yield associated with a specifictarget draft at least once over a specified timehorizon.It is common practice to make use of the recurrenceinterval concept to quantify risk of failure of a waterresource system. Typical recurrence intervals associatedwith large systems are 1 in 20, 1 in 50, 1 in 100and 1 in 200 years. The probability of failing in aparticular year, the annual risk of failure, is thereciprocal of the recurrence interval. Therefore, a2 per cent probability of failure in any one year isequivalent to a recurrence interval of 50 years.Thus:R = 1/T (4.4)where R indicates the annual risk of failure and Tindicates the recurrence interval of failure.The probability of successfully meeting the requirementsfor water in a particular year, the annualprobability, is simply one minus the annual risk of

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-15failure. Annual reliability of supply is thereforerelated to the recurrence interval of failure by thefollowing relationship:r = 1 – 1/T (4.5)where r is the annual reliability of supply.The long-term risk of failure is related to annual riskof failure by the Bernoulli probability relationship:R n= 1 – (1 – R) n = 1 – (1 – 1/T) n (4.6)Target draft (10 5 m 3 yr –1 )Derived from 41 sequences,64 years’ duration eachBase yield (10 5 m 3 yr –1 )where R nis a long-term risk of failure and n representsa planning period (length of sequence) inyears.Although some assessment of the risk of failuremay be made from analyses of an historicalstreamflow sequence, the confidence that can beattached thereto is normally not very high, unlessexceptionally long records exist. Stochasticallygenerated streamflow sequences are thereforeemployed as a means of increasing the samplesize of possible configuration of streamflowsequences in order to obtain improved statisticalassessment.4.2.5.3 Draft–yield reliability characteristicsThrough the analyses of a large number of stochasticallygenerated streamflow sequences, typicallybetween 200 and 2 000 sequences of the same durationas the historic sequence, a probabilisticYield (10 5 m 3 yr –1 )Target draft (10 5 m 3 yr –1 )Figure II.4.6. Comparison of long-term yieldcharacteristicsExceedance probability of base yield(as percentage of sequences analysed)Figure II.4.7. Family of long-term draft–yieldreliability characteristicsassessment of the characteristic behaviour of awater resource system can be obtained.While many well-proven stochastic hydrologicalmodels have been developed, extensive tests andre-sampling need to be performed to ensure thatthe basic parameters of the historical records, onwhich the models are calibrated, are well preservedfor each point of interest. The validation of modelsis particularly important in semi-arid and arid areaswhere large variations in streamflow occur.Figure II.4.6 reflects the addition of probabilisticinformation on the basic draft–yield responsediagram. The partial box plots indicate thesampled distribution of base yields resulting fromthe analysis of a large number of generatedstochastic sequences, at target drafts of 70, 80 and90 million m 3 per year. The shape and somewhatsteeper decline of the historic base yield line,compared with the 1:100 probabilistic base yieldline, are attributable to the specific configuration ofthe historic critical period, whereas the probabilisticline displays a combined value from the analysisof a large number of inflow sequences.Additional perspective can be gained by presentingthe draft–yield reliability characteristics as shownin Figure II.4.7. These curves also form the basis forthe assessment and management of complex multireservoirsystems as described in the sections tofollow.For the example in Figure II.4.7, should a yield of60 million m 3 per year be required from a system ata risk of failure not exceeding 1/100 years, itcan be achieved by imposing a target draft of

II.4-16GUIDE TO HYDROLOGICAL PRACTICES82 million m 3 per year on the system (see dottedline). The additional 22 million m 3 per year willthen be available at a risk of about 1/80 years. Theadditional water can be used for applications inwhich a lower assurance of supply is required, suchas the generation of secondary hydropower or thesupport of adjoining or other water resourcesystems. Alternatively, the storage can be reducedso that a firm yield of 60 million m 3 per year can beobtained at the specified 1/100 year risk of failure.4.2.5.4 Short-term yield characteristicsWhereas the long-term yield-reliability curvescapture the long-term yield capabilities of awater resource system and provide perspectiveon the long-term average behaviour thereof,they do not contain sufficient information tomake short-term operational decisions. Therethe influence of the ruling state of storage is ofparamount importance. However, decisions withrespect to real-time water allocations cannot bebased solely on the current situation, but shouldalso account for safeguarding the supply forsome period into the future. The duration of thissafeguarding period should be a few time stepslonger than the time step between major operationaldecisions.Short-term draft–yield reliability characteristics aredeveloped in the same way as the long-term familyof curves, except that short-term curves also relateto a specific starting storage. Curves therefore needto be developed for a range of starting storages.Because of the shorter duration of the sequencesused, typically two to five years, many sequencesmay not span a critical period. Therefore, significantlymore short-term sequences need to beanalysed to achieve convergence than for the longtermanalyses. More detail on the practicalapplication of short-term characteristic curves inthe real-time operation of water resources systemsis given by Basson and others (1994).4.2.5.5 Reservoir filling timesWhen a new dam is built, a certain stage of storagein the reservoir must be reached before the full yieldcan reliably be abstracted from it. In semi-arid andarid areas, as well as where water resources developmenthas practically reached its full potential, itmay take several years after the start of impoundmentto reach the first filling of the reservoir, evenif no water is abstracted during this period. This canhave a major impact on the planned developmentphasing, as well as on the economic feasibility of aproject.Probabilistic projections of filling times for newreservoirs can be obtained by determining the reservoirtrajectories for a large number of stochasticinflow sequences, starting empty. A practical durationfor the analyses should be selected, whilevarious options for incrementally imposing drafton the reservoir may also be tested. Figure II.4.8shows a probabilistic assessment of the filling timefor a reservoir.4.2.6 Multi-purpose reservoirs andoperating rulesMost storage reservoirs serve a number ofpurposes, as shown in 4.1. It is generally not practicableto allocate a fixed portion of storage foreach purpose. In most cases, such an allocation isrestricted to emergency purposes. For example, abuffer zone is often created immediately abovethe dead-storage zone and is reserved for use inexceptional circumstances, such as flushing thedownstream river section in case of accidentalcontamination, emergency water supply to dealwith sudden health hazards or fire fighting.However, most purposes are served from the samestorage and their requirements are accommodatedby complex release rules for reservoiroperations. Different users will require differentquantities of water at different times, with differentassurances of supply.Reservoir releases are often formulated in terms ofrule curves that indicate the rate of release as afunction of the ruling or instantaneous storageand the time of the year. Different assurance ofsupply requirements also imply that different categoriesof users have different tolerances to copeVolume in storage (10 6 m 3 )180160140120100806040200Full supply capacityMinimum %1%5%25%50%75%95%99%Maximum %6 12 18 24 30 36 42 48 54 60Time (months)Figure II.4.8. Probabilistic assessment of reservoirfilling time

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-17with some degree of failure in their water supply.In many parts of the world it is often not feasiblefor a project to be developed to meet the fullrequirements for water all the time. Strategies arethen introduced to incrementally curtail thesupply of water to some users when critical levelsof storage are reached.Some uses of a reservoir do not require therelease or abstraction of water. However, certainstorage limits at specific times of the year, whichmay impact on the yield characteristics of thereservoir or water resource system, should beobserved, for example, when reservoirs are usedfor flood control, recreation, salinity control andwhen environmental considerations areinvolved.The design and operation of multi-purpose reservoirsrequire complex analyses, which are generallycarried out by iterative methods that involveadjustments of the rule curves and evaluation ofthe effects on all individual purposes in order tooptimize water resource system management.Formal optimization techniques can also beemployed to find the best solution among certaintrade-offs. It is important that all potential usesand users of a reservoir be taken into accountduring the planning stages of a project and thatconsideration be given to the real-time operatingrules in advance of the construction. Where multipurposereservoirs or complex water resourcesystems are involved (see 4.2.7), it is advisable thatthe operating rules be developed as part of theplanning process.Examples of rule curves and approaches to real-timeoperating rules can be found in Box and Jenkins,1970; Basson and others, 1994; Loucks and others,1981 and Svanidze, 1977.4.2.7 Multi-reservoir water resourcesystemsOwing to the high degree of water resourcesdevelopment in many parts of the world and itssteady growth in others, the occurrence of multiplereservoirs in a basin is becoming more andmore common. These may be in a series downstreamof one another on the same river, inparallel on branches of the river, or variouscombinations thereof in a catchment or riverbasin. Reservoirs in adjoining catchments ordifferent basins may also be linked togetherthrough the transfer of water, resulting in evenlarger and more complex water resource systems,such as shown in Figure II.4.9.Where two or more reservoirs are linked togetherthrough their location on the same river system orvia transfers, one will impact on the other, even ifonly with respect to shared downstream releaserequirements. Such reservoirs are inherently part ofthe same system, and need to be recognized as suchin the management of the resource. In manyinstances, the introduction of inter-reservoir operatingrules may be required. Any new additions alsoneed to be evaluated and later managed in thecontext of the overall system.Where initial single reservoir projects develop intomulti-reservoir systems over time, the operation ofexisting components may have to be changed.Major changes are often difficult to implementbecause of many legal, political, economic andphysical constraints. Accordingly, the level of optimizationthat can be reached in practice in suchcases is generally low.Where greater flexibility exists or can be added,significant benefits may be achieved from the operationof reservoirs as one interconnected system.Generally, this is further enhanced where differentbasins are linked together through the transfer ofwater. Individual reservoirs or sub-components of asystem may, for example, be operated at a targetdraft which is in excess of the firm yield of the reservoiror subsystem, but with the knowledge that itcan be supported from other parts of the systemduring periods of deficient yield. In this way anoverall yield can be obtained which is greater thanthe sum total of the firm yields of the componentparts of the system.It is strongly recommended to add a probabilisticdimension to the management of multi-reservoirwater resource systems. This requires that stochasticstreamflow sequences be generated for eachpoint of interest in the system. Of specific importancein this regard is that the cross-correlationamong observed streamflow sequences at therespective points be carefully preserved. Much ofthe confidence related to the probabilistic managementof water resource systems is dependent on theaccurate replication of these characteristics in thegenerated sequences.The multi-dimensional problem presented by theprobabilistic management of water resources as oneinterconnected system is comprehensively coveredin the literature. It is evident that determination ofthe yield characteristics, as well as operationalmanagement of multi-reservoir water resourcesystems, can be very complex and can generally bedone solely with the aid of sophisticated computer

II.4-18GUIDE TO HYDROLOGICAL PRACTICESFigure II.4.9. Multi-reservoir water resource system extending over several river basinsmodels. Several models have been developed, mostof which can be obtained from the relevant organizationsor institutions, normally subject to someform of licence or agreement. Reference may bemade to Hatch Energy, Canada (www.hatchenergy.com); BKS Group (www.bks.co.za) and Departmentof Water Affairs and Forestry, South Africa (www.dwaf.gov.za); Danish Hydrological Institute,Denmark (www.dhisoftware.com); HydrologicalEngineering Centre, US Army Corps of Engineers(www.hec.usace.army.mil) and Deltares,Netherlands (www.wldelft.nl).4.2.8 Incidental effects of reservoirsThe purpose of this section is to create a generalawareness of the incidental effects of reservoirs;however, the subject is not addressed in great detail.The section focuses on the effect of impoundmentin reservoirs created by dams, not with the directimpacts of dam structures or hydropower stations,such as the creation of barriers to fish migration.The important social impacts of reservoirs are notaddressed either.4.2.8.1 Effects on hydraulic and hydrologicalregimesThe construction of a dam causes changes in thehydraulic and hydrological regimes downstream.Consumptive uses of water reduce the mean flow,while reservoir regulation changes the seasonaldistribution of flow and generally reduces its variability.The detention of water in the reservoir causessedimentation and results in released water withgreater transport capacity than the inflow, whichcan cause erosion below the reservoir. The decreasein hydraulic gradient may cause backwater andsedimentation problems in the river channelupstream of the reservoir.4.2.8.2 Environmental effectsEnvironmental effects are of increasing concern inthe planning and management of water resourcesprojects.The construction of reservoirs generally has a veryimportant impact on the environment. Where the

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-19storage volume is large in relation to annual runoffand there is a high water nutrient load level,eutrophication can have a significant impact on thequality of the water as a result of long residencetimes. The reservoir has a major effect on thetemperature and oxygen content of the releasewater. Less turbid water resulting from sedimentdeposition allows for deeper light penetrationwhich may cause algal blooms. The regulation offlow is also associated with a change in the natureof land use and increased water use downstream ofthe reservoir. This generally results in an increase inthe amount of wastewater produced, which maylower the quality of water in the receiving river.Changes of this nature are a major concern.However, reservoirs also cause changes with beneficialeffects. In many cases, if managed appropriately,the environment in the vicinity of reservoirs anddownstream may be greatly improved by providingrecreational, aesthetic, ecological and healthbenefits.It is of primary importance to provide monitoringfacilities for measuring environmental factors bothbefore and after construction and to assess continuouslyall environmental effects of storagereservoirs.4.2.8.3 Environmental flow requirementsAs a means of mitigating the impacts of reservoirson downstream aquatic life, releases are made to atleast partly recreate some characteristics of thenatural flow regime necessary for maintaininghealthy ecosystems. Such releases need to beallowed for during the planning phases of a projectand in determining the yield characteristics of awater resource system. Environmental flow requirementscan have a major impact on the abstractableyield from a system, particularly where a highconservation status of a river needs to be maintained.Several cases have also been recorded where,because of the growing awareness and appreciationof environmental issues, allowable abstractions hadto be substantially reduced in favour of larger environmentalreleases.The determination of environmental water requirementsis a specialist field of its own, and is beyondthe scope of this chapter (see Volume I, Chapter 7,and Chapter 3 of the present volume). Methodshave, however, been developed which can be usedby water resource practitioners to obtain an approximationof environmental water requirements forinitial planning purposes (Hughes and Hannart,2003).4.2.8.4 Other effectsBackwater effects produced by the impoundmentin reservoirs, as well as fluctuations in reservoirlevel, such as may be caused by flood flows, windset-up, wave action and periodic undulations ofthe water surface (seiches), may be reflected inshort-term variations in local water balance calculations.However, these effects relate mainly todesign aspects and the safety of the dam structure,as well as the safety of people and developmentsin the immediate proximity of the reservoir basin,and are therefore not considered further in thissection.4.2.9 Remote-sensing estimates ofreservoir capacityThe delineation of surface water bodies and theinventory of surface water supplies, including lakes,ponds and reservoirs, have historically been developedon the basis of maps and photo interpretationtechniques, but digital multi-spectral data haverecently been used as well. These data can besubjected to automated analysis so as to achieverepetitive, rapid results that in many cases also meetaccuracy requirements. In general, the accuracy ofdetecting and measuring water bodies has beenlargely a function of proper identification of waterand sensor spatial resolution. Identification problemsinvolve confusion with areas with similarappearance, such as cloud shadows, dark soils andurban areas. However, aerial photograph interpretationcan be used to minimize these errors and checkthe initial results. Therefore, for extremely accuratework, aerial photographs still provide the best datasources. Satellite data also provide good sources fordetermining morphometric parameters, such aslength, width and surface water area for differentelevations, if the resolution is suitable for thespecific use.From a remote-sensing perspective, water has a relativelylow reflectance, especially in both thenear-infrared and visible portions of the electromagneticspectrum. This will help to separate urbanareas, fields and sometimes cloud shadows, andambiguities produced by variations in atmospherictransmission (Engman and Gurney, 1991). The useof Thematic Mapper data with its 30-m nominalresolution will increase these accuracies.Furthermore, data from the SPOT satellite systemare expected to yield improved accuracy.Remote-sensing, for the most part, can only determinethe surface of the water and cannot measurethe volume directly. Mapping surface water area in

II.4-20GUIDE TO HYDROLOGICAL PRACTICESreservoirs can be used to estimate the volume ofwater in storage. The procedures that are used toestimate lake volumes depend on an empirical relationshipbetween surface area or shoreline lengthand volume. Either an area–volume relationshipmay be developed or topographic features can beused to estimate the water stage in the reservoir andthen relate the stage to water volume. Governmentagencies can use this approach not only to supplementdata they obtain for the reservoirs theymanage, but also to maintain an awareness of reservoirsthey do not control but which may affecttheir own management strategies under extremeconditions such as major flooding. These techniquesclearly demand high spatial resolution dataexcept under extreme flood conditions (Engmanand Gurney, 1991).4.2.10 Climate changeThere is growing evidence that global temperaturesare rising and that the rate of increase may besubstantially higher than has occurred in the past(IPCC, 2001). Some global circulation modelssuggest that this could cause changes in annualprecipitation and increase the variability of climatein certain regions. Scenario analyses for assessingthe potential impacts of climate change on streamflowindicate that in some areas streamflow coulddecrease by as much as 10 per cent by the year 2015(Schulze and Perks, 2000).Such changes in climate would have significantimpacts, not only on the yield characteristics ofwater resource systems, but also on the requirementsfor water to be abstracted from the systems.It is wise, therefore, to anticipate the possibility ofclimate change and perform scenario analyses forareas that could be vulnerable so as to assess thepotential impacts that climate change might bring.Although it is appropriate that the potential impactsof climate change be considered in the long-termplanning of water resources systems, a balanceshould be sought between preparedness and possibleoverreaction to prevent valuable resources frombeing wasted.Sensitivity analyses, focused on how the yield characteristicsof water resource systems, could beaffected by climate change, can be performed byincrementally changing the mean and/or standarddeviation with respect to the synthetic generationof streamflow. Indications of what may be regardedas a realistic extent of such changes can be derivedfrom scenario analyses with the aid of global circulationmodels, but will probably largely remainsubject to personal judgement.4.3 FLOOD MANAGEMENT[HOMS 181, J04, J10, J15, K10,K15, K22, K45]4.3.1 GeneralA flood is a “rise, generally brief, in the water levelsin a stream to a peak from which the water levelreceded at a slower rate” (UNESCO/WMO, 1992).Some floods overflow the normal confines of astream or other body of water and cause floodingover areas which are not normally submerged.Floods, high or low, are part of the naturalhydrological cycle and are generally an outcome ofa complex interaction between natural randomprocesses in the form of precipitation andtemperatures with the basin or watershedcharacteristics. In general, the magnitude of a flooddepends on the following factors:(a) Volume, spatial distribution, intensity andduration of rainfall and snowmelt over thecatchment;(b) Catchment and weather conditions prior to therainfall event;(c) Ground conditions such as land use, topographyand so forth;(d) The capacity of the watercourse to convey therunoff (including that due to ice jams or logjams);(e) Impact of tidal or storm surges.Flood plains offer many advantages for humansettlement and socio-economic developmentbecause of their proximity to rivers that providerich soils, abundant water supplies and a means oftransport. Floods also replenish wetlands, rechargegroundwater, and support fisheries and agriculturalsystems, thereby adding to the attractiveness offlood plains for human settlement and economicactivities. At the same time, flood hazards producethe most adverse impacts on the economy andsafety of people. Floods continue to lead all naturaldisasters in terms of the number of people affectedand resultant economic losses (Munich Re, 2006).The struggle of humankind against this naturalhazard is as old as the history of human settlement.Over recent decades, this struggle has seen a gradualshift from flood control to flood management. Thischapter provides an overview of efforts that can bemade to mitigate the adverse impacts of floodswhile making use of the flood plains.4.3.2 Flood management strategiesFlood control refers to the specific process of providingand operating structures designed to eliminateor minimize the damaging effects of floods by

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-21detaining, constraining or diverting flood flows upto an economically based design limit (ICID, 1996;Framji and Garg, 1978). However, flood managementrefers to the overall process of preventing andmitigating the extent of flooding and reducing theflood risks in a holistic manner. Flood risks, whichcan be defined as the expected losses from floodevents spread over a specified time period, are aconstruct of the following factors:(a) Magnitude of the flood hazard expressed interms of frequency and severity (depths ofinundation and related velocities);(b) Exposure of human activities to flooding;(c) Vulnerability of the elements at risk.Providing absolute protection to flood-prone areasfor all magnitudes of floods is neither possiblefrom a practical point of view nor economicallyviable. Hence, a practical approach to floodmanagement would be to provide a reasonabledegree of protection against flood risks at anacceptable economic cost through a combinationof structural and non-structural measures. Overthe years, flood protection measures have playedan important role in safeguarding both people andsocio-economic development from flooding.However, during the last decade or so, these measureshave been complemented with non-structuralmeasures such as flood forecasting and land-useregulation in response to a perceived need for aparadigm shift from flood control to floodmanagement.There are four major flood management strategiesaimed at reducing flood risks:(a) Modifying flood characteristics;(b) Changing society’s susceptibility to flooddamage;(c) Reducing the loss burden per capita;(d) Bearing the loss.Flood modification methods aim at changing thevolume of runoff, the time taken to attain the peak,the duration of the flood, the extent of the areasusceptible to flooding, the velocity and depth offlood waters and/or the amounts of sediments andpollutants carried by the floods. These methodsinvolve flood protection by means of physicalcontrols such as dams and reservoirs, levees andembankments, channel modification and flowdiversion and catchment treatment.Measures can be taken that reduce the susceptibilityof economic activities to damage, with certainactivities focused on the flood plain. These includeland-use regulation, flood-proofing, flood forecastingand flood warning.Reducing the loss burden consists of actionsdesigned to modify the incidence of the losses percapita, either by spreading them over a largersegment of the community than that which isimmediately affected or spreading them moreevenly over time. This is a strategy for reducing thelosses by means of actions planned to assist theindividuals and the community in the preparatory,survival and recovery phase of floods, such as emergencypreparedness, evacuation, flood fighting,post-flood recovery and insurance programmes(these measures are complementary to thosediscussed in the previous two items).Bearing the loss denotes living with floods. Withthe growing emphasis on considering the wholerange of responses to flood hazards in cost–benefitterms, bearing the loss can often be considered themost acceptable solution.The development of policies, strategies and plans tocombat the risks associated with flooding or anynatural hazard should be based on a comprehensiveassessment of the risks involved. This requires anintegrated approach whereby a wide range of floodmanagement measures should be considered. It isnecessary to look at the overall situation, comparethe available options and select a strategy that ismost appropriate to a particular situation. Whilerecognizing the pros and cons of various structuraland non-structural measures, a good combinationof both types of measures needs to be evaluated,adopted and implemented. For example, a levee inone part of town may be positively supplementedby land-use adjustments in an unprotected floodwayarea and by structural adjustments in a sparselybuilt-up sector, or flood control by using reservoirsmay be combined with land-use regulations.4.3.3 Integrated flood managementTraditionally, flood management has focused ondefensive practices. It is widely recognized, however,that there is need for a shift from defensive actionto the proactive management of risks. Integratedflood management, designed to integrate land andwater resources development in a river basin withinthe context of integrated water resources management,seeks to manage floods in such a way as tomaximize the net benefits from flood plains whileminimizing the loss of life from flooding (WMO,APFM, 2004). Thus, occasional flood losses can beaccepted in favour of a long-term increase in theefficient use of flood-prone areas. There are fiveobjectives in integrated flood management:(a) Manage the water cycle in so far as it relates toland, as a whole;

II.4-22GUIDE TO HYDROLOGICAL PRACTICES(b) Integrate land and water management;(c) Adopt the best mix of strategies;(d) Ensure a participatory approach;(e) Adopt integrated hazard managementapproaches.For a detailed discussion of these objectives, pleaserefer to WMO (APFM), 2004.For flood management to be carried out within thecontext of integrated water resources management,river basins should be considered integratedsystems. The multi-faceted measures include, forexample, socio-economic activities, land-usepatterns, hydromorphological processes, publicawareness, education, communication andstakeholder involvement with transparent decisionmaking.These need to be recognized as constituentparts of these systems that are duly embedded innon-structural measures.Flood management is an interdisciplinary pursuitinvolving different sectors of the economy and thevarious departments and institutions which havean impact on the magnitude of floods, as well asthe implementation of flood management measures.For this, the linkages between various relevantsectors become very important, and the mostimportant key is cooperation and coordinationacross institutional boundaries. While themandates of many institutions may cover onlypart of the river basin or one sector, others extendwell beyond the basin boundary. Effective communicationacross institutional and disciplinaryboundaries is at the core of integration, which cantake place only if there is a clear understanding ofcommon goals. Emphasis, therefore, should beplaced on the adoption of flexible strategies suitedto each flood-prone region, which is characterizedby various physical, social, cultural and economicsituations. Further, it is important to evaluate thedifferent options and their relative advantages anddisadvantages.Loss of life can be avoided if reasonably accurateand reliable forecasts are provided to flood-plainoccupants in a timely manner. However, this has tobe supported with adequate preparedness measuresand response mechanisms designed to vacate peoplefrom the threatened zones. Flood hazard maps, alsoreferred to as flood maps, flood risk maps or floodplainzoning maps, show areas likely to be floodedwith a given probability and provide a long-termadvance warning that serves as a basis for helpingpeople to make their own decisions as to whetherand where to live and invest in the flood plain(WMO, 2006a). These tools play an important rolein building awareness among various stakeholdersof the risks of flooding and help them organizeflood response activities. Flood-plain zoning, whichcan be one further step, can be of great value, but italso has limitations because of the difficulties ofenforcing the related rules and regulations – particularlyin developing economies with populationpressure.4.3.4 Structural measures4.3.4.1 Design floodsA design flood is defined as the flood hydrographor the instantaneous peak discharge adopted forthe design of a hydraulic structure or river controlafter accounting for political, social, economic andhydrological factors. It is the maximum floodagainst which the project is protected and its selectioninvolves the selection of safety criteria andthe estimation of the flood magnitude that meetsthese criteria. This subject is covered in detail in5.10.Although a flood control structure is installed tocontrol future floods, its design is generally basedon analyses of past floods. Such an extrapolation ofthe past hydrological series for the future may notalways be appropriate, however, owing to thechange in meteorological events or in the changein the hydrological response of the basin.Anthropogenic influences due to the growth ofpopulation and higher standards of living mayresult in intensified land development. The extensionand intensification of urbanization oftencontribute to increased volumes and peak flows ofsurface runoff and sediment transport. Deforestationmay result in an increase in sediment yield anddestabilization of river morphology. Forests may betransformed into agricultural lands and the drainageof agricultural lands improved. An assessmentof the hydrological effects of upstream changes canbe made by using deterministic hydrological models(see Chapter 6) to evaluate their impact on the riskof downstream floods.A number of studies on the potential impacts ofclimate change on flooding have been carried outas part of the work of the Intergovernmental Panelon Climate Change (IPCC, 2007). At this time, it isnot possible to predict potential increases in floodpeaks due to climate change for specific basins withthe degree of certainty necessary for their incorporationinto the planning and design process.However, adaptive management techniques, suchas revision of criteria for determining the freeboardon levees and other works or judiciously modified

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-23operating procedures for control structures, holdthe promise of accommodating the potentialincrease of extremes caused by climate change.4.3.4.2 Flood retention reservoirsThe over-bank spilling of a river resulting in floodingdoes not generally occur for long periods, evenduring the flood season. High-magnitude floods arecaused by severe storms associated with extrememeteorological systems such as cyclones and intensemonsoons, and subside within a reasonable period.Depending on the catchment characteristics andstorm track, flood discharge fluctuation due to heavyprecipitation, followed by a relatively drier spell,can be used to advantage to moderate the floodthrough a variety of storages during high discharge.Storage is generally provided behind a dam on anupstream reach of the river, but distributed storagebasins on the flood plains can also be used.4.3.4.2.1 Flood storage capacity of reservoirsThe volume of storage that must be provided toretain flood waters in a reservoir depends on thefollowing factors:(a) Volume, peak flow, duration and other characteristicsof the upstream flood that is to bemoderated;(b) Storage requirements to meet various otherwater demands;(c) Carrying capacity of the downstream channel;(d) Extent of flood moderation required.If floods are highly variable over the year and occuronly during a certain season, the reservoir may haveto play a multi-purpose role to meet various waterdemands in addition to flood moderation. If so,reservoir capacity will be fixed mainly on the basisof other water demands, with only a certain storagereserved specifically for flood moderation duringthe flood season. In such cases reservoir levels aredrawn down before the flood season begins and arerefilled as the season passes. Flood storage can beprovided either in on-stream or off-stream reservoirs.If such dedicated storage cannot be provided,some flood abatement can be achieved through theuse of carefully designed reservoir operationschedules.The multiple uses of reservoirs are also consideredunder 4.2.6.4.3.4.2.2 Design considerationsFlood abatement is achieved by detaining and laterreleasing a portion of the peak flood flow. Theamount of storage required, or detention storage, isgenerally specified as that part of the reservoir storagethat can produce a given reduction in the floodpeak of a given magnitude or of a given probabilityof occurrence. The following basic types of storagecan be distinguished:(a) Regulated storage, either in an on-stream or anoff-stream reservoir;(b) Unregulated storage in an on-stream reservoir;(c) Unregulated storage in an off-stream reservoir.The storage capacity needed to achieve a giveneffect will depend on the type of storage used. Theflood-transformation effects of each type of storageaiming for the same flood-peak reduction are shownin Figure II.4.10 and are discussed in the followingsubsections. In practice, the effect of a flood controlreservoir is generally a combination of regulatedand unregulated storage.4.3.4.2.3 Regulated detention storageFull control over the flood detention storage of areservoir provides the highest efficiency of floodmitigation because water storage can only beginafter the highest permissible flow, also known asthe non-damaging flow, has been reached downstreamfrom the reservoir. Therefore, only thatportion of floodwater that is apt to cause damage isstored.Control over storage is achieved by the regulationof gated outlets in the case of an on-stream reservoirand of gated intakes and outlets in the case of anoff-stream reservoir. In an on-stream reservoir, fullcontrol is achieved only if the outlet has a sufficientcapacity to release the non-damaging flow whenreservoir storage is at its minimum, and if the releaseof water from the detention storage can be fullyregulated. In an off-stream reservoir, full control isachieved only if the intake has a sufficient capacityto prevent the rise of flow in the downstreamsection of the river above the non-damaging flow,Streamflow m 3 s –1Q N0Uncontrolled flowControlled flow(a) (b) (c)TimeStorage required to reduce flood peak to flow Q NFigure II.4.10. Effects of reservoirs on floods –regulated storage (a), unregulated on-streamstorage (b) and unregulated off-stream storage (c)

II.4-24GUIDE TO HYDROLOGICAL PRACTICESand if the release of the detained water can also beregulated.The design flood for the determination of the flooddetention storage capacity of a reservoir need notto be the same as the one used for the design of itsspillway because the dam’s safety requirementsgenerally differ from flood protection requirementsdownstream from the reservoir.4.3.4.2.4 Unregulated on-stream detentionstorageThe storage above a fixed spillway crest of an onstreamreservoir is generally regarded as unregulatedfor design purposes, even though it may be partiallyregulated by release through gated outlets andturbines. However, for the design of unregulateddetention storage, these releases are eitherconsidered constant during the passage of a flood,or the outlets are considered closed. The formercondition is generally adopted for the assessmentof the normal downstream flood control effects ofthe reservoir, while the latter condition is appliedfor the assessment of the dam safety.Unregulated detention storage plays an importantrole in the safety of a dam against overtopping. Itsdesign is interlinked with the design of the damspillway and must be based on the same designflood as the spillway itself. Safety considerations inspillway design require that the reservoir be regardedas filled up to the spillway crest at the beginning ofthe design flood.As the comparison of parts (a) and (b) ofFigure II.4.10 indicates, unregulated storage is lessefficient in flood-peak reduction than regulatedstorage. This is because unregulated storage beginsfilling even before it is needed.4.3.4.2.5 Unregulated off-stream detentionstorageUnregulated off-stream detention storage arises inoff-stream reservoirs, sometimes called poldersbecause of their resemblance to real polders. Theyare constructed by enclosing a part of a flood plainwithin a dyke whose crest at the upstream end islowered to form a sill, thus providing an intake tothe enclosure. When the river stage at the upstreamend rises above the sill crest, the polder starts fillingby overflow over the sill. The fact that the riverbypasses the reservoir makes the unregulated offstreamstorage more efficient than unregulatedon-stream storage because unnecessary filling startslater. (See parts (b) and (c) of Figure II.4.10.)4.3.4.2.6 Operation considerations for designpurposesFlood detention storage is frequently provided inmulti-purpose on-stream reservoirs with gatedoutlets with a capacity sufficient to provide a highdegree of control of the storage. These reservoirsalways have some incidental ungated flood detentionstorage and, in many cases, part of the gatedstorage is reserved for flood detention. In addition,storage designated for other uses may occasionallybe used to moderate floods. Although this diversityoffers flexibility, it makes the flood reductionstrongly dependent on the mode of reservoir operation.Therefore, in such cases, it is necessary toanalyse many different operation modes duringthe early stages of design because the results affectthe selection of the design variables for theproject.Controlled storage reservoirs are operated by regulationschedules, or set of rule curves, which aim atimpounding part of the inflowing flood above aspecified safe amount during the rising stage. Whenthe storage capacity is fully used, and depending onthe inflow, the outflow is increased to ensure thatthe design reservoir levels are not exceeded.Thereafter, the impounded flood water is releasedin such a manner as to empty the reservoir within areasonable time so that it is available for receivingthe next flood, all the while keeping the rate ofrelease as far as possible within the safe limits forthe lower reaches. Such an operation should beaccompanied by flood warnings to the downstreamcommunities.In multi-purpose reservoirs, interests such as irrigationhydropower generation and flood controlgenerally compete with one another even when thereservoir is owned by a single State or agency. Thisconflict may be heightened when more than onecountry is involved. Irrigation and hydropowerneeds generally dictate the filling up of reservoirs assoon as possible and their retention at as high alevel as possible. To achieve flood moderation, thereservoir levels should be kept as low as possibleand the reservoir depleted as soon as possible aftera flood so as to be of use for flood absorption duringthe next flood event. A reservoir is more effectivefor flood moderation if, apart from the incidentalstorage available in reservoir on a river, storagespace is allocated for flood detention and is notencroached upon. Multiple uses of reservoir storagetherefore imply a compromise, which inevitablyresults in less than the maximum possible benefitsfor any one single user but which realize the maximumbenefit for the project as a whole. (See 4.2.6.)

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-25The availability of accurate and timely inflow forecastsis essential to optimize the overall benefitsderived from a reservoir. In particular, it will beimportant to forecast the arrival of two or morefloods in close succession, with the possibility thatthe detention storage filled by one flood may not beemptied before the arrival of the next. If such forecastsare missing or inadequate, the effectiveness ofmulti-purpose operation can be hampered and theability to moderate the flood will be reduced.Continued and effective moderation of high floodsby reservoirs, over many years, has a tendency tobring a false sense of security among those who liveand work in the downstream reaches. This may leadto encroachments on the flood plains and developmenton the riverbanks and in turn, a reduction inthe conveyance of the channel downstream. Thiscan result in serious damage when a large flood hasto be let down from the reservoir. Flushing doses,not exceeding the discharge against which protectionis given in the downstream valley, should bereleased in order to keep the river channel alive anddeter encroachment in the flood plains. All suchreleases should be accompanied by an advancedflood warning.It is therefore essential to formulate and use specificrule curves or preset procedures for the operation ofmultiple-purpose reservoirs in order to maximizethe benefits from the project while ensuring thesafety of the structures and downstreamcommunities.4.3.4.2.7 Sedimentation effectsThe deposition of sediment in a reservoir reduces areservoir’s storage capacity and performance.Reservoir design should allow for allocation of partof its storage capacity for sediment deposition toprevent premature reduction of its active storagecapacity. However, designated storage can be inadequate,affecting flood detention storage significantly.The upstream part of the reservoir may be affectedfirst by the sedimentation process. Thus, a reservoir’sflood moderation efficiency may decreasewith time. This should be considered in long-termflood protection planning so that timely alternativescan be developed and appropriate levels ofprotection provided for the system. (See 4.2.2.2.)4.3.4.3 Other structural measures4.3.4.3.1 Bypass and diversion channelsDiversion of river water may be employed to keepthe downstream discharges within the conveyanceof a river system. Flow may be diverted all or in partinto a natural or artificially constructed channellying within, or in some cases outside, the river’sflood plains. The diversion may move water fromone river to another, to a depression or to the sea,or it may be returned to the same river channelsome distance downstream. Diversion of floodwaterfrom one river to another involves the followinghydrological considerations:(a) Determination of a design-flood hydrographfor both rivers;(b) Separation of the part of the flood hydrographto be diverted;(b) Flood routing of the diverted flow through thediversion channel;(c) Combination of the diverted flow with floodswhich may occur in the receiving river;(d) Estimation of the revised flood frequenciesat the downstream segments of the riversconcerned.Care should be taken to evaluate the phasing of thesuperimposed floods in the receiving river as well asthe backwater effect that may cause an increase inthe flood risk in the reach upstream from the diversion’sdischarge point in the receiving river.4.3.4.3.2 Drainage improvement and channelmodificationCongestion of surface water drainage due to inadequacyof natural or artificial drainage systems resultsin flooding in areas with moderate ground slopes.In such cases, effective flood management can beachieved by increasing the capacity of the existingdrainage channel or by constructing supplementarychannels for accelerated evacuation offloodwater. Similarly, channel modifications generallyaim at boosting channel conveyance capacityby deepening and widening the channel, cuttingmeanders, shortening channel length, clearingvegetation and possibly lining a channel to reduceits resistance to flow. This results in increased flowvelocity and lower water levels with flood reductionalong the modified reach.It is important to note, however, that channelmodification and drainage improvement causeincreases in flood peaks downstream. The effects ofsuch works can best be assessed by hydraulic routingmethods (6.3.6) with proper consideration of theinteraction between the floods in the main channeland in the tributaries downstream. The possibilityof an increase in magnitude and duration offlooding in the downstream area should beconsidered when planning such schemes throughhydraulic modelling of the entire drainage system

II.4-26GUIDE TO HYDROLOGICAL PRACTICES(6.3.6). An incidental effect of channel modificationmay be increased scouring in the modified reachand sediment deposition downstream.Opposite effects can be achieved by reducing thechannel capacity by various river-training structureswhich, by slowing the flow, can cause increasedflooding upstream and flood reductiondownstream.4.3.4.3.3 Levees and floodwallsThe oldest, most common and quickly constructedmeans of flood protection, which is often economical,is a system of levees, also called embankmentsor dykes. Levees are constructed either on riverbanksin a general direction parallel to the flow ofthe river or surrounding riparian areas so that theycan serve as artificial high riverbanks during highfloods and prevent flooding. Levees are constructedmainly from earth and must be resistant to hydrostaticpressure from floods, erosion, piping failureand seepage. Resistance can be achieved by buildinglevees with a broad base. As a result, evenmoderately high levees occupy a large base area,and in terms of land costs, can be prohibitivelycostly in urban and industrial locations. In developedareas, where adequate space is not available orland is too expensive for an earthen embankment,concrete or masonry floodwalls may be a moreeconomical, socially acceptable option. River-trainingworks such as spurs, studs and revetments aresometimes necessary in combination with thelevees to protect them from flooding. To achieveproper levee design, attention should be paid to thefollowing factors:(a) Levee alignment;(b) Design flood levels;(c) Design freeboards;(d) Structural design of levees;(e) Drainage sluice location and design.The height of a levee system is determined in sucha manner as to provide the area concerned with acertain degree of protection defined according tothe economic value of the protected area and tolocal or national decisions as to what is regarded asacceptable risk. For further information on risk, see4.2.5.2 and 5.10.8. The design is generally set interms of protection from a design flood with acertain probability of occurrence within specificperiods, for example, the 1-, 10-, 25-, 50-, 100- or1 000-year flood. Design water levels should becalculated on the basis of the hydraulic conditionsin the entire basin. On rivers where human activitiesinfluence the water regime (upstream reservoirs,levees or barrages), their effect should be consideredand, on rivers subject to frequent ice jams or landslides,water levels should be calculated accordingto the highest backwater levels caused by downstreamjams. Construction of high levees tends tobe unattractive in view of the cost considerations.Another consideration is the potential damage thatresults when levees are overtopped. Design waterlevels in ice-prone rivers should be calculated onthe basis of ice-free observations if the flow regimeis natural.Freeboards above the design flood level are addedto ensure that design floods do not overtop thelevee; uncertainties in design flood calculations,including those due to likely climate change, areaccounted for; seepage does not cause significantflow within the body of the levee to cause pipingand waves do not spill over the crest of the levee.Depending on wave conditions and the slope of thelevee on the waterside, the freeboard shouldnormally be in the range of one to two metres.Freeboards can be provided by building floodwallson the crest of the levee to reduce costs. The loadingof the levees, not only in terms of force, but alsoin relation to their susceptibility to seepage, dependson the duration of the floods. Thus a statisticalstudy of the duration of certain water levels mayhelp to design and construct seepage-resistantembankments. Drainage sluices, service roads onthe crest or on the toe, fuse plugs and toe drains areexamples of important components to be consideredas part of the design of the levees.The alignment of the levees and the width of theunprotected flood plains is governed by and influencesthe upstream and downstream hydraulicconveyance conditions of the channel. The locationof the flood levees should consider the effect ofthe spacing between the embankments on the newwater levels upstream due to the loss of valley storage,otherwise available for flood moderation. Veryclose embankment spacing may cause an unacceptablerise in water levels in the upstream sectionsand abnormal sand deposition in the upstream ordownstream reach. The loss of valley storage can bekept to a minimum if the flood plain on one sidecan be kept at a lower level or may be left unprotected,depending on the situation. Such asolution is possible only if one of the flood plainson one side have a lower economic value than thoseon the other side.The risk of levee breaches cannot be eliminatedcompletely. Fuse plugs should, therefore, beprovided in long levees to save the protected areaswith high economic values, at the cost of floodingless economically important areas such as farmland.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-27The area that would be inundated by spillagethrough breaches should be identified on the basisof previous experience, supplemented by hydraulicstudies as necessary. Emergency plans should bedevised and warnings should be issued prior to andduring major events when such areas are likely tobecome inundated. Emergency planning for potentialbreaches of embankments forms a vitalcomponent of the integrated flood managementapproach.4.3.5 Non-structural measures4.3.5.1 GeneralStructural measures alone cannot completelyeliminate flood risks. They may even result ingenerating a false sense of security leading toinappropriate land use in the areas that aredirectly protected and often in adjacent areas. Toreduce flood risk, the vulnerability of economicactivities to adverse impacts of flooding alsoneeds to be addressed.Non-structural measures broadly reduce vulnerabilityto flooding. They may constitute planningmeasures and response measures. Flood-plainmapping, land-use planning and regulation, floodforecasting, flood-proofing and insurance are planningand preparednes measures that are to beimplemented prior to the onset of floods. Responsemeasures are actions to be taken during and afterthe flooding; these include fighting floods, emergencyevacuation and economic recoveryassistance.4.3.5.2 Land-use planning and catchmentmanagementLand-use planning aims at reducing the risk causedby flooding by addressing the magnitude, exposureand vulnerability of people and their economicactivities. Catchment management consists ofactions that affect the hydrological process andaim to modify the way or rate in which rainfall istransformed into streamflow, especially floods.Catchment management measures include theintroduction of vegetation and crops that protectthe soil, the prohibition of cultivation and grazingon steep slopes, reforestation, better forest managementand control of shifting cultivation inconjunction with minor engineering works suchas check dams, trenches and contour bunds.Catchment management measures can have asignificant impact on small floods and small catchments,but they are much less effective on largercatchments. An important contribution of watershedmanagement is the reduction in silt loading inrivers of aggrading nature. Urbanization caused byland-use change has a significant impact on themagnitude of floods, reducing the time of concentration,and increasing flood peaks, particularly incatchments up to 100 km². Regulating land usethrough building by-laws can help control urbanizationso that it does not seriously affect thehydrological response characteristics of the catchmentsconcerned.4.3.5.3 Flood-plain regulationThe flood plain is an integral part of the riversystem which allows the passage of flood flows.When the flood plain is not occupied by water, itforms part of the land system offering possibilitiesfor various economic activities. Integrated floodmanagement should implement patterns of landuse which, while taking advantage of the benefitsoffered by flood plains, reduce to a minimum thedamage suffered during the inevitable periods offlooding.Overdevelopment of the flood plain is the maincause of increasing loss of life and flood damage.Therefore, the most desirable approach is to assessthe risks due to flooding, identify them for theinformation of all stakeholders and, whererequired, regulate and even prohibit new developmentin the flood plains by land-use planning andrelated regulatory measures. However, those developmentsthat are permitted must carry outflood-proof measures for existing and new structuresand sometimes attempt to relocate theexisting development elsewhere. Where the extentof present development is substantial, or the floodplain is essential for food production or other keyeconomic activities, alternate strategies such asflood-proofing and protection can be considered.Redevelopment and resurrection of an area badlyaffected by floods can involve permanent alterationof the uses of the land as the only economicallyfeasible alternative, such as resettlement in lesshazard-prone areas.Accordingly, the flood plain may be mapped toshow the extent of likely flooding due to floods ofdifferent return periods, (for example 1 in 10, 25,50 and 100 years) by hydraulic routing of designfloods of different frequencies through the floodplain and determining the corresponding floodlevels, discharges and areas inundated. The resultscan be drawn onto topographic maps at a scale of1:20 000 or 1:10 000 or even larger, with contourintervals, depending on the topography.

II.4-28GUIDE TO HYDROLOGICAL PRACTICESThe unique capabilities of satellites to providecomprehensive coverage of large areas at regulartime intervals with quick turn-around times havebeen valuable in monitoring and mapping pastflood events and, therefore, providing informationon the flood dynamics for major rivers. Floodedareas, extending to several thousands of squarekilometres, can be mapped effectively using satellitedata. Multi-temporal satellite data have beenused with digital elevation models to identify floodinundation areas, even including flooding undervegetative canopies and, used in conjunction withgeographical information systems and terrainmodelling, help to identify sections of the inundatedflood plains, together with information suchas the related water quality.With the evolution of flood-plain mapping andzoning, appropriate legal and administrative protocolsshould be developed, including flood-plainregulation and zoning based on by-laws, subdivisionregulations, building codes and landdevelopment policies (WMO, 2006b).4.3.5.4 Flood forecasting and warningFlood forecasting enables society to ascertain thefuture states of hydrological phenomena, especiallyas to when the river might inundate its flood plain,to what extent and for how long. Flood forecastsformulated and issued sufficiently in advance allowauthorities to respond well in advance by, for example,operating dams; opening or closing gates;making anticipatory releases to enlarge storagecapacity; issuing preventive instructions, such asbans on navigation and fishing; invoking emergencymeasures, such as announcing generalizedalerts; mobilizing evacuation of and assistance tothe population in high-risk areas or orderingplanned breaches of flood dykes. To do so, it isessential to develop and operate flood forecastingand warning systems (see Chapter 7), which wouldindicate, with sufficient lead time, the expectedextent and duration of flooding.Flood forecasting involves continuous system monitoringand operation, regardless of the frequency ofuse. If it is to be economical, such a system should,wherever applicable, implement a multi-hazardapproach, thereby combining the flood-warningsystem with other activities, such routine dailyweather forecasting, regular hydrometric measurementsand navigational traffic control.Of utmost importance after the formulation ofthe forecast, is its dissemination to the users oraudience concerned as a warning transmitted bytelephone, facsimile, radio/wireless/TV bulletins,telegrammes, electronic mail and other mediasystems (see Chapter 7) for which a robust communicationsystem should be used and wellmaintained.4.3.5.5 Flood insurance and other economicinstrumentsThe principal objective of flood insurance is tospread the economic costs of flood damages so thatthey become more manageable for society.Insurance, unless tied to premium increases onexposure to risk, may not result in reducing theoverall losses to society. Flood insurance differsfrom the other tools for managing flood losses inthat, whereas other tools are geared to reducing thecost of flood damage from each flood, insurancedistributes the losses over time and space. It placesthe burden on those who enjoy the benefit of floodplainlocation rather than making the burden thesole responsibility of the government.Risk perception studies carried out in the UnitedStates show that without a mandatory componentto an insurance system, people tend to perceiveflooding as a low risk and therefore do not buycoverage. It is important that policies be consistent,as some people may not purchase insurance ifhistory has shown that the government providesrelief to all, regardless of insurance coverage. Inaddition, a purely voluntary insurance scheme maynot yield sufficient funds to cover future compensationclaims. However, it has proved difficult to makeinsurance mandatory – and therefore effective –unless it is preceeded by a major educationalcampaign. Insurance rates may be tied to risk, withoccupants being potentially able to reduce their riskof exposure, for example by flood-proofing theirproperty. Insurance is an option that should beconsidered but, for the time being, it is probably nota feasible alternative in many developing countries.Flood insurance is available in a few countries withwell-established insurance markets, such asGermany, Japan, the United Kingdom of GreatBritain and Northern Ireland, and the United States.There is considerable diversity in the way in whichflood insurance is provided, as well as in the methodsused to determine premiums. For insuranceschemes to be successful, there needs to be a cleardefinition of the risk, as premiums should reflectthe degree of risk at a given location in the floodplain established on the basis of flood frequencyand hydraulic modelling. If possible, flood insuranceshould be considered complementary to aflood-plain zoning programme. There is no single

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-29model of an optimal flood insurance programmefor all countries.4.3.5.6 Flood-proofingFlood-proofing is defined as follows:A combination of structural changes and/or adjustmentsincorporated into the design and/orconstruction and alteration of individual buildings,structures or properties subject to flooding primarily,for the reduction or elimination of flooddamages (USACE, 1995).An example of a specific action designed to floodproof a structure is the installation of barriers acrossall openings at ground level to prevent seepage ofwater and the entry of debris into the main structure.Such devices can be permanent or temporaryin design, with the latter being installed precedingthe onset of a forecast flood (Szöllösi-Nagy andZevenbergen, 2005). Flood-proofing can also beachieved by locating structures above the level ofthe design flood. Such structures could includehuman dwellings, animal shelters and public buildings,including temporary emergency shelters.4.3.6 Flood emergency managementNo matter what strategies are adopted to reduceflood risk, there will always be some residual risk.Whatever strategies are used to reduce risk fromflooding, whether through structural measures andflood embankments or non-structural measuressuch as reforestation, only partial safety can bepromised to those who inhabit the flood plain.When protection fails, damage can be more extensivebecause of the increased investments made inthe flood plain. For many societies throughout theworld, the cost of reducing risk by adopting highcoststructural measures or policies aimed atrelocating at-risk land use is simply unaffordable. Itis also possible that such measures may causedamage to the environment or run counter to theparticular development goals. An alternative strategyto be considered, even when structural measuresare in place, is to reduce vulnerability throughdisaster preparedness and flood emergency response.When flooding is inevitable, it is important to takemeasures that reduce the adverse impacts of such asituation on the lives of people affected. Floodemergency management is aimed at managing andminimizing the damaging effects of flooding. Theobjective is to prevent loss of human life and avoidthe exposure of critical activities by temporarilyshifting people and such activities away fromflood-prone areas, thereby reducing the negativeimpacts of flooding on the community. Flood emergencymanagement can be divided into threestages:(a) Preparedness: pre-flood measures to ensureeffective response;(b) Response: measures taken during the floodingto reduce adverse impacts;(c) Recovery: measures to help the affected communityrecover and rebuild after the event.Emergency management requires cooperationacross sectors and administrative levels. In additionto mobilization resources, it is vital to maintain acontinuous, timely and precise flow of informationflow in support of those handling the emergencysituation. Emergency response planning must becompleted well before the flood season and must bebased on clear technical and financial plansdesigned to match scenarios of flood hazards whichmay occur. These emergency management plansshould be the subject of regular review and revision.Lessons learned each flood year need to beincorporated into future plans. Important elementsof these plans include the following:(a) Assessment of flood risk and factors thatcontribute to losses caused by flooding;(b) Zoning of protected or unprotected areasaccording to flood risk;(c) Inventory of flood control or protectionsystems;(d) Analyses of technical means to counteractfailure of flood protection structures duringfloods;(e) Study of situations which might develop whensome elements in the flood protection systemfail;(f) Planning of second, third and subsequentdefence lines in the event of progressive failureof linear protection systems such as levees;(g) Estimate of costs of fighting floods in differentsituations;(h) Development of evacuation routes and plans,emergency shelter facilities and provision,medical facilities, and so forth.Key components of a flood-emergency responseplan include an early warning system, protection ofcritical infrastructures, assessment of immediateneeds and provision of safe shelters for the effectedpopulation, with adequate facilities for all ages andboth men and women.4.3.6.1 Emergency preparedness andresponseThe most critical element in flood damage reductionis emergency preparedness and response. As

II.4-30GUIDE TO HYDROLOGICAL PRACTICESoutlined in the previous section, detailed responseplans need to be prepared in advance and reviewedby the coordinating unit with all key agencies andstakeholders, with specific duties being assignedto each so that there will be no confusion underpressure. A coordination mechanism must beincluded in the plan, including provision forresponse committees, meeting venues and sourcesof information. Often this takes the form of anincident management centre where material,support staff and information such as maps andplans are available. The awareness of the affectedcommunity should be raised and maintained,with a thorough understanding of how to respondappropriately. This will be critical in achievingquick response in situations such as coordinatedevacuation from the affected area when disasterstrikes. Information on evacuation routes andemergency shelters should be available to all wellin advance. Emergency response teams shouldreceive training early on and their skills upgradedconstantly with mock emergency exercises carriedout on a regular basis.A key component of any emergency preparednessplan is an inventory of resources that can beaccessed. In the case of flooding, this could includeitems such as vehicles, buses, trucks, earth-movingequipment, pumps, covering and protecting materials,generators, construction materials and mobilecommunication equipment. Basic responsibility fordeveloping and implementing such plans generallylies with the administrative authority of the affectedarea. The same authority must also decide whenand how to evacuate the target population, ifnecessary.Action taken during floods to prevent damage aswell as divert floods from sensitive areas is generallyknown as flood fighting. This is an emergencymeasure aimed at mitigating flood impacts on societyand the environment. Flood fighting includesbuilding temporary levees with any material that isavailable, closing breaches with sand bags, movinggoods and equipment out of reach of the floodwaters,protecting immovable equipment with plasticsheets or grease, and so forth. When floods occur,water supply and sewerage are often disrupted withpotentially serious effects for the health of thepopulation. Therefore, flood fighting includeselements of infrastructure maintenance that arerelated to pubic health.4.3.6.2 Post-flood recoveryAfter the floodwaters have receded, those affectedwill require assistance to restore pre-existingconditions as soon and as far as possible. Examplesof the measures to be taken include the restorationof road and rail links, and the rehabilitation ofpower installations, public buildings, water supplyand sewerage systems merchandise and shoppingareas, industries, factories, poultries, fisheries,piggeries, tube wells and agricultural machinery,irrigation and drainage systems and structures.Action is required to pump water out of low-lyingareas and remove overlying sand and silt that willhave been deposited on flooded areas. On thewhole, efforts are required to provide a post-floodeconomic stimulus to flood-affected areas.The relevant administrative agency will provideflood-disaster relief in the form of financial andother aid to relieve the distress of flood victims. Atthe international level, the United Nations DisasterRelief Coordinator has funds to assist victims ofdisastrous floods and other natural hazards. In somecountries, permanent funds have been establishedfor this purpose and relief may take the form ofgrants, interest-free or low-interest loans and subsidies.Relief may extend to measures such as thesupply of free seeds and other agricultural inputs tofarmers. Often aid for flood victims is provided onan ad hoc basis by the government or voluntaryorganizations such as the local Red Cross or RedCrescent Society at the national level, and by theInternational Federation of Red Cross and RedCrescent Societies and the Office of the UnitedNations Disaster Relief Coordinator at the internationallevel. Some governments declare a tax holidayfor those affected, thereby further reducing theburden on them.After a major flood, it is very important and urgentto make an assessment of the causes and effects ofthe disaster and of the performance of emergencyactions, followed by recommendations that wouldimprove preparedness and reduce flood losses forthe next event. One thing is certain: there willalways be another flood event at some time in thefuture.4.4 IRRIGATION AND DRAINAGE[HOMS K70]4.4.1 IrrigationThe practice of irrigation as a means of producingfood has been around for over 5 000 years (Framji,1987). During the second half of the twentiethcentury, the total irrigated area in the worldincreased from about 115 million hectares to over

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-31270 million hectares. This has led to a more thantwofold increase in the world’s total food grainproduction (cereals, oilseeds and pulses), from1 763 million tonnes to 3 891 million tonnes.Irrigated land accounts for 20 per cent of the totalcrop area in the world and over 40 per cent of totalfood production. Irrigation, combined with the useof high-yielding crop varieties, which can be grownonly under irrigated conditions, has indeed been acrucial element in many countries’ struggles toachieve and maintain self-sufficiency in food grainproduction.Irrigation is the largest user of water, taking morethan 70 per cent of the world’s fresh water supply.Although irrigation is not a new practice, most irrigationsystems are operated inefficiently, withefficiency seldom exceeding 40 per cent. Historyabounds with examples of civilizations that owetheir success to well-planned and well-managedirrigation systems and those which met their downfallbecause of improper and inefficient managementof irrigation systems. Efficient management of irrigationsystems centres around maintaining anappropriate soil moisture regime in the plants’ rootzone to promote healthy plant growth. This requiresthe timely supply of adequate amounts of waterand removal of excess water from the root zone.Therefore, both irrigation and drainage are necessaryfor the proper management of water foragriculture.4.4.1.1 Why plants need irrigationWater is essential for plants in many ways:(a) Nearly 70 per cent of a plant is comprised ofwater;(b) Initially, water is required to soften the seedand its covering to facilitate emergence first ofthe root and then of the seedling above thesoil;(c) Water functions as a solvent and dissolves andtransmits through the plant roots nutrientssuch as nitrogen, phosphate and potassium,which are essential for healthy plant growth;(d) Water is the solvent for biochemical reactionsin plants, such as carbon fixation and photosynthesis;(e) The carbon, nitrogen, hydrogen and oxygenrequired for plant growth are derived fromwater and atmospheric air and make up mostof the body of the plant;(f) Roughly 95 per cent of the water absorbed bythe plants is transpired from the leaves and thestems. This process also helps to cool the plantsduring hot weather;(g) Without water, plants wilt and ultimately die.The soil is capable of retaining moisture with theforces of adsorption and surface tension. Any additionalmoisture that enters the soil medium, beyondthat held by these two forces, moves down throughthe soil pores under the influence of gravitationalforce. This process is known as percolation. A measureof the tenacity with which water is retained inthe soil, indicating the force required to extractwater from the soil, is referred to as the soil moisturetension. The amount of moisture in the soilmedium is referred to as the soil moisture content.The soil moisture content at which plants can nolonger extract water from the soil to meet theirevapotranspiration requirements is known as thewilting point. When the soil moisture falls to thislevel, plants wilt and die unless water is replenishedin the root zone. The amount of soil water availablebetween the moisture content at the field capacityand at the wilting point is referred to as the availablemoisture capacity.The purpose of crop irrigation is to ensure that anadequate water supply in the root zone at all times,in the range between field capacity and wiltingpoint. Soil moisture is affected by rainfall, irrigation,evapotranspiration, runoff, infiltration anddeep percolation. When all the interstices in thesoil are completely filled with water, the soil is saidto be at its saturation capacity. In this state, waterwill drain out of the soil root zone under the influenceof gravity until an equilibrium is reached. Thesoil is then said to be at field capacity. This stage isgenerally reached within one to three days afterirrigation or rainfall. Efficient irrigation returns rootzone to field capacity. Water applied in excess ofthis amount is considered wastage unless deliberatelydone for leaching purposes.4.4.1.2 Crop water requirementsCrop water requirement is defined as the depth ofwater needed to meet the water loss throughevapotranspiration of a disease-free crop growing inlarge fields under conditions which impose no soil,soil water and fertility conditions, thus achievingfull production potential under the given growingenvironment (Doorenboes and Pruit, 1977).The water requirement is crop- and location-specific,and is influenced by crop species, local climate andthe soil. It is estimated for a specified period of time,for example a week, month or growing season.The evapotranspiration requirement comprisesevaporation from the adjacent soil surface, evaporationfrom the intercepted water and transpirationfrom the stomata of the epidermis of the plant

II.4-32GUIDE TO HYDROLOGICAL PRACTICESsurface such as bark or leaves. In addition, water isalso required for the metabolic activities for plantgrowth. The total water required for healthy cropgrowth is referred to as the consumptive use.However, the water needed for the metabolic activityis very small – less than one per cent – comparedwith the evapotranspiration requirement and, assuch, the terms consumptive use and cropevapotranspiration are used interchangeably.Part of a crop’s water requirement is often met bythe local rainfall and a contribution from the soilmoisture storage, as well as through the capillaryrise of groundwater wherever the groundwatertable is nearer to the root zone. Only a portion ofthe local rainfall, called effective rainfall, is usedby the crop for its growth. Care must be taken withthe use of this term because effective rainfall meansdifferent things for practitioners of different disciplines.For a water resources engineer, effectiverainfall is the rainfall that reaches the storagereservoir as runoff, while for geohydrologists it isthe portion of the rainfall that contributes togroundwater storage. For an agronomist or farmer,however, it is the portion of the rainfall thatcontributes to meet the crop’s evapotranspirationrequirement. In terms of crop water requirements,effective rainfall is defined as that part of the rainfallwhich is useful directly or indirectly for cropproduction at the site where it falls, but withoutthe use of mechanical means. The remaining rainfalleither evaporates back into the atmosphere,runs off the soil surface, or is absorbed by the soilor percolates through the root zone. The amountof effective rainfall depends on various factorssuch as plant species, soil moisture conditions inthe root zone, climate and the time distribution ofrainfall. Details of the estimation of effective rainfallfor crop-soil-climatic-specific situations arediscussed in Irrigation and Drainage Paper 25 ofthe Food and Agriculture Organization of theUnited Nations (FAO)(Dastane, 1972).4.4.1.3 Determination of crop waterrequirementsOver many years, FAO has issued various guidelineson the estimation of crop water requirements. Inparticular, Irrigation and Drainage Paper 56 (Allenand others, 1998) contains a detailed computationof crop water requirements.In 1990, a panel of experts recommended to FAOthe adoption of the Penman–Monteith combinationmethod as the new standard for referenceevapotranspiration estimation. This method usesstandard climate data that can be easily measuredor derived from other commonly measured dataand is reported to provide consistent values for cropwater requirement calculations through the world.Basic definitions and concepts involved in thedetermination of crop water requirements are givenbriefly below.4.4.1.3.1 Evaporation and transpirationEvaporation is the process of converting liquidwater into water vapour and its removal from theevaporating surface. Evaporation occurs from lakes,rivers, wet surfaces, soils and vegetation.Transpiration is the process of vaporizing liquidwater contained in plant tissues and removing it tothe atmosphere. Crops predominately transpirethrough stomata. Nearly all the water taken up byplants is lost by transpiration and only a tiny fractionis used within the plant for its metabolicgrowth.4.4.1.3.2 EvapotranspirationEvaporation and transpiration occur simultaneouslyfrom a cropped area and it is very difficult todistinguish between the two. Hence the two arerepresented together by the term evapotranspiration(ET). As a rule, the units of evapotranspirationare expressed as mm per day. Evaporation from acropped soil depends mainly on the amount ofsolar radiation reaching the soil surface and varieswith the stage of crop growth. At sowing stage,nearly 100 per cent of evapotranspiration comprisesevaporation only, while with full crop cover, morethan 90 per cent of evapotranspiration comes fromtranspiration. The crop type, variety and growthstage should be considered when assessing theevapotranspiration from crops. Variations in cropheight, crop roughness, crop rooting characteristics,albedo, resistance to transpiration and groundcover lead to different evapotranspiration valuesfor crops under identical environmentalconditions.Three terms are used to express evapotranspiration:the reference evapotranspiration (ET 0), cropevapotranspiration under standard conditions(ET c) and crop evapotranspiration under nonstandardconditions (ET c). For more informationon evaporation and evapotranspiration, seeVolume I, Chapter 4, of the present Guide.4.4.1.3.3 Reference evapotranspirationThe evapotranspiration rate from a reference surfacethat is not short of water is called the referenceevapotranspiration, ET 0. The concept of the

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-33reference evapotranspiration facilitates the study ofthe evaporative demand of the atmosphere independentlyof soil factors, crop type, cropdevelopment and management practices. Thus theonly factors affecting reference evapotranspirationare climate parameters; hence it can be computedfrom observed or estimated weather data. The FAOPenman–Monteith method is recommended as thesole method for determining reference evapotranspiration.The method has been selected because itclosely approximates grass reference evapotranspirationat the location evaluated, is physically based,and explicitly incorporates both physiological andaerodynamic parameters. The method requiresradiation, air temperature, air humidity and windspeed data. Calculation procedures to deriveclimatic parameters from meteorological data arepresented. Procedures to estimate missing meteorologicalvariables required for calculating referenceevapotranspiration are also outlined. This allowsfor the estimation of reference evapotranspirationunder all circumstances, even where climate data ismissing. Relating evapotranspiration to a specificsurface provides a reference to which evapotranspirationfrom other surfaces can be related. Thisobviates the need to define a separate evapotranspirationfor each crop and stage of growth. Such areference value also facilitates a comparison ofvalues of reference evapotranspiration at differentlocations or in different seasons. The referencesurface is a hypothetical reference crop with specificcharacteristics. The reference crop is defined as:..a hypothetical crop with an assumed height of0.12 m, with a surface resistance of 70 s m –1 and analbedo of 0.23, closely resembling the evaporationfrom an extensive surface of green grass of uniformheight, actively growing and adequately watered...(Allen and others, 1998).Detailed calculations of the reference cropevapotranspiration are given in Chapter 4, Part A ofFAO Irrigation and Drainage Paper 56 (Allen andothers, 1998). Use of other denominations such aspotential evapotranspiration (PET) is discourageddue to certain ambiguities associated with suchterms.4.4.1.3.4 Crop evapotranspiration understandard conditionsThis refers to the evaporation demand of cropsthat are grown in large, adequately irrigated fieldsunder excellent management and environmentalconditions, and achieve full production under thegiven climatic conditions (Allen and others,1998).4.4.1.3.5 Crop evapotranspiration undernon-standard conditionsActual crop evapotranspiration is affected byfactors such as soil salinity, presence of hard pansin the subsoil, poor soil fertility and soil management,inadequate plant protection measures,ground cover, plant density and soil water content.Hence under such non-standard conditions, cropevapotranspiration under standard conditionsgenerally requires an adjustment. The predictionof the reduction in evapotranspiration caused bysoil water salinity may be achieved by combiningyield–salinity equations from FAO Irrigation andDrainage Paper 29 (Ayres and Westcot, 1985) withyield–evapotranspiration equations from Irrigationand Drainage Paper 33 (Doorenboes and Kassam,1979). These details are given in FAO Irrigationand Drainage Paper 56 (Allen and others, 1998).The crop evapotranspiration, ET cfor any particularcrop is determined by multiplying the referenceevapotranspiration ET 0, with a coefficient K c, calledthe crop coefficient (ET c/ET 0). The value of the cropcoefficient is crop specific and is dependent on thestage of growth of the crop and the prevailingweather conditions. The differences in the cropcanopy and aerodynamic resistance relative to thereference crop are also accounted for in the coefficient.As such, the coefficient serves as an aggregationof the physical and physiological differencesbetween crops.4.4.1.4 Irrigation requirementIrrigation is defined as the artificial supply ofwater to plants to ensure the healthy growth of acrop. It stands in contrast to the natural supplyfrom rainfall, soil moisture and capillary contributionfrom groundwater, among others. The netirrigation requirement is the amount of irrigationto be provided to the plant root zone afteraccounting for the contribution of rainfall, soilmoisture and capillary supply from groundwater.It should include any special requirements suchas those for the leaching of salts from the rootzone and, in the case of rice paddies and jute, thewater required for land preparation, standingwater requirements, percolation and periodicaldraining.Accordingly, the net irrigation requirements of acrop is equal to the evapotranspirational requirementET cat its root zone (the crop waterrequirement), plus special crop requirements suchas water required for leaching and land preparationminus effective rainfall, plus the contribution of

II.4-34GUIDE TO HYDROLOGICAL PRACTICESsoil moisture and the capillary supply fromgroundwater.Supplying irrigation to a crop inevitably incurssome water loss between the source and the rootzone as a result of the transport, distribution andapplication of the water. The efficiency of irrigationdepends on the efficiency of the conveyance system,the distribution system and the particular methodand timing of irrigation application. Accordingly,the total amount of irrigation water required,known as the gross irrigation requirement, isassessed as follows:Gross irrigationrequirement=Net irrigation requirementIrrigation efficiencywhere irrigation efficiency is the application efficiencytimes the distribution efficiency times theconveyance efficiency.The net and gross irrigation requirements can beassessed at the individual field or farm level aswell as at the level of command area of an outletor minor or major distributary branch or maincanal, or an irrigation project using the correspondingvalue of efficiency. The gross irrigationrequirement is expressed in terms of volume ofwater per hectare of cropped area over a specifiedperiod of time such as a week, month or cropseason.4.4.1.5 Irrigation systemsSources of irrigation water can be stored in manmadestorage reservoirs or lakes, groundwaterdeveloped locally through open wells or tubewells, or through diversion of natural channels. Inthe case of storage reservoirs formed behind damsand diversion weirs, irrigated areas may be far fromthe source and water may need to be transmittedthrough a large distribution network of canals andmajor and minor distributaries before it is deliveredto the farmers’ fields (see Figure II.4.11). Thedistribution network is generally minimal forwater sourced from river diversions, lakes andgroundwater wells, in descending order.Considerable seepage losses are associated withconveyance of water from the source to the fieldoutlets. Additional water losses are associated withdistribution of water below the outlets throughthe water courses and field channels. It is thereforeimportant that the irrigation efficiency used forassessing the gross irrigation requirement at thesource take account of conveyance and distributionlosses in the system.Irrigation is applied to crops by various methodsthat can be broadly classified as surface and subsurfacemethods. The irrigation methods can be furtherclassified as irrigation based on gravity flow or basedon pressurized water flow (see Figure II.4.12).Detailed descriptions and design procedures forCommand areaof main systemMain canalBranch canalMajor distributaryMinor distributaryField outletsBlockWater courseField channelOutletRainETRunoffRoot zonePerclationWatersupplyWater deliverysubsystemFieldsupplyWaterapplicationsubsystemFieldintakeWater use(cropping)subsystemSurfaceSubsurfaceoutflowWater removal(drainage)subsystemOutflowFigure II.4.11. Irrigation system

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-35Irrigation methodsGravity irrigationPressurized irrigationSurfaceSubsurfaceBorder Check basin FurrowSprinklerDripCorrugatedfurrowsDeepfurrowsSurgeirrigationRotating gunMicrosprinklerLow-energypressurizedapplicationFigure II.4.12. Broad classification of irrigation methodsthese irrigation systems are available in standardtextbooks on irrigation.4.4.1.6 Soil moistureDifferent methods of irrigation water applicationgive rise to different patterns of moisture availabilityin the crop root zone (see Figure II.4.13). Theirrigation method that produces uniform root zonemoisture content at or near the field capacitythroughout the crop season causes the least stressto the plant, thereby facilitating healthy growth.Modern irrigation control and application techniquesare based on monitoring the soil moisturestatus in the root zone. The soil moisture content atdifferent depths in the root zone can be determinedthrough gravimetric or volumetric methods.Modern tools such as the neutron moisture probeand the time domain reflectometer are beingincreasingly used for accurate monitoring of thesoil moisture content in the root zone (seeVolume I, 4.5). Information from soil moistureprobes can be directly entered into computerswhich assess the irrigation requirement and operateFieldcapacity(1/3 atm)Moisture contentWilting point(15 atm)0 5 10 15 20Daysatm – atmosphereDrip methodSprinklermethodSurfacemethodFigure II.4.13. Soil moisture regimes in differentirrigation methodsthe irrigation system automatically. Most of themodern automated irrigation control and operatingsystems are based on this approach.4.4.1.7 Irrigation scheduling of cropsIrrigation water is to be applied in such a way that,as far as possible, the crop water requirements aremet over time. As the crop water requirementschange in time with the stage of growth of the crop,as well as with the occurrence of rainfall, the supplyof irrigation water to a crop should follow a wellplannedschedule. Such a schedule should ensurethe application of the right amount of water to thecrop at the right time so as to obtain high yield ofgood quality produce, with high water use efficiencyand with least damage to the environment,all at a low cost of operation. Determining such aschedule is referred to as irrigation scheduling.There are several practices in vogue for irrigationscheduling. Irrigation scheduling procedures varydepending on whether there is an adequate supplyof water or the supply is limited.4.4.1.7.1 Irrigation scheduling under adequatewater supplyWhen an adequate supply of water is available, theobjective of irrigation scheduling is to eliminateperiods of water deficit so as to achieve the fullpotential crop yield. Irrigation doses are applied toreplenish the soil moisture whenever the soil watercontent of the root zone falls to a level at which itbegins to have an adverse impact on crop yield. Theagronomists or crop scientists seek to maximize theoutput of the individual crop fields for a given watersupply using the empirically derived scientificknowledge of crop response to available soil water.The main factors which govern the irrigation schedulein this case are climate, soil, type of crop and itsstage of growth. Numerous agronomic studies arereported in the literature describing irrigation

II.4-36GUIDE TO HYDROLOGICAL PRACTICESschedules for a large variety of cropping systemsand crops: The studies are generally focused ondetermining the depth and intervals of irrigationbased on crop water requirements during differentstages of the crop growth, soil moisture extractionpatterns from the root zone, optimal soil moistureregimes to be maintained in the root zone at variousgrowth stages and other factors.These procedures are generally guided by one of thefollowing criteria:(a) Critical growth stages of the crop (Prihar andothers, 1976);(b) Ratio of irrigation water applied to the cumulativevalue of pan evaporation (Prihar andothers, 1976);(c) Soil-water depletion (Rao and others, 1988aand 1988b; Hajilal and others, 1998);(d) Assessment of crop evapotranspiration usingclimate factors and crop factors derived fromexperiments conducted locally for each cropand based on the Penman–Montieth method(Doorenboes and Pruit, 1977; Allen and others,1998);(e) Soil-moisture tension values observed in thefield;(f) Visual crop features: plant drooping, changein leaf colour, rolling of leaves and so forth;(g) Plant indices such as relative leaf water potential,leaf water content and leaf water diffusionresistance.4.4.1.7.2 Irrigation scheduling under limitedwater supplyWhen water supplies are limited, crop water deficitsin some periods of the growing season are unavoidable.Crop response to deficits at different periodsof the growing season is not uniform and deficits insome critical periods of growth have a greateradverse impact on yield than in others. Therefore,under limited water supply conditions, the irrigationscheduling problem becomes one of distributingthe deficits over the crop growing season in such away that they have minimum impacts on cropyields. The problem is complex and any attempts atits solution require integration of information onsoil, growth stage of the crop and crop responses totimed inputs of water. Such a framework consistsessentially of the development and incorporationof the following three components:(a) A soil-water balance model to break down thewater inputs (rainfall and irrigation) into differentcomponents;(b) A dated water production function model ofcrops to relate crop yield to water used at differentperiods of crop growth;(c) An optimal irrigation programming model thatincorporates (a) and (b).4.4.1.7.3 Optimal irrigation schedulesIrrigation programming refers to the process ofdrawing up an optimal schedule of irrigationapplications for a crop during its growth periodunder specified conditions of water availability andclimate. This calls for exploration and evaluation ofthe effect of all possible irrigation regimes on thecrop yield and establishing an optimal regime usingan appropriate mathematical optimizationprogramme. The dated water production functionsthat relate the yield response to timed irrigationapplications over the crop season provide a meansfor this and incorporate effects of both timing andquantity of irrigation water application on cropyield. The time periods are either chosen to coincidewith the physiological growth stages of crops or aretaken as some convenient time period such as amonth or a week.The main difficulty addressed by these models isthat irrigation decisions in different intervals ofthe growing season are not independent. Each irrigationdecision is based on available soil moisture,crop status and available water supplies for theremaining period of the growing season.Information about all of these factors up to thetime of the decision must be utilized before thedecision to irrigate is made. Those decisions arethus multistage, sequential and state dependent.Basically, these models simulate the variousdynamic processes that lead to crop productionand yield and their changes in response to thechanges in environmental conditions of whichwater stress is one. Near-potential yields can beobtained by proper choice of irrigation scheduleseven under relatively high water supply deficits(Rao and Rees, 1992).Despite advances in irrigation scheduling models,human judgment and expertise will continue to bea major source of decision support in irrigationmanagement. The most appropriate irrigationschedule can be developed by using quantitativemodel predictions together with local knowledgeand experience of farmers and irrigation operators.Such a balance is possible by integrating modelsand heuristic knowledge in an expert systemframework.Many farmers grow a number of crops in the sameseason. In such situations, a limited water supplyimplies that water is not adequate to produce potentialyields of all the crops. This leads to competition

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-37for water between crops, both at the seasonal andintraseasonal level. The problem of multicropseasonal and intraseasonal allocation of water canbe solved by dividing the problem into two levels,seasonal and intraseasonal.4.4.1.7.4 Irrigation scheduling In real timeIrrigation schedules can be based on optimizationmodels for planning and design purposes. However,in real-time operation, both weather and watersupplies may be different from those assumed inderiving the planned crop irrigation schedules.Hence, the optimized schedules may need to bemodified to match the real-time information onweather and water supplies. Since the effect of eachdecision can be evaluated as crop yield only at theend of the season, irrigation decisions should bedeveloped in a sequential manner while goingforward in time. Therefore, an irrigation decision isto be made each week with the entire planninghorizon in mind.4.4.1.7.5 Use of medium-range weather forecastsin irrigation schedulingMedium-range weather forecasts provide informationabout the weather 3 to 10 days in advance andcan be used in agricultural management includingirrigation scheduling. Historical rainfall data can beused to examine the influence that the three-tofive-dayadvance information on rainfall will haveon the irrigation scheduling of crops. Whilemedium-range weather forecasts are useful for irrigationscheduling regarding shallow soils andsituations in which small irrigation depths areapplied frequently, they do not lead to significantwater savings for deep-rooted crops in soils withrelatively highly available water capacity.4.4.1.8 Irrigation methods4.4.1.8.1 Traditional irrigation methodsThe operational efficiency of traditional flow irrigationsystems is low; therefore, there is a great need foradopting modern, efficient irrigation methods.Alternate furrow irrigation, surge flow irrigation andpressurized irrigation systems (drip and sprinkler irrigation)are considered to be efficient technologies.4.4.1.8.2 Surge irrigationSurge flow irrigation is a recent surface irrigationmethod (Stringham and Keller, 1979; Stringham,1988). This is accomplished by surging the waterdown the furrows at timed intervals until the waterreaches the end of the furrow. In the past few years,field researchers have investigated significant newroles for surge flow that rely on its ability to distributewater uniformly, save water, reduce infiltrationand deep percolation losses, and control runoff anddrainage through surface systems. This method,which applies water uniformly, is used to create ashallow, uniform water profile and keep the waterat the root zone cutting down deep percolation. Ithas an efficiency of 85 per cent, and saves up to 25per cent in fertilizer costs.4.4.1.8.3 Pressurized irrigation systemsAs the water is conveyed through a pipe system, theconveyance losses are eliminated, resulting inhigher irrigation efficiencies. Pressurized systemsare recognized as achieving high water-use efficiencyand improved crop productivity with lowlabour inputs and adaptability to hilly terrain. Theyare suitable for water-scarce areas, can reduce frostattack and can readily apply water-soluble fertilizers.The system is well suited to canal, tank andgroundwater irrigated areas. All close-grown cropssuch as cereals, pulses, oil seeds, sugar cane, cottonand other plantation crops can be grown using thesprinkler irrigation method. An advantage of thesesystems is that undulating lands and shallow soilareas can be irrigated without having to level theland.4.4.1.8.4 Drip systemThe drip system of irrigation is a comparativelymodern method of water application. The initialinvestment is costly, but is the drip system is suitablefor situations calling for high water-use efficiencyand involving undulating terrain. Considerableexperimental research has been carried out over thepast 30 years to investigate water savings and yieldincreases, design of appropriate components andtheir materials, moisture distribution and irrigation,and fertilization under drip irrigation. Waterapplication efficiencies of 80 to 90 per cent can beachieved with this method.4.4.1.8.5 Sprinkler systemSprinkler systems distribute water in a manner similarto rainfall, so that the runoff and deep percolationlosses are minimized and uniformity of applicationis close to that obtained under rainfall conditions.4.4.1.8.6 Microsprinkler systemMicrosprinklers facilitate spraying of water underthe tree canopy around the root zone of the trees,

II.4-38GUIDE TO HYDROLOGICAL PRACTICESabout 30 cm high, and work under low pressure.This method is least affected by wind. The exactquantity of water required can be delivered daily toeach plant at the root zone. Water is given only tothe root zone area as in drip irrigation but unlikethe much wider distribution provided by sprinklerirrigation. This method is well suited to the wateringof trees, orchards and vegetable crops,particularly in combination with the use of localrenewable energy sources for pumping water.4.4.1.8.7 Low-energy precision applicationsystemsRecent innovations in microsprinkler systems arelow-energy precision application systems. In these,the laterals are equipped with drop tubes fittedwith very low pressure orifice emission devicescalled socks. Water is discharged just above theground surface into dead-end furrows or microbasins,thus preventing soil erosion and runoff. Thesesystems are not affected by wind forces and, inaddition to saving considerable energy, theyprovide uniformity of application and very highapplication efficiencies, in the order of 98 percent.4.4.1.9 Development of decision-supportsystems and use of geographicalinformation systems in irrigationIt is useful to link simulation models and systemmodels to spatial databases by means of a geographicalinformation system so as to develop expertdecision-support systems for conjunctive use andreal-time irrigation operation. This approach focuseson providing decision support to irrigation plannersand managers, enabling them to use routinelycollected spatial data and forecasts moreeffectively.4.4.1.9.1 Geographical information systems forspatial distribution of rechargeThe spatial distribution of recharge for variableweather, soil, land-use and water-supply conditionsover the command area of an irrigation project canbe assessed using a geographical informationsystem. A new coverage can be derived by superposingdigital maps of the command area with differentmap coverage, such as those for rainfall, groundwaterand cropping patterns. Each of the polygonalareas of this coverage will be homogeneous withrespect to all the coverage used. As such, these polygonscan be used as the basic units for water balancestudies and irrigation scheduling (Chowdary andothers, 2003).4.4.1.9.2 Development of decision-support systemsfor real-time irrigation managementDecision-support systems can be developed for thereal-time management of irrigation systems by suitablycombining the real-time data with thedecision-support-system scheme developed to planirrigation system management. A simple soil-waterbalance model can be used to assess the root zonesoil moisture condition and a simple canal flowmodel can be used to account for seepage losses.Based on this information and on knowledge of thewater available in the distributary and the mediumtermweather forecasts, it is possible to derive thebi-weekly irrigation requirements at the head of eachdistributary. This information can then be linked tothe geographical information system of the commandarea canal system to facilitate the following tasks:(a) Selecting the distributary of interest from thecanal network;(b) Running the field water balance model in realtime;(c) Drafting a report of the current water status;(d) Preparing a water indent for the irrigationrequirements at the head of the distributary.4.4.1.10 Conjunctive use of surface andgroundwater in irrigationConjunctive use refers to the integrated management ofsurface and groundwater resources in a harmoniousmanner so that the best use of both water sources isachieved to meet specified objectives in the area. Forimproved water-use efficiency in canal-irrigatedcommand areas, optimal and efficient utilization ofsurface water and groundwater becomes imperative andshould be ensured from the planning stage. For example,using surface water during the monsoon period andgroundwater during the non-monsoon period to irrigatethe same land mass is a type of conjunctive use. Similarly,seepage from the canals and percolation of irrigationwater both contribute to groundwater storage whichcan be withdrawn at a different point in time for irrigation.This is another, albeit inadvertent, example ofconjunctive use. Conjunctive use can help to achievethe following aims:(a) Increase the availability of water supply for irrigation;(b) Enhance sustainability of the long-term groundwaterregime equilibrium;(c) Improve regulation and facilitate the phaseddevelopment of a water resource, using thestorage space of the aquifer;(d) Provide flexibility in supply to match the waterdemand by smoothing peaks in surface watersupplies;(e) Reduce waterlogging and soil salinity.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-394.4.1.10.1 Guidelines for conjunctive useIrrigation planning for conjunctive use requiresconsideration of quantitative and qualitative aspectsof groundwater and surface water resources as wellas economic aspects. Putting conjunctive use intooperational practice requires the development ofguidelines (CWC, India, 1997) which may includethe following tasks:(a) Mapping of groundwater conditions and theirchanges in time and space;(b) Quantification of available groundwaterresources in the region based on detailed waterbalance studies;(c) Assessment of the additional recharge togroundwater;(d) Estimation of minimum desirable and maximumpermissible limits to additional extractionof groundwater for conjunctive usepurposes;(e) A broad water-use plan, based on existing wateravailability conditions;(f) Planning the regulated combined use of groundwaterand surface water in time and space;(g) Identifying and detailing the areas to be servedfrom the surface water and groundwater sourcesseparately or in combination;(h) Assessing the adverse socioeconomic impactsof conjunctive use in the long term.4.4.1.11 Use of marginal quality water forirrigationWater is considered suitable for irrigation when ithas no negative osmotic or specific toxic effects oncrop production, contains no solute affecting thechemical or hydraulic properties of soil and doesnot cause deterioration of groundwater or surfacewater. These adverse conditions are caused primarilyby salt accumulation in the root zone of plants.Accordingly, water of marginal quality may be usedduring stages of growth that are less sensitive topoor-quality water, especially salinity, and by ensuringthat there is no accumulation – or as little aspossible – of salts in the root zones. This can beprevented, either by leaching with a regular supplyof adequate water or by adopting special irrigationmethods. In situations of inadequate water availabilityor of water salinity, the drip and pitcherirrigation methods are the most appropriate. Thesemethods ensure that the salts do not accumulatenear the roots and maintain low soil-moisturetension, thus protecting the plants from adverseeffects.Different qualities of water can also be used in aridclimates by blending the marginal quality waterwith good-quality water in the supply system toproduce a predetermined quality to match the salttolerance of the crop, or through alternate irrigationwith good- and marginal-quality waters fromdifferent sources, such as canal water and salinegroundwater.Crops have different salinity tolerance levels. Whensalinity cannot be maintained at acceptable levelsby using the above methods, it is desirable to choosecrops or varieties that are tolerant to salinity, suchas vegetables, barley, sorghum, wheat and tomato,and to adopt suitable soil and water managementpractices, along with a judicious use of fertilizers.4.4.2 Agricultural drainageAgricultural drainage is the removal of dissolvedsalts and excess water from the root zone and landsurface to create more favourable plant growthconditions. Agricultural land drainage by surfaceand subsurface drainage systems was reportedlypractised by Egyptians and Greeks in prehistorictimes.For most irrigation projects throughout the world,drainage needs have not been adequately assessedand handled. Failure to realize the potential benefitsof irrigation projects is often attributable toinadequate attention to drainage. The cost of drainageis often significant and acts as a deterrent toinvestment in the initial planning stages and implementationof irrigation projects. The adverse effectsof inadequate drainage begin to appear only afterseveral years of operation of an irrigation system.Attention to drainage at this stage is generally toolate and hence ineffective.4.4.2.1 Purpose of agricultural drainageWaterlogging of agricultural land is, in a broadsense, the condition of saturation of the crop rootzone leading to restricted aeration, reduced oxygenlevels and increased carbon dioxide levels. Underhot, arid conditions, the evaporation process bringsup shallow subsurface water along with dissolvedsalts to the soil surface and can render the soil salineafter many years. Conditions of waterlogging andsoil salinity are detrimental to healthy crop growthand lead to reduced agricultural production. Thereasons for waterlogging and soil salinity are many,and include high rainfall, unfavourable topography,lack of natural drainage, low-permeabilitysoils, soils with hard pan at shallow depths, seawaterintrusion, high evaporation during long, hotand dry periods, and the presence of salts in thesoil. In addition, many anthropogenic activities

II.4-40GUIDE TO HYDROLOGICAL PRACTICESaccentuate the problem, such as the inappropriatemanagement of land and irrigation water, use ofpoor-quality water for irrigation, high seepage fromirrigation systems, adoption of unsuitable croppingpatterns and blockage of natural drains and outletsdue to the construction of roads, culverts, bridgesand railways. Most soils in arid regions containsome salts. India, Indonesia, Iraq, Egypt andPakistan, for example, have vast tracts of waterlogged,saline lands. Reclamation of such lands iscostly and has low economic returns.The objective of agricultural drainage is to improvethe physical and chemical environment of the landso as to enhance its productivity or maintain it at ahigh level. This is achieved by removing excesssurface and subsurface water, together with dissolvedsalts. The water to be removed may be excess waterapplied through irrigation, excess rainfall and seepagefrom conveyance or storage systems or irrigatedareas upstream. Most agricultural lands have somedegree of natural surface and subsurface drainage.Artificial drainage is achieved by installing surfaceand subsurface drainage systems to achieve thefolllowing:(a) Maintain a correct water and nutrient balancein the agricultural lands;(b) Remove excess water and stimulate healthycrop growth;(c) Restore root zone aeration;(d) Remove excess salts through surface disposalor leaching;(e) Increase the availability of applied nitrogenfertilizer by minimizing denitrification;(f) Reduce the specific heat of the soil-watermedium;(g) Lower the water table;(h) Increase the root zone from which nutrientscan be absorbed.fd – field drainld – lateral draincd – collector drainfdldcdSub-main drainFigure II.4.14. Drainage system layout showing thehierarchy of the components0.1%0.2%Main drain4.4.2.2 Types of drainageDrainage systems can be classified as surface, subsurfaceor vertical.4.4.2.2.1 Surface drainage systemsSurface drainage is the removal of excess water fromthe land surface through gravitational flow involvingmainly open drains and land grading to preventsurface water stagnation. The disposal of excesswater is achieved by installing a network of surfacedrains that link the area to be drained with themain outlet. A hierarchical pattern is usual in whichthe smallest component of the system is the fielddrain, followed by the lateral drain, the collectordrain, the sub-main and the main drain(see Figure II.4.14). In some situations, isolatedwaterlogged patches of land may be drained throughrandomly located drains. The field drains are small,temporary and shallow (

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-41clogged with colloidal material, obstructing the gravitationalflow into the drains and rendering the drainsineffective. Subsurface drainage systems in irrigatedlands in arid and semi- arid regions are suitable forleaching dissolved salts from the root zone. As subsurfacedrains are laid well below the land surface, thereis no loss of cultivable land area. However, while theinitial cost of a subsurface drainage system is greaterthan that of a surface drainage system, the maintenancecosts are practically negligible and theoperational life is much longer.4.4.2.2.3 Vertical drainage systemsA vertical drainage system involves the mechanicalpumping of the water through a shallow tube-wellsuitably designed and installed in the field. A multiplewell-point system, comprising a network ofclosely spaced shallow tube wells, can also be usedto provide drainage of a waterlogged region, particularlywhere salt water from deeper layers is likelyto be pumped up if a single tube well of higherdischarge capacity is used. The result of operatingsuch a system is that all the individual cones ofdepression will interlink and draw down the watertable under a larger area. Generally, all the wells arejoined to a single pump. The pumping rate isdecided according to the safe depth at which thegroundwater table is to be maintained. The systemcan drain excess water from depths of two metres,which is the normal limiting depth of subsurfacedrain systems. Vertical drainage calls for the expensiveconstruction of tube wells and a continuousenergy supply for pumping.Another form of vertical drainage involves inducementof evapotranspiration by planting appropriatevegetation over the area to be drained. Plantationssuch as eucalyptus, poplar and casuarinas, whichtranspire at a high rate, are being used for thispurpose. This is also referred to as biodrainage.Biodrainage is found to be especially appropriatefor landlocked areas where suitable outlets fordisposal of drainage water do not exist or are limitedin capacity. This recent technology requires furthertesting and evaluation to determine its suitabilityin specific situations.4.4.2.3 Design of agricultural drainagesystemsThis involves the following steps:(a) Surface drainage(i) Determining the quantity of excess waterto be drained;(ii) Deciding on the rate at which the excesswater is to be drained;(iii) Designing the physical components ofthe drainage system: selection of suitableoutlet location based on the knowledgeof existing outlets and disposal systems(natural streams), layout and sizes ofthe drains, design of outlet and ancillarycontrol structures;(b) Subsurface drainage(i) Determining the quantity of excess waterto be drained by finding the amount ofrecharge by rainfall or excess irrigation;(ii) Determining the hydraulic head understeady- and unsteady-state water tableconditions;(iii) Designing physical components of thedrainage system. This includes the determinationof the layout and the sizes ofthe drain pipes, the depth at which thepipes are to be located, slopes and alignmentof pipes, location and selection ofthe outlet and so forth.Pumping stationMain draincdfdSub-drainSub-drainfdfdcdfdcdcdcd cdThe design of a drainage system is based on theamount of water to be removed from an agriculturalarea in one day so as to avoid damage to thecrops due to waterlogging. It is referred to as thedrainage coefficient and is expressed in terms ofcentimetres per day or in litres per second perhectare. The value of the drainage coefficient is afunction of the rainfall characteristics, such asintensity and duration, the rate of runoff generated,the crop tolerance to excess water and the stage ofcrop growth.Natural drainfd – field draincd – collector drainFigure II.4.15. Subsurface drainage system layoutThe drainage coefficient is the key parameter in thedesign of surface and subsurface drainage systems.In the case of subsurface systems under steady-stateconditions, the coefficient has the same meaning asfor surface drainage systems. However, for unsteady

II.4-42GUIDE TO HYDROLOGICAL PRACTICESflow conditions, the concept of the drainage coefficientfor subsurface systems is different in the sensethat it is the rate at which the water table is to belowered.Accordingly, soil-water properties such as infiltration,saturated hydraulic conductivity anddrainable porosity play an important role in thedesign.The depth at which subsurface drains are placed isdecided on the basis of the maximum depth of theroot zone and the capillary rise of water in the soilwhich, in turn, depend on the soil texture.Details of design practice and operation of surfaceand subsurface drainage systems can be obtainedfrom standard text books on drainage. TheInternational Commission on Irrigation Drainagehas produced several publications on thesubject.4.4.3 Use of remote-sensing and generalinformation systems in irrigationand drainageRecent developments in remote-sensing technologyare proving valuable in the planning andmonitoring of irrigation and drainage systems.Remote-sensing can be used to identify land useand areas that are cropped, irrigated, waterloggedor flooded. It can also yield information on soilsalinity, crop water needs and stress, and cropyields. Information derived from remote-sensingtechniques, linked to a geographical informationsystem, is considered to be the future for planningand managing irrigation and drainagesystems.Landsat Thematic Mapper (TM) and SPOTMultispectral Scanner (MSS) data combined withradar measurements from the European remotesensingsatellite with synthetic aperture radar(ERS-1 SAR) may be used for obtaining informationon land use and crop areas. The temporalnormalized differential vegetation index (NDVI)can be used to monitor vegetal cover and cropgrowth. The low-resolution advanced very highresolution radiometer (AVHRR) satellite imagerycan be used operationally to estimate annual croparea, derive 10-day yield indicators and derivequantitative estimates of crop condition andproduction. Currently, the National MeteorologicalServices in a number of countries routinely provideNDVI maps, derived from the VIS, or visible, andNIR, or near-infrared, channels, on a monthlybasis for monitoring the vegetation growth andfor crop forecasting in support of real-time irrigationand drainage management.There is a great deal of literature on the use of satelliteremote-sensing applications in irrigationmanagement (Bastiaanssen, 1998; Musiake andothers, 1995; Vidal and Sagardoy, 1995; Kurtas andNorman, 1996).4.5 HYDROPOWER AND ENERGY-RELATED PROJECTS [HOMS K10, K15,K22, K45]4.5.1 GeneralMan has always exploited hydropower. The firstrecord of its application as a mechanical force goesback to Hercules who, in Greek mythology, deviateda river to clean the stables of Augias.The driving force of water has long been transformedinto mechanical force for use in mills andfactories. The advent of electricity at the end of thenineteenth century made it possible to transformthis hydraulic power into electric power, which ismore easily transmitted far from its source for useelsewhere. The use of hydroelectricity rose rapidlyduring the twentieth century and continues to havea promising future today.The recovery of the driving force of water is achievedprimarily in two ways:(a) Using the streamflow (speed of the water massflowing in the riverbed);(b) Using a drop in hydraulic head, that is, transformationof potential energy into kinetic energyby a change in altitude.Another energy use of water is its use as a coldsource for thermal power stations operated by coal,oil or nuclear fuels. Water is necessary in practicallyall the technical stages of thermoelectric energyproduction, from drilling exploratory bores inlayers of gas and oil, to the transformation of fossiland nuclear fuels into electric power in thermalpower stations.The fundamental difference between the productionof electricity of thermal origin and hydroelectricpower stations lies in water consumption. A thermalpower station will use water for cooling: part ofthis water will be evaporated by the energy productionsystem and part will be rejected at a temperaturehigher than that of the withdrawal. A hydroelectricpower station will return the same quantity of water

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-43to the natural environment, minus any loss byevaporation from reservoirs, but generally with asomewhat different hydrological regime.In river management, a series of various types ofpower stations along the river must be consideredat an early stage because of the potential for conflictamong the users regarding their needs. The impactsof these installations are broad: variations in flow,heating of the water, reduction in the number offish species, evaporation, diversions into or out ofthe catchment area, risks of pollution and so forth.4.5.2 HydropowerHydropower is renewable energy, derived originallyfrom the sun, which drives the water cycle, causingrivers to flow over millennia. Hydropower uses thisenergy without consuming water to any greatextent and can therefore be described as sustainableenergy as defined by the United Nations WorldCommission on Environment and Development:“...development that meets the needs of the presentwithout compromising the ability of future generationsto meet their own needs.”Hydroelectric schemes are diverse, not only as aresult of the different natural conditions to whichthey may be adapted, but also because of the diversityof circumstances related to power demand anduse. Hydroelectric power is frequently developed aspart of a multi-purpose project so that the projectmay involve the full range of water resources considerations,such as flood control, navigation,irrigation, municipal and industrial supplies, recreation,and fish and wildlife enhancement. Furtherinformation on multi-purpose projects is availablein 4.1.A project is rarely restricted to a local area. In mostinstances, it deals with an entire river basin, entailingregional, national and internationalconsiderations. In considering any magnitude ofdevelopment, the planning phase must take intoaccount all water resources needs of the region andthe ways in which such needs are to be met. Theeffects of a hydroelectric development project onthe resources and various needs in a region, and itscapacity to meet those needs, must be carefullyevaluated.Although hydroelectric projects have becomeincreasingly large during the past century, smallhydroelectric plants of up to a few megawatts caneconomically exploit the energy at potential siteson small streams, or they can often be integratedinto existing dams or artificial waterways.4.5.2.1 Advantages and disadvantages andimpact on the environment4.5.2.1.1 AdvantagesAlthough hydroelectric installations throughoutthe world meet around 20 per cent of global demandfor electrical energy, their output is proportionallygreater than that of other sources. They use energy,the supply of which, in almost all countries, isprone to risks associated with climate variabilityand change, but not to political or economic risks.Hydroelectric energy is especially significant as aneconomic stimulus in developing countries and asan important part of complex power systems inmore industrialized countries. Its importance willnot diminish for the following reasons:(a) It is derived from a continuously renewableresource powered by the energy of the sun;(b) It is non-polluting – significant heat or noxiousor greenhouse gases are not released in itsproduction;(c) Hydroelectric plant efficiencies can approach95 per cent, whereas fossil-fuel-fired thermalplants attain efficiencies of only 30 to 40 percent;(d) Hydroelectric plants have a long, useful life, ifproperly maintained;(e) Hydroelectric technology is a mature technologyoffering reliable and flexible operation, andits equipment can be readily adapted to siteconditions;(f) Water in storage provides a means of storingenergy and may be available for other purposes;(g) Hydroelectric plants are capable of respondingwithin minutes to changes in electricaldemands;(h) Hydroelectric generation has no fuel costs,and low operating and maintenance costsmean that it is essentially inflation proof;(i) It replaces the use of fuels which wouldotherwise have to be imported or, ifproduced nationally, could be exported,thereby improving a country’s balance ofpayments;(j) It generates a source of employment during itsconstruction, exploitation and maintenance,and helps reactivate regional and nationaleconomies.4.5.2.1.2 DisadvantagesHydroelectric energy does, however, have somedisadvantages, as follows:(a) Capital costs are relatively high;(b) There is only a limited possibility for a stageby-stageconstruction, possibly meeting a

II.4-44GUIDE TO HYDROLOGICAL PRACTICESgrowing demand for electricity, especiallybecause the largest investment must be madeat the beginning of civil engineering workson the river;(c) Production is often far from the centres ofconsumption;(d) Construction of hydroelectric energy plants is alengthy undertaking;(e) The rivers and lakes concerned are notprivate property and the decision to develophydropower must be taken at the nationallevel, involving thorny political negotiations– planning, construction and returnon investment may extend over severaldecades;(f) Potential destruction of natural habitats andthe loss of plant and animal species.4.5.2.1.3 Environmental impactA hydroelectric power installation clearly has animpact on the environment, as described in 4.2.8,and more specifically in 4.2.8.3.In particular, it can have the following impacts:(a) A modification of the river’s flow regime;(b) A fill of stored water volumes of one part of theyear on another;(c) Unnaturally rapid variations in streamflow;(d) Flooding of upstream areas.It will therefore be necessary to assess the other usesof water upstream and downstream from theplanned installation so as to take them into considerationin the design and operation of theinstallation.(c) The downstream part of a river will often bebroad and feature a shallow slope, but will havea constant flow which suits the installation oflow-head or run-of-river power plants;(d) The last category of installation includespumped storage plants.4.5.2.2.1 Power of an installationHydroelectric energy is developed by transformingenergy in water that falls from a higher level to alower level into mechanical energy on the turbinegenerator shaft and into electrical energy throughthe generator rotor and stator. The power potentialof a site in kWh is a function of the discharge andof the head as indicated below, and its exact expressionis as follows:P = 9.81ηQH (in kW) (4.7)where η is plant efficiency, Q is discharge in m 3 s –1 ,and H is the net head (fall) in metres, that is, thetotal height between the upstream level and thedownstream level.It is therefore necessary to know the exact powerthat will be produced and to have a clear definitionof the project components. Within the frameworkof a preparatory project and pre-dimensioning, thefollowing formula could be used, giving a very goodpower approximation:P = 8.5QH (4.8)where Q is the discharge (in m 3 s –1 , ) and H the nethead (in metres).4.5.2.2 Types of installationsIt is somewhat difficult to classify hydroelectricinstallations because they are all unique and they areadapted in each case to a river’s geomorphology, itshydraulic regime and the consumption needs of anarea or a country as a whole.The following is an attempt to classify them accordingto their position on a river and their type ofoperation:(a) The upstream part of a river generally has asteep channel (see the hypsometric curves inFigure I.2.21, Volume I, Chapter 2) and highlyvariable and seasonal low flow. Accordingly, ahigh-head power plant will generally be installed;(b) The middle reaches of a river have only amoderate slope, but a steadier flow. This willlead to the installation of a medium-headpower plant;4.5.2.2.2 High-head power plantsThese power plants (see Figure II.4.16) are characterizedby a specific hydrology because of theirlocation close to the source of the river, often inhigh mountains with a small catchment area. TheP = HQFigure II.4.16. High-head power plant

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-45hydrological regime will be typical of such locations:a highly variable flow according to the season,directly influenced by strong mountain rains and, ifat higher altitudes, by the snowmelt. As a result,there are periods of very high flow and periods ofvery low flow. The dam must therefore be able tostore water at the times of high flow so that it canbe used when there is a demand for electricity. Ifthe high flows come from snowmelt, the reservoirmust be sufficiently large to store all the water heldupstream in the snow pack.The means of calculating the dimensions of thereservoirs are described in 4.2. Because of the steepslope of the river, it can be arranged in such amanner that a large difference in head existsbetween the reservoir level and the turbines withoutthe need to transport the water long distances.It will then be possible to generate a large amountof power, in spite of comparatively low flows (seeequation 4.7). This relation for high dams can beexpressed in the following manner:P = 9.81ηQH (4.9)Installations of this type can store water and thustransfer it from one season to another and can havea complementary use in maintaining flow ratesduring periods of low flow to match electricitydemand.4.5.2.2.3 Medium-head power plantsIn the middle reaches of a river, the flow is alreadymore regular than upstream and the slope is stillsufficient to provide a useful head in the order of40 to 100 metres. It is therefore possible to installa dam that will allow some of the flow to be storedduring times of low electricity consumption: atnight, hours of low activity or on non-workingdays (see Figure II.4.17). Moreover, during periodsof high demand, stored water can be releasedin order to create a flow through the turbines thatis above-normal streamflow. This type of operationcan operate on a daily, weekly or monthlybasis.The power of such an installation can be presentedin the following manner:P = 9.81ηQH (4.10)4.5.2.2.4 Run-of-river power plantsThis type of installation, also known as a low-headpower plant (see Figure II.4.18), provides no waterstorage for later use and energy production is fullydependent on the current flow in the river. All or apart of the flow passes through the turbines and isreturned immediately to the river. There is thus nomodification of river flow. If a turbine is stoppedbecause there is no demand for electricity or itneeds to be repaired, the flow must be maintainedand diverted through an alternative route usingvalves or a bypass channel. Water thus diverted willbe lost for the supply of electricity. Since run-ofriverpower plants operate permanently, a detailedstudy of the river regimen is necessary to dimensionthe turbines and other characteristics of theinstallation.Floating mills are increasingly being used on largerivers. These are made up of water wheels, whichoperate electric alternators. The mills are installedon barges and are positioned on the river usingcables and winches. The advantage is that they riseand fall with the level of the river, they can bebrought back to the banks for maintenance andduring floods and can always be positioned at locationsof maximum flow. Moreover, the investmentis modest and the small electric generating unitsthat are required can be built locally. There is noneed to construct civil engineering works on theriver and only the winches need to be anchored onthe banks. The great disadvantage is the yield of thepaddle wheels, which is only 30 to 50 per cent.P = HQP = HQFigure II.4.17. Medium-head power plantFigure II.4.18. Low-head power plant

II.4-46GUIDE TO HYDROLOGICAL PRACTICES4.5.2.2.5 Pumped storage power plantsIt is impossible to store large quantities of electricity.Thus the electricity produced at times of weakdemand can be used to pump water and to store itin a reservoir at a good height above the river. Whendemand rises again, it is then possible to release thewater through turbines to produce electricity. Thetotal yield of the operation is approximately 70 percent, but it can be profitable if the energy used forpumping would otherwise have been lost becausethe low-head turbines on the river would have beenstopped. Plants of this kind resemble high-headpower plants and often the pumps are reversibleand also serve as turbines. The yield may not behigh, but it can be important because of its flexibilitywithin the overall generating capacity of a regionor country.4.5.2.3 Structure of a hydropower plantA hydroelectric power plant comprises several structureswhich are, from upstream to downstream, asfollows: the intake, headrace, penstock, powerhouse,tailrace or discharge water passage, and related structuressuch as fish ladders and a system for providingcompensation water. See Figure II.4.19.4.5.2.3.1 IntakeThe intake (see Figures II.4.20 and II.4.21) is necessaryto divert water from the river and direct it to theturbines. The intake is necessarily located near theriverbed, and is frequently incorporated in a dam.Fish laddersDam and intakeHeadraceCompensationwaterThe principal rules to be observed are as follows:(a) The position in the river must be so that floatingobjects do not block the intake;(b) The intake must be provided with protectivegrids to prevent objects from entering theturbines and fish from being trapped in thepowerhouse. In the latter case, the spacing ofthe bars must be based on local regulations andmay be only a few centimetres;(c) The surface area of the grids must be such thatit allows the flow to pass without creating toogreat a loss in pressure.As necessary, a system should be installedto allow the removal of floating objects thatfrequently accumulate upstream of the intake.4.5.2.3.2 HeadracesRiverPowerhousePenstockDischargeFigure II.4.19. Diagram of a hydropower plantThe headrace directs water to the power plant,which is often far from the intake, in order toScreencleanerIntakeFlowPowerhouseCraneGeneratorPenstockTurbineDrafttubeScrollcaseFlowTail race water levelFigure II.4.20. Example of intake in a dam

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-47benefit from as great as possible a difference inheight between the level of the intake and that ofthe discharge point into the river. They are eitheropen-top or covered canals, or tunnels with eitheropen or closed conduit flow.Open channel flow is possible only if the headraceleads from the top of the dam; the first partof the headrace depends to a large extent on thetopography of the site. In general, the slope ofopen channels is gentle and the water velocity islimited to about 2 m s –1 . Open channels oftenlead to a pressure pipe, or penstock (seeFigure II.4.21), which guides the water from thechannel or tunnel to the turbine down a verysteep slope. All of these parts are equipped withvalves to cut the water flow and to isolate themfrom the river for inspection and maintenancepurposes. Additional works, such as surge tanksor standpipes, are installed to accommodate accidentalexcessive pressure rises.Pressure pipes are often made of metal, but can alsobe made of reinforced concrete, pre-stressedconcrete or, as in the past, wooden planks assembledin a barrel-like form.4.5.2.3.3 PowerhouseThe powerhouse is the building or the undergroundexcavation that contains the generating units: theturbines and alternators. It must be adapted to thesize of the generating units and, in most cases,should include maintenance or repair shops. Ingeneral, electric units, such as voltage transformersand the terminals of the electric cables feedingthe network, are coupled with the powerhouse.A plan of a typical powerhouse is provided inFigure II.4.22.4.5.2.3.4 Tail race or discharge water passageTail races are needed to return the water to the riverafter it has passed through the turbines. The partwhich connects the powerhouse to the riverdepends primarily on the type of turbine and thuson the fall in head. These discharge passages veryoften include a gate to isolate them from the riverin case of an emergency. Where a number of powerplants are installed in a series along a river, the tailrace of one could be used directly as the intake forthe power plant downstream.4.5.2.3.5 Related structuresIn hydroelectric power plants, it is often necessaryto build structures which are not directly requiredfor the production of electricity production, but arenecessary for water management and the observanceof regulations. Examples include thefollowing:(a) Fish ladders (see Figure II.4.23) – to allow migratingfish to bypass dams and not be crushed bythe turbines;(b) Compensation water systems – to return a partof the discharge of the river directly to thefoot of the dam. This prevents the reach ofthe river between the intake and the dischargeof the tail race – the short-circuited section– from becoming dry so as to preserve theaquatic life of the river and allow other uses ofwater along the short-circuited section. In thissection it is normal to maintain a permanentflow in the river, the rate being set accordingRiverchannelbottomLower Snake River Juvenile Migration Feasibility StudyLower Granite Powerhouse StationFigure II.4.21. Example of a partially buriedpenstockFigure II.4.22. Plan of a typical powerhouse

II.4-48GUIDE TO HYDROLOGICAL PRACTICESFigure II.4.23. Example of a pool and weir fishladder, allowing fish to bypass the dam and reachthe higher water level through a series of lowheadweirsto country regulations or the needs of otherwater users.4.5.2.3.6 Special provisionsIt is becoming increasingly frequent for mediumto small power plants to incorporate all thesestructures within one barrage power station. Thishas the advantage of decreasing the length of theheadraces and reducing the short-circuited sectionto zero. Such installations can pass a constant flowin the river while continuously generating electricity.However, this is not possible without theconstruction of a dam and a river geometry thatgives rise to a desirable head difference at theselected location.4.5.2.4 Power plant flow determinationA reliable estimate of the energy that can be generatedat a selected site depends to a large extent onthe type of power station that is going to beconstructed and on the hydrology of the upstreamriver basin. The hydrological study of the river atthe power plant location should be as exhaustive aspossible, and should include the following information,which makes it possible to determine powerplant streamflow requirements:(a) Daily and monthly streamflow data for anextended period of time – more than 10 years,preferably 30 years, if possible;(b) Flow-duration curve or flow-frequency curve;(c) Historical records of floods near the site;(d) Computed design flood;(e) Mean annual discharge;(f) Minimum annual flow;(g) Minimum-flow requirements downstream fromthe site;(h) Streamflow diversions upstream from the damor intake works;(i) Drainage areas;(j) Evaporation losses from proposed reservoirsurfaces;(k) Stage–discharge relationship immediatelybelow proposed site;(l) Spillway design-flood hydrograph;(m) Dam, spillway and outlet rating curves;(n) Project purposes, available storage and potentialoperating rules;(o) Seepage losses, fish bypass requirements andother diversions from storage;(p) Reservoir elevation-duration information;(q) Annual peak-discharge data to assess risks associatedwith spillway design.The flow-frequency curve, or flow-duration curve,illustrated in Figure II.4.24, classifies the daily averageflows of an average hydrological year (seeChapter 5) and is widely used at the preparatoryproject stage. It indicates the number of days in theyear, or the annual return frequency, for which agiven flow is reached or exceeded, making it possibleto estimate the potential production of ahydroelectric power station as a function of streamflowreliability. In turn, an estimate can be made ofthe profitability of the investment.By using this approach, different productionstrategies can be simulated based on the physicaldesign of the dam and the number of turbinesinstalled. In addition, requirements of otherdemands on the water, such as minimum-flow,irrigation or drinking water supply, can be takeninto account.Flow m 3 s –190080070060050040030020010001.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00Probability of exceedanceFigure II.4.24. Flow-frequency curve of dailyinterannual streamflow data

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-49The direct use of this curve is simple, as shown inFigure II.4.25: For a minimum-flow requirement of50 m³ s –1 (thick line), and a required power plantstreamflow of 180 m 3 s –1 (dashed line), the flowthat can be used by the power plant is limited bythese two lines, based on the flow-frequency curvein blue.In other words, this figure indicates that the powerplant will be able to function for 92 per cent of theyear, or 336 days a year, when the river discharge ishigher than 50 m 3 s –1 . However, for 40 per cent ofthe time, or 146 days a year, the river flow will begreater than can be fed through the turbines; therefore,water will have to be discharged over orthrough the spillway unless the reservoir has a largeenough capacity to store the excess water.If such a curve is established solely on the basis ofmonthly mean flows, it can be useful for studyingthe effect of the dam on high and low flows, but thevolume that can actually be used by the turbineswill be over-estimated, and in turn the profitabilityof the system. This error is common and causesserious problems for those who invest in suchsystems; hence the importance of collecting andusing data on mean daily flows.Furthermore, since these curves are interannualmeans, they give a picture only of the mean flowon a given day. The usable water volume will varyenormously from year to year, as river flow dependson precipitation, which is not regular from year toyear. Therefore, the first years of production mightbe dry years and the financial amortization mightbe delayed during the first period. Investors shouldbe aware of these risks and include them in an estimationof overall costs. To this effect, a simulationof the operation of the power station by usingdaily data of the past several years is oftennecessary.Flow m 3 s –1900800700600500400300200100Flow-frequency curveMinimum flowRequired flow01.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00Figure II.4.25. Example of the use of a flowfrequencycurveThe recommendations and methods presented inthis Guide provide a good basis for studying thehydrological characteristics of an area. In particular,if river flow data are not available, methods forderiving data from rainfall-runoff models or fromneighbouring basins may be used, but they introduceadditional uncertainty which should be takeninto account when estimating the potential productionand hence return on investment.Other techniques used in the analysis of hydrologicaldata to extract relevant information for designpurposes are given in Chapter 6. The HydrologicalOperational Multipurpose System (HOMS)Reference Manual (see Section K) provides informationon the availability of software packages for theapplication of these techniques.4.5.2.5 Determination of the headThe head of a hydroelectric plant is determined bythe geographical and topographic characteristics ofthe river. It is important to distinguish between thefollowing factors:(a) The gross head, which is the difference betweenthe water level upstream of the intake and thewater level in the river at the point of dischargedownstream or, in certain cases, such as whenpartially submerged runner-wheel turbines areinvolved, at the wheel axle;(b) The net head, which is the pressure of thewater as it flows into the turbine. It is derivedby subtracting from the gross head any energylosses, including the losses of potential energydue to friction caused by components such asgrids, valves and pipelines. These energy lossesare a function of water velocity in the inletsand can reach several tens of metres where highheads are concerned.Therefore, the net head, which determines thepower of the machine, varies constantly. It dependson the following parameters:(a) The surface level of upstream water, which candiffer according to the season for many powerplants and dams;(b) The flow used by the turbines, which variesaccording to the demand for electricity;(c) The river discharge or flow, which can raise thelevel downstream, for example during floodswhen the spillway passes large flows. Thishappens even in low-head power plants, wherethere may not be enough gross head to makethe turbines rotate during floods.The net head is necessarily lower than the grosshead, and it can only be calculated once the

II.4-50GUIDE TO HYDROLOGICAL PRACTICESengineers have decided on the actual elements tobe installed. The proposed preparatory projectformula (see equation 4.8) considers only theaverage energy losses observed on a large number ofpower plants. For a precise calculation of the powerproduced, it is often necessary to simulate thefactory’s operation on a daily basis throughout theyear and determine the average net head.Therefore, the power of a hydroelectric power plantvaries significantly according to the prevailinghydraulic conditions but, by convention, the maximumcapacity of a power plant is always given asthe power generated at maximum flow under thelargest net head.4.5.2.6 Production of a generating plantThe energy produced by a hydroelectric powerplant is the most important factor to determinebecause it makes it possible to estimate the annualincome of the system and hence its financialviability. Production is deduced from the powerin kWh:E = Σ P i.t i(4.11)where E is energy in kWh and P iis the power of theplant during the time period t i.The power results from the net head at a givenmoment with the river discharge at the samemoment. This calculation can be made using themean net head associated with the mean flow overa period which is hydrologically relativelyhomogeneous.To calculate the income, it is necessary not only toevaluate the quantity of electric power produced,but also its production schedule as related to theselling prices of electricity because the price varieson the markets.It is therefore advisable to consider setting fixedprices for certain periods in the concerned area andto carry out a simulation of production accordingto each price period in accordance with the usableflow at the same periods. Although done by economists,the survey is based on the outputs offorecasting models developed by hydrologists.In preparatory project studies, and in areas wherethe number of hydropower stations to be consideredis limited, production over a given period canbe estimated by using the following formula:E = 8 AH/3600 (4.12)where E is the energy in kWh produced during acertain period, A is the volume of usable water inm³ during the chosen period and H is the gross headin metres.This simplified formula takes into account the averageenergy losses of power plants, as well as themean efficiency of all elements installed. It generallygives a precision of about 5 per cent.4.5.2.7 Water qualityWater quality is generally not a major concern inhydroelectric projects, although they can have aneffect on it. Various studies and recent experimentsshow that certain lakes can become eutrophic,either because they have been used as a recipient ofurban or industrial wastewater, or because, at thetime of dam construction, the flooded zone wasneither cleaned nor deforested. The decaying vegetationcan cause a significant reduction in dissolvedoxygen, severely limiting aquatic life for manyyears. This can be a very important consideration ifpisciculture or other water activities are practised inthe area of the establishment.As a result, water may become so acidic and corrosivethat it may attack the runner blades and otherparts of the turbine machinery (see 4.9, in particular4.9.2.2). A more serious effect can be thedischarge of de-oxygenated water into the riverdownstream, which can destroy fauna and floraseveral kilometres deep, bringing all fishing activityto a halt.Another risk is related to sedimentation in the reservoirs:sediments brought into the reservoir by theriver settle to the bottom because of the low speedof water and then undergo decantation at thebottom of the reservoir. These sediments cancontain pollutants such as heavy metals – lead,arsenic, copper – which concentrate in the reservoirand can reach dangerous levels.In certain cases, by draining the reservoir underflooding conditions, sediments can be cleaned outand returned to the river. However, such an operationshould be studied with care to ensure that itdoes not pose a threat to the downstream reaches ofthe river.4.5.2.8 Hydroelectric project stagesWhen constructing a hydroelectric power plant, itis essential to proceed on the basis of a clear plan inorder not to omit any important details, andto correctly evaluate the profitability of the

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-51investments which will be substantial, especially ifa dam is necessary. Small systems, for example arun-of-river power plant with a simple dam,may have a financial amortization period of 8 to10 years, compared with 30 to 50 years for largesystems. Thus the quality of the study depends ontwo major criteria:(a) The hydrological study, on the basis of whichthe potential output and annual risks areassessed;(b) The geophysical study, which is used to locatethe best site in order to have the greatest possiblehead in the selected zone.4.5.2.8.1 Hydrological studyAs previously stated, this study must be as completeas possible, outlining the methods used to determinethe flows and other characteristics so that theuncertainty related to the evaluation can beassessed. It is necessary to have daily or, at a veryminimum, weekly outputs. It is also essential toknow the risk involved in using these averagesbecause they can mask a flood that might be toomassive to pass through the turbines and will haveto pass through the spillways. Indeed a monthlyflow of 100 m 3 s –1 might be the result of a meanflow of 30 m 3 s –1 for 28 days and a flow of1 080 m 3 s –1 for two days, or even of a slowly varyingflow ranging from 120 m 3 s –1 to 80 m 3 s –1 . Thedifference will have a major impact on the energythat can be generated in a month.It is important to conduct a careful study of floodfrequency in order to dimension the works that willbe needed to handle such high flows withoutdamage to the dam or power plant. It is also necessaryto evaluate the project flood, which is themaximum flood that will be passed without anydamage to the work; a larger flood will be likely tocause serious damage to the installations. In manycountries project floods are defined in regulationsand computed according to downstream risks.In carrying out a flood study, it will not only benecessary to calculate the flow to dimension thespillways, but to locate all the high-tension electricalinstallations, including the power station itself.Unfortunately, as a result of inadequate flood studies,power plants are sometimes submerged, andelectrical installations destroyed, by floods of afrequency of only a few tens of years.Finally, to be complete, a hydrological study mustconsider the various uses of water and how theproject will make it possible to respect them. Itmust also take into account the various problems offlow related to existing installations such as bridges,mills, dams and fords in the reservoir’s zone ofinfluence and downstream of the power plant.Clearly, as the project takes shape, an increasingamount of information must be provided. This willrequire the expenditure of significant funds at thepreparatory project stage. If these funds are notmade available and an adequate hydrological studyis not undertaken, the profitability of the wholeproject will be at stake, with a likely loss of majorinvestments at stake.When computing the necessary hydrologicalelements, it is important to remember the flowsthat will have to be assigned to other water usersand to determine jointly whether the flows can passthrough the turbines and thus generate electricity.If so, it must be decided when and how, or whetherthey must be diverted upstream of the headrace.The outcome of these studies can change considerablythe project’s financial viability. In general,these studies involve regional or nationalgovernments.4.5.2.8.2 Geophysical studyAlong with the hydrological study, it is necessary toobtain as much information as possible to evaluatethe potential head. Plans, existing surveys ormedium-scale maps are generally used to carry outgeophysical studies, but should be supplementedby field surveys. The information collected in thefield is essential to determine the position of thedam and of the intake, to choose the best locationfor the power plant, and decide on the mosteconomic means of connecting the two.Such field visits are also vital for locating traces ofold floods, identifying different uses of water in thearea and determining whether the potential locationof the reservoir includes zones where watermight be lost through infiltration or zones wherefauna or flora will need to preserved.4.5.3 Operation of a hydroelectric systemThe operation of a hydroelectric system is verycomplex. It is defined by its generation capacityand the demand for power supply. It is necessary tofind a balance between present and future powergeneration, because the generation of large amountsof electricity in the present can lead to a energyproductiondeficit in the future. However, a lowlevel of generation in the present may lead to theexcess storage of water, which will need to bereleased later. Therefore, it is necessary to use adesign procedure that will optimize the use of water,

II.4-52GUIDE TO HYDROLOGICAL PRACTICESmaximizing benefits and minimizing costs. See 4.2for further information.4.5.4 Other projects related to energyproductionAs stated at the start of this section, while the principaluse of water in the generation of electricenergy is through hydroelectricity, water is alsoessential in the production of thermal energy. Aguide to the quantity and quality of water necessaryfor various thermoelectric energy-generating processesis provided below.4.5.4.1 Production of energy from fossil ornuclear fuelsThe use of water in the production of electricity isidentical to fossil or nuclear fuels. All thermalpowerhouses use water for the production of vapourand for the cooling system, and, to a lesser extent,for general services such as for drinking. Rivers andlakes serve as the cold source necessary for theCarnot cycle.The volume used depends essentially on thecharacteristics of the system used for coolingcondensationand evacuation of heat. Water as acoolant in the condenser is the most important useand the necessary quantity for this purpose is in theorder of 0.032 to 044 m 3 . s –1 . MW –1 on the basis ofan increase of temperature of 8°C. The principalmeans of dissipating the residual heat are drycooling towers and the direct discharge into riversof the effluents from the heat exchanger. Theapplication of regulations designed to limit excessivewarming of rivers has resulted in a reduction in theuse of the direct discharge into rivers. Evaporationcooling towers are the largest water consumers,discharging only the condensed water into theriver. Dry cooling towers disperse the residual heatof the plant directly into the atmosphere by meansof thermal exchangers cooled by air, without theaddition of heat to the natural water bodies, andwithout their consumptive use. Thus, the plantsthat use this system need a larger amount of fueland an additional plant investment.In the case of power plants using coal dust as fuel,water is also needed to transport the ashes. Thisdemands about 0.00095 m 3 . s –1 . MW –1 , and desulphurizationof the combustion gasses with ademand of about 0.0000019 m 3 . s –1 . MW –1 .As in any other complex system, nuclear plantsare exposed to numerous unpredictable problemsthat can interfere with their normal functioningand in extreme cases can endanger the healthand security of the population. The possibleoccurrence of serious accidents is undoubtedlyvery low because severe safety and safeguardmeasures are implicit in nuclear plant design(IAEA, 1981). WMO (1981) describes the differenttypes of nuclear power plant and analyses theproblems connected with hydrology and waterresources that should be considered in the planning,design, exploitation and the shut-down ofnuclear plants.This publication contains some examples of thetechnologies used to address important questionsat varying levels of complexity. Both high and lowflows are of special importance for the managementand security of a nuclear plant. It is essential thatemergency cooling of the nuclear core, the coolingof the used fuel and the final heat sink have a reliablewater supply. Protection against flooding is alsovery important, regardless of the type of powerplant, because it can interrupt normal operationsand especially if it affects two or several systems,thereby reducing the effectiveness of emergencysafety systems. Therefore, it is generally necessaryto apply the best available hydrological forecastingsystem to the basin upstream of a power plant andto perform periodic revisions of the hydrologicalanalyses and assumptions made in the planning ofthe plant.In most energy-related projects, considerations relativeto water quality do not control the feasibility ofthe project, but they can influence its size, the typeof procedures used, the choice of location and otherfactors. The composition of groundwater fromdifferent sources varies considerably in terms ofdissolved salts and gases. Surface waters generallycontain suspended load and often, dissolved orsuspended organic matter, originating from rottingvegetation or from wastewater. The growing use ofsynthetic detergents, some of which cannot easilybe destroyed in wastewater treatment processes,results in the presence of measurable quantities ofthese chemicals, even in drinking water supplyreservoirs.Rainwater can have a low pH and be potentiallycorrosive in industrial zones because of coal dustand oil particles. If these are carried by the wind,the impact can be significant even at a great distancefrom the emission sources. Most waters, however,can be treated to be used in the cooling by condensers,transport of ash and desulphurization ofcombustion gasses. Nevertheless, boiler feedrequires pure water, without any trace of dissolvedsalts. The cost of preparing pure water generally

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-53increases with the quantity of salts dissolved in thewater.To a large extent, radioactive waste from nuclearpower plants is caused by events such as leakage,blowdown, maintenance and fuel restocking. Thewater that circulates through the reactor is used asa heat source and the products of the corrosioncreated in the system are the principal source ofradioactive isotopes in water in the reactor. It isessential that the water used for cooling and feedwater be exceptionally pure, as all salts or otherimpurities contained in the water can captureneutrons and make them radioactive. The productsof fission within the fuel elements constituteanother potential spring of radioactive isotopes inthe reactor water. Therefore, the quantity of radioactiveisotopes present in the reactor water dependson the rate of corrosion, any failure in the coatingof fuel elements and its rate of elimination bycondensation or by cleaning of the reactor. Thepossible presence of radioactive isotopes in watercalls for special precautions to be taken in wastetreatment.In the primary circulation system, great attentionshould be given to maintaining water at a highstandard of purity in order to minimize the accumulationof excessive radioactivity due to impuritiesor corrosion products. There may be no loss ofprimary water, but some of it is extracted, purifiedand recycled. The possible danger of corrosionunder pressure implies that the boiler water containsvery low concentrations of oxygen and chlorides.To reach this level of purity, the water used in theprimary circulation system must undergo deaerationand evaporation treatments so as to reduce thelevels of oxygen and chloride to less than 0.003 and0.3 mg/l, respectively.4.5.4.2 Coal extractionThe extraction of coal, whether from open or undergroundmines, uses only a small quantity of water.In fact, the infiltration of water underground canbe an obstacle to mining activity and may requireconsiderable effort and investment to remove it.The production of coal dust makes use of largequantities of water for washing the dust but recyclingsystems are generally employed.Activated carbon sludge technology has been usedsince the beginning of the twentieth century. Thetransport of coal sludge can be economical for highvolumes or over large distances, but after the separationof the pulverized coal dust, the water must be treatedbefore being discharged into a natural watercourse.Wastewater treatment facilities will depend on thequality of the coal dust to be transported – its contentin sulphur, ashes and minerals – on the chemicaladditives necessary to inhibit corrosion in pipes andequipment, as well as on the chemicals used ascoagulation agents in the procedure.Wastewater from coal mines sites contain a widerange of metals, suspended solids and sulphatesoriginating from pyrites and/or from marcasite,which are commonly associated with coal deposits,and from schist and sandstone. If exposed to theatmosphere, these minerals form sulphuric acidand compounds of ferric hydroxide. Whether insettling ponds, slag heap of waste rock or whereverit is stored, coal dust can therefore produce aciddrainage. The impact on the receiving waters willbe to produce a high degree of acidity (pH of 2 to 4)and high concentrations of aluminium, sulphate,irons and trace amounts of heavy metals.4.5.4.3 Uranium extractionLittle water is used in underground or open uraniummines and what is used is mainly as drinking water.The total use of water during the tertiary crushingof uranium is also small and it is mainly used tolubricate the crusher.Uranium concentration generates both radioactiveand non-radioactive waste and effluents. Solid,liquid and gaseous effluents may be discharged intothe environment in large or small amounts, accordingto the procedure in place to check and controlthe release of the waste.4.5.4.4 Petroleum productionWater supply availability and cost, together withenergy conservation and environmental concerns,have an impact on petroleum processing. Modernrefineries are designed so as to reduce water demandto some two per cent of what it was for the olderrefinery systems and procedures. Currently, greatimportance is given to air-cooling in place of watercoolingand to the multiple uses of water, includingrecycling. The level of water utilization thereforedepends on the age of the refinery and tends to bedirectly proportional to the capacity and complexityof the refineries. The demand for water canfluctuate between 0.1 and 3 m 3 bbl –1 according tothe size and complexity, and the processes used bythe refinery.Effluents from petroleum production and the refiningprocess need to be treated before being releasedinto natural watercourses. Such treatment mostly

II.4-54GUIDE TO HYDROLOGICAL PRACTICESinvolves the use of settling tanks and the separationof petroleum from water. Because of the high quantitiesof water required by some procedures,recycling has become necessary in new refineries.4.5.4.5 Methanol productionThe conversion efficiency for producing methanolfuel from wood or natural gas is approximately60 per cent. Therefore, a large proportion of theheat content of the original carbon-rich sourcematerials must be rejected during the process.Approximately half of the heat loss can be rejectedvia an evaporation cooler, requiring approximately3 m 3 of water to be evaporated for every tonne ofmethanol produced. Alternatively, if direct coolingis used, and a 10°C temperature rise is permitted,170 m 3 of water would be passed through the heatexchanger to remove this heat with an inducedevaporation loss of 1.5 m 3 /tonne of product. Clearly,if water is scarce or costly, the design must includea means of eliminating heat that is efficient in itsuse of water.4.6 NAVIGATION AND RIVER TRAINING4.6.1 Application of hydrology tonavigationRivers are characteristic landscape features and partof the natural, cultural and economic environment.Besides their function as navigable waterways, theyhave great significance in terms of the nationaleconomy and ecology.During the early developmental stages of navigation,transport facilities were primarilydependent on the characteristics of the rivers orriver reaches concerned. Over time, the need forincreased transportation capacity led to thedevelopment of uniform navigation conditionsby means of river canalization or river training,which allowed long-distance transport on everlarger ships without frequent and expensivetranshipment.Since the early times of river navigation, depth andwidth have been the basic parameters of waterways.There are different concepts of waterway development.According to classical river-regime theory,river engineering based on hydrological characteristicsis preferred when dealing with free-flowingand strongly meandering flatland rivers, whilehydraulic engineering is the method of choicewhere steeper river reaches, including those withreinforced embankments, are concerned. Thenumber of parameters that can be taken intoaccount depends solely on the computer capacityavailable. Increasing emphasis is now being placedon the interaction between ship design – form,draught, mode of propulsion – and the structureand routing of waterways. As regards hydrologicalfeatures, a general differentiation must be madebetween free-flowing and impounded or canalizedriver reaches or artificial canals. Hydrological–hydraulic parameters and the features of theinteraction between ships and the waterway characterizeand define the quality of any navigablewaterway.Some factors that influence navigation remainmore or less constant over long periods and canbe described by well-defined parameters. Otherfactors, however, characterize the temporallyvariable navigation conditions that depend onthe streamflow regime of the river, particularlyon events such as floods and low-flow periods. Anexample of an event with negative consequenceswas the prolonged low-flow period in the Rhineriver in August 2003 (see Figure II.4.26). Anotherkey factor is the upstream catchment of theriver: its type and size and flow over the course ofthe year.Hydrology plays a key role in two primary aspectsof river navigation:(a) The characterization of river reaches withrespect to the types of vessels that regularly usethem for navigation (for example, waterwayclassification according to Figure II.4.27);(b) The current hydrological conditions thatcontrol the operation of vessels as a function ofnavigable depth or equivalent water levels, forexample.Figure II.4.26. Low flow in the Rhine river inAugust 2003 hinders navigation

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-55Figure II.4.27. Example of classification of waterwaysThese two aspects are discussed in greater detail inthe following sections.4.6.1.1 Application of hydrologicaldata to the characterization ofwaterwaysThe goal of an investigation of any waterway inrelation to its potential for navigation is to determinethe seasonal probabilities of navigability forvarious categories of vessels on the individualreaches of the waterway. This may be accomplished,for instance, by using a system of categories that arebased on relevant parameters, as those defined bythe United Nations Economic Commission forEurope. The definitions of several of these parametersare given below:Minimum depth of waterway (h) – The minimumdepth at the navigable low stage that ensures therequired width of the waterway.HNHKHKhBHhLWaterway or fairway – The part of the river that ispassable by ships and ship caravans, marked bymeans of navigation signs (buoyage).HNHNavigation clearance – The complexity of factorscharacterizing the depth, width, height and sinuosityof the waterway required for regular and safenavigation by vessels of given dimensions (seeFigure II.4.28).Note: See definitions, 4.6.1.1.Figure II.4.28. Geometrical elements of awaterway

II.4-56GUIDE TO HYDROLOGICAL PRACTICESMinimum width of waterway (B) – The minimumwidth at the navigable low stage that ensures therequired depth of the waterway.Prescribed vertical clearance (H) – The minimumvertical difference across the entire width of thewaterway between the lower edge of any structure,for example a bridge, and the navigable highstage.Minimum sinuosity radius (R) – The prescribedlower limit of the sinuosity radius of a river bendmeasured to the axis of the waterway during navigablelow stage.Navigable low stage (HK) – The critical stage ensuringthe prescribed value of water depth and width.Navigable high stage (HN) – The highest criticalstage generally ensuring the prescribed clearance.Navigation water demand – The streamflow that isneeded to ensure the depth required for safety andease of navigation.Minimum streamflow for navigation – The streamflowensuring the navigable low stage in a givencross-section.Maximum streamflow for navigation – The streamflowensuring the navigable high stage in a givencross-section.Navigation season – The part of the year duringwhich navigation is not hampered by ice.Ford – The transition reach with small depthbetween two bends of a river (as used in thiscontext).given river, to its minimum stage. The sinuosityradius can be determined graphically from acontour map of appropriate scale with sufficientaccuracy.In order to investigate the possibility of navigationon a river, it is necessary to carry out the abovementionedprocedure for several values of minimumnavigation width so that the navigation category ofthe natural river or waterway classification can bespecified.4.6.1.1.2 Hydrological parametersIn order to determine the degree to which therunoff regime corresponds to the navigable lowstage, it is necessary to compute flow hydrographsand duration curves of water stages or flowdischarges at defined cross-sections.The flow hydrographs should, if possible, bedetermined from daily data of a time series witha minimum length of 30 years and a wide rangeof probabilities (see Figure II.4.29). In addition,they should be computed for a number of probabilitiesof exceedance. The periods during whichthe prescribed minimum depth of waterway isexpected with a given probability can be determinedby superimposing the levels of navigablelow stage on these curves. The durations of these102.5102.0101.5Principal shallow – The shallowest section along agiven navigation reach.The procedures for describing these parameters areexplained in further detail below.4.6.1.1.1 Geometric parameters101.0100.5HK80%90%95%99%The determination of the depth and width availablefor navigation requires a closely spaced seriesof observation cross-sections (for example echosounder measurements) along the river. The minimumstage at which the minimum navigablewidth is still available has to be identified for eachcross-section. The navigable low stage for eachcross-section is determined by adding the minimumnavigation depth, as recommended for the100.0I II III IV V VI VII VIII IX X XI XIIFigure II.4.29. Average flow hydrographs in crosssectionat kilometre 1 695 on the Danube

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-57periods can then be obtained by computingdurations and/or probabilities. As the homogeneityof stage data is not always guaranteed, theduration of the relevant discharges should bedetermined first and then converted to stagedata by means of a valid stage–discharge relationship(see Volume I, Chapter 5). It is possibleto find the minimum duration of the navigablelow stage along the given river reach by comparingthe navigable low stage with the flowhydrographs in various cross-sections. For example,according to investigations carried out onthe Danube river, the navigable low stage correspondsto the water stage of 94 per cent duration,as computed for the series of ice-free stage data(Figure II.4.30).In temperate and Arctic climatic zones, the lengthof the navigation season is primarily determined bythe ice regimes of the rivers. On the basis of observeddata of the various ice phenomena such as ice drift,complete freezing, ice break-up, and ice cessation(Volume II, Chapter 6), the values of the variousphenomena expected with given probabilities canbe computed, and the durations of forced interruptionsof navigation by river ice can be estimated.The results of such a calculation for the Hungarianreach of the Danube river are shown inFigure II.4.31.In order to ensure the efficient operation of icebreakers(breaking, pitching, shredding), it isnecessary to obtain and analyse time series of observationsof ice thickness. Here it is particularlyimportant to identify the times when it is worthwhileto continue or commence ice-breaking so asto keep the fairway clear of ice, and when suchefforts should be abandoned as uneconomical.These times depend heavily on the meteorologicalconditions controlling the formation and break-upof ice.4.6.1.1.3 Hydraulic parametersThe investigation of the flow regime, as describedin the foregoing subsection, can only be carried outfor selected, relatively stable cross-sections.Therefore it is necessary to estimate the navigableFlow discharge, Q/m 3 s –17 000Ouma6 0005 0004 0003 000Q(H)Mohacs2 000Q 94%: 1 135 m 3 s –1H 94%: 217 cm1 0000 10 20 30 40 50 60 70 80 90 100Duration (%)100 200 300 400 500 600 700Water stage H (cm)Kilometre Duration (days)Duration (days) Frequency (per cent)100806040200140120100806040200120100806040200Bratislava1 850DunaremeteFrequency of ice drift and complete freezing1 800GönyüProbability of duration of icy daysProbability of duration of stable ice coverVágKomaramDunaalmasMosoni-Duna1 750EsztergomGaramIpoly1 700Vác1%20%50%80%99%1%20%50%80%99%Budapest1 650ErcsiAdony1 600DunaújvárosDunaföldvár1 550Dombori1 500SióBajaMohács1 450Figure II.4.30. Determination of navigable waterstage and flow of a given durationFigure II.4.31. Ice conditions along the Hungarianreach of the Danube river

II.4-58GUIDE TO HYDROLOGICAL PRACTICESlow and high stages by interpolation for the riverreaches between these cross-sections. The mostreliable method of interpolation, especially in thecase of the navigable low stage, is the developmentof water-level profiles. This requires knowledge ofhydraulic parameters such as the slopes and theroughness of the various river reaches concerned(see 6.3.6).4.6.1.2 Application of hydrological data tooperational navigationInland navigation is a complex economic activitythat is highly dependent on natural factors. Withoutreliable knowledge on the state of the riverbed, thestreamflow, the ice regime, and their expected variabilityover time, and the planning and operationof navigation activities would be seriouslyhampered. In order to provide this information, itis necessary to continuously collect data on thehydrological regime, predict expected changes andtransfer regularly these data and forecasts to potentialusers. In many cases, this is still done in theconventional manner with the support of NationalHydrological Services. Recently, however, themodelling systems and information services havebecome more and more routine and are often useddirectly by the navigation services themselves, forexample the use of the Electronic WaterwayInformation System on the Rhine river.4.6.1.2.1 Data collectionNavigation utilizes a wide range of data collected byHydrological Services. These include:(a) Data collected on the river basin, such as topography,vegetation, land use and precipitation.This is done in close cooperation with NationalMeteorological Services and regional planningauthorities;(b) Data collected at gauging stations: stage,streamflow, water temperature, air temperature,suspended-sediment load, bed load andice phenomena and so forth;(c) Physiographic data collected along riverreaches, such as variations in the river-course,bed structures, fords and their depths, flowdirection and velocity, water-surface profilesand ice phenomena.For most of the data required for navigation, theobservation methods are those used in general practice(see Volume I, Chapter 2), although differencesarise primarily in connection with measurementsmade at gauging stations and observations carriedout along the river sections between gaugingstations.The transitions between river bends of opposingorientation often contain shallow sections whichconstitute the most critical points in the longitudinalcourse of natural waterways. Accordingly, depthmeasurements of these shallows should beconducted frequently whenever the water depthabove the shallow does not reach the prescribedvalue. The depths should be measured along thecrest of the shallow section. As a result of thesemeasurements, the navigable width of the waterwaymay be determined for the shallow river reach.The length of the river reach in which the waterdepth is less than the minimum navigable depthshould be marked.Knowledge of the direction and velocity of flow isrequired to enable reliable manoeuvres of bargetrains through critical reaches such as shallowreaches, inlets and outlets, as well as in the headwatersand tailwaters of ship locks. The surfacevelocity is measured by means of floats, while thedirection and velocity of currents within thewater body are measured by current metresequipped with direction finders. The latest techniqueuses the acoustic doppler current profilerprinciple that makes it possible to measure orcompute all parameters needed at any point of across-profile.Standard ice observations made as part of routineprogrammes at the gauging cross-sections are notsufficient for safe flood discharge and navigation.They must be complemented with respect to theplaces where observations are made and thephenomena that are observed. The observationsmust be extended to river reaches between thegauging stations so that an observation point isestablished at least at every 5 to 10 kilometres. Themost important task is to observe river reaches regularly,particularly for ice jams. During periods ofdrifting ice and at times of freeze-over and breakup,observations should be made daily, while duringthe period of fixed ice cover and unchanged flowregime, observations made every 5 to 10 days maybe satisfactory. The reliability of ground observationsmay be enhanced and supplementedconsiderably by aerial surveys and photos. It isrecommended that ice maps be drawn at least every5 to 10 days and disseminated among the competentauthorities and users.Ice predictions for navigation require observationsof the first crystallized formations, and then thedevelopment of brink ice. Where hydraulic conditionssupport the forming of frazzle ice, its densityshould be characterized according to the followingthree steps: 0–33 per cent, 34–67 per cent and

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-5968–100 per cent of the depth of the river. Thedensity of drifting ice is characterized according tothe percentage of the surface area of the river that itcovers 0–10 per cent, 11–20 per cent, and so forth,up to 91–100 per cent.4.6.1.2.2 ForecastingThe efficiency and safety of inland navigationdepend on the reliability of hydrological data andof the forecasts of water stages under low-flow andflood-flow conditions, ice phenomena and waterdepths at narrows and shallow sections. There is aneed for both short- and long-term forecasts. Thoseresponsible for navigation are naturally interestedin forecasts of flow rates along the navigablestretches of rivers.In addition to the general methods of hydrologicalforecasting (see Chapter 7), navigation often usesmonthly forecasts that are compiled by taking intoaccount the water volume stored in the rivernetwork, both surface water and groundwater.Because navigation is particularly sensitive to thereliability of stage forecasts during low-flow periods,the confidence bands of the forecasts shouldbe narrow. For example, the following values areapplied for the Danube river:Probability of exceedance Width of confidence band60–70% 50 cm70–80% 40 cm80–100% 30 cm4.6.1.2.3 Transmission of data and forecastsThe data collected along a navigable river and theforecasts based thereon can only be utilized if theyreach the navigation companies, the shipmastersand the waterways administration in a timelymanner.To ensure this, a well-organized system for thecollection and transmission of information is indispensable.For instance, in Germany, use is made ofthe nautical information radio, or NIF. Such asystem is of particular importance on internationalrivers such as the Danube, which flows througheight countries. In conformity with the recommendationsof the Danube Commission, the datacollected in the Danube Basin are transferred dailyby telex. In order to avoid errors, internationallyagreed codes (see Volume I, Chapter 2) have beenadopted for data transfer. Announcements reachthe shipmasters partly by radio and partly in theform of daily hydrological bulletins.4.6.1.3 Navigation on lakes and canalsNavigation on lakes and canals differs considerablyfrom navigation on rivers:(a) The importance of the physiographic andhydrological regimes for ensuring navigationconditions is considerably lower becausecontrol structures provide stability of theseconditions;(b) On lakes and impoundments, the duration ofice cover is longer and hence the navigationseason becomes shorter;(c) While problems due to shallows are reduced orfully eliminated, problems caused by silting atheads of reservoirs or ship locks and in harbourbasins can arise locally;(d) Wind impact on navigation increases on lakesand impoundments;(e) There is a greater dependence of navigationoperations on the operation rules of locks andother structures.The safety of navigation on lakes and canals requiresan expanded range of observations:(a) On the shores of lakes and river impoundments,wind-measuring stations and warningfacilities should be established and operated;(b) In order to minimize siltation by technicalmeans, the amounts of sediment entering andleaving impoundments should be measuredsystematically to yield a sediment balance;(c) As barrages create favourable conditions forfrazzle-ice formation, regular observationsshould be carried out in the vicinity of thesestructures;(d) Automated stage recorders should be installedat the cross-sections that are particularly difficultfor navigation, for example weirs, inlets,and outlets.In order to be useful, these data must be checked forplausibility and documented, and should be sent tousers, such as shipmasters, in a timely manner.4.6.2 Classification of river trainingRiver training, river regulation and waterway maintenanceare continuous activities aiming to facilitatenavigation, protect riverbanks and riparian dwellers,and support flood control. Rivers in their naturalstate often change their beds and, in doing so, causedegradation of the channel and hinder navigation.The discharge of ice and floods show a differentiatedpicture in this case, depending heavily onriparian land uses and the availability of open land.River training strives to make the river form its ownbed with reasonably constant geometrical and

II.4-60GUIDE TO HYDROLOGICAL PRACTICEShydraulic conditions, but it also produces a numberof undesirable consequences of a socio-economicand ecological nature.Depending on the purpose to be served, river-trainingworks may be classified as high-water training,low-water training and mean-water training.High-water training, also known as flood-bed regulationand training for discharge, is aimed at therapid discharge of maximum floods. It is mainlyconcerned with the most suitable alignment andheight of marginal embankments for the dischargeof floods and may also include other schemes ofchannel improvement for the same purpose. Landuseregulations governing flood plains haveessentially the same goal as locally restricted floodcontrol measures, namely the discharge of floodswithout significant damage or loss of life.Low-water training is designed to provide minimumwater depth for navigation during thelow-water season. This is achieved by contractingthe width of the channel at low water and is generallycarried out with groynes. Low-water training isalso known as training for depth.Mean-water training or mean-bed regulation is themost important of all. Any effort to alter the rivercross-section and alignment must be designed inaccordance with that stage of the river at which themaximum movement of sediment takes place overa period of a year or more. Although high stages offlow lead to maximum bed activity, such stages aremaintained for a short duration; however, there islittle movement of sediment at the lower stagesthat persist for a large percentage of time. In betweenthe two, there is a stage at which the combinedeffect of forces causing sediment movement andthe time for which such forces are maintained is ata maximum. This stage, somewhere near the meanwater level, is the most important with regard toinfluencing the configuration of the river. Meanwatertraining is concerned with the efficientmovement of the sediment load of the river andmay therefore be called training for sediment.Mean-water training establishes the basis on whichthe former two are to be planned (Singh, 1989).Most commonly used river-training works includeguide banks, groynes or spurs and studs, cut-offs,revetments, vegetative protection, gabions andwalls.Figure II.4.32 offers a schematic overview of differentaspects of river morphology withmorphodynamic processes, including boundaryconditions, which show influencing factors andphysical processes.In addition to hydrological data, a great number ofother physical, geographical, morphological, meteorologicaland hydraulic data and/or relationshipsare required for the design and success of rivertrainingmeasures. The scope of this Guide does notpermit a detailed explanation of many of these variables.Here only the aspects with special relevanceto hydrology are discussed.4.6.3 Erosive forces due to channel flowIn a wide, straight channel exhibiting two-dimensionaluniform flow, the shear stress (τ) on the bedcaused by the flow is given by the followingequation:τ = γdS (4.13)where γ is the specific weight of water, d the flowdepth and S the water surface slope (see 4.8). Underuniform flow conditions, flow depth can be determinedby a flow resistance equation, such asColebrook–White’s or Manning’s. The energy slopeand water surface slope are equal to the bed slope.This in turn is generally fixed by topographicalcontrols. Bed shear stress thus varies with flowdepth and channel gradient and reaches a maximumat peak discharge.In a straight channel of finite width, the flowpattern and related velocity distribution are affectedby bank friction and the boundary shear stressvaries accordingly. Maximum shear stress occursbelow the maximum velocity filament and subsidiarypeaks in the zone of down-welling near thebanks. The maximum shear stress on the slopingbanks is generally about 0.8 γdS.Rivers tend to have three-dimensional flow patterns,owing to local variations in channel cross-section,such as pools and riffles, and plan geometry. Therelated secondary flow distorts the main velocityand shear stress distributions. In this case, equation4.13 gives only the approximate average shearstress.The secondary flow in a meandering channel causespeak shear stresses at the base of the outer bank ofthe bend. Measurements of shear stress distributionsin meander bends suggest that the ratio of maximumto average shear stress is a function of the ratio ofchannel width to curvature, bank roughness andthe presence of meander bends upstream (Apmann,1972). Maximum shear stress values can be up to

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-61three times the average in the upstream approach.Clearly, high local shear stresses at meander bendsproduce corresponding bank erosion and bedscour.Any localized feature, such as a bridge or weir, canadversely affect the general flow pattern and causelocalized erosion (Neill, 1973). The effects maymodify the velocity distribution to introduce athree-dimensional flow field or may increaseturbulence. Highly localized erosive activity islikely to remain downstream of the feature untilthe flow pattern has readjusted to that of thechannel.In general, it is not yet possible to model numericallythe boundary shear stress in strongly threedimensionalflows; neither is it generally feasible totake field measurements of design conditions.Therefore, physical models are commonly used toinvestigate the flow pattern and design parameters(CIWEM, 1989).Short description: Reshaping of the river bed through current action under considerationof the bed geometry, bottom substrate and sediment yield, includingimpacts through anthropogenic interventions and navigationSub-areas:AGeometricstructuresBBottomsubstrateCSedimenttransportRiverbedFloodplainsComposition Structure Bed loadSuspendedloadWash loadNavigationFloodprotectionColonization Stability Bed-formingprocessContaminantsInteractions through:DMorphodynamicprocessesChanges throughrelocation ofmaterial, erosionand aggregationCaused and influenced by:Boundary conditionsInfluencing factorsPhysical processes- Precipitation, temperature- Runoff hydrographs- Geology- Landscape or vegetation- Type of water body- Degree of river training- Constructive interventions- Maintenance activities- Dredging, dredgedmaterial dumping- Navigation- Land use- Water resourcesmanagement- Flow velocity- Bottom shear stress- Turbulences- Flow dynamics- Secondary currentsFigure II.4.32. Aspects of river morphology – a schematic overview(German Federal Institute of Hydrology)

II.4-62GUIDE TO HYDROLOGICAL PRACTICES4.6.4 Erosive forces caused by wavesand craftWave action sets up an unsteady flow field at thebank, which can cause erosion through a combinationof the following factors:(a) Shear stresses caused by run-up and downrush;(b) Direct impact of flow onto the bank;(c) Related seepage flow response in the bank tounsteady external boundary conditions.The water motion produced by a boat depends onthe size and geometry of the waterway, and theboat’s shape, size, speed and sailing line. Thecomponents of water motion can be divided intoprimary and secondary waves and the screw race.The effect of water level drawdown, together withwaves and the return current, can cause seriousbank erosion, particularly if the blockage factor ishigh. On a sloping bank, this often manifests itselfas a characteristic S shape in the bank profile ataround water level (CIWEM, 1989).In general, the erosive action of the screw race isminor compared with the above effects when thecraft is underway, but serious erosion can be causedwhen a craft is manoeuvring close to the bank orstarting off. Velocities caused by propeller actionare dependent on the propulsion system, installedengine power and duration of applied power(Prosser, 1986).Field measurements of the water-level drawdown,and waves and currents produced by passing craftare the best means of determining bank protectioncriteria. In the absence of such data, values can beestimated using the procedures described by PIANC(1987). Craft under 40 tonnes navigating in smallcanals and rivers in the United Kingdom canproduce waves of up to 0.4 metres high, butcurrents of up to 3 m s –1 can be produced (CIWEM,1989).are differences, most have the following points incommon:(a) One of the components of meandering is valleyfill with sediment movement;(b) Natural rivers strive to achieve or maintain astate of dynamic equilibrium;(c) The nature of meandering, the developmentdegree of bends and the frequency of theiroccurrence vary from river to river.The primary task of river training is to find an optimal,self-stabilizing river course that is adapted toits particular nature. The artificial bends should beselected so that a new dynamic equilibrium can beestablished. To do so, it is indispensable to studythe bends that are still in a natural state so as tobecome familiar with the river regime.The sinuosity of river bends can be characterized insimple terms as a series of circular arcs (seeFigure II.4.33). The following parameters must bedetermined:L – Arc length, as measured along the central line,between the two turning points;H – Bend length;A – Bend amplitude;R – Bend sinuosity or radius;α – Central angle of the river bend.Depending on the degree of its development, a riverbend can be:(a) A straight reach;(b) A false bend, when the straight line connectingthe two neighbouring turning points doesnot intersect the convex bank line, but remainsbetween the two bank lines;(c) A true bend, which in turn may be:(i) An underdeveloped bend, if in each ofthe two neighbouring inflexion crosssections,there is at least one point fromwhich that of the other section is visible;(ii) A developed bend, if 1.2 H < L < 1.4 H andα i< 120°.4.6.5 Evolution and characterization ofriver bendsLNatural watercourses generally tend to form irregularlyvarying channels in their flood flow beds andflood plains. This phenomenon is explained by thefact that each river is a system striving for dynamicequilibrium, in which one of the components ofchange – in addition to the river slope – is theformation of river bends or meanders.RHαARMany theories have been offered to explain thephysical reasons for meandering. Although thereFigure II.4.33. Definition sketch of river bendparameters

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-63The sinuosity characteristics of Figure II.4.33 can beplotted as a longitudinal profile or can be investigatedas random variables by statistical methods.The geometrical characteristics of the riverbed ineach cross-section are the following:(a) Area of the cross-section (F);(b) Cross-section width (B);(c) Wetted perimeter (P);(d) Hydraulic radius (R = F/P);(e) Average water depth (H = F/B).The geometrical characteristics of the riverbedchange both in time and along the river. On thebasis of periodic riverbed surveys, the geometricalcharacteristics can be investigated either as functionsof the water stage or, with relativefrequencies, of the various variables computedfor different river reaches. Figure II.4.34 is anexample showing the width variation of the crosssectionalong the Danube river downstream ofBudapest.4.6.6 Determination of design dischargesand stages4.6.6.1 Determination of the designdischarge for flood-bed regulationCharacteristic flood data can be determined andflood discharges, with various probabilities, can becomputed by using the methods described inFigure II.4.35. A groyne system on the river Elbe,at km 477. The gravel bank on the inside does notobstruct navigation.Chapter 5. The outputs of these computations arethe basic data necessary for selecting the designdischarge for flood-bed regulation.In present practice, the design discharge is given asa magnitude of a given probability, or a given averagereturn period, of the ice-free annual peakdischarges. The probability depends on the demographicand economic conditions of the area to beprotected.Width of water surface, B (m)750700650600550500450400350300Q = 1 000 ms –32500 10 20 30 40 50 60 70 80 90 100Relative frequency (per cent)Budapest – southern frontierBudapest – DunaföldvárDunaföldvár – SióSió – southern frontierQ = 2 500 ms –30 10 20 30 40 50 60 70 80 90 100Relative frequency (per cent)Figure II.4.34. Width variation of the cross-sectionof the Danube river4.6.6.2 Determination of the designdischarge for mean-bed regulationThe dimensions of the mean bed are related closelyto the flow and sediment regimes. Both regimes,and consequently, the evolution of the riverbed, areprocesses that are changing in time. The task is todetermine the effective, or design discharge thathas the greatest impacts on the natural and/orplanned dimensions of the riverbed. Figure II.4.35shows a groyne-fixed river flow, which should bethe natural line of a meandering river.Each of the geometrical parameters of the riverbedmay vary in a different manner, depending on theduration of the various discharges. Thus, onedischarge value will be the dominant one withrespect to the width of the mean-flow bed, whileanother will be dominant for its depth. For each ofthe geometrical parameters, a discharge value canbe found whose effect on that parameter will be thestrongest, but there will be no single discharge thatwill equally form all riverbed variables and optimizethem.

II.4-64GUIDE TO HYDROLOGICAL PRACTICESSince the sediment regime plays an important rolein riverbed formation or design, the characteristicsof sediment transport should be considered. See4.8.One method for determining the design dischargeQ Dat a given cross-section of a river may be appliedgraphically or numerically (see Figure II.4.36): thevertical axis of an orthogonal coordinate systemindicates water stage H(m) and the horizontal axis iscalibrated for four different variables – water stagefrequency f (m –1 ), flow discharge Q (m 3 s –1 ), averageflow velocity v (m s –1 ) and the product P = Δf Q v(m 4 s –2 ), where Δf is dimensionless (as Δ f = Δ f(H) =[m][m –1 ]). In this coordinate system, the curvesrepresenting the relationships Q(H), v(H) and f (H)are plotted first.While Q(H) and v(H) are generally concave curves,as shown in Figure II.4.36, f(H) is a more or lessasymmetric histogram, or bell-shaped curve,whose basis is the vertical H axis, and the areaenclosed between the f(H) curve and the H axis isunity. The H axis may then be subdivided into asufficient number of equally spaced intervals ofΔH (m) within the area between the minimum andmaximum water stages recorded. At the mediumstage H iof each interval Δ H i, the values Q i= Q(H i)(m 3 s –1 ), v i= v(H i) (m s –1 ) and f i= f(H i) (m –1 ) areread from the respective curves and the productsΔ f i= Δ H i...f iare computed. Finally, for each waterstage H i, the product P i= Q i. v i. Δ f i(m 4 s –2 ) iscalculated. This product is proportional to thekinetic energy of the flowing water, and the locationof the resultant P Dof the parallel (horizontal)powers, P iis determined, for example, by using thegraphical funicular polygon method or the numericalmomentum equation, both of which areH (m)Q iwell-known methods in statics. At the water stageH D, corresponding to the resulting power P D, therequired value of the design discharge Q D= Q (H D)can be read from the curve Q(H). The results thusobtained should be checked in river reaches thatare presumed to be stable.4.7 URBAN WATER RESOURCESMANAGEMENT[HOMS I26, I81, K22, K70]4.7.1 GeneralUrban water management is a broad term coveringthe management of water use, water conservationand impacts on the aquatic environment in urbanareas. Urban development has an impact on waterand the environment. Integrated urban watermanagement is the development of water facilitiesby using approaches that combine urban planningand sustainable development. As part of urbanplanning, integrated urban water management isrecognized as the most appropriate mechanism forproviding infrastructure and services for watersupply and the management of urban wastewaters,including storm water runoff.4.7.1.1 Water sources and impactsThe design, maintenance and management ofstorm drainage systems is highly dependent on theorigin of the water which, in an urban area, may beany of the following:(a) Runoff from upstream areas;(b) Runoff from adjacent areas;(c) Baseflow from groundwater;(d) Runoff from rainfall over the area considered;(e) Tides and surges;(f) Wastewater (sanitary, industrial and so forth).v iH iΔH if (H)f iΔf iv (H)Q (H)Flooding caused by runoff from rural areas or fromhigh groundwater levels is considered in otherchapters. Chapter 4 focuses on the design andmanagement of urban drainage systems to dealwith surface runoff from local rainfall and its interactionwith receiving water bodies.P i = Q i vi Δf iQ (m 3 s –1 )v (m s –1 )f (m –1 )Q (m 4 s –2 )Figure II.4.36. Graphical method used todetermine design dischargeMunicipal and industrial water supply and managementare related to urban drainage because they arethe source of polluted domestic and industrial wastewater.Daily variations in the quantity and quality ofwastewater from these sources need to be monitoredbecause they serve as inputs to the following tasks:(a) Drainage-system design, maintenance andrehabilitation;

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-65(b) Design and management of wastewater treatmentplants;(c) Assessment of the impacts of polluted andtreated water on receiving water bodies.The monitoring and management of groundwater inurban water areas are very important because of thevariety of ways in which human activities interactwith the balance and quality of groundwater, whichis often a major source of drinking water for urbanareas. However, groundwater recharge in urban areasis generally reduced because of the increased percentagesof impervious areas that cause lower infiltrationrates and faster surface runoff. Furthermore, groundwaterin urban areas is subject to pollution from bothpoint and non-point sources.4.7.1.2 Goals of integrated urban watermanagementThe goals of integrated urban water managementare as follows:(a) Provide good-quality water in adequate quantitiesand meet domestic and commercialpurposes under optimal economic conditions;(b) Minimize pollution and other adverse effectson the environment, including adverse groundwaterlevel changes;(c) Minimize the costs of floods and damage causedby storms through adequate storm drainagebased on the combination of improved drainagenetworks, real-time control of auxiliarystructures (retention and detention basins,pumping stations and the like) and warningsystems;(d) Minimize the adverse effects of treated oruntreated urban waters (domestic, industrialand storm) on receiving water bodies.Managing urban drainage systems to meet thesegoals involves the following tasks:(a) Evaluating the impact of urban developmenton the discharge and water quality of the basinunder alternative scenarios and for differentreturn periods;(b) Designing and implementing control measuresand storm-drainage practices to reduce theimpacts;(c) Implementing these measures through soundmanagement.4.7.2 Urban development impactsUrban drainage catchments differ from rural catchmentsin many respects:(a) Land-use patterns are different and generallybetter documented than in natural catchments;(b) The percentage of impervious areas is higher;(c) Unless special techniques for runoff reductionare applied, floods are generated rapidly withhigher peaks;(d) Water is drained from the catchments by acombination of surface collectors and undergrounddrainage systems;(e) Urban drainage basin areas are generally small,although in large metropolitan areas they canbe large, featuring complex systems of buriedpipes, pumping stations and, in recent years,large underground storage facilities.Urban development changes land use (seeFigure II.4.37), sharply increasing the percentage ofimpermeable area, such as roofs, streets and parkinglots. It also introduces man-made drainage,such as conduits and channels, which modify thehydrological cycle by increasing overland flow anddecreasing groundwater flow. Under this scenario,peak discharges increase (Figure II.4.37 (b)), as doesthe frequency of flooding. The higher flows fromurban surfaces can carry with them greater loads oftotal solids, such as sediments and garbage, andpollution, which then degrade the water quality ofthe receiving waters.Where an urban area is already developed, the solidsproduced in the basin come mainly from sedimentsand solid wastes washed from urban surfaces. Inthis case, total solids are a function of the frequencyof solid waste collection and the cleaning of streetA(a)(b)FlowA’Existing land useFuture urbanizationAfter urbanizationΔS – Flood level increase due tofuture urbanizationBefore urbanizationTimeSection A-A’Figure II.4.37. Hydrological impacts of landdevelopment: (a) Land-use change causes anincrease in flow depth and cross-section and(b) urbanization causes hydrograph change.ΔS

II.4-66GUIDE TO HYDROLOGICAL PRACTICESsurfaces, as well as hydrological factors such as thefrequency of rainfall events.On rainy days, the surface wash load is derivedmainly from litter and other surface contaminants.Table II.4.2 shows the variation of some water qualityparameters for different land uses, as measuredin cities in the United States.Many diseases can be traced to poor water management.In the humid tropics, diseases and symptomsrelated to poor water supply, sanitation and drainageinclude diarrhoea, cholera, malaria, dengue andleptospirosis. The environmental conditions relatedto drainage which help to spread malaria are stagnantwater, deforestation, soil erosion and flooding.Dengue is a disease found in warm climate which isspread by mosquitoes that live in clean, stagnantwater that may be kept in or near homes (tyres,vases, and so forth) during the rainy season. Pondsor on-site detention systems should be carefullydesigned and monitored in such climates to avoidmaintaining an environment favourable to thiskind of disease.4.7.3 Urban storm drainage designThe main design components of sewer systems aregutters, conduits, channels and detention or retentionelements. The hydrological design of theseparts is based on the calculation of the design maximumdischarge or the hydrograph which integratesboth the flood peak and volume. The methods usedin design are generally based on assumptionsregarding the rainfall-runoff relationship. There aretwo major methods:(a) The rational method, which estimates only thepeak discharge and assumes that the proportionof the rainfall that runs off is constant and thatthe rainfall intensity for any given duration isalso a constant. These are reasonable assumptionsfor small basins of less than 2 km 2 ;(b) Flood hydrograph estimation, which computesthe peak and volume of the flood event. This islikely to be important for reservoirs and takesinto account large urban basins.The main inputs of these methods in estimatingthe maximum discharge and its volume are designrainfall, land use within the upstream basin andother characteristics of that basin.4.7.3.1 Design rainfallStorms over urban areas are, as elsewhere, stochasticin nature. Therefore the design of the drainagesystem is based on storms of certain return periods.Rainfall depth for a certain return period is normallytaken from rainfall intensity–duration–frequencycurves that have been established for many cities(see 5.7). The choice of the return period for thedesign storm to be used as input for rainfall-runoffanalysis depends on the importance of the area tobe protected and the possible damage that might becaused by flooding.Rainfall intensity varies greatly from a temperateclimate to a humid tropical climate. Figure II.4.38shows the one-hour duration rainfall of 11 gaugesin a humid tropical region of Amazonia comparedwith the mean of the rainfall for gauges outsidethis region (subtropical and temperate). For thesame return period, the difference in rainfall intensityis about 25 per cent, which might translate intoa 50 or even 100 per cent increase in the designflood peak, depending on the design method used.Table II.4.2. Median event mean concentration for Nationwide Urban RunoffProgram, United States (Environmental Protection Agency, 1983)Constituent (mg/l) Residential Mixed Commercial Non-urbanBiochemical oxygen demand (BOD) 10 7.8 9.3 –Chemical oxygen demand (COD) 73 65 57 40Total suspended solids (TSS) 101 67 69 70Lead (Pb) 0.144 0.114 0.104 0.03Total copper (cu) 0.033 0.027 0.029 –Total zinc (Zn) 0.135 0.154 0.226 0.195Total Kjeldahl nitrogen (TKN) 1.900 1.29 1.180 0.965Nitrite (NO 2) and nitrate (NO 3) 0.736 0.558 0.572 0.543Total phosphorus (Tp) 0.383 0.263 0.670 0.121Soluble phosphorus (Sp) 0.143 0.056 0.080 0.026

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-67Managing urban drainage in such humid environmentsmay require the use of a more frequent returnperiod for design, with a consequently higher riskof inundation.4.7.3.2 Basin developmentThe size and level of development of an urban areamust be considered when modelling an urbandrainage system. In addition, the modelling willneed to be undertaken on a range of scales. Thesystem is generally a combination of minor andmajor drainage flow networks. The former servesthe drainage of small areas (≤2 km 2 ), such as sitedevelopments or condominiums, whereas majordrainage is composed of large trunk drains and/ormajor urban streams. The upstream basins of thesemain streams may include both urban and nonurbanareas.In developed urban areas, the drainage system iswell defined while, in undeveloped areas, naturaldrainage still works. When studying futurescenarios for basins in undeveloped areas, it isnecessary to derive the general outline of a futuredrainage system from the urban developmentplan.In countries where cities are not growing becausethe population has stabilized, which is the case insome European cities, future urban scenarios aremainly related to the improvement of existingdrainage and water quality. In most developingcountries, however, urban development isdynamic and often uncontrolled. It is very difficultto manage the potential impacts of thisdevelopment on the runoff in order to avoidP mm/h110100908070605040Latitude 20°S–30°S30Humid tropics20101 10 100Return period, yearsFigure II.4.38. Comparison of the mean maximumrainfall of one-hour duration in the humid tropicsand within latitudes 20°S and 30°S in Brazil(Tucci, 2001)environmental degradation and increased flooddamage.Any computation of discharge in urban basins mustbe based on existing major and minor drainagesystems, plus an analysis of likely or planned futuredevelopment scenarios. The computation of adesign peak discharge for small drainage areasgenerally involves the rational method, despite itslimitations with regard to the spatial and temporalvariability of hydrological processes (Heaney andothers, 2002). The approach to be used for themajor components of the drainage system willdepend on whether the basin is developed orundeveloped.Undeveloped or developing areas: In a basinwhich is currently undeveloped or is subject toincreasing urbanization, undeveloped areas willnot have intricately planned streets and minordrainage systems, but will have an urban planbased on population density in the form of anurban master plan. Empirical relationshipsbetween population density and impervious areascan be developed to help set a design figure forthe percentage of impervious areas (AI). Such arelationship has been derived for three majorBrazilian cities, São Paulo, Curitiba and PortoAlegre (Campana and Tucci, 1994), which resultsin the following equation:AI = 0.489 DH (4.14)where DH is urban density expressed as inhabitantsper hectare. This relationship is valid in the Braziliancontext as long as DH < 130 inhabitants/hectare.Above this density, an impervious area of 65 percent is assumed. It was developed for areas greaterthan 2 km 2 since for smaller areas there could besome distortion.4.7.3.3 Design peak flowThe design peak flow can be estimated by using thefollowing procedures:(a) Flood frequency based on a flow series ofadequate length;(b) Empirical equations based on a regional floodfrequency analysis;(c) Design rainfall fed into a rainfall-runoff modelin order to estimate the discharge.In (a), stationary and representative peak flowsamples are needed. Such data are not alwaysavailable and there can be difficulties in achievinga stationary sample of flow events because ofthe continuing urbanization of the basin. As

II.4-68GUIDE TO HYDROLOGICAL PRACTICESregards (b), empirical equations must be developedfor specific regions based on the regionaldata. It is generally recommended that theseequations not be used outside the region inwhich they were developed. The rainfall-runoffprocedure, (c), is the method most used in estimatingthe peak and hydrograph of rainfall witha selected return period. This approach offerssimple methods which are used for small basins.These compute only the peak flow, as is the casewith the rational method described below. Othermethods attempt to estimate both the peak andthe flow distribution in time, thus yielding adesign hydrograph. These are also described morefully below.4.7.3.3.1 The rational methodIn small basins, the simple rainfall–peak flow relationshipknown as the rational method may beused. This uses the following simple equation forthe peak flow:Q = 0.278 . C . I . A (4.15)where Q is the discharge in m 3 s –1 , C is the runoffcoefficient; I is the rainfall intensity in mm/h and Ais the basin area in km 2 . The rainfall intensity isselected according to the return period T (generally2–10 years in minor drainage systems) and the rainfallduration t. T is based on the design decision andthe characteristics of the system modelled. In therational method, t is equal to the time of concentrationof the basin.Time of concentration (t c) is the sum of the time ittakes for the water to flow over the basin surfaceuntil it reaches the inlet (t b) and the travel timethrough conduits and natural and constructedchannels (t r):t c= t b+ t r(4.16)The value of t bcan be estimated by empiricalequations developed for surface flow. The flow isgenerally less than 60 m long; if longer, it tendsto be concentrated in a swale, gutter or a smallnatural channel. It can be estimated by:H (m)(4.17)where t bis in minutes, C 5is the runoff coefficientused for a five-year return period, L is the length ofthe overland flow in metres and S is the averagebasin slope in per cent (SCS, 1975).The travel time through the system of natural andartificial canals and conduits can be calculated byestimating its velocity using the Manning equationso that:n X it r = ∑V (4.18)i=1 iwhere X iand V iare the length of reach i of thesystem and velocity through it and n is the numberof reaches.Runoff coefficient: This coefficient (C) is presentedas the ratio of the total overland flow to the totaldesign rainfall over the basin. It is a function ofrainfall intensity, the spatial and temporal distributionof the rainfall, the extent of urbanizationand soil characteristics, among other factors. Theevaluation of a mean value of C for a drainage areais a highly simplified but pragmatic representationof its water balance and the impact ofurbanization.In design, this coefficient is estimated using tablespresented in the literature (ASCE,1992) such asthose reproduced in Tables II.4.3 and II.4.4. Thecoefficient can be modified for return periodsgreater than ten years multiplied by the coefficientCf presented in Table II.4.5.The runoff coefficient of a basin can be estimatedfrom the proportion of pervious andTable II.4.3. Normal range of runoff coefficients(ASCE,1992)Surface characterRunoff coefficient CPavement– Asphalt and concrete 0.70–0.95– Brick 0.70–0.85– Roofs 0.75–0.95Lawns, sandy soil– Flat (2%) 0.05–0.10– Average (2–7%) 0.10–0.15– Steep (>7%) 0.15–0.20Lawns, heavy soil– Flat (2%) 0.13–0.17– Average (2–7%) 0.18–0.22– Steep (>7%) 0.25–0.35Note: Ranges of C values are typical for return periodsof 2–10 years.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-69Table II.4.4. Typical composite runoff coefficientby land use (ASCE,1992)Description of the areaBusinessRunoff coefficient C– Downtown 0.70–0.95– Neighbourhood 0.50–0.70Residential– Single family homes 0.30–0.50– Multi-units, detached 0.40–0.60– Multi-units attached 0.60–0.75– Residential (suburban) 0.25–0.40– Apartments 0.50–0.70Industrial– Light 0.50–0.80– Heavy 0.60–0.90– Parks, cemeteries 0.10–0.25– Playgounds 0.20–0.35– Railroad yards 0.20–0.35– Unimproved 0.10–0.30Note: Ranges of C values are typical for return periodsof 2–10 years.Table II.4.5. Flow coefficient correction factor(Wright-MacLaughin Engineers, 1969)Return period in years Coefficient Cf2–10 1.0025 1.1050 1.20100 1.25series and the values of the coefficients may beunderstood as relating to a two-year return period(Urbonas and Roesner, 1992). In this equation, if AIis considered to be 1.0, then C is 0.95, that is, animpervious coefficient of 0.95 with losses of 5 percent. This could be caused by depression storage,evaporation from warm surfaces, antecedent moistureconditions, infiltration at surface junctionsand data uncertainties.In Brazil (Tucci, 2001), 11 basins have been usedwith areas ranging from 3.4 km 2 to 106 km 2 and1 to 51 per cent of impervious area, resulting in:C = 0.047 + 0.9 AI (4.21)with an R 2 of 0.92. The latter equation has coefficientsvery close to those of 4.20.This coefficient changes in each flood, dependingon the rainfall intensity characteristics and initialsoil moisture conditions. In a rural basin, C pmayvary greatly; therefore, it is important to recognizethat these equations were developed according tomean values and represent only the mean conditionsfrom the recorded data.Figure II.4.39 shows the runoff coefficients formany events for two scenarios in the Diluvio Basin:1979–1982, with 19.7 per cent of impervious areas;and 1995–1997, with 40 per cent of imperviousareas. This demonstrates that there is a relationshipbetween the runoff coefficient and peak discharge:both rise with increasing rainfall.0.700.60impervious areas. A weighted coefficient can becalculated by:C = C p+ (C i– C p)AI (4.19)C0.500.400.30where C pis the coefficient for pervious areas and C iisthe coefficient for impervious areas; AI = A i/A trepresentsthe ratio of impervious areas to the total area.The application of this type of equation to 44 smallurban basins in the United States produced thefollowing relationship (Schueler, 1987):0.200.100.0079–8296–9720 40 60 80Peak flow m s –3C = 0.05 + 0.9 AI (4.20)with a correlation coefficient of R 2 equal to 0.71.The hydrological data used were from two-yearFigure II.4.39. Runoff coefficient (C) for twoscenarios: urbanization of 1979–1982 and urbanizationof 1995–1997. Diluvio Creek in Porto Alegre(Santos, 1998)

II.4-70GUIDE TO HYDROLOGICAL PRACTICESQuantitative aspects of determining urban runoffare discussed in more detail in 5.10.4.7.3.4 Hydrograph methods: storm watersimulation modelsThese methods are based on rainfall-runoff modelswhich calculate the flood hydrograph from therainfall for a selected return period and certain timeand space distribution. In addition to the rainfall,the initial state of the model variables, modelparameters and other basin characteristics mustalso be known. The return period of the flood isgenerally assumed to be the same as that of therainfall. However, approaches can be used to allowthe values of other inputs, such as rainfall lossesand temporal patterns, to vary stochastically.These models generally contain two major modules:a hydrological module and a hydraulic module. Thehydrological module is used to calculate the overlandflow volume in time. The hydraulic module isused to calculate the transport of this volumethrough streets, conduits, channels and reservoirsin time and space.The hydrological module employs the followingtools:(a) Coefficients, as in the rational method;(b) Infiltration equations such as those of Hortonneand Green and Ampt;(c) Empirical relationships, such as those developedby the United States Department of AgricultureSoil Conservation Service (SCS), now knownas the Natural Resources Conservation Service.The first group of methods is biased whenthe model is applied to a magnitude of rainfallthat is for larger or smaller than that used inits development. The second and third groupsare more reliable and found in models such asthat of the SCS (SCS, 1975) and HEC-1 (Feldman,1995). The main simplifications of thesemodels include a uniform space distribution ofparameters and rainfall in each sub-basin.The hydraulic module can be represented by thefollowing types of equation:(a) Storage and kinematic wave transport equations,which feature two major simplifications:they are used for free surface flow in pipes andchannels, but do not take into account backwatereffects, which are very common in urbanenvironments;(b) Diffusion and hydrodynamic equations forfree surface flow: this type of model takes intoaccount the backwater effects but cannot beused for flow under pressure, which occurswhen the flow is greater than the designconditions;(c) Hydrodynamic equations for systems with pipeflow under pressure and free surface otherwise.This model is mainly used for the simulationof flood scenarios or events above the designconditions.These models aim to reflect differences in imperviousareas across the sub-basins, overland flowcharacteristics, different times of concentration forsub-basins and flood-routing effects through themain channels and streams. Where urban developmentis dynamic over time, the model should beused to evaluate the impact of changes in urbandensity derived from future planning scenariosusing relationships between impervious areas andurban density (Tucci, 2001).Examples of models used for this purpose are Mouse(DHI, 1990), Hydroworks (HR Wallingford –Wallingford Software) and storm water managementmodel, or SWWM (Huber, 1995).Information on hydrological and urban characteristicsis required to estimate model parameters andreduce planning and project uncertainties. Duringthe 1970s and 1980s, there were significant advancesin the methods or procedures for measuring rainfalland runoff (Maksimovic and Radojkovic, 1986),which have enabled the development and calibrationof complex, often physically based, models forrainfall and runoff analysis and storm drainagesystem design (Yen, 1986). Although the drainagesystems are generally designed to provide floodprotection from storms of a specific probability,most of the present-day models can simulate theconsequences of surcharged flow combined withsurface flow on the streets (open-channel flow).More on such models can be found in 5.10.5.Most urban studies in developing countries must beperformed without the use of recorded hydrologicaldata because the relevant data are either difficult toobtain or do not exist. Therefore, there is an urgentneed for improved hydrological data collection inurban environments, especially in the humid tropics.Without these data, model parameters maypresent great uncertainties, which may result inhigher urban drainage construction costs due to theoversizing of infrastructure, or costs associated withflooding caused by undersized drainage.4.7.3.5 Water qualityWater quality models generally have a quantitativemodule simulating the discharges resulting from

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-71rainfall, and a water quality module simulating thevariation in water quality as expressed by parameterssuch as biochemical oxygen demand, nitrogenand phosphorus. The water quality module generallyinvolves the following steps: pollution loadevaluation, and transport, retention and control ofthe pollutant. Some of the models that have waterquality components include SWWM, Mouse andStorm (HEC, 1977). The main difficulty with waterquality simulation and evaluation is the lack ofobserved data for fitting model parameters.Consequently, validation is generally based oncomparison with published information from elsewhere.Uncertainty analysis can be used to betterunderstand the limits of the impacts and controlmeasures needed to help make decisions for themanagement of urban developments.4.7.4 Urban drainage control measuresThe main goals of urban drainage are to decreasethe frequency of flooding and improve water quality.Urban storm water management is mainlyconcerned with the distribution of the volume ofwater in time and space within the urban basin,taking account of urban development, hydraulicnetworks and environmental conditions (Urbonasand Stahre, 1993).The key control measures are either structural ornon-structural.Structural measures: works designed to control theimpact of floods on a major drainage system withina given urban development scenario. They aregenerally channel improvements and retentionponds.Non-structural measures: land use and other regulationsdesigned to limit the threat of flooding andflood warning, including the real-time forecastingof rainfall and of the likely impact of the forecastflood. Urban drainage regulations can be used tolimit peak discharge downstream and reduce thedegradation of water quality, taking into accountsocial and economic conditions. Basic features ofthis type of regulation are to keep the peakdischarge from the new development equal to orbelow the pre-development scenario and to setlimits on impervious surfaces in each development.Public participation is essential for thedevelopment of effective regulations, whichshould include awareness-raising and educationalprogrammes.Source control measures for new developmentshave been included in the regulations of manycountries (Urbonas and Stahre, 1993). Sourcecontrol involves the provision of measures for storagenear the location of the source of runoff,decreasing the need for conveyance increase downstream(Urbonas and Stahre, 1993). Some sourcecontrol facilities are permeable pavements andparking areas, infiltration basins and trenches.4.7.5 Urban drainage management4.7.5.1 PrinciplesExperience gained from urban drainage planning inmany countries has led to the establishment ofsome general urban drainage management principles(Urbonas and Stahre, 1993):(a) Management should be based on an urbandrainage master plan for the municipality;(b) Public participation in urban drainage managementshould be increased;(c) Urban drainage control scenarios should takeaccount of future city developments;(d) Urban drainage development should be basedon cost recovery for investments;(e) An evaluation of flood-control measures shouldbe undertaken for the whole basin, not only forspecific flow sections;(f) Flood-control measures should not transfer theflood impact to downstream reaches but shouldgive priority to source control measures;(g) More emphasis should be given to non-structuralmeasures for flood-plain control, such asflood zoning, insurance and real-time floodforecasting;(h) Steps should be taken to reduce the impact ofurban surface wash-off and other related urbandrainage water quality problems.In many developing countries, urban drainage practicesdo not fulfil these principles. The main causesare the following:(a) Urban development in the cities occurs too fastand unpredictably. Generally this developmentstarts downstream and moves upstream whichincreases the potential for negative impacts;(b) Peri-urban areas are generally developed withouttaking into account the city’s regulations,or there are no city regulations;(c) Peri-urban and risk areas – flood plains andhillside slopes – are occupied by low-incomefamilies and have no established infrastructure.Spontaneous housing development in riskyareas in the humid tropics may be found inthe following cities: on land prone to flooding– Bangkok, Bombay, Guayaquil, Lagos, Monrovia,Port Moresby and Recife; on hill slopesprone to landslides – Caracas, Guatemala City,

II.4-72GUIDE TO HYDROLOGICAL PRACTICESLa Paz, Rio de Janeiro and Salvador da Bahia(WHO, 1988);(d) Lack of appropriate garbage collection anddisposal leads to pollution of the water andclogging of the drains. Some African countrieshave no urban drainage, and when systematicdrainage does exist, it is often filled withgarbage and sediments (Desbordes and Servat,1988);(e) Lack of institutional organization as a basis fordeveloping urban drainage at the municipallevel, hence no power of regulation, no capacity-buildingand weak administration.4.7.5.2 Management practicesThe main urban drainage policy requirements maybe summarized as follows:(a) Regulation should ensure that urbanization willnot allow flood flows of a given return period toincrease within the basin;(b) Urban space should be reserved for detention(Figure II.4.40) or parks built within the riverboundaries for storing flood volumes, sedimentand trash detention and water quality improvement.If some of the impact of upstream urbanizationcannot be controlled due to a lack oflaw enforcement, urban drainage policy maybe used to limit to a minimum the transferdownstream of the impact. Instead of havingthe solid waste and sediments distributed inconduits or along the rivers and channels, theycan be retained in specific places for cleaning,reducing maintenance costs. However, this isnot always the best solution; therefore, eachcase should be evaluated on the basis of localconditions. Further guidelines for this formof integrated land and water management inGreen areaDetention and parksFigure II.4.40. Detention for urban drainagecontrol, planning stage (Tucci, 2001)urban areas have been prepared, for example,by Lawrence (2001).(c) When the solution for flood control in themajor drainage system is the use of stormsewers or increased channel capacity, the planor design has to evaluate and limit the downstreamimpact of storm sewers or expandedchannel capacity.Based on these principles, urban drainage managementshould incorporate the following features:(a) Prevention: planning urban space by takinginto account urban drainage flood-plain areasin city development. Source control and nonstructuralmeasures are the main choices at thisplanning stage;(b) Permanent institutional elements: regulationof minor drainage taking into account theincrease in peak flow; regulation of land use inflood plains; tax incentives to protect conservationareas and existing drainage control areas;public procedures to check and enforce regulationsbased on local conditions; increased lawenforcement at the site level when the area isalready partially developed;(c) Capacity-building: improve the technicalcapacity of local and state government personnel;create better working conditions so thatskilled professionals can remain on the job;production of a city urban drainage manual;operation of a technical education programmefor architects and engineers; general educationof the population regarding relevant issues;(d) Public participation: use public opinion pollsas part of a campaign to involve the generalpublic in the planning of urban drainage facilities,taking into account local requirements;consult the public through representatives ofnon-governmental organizations with regardto urban drainage plans and projects at allstages of development; increase public awarenessof the impact of urbanization on urbandrainage;(e) More hydrological data: the lack of adequatehydrological and physiographic data is achronic problem in the urban areas of developingcountries resulting in the design of projectscharacterized by high cost or underperformance.A programme of data acquisition anddevelopment of methodologies for the use ofdata in the production of information for urbandrainage is essential for sound urban drainageplanning;(f) Impact control: Structural measures for urbanflood control may be developed sub-basinby sub-basin so as to decrease the impacts ofurbanization with regard to water quantity

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-73and quality. In the planning process, rainfallrunoffand water-quality models can be used toassess the efficiency of the controls measures.The associated costs are generally distributed tothe basin population based on the imperviousarea of their property.4.7.6 Remote-sensing estimates for landuseRemote-sensing techniques play an important rolein urban drainage design, particularly for the estimationof land use. This is discussed in Volume I,Chapters 2 and 4.4.8 SEDIMENT TRANSPORT AND RIVERCHANNEL MORPHOLOGY[HOMS I09, K65]4.8.1 GeneralThe transport of sediment by water flowing in riversand channels is an important factor in the planning,design and operation of water managementprojects. It affects the life of storage reservoirs, thestability and conveyance of river channels, thedesign of structures that are in contact with theflowing water and the suitability of the watersupplies for various uses. A proper assessment of theeffects of sediment transport, and of the measuresthat may be necessary for its control, require knowledgeof the processes of sediment erosion,transportation and deposition, and of their interactionwith the hydrological processes in thecatchments concerned.This section is devoted to these erosion and sedimentationprocesses and their role in determiningriver channel morphology, while 4.10 addressesquestions concerning hydroecology, in which channelmorphology is a key factor.4.8.2 Catchment erosionAgents of erosion include wind, ice and gravity, butthe most efficient one is running water. The processesby which water degrades the soil arecomplicated and depend on rainfall and soil properties,land slope, vegetation cover, agriculturalpractices and urbanization. The last two factorsaccount for the most important effects of man’sactivities on erosion.Empirical equations have been developed for thedetermination of soil loss or sheet erosion fromagricultural lands. One was developed by Musgravefor conditions prevailing in the United States(Chow, 1964). It was subsequently amended toapply to a wider range of conditions to yield theUniversal Soil Loss Equation. This was then furtherdeveloped to include erosion caused by constructionand building. The result is the Revised UniversalSoil Loss Equation:A = R . K . LS . P . C (4.22)where A is soil loss in tonnes per hectare per year, Ris a rainfall erosivity factor; K is a soil-erodibilityfactor; LS a topographic factor composed of L, afactor dependent on the length of the slope, and S,the slope of the land surface; P is a conservationpracticesfactor; and C is a cover factor. Each factoris evaluated by using maps and tables derived fromempirical data for the particular location andconditions.Bare land and badlands may develop gullies withrates of advance that can be computed by empiricalformulae containing parameters such as the drainagearea of the gully, approach channel slope,rainfall depth and clay content of the eroding soil.4.8.3 Channel erosionChannel erosion is caused by the forces of theconcentrated flow of water. Its rate depends on thehydraulic characteristics of channel flow and theinherent erodibility of channel materials. In noncohesivematerials, the resistance to erosion isaffected by the size, shape and specific gravity ofthe particles and the slope of the bed. In cohesivematerials, it also depends on the bonding agents.The relationships between the hydraulic variablesand the parameters influencing the erodibility ofchannels are not fully understood and are oftenexpressed by empirical formulae (Chow, 1964),(Maidment, 1992). Stream and river control workscan accelerate channel erosion locally if they causean increase in channel depth or flow velocity,change the direction of the flow, or reduce the naturalsediment load. The latter effect occurs frequentlybelow dams and may persist for many kilometresdownstream. Procedures for measuring and computingbed material, suspended sediment dischargeand sedimentation are discussed in Volume I, 5.5.4.8.4 River systemsRivers are formed along more or less defined channels,draining from the land the water that runsoff from precipitation and from the melting ofsnow at high altitudes. They developed over the

II.4-74GUIDE TO HYDROLOGICAL PRACTICESages. Along with water, they also convey sediment,washed down from the catchments and erodedfrom their own bed and banks. River systems andriver processes are complex. For example, theinputs to a river reach are the water and sedimentdischarge, and the primary responses are thewidth, depth and velocity of flow, sedimentdischarge through the reach and the rate of sedimentand water storage, which could be plus orminus, in the reach under consideration. Bedroughness and friction factors may be regarded assecondary factors; their values are interrelatedwith the depth and velocity of flow, sedimenttransport rate and to some extent, with the rate ofscour and deposition.Over geological time, a river evolves in such a waythat it can in the long run transport the sedimentdelivered to it with available water runoff. Mostnatural channels are considered to be in regimeflow when the major dimensions of their channelsremain essentially constant over an extended periodof time. The condition of regime flow does notpreclude the shifting of channel alignment byerosion and rebuilding of the banks, but it requiresbalance between these factors. It requires that thesediment discharged from any given reach be equalto that which is introduced into the next reach.However, this does not mean that there is an invariantrelationship between sediment discharge andwater discharge. For most mobile bed streams, therewill be a range of discharge values within which thestream can adjust with as much as a tenfold variationin sediment discharge by variation in bedforms – ripples and dunes. A concurrent variation isflow depth and velocity, without any appreciablechanges in slope, channel width or average bedelevation. A stream may vary its channel dimensionslocally, in time or space, without interferingwith regime flow as long as these variations fluctuatearound a balanced average. Indeed, this is howa river adjusts downstream of a confluence with atributary which has different characteristics of sedimenttransport.change until equilibrium is established betweensediment inflow and discharge.4.8.4.2 River channel patternsThere are three main types of river channel patterns:straight, braided and meandering. These characteristicsare considered from a plan view. Numerousfactors influence whether a stream takes one formor other and their relationships are still notcompletely understood.4.8.4.3 Straight channelsStraight channels are those that have a straightalignment. They generally occur when a channelslope is similar to a valley slope, or where steepslopes produce relatively high velocities. In thelatter case, it is possible that the straight alignmentresults primarily from momentum that discouragesturning.4.8.4.4 Braided channelsA distinctive characteristic of braiding is multiplechannels. There are, however, two types of multiplechannel stream: one is the interlaced multi-channelstream separated by islands at low stages giving theappearance of braided hair. At high-flow stages, theislands may become submerged and the streammay flow from high bank to high bank(Figure II.4.41a). Another type of multiple channelstream is the distributary type found in deltas ordebris cones. (Figure II.4.41b) These are generally(a)4.8.4.1 Aggrading and degrading streamsIn certain sections of some streams where theamount of sediment introduced exceeds that whichthe stream can transport, the excess must be deposited.The stream bed is thus built up or aggraded.Conversely, if the rate at which sediment is introducedto a stream is less than its transport capacity,and its channel bed and banks are erodible, thestream will erode bed and banks to supply the deficiency.The major dimensions of aggrading ordegrading channels remain in a constant state of(b)Figure II.4.41. Braided channel (a) and distributarychannel (b)

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-75aggrading channels that divide to follow separatecourses, which finally disappear into sheet flow atthe coast.4.8.4.5 Meandering channelsMeandering channels follow a winding or tortuouscourse. They tend to shift continuously by localerosion and rebuilding of banks. Most problemsarising in channel control concern meanderingstreams, as bank erosion occurs frequently withthese. As solutions to these problems depend onknowledge of channel characteristics, numerousauthors have studied and produced voluminouswork on meandering rivers.The basic stream meander is essentially a sinusoidalcurve as shown in Figure II.4.42a. It is adynamic form, tending constantly to shift its positionby erosion of the concave bank and depositionalong the convex bank of the bends. Under idealconditions, a meander system will migrate downstreamin an orderly progression along a centralaxis (Vanoni, 1975). The primary dimensions of ameandering system are its length, width and tortuosityratio – also known as its sinuosity. Fiveprimary factors determine these dimensions: valleyslope, bank full discharge, bed load, transverseoscillations and degree of erodibility of thealluvium.Ideal meandering systems seldom exist in nature.Individual meanders and overall systems ofnatural meandering streams tend to becomeMeander widthLoop cut-offMeander length(a)(b)SucessivechannelpositionsResistant bankOxbow lakeChute cut-offFigure II.4.42. Meander channel (a) anddeformed channel (b)distorted. A typical meandering streamFigure II.4.42b) is formed of numerous irregularbends of varying size and shape that resemble anideal meander pattern only in respect to thealternating direction and continuing migrationof the bends.4.8.5 Flow regimes and bed formsWhen the average shear on the bed of an alluvialchannel exceeds the critical shear stress for the bedmaterial, the material forming the bed starts tomove, thereby disturbing the initial smooth bed.The nature of the bed and water surface change asthe characteristics of flow and sediment change.The types of bed and water surface are classifiedaccording to their characteristics and are calledregimes of flow (Garde and Ranga Raju, 2000). Bedform has an effect on flow resistance, sedimenttransport and turbulence.4.8.5.1 Process of bed formsUndulation and deformations of the mobile bed ofchannels are called bed forms. According to oneschool of thought, a small disturbance on aninitially flat bed can, under certain conditions,affect the flow and local transport rate of sedimentleading to the formation of troughs and crests. Beddeformation caused in that manner accentuates thedisturbance, which in turn increases the rate oflocal scour in troughs and deposition over crests,leading to the formation of ripples and dunes. Thegrowth of bed forms thus continues until a stage isreached when factors associated with the increasedsize of bed forms intervene and limit furthergrowth.Ripples and dunes thus achieve their optimum size.This is known as a lower regime flow and beginswith the start of motion. The resistance to flow islarge and sediment transport is small. The bed formis either ripples or dunes or some combinationthereof. Resistance to flow is caused mainly by theform of the roughness. Plane bed, ripples and dunesare the bed forms in this range.Under certain other conditions, the local sedimenttransport rate works in such a way that the size oftroughs and crests is diminished, leading to theestablishment of a flat bed. This is called upperregime flow. In upper regime flow, resistance to flowis relatively low and sediment transport is high. Theusual bed forms are antidunes and a plane bed.Hence it is desirable to know beforehand whatregime would prevail in a stream for a known flowcondition.

II.4-76GUIDE TO HYDROLOGICAL PRACTICESThe transition zone encompasses the bed formsthat occur during the passage from lower regime toupper regime. This transition is not unique in looseboundary hydraulics. Figure II.4.43 shows bedforms arranged in increasing order of sedimenttransport rate. The processes concerned are describedbelow (Simons and Richardson, 1961; Van-Rijn,1984).4.8.5.2 Plane bedWhen average shear stress on the bed is less thanthe critical shear for material on the bed, no movementof material occurs on the bed. This regime iscalled plane bed without motion of sediment particles,and the laws of open channel flow on a rigidbed would be applicable in such a case.4.8.5.3 RipplesWhen the flow and hence shear on the bed areincreased, the bed deforms in small three-dimensionalundulations called ripples. They are triangularin shape with a flat upstream shape and steep downstreamface. With ripples on the bed, materialmoves primarily along the bed as bed load. Water isclear and the water surface is choppy. Ripples moveslowly in a downstream direction.4.8.5.4 DunesAs the discharge is further increased, ripples growin size and are called dunes. Dunes are also triangularin shape but much larger than ripples. Flowseparates behind each dune and produces a wakeand high turbulence. As a result, there is greaterenergy loss and further material is thrown intosuspension. Dunes also move in a downstreamdirection. The Froude number is much less thanunity. The water surface appears to boil. This regimeis known as ripples and dunes.4.8.5.5 Transition regimeIf the discharge is increased further, the dune lengthgrows, height decreases and the dunes are partlywashed out. The Froude number is 0.8 or close tounity. Water surface waves are in phase with thepartly washed-out dunes. This is known as a transitionregime and is generally very unstable.(a) Typical ripple pattern, Fr

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-774.8.5.8 Boundary roughness caused by bedformThe effect of bed form is to increase boundaryroughness. Variation in the roughness coefficient(n) in the Manning formula (see Table II.4.6) variesfrom 0.010 to 0.030 depending on flow regime,which means that lower regime flow (Fr = 0.15 to0.37) would have an n value in the range 0.01 to0.013. Rivers change their bed form during floodsand hence their roughness coefficients. It is reportedthat the Darcy–Weisbach friction factors for ripplesand dunes bed are 4.5 to 8.7 times greater thanthose for flow over a flat bed with immobile sandgrain roughness. If dunes change to transition orflat bed, however, the roughness coefficient candiminish.Table II.4.6. Variation of Manning’s roughnesscoefficient n with different bed forms (Simons andRichardson, 1961), expressed in m –1/3 sBed form Regime ApproximateFroudenumber (Fr)Approximaten valueRipples Lower 0.14~0.37 0.018~0.30Dunes Lower 0.28~0.65 0.020~0.040Transitions Transitions 0.55~0.92 0.014~0.030Flat bed Upper 0.70~0.92 0.010~0.030Antidunes Upper >1.0 0.010~0.030The roughness coefficient n is analogous to the frictionfactor f in the Darcy–Weisbach formula:V =⎛⎝8gf0.5⎞ 0.5(R⎠ hS)(4.23)where V is mean velocity, g is acceleration due togravity, R his hydraulic mean radius and S is channelslope.A relation between friction f and the roughnesscoefficient n is given by:n = µ . f 1/2 . R h1/6(4.24)where R his hydraulic radius [m] and µ = 18g≅ 0.113 (for the SI system).This results from the identity:VV *=⎛ 8⎝ f0.5⎞⎠= R 1/6hn g=Cg(4.25)where C is Chezy’s coefficient, V is flow velocity,and V * = g ⋅ R h⋅ S is the shear velocity.4.8.5.9 Resistance relationshipRivers are dynamic in nature and their behaviour isaffected by natural events such as earthquakes,landslides or changes in the local climate. Theimportance of bed form in discharge measurementsis well known as it affects alluvial roughness andthereby the resistance relationship. Methods topredict alluvial roughness are available in the literature(Yalin and Ferreria da Silva, 2001).4.8.5.10 Prediction of bed formsCriteria have been proposed by several investigatorsfor ascertaining the kind of bed form such asripples, dunes, transition, flat bed and antidunes.Most of these criteria are however based on flumedata, and certain difficulties arise when they areapplied to rivers.Depths and velocities across rivers are rarely uniformand hence bed forms observed in different parts ofthe sections can also be different. Owing to thethree-dimensional topography of bed forms, theircharacterization is possible only in a statistical senseand it may be necessary to use spectral density functionsto obtain a meaningful representation of thegeometry. Bed forms take some time to change as aresult of changes in discharge and always lag behindby a certain time. In addition, the type of bed formis dependent, not only on depth, slope and bedmaterial size, but also sediment supply. In a flumewith a low sediment supply, the bed form increasesin size from ripples to dunes with increase in velocity.However, in a river with increase in discharge,velocity increases along with the sediment concentration,resulting in a reduced bed form size androughness coefficient. The variation in Manningco-efficient n in a flume with increase in dischargeis not comparable to that in river. Because of thesedifficulties, the current advice is to observe the bedform in a river at the required location by means ofan echo sounder and to repeat these observations atdifferent flood stages.4.8.6 Transportation of sediments inchannelsWhen the shear stress on the bed exceeds the criticalshear stress for the given material, material onthe channel bed starts moving. Depending on thegradation of channel bed material, sediment istransported near the bed by contact, saltation or insuspension. Sediment transport, sediment load and

II.4-78GUIDE TO HYDROLOGICAL PRACTICESsediment discharge are commonly used terms inriver-engineering, which divides the sediment loadinto three: bed load, suspended load and total load.The term wash load is also used. This is related tocatchments and is composed of particles whose sizeis finer than those found in the stream bed. For adetailed discussion on the following formulae andother sediment transport equations, please refer tothe Manual on Sediment Management and Measurement(WMO-No. 948). Various measurement techniquesare discussed in the Manual on OperationalMethods for the Measurement of Sediment Transport(WMO-No. 686).4.8.6.1 Suspended sediment transportFine, or suspended, sediments transported inrivers, originate mainly from the topsoil of thecatchments and from the banks of the channels.However, fine sediments also originate fromsewage and other return flows. Such sedimentscomprise about one third of the suspended-sedimentload in the lower Rhine river, for example. Alarge portion of the transported material settles onto the flood plains (Guy, 1970), especially upstreamof hydraulic structures. The settled material undergoescompaction and other physical and chemicalchanges that can sometimes prevent its re-erosionby flows that would have otherwise carried it. Adecrease in the mean annual sediment transportedper unit area of the catchment is generally foundas the area of the catchment increases. The concentrationof suspended sediment in runoff is describedby formulae such as that of the National ResearchCouncil (1973), Negev (1972) and Beschta (1987):log c s= C log Q + B (4.26)in which c sis the concentration expressed in weightper unit volume of water, Q is the water discharge,C is a dimensionless coefficient and B is a functionof the rainfall depth, the antecedent discharge orsome other meteorological or hydrologicalvariable.The concentration of suspended sediment varieswithin the channel cross-section. It is relativelyhigh in the lower portion and may also be laterallynon-uniform so that it will need to be sampled atvarious points or along several verticals of the crosssectionto obtain its mean. The mean concentrationshould be evaluated to compute the total sedimentweight-per-unit time when multiplied by the waterdischarge. The graph of suspended sediment againsttime generally has a peak that does not occur simultaneouslywith the peak discharge. This lag is aresult of the specific conditions in a watershed andno generally applicable method has yet been foundto evaluate this difference.4.8.6.2 Bed-load transportCoarse sediments, or bed load, move by sliding,rolling and bouncing along channels and areconcentrated at or near the channel bed. The variablesthat govern transport are the size and shape ofthe particles and the hydraulic properties of theflow. As described in 4.8.5, the channel bed assumesdifferent configurations exerting resistance to awide-ranging flow of water and assumes a maximumvalue for the dune configuration. An empiricalformula for the rate of coarse sediment transportproposed by Du Boys in 1879 (Chow, 1964) is stillwidely used today in many models. The formula isas follows:q s= c τ 0γ⎛⎝τ 0γ− τ cγ⎞⎠ (4.27)where q sis the sediment transport rate per unitwidth of the channel in kg s –1 m –1 , τ o= γ R hS eis theshear stress at the channel bed in kg m -2 , τ cis anempirical value for the minimum τ orequired fortransporting the sediments considered, γ is thedensity of the water in kg m –3 , c is a dimensionalcoefficient in kg m –3 s –1 , S eis the energy slope of thewater and R his the hydraulic radius in metres which,for wide rivers, may be replaced by the mean depthof water. Values of the coefficients for equation 4.27are given in Table II.4.7 (Chow, 1964).Table II.4.7. Typical values of c and τ cparametersClassification Mean diameter c(kg m –3 s –1 ) τ c(kg/m 2 )(mm)Fine sand 1/8 8 370 000 0.0792Medium sand 1/4 4 990 000 0.0841Coarse sand 1/2 2 990 000 0.1051Very coarse sand 1 1 780 000 0.1545Gravel 2 1 059 000 0.251Gravel 4 638 000 0.435A more theoretically based formula was developedby Meyer-Peter in 1934 (Chow, 1964):q s=⎧⎨⎩(γ q ) 2/3 ⋅ S e − A ⋅ dB3/2⎫⎬⎭(4.28)where q is the water discharge per unit width of thechannel in m 2 s –l , γ is the specific weight of water inkg m –3 , S eis the energy slope, d is the representative

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-79grain size in metres, q sis the bed-load discharge perunit width of the channel in kg m –l s –l , B is a dimensionlessconstant that assumes the value of 0.40 ina consistent unit system and A is a dimensionalconstant that assumes the value of 17.0 in the SIsystem of units. If the transported sediments are ofdiverse sizes, d replaces d 35, which is the mesh sizethrough which 35 per cent of the weight of the bedload would pass. Equation 4.28 yields results thatare reliable, particularly for sand-bed channels.A second version of this formula has been developedto take into account the effects of dunes(Meyer-Peter and Müller, 1948):q s = 8 (β⋅τ * − 0.047) 3 ( s − 1)g ⋅ d 3(4.29)where q sis the bed-load discharge per unit width ofthe channel in m 3 s –l m –l , expressed in grains volume(without the empty spaces), g is gravitational acceleration,d is the representative grain size and τ * isthe dimensionless boundary shear stress, which isexpressed by:τ * R h ⋅ S= (4.30)( s − 1) ⋅ dwhere R his the hydraulic radius, S the slope of thechannel and s is an adimensional factor, givenby s = γ s/γ, γ sis the specific weight of the sediment,and γ is the specific weight of the fluid. Finally, thecoefficient β is a function of two Strickler’s numbersof the channel and of the grains (that is, after formdrag caused by bed forms has been excluded) and isexpressed by K f β =⎛ ⎞3/2 .⎝ K grains ⎠Strickler number K is the same as 1/n where n is theroughness coefficient in Table II.4.6.4.8.6.3 Total load formulaeTotal load carried by the stream is the sum of bedload, suspended load and wash load. However it isdifficult to relate wash load to flow conditions.Wash load is generally absent in flume experiments;therefore, total load would be bed material loadplus suspended load.Relationships for estimating total load can bebroken down into microscopic and macroscopicmethods. In the former, bed load and suspendedload are calculated separately, then added. Einstein’smethod can be cited as an example (Vanoni, 1975).Macroscopic methods are based on the premisethat, since the suspended load and bed load areessentially dependent on the same flow parameters,there is no need to estimate each separately. Instead,the total transport rate can be related to flowingfluid and sediment characteristics. The Engelundand Hansen (1967) method can be mentioned here,as it is simple:q s= 0.05 ⋅( s − 1) ⋅ g ⋅ d 3 ⋅⎛⎝K f2 ⋅ Rh1/3g⎞⎠ ⋅ (τ* ) 5/2 (4.31)where q sis the rate of sediment transport inm 3 s –l m –l , while s, d, g, K f, Rh and τ * are the same asin equation 4.30.Another formula for total load is that of Van Rijn(1984). The advantage of the Van Rijn method isthat it allows a separate calculation of the bed loadtransport and the suspended sediment transport.While suspended sediment is a complex topic, andtherefore not included in this publication, the bedload transport formula is provided as follows:q = 0.053 ⋅ T 2.1b 0.3D ⋅ ( s − 1) ⋅ g ⋅ d 350*(4.32)where q bis the bed load transport rate per unitwidth, s is the same as in equation 4.30, g is thegravitational acceleration, d 50is the representativebed material size, T is the transport stageparameter:T = (U 2 2− U* *cr )2U *cr(4.33)D *= d 50[(s–1) g/ν 2 ] 1/3 , ν is the kinematic viscosityand U *is the bed shear velocity given by:U * =C fC graing ⋅ R h ⋅ S (4.34)where S is the slope, R his the hydraulic radius, C fand C grainare the Chezy coefficients of the channeland grains, respectively, and U *cris the critical bedshear velocity, given by the Shields diagram (seeManual on Sediment Management and Measurement(WMO-No. 948), 3.2).4.8.6.4 Sediment transport on steep slopesand debris flowIn a steep catchment, different sediment transportprocesses may occur. During a flood event, dischargecan increase to such a level as to destroy the armourlayer of stream or torrent bed. As a result, fluvialtransport of the bed material will start. In addition,sediment may be supplied to the channelfrom slope failures; thus sediment availabilitycould be sufficient for the flow to move sediment inrates close to its transport capacity. At very high

II.4-80GUIDE TO HYDROLOGICAL PRACTICESsediment concentration, sediment moves in sloughsand flow becomes unsteady. At the front of thewave flow, the particles are more or less uniformlydistributed over the flow depth, while the mixturebehind the front may become more diluted. At theend the coarse particles are concentrated near thebed (Rickenmann, 1991).Erosion can increase as a result of changes broughtabout by earthquakes or lava eruptions. Ash, clay,coarse gravel, boulders, trees, loose rocks andanthropogenic material are transported by flowingwater in terrain such as debris or mud flows. Debrisflows, like flash floods, are fast moving and occur ina wide range of environments. A debris flow has theconsistency of wet concrete and moves at highspeeds of 15 m s –1 or even faster. Debris flowcommonly occurs in gently sloping alluvial fans,cone- to fan-shaped land forms created over thousandsto millions of years by the deposition oferoded sediment at the base of mountains ranges.The measurement of sediment transport and debrisflow in such cases is very difficult, but research studiesare being carried out in France, China, Japan,the United States and the Russian Federation, wherelarge areas are affected by intense erosion. Gravelbed transport, debris flow and mud flow are topicsof recent interest and the work of Thorne and others(1987), Coussot (1997), Zhaohui Wan and ZhaoyinWang (1994) provide further information.4.8.6.5 Sediment transport in gravel-bedriversIn mountain rivers, bed-load discharge accounts fora relatively large proportion of the total discharge.Estimates of sediment transport in gravel-bed riversare limited due to problems associated withsampling of bed load and bed material in the field,extreme non-homogeneity of the bed material andnon-equilibrium bed-load transport. On the basisof flume data using sediment sizes up to 29 mm andslope up to 20 per cent, an equation has beenderived (Smart, 1984) for bed-load transport asfollows:q BV3 1/2⎡⎣g ( s − 1)d a⎤ = 4 ⋅ ⎛ d 90 ⎞⎝ d 30 ⎠⎦⋅ S 0.6 ⋅⎛⎝⎞VV *⎠ ⋅τ*0.5 (τ * −τ * c )0.2(4.35)where q BVis volumetric bed-load transport per unitwidth; S is the channel slope; g is gravitationalacceleration; d ais the arithmetic mean size; d 90andd 30indicate bed material size finer than 90 and 30per cent, respectively; V is average velocity, V *isshear velocity of flow, τ* is adimensional shearstress (as in equation 4.29) and τ* c, dimensionlesscritical shear stress, corrected to take into accountthe slope, expressed by:⎛⎝τ * c = τ * 0c ⋅ cos α 1 − tan αtan ϕ⎞⎠(4.36)where τ* 0cis the critical Shields parameter, α is theangle of the slope so that S = tan α, and ϕ is theangle of repose of a submerged bed material.Sediment transport in rivers is a subject that hasinterested many researchers for which other texts(Raudkivi, 1998; Yalin, 1992) could also be used forreference. These formulae are based on empirical,semi-theoretical and theoretical equations, checkedwith laboratory data. Owing to the non-availabilityof reliable data from natural streams, however, littlefield data was used. Results of these formulae oftendiffer enormously. It is difficult to ascertain whichformula yields the most realistic results. Selectionof a particular formula or set of formulae requirescalibration with observed data for a particular riversystem.4.8.7 SedimentationSuspended sediments are deposited according totheir settling velocity. A relationship between thegrain size and the settling velocity is shown inFigure II.4.44. Coarse sediments deposit first, theninterfere with the channel conveyance and maycause additional river meanders and distributaries.Sediment entering reservoirs deposits and may formdeltas in the upstream part of reservoirs. The depositedsediment may later be moved to deeper parts ofthe reservoir by hydraulic processes within thewater body. As the area of flowing water expands,the depth and velocity decrease; eventually evenfine sediments begin to deposit. A significantconcentration of suspended sediment may remainin the water column for several days after its arrival.This may interfere with the use of the stored waterfor water supply, recreation and other purposes.Not all of the sediment will be deposited in reservoirs,however. A large portion of it may remain inthe upper zones of the catchment, some may bedeposited upstream of the reservoirs and thereleased water carries some downstream of the dam.The sediment-trapping efficiency of a reservoirdepends on its own hydraulic properties, those ofthe outlet and the nature of the sediment.The density of newly deposited sediments isrelatively low but increases with time. The organic

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-81component in sediment may undergo changes thatmay reduce sediment volume and enhancebiochemical processes in the stored water. Forfurther information, see 4.9 and 4.10.measures. They include channel improvement andstabilization works, reservoirs, debris and sedimentbasins, levees, dykes, floodways and floodwaterdiversions.4.8.8 Sediment control measuresSediment control measures fall into two broad categories:land-treatment measures for watershedprotection and structural measures. Detaileddescriptions are provided by Vanoni (1975). Theaim of land-treatment measures is to reduce erosionin the watershed, and thus the rate of sedimentformation, by improving the protective cover onthe soil surface, diminishing surface runoff andincreasing infiltration rates. These measures includethe following:(a) Land management based on agronomy andforestry, such as the use of crop rotation andthe exclusion of grazing on critical runoff andsediment-producing areas;(b) Appropriate field practices such as contourfarming on sloping land, the development ofgradient terraces on steep slopes and the gradingand lining of natural waterways, irrigationand drainage ditches, and depressions.Structural measures are aimed at providing protectionbeyond that afforded by land-managementV s cm s –11 000100101.010 –110 –2 0.01 0.1 1.0 10 100 1 000Grain size, mmSettling velocity, cm s –1Figure II.4.44. Settling velocity of quartz grains4.9 WATER QUALITY AND THECONSERVATION OF AQUATICECOSYSTEMS [HOMS K55]4.9.1 GeneralWater resources projects should be designed andoperated in an environmentally friendly way. Inaddition, they should comply with water-qualitystandards, thus avoiding detrimental effects onaquatic ecosystems and on water quality. This is thesubject of this subsection of the Guide. The nextsubsection, 4.10, addresses the broader question ofthe environmental management of rivers in thecontext of river morphology and ecology, focusingon the main impacts that water resources projectscan have on river ecosystems and the methodscommonly applied to reverse or mitigate them.There are close relationships between somequantitative characteristics of water bodies, such asthe flow regime and dilution capacity in rivers, orthe flushing time and stratification patterns inlakes, and their ecological functioning and waterquality. As water resources projects generally altersome of these quantitative characteristics, it shouldbe possible to estimate or predict the environmentalimpacts when these relationships are wellunderstood and defined. Unfortunately, suchrelationships can be very complex, and in somecases are only known in qualitative terms. Moreover,the data required to parameterize them are rarelyavailable in practice. Therefore, it is natural thatonly rough estimates can be made of water qualityand a project’s environmental impacts.Some measures to protect water quality and aquaticecosystems were recommended by the UnitedNations at the International Conference on Waterand the Environment: Development Issues for theTwenty-first Century (United Nations, 1992).Recommendations for the environmentally friendlydesign and operation of water resources projects, aswell as mitigation, rehabilitation, and restorationof existing projects and further detailed referencescan be found in Petts (1984), Gore and Petts (1989),and the World Commission on Dams (2000).Recommendations relating to hydropower and irrigationdams are available in Brookes (1988) andGardiner (1991), those regarding channelization

II.4-82GUIDE TO HYDROLOGICAL PRACTICESprojects, in Brookes and Shields (1996), and waterresources in rivers, in Cowx and Welcomme (1998)and WMO/GWP (2006). Thomann and Mueller(1987), and Chin (2006) offer good introductionsto the topic of water quality in rivers, lakes andreservoirs.4.9.2 Relationships between waterquantity and water quality4.9.2.1 Streams and riversA significant proportion of the variability in riverwater quality can be related to variations in riverflow. The effects of changes in river discharge onthe concentration and load of substances are numerous,and may counteract each other. An increase ofriver flow generally leads to the followingdevelopments:(a) Enhanced dilution of pollutants enteringwith wastewaters;(b) An increase in suspended solids derived fromsurface runoff and disturbance of bottomsediments;(c) The release of materials adsorbed by, orprecipitated in, sediments such as phosphatesand heavy metals;(d) Higher demand for biochemical oxygencaused by stirring up reducing substancesfrom the riverbed;(e) Decreased ratio of groundwater to surfacerunoff in the river flow, generally resultingin a lower pH;(f) The washing out, and subsequent reductionof benthic organisms and in residencetimes;(g) Attenuated effects of sudden inputs ofpollutants;(h) The reduced absorption of solar radiationand a related decline in water temperatureand photosynthetic activity;(i) Greater turbulence and better aeration leadingto higher levels of dissolved oxygen inconjunction with lower temperatures.The sequence and time of occurrence of high flowsare critical in determining the extent of many ofthese effects. A second flood wave, following shortlyafter a first, may contribute little to the effects ofthe first flood. Thaw and rain after a long period offrost may lead to a sudden influx of road de-icingsalts and may cause significant sodium and chloridepeaks, despite the rise in flow (see earlierreference to these factors in 4.7). The land use, soiltype, land cover and other characteristics of theportion of the basin in which the flood-generatingrunoff originates are other factors affecting themagnitude of water quality changes caused by highflows.When the rise in river flow results in significantflood-plain inundation, a number of additionalwater quality effects may follow. Most significantamong them are the following:(a) Flood attenuation related to additional valleyand bank storage, leading to a reduction indownstream flood flow, hence lowering thevarious effects listed previously under (a) to (i);(b) Increase in the water surface-to-volume ratio,resulting in expanded opportunities for solarradiationabsorption and increases in watertemperature and photosynthetic activity;(c) Reduced flow velocity in the flood plain, leadingto decreased re-aeration and depositionof potentially contaminated suspended solidsoutside the main river channel;(d) Intensive contact with previously depositedsediment, various types of soil structures,dumps, wastewater treatment plants, industrialchemicals and so on, that can lead to riverpollution.In general, low-water periods produce oppositeeffects to those caused by flow increases. Further,low-flow periods are often accompanied by a relativelyhigh diurnal variation in water qualitycharacteristics, for example, dissolved oxygen,carbon dioxide, pH and temperature. In aridclimates, the effect of evaporation on the concentrationof various substances in the water can besignificant. In cold climates, low-water periods inwinter may also be periods of oxygen deficit wheneverthe ice cover interferes with the re-aerationprocess.4.9.2.2 Large lakes and reservoirsThermal stratification is a result of natural factors.However, thermal pollution and increased watertemperatures caused by flow reduction can be acausal or contributing factor (see 4.9.5.4).Figure II.4.45 shows a representative profile of thesummer stratification in a large storage reservoir.Thermal stratification can lead to dissolved oxygenstratification, particularly in nutrient-rich meso andeutrophic lakes and reservoirs, as well as to thestratification of other dissolved substances. In theepilimnion or upper layer of water, the water iswarmer in summer and its quality is generallybetter. In the upper layer, one may expect reducedsilicate content following increases in diatom abundance,decreased hardness from direct inputs ofprecipitation water and, most importantly, increaseddissolved oxygen caused by atmospheric exchange

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-83and photosynthesis by phytoplankton andmacrophytes.In the hypolimnion, or lower layer of water, thewater is colder in summer and has a reduced concentrationof dissolved oxygen. Various potentiallyharmful substances frequently accumulate in thislayer owing to deposition on the bottom, adsorptionon sediment and ingestion by living organismswhich, when they die, decompose on the bottom ofthe lake (see 4.9.5.3). Anaerobic decomposition ofalgae and other organisms may occur in thehypolimnion. One may expect the hypolimnion toshow trends of an increasing concentration ofammonia and hydrogen sulphide, a reduction innitrate and sulphate concentration, an accumulationof sediment and occasionally of heavy metals,and a periodic increase in iron, manganese andphosphate concentrations.During the turnover caused by the seasonal coolingof the surface layer of the lake, a convective circulationtakes place, resulting in vertical mixing of thelake and a uniform temperature. In deep lakes andreservoirs with a large hypolimnion volume, theseturnover events can lead to fish-kills and other problems,because a large volume of low-quality water ismixed with the higher-quality epilimnetic water.In addition to the aforementioned effects, thefollowing developments can be expected:(a) In large lakes and reservoirs, organic matter isbiodegraded to a large extent because of longresidence times;(b) Variations in lake water quality are dampenedout for the same reason;(c) Water quality in the rivers flowing out of areservoir depends largely on the occurrence of7°CDamWater surfaceEpilimnion 35°CMetalimnion 10°C to 35°C(thermocline)Hypolimnion 10°CInflowFigure II.4.45. Representative profile showingsummer stratification in a large storage reservoirwith a high damstratification and the depth at which the intakestructure is located, since rivers flowing out ofnatural, unregulated lakes draw epilimneticwater.4.9.3 Effects of water resources projectson water quality in streams andrivers4.9.3.1 Dams and weirsDams – and to a lesser extent, weirs – generally havethe following effects on water quality in theupstream reach of a river by raising upstream waterlevels:(a) Intensification of self-purification processesbecause of increased residence time in the reachand more deposition of suspended solids, whichresults in increased solar-radiation absorptionand changes in the sediment characteristics ofthe riverbed;(b) A rise in water temperature and phytoplanktonproduction, greater oxygen consumption andincreased day-night fluctuations in oxygen, pHand carbon dioxide as a result of (a).Fish migration may be disturbed both by the physicalbarrier and changes in water quality. Changes instream bank or shoreline vegetation, which aregoverned by local topography, climate and waterlevelvariation, may also affect water quality. Forexample, water turbidity may be increased in reservoirswith fluctuating levels. In cold climates, damsand weirs create favourable conditions for anextended duration of ice cover in upstream reaches.This leads to decreased re-aeration. Further effectswhere large storage volumes are concerned mayresult from thermal stratification. Increased pollutionretained in the reservoir may lead toeutrophication and anaerobic conditions (see4.9.5.1 and 4.9.5.2, respectively).The effects of a dam or weir on water quality in thedownstream river reach depend on the water residencetime in the impoundment, where this iscalculated as the ratio of storage volume to streamflow.They also depend on the stratification anddam design and operation, particularly the depth atwhich intake structures are located in relation tothe hypolimnion. The most important effects of adam or weir are as follows:(a) Reductions in suspended solid load, pollutionload and turbidity;(b) Changes in the chemical characteristics of thewater – often a lower concentration of dissolvedoxygen and nitrates – and increases in phosphate,carbon dioxide and hydrogen sulphide,

II.4-84GUIDE TO HYDROLOGICAL PRACTICESthe latter particularly when anaerobic conditionsprevail upstream;(c) Lower summer water temperatures and higherwinter water temperatures, with major effectson invertebrate and fish communities downstream;(d) Less day-night temperature fluctuations towhich the river flora and fauna must adapt.4.9.3.2 River training worksRiver training generally involves a deepening andstraightening of the river channel for variouspurposes, including navigation, flood control, landuseimprovement and erosion protection. Thisresults in changes in the geometric and hydrauliccharacteristics of the river channel, and in somecases, of the flood plain as well. For further information,see 4.6.When river training is done for navigationalpurposes, it generally involves the construction ofnavigation weirs and locks. In addition to theeffects of weirs (4.9.3.1), the training works andthe operation of navigation canals lead to increasedturbidity and mixing of the water and aerationfrom the mechanical effects of the moving boats.However, the boats are a source of routine andaccidental pollution, and can re-suspend contaminatedsediments from the bottom. The dredging ofnavigation lanes can also cause similar problems.In other cases, river-training works lead to a reductionin self-purification processes because thestraightening of the banks eliminates stagnantwater zones both as areas of self-purification andas a favourable environment for animal and plantlife. The reduced surface-to-volume ratio leads to areduction in the solar-radiation absorption and reaeration.The loss in re-aeration may be partlycompensated when river training produces higherwater velocities.4.9.3.3 Flow reduction and augmentationIn addition to the flow-regulation effects of dams,many water resources projects involve downstreamflow reductions resulting from diversions for variouswater supply purposes or augmentation frominputs of water coming from sources outside thebasin.When the extracted water undergoes treatment andthe resulting sludge and residues are returned to thedonor river, or when water is diverted from the lesspolluted portions of the river cross-section, thediversion effects are equivalent to a reduction inflow or to a pollution input (see 4.9.5). The disposalof sludge and residue is generally the focus of regulationsand legislation relating to the quality ofeffluents. These differ widely from country tocountry.The effects of flow augmentation depend mainlyon the quality of the additional water as comparedto that of the river water. An addition of water ofpoorer quality is equivalent to a pollution input, asis the net effect of a project diverting water from acleaner tributary.4.9.4 Effects of water resources projectson water quality in large lakes andreservoirsWater quality in large lakes and reservoirs may beimproved or degraded by water resources projects.Where such projects involve a withdrawal of waterof a better-than-average quality, for example fromthe epilimnion, they will generally worsen thewater quality in the lake. The same is true whenwater of poorer-than-usual quality is pumped intothe lake or reservoir. As explained in 4.9.2, thequality of the water flowing out of a reservoirdepends on whether or not stratification occurs,and on the depth at which the intakes arelocated.Water quality in a large reservoir depends to a largeextent on the characteristics of the underlyingterrain before flooding and on the treatmentapplied to it. If the future reservoir bottom iscovered by soil with rich organic content orhumus, the latter is leached after the reservoir isfilled and accelerated eutrophication may result(see 4.9.5.1). This may be avoided by the removalof vegetation and soil prior to flooding, althoughthis is a costly operation.4.9.5 Water quality changes caused bypollution4.9.5.1 EutrophicationOne of the most common forms of pollution isexcessive concentrations of nitrogen and phosphorusnutrients originating in urban wastewaters orrural runoff. This generally results in the rampantgrowth of algae, particularly in areas with low watervelocity. The subsequent decrease in dissolvedoxygen concentrations can lead to significantreductions or even the disappearance of a numberof plant and animal species. This is known aseutrophication. It is a natural process that marksthe maturing and ageing of lakes. However, underconditions not involving man’s activity, this may

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-85take hundreds or thousands of years, depending onlake size, hydrological conditions and land cover inthe basin. Civilization is responsible for acceleratedcultural eutrophication in a large number of lakesall over the world.Eutrophication and its causative factors are a majorwater quality problem. Considerable researchdevoted to the study of cultural eutrophication hasresulted in the availability of many quantitativecriteria and models for assessing its development.Further, different methods have been developed tohelp improve the condition of culturally eutrophiedlakes. Although the lack of many elements can limitprimary productivity, nitrogen and phosphorus arethe ones most likely to limit the growth of algae innatural waters. In some countries, attempts to stopthe advance of eutrophication have been made bybanning the use of phosphorus compounds indetergents and introducing advanced, tertiary treatmentprocesses for the removal of phosphorus andnitrogen.The effects of eutrophication are reflected in strikingchanges in the affected lake ecosystems. Highlypolluted environments have few species. Whenpollution is caused by toxic substances, the numberof individuals surviving in each species is low,sometimes extremely low. However, when there isan excess of nutrients, a handful of species canreach large population numbers, owing to theirincreased productivity, but this is always matchedby a decrease in diversity, because of the exterminationof many other species that cannot withstanddeteriorating environmental conditions.Further details on cultural eutrophication andpossible solutions can be found in Henderson-Sellers and Markland (1987), Harper (1991) andRyding and Rast (1989).4.9.5.2 Organic matter and self-purificationA large proportion of polluting substances ofmunicipal, industrial and, particularly, agriculturalorigin consists of organic matter. A number ofphenomena occurring in natural waters tends totransform this organic matter into more or lessinnocuous inorganic nutrients, a process known asself-purification. Some of these nutrients are recycledby algae and other producers, which generatesecondary organic pollution upon dying, such as ineutrophication. Before the biological degradationleading to self-purification can take place, theorganic substances dissolved in the water must beadsorbed and concentrated on the surface of solidparticles. Adsorption can take place on the solidparticles on the river bottom, banks and macrophytes,and on suspended solids.Most biological degradation is associated withoxygen consumption, which is the key factor in theself-purification process. When oxygen consumptionin water proceeds so rapidly that it exceeds therate by which oxygen is replenished from the air orby oxygen-producing biological activities, namelyphotosynthesis, the aerobic self-purification capacityof the water body is exceeded. This occurs whenone or more of the following conditions occur:(a) The load of organic matter exceeds the selfpurificationcapacity;(b) Biological degradation processes are acceleratedby certain factors, for example temperaturerise;(c) Oxygen replenishment is diminished by thermalstratification, ice cover or other causes.When the self-purification capacity is exceeded,water anoxia occurs, and the decomposition of theorganic matter generally continues under anaerobicconditions. This kills most metacellularorganisms and interferes with many uses of thewater body. The use of water for recreation and fisheriesis impossible under such conditions and itmay be much less desirable for other uses, such aswater supply.4.9.5.3 Adsorption and accumulation ofpollutantsSome harmful substances are adsorbed on organicand inorganic suspended solids. When the lattersettle on the bottom, these toxic substances aretemporarily removed from the main body of water.Organisms are also capable of concentrating anumber of organic and inorganic pollutantsthrough biochemical processes. For example, theconcentration of some pesticides in aquatic organismscan reach levels up to 300 000 times higherthan those found in the corresponding waterenvironment.However, owing to physical and biological processes,substances absorbed and accumulated byorganisms may be returned subsequently into thewater body in solution or in particulate form. Theconcentration of pollutants by different levels ofthe organisms is of particular significance becausethey are in the food chain and bioconcentratedpollutants are passed from one level of organism toanother in increasingly higher concentrations. Sucha process, called biomagnification, is responsiblefor mercury poisoning related to the well-knownMinamata disease.

II.4-86GUIDE TO HYDROLOGICAL PRACTICES4.9.5.4 Thermal pollutionThermal pollution is defined as an increase of thetemperature of a water body over the natural levelcaused by the release of industrial or municipalwastewater – in particular, cooling water fromnuclear and thermal power plants and other industrialprocesses.The effects of thermal pollution on water qualityare complex and relate to the effects of highertemperatures on the viscosity of water, itsdecreased solubility for oxygen and increasedchemical and biological activity. Thermal pollutionmay also be a contributing factor in thermalstratification. As a result of thermal pollution, theperiod of biological productivity is lengthened,which leads to an increased load of organic pollution.In addition, certain species of green algaeare replaced by blue-green algae, which transmitto the water undesirable characteristics of smell,taste and toxicity.As mentioned previously, self-purification processesare accelerated by higher temperatures, and thus bythermal pollution, to the extent that acute oxygendeficits may occasionally occur. In winter, ice formationis delayed by thermal pollution and thisbroadens the possibility of re-aeration. Becauseaquatic animals are ectotherms, that is, coldblooded,water temperature is a vital influence intheir growth, reproduction, and survival. Mostaquatic invertebrates and fish are adapted to narrowtemperature regimes; any departures from the naturalstate caused by thermal pollution or reservoirreleases of cold hypolimnetic waters can exterminatespecies from a river reach.4.9.6 Measures to reduce effects ofpollution on water qualitySuch measures can generally be divided into twogroups: preventive and corrective. Whenever feasible,preventive measures should be applied becausethey are more economical.4.9.6.1 Preventive measuresPreventive measures consist primarily in removingpollutants at the source. Treating wastewaters,changing industrial processes, altering the chemicalcomposition of certain industrial products byeliminating phosphorus compounds from detergentsand artificially cooling industrial wastewatersare means of doing so. If pollution originates fromdiffuse sources such as pesticides, herbicides, fertilizersand uncontrolled urban waste, and is washedinto a river or lake from the land surface, pollutionabatement can be achieved only by changing thepractices that lead to the uncontrolled spreading ofpollutants and adopting measures to reduce runoffand soil erosion.Significant pollution is generated through soilerosion. Its prevention requires adequate forestrymanagement and construction and farming practices.Finally, pollution from leachates, originatingfrom garbage dumps, may be significant at the locallevel. This can be avoided by ensuring that suchdumps are appropriately located and designed.4.9.6.2 Corrective measuresReducing pollution in water bodies after pollutantshave reached them is often difficult and costly. Inmost cases, it is only possible to treat the waterdiverted from the water body for specific purposes,such as for domestic or industrial water supply.However, in special circumstances, remedial workcan be carried out for the whole water body. Whererivers are concerned, remedial measures consistmainly of artificial re-aeration or oxygenation, orthe dredging of settled pollutants. Measures targetinglakes and reservoirs include the following:(a) Emptying the lake regularly between lateautumn and early spring so as to expose organicmatter directly to the air and permit aerobicdecomposition of the organic matter. This ismore feasible for reservoirs and small pondsthan for large natural lakes;(b) Dredging the bottom of the lake mechanicallyor by suction in the areas that contain thehighest concentrations of organic and pollutingmatter. Disposal of this material can be achallenge, however;(c) Forced re-aeration by compressed air in the deoxygenatedlayers;(d) Harvesting and disposing of organic matterproduced in the form of algal blooms, excessiveplant growth, undesirable fish and so forth.4.10 HYDROECOLOGY [HOMS K55]4.10.1 IntroductionSome of the effects that water resources projects canhave on water quality were discussed in 4.9.However, any attempt at defining the quality of ariver has to comprise much more than its waterquality alone. Indeed, clean waters are a necessarybut certainly not sufficient condition to ensure thata river ecosystem is in good ecological health. In

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-87the following discussion, the focus is no longer onwater chemistry, under the assumption that thereare no water quality problems, and concentratesinstead on other aspects of river quality. These arerelated to the river’s physical structure, habitatavailability and biodiversity, the natural processesthat determine them and how these aspects may beaffected by building and operating water resourcesprojects. The discussion starts by considering thepressing need for adequate environmental managementof rivers, as well as its objectives, and definingsome basic terms. Then, some fundamental conceptsof river morphology and ecology are considered,focusing on the processes more than on the organisms.This is necessary in order to understand thefinal two sections, which describe the main impactsthat water resources projects can have on riverecosystems, as well as the methods commonlyapplied to reverse or mitigate them.Hydraulic engineers and hydrologists are the professionalsin charge of designing and operating waterresources projects and, as such, they should beinvolved in any interdisciplinary teams responsiblefor river management and restoration. Referencecan be made to Meier (1998a) and WMO/GWP(2006) for short reviews of river ecology, to Jeffriesand Mills (1995) and Cushing and Allan (2001) forbasic but complete introductions to the topic andto Allan (1995) for an exhaustive review. A goodintroduction to the subject of river and lake restorationis given by the National Research Council(1992). Cowx and Welcomme (1998), the FederalInteragency Stream Restoration Working Group(FISRWG, 1998), Calow and Petts (1994), Boon andothers (1992), and Harper and Ferguson (1995) arecompilations that cover in detail most aspects ofriver management and restoration. Petts andAmoros (1996) provide an integrative vision ofecological change in river systems. Morisawa (1985),Leopold (1994) and Schumm (2005) provide goodintroductions to river morphology.This section deals specifically with the ecologicalimpacts of water resources projects on rivers, themain effects on lakes having already been discussedin 4.9.4.4.10.2 Environmental management ofrivers4.10.2.1 An urgent needMost rivers in the world have suffered widespreadenvironmental degradation caused by dams, pollution,water diversions, intensive land-use patterns,channelization or river training, flood-plaindevelopment, introduction of exotic species and soforth. Owing to these and other human-causedchanges, a larger proportion of organisms is extinctor imperilled in freshwaters than in any other typeof ecosystem (Angermeier and Karr, 1994), and theeconomic, ecological, recreational and aestheticvalue of many running waters has been sharplyreduced. In essence, freshwater ecosystems can beconsidered to be “biological assets [that are] bothdisproportionately rich and disproportionatelyimperilled” (Abramovitz, 1995).Countering this downward trend in river qualityrequires sound environmental management. Thisinvolves designing and operating new waterresources projects in a manner as environmentallyfriendly as possible to mitigate the impacts of existing,older projects and restore degraded rivers. Themain objective of such measures should be to maintainthe ecological conditions of healthy, unimpairedrivers and improve them in affected fluvial ecosystems,returning them to higher levels of normalityso that they can sustain the full suite of originalorganisms and habitats, as well as supply goods andservices to society.4.10.2.2 Environmental managementobjectivesWhat does it mean to maintain and improve theecological conditions of a river? Many haveunderstood this to mean to increase productivityand/or biodiversity, but these are veryanthropocentric concepts; “more is better” is notreally applicable to natural systems. For example, apristine, ultraoligotrophic alpine lake and its streamoutlet have very low concentrations of nutrientsand are thus quite sterile environments. Still, theyare unimpaired aquatic systems, which cannot beimproved. Indeed, improving the lake by making itmore productive, for example by fertilizing itswaters, would cause eutrophication, with theconsequences described in 4.9.5.1. It is clear fromthe foregoing discussion that the purpose ofenvironmental management of aquatic ecosystemscannot be to increase their productivity.In simple terms, biodiversity is the variety oforganisms and their habitats that can be found inan ecosystem. This concept has more intuitiveappeal as an adequate objective for river managementand restoration. However, many water bodieshave been degraded by the introduction of exoticspecies; they might have a higher biodiversity butare not better systems for it. It should be clear thatthe concepts of naturalness and belonging to aplace must be involved in the objective of

II.4-88GUIDE TO HYDROLOGICAL PRACTICESenvironmental management; these are explicit inKarr’s (1996) definition of ecological integrity:... the capacity to support and maintain a balanced,integrated, adaptive ecosystem, having the fullrange of elements (genes, species, assemblages) andprocesses expected in the natural habitat of aregion...A river corridor with high ecological integrityshould reflect the unimpaired, original conditionsin an area, including the presence of all appropriateelements – species, flood-plain ponds and wetlands,for example – and the occurrence of all naturalprocesses such as floods and lateral migration. Theseconditions should be characterized by little or noinfluence of human actions – conditions such asthose found in national parks. An ecosystem withhigh integrity reflects natural evolutionary andbiogeographic processes (Angermeier and Karr,1994). Restoring a river to high levels of ecologicalintegrity may be an impossible objective because ofeconomic, social, political or technologicalconstraints. If so, lower levels of integrity must besought. Some call this intermediate goal rehabilitationor renaturalization; it can also be thought of aspartial restoration.Some rivers have been modified for such a longtime, or so intensively, so that little or nothingnatural remains about them. Other systems, such asa series of hydropower reservoirs, are or will becontinuously managed. These sites cannot be trulyrestored, eliminating ecological integrity as amanagement goal. However, one can still strive forecological health, defined by Karr (1996) asfollows:An ecosystem is healthy when it performs all of itsfunctions normally and properly; it is resilient, ableto recover from many stresses, and requires minimaloutside care. Ecological health describes thegoal for conditions at a site that is managed orotherwise intensively used. Healthy use of a siteshould not degrade it for future use, or degradeareas beyond the site.To assess ecological integrity and health, it is necessaryto select a benchmark state against which otherstates can be compared and a variety of measurableecological indicators. For example, native biodiversityis an important indicator of ecological integrity.Once a restoration goal – a benchmark state – hasbeen selected, the degree of success can be appraisedby comparing measured values of the indicatorswith values for the benchmark. Meier (1998)provides a short summary on the meaning andobjectives of river restoration. Karr and Chu (1999)give a basic introduction to the use of multimetricbiotic indices to assess ecological integrity andhealth.4.10.2.3 The bases for environmental rivermanagementIt is clearly much easier, economical and effectiveto conserve rivers by maintaining their existingecological integrity than causing undue environmentaldegradation and then attempting to reverseit with restoration measures.All aspects of the environmental management ofrivers should be based on sound ecological principles.This is easier said than done, as rivers arehighly complex natural systems that are structuredby many different physical and biological drivingforces. Thus, a good understanding of their behaviourrequires a strong background in hydrology,hydraulics, fluvial geomorphology and stream ecology.Environmental river management is aninterdisciplinary endeavour, generally undertakenby teams composed of engineers, physical scientistsand biologists, plus social scientists, economistsand managers. Water resources projects can causevaried impacts on the fluvial environment, whichcannot be understood and therefore cannot beavoided or mitigated, without some basic knowledgeof river behaviour.Because interdisciplinary river studies are relativelyrecent, there is still some confusion regarding themost basic definitions. Dunbar and Acreman (2001)define hydroecology as:the linkage of knowledge from hydrological,hydraulic, geomorphological and biological/ecologicalsciences to predict the response of freshwaterbiota and ecosystems to variation of abiotic factorsover a range of spatial and temporal scales.This is precisely the subject of this subsection of theGuide. However, referring to similar issues, Zalewski(2000) defines ecohydrology as:the study of the functional interrelationshipsbetween hydrology and biota at the catchmentscale … a new approach to achieving sustainablemanagement of water.As Nuttle (2002) points out, however, Zalewski’sdefinition would imply that ecohydrology is bothscience and management at the same time. Nuttlecorrectly notes that a holistic approach to watermanagement depends on the integration of

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-89hydrological and ecological science, but that manyother factors, in addition to scientific knowledge,are involved in managing water. He goes on todefine ecohydrology as follows:the sub-discipline shared by the ecological andhydrological sciences that is concerned with theeffects of hydrological processes on the distribution,structure, and function of ecosystems, and onthe effects of biotic processes on elements of thewater cycle.Both definitions suffer from excessive generalityand are too inclusive. Ecohydrology cannot encompasseverything that has to do with both water andecology. Note also that the term ecohydrology hasbeen used in a much more restricted context, havingto do with the role of transpiration of terrestrialplants in the global water cycle.It is therefore preferable to only use the termhydroecology in the sense proposed by Dunbarand Acreman (2001) because it better conveysthe fact that it is hydrological and that otherphysical processes partly drive and structurefreshwater ecosystems, but still within a clearunderstanding that the main focus is on theecological integrity of these systems. In otherwords, it is about the effects that hydrology hason the ecology of rivers and lakes, and notvice-versa.4.10.3 Basic notions of river morphologyand ecology4.10.3.1 The components and extent of fluvialecosystemsA river ecosystem consists of many interactingorganisms of different species, the biota, that live ina physical setting, the abiotic environment. Theseorganisms need food sources in order to stay alive,grow and reproduce, and a place to live in the physicalenvironment: a habitat. They are also subject tomutual biotic interactions, for example, predation(acting either as prey or predator) and competition(fighting for limiting resources, such as space orfood).It is fundamental to emphasize from the beginningthat in terms of processes and behaviour, ariver comprises much more than the layman’sconcept of a wet channel as seen during lowflowperiods. In effect, it also incorporatesriverbed materials, streambanks and the completeflood plain. The flood plain is the largely horizontalalluvial landform adjacent to a riverchannel, separated from it by banks. It isconstructed by the river from sediment in thepresent climate and flow regime, and is inundatedduring moderate flood events. To makethis distinction clear, the terms river corridor orriver system can be used. Figure II.4.46 shows aFigure II.4.46. The concept of a river corridor in the case of a meandering alluvial stream. The rivercorridor includes all of the landforms shown, as well as the river’s flood plain (Federal InteragencyStream Restoration Working Group, 1998).

II.4-90GUIDE TO HYDROLOGICAL PRACTICESreach of a meandering river, illustrating some ofthese features.Thus, a river corridor has diffuse boundaries withthe terrestrial and groundwater systems, the riparianand hyporheic zones, respectively. It includesbars, side arms, flood-plain lakes and all otherfeatures created by fluvial processes within the floodplain. These channel and flood-plain featureschange with time. Therefore, a fluvial ecosystemcan be considered to have three spatial dimensions:longitudinal, in the downstream direction, lateral,into the flood plain and vertical, into the alluvialsediments, all of which vary with time (Stanfordand others, 1996).4.10.3.2 Fluvial landformsThe vast majority of river reaches are alluvial, thatis, they were formed over unconsolidated sedimentsthat were previously transported and deposited bythe stream flow. Non-alluvial rivers are thosebounded by bedrock and/or laterally confined byvalley walls, so that they are not free to adjust theirshape. As shown in Figure II.4.47, an alluviallandscape is determined by the interaction betweenthe hydrological regime or the pattern of flowvariability, the sediment load and calibre, the coarseregime of woody debris or tree logs, bed and bankmaterials and flood-plain vegetation for a givenvalley slope. Thus, the water, sediment and largewoody debris coming into an alluvial reach interactamong themselves, and also with the reach bankand bed materials and flood-plain vegetation. Bydoing so, they continuously modify the river’smovable sediment boundary through erosion anddeposition, shaping a dynamic, changing channel,with a given style or pattern. Most rivers are inregime, also referred to as steady state, or dynamicequilibrium, indicating that they are not sufferingaggradational or degradational trends. In otherwords, even though they may keep moving about,their form does not change statistically with time,so that they always look the same.The currently accepted view among river ecologists,for example, Stanford and others (1996), isthat the community structure in flood-prone riversystems – the species present in the river corridor –is mainly determined by the dynamics imposed bythese physical, hydrogeomorphic processes, notby biotic interaction. The opposite occurs inlakes.Especially where the local climate and river hydrologyallow for perennial flows and the occurrence ofwoody vegetation, flood-plain corridors of alluvialrivers are among the most dynamic, complex,diverse and productive – as well as endangered –ecosystems on Earth. The number of differentspecies of trees, plants, fishes, invertebrates, birds,mammals and so forth that can live in anintact flood-plain reach such as the one shown inFigure II.4.48 is immense.Climateor weatherregimeGeology– Lithology– TopographyLarge-scale rivermorpohologyNaturalvegetation/land useSoil characterWood regimeFlow regimeSediment regime– Bed material load– WashloadRiver/flood-painstyle (pattern)andshiftinghabitatmosaic(habitat regime)resulting from theinteraction betweenthe flow, sediment,and wood regimes,and the bed andbank materials aswell as the riparianvegetation alongthe reachSpeciesassemblage(dynamic)Upstream watershedvariablesEcological regimeReach variablesFigure II.4.47. Interaction among the catchment variables that determine the flow, sediment and woodregimes imposed on a reach from upstream. This, in turn, controls its morphological and habitat regimes– the river/flood-plain style and the shifting habitat mosaic, respectively – which together define theecological regime of the river corridor ecosystem.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-91As shown on the left-hand side of Figure II.4.47,the only independent variables in a given riverbasin, at the longest timescales, are its geology(or physiography: lithology and topography) andclimate. The local temperature and rainfallregimes cause weathering of the exposed rocks,determining the character of the soil and the typeof vegetation, if any, that can grow within thebasin. Together, acting through the streamnetwork, all of these variables prescribe thedischarge, sediment and large wood regimes forthe reach located downstream. They also drivethe load of organic detritus (leaves, twigs andorganic silt, commonly referred to as particulateorganic matter), dissolved matter fluxes such assolutes and the stream temperature regime.Human influences, including land-use changes,dam building and flood control measures, havedramatically altered all of these natural regimesin many rivers of the world.The study and management of alluvial rivers requirean understanding of their variability in space andtime, involving the following three considerations(Schumm, 2005):(a) There is a continuous spectrum of river typesor styles: meandering, single-thread sinuous,wandering or braided. These styles dependmainly on the flow, flood sediment load andsize, regimes, geologic history (particularly thevalley slope), vegetation and the occurrence ofprevious conditioning events. Different stylesof river employ different mechanisms to buildand interact with their flood plains accordingto various hydrological and geomorphologicalprocesses, resulting in distinct patterns oftemporal and spatial morphological variability,both at the surface of the river corridor andbelow it, within the alluvial aquifer;(b) Rivers change over longer timescales as a resultof climate or hydrological variability;(c) There can be a considerable amount ofvariability within any given reach as a result oflocal geomorphic and geological controls suchas tributaries, bank material variability andvegetation.Over time, driven by the flow regime, mostly byperiodic flooding, the channel moves across thevalley floor, reworking the bed and flood-plain sediments,thus destroying but also creating side arms,wetlands, ponds and a host of other riverine landscapefeatures, which are quickly colonized byriparian vegetation. Figures II.4.46 and II.4.48 showsuch processes acting in an alluvial river corridor, inthis case for streams with a meandering style. Inthis manner, the fluvial processes of erosion andsedimentation, interacting with vegetation growth,continuously modify not only the main wet channelbut indeed the entire river corridor, eventhough, from a distance, the landscape might seemunchanged, because it is in regime. This simple factexplains why changes in the flow, sediment andlarge wood regimes, often caused by water resourcesprojects, can cause wide-reaching impacts in thedownstream river corridor ecosystems: they areFigure II.4.48. Pristine river corridor of the Palena river in Chilean Patagonia, an active meanderingsingle-thread sinuous system of high ecological integrity. Note the diversity of forms, water depths andvelocities, ages of vegetation stands and the abundance of large woody debris in the channel. Water isoff colour because of glacial melt contributions. The patch of younger vegetation seen on the meanderpoint bar at the right of the picture is indicated with an arrow.

II.4-92GUIDE TO HYDROLOGICAL PRACTICESalterations in the three main ingredients of a river’sfunctioning and, as such, should cause a change inthe resulting landscape regime.As mentioned above, rivers of different style moveabout and create their flood plains through differentmechanisms. For example, meandering riversmigrate laterally by eroding the existing flood-plainmaterial on the outer side of bends and at the sametime depositing sediment on the point bar formedon the inner side, a process known as lateral accretion;they tend to create oxbow flood-plain lakes(see Figures II.4.46 and II.4.48). In contrast, wanderinggravel-bed rivers create mid-channel sedimentbars, which can be colonized by vegetation. Duringhigh discharges, the vegetation traps fine sediment,thus raising the surface of the bar by vertical accretionuntil it becomes an established island, whichlater becomes part of the flood plain when the riverabandons one of its adjacent side channels. Barssplit the flow, thus creating multiple channels oranabranches.River ecologists have found clear relationshipsbetween river type or pattern and some importantecological indicators of ecosystem health, suchas habitat complexity and biodiversity (seeFigure II.4.49). This is why it is so important in rivermanagement and restoration to consider the relationshipbetween hydrogeomorphologic processesand river styles.4.10.3.3 River morphology drives riverecologyAs previously noted, in unaltered river systems,hydrogeomorphic processes create a complexenvironment, which is highly heterogeneous,both spatially and temporally. This changingmosaic of in-channel, flood-plain and hyporheicor underground habitat patches provides sustenanceand a place to live for many differentspecies of plants and animals, both aquatic andriparian, whose life cycles have evolved inresponse to the highly dynamic and heterogeneousenvironment (Stanford and others, 1996).Thus, the river or flood-plain style can be consideredto be a geomorphological template,determining not only the plan form of the channel,but also the riverine landscape dynamicswithin the flood plain, thereby structuring thehabitat template available to organisms.Hydrology and morphology interact, setting thestage for riparian vegetation to colonize and drivingthe river corridor ecology through theestablishment of a shifting habitat mosaic.HighStreampowerBar formationAnastomosingBraidingCoalescing barsChannel meanderingor braidingBank erosionMain and secondarychannel evolutionIsland formationChannel avulsionMaximumSpring brook or pondBackwater formationMaximumBraidingMaximumMeanderprogressionPoint bar with seriesof scroll barsMaximumDecreasing avulsionMaximumTortuous meanders andcut-off oxbow lakesChannel wideningMaximumLarge woodLowFastRate of change for the shiffting habitat mosaicSlowFigure II.4.49. Continuum of river styles associated with increasing stream power, indicating the varyingrate of change for the shifting habitat mosaic, as well as some relevant hydrogeomorphic andbiogeomorphic processes (simplified from Lorang and Hauer, 2006)

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-93Intact river corridors are highly diverse and includea suite of aquatic, riparian and hyporheic habitatpatches. The habitat diversity in a reach dependson the river or flood-plain style, which determinesthe rate at which habitat patches change in spaceand time. The spatial variability is reflected in theavailability of deep and shallow waters, sandy andgravelly underground environments, old forest andyoung vegetation patches, fast and slow currents,cold or clear springbrooks and warmer or moreturbid main channels, dry gravel bars and siltywetlands and the like. The temporal variability offluvial habitats is tied to different timescales: diel(day-night) cycles affect water temperature; at theseasonal scale, flooding cycles result in inundationof the flood plain and reconnection of the riverwith its lateral aquatic habitats (ponds, wetlands,side arms), while deciduous trees shed their leaves,contributing a pulse of organic detritus; at geomorphologicaltimescales, which depend on river style,landforms are created and destroyed, thus reshapingthe habitat mosaic. Note that the concept of ashifting habitat mosaic implicitly incorporates boththe temporal (shifting) and spatial (mosaic) variabilityinherent to natural river systems.Ecologists have shown that the greater the habitatcomplexity in a river corridor, the greater the biodiversitythat it can sustain. Indeed, in order tosurvive, grow, and reproduce, organisms need foodand a place to live – a habitat – within the physicalenvironment they inhabit. Not only are theserequirements particular to each species, but a givenspecies can have different dietary and habitat needsat different stages of life, for example, a brown troutegg, fry and adult; or a nesting, versus a juvenileduck. The key here is that a certain river reach mustsupply the whole range of habitat needs for a speciesto permanently reside in it. This explains whydiverse, complex environments are able to sustain amuch higher diversity of organisms than uniformenvironments.Organisms of most species not only have varyinghabitat requirements as they age, but also at differenttimes of the day or during seasonal cycles. Thisimplies that individuals must be able to movebetween different habitat patches. They mightmove once in a lifetime, along the longitudinaldimension of the river, as is the case with somespecies of salmon migrating from the ocean towardsheadwaters, or on a daily basis, for example, whenan individual switches between a feeding positionin a riffle and its resting position under an undercutbank. Movement can take place along the transversalor lateral dimension, for example when fishspecies use lateral habitats to spawn, as shown inFigure II.4.50. Many species have patchy spatialdistributions, with few individuals per population.Movement between linked patches, which requireshigh connectivity, is keeping such species frombecoming locally extinct.4.10.4 Ecological impacts of waterresources projects4.10.4.1 The importance of change,heterogeneity and connectivityChange created by the disturbance regime isa fundamental component of a healthy river ecosystemthat needs to be maintained when deciding onthe design and operation of water resources projects,or reinstated if one wants to mitigate the impact ofexisting works. Indeed, in many cases, excellentrestoration results can be achieved by simply removingthe impact-causing factors: reinstating theoriginal flow, sediment and wood regimes to theriver without the need for further manipulation.If a highly diverse river corridor is not allowed tochange, for example by preventing the occurrenceof floods or by separating the flood plainfrom the main channel by embankments, oldpatches will no longer be destroyed and newhabitats will not be created, impeding the recruitmentof seedlings. This will result in progressiveaging of the flood-plain forests as the existingvegetation stands mature and take over the originallyheterogeneous fluvial landscape. The finalresult will be a uniform river corridor, whichsustains a lower biodiversity. This example illustratesthat not only the river’s morphology, butalso its habitat availability can be considered toBreamSilver breamBleakRoachChubRoachBleakChubStone loachBarbelNaseGudgeonDaceBrown troutGraylingPikeTenchPond loachTenchRuddBlack bullheadPumpkinseedFigure II.4.50. Use of spawning habitat by differentfish species in the wandering upper Rhoneriver, France (Roux and Copp, 1996)

II.4-94GUIDE TO HYDROLOGICAL PRACTICESbe in a state of regime, or dynamic equilibrium,where individual patches are continuously changing.However, the overall availability of differenthabitat types remains more or less the same overa reach. The conceptual model of the regimebehaviour of habitat in a river corridor wasdescribed earlier as the shifting habitat mosaic.Water resources projects tend to stabilize, oversimplifyand disconnect river corridors, resulting inspatially homogeneous conditions that are unableto provide varied habitat features for a diverse rangeof species. For example, rivers are channelizedbetween levees, resulting in uniform trapezoidalcross-sections, with no variability in depths orvelocities, and in severed connections between themain channel and the flood-plain features that areleft dry behind the dykes when the river floods. Thealluvial reach shown in Figure II.4.50 shows a varietyof aquatic environments used for spawning by awide range of fish species. If this reach were to bechannelized for navigation, it would end up as auniformly deep, narrow, single-thread channel,with much reduced heterogeneity. Many habitatswould be lost in the lateral channels and floodplain, resulting in a sharp decrease in fishbiodiversity.Unfortunately, hydraulic works generally affect ariver’s temporal dynamics, too, impeding change.For example, when floods are regulated by dams,hydrogeomorphic processes can no longer reworkthe flood-plain sediments, with the aforementionedconsequences.In high-energy fluvial systems that have large slopesand/or floods in relation to their sediment size, thedisturbance pattern, that is, the rate at which habitatsare created and destroyed, may be too fast toallow for high biodiversity. Typical examples aregravel and sand-bed braided rivers, where bars andislands have such a high turnover rate that most ofthe in-channel habitat patches are relatively recentor young. In contrast, a low-energy meanderingriver, which migrates laterally at a slow rate, mighthave a large proportion of its flood plain undermature vegetation, with little availability for youngpatches. Such a system will also be too homogeneousto sustain a large biodiversity. Indeed, it hasbeen hypothesized that diversity is maximized influvial ecosystems with an intermediate rate ofdisturbance: those with intermediate energy. Thiscorresponds to river styles between single-threadsinuous and wandering, that show a low tendencyto braiding, a high level of anabranching or formationof multiple channels, moderate meandering,high retention of large woody debris and a strongtendency to form bars, as shown in Figure II.4.49.In other words, too much change – continuouslyresetting the system, or too little change – allowingone type of habitat to dominate over all others –results in decreased diversity.Therefore, clean water is only one of many ingredientsof a healthy river ecosystem.4.10.4.2 Some recurrent impactsMost water resources projects include channelizationor river training, regulation and diversion,altering the flow, sediment and wood regimes andcutting off the ecological connectivity along ariver’s spatial dimensions, thus decreasing theecological integrity of the fluvial system. The effectsof other human-induced changes, such as pointpollution and overfishing, are generally reversibleand the mitigation strategies are obvious, except forextinction which is rarely reversible. Definitivesolutions are hard or even impossible to achieve inthe cases of diffuse or non-point pollution and theinvasion by exotic species, such as the introductionof the parasitic sea lamprey in the Great Lakeswatersheds.The effects of dams, diversions, and channelizationworks show recurrent patterns worldwide (Stanfordand others, 1996; Brookes, 1988; Petts, 1984). Theseinclude the following effects:(a) Habitat diversity and connectivity are reduced:Flow, sediment and large wood regimes arealtered, affecting the fluvial dynamics thatcreate heterogeneous in-channel and floodplainhabitat patches. The longitudinal connectivityis interrupted by dam barriers leading tofish passage problems, for example. Seasonalflow variability is reduced, but hourly or dailydischarges can fluctuate wildly. The naturaltemperature regime is lost because of hypolimneticreleases. Channelization proceduresdisconnect the wetted channel from its floodplain, altering baseflow or groundwater interaction,degrading riparian habitats, impedingseasonal flood-plain inundation – and thusreconnection to the channel – and creatingan homogeneous wetted channel. Dewateringsevers the longitudinal dimension andcan cause high mortality of aquatic organismsthrough stranding. The lack of flooding allowsvegetation to encroach upon the channel andthen mature, resulting in less diverse riparianzones. In short, hydraulic works create discontinuitiesalong the river’s spatial dimensionsand homogenize channel and flood-plain habitatconditions;

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-95(b) Native diversity decreases while exotic speciesproliferate: The altered hydrological, sedimentand temperature regimes do not provideadequate environmental conditions for mostnative species. However, the homogenizationof habitats allows exotic species to competebetter. For example, in the United States, thenative Colorado river fish species were adaptedto extreme turbidities, flows and temperatureregimes. Because of their adaptation, they faredwell where no exotic species could survive.When dams were built, however, they regulatedthe flow conditions and started releasing cold,clear hypolimnetic waters. As a result, the nonnativerainbow trout was able to invade andoutcompete the native species, driving them tothe brink of extinction.Ecosystem productivity can often be enhanced bythe changes, for example when a highly variableflow regime is regulated into a constant dischargeyear-round, or when dams release clear, nutrientladenwaters from the bottom of reservoirs. In thiscase, a handful of species can reach large populationnumbers, but this is always matched by adecrease in diversity owing to the extinction ofmany other, rarer, species that depended for theirsurvival on the temporal variability of the flowsand the associated spatial variability of the habitat.The ecological impact of water resources projectsis not always predictable quantitatively becausethe relationship between hydrology, morphologyand ecology, namely hydroecology, is not at allsimple. Certain impacts can be mitigated if theright design and operational procedures areadopted (see Petts, 1984; Brookes, 1988; Gore andPetts, 1989; Gardiner, 1991; National ResearchCouncil, 1992; Cowx and Welcomme, 1998). Forexample, selective multidepth withdrawal structurescan alleviate water quality problems and helpmaintain the original temperature regime downstreamof dams. Difficult societal and economicdecisions can be involved, as is the case when acomplete flow regime, including extremes of floodplaininundation and low-flow periods, must bedetermined or when it is desirable to allow lateralmigration of a river in order to re-establish a shiftinghabitat mosaic.4.10.5 Mitigation of ecological impactsThe most important conclusion of the abovesummary of hydroecology is as follows: ecologicallyhealthy river corridors are very complex landscapesthat depend on continued change to maintain theirshifting habitat mosaic and connectivity, and thustheir natural communities. Continued change isproduced by hydrogeomorphic forces linked toflooding disturbances.However, most of what is generally referred to asthe environmental management of rivers does notcentre around these fundamental scientific conceptsand how they can be used to attempt effectiveconservation or restoration of rivers. Instead, itfocuses constantly on two technical aspects which,in light of the complexity of river corridor systems,are of relatively minor importance, namely therestoration of the physical habitat by placing structuresin rivers, and the determination of minimuminstream flows.Most of these approaches have ignored some of thebasic principles of river behaviour. Restoring habitatby locating fixed structures in a channel goesagainst the natural tendency of a river to moveabout. Habitat is essentially dynamic – shifting, notfixed. As a river changes, it manufactures habitat.Also, this type of habitat enhancement technique isgenerally geared exclusively towards fish. Of course,such knowledge can still be useful when rehabilitatingstreams that cannot be allowed to migratelaterally, for instance in urban settings. Examples ofand further references to this approach can befound in Cowx and Welcomme (1998).Minimum instream flows (see 4.6.2.3.5), which inmany countries are called ecological or environmentalflows, have not generally led to much morethan what the name implies: a minimal, year-roundconstant flow to maintain a semblance of an aquaticecosystem. However, there is certainly scope forthese methodologies to consider some of the aspectsthat have been described previously as being fundamentalfor a river system to maintain or recover ahigh level of ecological health or integrity. In fact,as instream flow methodologies set levels of flowconsidered to be adequate for different purposes, ifthe right purposes are taken into account and themodels based on theory or field data can representthe relationships with flow, good results can beexpected.One of the main problems is that most instreamflow models were based on the end results of thecausal chain, rather than the processes that createdhabitat in the first place. Until the late 1990s, therewere a variety of approaches in use. Hydrologicalmethodologies prescribed simple percentages ormore complicated functions of the available flow.Hydraulic methods attempted to preserve a proportionof the available wet habitat based on conceptsof marginality. Habitat models computed available

II.4-96GUIDE TO HYDROLOGICAL PRACTICEShabitat for a certain life stage of a given speciesbased on habitat suitability criteria. Hydraulicmodelling of a reach under varying flows was yetanother approach. Jowett (1997) compares themand offers detailed references.The building block methodology (Tharme andKing, 1998) was the first of a new series of instreamflow models that have been labelled holistic methodologiesin that they address the flow needs of theentire riverine ecosystem, based on explicit linksbetween changes in hydrological regime and theconsequences for the biophysical environment.This approach uses findings from biological studiesin order to recommend levels of flow to meet variousecological criteria during the year, such asconnectivity with lateral spawning habitats andfish migrations. The different flows needed duringspecific months or seasons of the year are then usedas building blocks to form the overall instream flowhydrograph. Interannual variability can also beadded by specifying instream flow hydrographs fordry, average and wet years. Tharme (2003) presentsa global assessment of instream flow methodologies,comparing holistic approaches with theprevious three types and giving an inclusive listing,with references, of the 207 individual techniquesdeveloped at the time.Holistic methodologies are clearly top-downapproaches, unlike habitat models, which by definitionare bottom-up or reductionistic. If thehydrological, hydraulic and habitat models can besaid to focus on the lowest levels of the ecologicalchain, the symptoms, holistic methodologies canbe seen as focusing on the intermediate levels ofcausality. Caution must be exercised simply becauseflows do not explain all of the ecological variancein a river reach. As previously noted, the interactionbetween flow, sediment and wood regimes, aswell as reach materials and vegetation, determinethe river’s morphologic style, and thus the shiftinghabitat mosaic. The temperature regime is also relevant,as aquatic organisms are mostly ectotherms.Thus, any application of a holistic instream flowmethodology in a degraded reach must also considerthe restitution of more natural sediment, wood andtemperature patterns.Since the fundamental process attribute of a riverecosystem is the shifting habitat mosaic – whichdepends on the geomorphic template given by theriver style – it should be possible to use instreamflow methodologies to focus on these causativemechanisms, rather than looking farther along thecausal chain. While it must be recognized that riverstyle and the subsequent shifting habitat mosaicmight be determined mainly by flows above acertain threshold and organisms in the reach needto survive there year round, some work is indeedbeing undertaken in that direction. For example,Lorang and others (2005) used remotely sensedimagery to evaluate geomorphic work in a distributedfashion – pixel by pixel – across a flood-plainreach of a gravel-bed river. They did this over arange of flows, carrying out data-based modellingof stream power at each pixel, as a function of flow.Coupling this type of research with sediment transportand hydraulic models could lead tohydroecological methods that would assess themagnitude and duration of the flows needed toperform sufficient work in order to maintain theriver style and shifting habitat mosaic in a riverreach.This offers exciting possibilities for the futurebecause geomorphic work requires the integrationof variables that can be obtained by combiningmagnitude and duration in different ways, thusproviding the much sought-after flexibility. Also,such types of result could easily be added as anextra building block into holistic instream flowmethodologies, ensuring the maintenance of themain drivers of ecological health and integrity inrivers.References and further readingAbramovitz, J.N., 1995: Freshwater failures: The crises onfive continents. Washington, DC, World Watch 8,pp. 27–35.Allan, J.D., 1995: Stream Ecology: Structure and Functionof Running Waters. London, Chapman & Hall.Allen G. Richard, Luis S. Pereira, Dirk Raes and MartinSmith, 1998: Crop Evapotranspiration – Guidelines forComputing Crop Water Requirements. FAO Irrigationand Drainage Paper 56. Rome, Food and AgricultureOrganization of the United Nations.Angermeier, P.L. and J.R. Karr, 1994: Biologicalintegrity versus biological diversity as policydirectives: Protecting biotic resources. Bioscience,44(10):690–697.Apmann, R.P., 1972: Flow processes in open channelbands. Journal of the Hydraulics Division. Proceedingsof the American Society of Civil Engineers,98:785–809.ASCE, 1992: Design and Construction of UrbanStormwater Management Systems. Urban WaterResources Research Council of The American Societyof Civil Engineers and the Water EnvironmentFederation. 715 pp.Ayres, R.S. and D.W. Westcot, 1985: Water Quality forAgriculture. FAO Irrigation and Drainage Paper 29.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-97Rome, Food and Agriculture Organization of theUnited Nations.Basson, M.S., C. Triebel and J.. Van Rooyen, 1988:Analysis of a multi-baisn water resource system:A case study of the Vall River system. IWRA VI thWorld Congress on Water Rsources, Ottawa,Canada.Basson, M.S., R.B. Allen, G.G.S Pegram and J.A. VanRooyen, 1994: Probabilistic Management of WaterResource and Hydropower Systems. Highlands Ranch,Colorado,Water Resource Publications.Bastiaanssen, W.G.M., 1998: Remote-sensing inwater resources management: the state of theart. Colombo, International Water ManagementInstitute, 118 pp.Beschta, 1987: Conceptual models of sediment transportin streams. In: Sediment Transport in Gravel-Bed Rivers(C.R. Thorne, J.C. Bathurst and R.D. Hey, eds). JohnWiley & Sons, pp. 396–397.Boon, P.J., P. Calow and G.E. Petts, 1992: RiverConservation and Management. Chichester, Wiley.Box, G.E.P. and G.M. Jenkins, 1970: Time SeriesAnalysis: Forecasting and Control. San Francisco,Holden-Day.Brookes, A., 1988: Channelized Rivers: Perspectives forEnvironmental Management. Chichester, Wiley.Brookes, A. and F.D. Shields, 1996: River ChannelRestoration: Guiding Principles for Sustainable Projects.Chichester, Wiley.Calow, P. and G.E. Petts, 1994: The Rivers Handbook,Volumes 1 and 2. Oxford, Blackwell.Campana, N. and C.E.M. Tucci, 1994: Estimativa deárea impermeável de macro-bacias urbanas. RevistaBrasileira de Enghenharia – Caderno de RecursosHídricos 12:2, December.Chin, D.A., 2006: Water Quality Engineering in NaturalSystems. New York, Wiley–Interscience.Chow, V.T. (ed.), 1964: Handbook of Applied Hydrology.New York, McGraw-Hill.Chowdary, V.M., N.H. Rao and P.B.S. Sarma, 2003:GIS-based decision support system for groundwaterassessment in large irrigation project areas.Agricultural Water Management, 62:229–252.CIWEM (Chartered Institution of Water andEnvironmental Management), 1989: RiverEngineering – Part II, Structures and Coastal DefenceWorks. London.Coussot, P., 1997: Mudflow Rheology and Dynamics,IAHR Monograph Series. Rotterdam, A.A. Balkema,255 pp.Cowx, I.G. and R.L. Welcomme (eds), 1998:Rehabilitation of Rivers for Fish. Oxford, Fishing NewsBooks (Blackwell Science).Cushing, C.E. and J.D. Allan, 2001: Streams: Their Ecologyand Life. San Diego, California, Academic Press.CWC (India), 1997: Draft guidelines for planning ofconjunctive use in irrigation projects. CentralWater Commission, Ministry of Water Resources,Government of India.Danish Hydraulics Institute (DHI), 1990: MOUSEModelling of Urban Sewer Systems onMicrocomputers. Users Guide and TechnicalReference. Edition 3.0.Dastane, N.G., 1972: Effective Rainfall. FAO Irrigationand Drainage Paper 25. Rome, Food and AgricultureOrganizaiton of the United Nations.Desbordes, M and E. Servat, 1988: ‘Towards a specificapproach of urban hydrology in Africa’ In:Hydrological Processes and Water Management inUrban Areas. Proceedings Duisburg Conf., UrbanWater 88, April 231–237.Doorenbos, J. and W.O. Pruitt, 1977: Crop WaterRequirements. FAO Irrigation and Drainage Paper 24.Rome, Food and Agriculture Organization of theUnited Nations.Doorenbos, J. and A.H. Kassam, 1979: Yield Response toWater. FAO Irrigation and Drainage Paper 33. Rome,Food and Agriculture Organization of the UnitedNations.Dunbar, M.J. and M.C. Acreman, 2001: Applied hydroecologicalscience for the twenty-first century. In:Hydro-Ecology: Linking Hydrology and Aquatic Ecology(M.C. Acreman, ed.). IAHS Publication No. 266.Wallingford, IAHS Press, pp. 1–17.Engelund, F. and E. Hansen, 1967: A Monograph onSediment Transport in Alluvial Streams. Denmark,Teknisk Forlag.Engman, E.T. and R.J. Gurney, 1991: Remote-sensing inHydrology. London, Chapman and Hall, 225 pp.Federal Interagency Stream Restoration WorkingGroup (FISRWG), 1998: Stream Corridor Restoration:Principles, Processes, and Practices. GPO Item No.0120-A; SuDocs No. A 57.6/2:EN 3/PT.653.Feldman, A.D., 1995: HEC-1 flood hydrograph package.In: Computer Models of Watershed Hydrology (Vijay P.Singh, ed.). Fort Collins, Colorado, Water ResourcesPublications.Framji, K.K. (ed.), 1987: Improvement in irrigationmanagement with special reference to developingcountries. In: State-of-the-Art: Irrigation, Drainageand Flood Control, No. 4. New Delhi, InternationalCommission on Irrigation and Drainage.Framji, K.K. and B.C. Garg, 1978: Flood Control in theWorld – A Global Review, Volume II. InternationalCommission on Irrigation and Drainage.Garde, R.J. and K.G. Ranga Raju, 2000: Mechanicsof Sediment Transportation and Alluvial StreamProblems. New Delhi, New Age International (P) Ltd.Publications, 696 pp.Gardiner, J.L. (ed.), 1991: River Projects and Conservation:A Manual for Holistic Appraisal. Chichester, Wiley.Gore, J.A. and G.E. Petts (eds), 1989: Alternatives inRegulated River Management. Boca Ratonne, Florida,CRC Press.

II.4-98GUIDE TO HYDROLOGICAL PRACTICESGraf, W.H., 1984: Hydraulics of Sediment Transport.Littleton, Colorado, Water Resources Publications,513 pp.Guy, H.P., 1970: Fluvial Sediment Concepts. ChapterC1, Book 3, Techniques of Water ResourcesInvestigations. Washington, DC, United StatesGeological Survey.Hajilal, M.S., N.H. Rao and P.B.S. Sarma, 1998: Planningintra-seasonal irrigation requirements in largeprojects. Agricultural Water Management,37(2):163–182.Harper, D.M., 1991: Eutrophication of Freshwaters:Principles, Problems, and Restoration. London,Chapman-Hall.Harper, D.M. and A.J.D. Ferguson, 1995: The EcologicalBasis for River Management. Chichester, Wiley.Heaney, J.P, D. Sample and L. Wright, 2002: Costs ofurban stormwater control. USEPA Report EPA/600/R-02/021, January 2002. Cincinnati, Ohio, UnitedStates Environmental Protection Agency.Henderson-Sellers, B. and H.R. Markland, 1987:Decaying Lakes: The Origins and Control of CulturalEutrophication. Chichester, Wiley.Hipel, W.H., A.I. McLeod and W.C. Lennox, 1977:Advances in Box–Jenkins Modelling. Water ResourcesResearch, 13(3):567–575, http://fisher.stats.uwo.ca/faculty/aim/vita/pdf/Advances1.pdf.Hughes, D.A. and P. Hannart, 2003: A desktop modelused to provide an initial estimate of the ecologicalinstream flow requirements of rivers in South Africa.Journal of Hydrology, 270(3–4):167–181.Hydrologic Engineering Center (HEC), 1977: Storage,Treatment, Overflow, Runoff Model STORM: User’smanual, Generalized Computer Program 723-S8-l7520.Davis, California, Hydrologic Engineering Center,Corps of Engineers.Hynes, H.B., 1972: The Ecology of Running Waters.Toronto, University of Toronto Press.Intergovernmental Panel on Climate Change (IPCC),2001: Climate Change 2001: The Scientific Basis. NewYork: Cambridge University Press.———, 2007: Fourth Assessment Report – Climate Change2007. Geneva.International Atomic Energy Agency (IAEA), 1981:Ultimate Heat Sink and Directly Associated GeatTransport Systems for Nuclear Power Plants: A SafetyGuide. Safety Series No. 50-SG-D6. Vienna.International Commission on Irrigation and Drainage(ICID), 1996: Manual on Non-structural Approaches toFlood Management. Delhi.Jeffries, M.J. and D.H. Mills, 1995: Freshwater Ecology:Principles and Applications. Chichester, Wiley.Jowett, I.G., 1997: Instream flow methods: A comparisonof approaches. Regulated Rivers: Research andManagement, 13:115–127.Karr, J.R., 1996: Ecological integrity and ecologicalhealth are not the same. In: Engineering WithinEcological Constraints (P.C. Schulze, ed.). Washington,DC,National Academy of Engineering, pp. 97–109.Karr, J.R. and E.W. Chu, 1999: Restoring Life in RunningWaters. Washington, DC, Island Press.Kurtas W.P. and J.M. Norman, 1996: Use of remotesensingfor evapotranspiration monitoring over landsurfaces. Journal of Hydrological Sciences, 41:495–516.Lawrence, I., 2001: Integrated urban land and watermanagement – Planning and design guidelines.Cooperative Research Centre for Freshwater Ecology,University of Canberra, ACT 2601. Technical Report1/2001. February 2001. Australia.Leopold, L.B., 1994: A View of the River. Cambridge,Massachusetts, Harvard University Press.Lorang, M.S and F.R. Hauer, 2006: Fluvial geomorphicprocesses. In: Methods in Stream Ecology (F.R. Hauerand G.A. Lamberti, eds). Second edition, Chapter 7.San Diego, Academic Press/Elsevier.Lorang, M.S., D.C. Whited, F.R. Hauer, J.S. Kimball andJ.A. Stanford, 2005: Using airborne multispectralimagery to evaluate geomorphic work across floodplains of gravel-bed rivers. Ecological Applications,15(4):1209–1222.Loucks, D.P., J.R. Stedinger and D.A. Haith, 1981:Water Resources Systems Planning and Analysis.Prentice-Hall.Maidment, D.R. (ed.), 1992: Handbook of Hydrology. NewYork, McGraw Hill.Maksimovic, C. and M. Radojkovic, 1986: Urban DrainageCatchments – Selected Worldwide Rainfall-Runoff Datafrom Experimental Catchments. Oxford, PergamonPress.Meier, C.I. 1998. The ecological basis of river restoration:1. River ecology for hydraulic engineers. 2.Defining restoration from an ecological perspective.Proceedings of the Conference EngineeringApproaches to Ecosystem Restoration, Denver,Colorado. Reston, Virginia, American Society ofCivil Engineers.Morisawa, M.E., 1985: Rivers: Form and Process.Geomorphology Texts, No. 7. London, Longman.Munich Re, 2006: Annual Review: Natural Catastrophes2005, Knowledge series, Topics Geo. Munich. http://www.munichre.com/publications/302-04772_en.pdf.Musiake, K., T. Oki, T. Nakaegawa and K. Wakasa,1995: Verification experiment of extraction of soilmoisture information using SAR mounted on JERS-1and ERS-1, Final report of JERS-1/ERS-1 SystemsVerification Programme–Ministry of InternationalTrade and Industry, Kyoto, pp. 617–624.National Research Council, 1973: Proceedings of theNinth Hydrology Symposium on Fluvial Processesand Sedimentation, 8–9 May 1973, Edmonton.Ottawa, Ontario, Inland Waters Directorate,Department of the Environment.———, 1992: Restoration of Aquatic Ecosystems.Washington, DC, National Academy Press.

CHAPTER 4. APPLICATIONS TO WATER MANAGEMENTII.4-99Negev, M., 1972: Suspended Sediment Discharge inWestern Watersheds of Israel. Hydrological Paper 14.Jerusalem, Hydrological Service.Neill, C.R., 1973: Guide to Bridge Hydraulics. Roads andTransportation Association of Canada. Toronto,University of Toronto Press.Nuttle, W.K., 2002: Is ecohydrology one idea or many?Hydrological Sciences Journal, 47:805-807.Odum, E. and G.W. Barrett, 2004: Fundamentals ofEcology. Fifth edition. Brooks Cole.Pegram, G.G.S. and R.S. McKenzie, 1991: Syntheticstreamflow generation in the Vaal River systemstudy. Proceedings of the South African Institutionof Civil Engineering, January, pp. 15–24.Permanent International Association of NavigationCongress (PIANC), 1987: Guidelines for the designand construction of flexible revetments incorporatinggeotextiles for inland waterways, Reportof Working Group 4 to Permanent TechnicalCommittee 1. Brussels.Petts, G.E., 1984: Impounded Rivers: Perspectives forEcological Management. Chichester, Wiley.Petts, G.E. and C. Amoros (eds), 1996: FluvialHydrosystems. London: Chapman and Hall.Prihar, S.S., K.L. Khera, K.S Sandhu and B.S. Sandhu,1976: Comparison of irrigation schedules based onpan evaporation and growth stages in winter wheat,Agronomy Journal, 68:650–653.Prosser, M.J., 1986: Propeller-Induced Scour, ReportRR2570, British Hydromechanics ResearchAssociation. Bedford, Cranfield.Rao, N.H., P.B.S. Sarma and Subhash Chander, 1988a: Asimple dated water-production function for use inirrigated agriculture. Agricultural Water Management,13(1):25–32.Rao, N.H., P.B.S. Sarma and Subhash Chander, 1988b:Irrigation scheduling under a limited water supply.Agricultural Water Management, 15(2):165–175.Rao, N.H. and D.H. Rees, 1992: Irrigation schedulingof rice with a crop growth simulation model.Agricultural Systems, 39(2):115–132.Raudkivi A.J., 1998: Loose Boundary Hydraulics.Rotterdam, A.A. Balkema, 496 pp.Rickenmann D., 1991: Hyperconcentrated flow and sedimenttransport at steep slopes. Journal of HydraulicEngineering, 117(11):1419–1439.Roux, A.L. and G.H. Copp. 1996. Fish populations inrivers. In: Fluvial Systems (G.E. Petts and C. Amoros,eds). Chapter 8, pp. 167–183. London: Chapman &Hall.Ryding, S.O. and W. Rast (eds), 1989: The Control ofEutrophication of Lakes and Reservoirs. Man and theBiosphere Series. Paris: UNESCO.Santos, R.J.L., 1998: Incerteza na estimativa dohidrograma de projeto com base em modeloshidrológicos. MSc thesis. Porto Alegre, Institute ofHydraulic Research.Schulze, R.E. and L.A. Perks, 2000: Assessment of theImpact of Climate Change on Hydrology andWater Resources in South Africa. ACRU ReportNo. 33, School of Bioresources Engineering andEnvironmental Hydrology, University of Natal,South Africa.Schumm, S.A., 2005: River Variability and Complexity.Cambridge, Cambridge University Press.SCS, 1975: Urban Hydrology for Small Watersheds.Technical Release 55. Washington, United StatesDepartment of Agriculture.Simons, D.B. and E.V. Richardson, 1961: Forms ofbed roughness in alluvial channel. Journal of theHydraulics Division, 87(3):801–105.Singh, B., 1989: Fundamentals of Irrigation Engineering,Roorkee, Nem Chand & Bros.Smart, G.M., 1984: Sediment transport formula forsteep channels. Journal of Hydraulic Engineering,(110)3:267–276.Stanford, J.A., J.V. Ward, W.J. Liss, C.A. Frissel, R.N.Williams, J.A. Lichatowich and C.C. Coutant, 1996:A general protocol for restoration of regulatedrivers. Regulated Rivers: Research and Management,12(4–5):391–413.Stringham, G.E., 1988: Surge Flow Irrigation. UtahAgricultural Experiment Station Research Bulletin,515 pp.Stringham, G.E. and J. Keller, 1979: Surge flow forautomatic irrigation. ASCE Irrigation and DrainageDivision Speciality Conference, Albuquerque, NewMexico, pp. 131–142.Svanidze, G.G., 1977: Mathematicheskoe modelirovaniegidrologicheskikh ryadov (Mathematical modellingof hydrological series). Leningrad, Gidrometeoizdat.Also published in English: 1980, Water ResourcesPublications LLC.Szöllösi-Nagy, A. and C. Zevenbergen (eds), 2005: UrbanFlood Management. London: A.A. Balkema Publishers.Tharme, R.E., 2003: A global perspective on environmentalflow assessment: emerging trends in thedevelopment and application of environmentalflow methodologies for rivers. River Research andApplications, 19(5–6):397–441.Tharme, R.E. and J.M. King, 1998. Development ofthe building block methodology for instream flowassessments, and supporting research on the effectsof different magnitude flows on riverine ecosystems.Water Research Commission Report No. 576/1/98,South Africa.Thomann, R.V. and J.A. Mueller, 1987: Principles ofSurface Water Quality Modeling and Control. HarperCollins Publishers.Thorne C. R., J.C. Bathurst and R.D. Hey, 1987: SedimentTransport in Gravel-Bed Rivers. John Wiley and Sons.Tucci, C.E.M. (ed.), 2001: Urban Drainage in HumidTropics. Volume 1 of Urban Drainage in SpecificClimates (C. Maksimovic, ed.). IHP-V Technical

II.4-100GUIDE TO HYDROLOGICAL PRACTICESDocuments in Hydrology No. 40, . Paris, UNESCO.http://unesdoc.unesco.org/images/0012/001228/122848eo.pdf.United Nations, 1977: Mar del Plata Action Plan. UnitedNations Water Conference, Argentina.United Nations 1987: Our Common Future. WorldCommission on Environment and Development.Oxford, Oxford University Press.United Nations, 1992: International Conference onWater and the Environment: Development Issuesfor the Twenty-first Century, 26–31 January 1992,Dublin, Ireland.United Nations, 2000: Millennium Development Goals,Millennium Declaration. New York.United Nations Educational, Scientific and CulturalOrganization/World Meteorological Organization(UNESCO/WMO), 1992: International Glossary ofHydrology. Third edition.———, 1997: Water Resources Assessment Activities:Handbook for Review of National Capabilities.Urbonas, B. and P. Stahre, 1993: Stormwater BestManagement Practices and Detention. EnglewoodCliffs, New Jersey, Prentice Hall.US Army Corps of Engineers (USACE), 1995: FloodProofing Regulations. https://www.nwo.usace.army.mil/nfpc/fpreg/ace6.htm.Van Rijn, L.C., 1984: Sediment transport, part III: Bedforms and alluvial roughness. Journal of HydraulicEngineering, 110(12):1732–1754.Vanoni, V.A. (ed.), 1975: Sedimentation Engineering. ASCEManual 54. New York, American Society of CivilEngineers.Vidal, A. and J.A. Sagardoy (eds), 1995: Use of Remote-Sensing Techniques in Irrigation and Drainage. WaterReports 4. Proceedings of the Expert Consultation,Montpellier, France, 2–4 November 1993. Rome,Food and Agriculture Organization of the UnitedNations, 202 pp.World Commission on Dams (WCD), 2000: Dams andDevelopment. A New Framework for Decisionmaking.The report of the World Commission onDams. London, Earthscan Publications.World Health Organization, 1988: Urbanization and itsImplications for Child Health: Potential for Action.Geneva.World Meteorological Organization, 1981: Meteorologicaland Hydrological Aspects of Siting and Operation ofNuclear Power Plants. Volume II – HydrologicalAspects. Technical Note No. 170, WMO-No. 550,Geneva.———, 1989: (Long, Yuqian), Manual on OperationalMethods for the Measurement of Sediment Transport.WMO-No. 686, Geneva.———, 1992: Hydrological Aspects of Accidental Pollution ofWater Bodies, Operational Hydrology Report No. 37,WMO No. 754, Geneva.———, 1999: Contaminants in Rivers and Streams– Prediction of Travel Time and lLongitudinalDispersion. Operational Hydrology Report No. 45,WMO-No. 886, Geneva.———, 2003: Manual on Sediment Management andMeasurement. Operational Hydrology Report No. 47,WMO-No. 948, Geneva.———, APFM, 2004: Integrated Flood ManagementConcept Paper, APFM Technical Document No. 1,second edition. Associated Programme on FloodManagement, Geneva.———, APFM, 2006a: Social Aspects and StakeholderInvolvement in Integrated Flood Management, APFMTechnical Document No. 4, Flood ManagementPolicy Series. Associated Programme on FloodManagement, Geneva.———, APFM, 2006b: Legal and Institutional Aspectsof Integrated Flood Management, APFM TechnicalDocument No. 2, Flood Management Policy Series.Associated Programme on Flood Management,Geneva.———/GWP, 2006: Environmental Aspects of IntegratedFlood Management. APFM Technical DocumentNo. 3, Flood Management Policy Series. WorldMeteorological Organization/Global WaterPartnership, Geneva.———/United Nations Educational, Scientific andCultural Organization, 1991: Progress in the implementationof the Mar del Plata Action Plan and astrategy for the 1990s. Report on Water ResourcesAssessment.Wright-McLaughlin Engineers, 1969: Urban Storm DrainageCriteria Manual. Two volumes. Denver, Colorado.Yalin M.S., 1992: River Mechanics. Oxford, PergamonPress, 220 pp.Yalin, M.S. and A.M. Ferreria da Silva, 2001: FluvialProcesses. Monograph. Delft, InternationalAssociation of Hydraulic Engineering and Research.Yen, B.C., 1986: Rainfall-runoff process on urban catchmentsand its modelling. In: Proceedings of the InternationalSymposium on Comparison of Urban Drainage Models withReal Catchments Data, 9–11 April 1986, Dubrovnik(C. Maksimovic and Radojkovic, eds).Zalewski, M., 2000: Ecohydrology: The scientificbackground to use ecosystem properties as managementtools toward sustainability of water resources.Ecological Engineering, 16:1–8.Zhaohui Wan and Zhaoyin Wang, 1994:Hyperconcentrated Flow, IAHR Monograph Series,Rotterdam, A.A. Balkema, 290 pp.

CHAPTER 5EXTREME VALUE ANALYSIS5.1 INTRODUCTIONThe purpose of frequency analysis is to analyse pastrecords of hydrologic variables so as to estimatefuture occurrence probabilities. The data used inthe analysis must be evaluated in terms of the objectives,length of records available and completenessof records. It must also satisfy certain statisticalcriteria such as randomness, independence, homogeneityand stationarity. A frequency analysis canbe performed using single-site data, regional data orboth. It can also include historical information andreflect physical constraints.Because hydrological phenomena are characterizedby great variability, randomness and uncertainty, itshould, therefore be recognized that statistical analysisof hydrological data will not always yield a trueanswer. The sources of uncertainty in frequencyanalysis include representativness of the analyticalapproach, selection of the probability distributionand estimation of parameters.Hydrological analysis is generally based on wellestablishedprinciples of hydrodynamics,thermodynamics and statistics. However, thecentral problem in hydrological analysis is theapplication of these principles in a natural environmentthat is non-homogeneous, sparsely sampledand only partially understood. The events sampledare usually unplanned and uncontrolled. Analysesare performed to obtain spatial and temporal informationabout hydrological variables, regionalgeneralizations and relationships among the variables.Analyses can be performed using deterministic,parametric, probabilistic and stochastic methods.An analysis based on the deterministic approachfollows the laws that describe physical and chemicalprocesses. In the parametric approach, ananalysis is performed by intercomparison of hydrologicaldata recorded at different locations andtimes. In the probabilistic approach, the frequencyof occurrence of different magnitudes of hydrologicalvariables is analysed. In the stochastic approach,both the sequential order and the frequency ofoccurrence of different magnitudes are analysedoften using time-series methods. Evidence continuesto accumulate documenting the dynamic andnonlinear character of the hydrological cycle. Inthe case of extreme events, our major interest is notin what has occurred, but the likelihood that furtherextreme and damaging events will occur at somepoint in the future.The occurrence of many extreme events in hydrologycannot be forecasted on the basis of deterministicinformation with sufficient skill and lead time. Insuch cases, a probabilistic approach is required toincorporate the effects of such phenomena intodecisions. If the occurrences can be assumed to beindependent in time, in that the timing and magnitudeof an event bears no relation to precedingevents, then frequency analysis can be used todescribe the likelihood of any one or a combinationof events over the time horizon of a decision.Hydrological phenomena commonly described byfrequency analysis include storm precipitation(5.7), low flows (5.8) and annual flood maxima(5.9).Both the detail and precision of the analysis shouldbe consistent with the quality and samplingadequacy of the available data and with the accuracyrequired by the application of the analysis.Consideration should be given to the relationshipbetween the cost and time devoted to an analysisand to the benefits expected. Traditionally, graphicaland very simple computational methods haveproven more cost effective than more sophisticatedmethods, and they may be sufficiently accurate forthe data and purposes involved. However, the widespreadavailability of personal computingequipment, with general-purpose statistical softwareand computation environments such asspreadsheets, has largely replaced hand computationalprocedures. A major advantage of the moderncomputational environment is that it shouldimprove an agency’s ability to store, retrieve andanalyse data. Further, the graphical capabilities ofpersonal computers should greatly enhance theability of hydrologists to review and understandtheir data, as well as the results and the computationsthat they perform.5.2 STATISTICAL SERIES AND RETURNPERIODS [HOMS H83]In frequency analysis, a series is a convenientsequence of data, such as hourly, daily, seasonal orannual observations of a hydrological variable. If

II.5-2GUIDE TO HYDROLOGICAL PRACTICESthe record of these observations contains all theevents that occurred within a given period, theseries is called a complete duration series. Forconvenience, the record often contains only eventsof magnitude above a pre-selected base or thresholdlevel; this is called a partial duration series orpeaks-over-threshold series. A series that containsonly the event with the largest magnitude thatoccurred in each year is called an annual maximumseries.The use of the annual maximum series is verycommon in frequency analyses for two reasons.The first is for convenience, as most data areprocessed in such a way that the annual series isreadily available. The second is that there is asimple theoretical basis for extrapolating thefrequency of annual series data beyond the rangeof observation. With partial series data, suchtheory is not as simple because one must considerthe arrival process of floods within a year andthe distribution of the magnitude of floods whenthey do occur. Another problem with partialduration series is the lack of independence ofevents that might follow one another in closesequence, as well as seasonal effects. However, ifthe arrival rate for peaks over the threshold islarge enough and can be modelled by simpletwo-parameter distributions, for example 1.65for the Poisson arrival with exponential exceedancesmodel, it should yield more accurateestimates of flood quantiles than the correspondingannual flood frequency analyses. However,when fitting a three-parameter distribution,such as the generalized Pareto distribution forexceedances with Poisson arrivals, there appearsto be no advantage in using a partial durationseries no matter how many floods are recordedon average each year (Martins and Stedinger,2000). It should not be a surprise that recordingthe value of a great many small events does nottell us much about the risk of very large eventsoccurring unless the structure of the model isfairly rigid.A limitation of annual series data is that each yearis represented by only one event. The secondhighest event in a particular year may be higherthan the highest in some other years, yet it wouldnot be contained in the series. The use of partialduration series can address this issue because allpeaks above the specified threshold areconsidered.The complete duration series may be required forthe stochastic approach in which independence isnot required. It may also serve for a probabilisticanalysis of data from arid regions where the eventsare rare and almost independent.The return period T of a given level is the averagenumber of years within which the event is expectedto be equalled or exceeded only once. The returnperiod is equal to the reciprocal of the probabilityof exceedance in a single year. If the annual exceedanceprobability is denoted 1/T a, where T ais theannual return period, the relationship betweenthe annual return period and the return period inthe partial duration series can be expressed asfollows:1/T a= 1 – exp {– λ q e} = 1 – exp {– l/T p} (5.1)where T p= 1/(λ qe) is the average return period inthe partial duration series with λ being the arrivalrate for peaks over the threshold and q eis the probabilitythat when such a flood occurs, it exceeds thelevel of concern. This equation can be solved for T pto obtain:T p= 1 / ln [1 – 1/T a] (5.2)T pis less than T abecause more than one event canoccur per year in a partial duration series. Forreturn periods exceeding ten years, the differencesin return periods obtained with the annual andpartial series is inconsequential. Table II.5.1compares the return periods for an annual maximumseries and a partial duration series. Thisformula is based on the assumption that floods inthe partial duration series occur independently intime and at a constant rate; relaxation of thatassumption yields different relationships (Robsonand Reed, 1999). NERC (1975) observes that theactual probabilistic model for arrivals with regardto large return period events is not particularlyimportant, provided that different models yieldthe same average number of arrivals per year (seealso Cunnane, 1989).Table II.5.1. Corresponding return periods forannual and partial seriesPartial series0.501.001.452.005.0010.00Annual series1.161.582.002.545.5210.50

CHAPTER 5. EXTREME VALUE ANALYSISII.5-35.3 PROBABILITY DISTRIBUTIONS USEDIN HYDROLOGY [HOMS H83, X00]Probability distributions are used in a wide varietyof hydrological studies, including studies ofextreme high and low flows, droughts, reservoirvolumes, rainfall quantities and in time-seriesmodels. Table II.5.2 lists the most commonly useddistributions in hydrology. Their mathematicaldefinitions are given in a number of references(Kite, 1988; Cunnane, 1989; Bobee and Ashkar,1991; Stedinger and others, 1993; Clark, 1994;Kottegoda and Rosso, 1997 and Hosking andWallis, 1997).Numerous probability distributions have beenintroduced in the literature to model hydrologicalphenomena such as extreme events. Despite intensiveresearch and study, no particular model isconsidered superior for all practical applications.The user should, therefore, screen available modelsin the light of the problem to be solved and thenature of the available data. Consequently, onlysome distributions that are in common use arereviewed in this chapter. The contending distributionsthat fit the observed data satisfactorily usuallydiffer significantly in the tail of the distribution,especially when extrapolation is involved. Nogeneral guidance is available for extrapolatingdistributions, particularly beyond twice the availablerecord length. The decision regarding whichdistribution to use should be based on the comparisonof the suitability of several candidatedistributions. The advantages and disadvantages ofthe various methods that can be used for this objectiveare discussed in 5.6.Annual totals, such as flow volumes or rainfalldepths, tend to be normally distributed or almostso because of the forces described by the centrallimit theorem of statistics. Monthly and weeklytotals are less symmetric, displaying a definiteskewness that is mostly positive and cannot usuallybe modelled by the normal distribution. Annualextremes – high or low – and peaks over a thresholdtend to have skewed distributions. The part of asample that lies near the mean of the distributioncan often be described well by a variety ofdistributions. However, the individual distributionscan differ significantly and very noticeably fromone another in the values estimated for large returnperiods, as well as very small cumulativeprobabilities. As hydraulic design is often based onestimates of large recurrence-interval events, it isimportant to be able to determine them as accuratelyas possible. Hence, the selection of the distributionis very important for such cases. The choice ofdistributions is discussed in the references citedabove, which include discussions on the methodsavailable for choosing between distributions. Thisis also discussed in 5.6.Generally, mathematical distributions having threeparameters, such as those appearing in Table II.5.2,are selected so as to make the distribution matchesthe available data more consistently. In some casesan empirical distribution can be used to describethe data, thereby avoiding the use of mathematicalparametric distributions.Use of a mathematical distribution has severaladvantages:(a) It presents a smooth and consistent interpretationof the empirical distribution. As a result,quantiles and other statistics computed usingthe fitted distribution should be more accuratethan those computed with the empiricaldistribution;(b) It provides a compact and easy-to-use representationof the data;(c) It is likely to provide a more realistic descriptionof the range and likelihood of values thatthe random variable may assume. For example,by using the empirical distribution, it is implicitlyassumed that no values larger or smallerthan the sample maximum or minimumcan occur. For most situations this is entirelyunreasonable.There are several fundamental issues that arise inselecting a distribution for frequency analysis(Stedinger and others, 1993):(a) What is the true distribution from which theobservations are drawn?(b) Is a proposed flood distribution consistent withavailable data for a particular site?(c) What distribution should be used to obtainreasonably accurate and robust estimates offlood quantiles and flood risk for hydrologicaldesign purposes?Unfortunately, the answer to the first question willnever be known, and it might not be much help ifit were. The true distribution of the data could beincredibly complex with more parameters than ahydrologist could ever hope to estimate. Thus, theaim is to establish a good, but simple approximationof the true distribution of the events. Standardgoodness-of-fit statistics and probability plots can,at least in part, address the second question, as theywill sometimes show that particular distributionsare not consistent with the available data. Theremay be pragmatic considerations to prevent the useof a distribution for a particular sample. For example,

II.5-4GUIDE TO HYDROLOGICAL PRACTICESthe distribution may be upper bounded at what isconsidered to be an unreliable low value, therebynot providing an acceptable estimate of extremeconditions. As a practical matter, many nationalagencies look at the problem from the point of viewof the third question: What distribution coupledwith a reasonable fitting procedure will yield goodestimates of risk in their region of the world? Thus,the aim is not to seek absolute truths. Instead, thegoal is to develop practical procedures which, withthe data in hand or data that can be collected, willprovide a good approximation of the frequencyrelationships of interest. Over the past four decades,various distributions have been introduced for usein hydrological frequency analysis. The followingsection provides an overview of some of thesedistributions.5.3.1 Normal family: N, LN and LN35.3.1.1 Normal distributionThe normal distribution (N) is useful in hydrologyfor describing well-behaved phenomena, such asthe total annual flow. The probability density functionfor a normal random variable X is given inTable II.5.2, and it is unbounded both above andbelow, with mean µ xand variance σ 2 x. The normaldistribution’s skewness coefficient is zero, becausethe distribution is symmetric. The cumulativedistribution function (CDF) of the normal distributionis not available in closed form, but books onstatistics include tables of the standardized normalvariate z p. The quantity z pis an example of afrequency factor because the p th quantile x pof anynormal distribution with mean µ and variance σ 2can be written as follows:x p= µ + σ z p(5.3)5.3.1.2 Log-normal distributionIn general, flood distributions are positively skewedand not properly described by a normal distribution.In many cases the random variablecorresponding to the logarithm of the flood flowswill be adequately described by a normal distribution.The resulting two-parameter log-normal (LN)distribution has the probability-density functiongiven in Table II.5.2. Often, the logarithms of arandom variable X are not distributed normally. Insuch cases, introducing a boundary parameter ζbefore taking logarithms can solve this problem,yielding a three-parameter log-normal distribution(LN3) (Stedinger and others, 1993) so that:Y = ln [X – ζ] (5.4)would have a normal distribution. Thus:X = ζ + exp (Y) (5.5)has a LN3 distribution. In terms of the frequencyfactors of the standard normal distribution z p, thequantiles of a log-normal distribution are asfollows:x p= ζ + exp (µ Y+ σ Yz p) (5.6)where µ Yand σ Yare the mean and standard deviationof Y. Parameter estimation procedures arecompared by Stedinger (1980).5.3.2 Extreme value distributions:Gumbel, generalized extreme valueand WeibullGumbel (1958) defined three types of extreme valuedistributions which should describe the distributionof the largest or smallest value in a large sample.They have been widely used in hydrology to describethe largest flood or the lowest flow.5.3.2.1 Gumbel distributionAnnual floods correspond to the maximum of all ofthe flood flows that occur within a year. Thissuggests their distribution is likely to be a memberof a general class of extreme value (EV) distributionsdeveloped in Gumbel (1958). Let X 1,...,X nbe aset of annual maximum discharges and letX = max{X i}. If the X iare independent and identicallydistributed random variables unboundedabove, with an exponential-like upper tail, then forlarge n the variate X has an extreme value (EV)type I distribution or Gumbel distribution with cumulativedistribution function given in Table II.5.2.Landwehr and others (1979) and Clarke (1994)discuss estimation procedures and Hosking (1990)has shown that L–moments provide accurate quantileestimates for the small sample sizes typicallyavailable in hydrology.5.3.2.2 Generalized extreme valuedistributionThe generalized extreme value distribution spansthe three types of extreme value distributions formaxima. The Gumbel and generalized extremevalue distribution distributions are widely used forflood frequency analyses around the world(Cunnane, 1989). Table II.5.2 provides the cumulativedistribution function of the generalized extremevalue distribution.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-5Table II.5.2. Commonly used frequency distributions (after Stedinger and others, 1993)Distribution Probability density function and/or cumulative distribution function Range MomentsNormal f X(x) =12πσ X2⎡exp – 1 2⎛ x – μ X⎞ ⎤⎢ ⎜2 ⎝ σ⎟ ⎥ X ⎠⎣ ⎢ ⎦ ⎥– ∞ < x < ∞ µ Xand σ 2 X , γ X = 0Log-normal af X(x) =1x 2πσ exp ⎡– 1 2⎛ ln(x)–μ Y⎞ ⎤⎢ ⎜ ⎟ ⎥ 2Y2 ⎝ σ Y ⎠⎣ ⎢ ⎦ ⎥ 0 < x µX= exp [µ Y+ σ 2 Y /2]σ 2 X = µ2 X {exp [σ2 Y ] – 1}γ X= 3CV X+ CV 3 Xtype III for 0 < β : ξ < x and γ X= 2/√α (for 0 < β and ξ = 0: γ X= 2 (CV X) for β < 0: x < ξ and γ X= –2/√α _Pearson f X(x) = |β| [β(x – ξ)] α–1 exp [– β(x – ξ)]/ Γ(α) 0 < α µ X= ξ + α/β; σ 2 X = α/β2Log-Pearson f X(x) = |β| {β[ln(x) – ξ]} α–1 exp {– β[ln(x) – ξ]}/xΓ(α) See Stedinger and others (1993).type III for β < 0, 0 < x < exp (ξ); for 0 < β, exp (ξ) < x < ∞Exponential f X(x) = β exp {– β(x – ξ)} ξ < x for 0 < β µ X= ξ + 1/β; σ 2 X = 1/β2F X(x) = 1 – exp {– β(x – ξ)} γ X= 2Gumbel f X(x) = (1/α) exp {– (x–ξ)/α – exp [– (x–ξ)/α]} – ∞ < x < ∞ µ X= ξ + 0.5772 αF X(x) = exp {– exp [– (x–ξ)/α]} σ2 2 X = π2 α 2 /6 = 1.645α 2 ;γ X= 1.1396Generalized F (x) = exp {– [1 – κ(x–ξ)/α] 1/κ } (σ2exists for –0.5 < κ) µ = ξ + (α/κ) [1 – Γ(1+κ)]extreme value XXXwhen 0 < κ, x < (ξ+α/κ); κ < 0, (ξ+α/κ) < x σ 2 X = (α/κ)2 {Γ(1+2κ) – [Γ(1+κ)] 2 }Weibull f X(x) = (k/α) (x/α) k–1 exp [ – (x/α) k ] 0 < x ; 0 < k, α µ X= α Γ(1 + 1/k)]XF (x) = 1 – exp [ – (x/α) k ] σ 2 X = α2 {Γ(1 +2/k) – [Γ(1 +1/k)] 2 }Generalized y = [1 – κ(x–ξ)/α] 1/κ for κ ≠ 0 y = exp [–(x−ξ)/α] for κ = 0logistic f X(x) = (1/α) [y (1–κ) /(1 + y)] 2 for κ < 0, ξ + α/κ ≤ x < ∞ µ = ξ + α/[1/κ–π/sin(κπ)]F X(x) = 1/[1 + y] for 0 < κ , −∞ < x ≤ ξ + α/κ XSee Ahmad and others (1998) for σ 2 X .Generalized f (x) = (1/α) [1 – κ(x–ξ)/α] 1/κ–1 for κ < 0, ξ ≤ x < ∞ µ = ξ + α/(1+κ)Pareto XF X(x) = 1 – [1 – κ(x–ξ)/α] 1/κ for 0 < κ, ξ ≤ x ≤ ξ + α/κ Xσ 2 X = α2 /[(1+κ) 2 (1+2κ)](γ Xexists for κ > – 0.33) γ X= 2(1−κ)(1+2κ) 1/2 /(1+3κ)HalphenType A f X(x) =Type B f X(x) =1⎡2m v K v(2α) xv–1 exp – α⎛ x m + m ⎞⎢⎣ ⎝ x ⎠2m 2v ef v(α) x2v–1 exp – x 2⎛ ⎞⎝ m ⎠⎡⎣+α⎛ x ⎞⎝ m ⎠⎢ ⎢⎤⎡Type B –1 f X(x) = 2m2vef v(α) x–2v–1 exp –⎛ m 2⎞+α⎛ m ⎞⎤⎢ ⎝ x ⎠ ⎝ x ⎠a Here Y = ln(X). A three-parameter log-normal distribution with Y = ln(X – ξ) is also commonly used.b Kν = modified Bessel function, second kind.c efν (α) = exponential factorial function.⎣⎤⎦⎥ for x > 0; m > 0; α > 0; – ∞ < α < ∞ b⎥ for x > 0; m > 0; ν > 0; – ∞ < α < ∞ c See Marlat (1956).⎦⎥ for x > 0; m > 0; ν > 0; – ∞ < α < ∞ c⎦

II.5-6GUIDE TO HYDROLOGICAL PRACTICESThe Gumbel distribution is a special case of generalizedextreme value distribution corresponding toκ = 0. Here, x is a location parameter, α is a scaleparameter, and κ is the important shape parameter.For κ > 0 the distribution has a finite upper boundat ξ + α/κ; for κ < 0, the distribution has a thickerright-hand tail and is unbounded above.Hosking and others (1985) describe the L–momentprocedure that is effective with this distribution. L–moments have been the basis of many regional andindex-flood procedures that make use of the generalizedextreme value distribution (Hosking andWallis, 1997). More recently, Martins and Stedinger(2000) present generalized maximum likelihoodestimators for the generalized extreme value distributionthat are more accurate than L–momentestimators over the range of hydrological interest.5.3.2.3 Two-parameter Weibull distributionIf W iare the minimum streamflow in different daysof the year, then the annual minimum is the smallestof the W i, each of which is bounded below byzero. In this case the random variable X = min {W i}may be described well by the extreme value type IIIdistribution for minima, or the Weibull distribution(see Figure II.5.1 and Table II.5.2). For k < 1, theWeibull probability density goes to infinity as xapproaches zero, and decays slowly for large valuesof x. For k = 1, the Weibull distribution reduces tothe exponential distribution corresponding to γ = 2.For k > 1, the Weibull density function is like adensity function of Pearson type III distribution forsmall values of x and α P3= k, but decays to zero fasterfor large values of x. Parameter estimation methodsare discussed in Kite (1988).5.3.3 Pearson type III familyThe Pearson type III (P3) distributions are commonlyused to fit a sample of extreme hydrological data. Atheoretical description of this distribution can befound in Bobée and Ashkar (1991) and a summary inMaidment’s Handbook of Hydrology, Chapter 18(Stedinger and others, 1993). The notations of thatpublication are used in the following. The probabilitydensity function of the P3 distribution, given inTable II.5.2, is defined by three parameters: ζ (location),β (scale) and α (shape). The method of momentsconsidering mean, variance and coefficient of skewnessis used by the Interagency Advisory Committeeon Water Data (1982) to fit the P3 distribution todata. Caution should be exercised in using moments,as they may yield an upper bound which might besmaller than an observed flood. The method of maximumlikelihood can also be used (Pilon and Harvey,1992). This distribution can be used for both positivelyand negatively skewed samples.The log-Pearson type III distribution (LP3) describesa variable x whose logarithm y = log x is P3 distributed.This distribution was recommended for thedescription of floods in the United States of Americaby the United States Water Resources Council,initially in 1966 and then again by the InteragencyAdvisory Committee on Water Data in 1982. It wasalso adopted for use in Canada as one of severalother methods (Pilon and Harvey, 1992).5.3.4 Halphen family: types A, B and B –1This family of distributions was specifically designedto model floods and more generally, extremes. Theprobability density function of these distributions(Perreault and others, 1999a) are given inTable II.5.2. Perreault and others (1999b) presentedprocedures for estimating parameters, quantiles andconfidence intervals for the Halphen distributions.The Gamma and inverse Gamma (x is the inverseGamma distributed if y = 1/x follows Gamma distributions)are limiting cases of the Halphen distributions.Figure II.5.1. The probability density function forthe Pearson type III distribution with lower boundζ = 0, mean μ = 1 and coefficients of skewnessγ = 0.7, 1.4, 2.0 and 2.8 (corresponding to agamma distribution and shape parametersα = 8, 2, 1 and 0.5, respectively)Although the probability density function of theHalphen distributions are mathematically morecomplicated than the three-parameter distributionscurrently used in hydrometeorology, that shouldnot be a serious obstacle for their use in practice,since the Halphen distributions can be applied withthe aid of user-friendly software such as HYFRAN(www.ete.inrs.ca/activites/groupes/chaire_hydrol/hyfran.html).

CHAPTER 5. EXTREME VALUE ANALYSISII.5-75.3.5 Generalized logistic distributionThe generalized logistic distribution was introducedto the mainstream of the hydrological literature byHosking and Wallis (1997) and was proposed as thedistribution for flood frequency analysis in theUnited Kingdom (Robson and Reed, 1999). Theparameterization is similar to the generalizedextreme value distribution, and both have Paretoliketails for large values of x. The cumulativedistribution function of the generalized logisticdistribution is given in Table II.5.2, as is the rangeof the variable. Hosking and Wallis (1997) andRobson and Reed (1999) document how the threeparameters of this distribution can be obtainedfrom L–moment estimators.5.3.6 Generalized Pareto distributionThe generalized Pareto distribution has a verysimple mathematical form (see Table II.5.2) and isuseful for modelling events that exceed a specifiedlower bound at which the density function has amaximum (κ < 1). Examples include daily rainfalldepths and all floods above a modest threshold.Hosking and Wallis (1987) discuss alternative estimationprocedures. Often the value of the lowerbound can be determined by the physical constraintsof the situation, so that only two parameters needbe estimated. If the physical situation does notdictate the value of the lower bound, then thesmallest observation may suffice as an estimator ofthe lower bound for x.A very interesting relationship exists between thegeneralized Pareto distribution and the generalizedextreme value distribution. If peaks in a partialduration series arrive as in a Poisson process andhave magnitudes described by a generalized Paretodistribution, then the annual maxima greater thanthe partial duration series threshold have a generalizedextreme value distribution with the samevalue of κ (Stedinger and others, 1993). Wang(1991) and Martins and Stedinger (2001) explorethe relative efficiency of the two modellingframeworks.5.3.7 Non-parametric density estimationmethodThe non-parametric method does not require eitherthe assumption of the functional form of the overalldensity function, or the estimation of parametersbased on the mean, variance and skew. The nonparametrickernel density estimation requires theselection of a kernel function K, which is a probabilitydensity function, and the calculation of asmoothing factor H. Then, using a sample of Nobservations of the variable x, an approximation ofthe probability density function for the variable x isobtained by assigning each x ja probability of 1/Nand then using the kernel function to spread outthat probability around the value of each x jtoobtain the following equation:f ( x ) =1NHN∑Ki=1⎛ x – x i ⎞⎝ H ⎠(5.7)The principle of a kernel estimator as expressed bythe above equation is that a kernel of prescribedform, triangular, normal, or Gumbel distributionfunction is associated with each observation over aspecified scale, expressed by H. The weighted sumof these functions constitutes the non-parametricestimate of the density function. The optimal valueof H can be determined based on a cross-validationprocedure (Adamowski, 1985) and is available in acomputer software package (Pilon and others,1992).5.4 HYPOTHESIS TESTINGThe data series must meet certain statistical criteriasuch as randomness, independence, homogeneityand stationarity in order for the results of afrequency analysis to be theoretically valid. Thesestatistical criteria are explained in Table II.5.3,where appropriate statistical tests are indicated. Amore detailed description of many of these testscan be found in Helsel and Hirsch (1992). Wellknownstatistical parametric tests such as the t-testand the F-test are not included in the table becausehydrological data series often do not satisfy someconditions for strict applicability of these tests,particularly the assumption of normality, whichcan adversely impact upon the power of parametrictests (Yue and Pilon, 2004). The tests indicated inthe table are of a non-parametric type, which avoidsassumptions regarding the underlying parametricdistribution of the data. Care should be taken toverify the assumptions underlying the tests, asviolation may lead to unreliable results (Yue andothers, 2002a).Statistical tests can only indicate the significance ofthe observed test statistics and do not provideunequivocal findings. It is therefore important toclearly understand the interpretation of the resultsand to corroborate findings with physical evidenceof the causes, such as land use changes. When datado not satisfy the assumptions, then a transformationcan often be employed so that the transformed

II.5-8GUIDE TO HYDROLOGICAL PRACTICESTable II.5.3. Statistical tests and statistical criteria (after Watt, 1989)Criterion Explanation Applicable statistical testsRandomnessIndependenceIn a hydrologic context, randomness means essentially thatthe fluctuations of the variable arise from natural causes.For instance, flood flows appreciably altered by reservoiroperation are unnatural and therefore cannot be consideredas random, unless the effect of the regulation is removedfirst.Independence implies that no observation in the data serieshas any influence on any following observations. Even ifevents in a series are random, they may not be independent.Large natural storages, in a river basin, for example, maycause high flows to follow high flows and low flows tofollow low flows. The dependence varies with the intervalbetween successive elements of the series: dependenceamong successive daily flow values tends to be strong,while dependence between annual maximum values isgenerally weak. Likewise, the elements of annual seriesof short-duration rainfall may, in practice, be assumed tobe independent. In some cases, however, there may besignificant dependence even between annual maximumvalues, for example in the case of rivers flowing through verylarge storages such as the Great Lakes of North America.No suitable tests for hydrologicalseries are available.– Anderson as described in Chow(1964).– Spearman rank order serialcorrelation coefficient as describedin NERC (1975).HomogeneityStationarityHomogeneity means that all the elements of the dataseries originate from a single population. Elderton (1953)indicated that statistics are seldom obtained from strictlyhomogeneous material. For instance, a flood series thatcontains both snowmelt and rainfall floods may not behomogeneous; however, depending on the results of a test,it may be acceptable to treat it as such. When the variabilityof the hydrological phenomenon is too high, as in the case ofextreme precipitation, non-homogeneity tends to be difficultto decipher (Miller, 1972), but non-homogeneity in yearlyprecipitation sums is easier to detect.Stationarity means that, excluding random fluctuations,the data series is invariant with respect to time. Types ofnon-stationarity include trends, jumps and cycles. In floodanalysis, jumps are generally due to an abrupt change ina basin or river system, such as the construction of a dam.Trends may be caused by gradual changes in climaticconditions or in land use , such as urbanization. Cycles maybe associated with long-term climatic oscillations.Terry (1952).– Spearman rank correlationcoefficient test for trend (NERC,1975)– Wald–Wolfowitz (1943) test fortrend. No satisfactory method oftesting is available for long-periodcycles.– Mann–Kendall test for trend (Yueand others, 2002b)observations would meet the criteria required foranalysis. Caution is advised in interpolation andextrapolation when data do not meet theassumptions.5.4.1 Wald–Wolfowitz test forindependence and stationarityGiven the data sample of size N (x 1,....,x N), theWald–Wolfowitz test considers the statistic R sothat:When the elements of the sample are independent,R asymptotically follows asymptotically normaldistribution with mean and variance given by thefollowing equations:_R = (s 2 1 – s 2) / (N – 1) (5.9)Var (R) = (s 2 2 – s 4 ) / (N – 1) – R _2(5.10)+ (s 14– 4s 1 2 s 2+ 4s 1s 3+ s 22– 2s 4) / (N – 1) (N – 2)N−1R = x i x i+1 + x 1 x N∑i=1(5.8)with s r= Nm ’ r and m’ r is the rth moment of thesample about the origin.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-9_The quantity (R – R) / (Var (R)) 1/2 follows a standardizednormal distribution (mean 0 and variance 1)and can be used to test at level α the hypothesis ofindependence by comparing |n| with the standardnormal variate u α/2corresponding to a probabilityof exceedance α/2.5.4.2 Mann–Kendall test for trenddetectionThe Mann–Kendall test is a rank-based non-parametrictest for assessing the significance of a trend.The null hypothesis H 0is that a sample of dataordered chronologically is independent and identicallydistributed. The statistic S is defined as follows(Yue and others, 2002b):n−1 nS = ∑ sgn(x j − x i )(5.11)∑i=1 j=i+1wheresgn (x) ={01 if x > 0if x = 0 (5.12)–1 if x < 0When n ≥ 40, the statistic S is asymptoticallynormally distributed with mean 0 and variancegiven by the following equation:Var{ S}= 118⎡⎣∑n (n − 1)(2n + 5)− t (t − 1)(2t + 5)⎦t(5.13)where t is the size of a given tied group andt∑ is thesummation over all tied groups in the data sample.The standardized test statistic K is computed byusing the following equation:⎤For independent sample data without trend, the Pvalue should be equal to 0.5. For sample data withlarge positive trend, the P value should be close to1.0, whereas a large negative trend should yield a Pvalue close to 0.0. If the sample data are seriallycorrelated, then the data should be pre-whitenedand a correction applied to calculate the variance(Yue and others, 2002b).The slope of a trend is estimated as follows:β = median⎛⎝x i − x j ⎞i − j ⎠ , ∀j < i(5.16)where β is the estimate of the slope of the trend andx jis the j th observation. An upward trend is representedby a positive value of β and a downwardtrend is represented by a negative value of β.5.4.3 Mann–Whitney test forhomogeneity and stationarity(jumps)We now consider two samples of size p and q (withp ≤ q) the combined set of size N = p + q is ranked inincreasing order. The Mann–Whitney test considersthe following quantities:V = R – p(p+1) / 2 (5.17)W = pq – V (5.18)where R is the sum of the ranks of the elements ofthe first sample of size p in the combined series andV and W are calculated from R, p and q. V representsthe number of times that an item in sample 1follows in the ranking an item in sample 2; W canalso be computed in a similar way for sample 2following sample 1.S − 1Var (S)K = 0S + 1Var (s)⎧⎪⎨⎪⎩If S > 0If S = 0If S < 0(5.14)The test statistic, U, is defined by the smaller of Vand W. When N > 20, and p, q > 3, and under thenull hypothesis that the two samples come fromthe same population, U is approximately normallydistributed with mean:U – = pq/2 (5.19)The standardized statistic K follows the standardnormal distribution with mean zero andvariance of one. The probability value P of thestatistic K of sample data can be estimated usingthe normal cumulative distribution functionas:P =12πze −t2 /2 dt(5.15)∫−∞and variance:Var(U ) =⎡⎣pqN ( N− 1)⎤ ⎡⎦ ⎣N 3 − N−12∑T⎤⎦ (5.20)with T = (J 3 – J) / 12, where J is the number of observationstied at a given rank. The summation ΣT isover all groups of tied observations in both samplesof size p and q. For a test at a level of significance,

II.5-10GUIDE TO HYDROLOGICAL PRACTICESthe quantity |u| = |(U – U – )/Var(U) 1/2 | is comparedwith the standardized normal quantile u α/2correspondingto a probability of exceedance α /2.5.4.4 Sample size and length of recordThe definition of a stable distribution for estimatingfuture probabilities of occurrence of ahydrological phenomenon requires that the lengthof record or sample size must be sufficiently long.In estimating daily extreme precipitation, Sevrukand Geiger (1981) report that the length of recordneeded to obtain a stable distribution is related tothe general humidity of the region and its physiographicconditions that determine the variability ofthe daily precipitation sum. As indicated inTable II.5.3, when the variability of the hydrologicalphenomenon is too high, difficulties in testingthe homogeneity of the hydrological series canarise. When the coefficient of variation of a sampledrawn from a skewed distribution is large (largevariability), the standard error of the sample coefficientof skewness which is used to fit the assumeddistribution will also be large. Sevruk and Geiger(1981) argue that for extreme precipitationfrequency analysis a 25-year period of record maybe sufficient in humid regions such as the northernRussian Federation, but even a 50-year period is notadequate in other regions where a distinct periodicfluctuation of precipitation exists. According tothese authors, a record of 40 to 50 years is, ingeneral, satisfactory for extreme precipitationfrequency analysis. Yue and others (2002a) and Yueand Pilon (2004) show, as well, how statistical characteristicsof the sample and record length canimpact upon the power of common statisticaltests.United States Water Resources Council (1981). Toapply this test, the assumption must be made thatthe logarithms or some other function of the hydrologicalseries are normally distributed because thetest is applicable only to samples from a normalpopulation. It is common to make the simpleassumption used by the United States WaterResources Council that the logarithms of the samplevalues are normally distributed. To apply the Grubbsand Beck test, the following two quantiles arecalculated:X H= exp (x _ + K Ns) (5.21)X L= exp (x _ + K Ns) (5.22)where x _ and s are the mean and standard deviationof the natural logarithms of the sample, respectively,and K Nis the Grubbs and Beck statistictabulated for various sample sizes and significancelevels. At the 10 per cent significance level, thefollowing polynomial approximation proposed byPilon and Harvey (1992) can be used for estimatingthe tabulated values:K(N) = –3.62201 + 6.2844N ¼– 2.49835N ½ + 0.491436N ¾ – 0.037911N(5.23)where N is the sample size. In applying the Grubbsand Beck test, any sample values greater than X Hare considered to be high outliers and those lessthan X Lare considered to be low outliers. For5≤N≤150, K(N) can be computed from the followingequation (Stedinger and others, 1993):K(N) = –0.9043 + 3.345 √log (N)– 0.4046 log (N)(5.24)5.4.5 Grubbs and Beck test for detectionof outliersAn outlier is defined as a data point that is far fromthe bulk of the data. The presence of outliers in adata sample can cause difficulties when attemptingto fit a distribution to the sample. There may existhigh or low outliers, or both, in a sample, and thesecan have different impacts on the frequency analysis.Although the problem of treating outliers is stillsubject to much discussion, certain procedures havebeen used in hydrology for their identification andtreatment, such as those described by the UnitedStates Water Resources Council (1981) for floodfrequency analysis or by Sevruk and Geiger (1981)for extreme precipitation.The Grubbs and Beck test for the detection ofoutliers is the test that is recommended by the5.4.6 Bayesian proceduresWhile the frequency estimation of probability isbased on the idea of an experiment that can berepeated several times, the Bayesian approach isbased on a personal assessment of probability andprovides an opportunity to take into account anyinformation that is available, by means of the priordistribution. Unlike classical models, Bayesianmodels consider the parameters of the problem asrandom variables rather than fixed values. Forexample, in the case of the detection of shifts in themean of a time series, classical statistical methodsassume knowledge of the time of the possible shift.The Bayesian approach, however, does not makeany assumptions concerning knowledge of the timeof the shift. This allows the approach to make inferenceson its characteristics, such as the changepoint and the amount of shift.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-11Perreault and others (1999c) and other authors havepresented Bayesian models for the detection of asingle shift in the mean. Perreault and others (2000)presented a method for a change in variability andapplied it to hydrological data, while Asselin andothers (1999) presented a bivariate Bayesian modelfor the detection of a systematic change in themean. A complete description of the Bayesian statisticalinference theory is presented in Box and Tiao(1973).5.5 POPULATION STATISTICS ANDPARAMETER ESTIMATIONAssuming that extreme events are described properlyby some family of distributions, a hydrologist’stask is to estimate the parameters of the distributionso that required quantiles and expectationscan be calculated with the fitted model. The statisticaland hydrological literature contains manymethods and philosophies for estimating theparameters of different distributions: those mostcommonly employed are outlined below.5.5.1 Parameter calculation methodsPerhaps the simplest approach is the method ofmoments, which computes estimates of the parametersso that the theoretical moments of adistribution match the computed sample moments.The recommended procedure for federal agenciesin the United States (Thomas, 1985; InteragencyAdvisory Committee on Water Data, 1982) usesthe moments of the logarithms of the floods flowsX = log Q.A variation on the method of moments, which hasproved effective in hydrology with the generalizedextreme value distribution, is the method of probability-weightedmoments or equivalentlyL–moments (Hosking and others, 1985; Hoskingand Wallis, 1997). Probability-weighted momentsor the corresponding L–moments provide a differentway to summarize the statistical properties ofhydrological datasets (Hosking, 1990). An advantageof L–moment estimators are that they are linearcombinations of the observations and thus do notinvolve squaring and cubing the observations. As aresult, the L-coefficient of variation and L-skewnessare almost unbiased, whereas the product-momentestimators of the coefficient of variation and coefficientof skewness are highly biased and highlyvariable (Vogel and Fennessey, 1993). This is ofparticular value for regionalization procedures,which is further discussed in 5.9.L–moments are another way to summarize thestatistical properties of hydrological data based onlinear combinations of the original data (Hosking,1990). Recently, hydrologists have found thatregionalization methods that use L–moments aresuperior to methods that use traditional moments.They have also worked well for fitting some distributionswith on-site data (Hosking and others,1985). The first L–moment is the arithmetic mean:λ 1= E[X] (5.25)Let X (i|n)be the i th largest observation in a sample ofsize n (i = 1 corresponds to the largest). Then, forany distribution, the second L–moment is a descriptionof scale based on the expected differencebetween two randomly selected observations:λ 2= (1/2) E[X (1|2)– X (2|2)] (5.26)Similarly, L–moment measures of skewness andkurtosis use:λ 3= (1/3) E[X (1|3)– 2 X (2|3)+ X (3|3)] (5.27)λ 4= (1/4) E[X (1|4)– 3 X (2|4)+ 3 X (3|4)– X (4|4)] (5.28)Just as product moments can be used to definedimensionless coefficients of variation and skewness,L–moments can be used to define adimensionless L-coefficient of variation and anL–coefficient of skewness (Table II.5.4). L–momentestimators have often been computed based on anintermediate statistics called probability-weightedmoments (Hosking, 1990; Hosking and Wallis,1997; Stedinger and others, 1993). Many early studiesused probability-weighted moment estimatorsbased on plotting positions (Hosking and others,Table II.5.4. Dimensionless statistics used todescribe distributions (product–moment andL–moment ratios)Name Denotation DefinitionProduct–moment ratiosCoefficient of variation CV Xσ X/µ XCoefficient of skewness a γ XE[(X – µ X) 3 ] / σ3XCoefficient of kurtosis b – E[(X – µ X) 4 ] / σ4XL–moment ratios cL–coefficient of variation L–CV, τ 2λ 2/λ 1L–coefficient of skewness L–skewness, τ 3λ 3/λ 2L–coefficient of kurtosis L–kurtosis, τ 4λ 4/λ 2a Some texts define β1 = [γ ] 2 as a measure of skewness.b x Some texts define the kurtosis as {E[(X – μx ) 4 ]/σ 4 – 3}; others use the term excessxkurtosis for this difference because the normal distribution has a kurtosis of 3.c Hosking (1990) uses τ instead of τ2 to represent the L–CV ratio.

II.5-12GUIDE TO HYDROLOGICAL PRACTICES1985); these were later found to lack the consistencyand invariance required of such estimators(Hosking and Wallis, 1995; Fill and Stedinger, 1995),so that subsequent work has shifted to use the unbiasedprobability-weighted moment estimators.Direct estimation of unbiased L–moments from asample is described by Wang (1996).A method that has very strong statistical motivationis maximum likelihood. It chooses theparameters which make a fitted distribution asconsistent as possible, in a statistical sense, with theobserved sample. Maximum likelihood estimatorsare discussed in general statistics textbooks and arerecommended for use with historical and paleofloodrecords because of their ability to makeparticularly efficient use of censored and categoricaldatasets.Non-parametric methods can be employed to estimatethe flood–flow frequency relationship,offering the advantage that one need not assumethat floods are drawn from a particular parametricfamily of distributions. These methods have beenadopted in Canada (Pilon and Harvey, 1992).5.5.2 Use of logarithmic transformationsWhen data vary widely in magnitude, whichfrequently occurs in water quality monitoring, thesample product-moments of the logarithms of dataare often employed to summarize the characteristicsof a dataset or to estimate distributionparameters. A logarithmic transformation is aneffective vehicle for normalizing values that vary byorder of magnitude, and for preventing occasionallylarge values from dominating the calculation ofproduct-moment estimators. However, the dangerof using logarithmic transformations is that unusuallysmall observations or low outliers are givengreatly increased weight. This is of concern whenlarge events are of interest and when small valuesare poorly measured. Small values may reflectrounding errors or may be reported as zero whenthey fall below a certain threshold.5.5.3 Historical informationIn addition to a relatively brief period of systematicmeasurements, there may be additionalhistorical information available that pertains, forexample, to the magnitude of floods prior to thecommencement of the systematic collection ofrecords. A gauging station might have only 20years of measurement records as of 1992, yet itmight be known that in 1900 a flood occurredwith a peak which exceeded any flood measuredand was also the greatest flood since the communitywas established in 1860. The magnitude ofthis flood and the knowledge that the other floodsfrom 1860 to 1992 were less than the flood of 1900should and can be used in the frequency analysis.In other instances, it may only be known that acertain number of floods from 1860 to 1972exceeded a certain threshold. This is also historicalinformation and should be included in thefrequency analysis. Different processes generatehistorical and physical paleoflood records. Floodsleaving a high-water mark are the largest to haveoccurred during the corresponding period, whereasslackwater sediment deposits in protected areascan provide evidence of the magnitude of a numberof large floods.Apart from the routine monitoring of streamflow,certain floods may be recorded simply becausethey exceed a perception level and have sufficientlydisrupted human activities for theiroccurrence to have been noted, or for the resultantphysical or botanical damage to be available withwhich to document the event (Stedinger and Baker,1987; Wohl, 2000). Several methods can be used toincorporate historical information into the estimationof the parameters of the mathematicaldistribution function. They are historicallyadjusted weighted moments, maximum likelihood,the expected moments algorithm and thenon-parametric method (Cohn and others, 2001;England and others, 2003; Griffis and others,2004). It has been shown that the maximum likelihoodmethod makes more efficient use of theadditional information than historically weightedmoments. Maximum likelihood estimators andexpected moments algorithms are both very flexibleand appear to be equally efficient with the LP3distribution for which expected moments algorithmswere developed, though maximumlikelihood estimators often have convergenceproblems with those distributions.5.5.4 Record augmentationIt is often possible to effectively extend a shortrecord using a longer record from a nearby stationwith which observations in the the short record arehighly correlated. In particular, a long series from anearby station can be used to improve estimates ofthe mean and variance of the events that occur atthe short-record site. For this purpose, it is notnecessary to actually construct the extended series;one only needs the improved estimates of themoments. This idea of record augmentation isdeveloped in Matalas and Jacobs (1964); see alsothe Interagency Advisory Committee on Water Data

CHAPTER 5. EXTREME VALUE ANALYSISII.5-13(1982), (Appendix 7). Recent improvements and adiscussion of the information gain are provided byVogel and Stedinger (1985). In other instances, alonger series can be created that will be employedin simulation or will be archived. The idea of usingrecord extension to ensure that generated flowshave the desired mean, variance and correlations isdeveloped by Hirsch (1982), Vogel and Stedinger(1985) and, where multivariates are concerned, byGrygier and others (1989).5.5.5 Analysis of mixed populationsA common problem in hydrology is that annualmaximum series are composed of events that mayarise from distinctly different processes. For example,precipitation may correspond to different stormtypes in different seasons, such as summer thunderstorms,winter frontal storms, and remnants oftropical hurricanes or snowmelt. Floods arisingfrom these different types of events may havedistinctly different distributions. Waylen and Woo(1982) examined summer runoff and winter snowmeltfloods separately. Vogel and Stedinger (1984)studied summer rainfall and winter ice-jam-affectedfloods. Hirschboeck and others (2000) consideredthe categorization of different floods above a specificthreshold into classes based on the prevailingsynoptic weather pattern; this resulted in a mixedpopulationflood analysis using a partial durationseries framework. In some mountainous regions insmall basins, summer thunderstorms produce thelargest floods of record, but snowmelt eventsproduce most of the maximum annual events. Insuch instances, as illustrated by Waylen and Woo(1982), separation of the flood record into separateseries can result in a better estimate of the probabilityof extreme events because the data describingphenomena that produce those large events isbetter represented in the analysis.Suppose that the annual maximum series M tis themaximum of the maximum summer event S tandthe maximum winter event W t:M t= max {S t, W t} (5.29)Here S tand W tmay be defined by a rigidly specifiedcalendar period, a loosely defined climatic period,or the physical and meteorological characteristicsbetween the phenomena that generated theobservations.If the magnitudes of the summer and winter eventsare statistically independent, meaning that knowingone has no effect on the conditional probabilitydistribution of the other, the probabilitydistribution for the annual maximum event M isgiven by (Stedinger and others, 1993):F M(m) = P[M = max(S, W) ≤ m] = F S(m) F W(m) (5.30)For two or more independent series of eventscontributing to an annual maximum, the distributionof the maximum is easily obtained. If severalstatistical-dependent processes contribute to anannual maximum, the distribution of the maximumis much more difficult to calculate from thedistributions of the individual series. An importantissue is that of deciding whether it is advisable tomodel several different component flood seriesseparately, or whether it is just as reasonable tomodel the composite maximum annual seriesdirectly. If several series are modelled, then moreparameters must be estimated, but more data areavailable if the annual maximum series or thepartial duration series for each type of event isavailable.The idea of the mixing of two distributions led tothe devolopment of a two-component extremevalue ß distribution by Rossi and others (1984),which corresponds to the maximum of two independentEV1 distributions. It can be thought of asthe maximum of two flood processes in a partialduration series, each with Poisson arrivals andexponentially distributed flood peaks. Generally,one of the two distributions is thought of as describingthe bulk of the data, and the other as the outlierdistribution. Because the model has four parameters,it is very flexible (Beran and others, 1986).Therefore, if only the annual maximum series areused, regional estimation methods are essential toresolve the values of all four parameters, makingregional two-component extreme value estimatorsan attractive option. The two-component extremevalue distribution has been successfully employedas the basis of index flood procedures (Frances,1998; Gabriele and Villani, 2002). The non-parametricdistribution (Adamowski, 1985) and Wakebydistribution (Pilon and Harvey, 1992) can also beused to model the mixture distribution.5.5.6 Frequency analysis and zerosLow-flow series often contain years with zero values,while some sites’ maximum series may also containzero values for some sites. In some arid areas, zeroflows are recorded more often than non-zero flows.Streamflows recorded as zero imply either that thestream was completely dry, or that the actualstreamflow was below a recording or detectionlimit. This implies that some low-flow series arecensored datasets. Zero values should not simply be

II.5-14GUIDE TO HYDROLOGICAL PRACTICESignored and do not necessarily reflect accuratemeasurements of the minimum flow in a channel.Based on the hydraulic configuration of a gaugeand on knowledge of the rating curve and recordingpolicies, it is possible to determine the lowestdischarge that can be reliably estimated and wouldnot be recorded as a zero. The plotting positionmethod and the conditional probability model arereasonable procedures for fitting a probability distributionwith datasets containing recorded zeros. Thegraphical plotting position approach, without aformal statistical model, is often sufficient for lowflowfrequency analyses. The low-flow frequencycurve can be defined visually and the parameters ofa parametric distribution can be estimated by usingprobability-plot regression as described by Kroll andStedinger (1996) and Stedinger and others (1993),or by using non-parametric methods.5.6 PROBABILITY PLOTS ANDGOODNESS-OF-FIT TESTS5.6.1 Plotting positions and probabilityplotInitial evaluation of the adequacy of a fitted probabilitydistribution is best done by generating aprobability plot of the observations. When thesorted observations are plotted against an appropriateprobability scale, except for sampling fluctuation,they fall approximately on a straight line.Such a plot serves both as an informative visualdisplay of the data and a check to determinewhether the fitted distribution is consistent withthe data.Such plots can be generated with special commerciallyavailable probability papers for somedistributions, including the normal, two-parameterlog-normal and Gumbel distributions, all of whichhave a fixed shape. Thanks to modern software,however, it is generally easier to generate such plotswithout the use of special papers (Stedinger andothers, 1993). The i th largest observed flood x (i)isplotted versus the estimated flood flow associatedwith the exceedance probability, or probabilityplottingposition q i, assigned to each ranked floodx (i); x (1)> x (2)> . . . > x (n). The exceedance probabilityof the i th largest flood x (i)can be estimated by any ofseveral reasonable formulae. Three commonly usedare the Weibull formula with p i= i / (n + 1), theCunnane formula with p i= (i – 0.40) / (n + 0.2), andthe Hazen formula with p i= (i – 0.5) / n. Cunnane(1978) and Adamowski (1981) provide a discussionof the plotting position issue. Plotting positions forrecords that contain historical information is developedin Hirsch and Stedinger (1987). Hydrologistsshould remember that the actual exceedance probabilityassociated with the largest observation in arandom sample has a mean of 1/(n+1) and a standarddeviation of nearly 1/(n+1) (Stedinger andothers, 1993); thus all of the plotting positions giveonly crude estimates of the relative range of exceedanceprobabilities that could be associated with thelargest events (Hirsch and Stedinger, 1987).5.6.2 Goodness-of-fit testsSeveral rigorous statistical tests are available and areuseful in hydrology to determine whether it isreasonable to conclude that a given set of observationswas drawn from a particular family ofdistributions (Stedinger and others, 1993). TheKolmogorov–Smirnov test provides bounds withinwhich every observation on a probability plotshould lie if the sample is actually drawn from theassumed distribution (Kottegoda and Rosso, 1997).The probability plot correlation test is a more effectivetest of whether a sample has been drawn froma postulated distribution (Vogel and Kroll, 1989;Chowdhury and others, 1991). Recently developedL–moments can be used to assess if a proposedGumbel-, generalized extreme value- or normaldistribution is consistent with a dataset (Hosking,1990; Chowdhury and others, 1991). Discussion ofthe development and interpretation of probabilityplots is provided by Stedinger and others, (1993)and Kottegoda and Rosso (1997).5.6.3 Information criteriaMany approaches have been suggested for thecomparison of flood distributions. Goodness-of-fittests have been applied to assess the suitability ofdifferent probability distributions for describingannual maximum flow series, and to evaluatesimulated samples in the case of simulation studies.These tests establish which distributions are,in general, the most appropriate for flood modeling.To assess the quality of a fitted model, Akaike(1974) introduced an information criterion calledAIC, which stands for Akaike information criterion.It can be adapted to many different situationsand consists in minimizing an information measure.The information criterion is defined asfollows:AIC(f) = –2log L(θ^, x) + 2k (5.31)where L(θ^, x) is the likelihood function, and k is thenumber of parameters. According to Akaike (1974),

CHAPTER 5. EXTREME VALUE ANALYSISII.5-15the model that better explains the data with theleast number of parameters is the one with thelowest Akaike information criterion. To select anappropriate model, some compromises between thegoodness of fit and the complexity of themodel must be accepted. Alone, the Akaike informationcriterion is not appropriate for modelselection.A Bayesian extension of the minimum Akaike informationcriterion concept is the Bayesian informationcriterion called BIC. It is defined as follows:BIC(f) = –2log L(θ^, x) + k log(n) (5.32)where L(θ^, x) the likelihood function, k is is thenumber of parameters and n is the sample size. TheBayesian information criterion is also a parsimonycriterion. Of all the models, the one with the lowestBayesian information criterion is considered to bebest. The Schwarz method (1978) is often used toobtain the Bayesian information criterion . However,this method can also be used to get an asymptoticapproximation of a Bayes factor. Furthermore, itcan be combined to an a priori probability distributionto obtain the a posteriori probability for eachdistribution of a given set of distributions. Bayesianinformation criteria have not yet been used muchin hydrology, however. The above-mentionedmethods, which merit broader use, are available inHYFRAN software. Ozga-Zielinska and others (1999)developed a computer package for calculatingdesign floods when a sufficiently long period ofrecord is available. There are many other computerpackages, including those listed in HOMS.5.7 RAINFALL FREQUENCY ANALYSIS[HOMS I26, K10, K15]The frequency of occurrence of rainfall of differentmagnitudes is important for various hydrologicalapplications. In particular, rainfall frequency analysesare used extensively to plan and designengineering works that control storm runoff, suchas dams, culverts, urban and agriculture drainagesystems. This is because, in most cases, good-qualityflow data of a length adequate for the reliableestimation of floods are generally limited or unavailableat the location of interest, while extensiveprecipitation records are often available. In general,there are two broad categories of approaches forestimating floods from precipitation data: thoseemploying the statistical analysis of precipitationdata and those based on the deterministic estimationof the so-called probable maximumprecipitation (PMP). While it has been used worldwidefor the design of various large hydraulicstructures, probable maximum precipitation doesnot provide probability estimates for risk assessmentwork. The main part of this section focuses,therefore, on statistical rainfall estimation methodsthat can provide both flood magnitudes and associatedprobabilities; the second part deals with theestimation of extreme rainfall. The theory andapplications of PMP have been well documented inhydrological and engineering literature such as theManual for Estimation of Probable MaximumPrecipitation (WMO-No. 332) and NRCC (1989); andare summarized in 5.7.5.6.The main objective of rainfall frequency analysis isto estimate the amount of precipitation falling at agiven point or over a given area for a specified durationand return period. Results of this analysis areoften summarized by intensity–duration–frequencyrelationships for a given site or are presented in theform of a precipitation frequency atlas, whichprovides rainfall accumulation depths for variousdurations and return periods over the region ofinterest. For instance, estimates of rainfall frequenciesfor various durations, ranging from 5 minutesto 10 days, and return periods from 1 to 100 yearsare available. Such data can be found for the UnitedStates in the US Weather Service and Atlas series ofthe National Oceanic and AtmosphericAdministration (Frederick and others, 1977), forAustralia in Australian Rainfall and Runoff: A Guideto Flood Estimation (Pilgrim, 1998), for Canada inthe Rainfall Frequency Atlas for Canada (Hogg andCarr, 1985) or in the Handbook on the Principles ofHydrology (Gray, 1973) and for the United Kingdomin the Flood Estimation Handbook (Institute ofHydrology, 1999).Basic considerations of frequency analysis of hydrologicaldata are discussed in 5.1 to 5.6, whereasspecial applications for rainfall analysis are coveredin 5.7. The statistical methods described hereinapply to storm or other short-duration rainfall data.Similar methods are used for flood peaks, floodvolumes, low flows, droughts and other extremeevents. In particular, the selection of distributiontypes for extremes of precipitation is discussed byWMO (1981).5.7.1 Assessment of rainfall data forfrequency analysisRainfall data used for frequency analysis are typicallyavailable in the form of annual maximumseries, or are converted to this form using continuousrecords of hourly or daily rainfall data. These

II.5-16GUIDE TO HYDROLOGICAL PRACTICESseries contain the largest rainfall in each completeyear of record. An alternative data format for precipitationfrequency studies is partial duration series,also referred to as peaks over threshold data, whichconsist of all large precipitation amounts abovecertain thresholds selected for different durations.The difference in design rainfall estimates usingannual maximum and partial duration series wasfound to be important for short return periods oftwo to five years but insignificant for long returnperiods of ten years or longer (Chow, 1964;Stedinger and others, 1993).As for any statistical analyses, both the quantityand quality of the data used are important. Theprecipitation data should be collected for a longperiod of time. A sufficiently long record of precipitationdata provides a reliable basis for frequencydeterminations. It is known that a data sample ofsize n, in the absence of a priori distributionalassumptions, can furnish information only aboutexceedance probabilities greater than approximately1/n (NRC, 1988). It is a common rule ofthumb to restrict extrapolation of at-site quantileestimates to return periods (years) of up to twice aslong as the record length (NRCC, 1989). Hence,long-term precipitation data are extremely valuablefor determining statistically based rainfall estimatesof reasonable reliability, especially for extreme rainfallswith high return periods, such as those greaterthan 100 years.The quality of precipitation data may affect itsusability and proper interpretation in floodfrequency analysis studies. Precipitation measurementsare subject to both random and systematicerrors (Sevruk, 1985). The random error is due toirregularities of topography and microclimaticalvariations around the gauge site. Random errors arealso caused by inadequate network density toaccount for the natural spatial variability of rainfall.The systematic error in point precipitation measurementis, however, believed to be the mostimportant source of error. The largest systematicerror component is considered to be the loss due towind field deformation above the orifice of elevatedprecipitation gauges. Other sources of systematicerror are wetting and evaporation losses of waterthat adheres to the funnel and measurementcontainer, and rain splash. A broader discussion ofsystematic errors and their correction is containedin Volume I, 3.3.6, of this Guide.As rainfall data are collected at fixed observationtimes, for example clock hours, they may notprovide the true maximum amounts for the selecteddurations. For example, studies of thousands ofstation-years of rainfall data indicate that multiplyingannual maximum hourly or daily rainfallamounts for a single fixed observational interval of1 to 24 hours by 1.13 will yield values close to thoseto be obtained from an analysis of true maxima.Lesser adjustments are required when maximumobserved amounts are determined from 2 or morefixed observational intervals as indicated inTable II.5.5 (NRCC, 1989). Thus, maximum 6- and24-hour amounts determined from 6 and24 consecutive fixed one-hour increments requireadjustment by factors of only 1.02 and 1.01, respectively.These adjustment factors should be appliedto the results of a frequency analysis of annualmaximum series to account for the problem of fixedobservational times (NRCC, 1989).Table II.5.5. Adjustment factor for dailyobservation frequencyNumber of 1 2 3–4 5–8 9–24 > 24observations/daysAdjustment 1.13 1.04 1.03 1.02 1.01 1.00factorFor frequency analysis studies, it is necessary tocheck precipitation data for outliers and consistency.As noted in 5.4.5, an outlier is an observationthat departs significantly from the general trend ofthe remaining data. Procedures for treating outliersrequire hydrological and mathematical judgment(Stedinger and others, 1993). In the context ofregional analysis of precipitation, the outlierscould provide critical information for describingthe upper tail of the rainfall distribution. Hence,high outliers are considered to be historical data ifsufficient information is available to indicate thatthese outlying observations are not due to measurementerrors. Regarding data inconsistency,there are many causes. Changes in gauging instrumentsor station environment may causeheterogeneity in precipitation time series. Datafrom the gauge sites located in forest areas maynot be compatible with those measured in openareas. Measurements in the valley and mountainstations and at various altitudes will not provideidentical information regarding precipitation characteristics.Therefore, care must be used in applyingand combining the precipitation data.5.7.2 At-site frequency analysis of rainfallA frequency analysis can be performed for a site forwhich sufficient rainfall data are available. Similarto flood frequency analysis, rainfall frequency

CHAPTER 5. EXTREME VALUE ANALYSISII.5-17analysis is also based on annual maximum series orpartial duration series (for example, Wang, 1991;Wilks, 1993). Arguments in favor of either of thesetechniques are contained in the literature (NRCC,1989; Stedinger and others, 1993). Owing to itssimpler structure, the annual maximum seriesbasedmethod is more popular. The partial durationanalysis, however, appears to be preferable for shortrecords, or where return periods shorter than twoyears are of interest. The choice of an appropriatetechnique should depend on the purpose of theanalysis and characteristics of the available data interms of both quantity and quality. Improvedreliability of the results can be generally achievedwith the use of sophisticated and comprehensiveanalysis methods. Virtually all hydrologicalestimates are subject to uncertainty. Therefore, it isoften advisable to produce estimates using two ormore independent methods and to perform asensitivity analysis to gain information regardingthe potential reliability of results.Briefly, the steps below should be followed to determinethe frequency distribution of annualmaximum rainfall for a given site:(a) Obtain a data sample and perform an assessmentof data quality based on hydrological andstatistical procedures;(b) Select a candidate distribution model for thedata and estimate the model parameters;(c) Evaluate the adequacy of the assumed modelin terms of its ability to represent the parentdistribution from which the data were drawn.The assessment of data quality is an importantstep in all statistical computations. The basicassumption in precipitation frequency analysis isthat the data are independent and identicallydistributed. As mentioned above, precipitationmeasurements could be subject to various sourcesof error, inconsistency and heterogeneity. Detailedexamination and verification of the raw data areneeded to identify invalid data in the record causedby instrument malfunction and/or human error.Standard statistical tests are available to verifyserial independence, stationarity and homogeneityof the data series (see 5.4).There is no general agreement as to which distributionor distributions should be used for precipitationfrequency analysis. A practical method for selectingan appropriate distribution is by examining thedata with the use of probability plots. Probabilityplots, which require the use of a plotting positionformula, are an effective tool to display graphicallythe empirical frequency distribution of the dataand to assess whether the fitted distribution appearsconsistent with the data. There are several plottingpositionformulae available in practice (see 5.6 andNguyen and others, 1989) among which the Hazen,Weibull, and Cunnane formulas are the most popular.The differences between these three formulaeare small for observations that are neither the largestnor the smallest; however, they can be significantfor the largest three or four values in the data series(Stedinger and others, 1993). An alternative methodfor making a good choice among different distributionsis based on the L–moment diagram (Stedingerand others, 1993).Common distributions that have been applied tothe analysis of annual maximum series include theGumbel, generalized extreme value, log-normal,and log-Pearson type III distributions. Among thesedistributions, the generalized extreme value and itsspecial form, the Gumbel distribution, have receiveddominant applications in modelling the annualmaximum rainfall series. The Gumbel distributionwas found, however, to underestimate the extremeprecipitation amounts (Wilks, 1993). Adamowskiand others, (1996) have shown that Canadianprecipitation intensity data for various durationsdo not appear to be drawn from a Gumbel distribution.Studies using rainfall data from tropical andnon-tropical climatic regions (Nguyen and others,2002; Zalina and others, 2002) also suggest that athree-parameter distribution can provide sufficientflexibility to represent extreme precipitation data.In particular, the generalized extreme value distributionhas been found to be the most convenient,since it requires a simpler method of parameter estimationand is more suitable for regional estimationof extreme rainfalls at sites with limited data orwith no data (Nguyen and others, 2002). When thereturn periods associated with frequency-basedrainfall estimates greatly exceed the length of recordavailable, discrepancies between commonly useddistributions tend to increase.Many methods for estimating distribution parametersare available in the hydrological and statisticalliterature. The simplest method is the method ofmoments that provides parameter estimates indicatingthat the theoretical moments are equal tothe computed sample moments. An alternativemethod for estimating parameters is based on thesample L–moments. These are found to be lessbiased than traditional moment estimators, and arethus better suited for use with small sample sizes.The L–moment method has proved effective in theestimation of the generalized extreme value distributionparameters (Stedinger and others, 1993).Another method is the method of maximum likelihood.This method provides estimators with very

II.5-18GUIDE TO HYDROLOGICAL PRACTICESgood statistical properties in large samples, but theestimators are often not available in closed formand thus must be computed using an iterativenumerical method.The reliability of precipitation frequency estimatesdepends on how well the fitted model representsthe parent distribution. Several goodness-of-fitcriteria can be used to test whether a selected distributionis consistent with a particular data sample(NRCC, 1989; Stedinger and others, 1993; ASCE,1996). As mentioned above, probability plots areextremely useful in the assessment of the adequacyof fitted distributions. The assessment is performedby plotting observed rainfall data versus plottingpositionestimates of exceedance probability on aspecialized plotting paper. The estimated distributionis plotted on the same graph. Goodness of fit isjudged by inspection. More rigorous statistical testssuch as the Kolmogorov–Smirnov, probability plotcorrelation and L–moment tests are available, allowingquantitative judgment of goodness of fit.However, the selection of the distribution that bestfits each dataset is not a recommended approachfor frequency analysis (Stedinger and others, 1993;ASCE, 1996). The use of the best-fitting distributionfor each data sample provides frequency estimatesthat are too sensitive to the sampling variations inthe data and the period of record available. Currentdistribution selection procedures adopted by manycountries are based on a combination of regionalizationof some parameters and split-sampleMonte-Carlo evaluations of different estimationmethods to find distribution-estimation procedurecombinations that give reliable quantile and riskestimates (Stedinger and others, 1993; ASCE,1996).5.7.3 Regional rainfall frequency analysisEven a long record may be a relatively smallsample of a climatic regime. A better measure ofthe regime at a station may be given by asmoothed map, which includes informationfrom nearby stations that can influence pointdata, and thus broadens the sample. The degreeof smoothing should be consistent with thespacing of observation stations and the samplingerror of the stations. Too little smoothing tendsto confound sampling error with spuriousregional variation.Rainfall frequency atlases have been produced byinterpolation and smoothing of at-site frequencyanalysis results. Regional frequency analysis,which involves data from many sites, has beenshown to reduce the uncertainties in quantileestimation of extreme events (Hosking and Wallis,1988). Similarly to regional flood analyses, thefollowing issues should be addressed whenconducting regional precipitation analyses: theselection and verification of homogeneous regions,and regional distribution parameters. Severalregional estimation methods have been suggested,among which identification of the regional probabilitydistribution and the estimation of theindex-flood procedure for use with the annualmaximum series are the most popular. For example,Schaefer (1990) used the index floodmethodology to conduct regional analyses ofannual maximum precipitation data in WashingtonState. It has been shown that climatically homogeneousregions can be defined based on the meanannual precipitation. Further, it was found thatthe coefficients of variation and skew of annualmaximum rainfalls vary systematically with themean annual precipitation. Hence, all sites withina homogeneous region could be characterized by aspecific three-parameter probability distribution,such as the generalized extreme value, havingfixed values of the coefficients of variation andskew. However, the use of mean annual precipitationas an index variable may not be appropriatefor other regions with different climatic or topographicconditions. For instance, the median ofannual maximum rainfalls at a site was recommendedas the index variable for regionalestimation of extreme rainfalls in the UnitedKingdom of Northern Ireland and Great Britain(Institute of Hydrology, 1999). In general, one ofthe main difficulties in the application of thistechnique is related to the definition of homogeneousregions. Various methods have beenproposed for determining regional homogeneity,but there is no generally accepted procedure inpractice (Fernandez Mill, 1995; Nguyen and others,2002).Another regional rainfall frequency analysis methodis the station-year method. This method attemptsto enlarge the sample size by pooling records froma number of stations into a single large sample ofsize equal to the number of station years of record.Hence, when applying the station-year method, itis not advisable to estimate rainfall amounts for asite for return periods that are much longer thanthe length of record at any of the stations. However,the method may yield more reliable estimates if thestations can be considered to be meteorologicallyhomogeneous, rather than using only the data originatingfrom one site. Further, the effect ofinterstation correlation should be investigatedbecause spatial correlation between samples tendsto significantly reduce the number of station years.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-19Duration Depth (mm) Location Date1 min8 min15 min20 min42 min1 h 00 min2 h 10 min2 h 45 min4 h 30 min6 h9 h10 h18 h 30 min24 h2 days3 days4 days5 days6 days7 days8 days9 days10 days11 days12 days13 days14 days15 days31 days2 months3 months4 months5 months6 months11 months1 year2 years3812619820630540148355978284010871400168918252467313037214301465350035286569260286299640164226432643393001276716369187382041222454229902646140768Table II.5.6. World’s greatest observed point rainfallsBarot, GuadeloupeFussen, BavariaPlumb Point, JamaicaCurtea-de-Arges, RomaniaHolt, Misssouri, United StatesShangdi, Nei Monggol, ChinaRockport, West Virginia, United StatesD’Hanis, Texas, United StatesSmethport, Pennsylvania, United StatesMuduocaidang, Nei Monggol, ChinaBelouve, Reunion IslandMuduocaidang, Nei Monggol, ChinaBelouve, Reunion IslandFoc Foc, Reunion IslandAurere, Reunion IslandAurere, Reunion IslandCherrapunji, IndiaCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCommerson, Reunion IslandCherrapunji, IndiaCherrapunji, IndiaCherrapunji, IndiaCherrapunji, IndiaCherrapunji, IndiaCherrapunji, IndiaCherrapunji, IndiaCherrapunji, IndiaCherrapunji, India26 November 197025 May 192012 May 19167 July 188922 June 19473 July 197518 July 188931 May 193518 July 19421 August 197728 February 19641 August 197728–29 February 19647–8 January 19667–9 April 19586–9 April 195812–15 September 197423–27 January 198022–27 January 198021–27 January 198020–27 January 198019–27 January 198018–27 January 198017–27 January 198016–27 January 198015–27 January 198015–28 January 198014–28 January 19801–31 July 1861June–July 1861May–July 1861April–July 1861April–August 1861April–September 1861January–November 1861August 1860–July 18611860–1861Revised: 29 November 1991, US National Weather Service, US Department of the Interior Bureau of Reclamation, Australian Bureau ofMeteorologyOwing to the latter and the spatial heterogeneity ofclimatic data, this approach is seldom used inpractice.5.7.4 Frequency analysis of area-averagedrainfallIn general, a catchment-average design rainfall isoften required for design flood estimation, especiallyfor large drainage basins. For instance, whenthe area of a basin exceeds about 25 km 2 , rainfallobservations at a single station, even if at the centreof the catchment, will usually be inadequate for thedesign of drainage works. All rainfall records withinthe catchment and its immediate surroundingsmust be analysed to take proper account of thespatial and temporal variation of rainfall over thebasin. For areas large enough for the average rainfalldepth to depart considerably from that at apoint, it has been found beneficial to convert pointvalues to areal values. Frequency values for areaaveragedprecipitation are generally obtained byapplying an areal correction factor (areal correctionfactor) to point precipitation values. There are manymethods of transformation point values to arealestimates, with different results being obtained inthe same network according to the method applied(Nguyen and others, 1981; Arnell and others, 1984;Niemzynowicz, 1982; Institute of Hydrology, 1999).The areal correction factor estimates depend on theraingauge network density and, consequently, onthe accuracy of estimating the mean precipitationover an area. Most of the procedures that areused for computing mean areal precipitation from

II.5-20GUIDE TO HYDROLOGICAL PRACTICESraingauge data, such as the arithmetic averagemethod, Thiessen polygon method and inverseddistance-squared method, give comparable resultsfor long time periods; but the differences in resultsamong the various methods increase as the timeperiod diminishes, as for daily rainfall. Densenetworks of raingauges have been used to developdepth-area-duration correction factors (Smith,1993; Institute of Hydrology, 1999). Areal correctionfactors depend on local climatologicalconditions and therefore, whenever possible,should be derived from local data. Validation isrequired if areal correction factors are to be used farfrom the location in which they were developed.As procedures developed for converting pointprecipitation frequency values to areal values aremostly empirical, alternative methods have beenproposed for directly carrying out areal precipitationfrequency analyses using stochastic models ofthe spatial and temporal distributions of rainfall(Bras and Rodriguez-Iturbe, 1985; Smith, 1993).5.7.5 Storm rainfall analysis forhydrological design applicationsFor design purposes, precipitation at a given site orover an area for a specified duration and returnperiod is commonly used in the estimation of floodpotential. The use of design precipitation to estimatefloods is particularly valuable in thosesituations where flood records are not available ornot long enough at the site of interest, or they arenot homogeneous due to changes of watershedcharacteristics such as urbanization and channelization.Furthermore, design problems usuallyrequire information on very rare hydrologicalevents: events with return periods much longerthan 100 years. Common storm rainfall analysistechniques that can be used to address these designproblems are discussed below.5.7.5.1 Maximum observed rainfallSome of the world’s largest recorded rainfallamounts for selected durations are given inTable II.5.6. These values, which represent thecurrent upper bounds on observed precipitation,are enveloped by the following approximateequation:P = 422T 0.475 (5.33)where P is the rainfall depth in millimetres, and T isthe duration in hours. Most locations in the worldwill never come close to receiving these extremerainfall amounts.5.7.5.2 Rainfall intensity or depth–duration–frequency relationshipsIn standard engineering practice, the results ofpoint-rainfall frequency analysis are oftensummarized by intensity–duration–frequencyrelationships or depth–duration–frequencyrelationships for each raingauge site with sufficientrainfall records. These relationships are commonlyavailable in both tabular and graphical form forrainfall intensities or depths at time intervalsranging from five minutes to two days and forreturn periods from two to one hundred years.Owing to the uncertainties involved inextrapolation, rainfall values are generally notprovided for return periods longer than roughlytwice the raingauge record. Empirical equationsexpressing intensity–duration–frequency anddepth–duration–frequency relationships havebeen developed. There are many such equationsappearing in the technical literature, of which thefollowing forms are the most typical:ai =t c (5.34)+ baTi =t c (5.35)+ bi = a ( t − b ) −c(5.36)i = a + b log T(1 + t ) c (5.37)where i is the average rainfall intensity, that is,depth per unit time, generally expressed in mm/hr,t is the rainfall duration in minutes or hours, T isthe return period in years, and a, b and c are coefficientsvarying with the location and return period.5.7.5.3 Temporal and spatial extrapolationof point rainfall estimatesA number of publications (NRCC, 1989; ASCE,1996; Pilgrim, 1998; Institute of Hydrology, 1999)provide mapped regional analysis of precipitationfrequencies for various return periods and durations.For instance, the US Weather Bureau providesa rainfall atlas that contains maps for the entireUnited States with contour lines of rainfall amountsfor durations varying from 30 minutes to 24 hoursand return periods from 2 to 100 years (Hershfield,1961). In addition to this atlas, the US NationalWeather Service has prepared isohyetal maps forrainfall events having durations from 5 to 60minutes and for return periods of 2, 10, and 100years for the eastern and central states (Frederick,and others, 1977). This set of maps is useful for

CHAPTER 5. EXTREME VALUE ANALYSISII.5-21estimating design rainfalls of short duration ordeveloping intensity–duration–frequencyrelationships.Quantile estimates of point rainfall for durationsand return periods not shown on the regional rainfallmaps can be obtained by interpolation. Forinstance, for the eastern and central regions of theUnited States, depths for 10- and 30-minutedurations for a given return period are computed byinterpolation from the available 5-, 15- and60-minute data for the same period (Frederick, andothers, 1977):P 10min= 0.41P 5min+ 0.59P 15min(5.38)P 30min= 0.51P 15min+ 0.49P 60min(5.39)For return periods other than 2 or 100 years, thefollowing equations are used:P Tyr= aP 2yr+ bP 100yr(5.40)in which a and b are empirical coefficients varyingwith return period values. Please note that theserelationships are merely for illustration purposes.Owing to the regional variation in such a relationship,its application should be based on climaticsimilarity between the regions of its derivation anduse.In the absence of short-duration rainfall data, eitherat a site or sufficiently nearby for interpolation, itmay be possible to estimate the rainfall regime fromany indirect data that may be available. Such datainclude mean annual precipitation and meanannual number of days with rain, which may beobtained from maps or otherwise estimated. For theUnited States, the average relationship of precipitationper precipitation day (mean annualprecipitation divided by days of precipitation witha base of one millimetre) to a 2-year 24-hour rainfallis as follows:Precipitation perprecipitationday (mm) 5 8 10 132-year 24-hourrainfall (mm) 36 56 79 107Again, the relationship given in this table is merelyfor illustration. Owing to the regional variation insuch a relationship, its application should be basedon climatic similarity between the regions of itsderivation and use.For durations of less than 24 hours, it is advisable toestimate the 1-hour rainfall frequency amountsfrom the 24-hour values, to interpolate for intermediatedurations and to extrapolate for durationsone hour. The 2-year 1-hour rainfall is related to the2-year 24-hour rainfall according to the meanannual number of days with thunderstorms. Studiesthat have included a wide range of climate indicatethe following relationship:Ratio of 2-year1-hour rainfall to 2-year24-hour rainfall 0.2 0.3 0.4 0.5Mean annualnumber ofthunderstorm days 1 8 16 24Rainfall-frequency values for durations of less thanone hour are often obtained by indirect estimation.Rainfall data for such short durations are seldomreadily available in convenient form for the compilationof annual or partial duration series for directfrequency analysis. Average ratios of rainfallamounts for 5, 10, 15 and 30 minutes to 1-houramounts, computed from hundreds of station-yearsof records, are often used to estimate rainfallfrequencydata for these short durations. Theseratios, which have an average error of less than 10per cent, are as follows:Duration (minutes) 5 10 15 30Ratio (n minutesto 60 minutes) 0.29 0.45 0.57 0.79Thus, for example, if the 10-year 1-hour rainfall is70 mm, the 10-year 15-minute rainfall is 57 percent of 70, or 40 mm.These ratios can yield erroneous results in someregions. For example, in regions where most of therainfall occurs in connection with thunderstorms,the above ratios would tend to yield values that aretoo low. However, in regions where most of therainfall results from orographic influences withlittle severe convective activity, the ratios mighttend to yield values that are too high. This variationhas been handled on a continental basis for Australia(Court, 1961; Hershfield, 1965), with a relationshipthat was developed by using a geographical locationand 1-hour rainfall intensity as variables. Therelationship is also dependant upon the averagerecurrence interval. When large quantities of rainfalldata for a region are to be subjected to frequencyanalysis, as is usual in the preparation of generalizedmaps, the compilation of annual series data for

II.5-22GUIDE TO HYDROLOGICAL PRACTICES98101189 77121071189 6610 8 765597 6 58446574654 333543224 3 23 222 11111100Minutes 20 30405060 80 100 120 150 180 240 300 360 8 10 12 14 16 18 20 22 24Hours 12 3 4 5 6Duration – 20 minutes to 6 hours Duration – 6 to 24 hoursNote: For 20-minute to 60-minute rainfalls, values are in mm per hour;for longer durations the values are in mm of depth.mm per hourFigure II.5.2. Rainfall–intensity and depth–durationrelationshipall durations is a challenging and tedious task. It iscustomary, therefore, to limit such compilations todata from a relatively small number of recordingstations with good records for at least ten years. Themeans of the annual series are then computed andused to prepare a diagram such as that given inFigure II.5.1, which permits the estimation of rainfallvalues for durations of up to 24 hours when the1- and 24-hour amounts are known. The diagonalline in Figure II.5.2 illustrates an example in which24-hour rainfall is about 73 mm and 1-hour rainfallis 22 mm. Values for other durations can be read offthe intersections of the diagonal. Thus, the amountfor 12 hours is 60 mm; for two hours it is 30 mm.Diagrams similar to Figure II.5.3 may be constructed(Miller and others, 1973) for interpolating betweenthe 2- and l00-year return periods. Such diagramsmust be based on good long-record stations if theyare to be reliable. As with the duration–interpolationdiagrams, they vary from region to region,where climatic regimes differ significantly. They areused in the same manner as the duration–interpolationdiagrams in that a diagonal is laid across theappropriate 2- and l00-year rainfall depths on theirrespective verticals, and depths for other returnperiods are read at the intersections of the diagonalwith the corresponding verticals.With the use of the above two types of interpolationdiagrams, only the 1- and 24-hour rainfall1211109mm depthamounts for the 2- and l00-year return periods needbe computed for most of the stations in the regionfor which the diagrams were developed. Thediagrams are then used to estimate other requiredvalues. Both types are subject to regional variations,and caution should be exercised in trying to applythe diagrams in regions other than those for whichthey were developed.Another method for estimating extreme rainfallquantiles for locations without rainfall data is basedon regional maps of rainfall statistics. For example,Environment Canada provides maps showingisolines of the mean and the standard deviation ofannual rainfall extremes for each region of Canadafor durations varying from 5 minutes to 24 hours(NRCC, 1989). Hence, if the Gumbel distribution isassumed to be acceptable for describing rainfallextreme distribution, the quantile estimate of rainfallfor a given return period at an ungauged locationcan be computed using the frequency factor methodand the corresponding interpolated values of rainfallstatistics. Similarly, for Australia, under theassumption of log-normal and log-Pearson type IIIdistributions for rainfall extremes, maps of regionalizedskewness along with available rainfallfrequency maps can be employed to derive intensity–duration–frequencycurves for any givenlocation using appropriate extrapolation and interpolationprocedures (Pilgrim, 1998).In summary, one of the main challenges for engineersand hydrologists is to obtain representativeRainfall depth (mm)484440363228242016128402 5 10 25 50 100Return period in years, partial duration seriesFigure II.5.3. Return-period interpolation diagram48444036322824201612840Rainfall depth (mm)

CHAPTER 5. EXTREME VALUE ANALYSISII.5-23information related to rainfall extremes at a givensite. Precipitation stations, however, are not typicallywithin close proximity to the site of interest,or they do not contain a sufficient period of rainfallrecords to allow a reliable estimation ofrainfall. The rainfall frequency maps should beexamined since they are sometimes based on theanalysis of limited data over rather restrictedareas, and the interpolation of rainfall characteristicsto other areas could lead to graveuncertainties. Appropriate regional rainfall analysisprocedures described in 5.7.3 should be used,especially for ungauged locations and for siteswith limited rainfall records.5.7.5.4 Mass rainfall curvesThe first step in a storm-rainfall study is to plotaccumulated values of rainfall versus time of day togive a mass curve, or integrated curve, for eachstation or for selected representative stations, ifthere are many. The mass curves for non-recordingstations are constructed by comparison with masscurves from recording stations by means of proportionalityfactors. In doing so, the movement of thestorm and any reports of the times of beginning,ending and heaviest rainfall should be taken intoaccount. Figure II.5.4 shows a typical set of masscurves from the storm of 31 March–2 April 1962 insouth-eastern Canada.The pertinent stations are then listed in a table andaccumulated values of rainfall are tabulated foreach station for a pre-selected time increment. A6-hour time increment is used in the present example,but other increments may serve equally well.For convenience, the stations should be listed inorder of decreasing magnitude of total storm rainfall.The next step is to examine the table and selectthe particular 6-hour period that has the largest6-hour rainfall increments. The values for this timeincrement are then listed. The period of maximumAccumulated precipitation (mm)Hours0 6 12 18 24 30 36 42 48 54 60 66 72 78 84 90 96 102108 114 120 126 132 136 144228 Alma141 Moncton133 St JohnTime 6 am 6 pm 6 am 6 pm 6 am 6 pm 6 am 6 pm 6 am 6 pm 6 am 6 pm0Date 31 1 2March–April 1962Figure II.5.4. Mass rainfall curves25020015010050Table II.5.7. Maximum average rainfall depth(mm) – storm of 31 March to 2 April 1962,south-eastern CanadaDuration (hours)Area (km 2 ) 6 12 18 24 4225 90 165 205 230 240100 85 155 190 215 2251 000 70 130 165 185 19010 000 50 90 115 140 145100 000 25 45 65 75 8512-hour rainfall is found in a similar way and itsrainfall is listed. The same operation is applied todefine the maximum 18-, 24-,…, n-hour increments.For periods embracing several 6-hourincrements, a considerable number of trials may berequired to find the period that includes the maximumrainfall for a particular duration.5.7.5.5 Depth–area–duration analysisStorm-rainfall analysis expresses the depth–area–duration characteristics of the rainfall from aparticular storm. The depth is defined for pertinentcombinations of enveloping area andduration, and is generally portrayed by tables orcurves. In the aggregate, such analyses provideuseful records for the design of flood control structuresand for research in quantitative precipitationforecasting.Individual point-rainfall observations are analysedjointly and in combination with other information.The rainfall data usually consist of observations ofdaily totals, interspersed with a few recorder measurementsthat contain short-term rainfall intensityinformation. Sometimes, these data are augmentedby observations obtained through special interviews,referred to as bucket surveys. Additionalinformation may come from synoptic weathermaps, radar, reports of rises of small streams andother sources. The procedure, which is summarizedin the following subsections, is described in detailin the Manual for Depth–Area–Duration Analysis ofStorm Precipitation (WMO-No. 237).Based on the tabulation of maximum rainfallincrements, isohyetal maps are prepared for eachduration, for example, 6 or 12 hours. Areasenclosed by each isohyet are then evaluated byusing a planimeter or by tallying grid points, andthe resulting values are plotted on a graph of areaversus depth, with a smooth curve drawn for each

Depth of precipitation (mm)II.5-24GUIDE TO HYDROLOGICAL PRACTICESmaximum precipitation for the design watershed.The main steps can be illustrated as follows:250200HighefficiencystormMoisturemaximizationTranspositionEnvelopingProbablemaximumprecipitationProbablemaximumflood(Total storm)10 100 1 000 10 000 100 000Area (km 2 )duration. A linear scale is commonly used fordepth and a logarithmic scale for area. The envelopingor maximum depth–area–duration data foreach increment of area and duration may be tabulatedas in Table II.5.7 from curves such as those inFigure II.5.5.5.7.5.6 Probable maximum precipitationThe term probable maximum precipitation, orPMP, is well established and is widely used to referto the quantity of precipitation that approachesthe physical upper limit of precipitation of a givenduration over a particular basin. The terms maximumpossible precipitation and extreme rainfallhave been used with roughly the same meaning.To ask how possible or how probable such precipitationis would be at best a rhetorical questionbecause the definition of probable maximum is anoperational one that is specified by the operationsperformed on the data.5.7.5.6.1 Basic methods of estimating probablemaximum precipitationThere are two methods for estimating probablemaximum precipitation: indirect and direct.5.7.5.6.2 Indirect type42 hours24 hours18 hours12 hours6 hoursFigure II.5.5. Enveloping depth–area–durationcurves150100The indirect type first estimates probable maximumprecipitation for the storm area, an area surroundedby isohyets, and then converts it into probable500High-efficiency storms are those for which the datasupport the assumption that their precipitationefficiency was near a maximum. The return periodof such storms, given by point data on the envelopingcurve, is usually more than 100 years.Moisture maximization is a procedure by which themoisture of a high efficiency storm is maximized.The increase is usually limited to 20–40 per centbecause there is an approximate physical upperlimit for the representative dewpoint, which is acritical factor, and this cannot exceed the highestwater temperature of the sea surface at the source ofwarm and wet air masses. In addition, this decreasesas one moves from the source of air masses to thedesign watershed.Transposition is a procedure which accounts formoving a high-efficiency storm from one locationto another within a meteorologically homogeneouszone. In essence, it replaces time with space in orderto increase the number of storm samples andprovide additional observed data.Enveloping refers to the use of a depth–area–durationrelationship drawn up on the basis of transposedstorms, thereby maximizing precipitation depths ofvarious area sizes and durations. This also compensatesfor a lack of observed data.5.7.5.6.3 Direct typeThe direct type estimates probable maximumprecipitation for the area directly encompassing theparticular project in the design watershed. Majorsteps include the following:Storm modelMaximizationProbablemaximumprecipitationProbablemaximumfloodThe storm model for a typical storm or for an idealstorm reflects the characteristics of the catastrophicprecipitation over the design watershed which islikely to pose the greatest threat of flooding for theproject. Such models can be classified as local, transposition,combination or inferential, depending ontheir source.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-25Local models are used for local storm maximizationand are selected from storm data observed in thedesign watershed. They can also be developed bysimulating historically extraordinary floods fromsurveys.Transposition models are derived by transposingactual storms in surrounding similar regions.Combination models are sequences of two or morestorms that are subject to spatial or temporal stormmaximization and are combined in accordancewith theories of synoptic meteorology.Inferential models are theoretical or physicalmodels which result from generalization and inference,using the three-dimensional spatial structureof storm weather systems within the design watershedwhereby major physical factors that affectprecipitation are expressed by a series of physicalequations. They mainly include convergencemodels and laminar models of the flow field orwind field.Maximization maximizes storm performance.When the storm model is that of a high-efficiencystorm, then only moisture maximization isperformed; otherwise both the moisture and powerfactors are maximized.The above four methods are applicable to both hillyregions and plains. The fourth method is in generalapplicable to area of under 4 000 km 2 and durationsshorter than 24 hours, whereas the other threemethods are independent of area size and duration,and work especially well for estimating probablemaximum precipitation for large watersheds largerthan 50 000 km 2 and durations greater than threedays.Probable maximum precipitation can also be estimatedby using the statistical estimation methodand the empirical formula method.5.7.5.6.4 Preliminary considerationsFor major structures, the cost of the spillway maybe a substantial proportion of the total project cost.Its design is therefore important enough to warranta very detailed study. However, in the preliminaryplanning stages, it is sufficient to use generalizedestimates of probable maximum precipitation ifthese are available for the area. Estimates of thistype for the United States have been published asmaps and diagrams in various issues of the USWeather Bureau Hydrometeorological Report series.Similar reports have been prepared by several othercountries for various parts of the world. The followingsteps should be taken when determiningprobable maximum precipitation:(a) Value basic data. Collect necessary hydrometeorological,geographic and orographic data,especially those related to extraordinary stormsor floods and corresponding meteorologicaldata, and asses their reliability;(b) Make full use of storm data. Such data forthe design watershed and its surroundingregions are the basis for calculating probablemaximum precipitation and are also one ofthe major factors influencing the precision ofresults;(c) Analyse characteristics and causes of largestorms in the design watershed in order toprovide a basis for determining methods forcalculating probable maximum precipitation,selecting indicators, maximizing and analysingthe reasonableness of results;(d) Fully understand the characteristics of themethods. Select two or more methods fromthose that are available for determining probablemaximum precipitation based on theconditions required for each method and thedesign requirements and data available for thewatershed. Complete the calculation separatelyand then select the final results by means of acomprehensive evaluation.5.7.5.6.5 Requirements for probable maximumprecipitationUnless depth–area–duration analyses applied to aproject basin have been constructed within thestorm-transposition zone, a number of individualstorm studies will be required to obtain estimates ofprobable maximum rainfall. Before these studiesare undertaken, the likely critical rainfall durationfor the particular design problem should be determined.The selection of an appropriate tentativerainfall duration design can help avoid the analysisof data that are not directly applicable to the projectand the subsequent need for analysis of additionaldata if too short a duration is adopted in the firstinstance.The approximate time of rise of flood hydrographsfor storms centring on different parts of the basinand the particular characteristics and proposedmethod of operation of the projected works shouldbe considered in selecting tentative design rainfallduration.The calculation undertaken should depend on thestorm characteristics and design requirements ofthe project (Ministry of Water Resources and

II.5-26GUIDE TO HYDROLOGICAL PRACTICESMinistry of Energy of the People’s Republic ofChina, 1995):(a) If a project design requirement calls for probablemaximum precipitation of a particularduration, only the storm volume and the mostsevere spatial or temporal distributions of thatduration need be calculated;(b) If the project calls for probable maximumprecipitation of several durations, probablemaximum precipitation should be determinedfor each of those durations.(c) If the project involves a series of reaches alonga river, as in a cascade of dams, then a seriesof probable maximum precipitation estimateswill need to be made, with attention being paidto coordination between the upper and lowerreaches. Regional estimates of probable maximumprecipitation should be in accordancewith the characteristics of observed storms;(d) For places where storm characteristics differamong seasons, probable maximum precipitationestimates should be made for summer,autumn, rainy seasons, typhoons and so forth.5.7.5.6.6 Selection of sub-basinsFor project sites with large drainage areas, it may benecessary to estimate the probable maximum rainfallfor some sub-basins and then compound theresultant probable maximum flood hydrographsfrom these sub-basins. To avoid subsequent unnecessaryor incomplete analyses of mean areal rainfalldepths during the storm studies, the sub-basins forwhich flood hydrographs are required should beselected before storm analyses are started. The selectionof sub-basins is influenced by the physicalcharacteristics of the basin and the availability andlocations of stream-gauging stations from whichthe sub-area flood hydrographs can be routed to theproject site.Three commonly used methods are summarizedbelow: the storm transposition method, the generalizedestimation method and the statisticalestimation method.5.7.5.6.7 Storm transposition methodThe basic assumption of storm transposition is thatthe region where the storm occurred – the stormsource – and the design region are similar in termsof geographic or orographic conditions and thesynoptic causes of storms. As a result, the structure– temperature, air pressure, wind power and spatialor temporal distributions – of a transposed storm isexpected to change little. It includes the two followingassumptions:(a) After transposition, the storm weather systemand the relative position of the storm areachange little;(b) After transposition, spatial or temporal distributions– the hyetograph and the isohyets – ofthe storm also change little.5.7.5.6.8 Selection of transposed objectsAnalyses should first be performed on the basis ofdata on observed catastrophic intense rainfall orfloods which were collected from the design watershedin order to understand the basic types ofcatastrophic rainfall or floods in the watershed andthen identify the storm types corresponding toprobable maximum flood, PMF, required by thedesign project. For example, if the event in questionis a tropical cyclone (typhoon, hurricane) or afrontal storm, the transposed object should beselected from among tropical cyclone storms orfrontal storms, respectively.5.7.5.6.9 Possibility of transpositionThis involves a study of whether the selected transposedobject is likely to occur in the designwatershed. There are three solutions:(a) Identifying meteorologically homogenouszones;(b) Setting transposition limits for a particularstorm;(c) Performing specific analyses on the designwatershed and comparing the similaritybetween the design watershed and the region ofthe storm source in terms of climate, weather,geography, orography and the like. The moresimilar these are, the more possible the transposition.5.7.5.6.10 Isohyetal map allocationIsohyetal map allocation moves the isohyetal mapof the transposed object to the design watershed,which raises questions such as where to put thestorm centre, whether to rotate the direction of thestorm axis – the direction of the major axis of theisohyetal map – and how to rotate it.The computations start with a study of the statisticsof the spatial distribution of actual storms, that is,finding common rules of central positions anddirections of axes of storms with weather causessimilar to those of the transposed object on thebasis of existing storm data, including thoseobserved, surveyed and recorded in the literature,and then making adjustments and decisions in relationto the particular circumstances of the project.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-27The transposed isohyets should suit the large-scaleorography of the design watershed as well as possible.The storm centre should match the small-scaleorography, such as that surrounding the riverchannel.5.7.5.6.11 Transposition correctionThe purpose of transposition correction is to estimatequantitative changes to rainfall caused bydifferences in conditions such as geometry, geographyand orography of the region. In other words,transposition correction typically includes thegeometric, geographic and orographic correctionsof the watershed. The geographic correction considersmoisture correction only, while the orographiccorrection includes moisture correction andpower correction. The geometric correction of thewatershed must be performed first for any stormtransposition.If the transposed object is very similar to the designwatershed with regard to the weather situation,orographic and geographic conditions are almostthe same and there is no obvious moisture obstaclein between, the storm isolines of the transposedobject may be moved to the design watershed withoutany change. Only a geometric correction of thewatershed is needed.If the two places are similar in terms of stormweather situation and different in terms oforographic and geographic conditions, and suchdifferences are not large enough to cause greatchanges to the storm mechanism, then powercorrection need not be considered. In this case,only moisture correction needs to be considered inaddition to the geometric correction of the watershed.This method is commonly used in plains andregions of low relief.If storms with different orographic conditions mustbe transposed because of actual conditions, mountainswill have some effects on the storm mechanism.In such cases, power correction needs to be consideredin addition to geometric and moisturecorrections of the watershed.Concern for the orientation of precipitationpatterns relative to basin orientations has resultedin special studies (WMO, 1986a; Hansen andothers, 1982).5.7.5.6.12 Storm maximizationIn storm transposition, selected transposed objectsare typically high-efficiency storms; therefore,only moisture maximization is needed whenmaximizing them. For such cases, maximizationmay be performed at the storm source before transpositiononly. Only after transposition correction,is the storm probable maximum precipitation.Maximization methods developed in the UnitedStates and adopted in a number of countries(Pilgrim, 1998) have been described by Weisner(1970) and in a number of publications of the USNational Weather Service, formerly the US WeatherBureau (see references in the Manual for Estimationof Probable Maximum Precipitation (WMO-No. 332,1986a).5.7.5.6.13 Generalized estimation methodThis method involves estimating probable maximumprecipitation for non-orographic regionsand orographic regions respectively. It is assumedthat precipitation in non-orographic regionsresults from the passing of weather systems, whilethat in orographic regions results from both thepassing of weather systems and orographic effects.Precipitation caused by weather systems is referredto as convergence rain, or convergence components,and those caused by orography are calledorographic rains, or orographic components.Precipitation generalization involves the generalizationof convergence rains, using thedepth–area–duration generalization of storms.This generalization method is applicable to both aparticular watershed and a large region thatincludes a lot of watersheds of various sizes. Forthe latter, it is called generalized or regional estimation.The content of generalization includesgeneralization of the depth–area–duration relationshipand the generalization of the spatial/temporal distributions of probable maximumprecipitation.Determining probable maximum precipitationusing the depth–area–duration generalized estimationmethod includes four steps:(a) Maximize actual large storms – only moisturemaximization being performed in most cases;(b) Transpose maximized storms to the studyregion;(c) Smoothen and fit envelope curves to data,including depth-duration smoothing, depthareasmoothing and combined depth–area–duration smoothing;(d) Apply the probable maximum rainfall on thestorm area to the design watershed so as todetermine the probable maximum storm onthe watershed area.

II.5-28GUIDE TO HYDROLOGICAL PRACTICESFor regional generalized estimation, regionalsmoothing should be added after step (c). A checkfor regional consistency involving numerouscomparisons has been described by Hansen andothers (1977) and in the Manual for Estimation ofProbable Maximum Precipitation (WMO-No. 332).The method is used to estimate probable maximumprecipitation for durations of 6 to 72 hours and forareas under 52 000 km 2 in plains and areas under13 000 km 2 in orographic regions in the UnitedStates. For orographic regions, the influence of thetopography should be considered in probable maximumprecipitation estimation. For other countriesor regions, the area sizes to which the method isapplicable need to be analysed, based on the actuallocal conditions.The method makes full use of maxima, includingthe largest rainfalls for various durations and areasof all the storm data in the particular region. Theresults of these calculations can be coordinated inthe region and the watershed.Now widely used in the United States, Australia,India and other countries, the generalized estimationmethod is described in the Manual for Estimationof Probable Maximum Precipitation (WMO-No. 332).Major results of the generalized estimation methodinclude the following:(a) The precipitation depth of probable maximumprecipitation: one is the enveloping curvemap of the depth–area–duration relationshipof probable maximum precipitation and theother is the probable maximum precipitationisoline map for several durations and areasizes;(b) The spatial distribution of probable maximumprecipitation: generalized as a set of concentric,similar ellipses;(c) The temporal distribution of probable maximumprecipitation: generalized as a singlepeak;(d) For orographic regions, there are also somecorrelograms or isoline maps of some parametersthat reflect orographic effects, which areused to convert probable maximum precipitationof convergence rains into probable maximumprecipitation for orographic regions.In principle, probable maximum precipitation forsmall watersheds may be determined using thestorm transposition method. Nonetheless, whenthe design region lacks the moisture and wind dataneeded for maximization, it will be very hard to usethe traditional storm transposition method. If anabstracted statistical value K mis transposed insteadof transposing directly the rainfall of a storm, theissue will be much simpler. K mmay be defined by:K m =X m − X n−1S n−1(5.41)where X mis the first item in the ranked observedseries, that is, the very large value,X – n–1is the average excluding the very large value,that is:n1X n−1 = ∑ X (5.42)n − 1 ii=2S n-1is the standard deviation excluding the verylarge value, that is:S n−1 =1n − 2n∑ ( X i − X i−1 ) (5.43)i=2Clearly, the more data that are used and the moreregions that are studied, then the enveloping valueof K mwill be closer to the value corresponding toprobable maximum precipitation.Hershfield (1965) collected data from more than2 600 rainfall stations, about 90 per cent of whichwere in the United States and developed a graphicalrelationship between enveloping values and meansof annual series of K mfor different durations(Figure II.5.6) for the use of hydrologists.K20 2015105 minutes155 minutes1 hour24 hours6 hours0 10 20 30Mean of annual maximum n-hour rainfall (mm)(6-hour curve interpolated from other durations)24 hours6 hours1 hour5.7.5.6.14 Statistical estimation methodThis is an approximate method for estimating probablemaximum precipitation for small watersheds,usually those under 1 000 km 2 . It is summarizedbelow.5100200 300 400 500 600Mean of annual maximum n-hour rainfall (mm)Figure II.5.6. K as a function of rainfall durationand mean of annual series

CHAPTER 5. EXTREME VALUE ANALYSISII.5-29When using Figure II.5.6 to determine K m, the averageX – n and S nare worked out based on rainfalldata from a particular station in the design watershedand the calculation is completed according tothe following equation:PMP = X – n + K m S n (5.44)The coefficient of variability is:C vn = S nX n(5.45)Therefore, equation 5.39 can be rewritten asfollows:PMP = (1 + K mC vn) X – n (5.46)Per cent of point rainfall for given area100908070605024 hours8 hours3 hours1 hour30 minutes0 125 250 375 500 625 750 875 1 000Area (km 2 )Figure II.5.8. Depth–area curvesAs illustrated by equation 5.46, determining probablemaximum precipitation with Hershfield’sstatistical estimation method is essentially a matterof transposing the statistical value K mof a very largestorm in a wide region and correcting it by usingthe storm average X – nand the coefficient of variabilityC vnfor the design watershed. The methodrequires that enough single-station, daily precipitationobservation series be available for the designwatershed.Maximum rainfalls needed are selected from amongrecords using a particular duration or durations(1 hour, 6 hours, 24 hours) each year and are organizedinto an annual series. The mean X – and thestandard deviation S nor the coefficient C vnof theseries are then calculated. The K value is determinedfrom Figure II.5.6 using the mean of the series. As aRainfall depth (mm)100806040200Return period (years)1.01 2 5 10 25 50 100 200617064+ 606265696663671.0 50 80 90 96 98 99 99.5Figure II.5.7. Example of an extreme probabilityplot+68Probability [ 100 M (N+1) ]-1.0 0 1.0 2.0 3.0 4.0 5.0 6.0Reduced variateresult, probable maximum precipitation can bedetermined according to equation 5.44 or 5.46.Care should be taken to ensure that the highest oneor two values in the annual series are consistentwith the other values comprising the series. If, forexample, the maximum value in a 30-year period istwice the second-highest value, it is clearly anoutlier. The easiest way to detect an outlier is toarrange the series in descending order and thencompute the return period of each value. Next, thevalues are plotted against their correspondingreturn periods on probability paper as shown inFigure II.5.7. If the maximum value of the series lieswell above the line delineated by the other items inthe series, it can be considered an outlier. An outliershould not be used to compute the mean or standarddeviation of the series. If used, the mean andstandard deviation should be adjusted as indicatedby Hershfield, who also provided an adjustment forlength of record. A complete, detailed descriptionof the entire procedure, including diagrams formaking the necessary adjustments, is given in theManual for Estimation of Probable MaximumPrecipitation (WMO-No. 332), Chapter 4.When the probable maximum precipitation is to beapplied to an area larger than about 25 km², itshould be reduced. No modification is considerednecessary for smaller areas. For larger areas, thepoint value is generally reduced by means of deptharea or area reduction curves similar to those ofFigure II.5.8.The statistical method described above may overestimatethe probable maximum precipitation inregions of heavy rainfall and in regions of frequentstorms of similar types. In regions of low rainfall

II.5-30GUIDE TO HYDROLOGICAL PRACTICESand where heavy rain-producing storms, such astropical cyclones, are rare but possible, the methodmay underestimate probable maximum precipitation.Values of K as high as 30 have been foundnecessary in order to exceed maximum observedpoint rainfall amounts in some regions. In somecountries, in particular the United States, wherestorm studies are the preferred source of data forprobable maximum precipitation determination,the statistical method has been used primarily as ameans of checking for consistency.5.7.5.6.15 Checking the plausability of probablemaximum precipitation estimatesIn principle, a variety of methods should be usedconcurrently to estimate probable maximumprecipitation. Results of those methods should thenbe analysed comprehensively to select the bestprobable maximum precipitation value. In the end,the plausibility of the selected probable maximumprecipitation should be checked from multipleperspectives so that the result is both maximal andpossible. In general terms, methods of checking therationality of probable maximum precipitationresults are the same as those for the plausibility ofprobable maximum flood results. As a result, methodsfor checking them are the same (see 5.10.2 orManual for Estimation of Probable MaximumPrecipitation (WMO-No. 332), Chapter 4.5.7.5.7 Design stormA design storm or design hyetograph is a rainfalltemporal pattern that is defined for used in thedesign of a hydraulic structure. A design hyetographor synthetic storm of specified exceedance probabilitycan be developed in the following way. Therainfall depth is obtained from the depth–duration–frequencyrelationship based on the specifiedprobability and duration. Next, an area adjustmentfactor is applied to the rainfall depth. Finally, amethod is used to distribute the rainfall depth overtime using available procedures (Wenzel, 1982;Arnell and others, 1984). Pilgrim and Cordery(1975) warn that approaches overly smoothing thetemporal patterns of rainfall are unsuited for designapplications because the time variability of rainfallintensity often has a significant effect on the designhydrograph. Two important points noted by Pilgrimand Cordery (1975) and Huff and Changnon (1964)are that the variability of intensities diminisheswith decreasing exceedance probability and themajority of extreme storms have multiple peaks ofhigh rainfall intensity. Depth–duration–frequencyrelationships can be regionalized using proceduresdescribed above.5.7.5.8 DroughtDrought is the low hydrological extreme resultingfrom perturbations in the hydrologic cycle over asufficiently long time to result in a significant waterdeficit. Local water resources become insufficient tosupport the established or normal activities of thearea. Droughts are interpreted and categorizedbroadly as meteorological, hydrological or agricultural.The meteorologist is concerned with droughtin the context of a period of below-normal precipitation.To a hydrologist, drought refers tobelow-average flow in streams or content in reservoirs,lakes, tanks, aquifers and soil moisture. To anagriculturist, drought means a prolonged shortageof soil moisture in the root zone.For meteorological drought, a useful means of analysisis based on the magnitude-span frequency. Asimple type of analysis would compare rainfalltotals for calendar months or pertinent seasonswith their corresponding normal values and assessseverity of drought based on negative departuresfrom normal values. To take into account the effectof time distribution of rainfall, an antecedentprecipitationindex may be used instead of totalrainfall. Another way to account for the month-tomonthcarry-over effects of rainfall for evaluatingseverity of meteorological drought is the Herbsttechnique (Herbst and others, 1966).The severity of agricultural drought may be judgedby the drought index, a means of summarizing andperiodically disseminating drought informationand crop-moisture conditions on a regional basis. Itcan be used for evaluating the drought hazard overa sizeable area or for periodic assessment of thecurrent extent and severity over a region.Hydrological drought severity is related to the severityof departure from the norm of low flows and soilmoisture in conjunction with excessive lowering ofgroundwater levels. In view of the considerable timelag between departures of precipitation and the pointat which these deficiencies become evident in surfacewater and groundwater, hydrological drought is evenfurther removed from the precipitation deficiencysince it is normally defined by the departure ofsurface and subsurface water supplies from someaverage condition at various points in time.5.7.5.9 Recent precipitation frequencyanalysis techniquesThe density of raingauges has been a significantlimitation in the development of precipitationfrequency analysis procedures. Radar provides a

CHAPTER 5. EXTREME VALUE ANALYSISII.5-31potentially important source of precipitation datafor frequency analyses. The most important advantageof radar for precipitation measurement is thecoverage radar provides of a large area with goodspatial and temporal resolutions, as small as 1 km 2and 5 minutes. With an effective range of 200 km,a single radar can cover an area of more than10 000 km 2 .Cluckie and others (1987) report a depth–area–durationanalysis of extreme events using hourly radarrainfall totals for five-km grid squares. The need tofirst correct and calibrate radar data is emphasized.Depth–area–duration analysis is performed onindividual storms as a means of classifying theirflood producing potential. Furthermore, Cluckieand Pessoa (1990) have used radar data from northwestEngland to characterize actual storms, whichhave then been maximized and transposed toobtain probable maximum precipitation estimatesfor catchments areas of interest (see 5.7.5.6 for adiscussion of probable maximum precipitation).Such an approach capitalizes on radar’s ability todelineate storms in space and time. In addition, aprogram called RADMAX implements the procedureand incorporates visualization of the stormtransposition step (Moore, 1993). Collier (1993)suggested the use of radar, and satellite data forcruder estimates to support probable maximumprecipitation estimation by using the storm modelapproach.Design problems generally require information onvery rare hydrological events, namely those withreturn periods much longer than 100 years.Traditional techniques for addressing these designproblems are mostly based on the use of probablemaximum precipitation. New frequency analysisprocedures, which exploit some of the tools ofprobable maximum precipitation, have been developedfor assessing rainfall magnitudes with verylong return periods. In particular, the NationalResearch Council (1988) recommended the stochasticstorm transposition techniques. In the probablemaximum precipitation application, storm transpositionis based on the assumption that there existmeteorologically homogeneous regions such that amajor storm occurring somewhere in the regioncould occur anywhere else in the region, with theprovision that there may be differences in the averageddepth of rainfall based on differences inmoisture potential. In the stochastic storm transpositionmethod, the frequency of occurrence ofstorms in the transposition region provides the linkfor obtaining frequency estimates of extreme stormmagnitudes. The stochastic storm transpositionprovides estimates of the annual exceedanceprobability of the average storm depth over thecatchment of interest. The estimate is based onregionalized storm characteristics such as maximumstorm centre depth, storm shape parameters,storm orientation, storm depth and spatial variability,and on an estimation of the joint probabilitydistribution of storm characteristics and stormoccurrence within a region. An advantage of thestochastic storm transposition method is that itexplicitly considers the morphology of the storms,including the spatial distribution of storm depthand its relation to the size and shape of the catchmentof interest (NRC, 1988).5.8 LOW-FLOW ANALYSES[HOMS I80, K10]5.8.1 GeneralInformation on the characteristics of low flows forstreams and rivers is important for planning, designand operation of water-related projects and waterresource systems. Such information is used indesigning wastewater treatment and storage facilitiesto ensure that releases do not exceed theassimilative capacity of receiving waterways, reservoirstorage design for multi-purpose systems andthe allocation of water for various purposes such asindustrial, agricultural, domestic and in-streamecological needs.Low-flow frequency analysis and flow-durationcurves are the two most commonly used analyticaltools to help assess the low-flow characteristics ofstreams, and these will be described in more detailin this section. Both approaches typically requireat-site continuous streamflow data for analysis,unless regional approaches are used to estimate atsitecharacteristics. Other characteristics that aresometimes useful include the amount of time orfrequency for which flows might be below a certainthreshold during a season and the volume of wateror deficit that might arise during the period inwhich flows are below a threshold. Statisticalapproaches can also be used to assess these aspects.Other approaches, such as passing historicalsequences of data or synthetically generatedsequences through a model of the river or reservoirsystem can provide additional, valuable detailedinformation for design purposes. The latterapproaches will not be covered in this Guide.Low flows are usually sustained by depletion ofgroundwater reserves or by surface discharge fromupstream bodies of water including lakes, wetlands

II.5-32GUIDE TO HYDROLOGICAL PRACTICESand glaciers. Low flows within a year or seasonmay result from different mechanisms forcing thehydrological response. Low flows in cold, northernclimates may occur due to the prolongedwinter period where precipitation occurring duringthis period is primarily in the form of snow, resultingin ever-decreasing flows until the occurrenceof the spring freshet. A second period that produceslow flows occurs during the warm season wherethere may be periods of significant evaporationand little precipitation. Depending on local climatologyand physiography, some basins mayproduce low flows resulting predominately fromone process or a combination of processes asdescribed above (Waylen and Woo, 1987). It isimportant to understand the processes producingthe low flows, as these may determine the analyticalapproaches taken to analyse their characteristicsand results.Anthropogenic intervention can greatly alter thenatural low-flow regime. For example, increasedextraction from surface water for irrigation mayoccur during periods of prolonged absence of rainfall,resulting in artificially suppressed flow values,compared with what naturally would haveoccurred. Significant extraction of groundwaterfor agricultural, industrial and human uses canreduce water-table levels and result in reducedstreamflow. A variety of other anthropogenic interventionscan occur within a basin and should beknown prior to proceeding with analyses of data.Such interventions can include upstream regulation,inter-basin transfers, return flows fromdomestic sewage systems that use groundwater asa water source and changes in land use, such asdeforestation, reforestation and urbanization.Such operations may cause increases or decreasesin flow rates (Institute of Hydrology, 1980;Smakhtin, 2001) and may well invalidate theassumptions commonly associated with theanalytical tools described below and in previoussections of this chapter.5.8.2 At-site low-flow frequency analysisInformation on low-flow frequency is obtainedfrom an analysis relating the probability of exceedingan event to its magnitude. Such relationshipscan be established for low flows of various durations,such as 1, 3, 7 or 14 days or other periods ordurations of interest. Commonly, non-parametricfrequency analysis or probability distributions areused to describe the frequency relationship ofobserved seasonal or annual low flows. As in thecase of flood flows, the parent distribution of lowflows is unknown.Various studies have been conducted to ascertainwhich distributions and which parameter estimationmethods may best represent the distribution oflow flows (see for example Nathan and McMahon,1990; Lawal and Watt, 1996; and Durrans andTomic, 2001). Results of the studies tend to differ, asthe same distributions, fitting methods or data arenot always used.Matalas (1963) analysed data for 34 sites in theUnited States using the Pearson type III (P3), thePearson type V (P5), the Gumbel type III (G3), whichis also known as the three-parameter Weibull (W3),and the three-parameter log-normal (LN3) distributions.He concluded that the G3 and P3 distributionsperformed equally well and tended to outperformthe other two distributions. According to Matalas(1963), the theoretical probability distributionshould have a lower boundary greater than or equalto zero, and he used this as one criterion in assessingthe acceptability of a distribution. Condie and Nix(1975) performed a similar analysis of data from 38Canadian rivers using the same probability distributionsas Matalas (1963). To ascertain the suitabilityof the distribution, they considered solutions inwhich the lower boundary parameter was greaterthan zero and smaller than the smallest observedflow. They recommended the use of the G3 distribution,with parameters estimated by maximumlikelihood, followed by the method of smallestobserved drought. Condie and Cheng (1982),furthering the work of Condie and Nix (1975),continued to recommend the use of the G3 distributionfor low-flow frequency analysis. In the latterstudy, they considered a negative lower boundary tobe acceptable. In such cases, they took the area ofthe density function from the negative lower boundaryto zero as representing the probability of theoccurrence of zero flows. They also verified that thelower boundary parameter was not larger than thesmallest member of the sample, as certain fittingmethods can provide such unrealistic results.Tasker (1987) showed that for 20 stations in Virginia,United States, using bootstrapping that the log-Pearson type III (LP3) and G3 distributions hadlower mean square errors in estimating the 7-day10-year (Q7,10) and 7-day 20-year (Q7,20) low flowsthan did the Box–Cox transformations or the log-Boughton methods. Vogel and Kroll (1989) analysedthe two-parameter log-normal (LN2) and twoparameterWeibull (W2) distributions fitted to datafrom 23 sites in Massachusetts, United States. Theyconcluded that the W2 distribution fitted poorly,while there was no evidence to reject the hypothesisthat the data were from a LN2 distribution. Inaddition, they analysed a variety of three-parameter

CHAPTER 5. EXTREME VALUE ANALYSISII.5-33distributions, namely the LN3, the LP3 and the G3.They found that the LP3 slightly outperformed theother three- and two-parameter distributions. Thesestudies indicate that the preferred frequency distributionvaries by region and there is no one frequencydistribution that clearly outperforms all others.Zaidman and others (2003) performed an analysisof 25 natural streams within the United Kingdomhaving more than 30 years of record. They deriveddata times series for durations of 1, 7, 30, 60, 90 and365 days for each of the basins. In turn, four threeparameterdistributions, namely the generalizedextreme value distribution, generalized logisticdistribution (GL), P3, and generalized Pareto distributionwere used to fit the data for each of the seriesand for each duration. Goodness-of-fit tests andjudgment were used to discern the results. The findingswere as follows:(a) The candidate distributions fit the observeddata points very well, with little quantitativeevidence to differentiate between them;(b) Certain distributions performed better thanothers with the distribution type varying withduration and basin characteristics;(c) The base flow index (Institute of Hydrology,1980) was very useful to quantify basingeology;(d) With regard to less permeable basins, the P3provided the best results for shorter durations,with the generalized extreme value surpassingthe P3 for longer durations;(e) For more permeable basins, the GL providedthe best results.It has been commonly observed (Nathan andMcMahon, 1990; Durrans and Tomic, 2001) thatthe largest flows within a series of minima are oftendescribed more effectively by a much steeper cumulativedistribution curve than would be used todescribe the subsequent lower flows. In response tothis phenomenon, approaches have been developedto fit only the lower portion or tail of thedistribution, rather than fitting the distribution tothe entire sample. Nathan and McMahon (1990)noted that a transition seems to occur where the“higher frequency flows are no longer considered”as low flows but represent more “normal conditions”.Approaches such as conditional probabilityadjustment (Condie and Cheng, 1982; Nathan andMcMahon, 1990), the application of censoringtheory (Kroll and Stedinger, 1996), mixture orcompound parametric models (Waylen and Woo,1987) and non-parametric frequency approaches(Adamowski, 1996; Guo and others, 1996) havebeen advocated to compensate for sample heterogeneity.Such approaches can also be used toperform an analysis when zero flow values arepresent in the sample.Durrans and Tomic (2001) explored the performanceof a number of methods that place an emphasison fitting only the tails of the distributions. Theyconcluded that the various methods performed“about as well as, if not better than, an estimationstrategy involving fitting” the entire dataset to theLN distribution using L–moments. In contrast tothis approach, for areas where annual or seasonallow-flow series may be generated by more than onemechanism and if these mechanisms can be identified,a mixture or compound parametric modelcould provide a more reasonable description of thedata. Alternatively, non-parametric frequency estimation,as proposed by Adamowski (1996) and Guoand others (1996), could be employed. Furthermore,it has been shown that non-parametric estimationprocedures provide estimates of low-flow quantilesas accurate as or more accurate than those producedby more traditional parametric approaches, namelythe LP3, W2 and W3 distributions, based on simulationexperiments with homogenous samples.Low-flow statistics are generally computed for periodsor durations of prescribed lengths, such as 1, 3,7, 14, 30, 60, 90, 120, 183 and 365 days. The lowflowdata for various durations are computed usinga moving average for the desired period. The movingaverage is the lowest arithmetically averaged flowof d consecutive days within a given year. As a rule,these values are computed over a hydrological orclimatic year rather than a calendar year. The hydrologicalyear is defined to start in a season when theflow is most likely to be high so that yearly lowflowperiods are not likely to be partitioned intodifferent years. Statistics such as the mean annuald-day minimum can be computed, as can the d-day,T-year low-flow statistic, commonly denoted asQd,T. In general, the specific d-day duration isselected according to agricultural, biological orengineering applications, which are usually relatedto the impact of the risk associated with the durationof low water availability on the system understudy. Computational methods for estimating theparameters of the distribution of the d-day seriesare similar to the methods described for floodfrequency analysis, with some minor variations,such as the parameter estimation method of smallestobserved drought for the G3 distribution.Two HOMS components are of particular interestfor estimating low-flow frequency statistics of d-daydurations. They are I80.2.03, the low-flow frequencyanalysis package, which allows testing of thehypotheses for randomness, homogeneity, trend

Discharge (m 3 s –1 )II.5-34GUIDE TO HYDROLOGICAL PRACTICESand independence, and I80.2.04, ProgramLOWSTATS, the low-flow statistical package.Limited analyses have been performed for durationsin excess of one year and the frequency ofthese multi-year flows have been determined usingplotting positions (Carswell and Bond, 1980;Paulson and others, 1991). Frequency analyses ofmulti-year low flows are important in water-supplystorage analysis where carry-over storage from yearto year is required to meet water-supply demands.HOMS component I80.2.05 Program DROUGHT,estimation of the probability of occurrence of n-month droughts, can be used to facilitate theanalysis of multi-year events.Examples of low-flow frequency curves for variousdurations are shown in Figure II.5.9. The low-flowdata are typically plotted with a logarithmic orarithmetic scale for the ordinate and a normal probabilityscale or Gumbel scale as the abscissa.Although few data samples will plot as a perfectstraight line, these types of paperare used to visuallyassess the overall fit of the model to the data.Special graph paper has been constructed to allowthe normal and Gumbel distribution to be drawn asa straight line. Methods have also been developedto change the scale of the abscissa for various threeparameterdistributions such that the cumulativedistribution function will plot as a straight line(Vogel and Kroll, 1989). This change of scale wouldbe valid for only one curve within the family ofcurves for the particular family of distributions, asthe skewness would most likely change with duration.The technique of adjusting the abscissa toreflect sample skewness is not commonly employedin the graphical representation of low-flow results.5.8.3 Low-flow frequency estimation atpartial-record sites using base-flowmeasurementsThe methods described thus far are valid for gaugedsites having sufficient data upon which to base a0.01 0.09 0.5 0.8 0.9 0.95 0.975 0.992010864327418321206030711.01 1.1 2 3 4 5 6 7 8 10 20 40 60 100Recurrence interval (years)Figure II.5.9. Frequency curves of annual low flowLength of period (days)frequency analysis: usually 10 years or more.However, discharge measurements made atungauged sites during times of low or base flow canbe used in conjunction with concurrent daily flowsat nearby gauged sites to estimate low-flowfrequency. Sites where only base-flow measurementsare available are referred to as partial-recordsites. A relation is established between the base-flowmeasurements at the partial-record site and concurrentdaily flows at the nearby gauged site. Thisrelation and low-flow characteristics at the gaugedsite are used to estimate d-day, T-year flows at thepartial-record site. The gauged site should havetopographic, climatic and geological characteristicssimilar to the partial-record site. In order to achievea linear relation, the logarithms of concurrent baseflowmeasurements y ~ , at the partial-record site, anddaily flows x ~ , at the gauged site, are normally usedto estimate the parameters of the linear relation.Such observations should be separated by significantstorm events so as to represent reasonablyindependent observations of the low-flow processes.At least 10 paired observations are needed todefine the relation between concurrentbase-flow measurements y ~ , and the daily flows x ~ .The analysis is based on the assumption or approximationthat the relation between y ~ and x ~ can bedescribed by:y ~ = a + bx ~ + e e ~ N(0,s 2 e ) (5.47)Where: a, b, and s 2 eare the constant, coefficient andvariance, respectively, of the linear regression equation.It is assumed that the residuals, e, areindependent and normally distributed. The estimatorsof the mean, M(y), and variance, S 2 (y), of theannual minimum d-day low flows at the partialrecordsite are as follows:M(y) = a + b M(x) (5.48)andS 2 (y) = b 2 S 2 (x) + S 2 e [1 – (S2 (x)/(L–1)S 2 (x ~ ))] (5.49)where M(x) and S 2 (x) are the estimators of the meanand variance of the annual minimum d-day lowflows at the gauged site, L is the number of baseflowmeasurement and S 2 (x ~ ) is the variance of theconcurrent daily flows at the gauged site.The primary assumption is that the relationshipbetween instantaneous base flows, as shown inequation 5.47, is the same as the relation betweenthe annual minimum d-day low flows at the twosites. This is a necessary assumption if theproposed method is to be used to estimate the

CHAPTER 5. EXTREME VALUE ANALYSISII.5-35d-day, T-year low flows at the partial-recordstation. While this approximation appears reasonablefor d-day means up to about seven days, itmay not be satisfactory for durations significantlylonger than this. Stedinger and Thomas (1985),and Thomas and Stedinger (1991) discuss the useof base-flow measurements to estimate low-flowcharacteristics at partial-record sites in the UnitedStates.The d-day, T-year low flow at the partial-record siteis estimated using the mean and variance given inequations 5.48 and 5.49. If a three-parameter distributionis used to estimate the d-day, T-year flow atthe partial record site, then the skewness of thegauged site is assumed to be the same that at thepartial-record site. As described earlier, the d-day, T-year low flow at the gauged site can be estimatedusing procedures described in 5.8.2. Stedinger andThomas (1985) explain why the d-day, T-year lowflow at the gauged site cannot simply be used as theindependent variable in equation 5.47. A loss ofvariance is associated with using least-squaresregression equations to estimate frequency estimatessuch as the d-day, T-year low flow. Inparticular, substituting the d-day, T-year low flow atthe gauged site in equation 5.47 would tend tooverestimate the d-day, T-year low flow at thepartial-record site. Stedinger and Thomas (1985)developed a procedure for obtaining an unbiasedestimate of the variance of the annual d-day flowsat the partial-record site using the relation in equation5.49 and the variances of the annual d-day lowflows and the concurrent daily flows at the gaugedsite.Stedinger and Thomas (1985) also developed aprocedure for estimating the standard error of thed-day, T-year low flow at partial-record stations.They illustrate that the standard error is a functionof the correlation between the base-flow measurementsand daily flows, the number of base-flowmeasurements made at the partial-record site, themagnitude of the departure of the d-day, T-year lowflow at the gauged site from the mean of the dailyflows used in equation 5.43 and the record lengthat the gauged site. Using data for 20 pairs of gaugingstations in the eastern United States, Stedingerand Thomas (1985) illustrated that standard errorsof about 30 percent can be achieved for partialrecordstations when the correlation coefficientsexceed about 0.7 and there are 15 base-flow measurementsand 25 years of record at the gauged site.Using data for 1 300 gauging station in the continentalUnited States, Reilly and Kroll (2003)demonstrated that the base-flow correlationapproach gave improved results over regionalregression models for 15 of the 18 major river basinsin the United States. Because the method utilizesat-site data, the base-flow correlation method generallyprovides more accurate estimates of d-day,T-year low flows than would the regional regressionmodels described in the next section.5.8.4 Regionalization of low-flowfrequency statisticsThe methods described thus far are valid for siteshaving sufficient data upon which to base afrequency analysis or for which base flow measurementsare available. Such sites should be relativelyfree of human intervention and should be of sufficientrecord length as to provide an accuraterepresentation of low-flow statistics for the basin.These statistics can be estimated for ungaugedbasins based on regionalization methods orthrough the abstraction of statistics from generatedtime series data obtained through statisticalor deterministic modeling. The first approach ismost commonly used to estimate the low-flowstatistic of interest, for example the seven-day,two-year low flow, Q7,2, at ungauged sites. Thestatistic of interest is regressed against a number ofindependent or explanatory variables. These independentvariables represent physical and climaticcharacteristics of the basin. Such approaches havebeen used with success for the estimation of designfloods, but it has been found to be much moredifficult to find accurate regression models to estimatelow-flow statistics (Vogel and Kroll, 1992;Waltemeyer, 2002).Regionalization generally entails the identificationof homogeneous regions over which a particularregression equation applies. Regionalization is anattempt to group basins geographically or in amultivariate space, which may not result ingeographically contiguous regions, based on physiographic,climatic or streamflow characteristics. Ingeneraly, the ability to define homogeneous regionsresults in increased predictive accuracy and moremeaningful physical models for the statistical estimationprocedure (Nathan and McMahon, 1992;Waltemeyer, 2002).HOMS component K10.2.04, regional analyses ofstreamflow characteristics, describes approaches fordeveloping regional relationships between streamflowand basin characteristics.Regional low-flow models are generally expressedin the following form:Q d,T= aX 1 b X 2 c X 3 d … (5.50)

II.5-36GUIDE TO HYDROLOGICAL PRACTICESwhere Q d,Tis the d-day, T-year low-flow statistic, theX iare basin physiographic or climatic characteristics,and a, b, c and d are parameters obtainedthrough multiple regression analysis (Weisberg,1980; Draper and Smith, 1981). Various low-flowstatistics are estimated from an at-site frequencyanalysis of the data from different sites within aregion. Basin and climatic characteristics are, inturn, derived from maps or from climatologicaldata (see Institute of Hydrology (1980), Vogel andKroll (1992) and Waltemeyer (2002). The parametersof the equation can be estimated using ordinary,weighted or generalized least squares techniques.Although the technique of generalized least squaresis more difficult to apply than ordinary least squares,Vogel and Kroll (1990) observed in their modelingof 23 basins in Massachusetts that the estimatedparameters and the t-ratios obtained using the twoapproaches were almost identical. However, themore complex approach provides information onthe composition of the error of prediction, allowingthe attribution of error to model error, measurementerror and sampling uncertainty. Vogel andKroll (1990) noted that model error was by far themajor component of the prediction error. Theiranalysis helps to emphasize the importance ofestablishing more physically meaningful statisticallybased model.Regional regression equations of the form of equation5.50 are applicable for regions where the d-day,T-year low flows are non-zero. Tasker (1991) hasdeveloped procedures for estimating low flows inregions where the d-day, T-year low flow may bezero. These techniques involve developing regionalrelationships with censored data and the use oflogistic regression to estimate the probability of thed-day, T-year being zero.Numerous basin and climatic characteristics havebeen used in regional regression equations toestimate low-flow statistics. Most models includea drainage area and a variable representingclimatic conditions, such as mean annual precipitation.Many other characteristics have beenconsidered, some of which are the mean watershedelevation, proportion of basin in forestcover, proportion of basin in lakes and swamps,average basin slope, drainage density, main channelslope and proportion of urban area. Giventhat low flows are normally a result of theprolonged absence of rainfall, it is commonlythought that their low-flow characteristics shouldbe closely related to the underlying geologicaland soil characteristics of the basin (Institute ofHydrology, 1980; Task Committee of theHydraulics Division, 1980).In certain cases, improved relationships have beenattained by including an explanatory variable representinga geological index. Such indexes have seenincreasing popularity and have led to increases inmodel performance. The base flow index (Instituteof Hydrology, 1980) could be considered to reflect,in part, basin geology and is the ratio of flow, generallyknown as baseflow, to the total flow. Accordingto Gustard and Irving (1994), a soil index can leadto improved prediction models.Another approach has been taken to improve linkagesbetween low-flow characteristics and recessioncurves or coefficients for the basin. Bingham(1982) defined a streamflow recession index, indays per log cycle of discharge depletion, at gaugedstreams in Alabama and Tennessee (United States)and then mapped the index according to computedindices at gauging stations and a geological mapfor use in estimating low-flow characteristics forungauged streams. Vogel and Kroll (1992) formulateda conceptual model of the form of equation5.50 to relate the unregulated flow of a basinduring recession periods to the basin’s characteristics.They regressed Q7,10 with three of the fivevariables of the conceptual model. Vogel and Krollfound dramatic increases in accuracy by inclusionof the three variables in the final regression model.The characteristics they considered were drainagearea, the base flow recession constant and the averagebasin slope. The final equation, althoughaccurate, cannot be used directly at an ungaugedsite without additional efforts being required toestimate the base flow constant. Vogel and Krollsuggest that this independent variable could beestimated from maps that would have to be developedor could be obtained through a modest andtargeted streamflow gauging program. They suggestthat the recession constant could be estimatedsimply by observing a few recession hydrographs.Other regional low-flow studies in the UnitedStates have used soils characteristics (Carpenterand Hayes, 1996) and the slope of the flow-durationcurve (Arihood and Glatfelter, 1991) asexplanatory variables in estimating low-flow characteristics.Arihood and Glatfelter (1986) mappedthe ratio of the 20-percent to 90-percent flow durationin Indiana for use in estimating low-flowcharacteristics for ungauged watersheds. Flowdurationcurves are discussed in the next sectionof this paper.5.8.5 Flow-duration curvesFlow-duration curves of daily discharge show thepercentage of days that the flow of a stream is

Discharge (m3 s-1)CHAPTER 5. EXTREME VALUE ANALYSISII.5-37greater than or equal to given amounts over a givenperiod of record. However, they provide no informationon the temporal sequences of the flows at asite or the probability of exceedance or nonexceedancein any given year. Even with this temporallimitation, flow-duration curves have a long historyof use in water resources planning and managementfor a variety of purposes. Some of the mostcommon uses of flow-duration curves are incomputing hydroelectric power potential for primepower and secondary power, water-supply and irrigationplanning, waste-load allocations and otherwater-quality management problems. Other usesinclude the determination of wastewater-treatmentplantcapacity, river and reservoir sedimentationstudies, instream flow requirements for habitatmanagement and the determination of optimalallocation of water withdrawals from reservoirs.They have also been found to be very simple anduseful for graphically illustrating flow characteristicsfrom flood to low flows for a basin. The shapeof the curve can vary from basin to basin, reflectingdifferences in physiography and climatology. Theyare also useful for illustrating impacts of interventionon water availability and can be used for a hostof other purposes.A flow-duration curve is usually constructedempirically by computing a series of ratios of thenumber of days in a streamflow record that havedischarges greater than or equal to preselectedvalues divided by the total number of days in therecord. The ratios, which are estimates of the probabilities,are plotted against their respectivedischarge values to construct the curve. A durationcurve of streamflow will generally plot as roughlya straight line on logarithmic probability paper,such as the one shown in Figure II.5.10. This typeof paper gives equal plotting accuracy at alldischarges so that differences in low-flow characteristicscan be discerned more precisely.Flow-duration curves are sometimes based onweekly or me. Such curves are usually less usefulthan a daily duration curve.Flow-duration curves can also be computed for eachyear, with the average or median of the annualbasedflow-duration curves representing the typicalcurve (Vogel and Fennessey, 1994). These allow thedevelopment of confidence intervals and returnperiods to be associated with the flow-durationcurve, and the resultant median annual flow-durationcurve is less sensitive to extreme periods ofobservations that may arise over the history of asite.The overall shape and, in particular, the shape ofthe lower portion of flow-duration curve is anindicator of the physiographic, geological andclimatic conditions of the basin. Of most interestin low-flow studies is the shape of the lower portionof the flow-duration curve. A low-sloping lowerportion implies that the basin is permeable andthat the response of the basin to rainfall is notflashy. In contrast, a higher-sloping lower curveimplies that the basin is less permeable and probablyprovides a flashy response for a given input ofrainfall. A basin with a higher permeability wouldalso tend to have a higher base flow index thanthe basin with lower permeability (Zaidman andothers, 2003).Regional relationships can be developed to provideestimates of flow duration for ungauged sites withina homogeneous region (Institute of Hydrology,1980; Fennessey and Vogel, 1990; Ries, 1994).Multiple regression models similar to those outlinedfor the estimation of low-flow statistics, such as theQ7,10, can also be developed for this purpose. Thedependent variable would be, for example, the100806050403020If the streamflow data are stationary, the derivedflow-duration curve should provide the longtermexceedance probabilities for the entirerange of flows, which is a useful planning tool.The tails of the flow-duration curve have beenfound to be sensitive to the number of yearsused to estimate the curve, which is a form ofsampling error. Additional details on constructionof flow-duration curves are available inother sources (see, for example, Searcy (1959),Institute of Hydrology (1980), and Vogel andFennessey (1994).108643210.5 1 2 5 10 20 50 80 90 95 98 99 99.599.899.9Per cent of time daily discharge exceeded that shownFigure II.5.10. Flow-duration curve of dailydischarge

II.5-38GUIDE TO HYDROLOGICAL PRACTICESvalue of the flow exceeded 95 per cent of the time,denoted as Q95 (Institute of Hydrology, 1980). Theindependent variables of such relationships are alsosimilar to those for other low-flow statistics andwould reflect basin characteristics and climaticconditions, such as the drainage area and long-termmean annual precipitation in the basin. HOMScomponent K10.2.05, regionalization of flow-durationcurves, or REGFLOW, can be used to estimateflow-duration curves. It can also be used to relatethese to geomorphological characteristics so thatflow-duration curves may be estimated for ungaugedbasins.5.9 FREQUENCY ANALYSIS OF FLOODFLOWS [HOMS H83, I81, K10, K15]In a number of cases, for example, in storage-reservoirdesign, it is necessary to establish the frequencyof flood volumes as well as peak flows. A multivariatestatistical analysis of flood hydrographs may beused in this case. A flood hydrograph may bedefined by means of the following randomvariables:Qmax is the maximum discharge during the floodperiod; V is the volume (in m 3 ) of the flood wave;and T is the duration of the flood period.By using another system of variables, a floodhydrograph may be defined by means of thesequences of discharges Q 1, Q 2, Q 3, ..., Q ncorrespondingto successive equal intervals of timeduring the flood period. Statistical analysis of therandom variables (Q, V, T) or (Q 1, ..., Q n) may beperformed by means of a multivariate probabilitydistribution. Some definitions and computationaltechniques connected with such probabilisticmodels may be found in Cavadias (1990). In thecase of flood characteristics, a power transformationor other methods may be used to normalizethe data. Alternatively, the frequency or probabilityof occurrence or non-occurrence of a floodvolume for an n-day period can be directly estimatedby performing a frequency analysis of thesite flow data or employing regionalizationmethods.The purpose of computing flood and rainfallfrequencies is to relate the magnitude of a flood orrainfall depth to its frequency or probability offuture occurrence. The key assumptions used toallow interpretation of the frequencies as probabilitiesare temporal independence of the elements ofthe analysed sample and stationarity of the record.For flood studies, the use of partial duration series ismore questionable than for rainfall, as the differentpeak floods during the year may be less independentthan the corresponding precipitation. However,if care is taken in the selection of the values exceedinga given threshold, a partial duration seriesanalysis should be suitable. The application offrequency analysis to a series of the annual floodmaxima – maximum annual series – is morecommon.The maximum annual series may be comprised ofeither daily maxima or instantaneous flood peaks.It is important to distinguish which of the two isrequired for the analysis. The relation of the twoseries at a site is dependent on the physicalcharacteristics of the watershed as well as theclimatologic factors causing the maxima of bothevents. For very small drainage areas, it is commonthat the two maxima do not occur on the same datenor as a result of the same climatic processes actingon the watershed.Thus, the simplest and most straightforwardapproach to estimate the frequency of large floodsis to use the record available at a site to fit one ofthe frequency distributions described in 5.1,employing an estimation procedure (see 5.5).Unfortunately, records are not always available atthe sites of interest and records may be too short toprovide reliable estimates of the rare floods ofconcern. Thus, most of the discussion in this sectionaddresses the use of information at more than onesite to estimate flood quantiles at sites which do nothave flood record.Caution must also be observed in computingfrequencies of floods: a clear distinction should bemade between stages and discharges. Naturalchanges in the stage–discharge relationship withtime or direct intervention in the channel mayrender many stage data non-homogeneous andunsuitable for frequency analysis. For most studies,it is preferable to work with discharges, and, ifnecessary, to then convert the results to stagefrequency using an appropriate stage–dischargerelationship. In certain cases, such as high stagescaused by ice jams, it may be more suitable to worksolely with stages for defining flood plains becausethe flow rate is not an issue.5.9.1 Regionalization of flood flowsFor a site that does not have a large number ofobservations in its maximum annual series,regional flood frequency analysis is recommendedfor the estimation of the flood quantiles. Even

CHAPTER 5. EXTREME VALUE ANALYSISII.5-39with 50 years of data it can be very difficult toregionalize the shape parameter of a distribution.As the record gets shorter, regionalizing the coefficientof variance should be considered. However,the point at which it becomes appropriate toregionalize depends on the homogeneity of theregions that can be constructed and the relativeaccuracy of at-site estimators, which depends uponthe at-site coefficient of variation and the skewnessof the flood distribution in the region. Twopopular regionalization procedures are the indexflood method and the regression-based procedures;Fill and Stedinger (1998) explore the combinationof the two. Regional procedures rely on data availablefrom other stations in the same hydrologicregion to obtain estimates of flood characteristicsat the site of interest. Cunnane (1988) indicatedthat a regional approach can produce more accurateflood estimates, even when a large number ofobservations are available at that site. In general,there are two steps in a regional flood frequencyprocedure:(a) The delineation of hydrologically homogeneousregions consisting of identification ofstations with similar behaviour;(b) Regional estimation, which involves informationtransfer from gauged sites to the site ofinterest within the same region.Homogeneous regions in can be defined in threedifferent ways, as illustrated by Figure II.5.11:(a) As fixed geographically contiguous regions;(b) As fixed geographically non-contiguousregions;(c) As neighbourhoods, where each target stationis associated with its own region.Regional flood estimation procedures can be definedby considering various combination techniques forthe determination of homogeneous regions and anumber of regional estimation methods (Stedingerand Tasker, 1986; Burn, 1990; Fill and Stedinger,1998; Pandey and Nguyen, 1999). GREHYS (1996a,1996b) presented the results of an inter-comparisonof various regional flood estimation proceduresobtained by coupling four homogeneous regiondelineation methods and seven regional estimationmethods. GREHYS (1996b) concluded that theneighborhood approach for the delineation ofgroups of hydrologically homogeneous basins issuperior to the fixed-region approach. Hydrologicalneighborhoods can be determined by using theregion-of-influence method (Burn, 1990) or canonicalcorrelation analysis (Cavadias, 1990; Ouardaand others, 1998). Regional flood estimation can becarried out using the index flood method or multipleregressions.5.9.2 Homogeneous region delineation5.9.2.1 Region-of-influence methodThe region-of-influence method (Burn, 1990),considers each site as the centre of its own region.The identification of a region of influence for atarget site is based on a Euclidian distance measurein a multidimensional attribute space. The setof attributes can be related to extreme flowRegion 1Region 2Region 3Region 4(a) Geographicallycontinuous regionsRegion 1Region 2Region 3Region 4(a) Non-contiguoushomogeneous regionsUngauged target siteNeighbouring stationNon-neighbouring station(c) Hydrologic neighbourhoodsFigure II.5.11. Approaches for the delineation of homogeneous regions (Ouarda and others, 2001)

II.5-40GUIDE TO HYDROLOGICAL PRACTICEScharacteristics of catchments. A weight function isdefined to reflect the relative importance of eachsite for regional estimation at the target site. In theoriginal approach, flow attributes are used todefine the region of influence, implying that thesite of interest must be gauged. For ungauged sites,climatological and physiographical informationmay be used as a surrogate for hydrologicalinformation. Hence, several versions of the regionof influence approach can be considered here,depending on whether the target site is gauged orungauged, and depending on the space of theattributes. Hydrological attributes that can beconsidered are the coefficient of variation ofmaximum floods and the ratio of the meanmaximum flow to the drainage area. Otherattributes include the longitude, the latitude andmeteorological attributes associated with floodevents such as the mean total annual precipitation,or the mean snow depth on the ground five daysbefore the spring flood.The weighted Euclidian distance in the attributespace, D ij, between two sites i and j is given by thefollowing equation:D ij =Mi∑ ω m (C mj 2− C m) (5.51)m=1where M is the number of attributes considered,and C miand C mjare the standardized values of theattribute m for sites i and j. The attributes are standardizedby division by their standard deviation overthe entire set of stations. The next step is to select athreshold value, ω on D ij, to define the limit ofinclusion of stations in the region of influence of atarget site.5.9.2.2 Canonical correlation analysisCanonical correlation analysis is a multivariatestatistical technique that allows a reduction in thedimensionality of linear dependence problemsbetween two groups of variables. This method canbe used to identify sites with flood regimes similarto the target site (Cavadias, 1990; Ouarda andothers, 1997).Ouarda and others (1997) demonstrated that themultiple regression method and the index floodmethod give equivalent results when combinedwith the canonical correlation analysis. Ouarda andothers (1999) presented an automated and transposableregional procedure based on canonicalcorrelation analysis and multiple regressions. Thegeneral methodology presented in Ouarda andothers (2000) allows the joint regional estimationof flood peaks and flood volumes. A more detaileddescription of the canonical correlation analysismethodology for regional frequency estimation isavailable in Ouarda and others (2001). A generaldescription of the method can be found in Muirhead(1982).5.9.3 Regional flood estimation methodsThe second step of regional analysis consists ininferring flood information, such as quantiles, atthe target site using data from the stations identifiedin the first step of the analysis. Regionalestimation can be accomplished using the indexfloodor regression methods.5.9.3.1 The index-flood procedureThe index-flood procedure consists of two majorsteps. The first is the development of the dimensionlessfrequency curve for a homogeneous region.The curve is derived from individual frequencyanalyses of all sites. The curve for each site is madedimensionless by dividing the curve by an index,such as the flood corresponding to the two-year or2.33-year return period or the mean. The mediandimensionless values are selected for the sites forvarious return periods. They are in turn plotted onprobability paper. The second step consists of thedevelopment of a relationship between the indexand the physical and climatological characteristicsof the watershed using regression-based procedures.The combination of the index with the dimensionlesscurve provides a frequency curve for anywatershed within the region.Much work has been done to extend these initialconcepts and assess the accuracy of index proceduresto determine various flood quantiles, forexample in Gabriele and Arnell (1991). Advanceshave been facilitated by the development of probability-weighted-moment(Greanwood and others,1979) and L–moment (Hosking, 1990) statistics.The need for analytical homogeneity tests can becircumvented by the use of Monte Carlo experiments.Homogeneity should and can be extendedfrom the slope of the curve, which is the coefficientof variation of the sample in Dalrymple’s approach,to include the skewness and kurtosis of the proposedregion. This leads to a more flexible index procedurethat allows higher moments of the region’sdata to indicate the potential underlying distribution.Heterogeneity of the lower moments can beassessed and potentially linked to characteristics ofthe watershed. Hosking and Wallis (1988) showthat “even when both heterogeneity and intersite

CHAPTER 5. EXTREME VALUE ANALYSISII.5-41dependence are present and the form of the[regional] flood-frequency distribution is mis-specified,regional flood frequency analysis is preferableto at-site analysis”. The index-flood method hasbeen found to be one of the most efficient regionalizationtechniques.5.9.3.2 Regression-based proceduresRegression techniques can be used to estimate themagnitude of a flood event that will occur on averageonce in Tr years, denoted Q TR, by using physicaland climatological watershed characteristics. Themagnitudes of flood events for various return periodsfor each gauging station are estimated by usinga preselected distribution from an at-site frequencyanalysis. In turn, characteristics for each watershedare derived from topographic maps or from generalizedclimatological data. The parameters of theequations that relate Q TRto the characteristics canbe obtained by using ordinary least squares,weighted least squares or generalized least squarestechniques. The latter two approaches have beenused to overcome the deficiencies in the assumptionsof ordinary least squares. Ordinary leastsquares regression procedures do not account forvariable errors in flood characteristics caused byunequal record lengths at gauging stations. Tasker(1980) proposed the use of weighted least squaresregression with the variance of the errors of theobserved flood characteristics estimated as aninverse function of the record length. Generalizedleast squares have been proposed because they canaccount for both the unequal reliability and thecorrelation of flood characteristics that existbetween sites. Using Monte Carlo simulation,Stedinger and Tasker (1985 and 1986) demonstratedthat the generalized least squares procedure providesmore accurate estimates of regression coefficients,better estimates of the accuracy of the regressioncoefficients and better estimates of the modelerror.The regional flood–frequency relationship developedby Benson (1962) for the north-eastern UnitedStates is as follows:Q TR= aA b Z c S d P e D f M g (5.52)where Q TRis the T-year annual peak discharge, A isthe drainage area, Z is the main channel slope, S isthe percent of surface storage area plus 0.5 per cent,P is the T-year rainfall intensity for a particular duration,D is the average January degrees below freezing,M is an orographic factor, and a, b, c, d, e, f and g areregression coefficients. Many independent variableswere tested to derive equation 5.52 and manydefinitions were tested for each variable. The goal isto obtain independent variables that are physicallyrelated to the dependent variable. Independentvariables that are related to a low return-periodflood may not be a driving force behind a higherreturn-period flood. A logarithmic transformationof equation 5.47 may be taken to create a linearadditive model for the regression procedures. Othertypes of transformations could be applied to thedependent and independent variables, but the logarithmictransformation remains the most popular.Both the signs and the magnitude of the coefficientsof the model should make hydrological sense. Forexample, the exponent d of the surface-storage termshould be negative because of the effect of suchstorage (lakes, reservoirs and so forth) in flatteningout flood peaks. Other exponents should be positivewith their magnitudes varying with the returnperiod. Care should be taken to ensure that not toomany independent variables are included in themodel. The variables included in the regressionmodel should be statistically significant at somepreselected and generally accepted level of significance(Draper and Smith, 1981).The resulting regression equation should be evaluatedto determine if it is regionally homogeneous.Residual errors of the regression should be plottedon topographic maps to check visually if geographicbiases are evident. If a bias in the estimation of theT-year annual peak discharge is geographicallyevident, then the Wilcoxon signed-rank test can beapplied to test its significance. The test provides anobjective method for checking the hypothesis thatthe median of the residuals in a sub-region is equalto the median residual of the parent region forwhich the regression equation was computed.Different homogeneous regions may be found fordifferent return periods. The homogeneous regionfor the relationship linking the index flood to thecharacteristics of the watershed need not coincidewith the homogeneous region for the characteristicsof the distribution of the index method, such asthe slope of the dimensionless curve.In practice, the power form function is the mostcommonly used model to describe the relationshipbetween the at-site estimates of flood quantiles Q Tand the hydrometeorological and basin characteristicsfor the region identified in the first step of theprocedure. A common procedure for the estimationof the parameters consists in lineralizing the powerrelationship by a logarithmic transformation, andthen estimating the parameters of the linearizingmodel by an ordinary least squares technique. Theusual procedure is therefore straightforward,because one can make use of multiple linear

II.5-42GUIDE TO HYDROLOGICAL PRACTICESregression techniques to identify the parameters ofa nonlinear model.An advantage of the multiple regression regionalestimation models is the flexibility in choosing thetype of distribution to represent the exceedances ateach site. The regression-based regional estimationmethod can also be applied using peaks-overthresholddata, in which case the generalized Pareto,exponential, and Weibull distributions can be used.Both the generalized Pareto distribution and theWeibull distribution contain the less flexible exponentialdistribution as a special case. In thepeaks-over-threshold approach, all flood peaksabove a prefixed threshold are considered. The lackof detailed guidelines for choosing the most appropriatethreshold constitutes a serious drawback ofthe method and is probably one reason why it isless used in practice than its annual flood seriescounterpart. For a review of various methods forthreshold selection, see Lang and others (1999).A regression-based method can also be performedusing non-parametric frequency analyis, whichdoes not require a priori distribution selection.Adamowski (1989) and Guo (1991) found that nonparametricmethods are particularly suitable formultimodal annual flood data following mixeddistributions. Non-parametric density estimationhas been used successfully in a regional framework(GREHYS, 1996b), including non-parametric regression(Gingras and others, 1995). As well, theL–moments technique can be used at all stages ofregional analysis including homogeneous regiondelineation and testing, identification and testingof regional distributions and quantile estimation(Hosking and Wallis, 1997).5.9.4 At-site and regional flow–duration–frequency approachMost of the regional flood frequency analysis literaturedescribes a flood event only by its instantaneouspeak or its maximum daily flow. When designing ahydraulic structure or mapping a flood plain, informationabout flood peaks is essential, but moreinformation may be desired. Indeed, flood severityis not only defined by the flood’s peak value, butalso by its volume and duration. The analysis offlow duration frequency, or QDF (Sherwood, 1994;Javelle, 2001), also known as flood durationfrequency or discharge deviation frequency, hasbeen proposed as an approach for a more thoroughdescription of a flood event. Flow–duration–frequency analysis is similar to theintensity–duration–frequency analysis commonlyutilized for rainfall (see 5.7 above). In this case,averaged discharges are computed over differentfixed durations D. For each duration, a frequencydistribution of maximum discharges is thenanalysed. Finally, a continuous formulation is fittedas a function of the return period (T) and the duration(D) over which discharges have been averaged.Javelle and others (2002) proposed a convergingflow–duration–frequency model based on theassumption of convergence between the differentdischarge distributions for small return periods.This formulation has been successfully tested forbasins located in France, Martinique and Canada.Javelle and others (2002) have also presented aregional flow–duration–frequency approach,combining the local flow–duration–frequencyformulation presented by Javelle (2001) and theindex flood method outlined in 5.9.3.1, which iscommonly used in regional flood frequency analysis.This regional model was developed by Javelleand others (2003) for 169 catchments in theCanadian provinces of Quebec and Ontario, and itwas used to define different types of flood behaviourand identify the corresponding geographic regions.Javelle and others (2003) showed that the parametersof the regional flow–duration–frequency modelprovide information about the flood dynamics.Unlike the intensity–duration–frequency analysisfor rainfall, flow–duration–frequency analysisremains under-utilized despite its strong potential.5.9.5 Combination of single-site andregional dataThe objective of procedures that aim to combinesingle-site and regional information is to improveupon at-site estimates that are based on a limitedseries of site data by using available informationfrom other sites. The need for such procedures isparticularly great in the estimation of extremehydrological phenomena where a combination oflimited site data and inference in the tails of probabilitydistributions conspire to destabilize suchestimators. A simple Bayesian approach presentedby Fortin and others (1998) combines local andregional quantile estimates knowing the varianceof estimation for each estimate. The United Stateshas guidelines for combining at-site quantile estimatesobtained by regional regression using thestandard error of each (Interagency AdvisoryCommittee on Water Data, 1982). The approachpresented by Kuczera (1982) and evaluated byLettenmaier and Potter (1985) is based on an empiricalBayes model that combines an at-site andregional variance and was shown to lead to substantialimprovements in performance over proceduresthat only used at-site information.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-43Clearly, regional hydrological information shouldbe of value in improving flood estimates, particularlywith regard to the shape and characteristics ofthe tail of the distribution, as these are hard toresolve with limited at-site datasets. For this reason,procedures adopted in many countries employsome combination of at-site skew, as well as the atsitemean and standard deviation so as to estimatea flood frequency distribution. In certain cases, onlythe skew is regionalized, and the regional skew isaverage with at-site skew. In the United Kingdom,the general procedure is to use an index flood procedurethat uses the at-site mean with a regionalgrowth curve to define flood risk at a gauged site(Robson and Reed, 1999), so that the value of twoparameters of the fitted logistic distribution aredetermined by regional data.Striking the right balance between the use ofregional information and at-site records to definethe frequency curve is a challenge. Clearly the lessdata one has at a site, the less confidence one has inat-site estimation of statistics, and the more theweight that should be placed on regional information.Theat-site standard deviation can also beweighted with a regional value (Kuczera, 1982;Lettenmaier and Potter, 1985) or the at-site meanand standard deviation cab be used with a regionalshape estimator (Lettenmaier and others, 1987).Region-of-influence ideas are appropriate here indefining the set of sites used for pooling. Usingregional estimators of the coefficient of variationand skewness based on different spatial averagingscales in a hierarchical approach (Gabriele andArnell, 1991) or regression to describe how a growthcurve or a shape parameter varies continuouslywith basin characteristics (Madsen and Rosbjerg,1997) are available options. The appropriate choicedepends upon the homogeneity or heterogeneity ofthe region and other flood distribution characteristics,the length of the record available at differentsites and the time an agency has to determine andunderstand those trade-offs. Stedinger and Lu(1995) illustrate some of the trade-offs among thenumber of regionalized parameters, the length ofrecord and the number of sites available, regionalheterogeneity and flood distributioncharacteristics.5.9.6 Flood frequency analysis andclimate variabilityThe foregoing discussion has for the most partembodied the traditional assumption that floodseries are a set of independent and identicallydistributed random variables. If they are not entirelyindependent but instead have some modest correlationfrom year to year, it has relatively little impacton the analysis and the bias of estimated floodquantiles. The more troubling concern is either atrend in the distribution of floods due to developmentand other changes in the basin, or what hasbeen called climate variability and climate change.All three of these effects can have a significantimpact on flood risk in a basin.The easiest of the three to deal with is when changesin the basin – particularly land cover, the drainagenetwork and channel characteristics – or theconstruction and operation of detention structureshave evolved over time. A traditional record ofannual maximum floods is no longer effective indescribing the risk of flooding under the newregime. The traditional approach to handlingchanges in channel characteristics and the operationof storage structures is to route a historicalrecord of natural flows through a hydraulic modelto generate a record of regulated flows, which canbe used as a basis for frequency analysis.Alternatively, a frequency analysis can be conductedon the natural flows and a design natural flowhydrograph can be constructed that is, in turn,routed through the hydraulic model based on theassumption that owing to operation of the facility,the exceedance probability of the design hydrographwould be unchanged because smaller and largerevents would have resulted in smaller and largerflood peaks downstream, respectively.For complicated systems involving several streamsor storage facilities, or for basins that have experiencedsignificant land-cover and land-use change,it is advisable to use historical or synthetic rainfalland temperature series to drive physically basedrainfall-runoff and hydraulic models. Such a studyallows the analyst to appropriately describe theoperation of different facilities, network and channelmodifications, as well as the likely effect ofland-cover and land-use changes.Dealing with climate variability and climate changeis a difficult problem (Jain and Lall, 2001). NRC(1998) makes the following observation:Evidence accumulates that climate has changed,is changing and will continue to do so with orwithout anthropogenic influences. The longheld,implicit assumption that we live in arelatively stable climate system is thus no longertenable.Changes in hydroclimatological variables, bothrainfall and runoff, over different timescales are

II.5-44GUIDE TO HYDROLOGICAL PRACTICESnow well documented for sites around the world(Hirschboeck and others 2000; Pilon and Yue, 2002;Pekarova and others, 2003). Two cases are immediatelyclear, those corresponding to climate variabilityand climate change.The first concern, climate variability, relates to suchprocesses as the El Nino-Southern Oscillation or theNorth Atlantic Oscillations, which result in asporadic variation in the risk of flooding over timeon the scale of decades. In cases where the record isrelatively, it would be hoped that such phenomenawould have passed through several phases resultingin a reasonable picture of the long-term averagerisk. With short records, such variations are moreproblematic. It is always good practice to attempt touse longer records from the same region in order toadd balance to the short record. If a composite or across-correlation between the short record andlonger records in the region is reasonably high,record augmentation methods described in 5.5.4can be used to develop a more balanced, long-termdescription of flood risk. However, with smallercatchments where year-to-year events are highlyvariable, it may not be effective to use simple recordaugmentation to correct distinct differences inflood risk between different periods because thecross-correlation between concurrent annual peakswill be too small.For operational concerns, an option would be toforecast variations in the El Nino-SouthernOscillation, or other indices, and unexplainedhydrological variations, so as to forecast more accuratelythe flood risk in the current and subsequentyears and advise water operations accordingly(Piechota and Dracup, 1999). However, for projectplanning purposes such short-term variations arelikely to be too short lived to affect the economicdesign of projects.The second climate concern would be climatechange in one direction or another that is notquickly reversed within a decade or two. Suchclimate change is on the scale of decades and is avery serious concern. Even mild upward trends canresult in substantial increases in the frequency offlooding above a specified threshold, as shown byPorparto and Ridolfi (1998) and Olsen and others(1999). It is clear that anthropogenic impacts arenow inevitable. The question is how soon and howsevere. Guidance is much more difficult to providefor this case because there is no clear consensus onhow fast the Earth is likely to warm from the releaseof greenhouse gases into the Earth’s atmosphereand what the impact of those changes will be onmeteorological processes at a regional or watershedscale. Generalized circulation models of the Earth’satmosphere have given some vision of how localclimates may change, but the inability of suchmodels to capture current meteorological processesat a regional or watershed scale yields limited confidencethat they will be able to predict accuratelythe rate and intensity of future change. However,the hydrological implications of different generalizedcirculation model scenarios are ofteninvestigated to provide a vision of what the futuremay hold (see Arnell and others (2001)). And, asArnell (2003) points out, the future will be the resultof both natural climate variability and climatechange.5.10 ESTIMATING DESIGN FLOODS[HOMS K10, K15, I81, K22]5.10.1 GeneralThe design flood is defined as the flood hydrographor the instantaneous peak discharge adopted forthe design of a hydraulic structure or river controlafter accounting for political, social, economic andhydrological factors. It is the maximum floodagainst which the project is protected; its selectioninvolves choosing safety criteria and estimating theflood magnitude that meets the criteria. The risk ofdamage occurring is equivalent to the probabilityof occurrence of floods larger than the design flood.The decisive factor in the determination of a designflood is that feature or parameter of the flood thatcan be identified as the major cause of potentialdamage. The decision as to which is the most relevantflood parameter for a particular project lieswith the planner and the designer and should bebased on an engineering analysis of the given situation.Decisive parameters usually include thefollowing:(a) Peak discharge in the case of culverts, stormsewers, bridge openings, spillways and outletsof weirs and small dams;(b) Peak stage in the case of levees, clearance underbridges, flood-plain zoning and the design ofroads and railways in river valleys;(c) Flood volume for the design of flood-controlreservoirs and, generally, for all cases whereflood attenuation by water storage can besignificant, such as for the design of spillwaycapacities and freeboards on dams;(d) Flood hydrograph shape in cases where superpositionof several floods must be considered,such as for flood protection downstream fromthe mouth of large tributaries or for reservoiroperation during floods.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-455.10.2 Design floodsThe following types of design flood are commonlyused in water-resource engineering practice (Singh,1992):(a) Spillway design flood – a term often used in damdesign to identify a flood that a spillway mustbe able to pass to provide the desired degree ofprotection for a dam;(b) Construction flood – a flood for which reasonableprecautions will be taken to avoid floodingof construction sites and thereby to preventdamage to a project during its construction;(c) Probable maximum flood – the largest floodthat may be expected at a site, taking intoaccount all pertinent factors of location, meteorology,hydrology and terrain (see 5.7). Itessentially has an infinite return period andcan be selected as the design flood to prevent amajor disaster;(d) Standard project flood – a flood resulting fromthe most severe combination of meteorologicaland hydrological conditions that are consideredreasonably characteristic of the geographicalregion involved, excluding extremely rarecombinations. It has a long but unspecifiedreturn period and may be selected as a designflood for structures of great importance;(e) Frequency-based flood – a flood determinedemploying frequency analysis of flood flowsor rainfall data by performing one of thefollowing:(i) frequency analysis of rainfall data to estimatea frequency-based design storm,which is then converted to design flood;(ii) frequency analysis of flood flows availableat the site to directly estimate the designflood;(iii) regional frequency analysis to estimatethe design flood.5.10.2.1 Magnitude and methods ofcomputationA design flood can be estimated by transformingthe design storm to design flood using, for example,the unit hydrograph concept or flood frequencyanalysis. The latter requires long-term streamflowdata at the site of interest. If streamflow data areunavailable or a hydrograph is required, then thedesign flood can be estimated using either a rainfallfrequency analysis coupled with a rainfall-runoffmodel or a rainfall-runoff method that may beeither data-based, or hypothetical or empirical.The rainfall information used for design flood estimationis referred to as the design storm and canbe classified as probable maximum precipitation, astandard project storm, or a frequency-basedstorm. For structures involving low-damage risk,such as culverts and secondary roads, the designflood may be calculated by empirical methods,given the typically low return period of such structuresand their relatively low importance. Forstructures or projects involving major potentialdamage, but without a risk of loss of life, designfloods should be computed, if possible, by methodsallowing an evaluation of their return periodsso that optimization methods can be used for theselection of the design flood magnitude. For situationsinvolving a risk of loss of life, the aim is toprovide maximum protection, and the maximumprobable flood or the standard project flood isusually adopted as the design flood. It is advisableto evaluate the reasonableness of the probablemaximum flood by comparing it with observedrainfalls and floods.Only a few of the more practical and popular methodsfor calculating floods have been described inthis chapter. There are many other methods, someof which have been developed for particular regions,such as those described by Maidment (1993) andKundziewicz and others (1993). For example, theGRADEX method (Guillot, 1993; Ozga-Zielinski,2002) is based on the combined use of rainfall andflow records. It assumes that the upper tail of theflood is near an exponential asymptote (gradient)of rainfall. The Flood Estimation Handbook proposesa procedure developed by the Centre for Ecologyand Hydrology in the United Kingdom thatcombines statistical analysis and modelling ofprecipitation time series to the hydrological simulationof discharge at catchment scale (www.nerc-wallingford.ac.uk).5.10.2.2 Design life of a project and designcriteriaIn the wide range of cases in which the design floodis selected by optimizing the relation between theexpected flood damage and the cost of flood-protectionmeasures, the resulting optimum level of thecalculated risk depends to a certain degree on thelength of the period over which the performance ofthe project is evaluated. This period is called thedesign life or planning horizon of the project and isdetermined in the project-planning stage on thebasis of the following four time spans:(a) Physical life, which ends when a facility canno longer physically perform its intendedfunction;(b) Economic life, which ends when the incrementalbenefits from continued use no longer exceedthe incremental cost of continued operation;

II.5-46GUIDE TO HYDROLOGICAL PRACTICES(c) The period of analysis, which is the length oftime over which a facility may be expected tofunction under conditions that can be relativelyaccurately foreseen at the time of theanalysis; any operation in the distant futurethat is subject to a high degree of uncertainty isexcluded from consideration;(d) The construction horizon, which is reachedwhen a facility is no longer expected to satisfyfuture demands, becoming functionallyobsolete.The optimum level of calculated risk, hence thedesign return period for each of these periods maybe different. The final selection of the design floodcannot be made without considering political,social, environmental and other quantifiablecriteria.In many cases, flood analysis criteria are oftenprescribed by regulations and not subject to negotiation.Different types of projects may requiredifferent types of criteria reflecting economic efficiencyand safety. Safety criteria can be specified interms of a return period, meteorological input and/or the maximum flood on record. The returnperiod (T), in years, that is to be used is often specifiedby the competent agency and may be related tospecified risk (R) or probability of failure (percent) over the service life (n) (in years) as given byT = 1/[1 – (1–R) 1/n ] (see 5.10.8).For example, when n = 2 and the acceptable risk isR = 0.02 per cent, then T = 99.5 years. A distinctionshould be made between specifying the criteria thatis to be met and specifying the computationalmethod to be used to estimate the design flood.When the computational method is not specifiedby the regulation, it must be selected and justifiedby the designer. It is advisable to ensure theadequacy of the design against given conditionsand intent of the project.5.10.2.3 Design floods for large reservoirsThe selection of design floods for the spillwaydesign of large storage reservoirs must be givenspecial attention because a reservoir may considerablychange the flood regime, both at thereservoir site and in the downstream section ofthe river.The basic flood-related effect of a reservoir is floodattenuation. Its estimation requires knowledge ofthe original flood hydrograph shape. When thehydrograph is not known, a hypothetical shape,often triangular, is assumed and fitted to the selectedflood volume and peak discharge. In evaluating theeffect of flood attenuation on the reduction of spillwaycapacity and freeboard of a dam, it is imperativeto adopt a conservative approach and to consideronly those effects that can be guaranteed at alltimes. Thus, only the effect of the ungated spillwayshould be considered. All gated outlets should beassumed to be closed and the reservoir filled to thecrest of the fixed spillway at the beginning of theflood.In addition to flood attenuation, the flood regimedownstream must be analysed carefully from thepoint of view of changes in the timing of floodpeaks, the effect of changes in the shape of floodhydrographs and the effects on the river channelcaused by an increased scouring tendency of thevirtually sediment-free water leaving the reservoirthrough the spillway.The type of dam structure must also be consideredbecause it is of prime importance in determiningthe vulnerability of the dam should overtoppingoccur. Vulnerability is highest for earthfill dams,which are in great danger of collapsing ifovertopped.5.10.2.4 Probable maximum floodProbable maximum flood is computed from probablemaximum precipitation (see 5.7) or from themost critical combination of maximum snowmelt(see 6.3.4) and rainfall, and it provides an indicationof the maximum possible flood that couldreasonably be expected for a given watershed. It isnot possible to quantify the term reasonable orassign a long but arbitrary return period to theprobable maximum flood. The concepts of probablemaximum precipitation and probable maximumflood are controversial. Nevertheless, it is necessaryto assess the potential impact of such extremeevents; therefore, numerical flood estimates arerequired for very extreme floods and are often usedin design practice.Probable maximum precipitation is analyticallyestimated as being the greatest depth of precipitationfor a given duration that is physically plausibleover a given watershed at a certain time of theyear, and its estimation involves the temporaldistribution of rainfall. The concepts and relatedmethodologies are described by WMO (1986a).The US Army Corps of Engineers (1985) has acomputer program, HMRS2, to compute probablemaximum precipitation, which can then be usedwith HEC-1 (see 5.10.5) to determine probablemaximum flood. WMO (1969) provides more

CHAPTER 5. EXTREME VALUE ANALYSISII.5-47details on estimation of maximum floods (see6.3.2).As rainfall usually accounts for a major portion ofprobable maximum flood runoff, special considerationmust be given to the conversion of rainfallto runoff. This conversion is done by deterministicrainfall-runoff models, but their applicationfor this purpose involves certain modificationsdesigned to accommodate the extreme magnitudeof the rainfall event that is being used asinput. The most important modifications are asfollows:(a) The effect of the initial soil-moisture conditionsand of the variation of the infiltrationrate during the rainfall on streamflow is greatlyreduced, compared to their effect in streamflowsimulation under normal conditions.Hence, the refined methods employed in mostmodels for estimating infiltration indices canbe considerably simplified. A common practiceis to use the minimum infiltration capacity, orthe maximum runoff coefficient, for a givensoil type and vegetation cover, throughout theentire storm;(b) When a unit hydrograph is used to transformthe maximum precipitation, it should beremembered that the validity of the underlyingassumption of linearity is limited to conditionssimilar to those for which the unit hydrographwas derived. An analysis of floods in a numberof basins (Singh, 1992) has shown that the peakordinates of unit hydrographs derived frommajor floods (greater than 125 mm of runoffover the basin area) are often 25 to 50 per centhigher than peak ordinates derived from minorfloods (25 to 50 mm of runoff). It is importantto bear in mind that the adjustment of the unithydrograph for the computation of the probablemaximum flood must be guided by thenecessity of making a conservative estimate:one that leads to the greater flood;(c) In the case of drainage basins larger than500 km 2 , or even smaller basins where theirdifferent parts have widely different runoffcharacteristics, it is generally necessary toderive separate unit hydrographs and probablemaximum floods for several sub-areas and toobtain the probable maximum flood for thewhole basin by routing the component floodsdownstream to the project site. It must beremembered that the same positioning of theisohyetal pattern of the design storm over thecatchment, which yields the maximum flood ifa single unit hydrograph is used for the wholecatchment, need not yield the maximum floodif the catchment is subdivided into severalsub-areas. Thus, for each different catchmentsubdivision, an optimal positioning of thedesign storm, that is, the position yielding themost unfavourable combination of the relevantparameters of the probable maximum flood,must be found separately subject to the restrictionsdue to orography, as discussed in 5.7. Theoptimal position of the design storm can beobtained as a result of sensitivity analysis.Although no specific return period can beassigned to the probable maximum flood, itsparameters should be compared with the respectivefrequency curves fitted to historical floodsto make sure that they have extremely longreturn periods and have been unequalled by anyhistorical flood event.5.10.2.5 Standard project floodA standard project flood is usually about 50 per centof a probable maximum flood (Singh, 1992). Itsdetermination is governed by considerations similarto those relevant to the probable maximumflood. The standard project flood is usually determinedby the transformation of the transposedlargest rainstorm observed in the region surroundingthe project, rather than from a meteorologicallymaximized rainstorm, as in the case with the probablemaximum flood. Nonetheless, the standardproject flood should represent a very rare event andshould not be exceeded by more than a few per centby the major floods experienced within the generalregion.5.10.3 Data preparationBasic data for determining design floods are therecords collected by regional or nationalHydrological and Meteorological Services. Thesedata exist in the form of stage recordings anddischarge measurements that form the basis for thecomputation of rating curves. As the magnitude ofthe design flood depends primarily on measurementsof high discharges, special attention shouldbe given to their evaluation and the extension ofrating curves.For a proper assessment of the flood regime, it isessential to obtain sufficient information on historicfloods. The basic element of such information isstage. In compiling information on flood stages,use can be made of traces of materials deposited byfloods, flood marks on bridges, buildings and riverbanks; recollection of long-time residents; photographstaken during floods; archived materials;articles in the press and memoirs. Paleoflood

II.5-48GUIDE TO HYDROLOGICAL PRACTICESinformation can also be considered (Viessman andLewis, 2003).To convert flood stages determined by such investigationsinto discharges, hydraulic computationsmust be based on reconstructed river cross-sections,longitudinal profiles, the slope of water surface andchannel roughness. All the known modifications ofthe river channel should be taken into account,such as dredging, embankments and channelstraightening. Owing to the limited accuracy of thereconstructed river characteristics, the applicationof the Manning and Chézy formulae is generallysatisfactory for hydraulic computations of this kind.Software such as HEC-RAS can facilitate theanalysis.5.10.4 Design flood computationtechniquesThe selection of computational techniques for thedetermination of design floods depends on thetype, quantity, and quality of available hydrologicaldata, as well as the type of design floodinformation. Owing to the complexity of the floodproducing process, the estimates are only approximations,and understanding of related issues isimportant to produce reliable estimates. There aremany methods, and the choice is often made on asubjective and intuitive basis. Some practical criteriafor the choice of the method can be found inPilgrim and Doran (1993) and details of manymethods are available in Pilgrim and Cordery(1993), Bedient and Huber (2002) and Viessmanand Lewis (2003).Depending on data availability and design requirements,the methods of estimating design floods canbe grouped into empirical, frequency-based andrainfall-runoff methods.To extract maximum information from scarce orinaccurate data, it is advisable to apply severaldifferent methods, compare the results and choosethe design parameters based on engineering judgment.Sensitivity analysis can be useful in makingthe final decision because it may show the impactof potential errors on the magnitude of the designvariable.5.10.4.1 Empirical methodsEmpirical flood formulae expressed as a floodenvelope curve may be used to provide a roughestimate of the upper limit of discharge for agiven site. A common type of formula expressesthe peak discharge Q (m 3 s –1 ) as a power functionof catchment’s area A (km 2 ) (Bedient and Huber,2002),Q = CA n (5.53)where coefficient C and exponent n vary withinwide limits and the values for a particular study canbe selected on the basis of empirical data.The application of empirical formulae is generallylimited to the region for which they have beendeveloped, and they should be used with greatcaution and only when a more accurate methodcannot be applied. Another drawback of empiricalformulae is the difficulty in assessing the returnperiod of the computed peak flow.An envelope curve enclosing maximum observedpeak flows can be plotted against catchment areasfor a large number of stations in a meteorologicallyand geomorphologically homogeneousregion. Such curves provide useful information,especially in cases where few data are available atany single station. Attempts have been made torefine the technique by constructing variousenvelopes related to different climatological and/or geomorphologic factors. However, the returnperiods of the peak flows remain unspecified.Uses of such formulae provide a rough estimateproviding only the order of magnitude of largeflood flows.5.10.4.2 Rainfall-runoff modelsDepending on whether the design flood is to besynthesized from precipitation and/or snowmelt orfrom known flood hydrographs at upstream points,the models of interest fall into two broadcategories:(a) Rainfall-runoff models, as described in 6.3.2;(b) Streamflow routing models, as described in6.3.5.Many rainfall-runoff relationships have been developedthat could apply to any region or watershedunder any set of conditions. However, these methodsshould be used with caution, as they are onlyapproximate and empirical. The most widely usedpractical methods are the unit hydrograph method(see 6.3.2.3), the rational method (see below), theSoil Conservation Service (SCS) method (see below)and conceptual models (see 5.10.5).5.10.4.2.1 Rational methodOne of the oldest and simplest rainfall-runoffformulae is the rational formula, which allows for

CHAPTER 5. EXTREME VALUE ANALYSISII.5-49the prediction of peak flow Q p(m 3 s –1 ) from thefollowing equation:Q p= 0.278CiA (5.54)where C is the runoff coefficient that is dimensionlessand selected according to the type of land usein the watershed, i is rainfall intensity (mm/hr) ofchosen frequency and for duration equal to thetime of concentration, and A is the watershed area(km 2 ). This method is often used in small urbanareas as well as for rough estimates in rural areas inthe absence of data for other methods. It is highlysensitive to rainfall assumptions and the selectionof C. Use of this method should be restricted tosmall areas; although the upper limit is not explicitlyestablished, it varies between 40 ha and500 ha.Because of its predominant use in urban areas, therational method is dealt with in more detail in4.7.5.10.4.2.2 Soil Conservation Service methodThe former US Department of Agriculture SoilConservation Service, now the National ResourceConservation Service, suggested an empiricalmodel for rainfall abstractions based on the potentialfor the soil to absorb a certain amount ofmoisture. On the basis of field observations, thepotential storage S was related to a curve numberCN varying between 0 and 100, which is a characteristicof the soil type, land use and the initialdegree of saturation known as the antecedentmoisture condition (AMC). The value of S isdefined by the empirical expression:S = 25.4⎛⎝1000CN⎞−10⎠ (millimetres) (5.55)The values of CN are given in Table II.5.8 as a functionof soil type (A, B, C, D), land use, hydrologicalcondition of the watershed and antecedent moisturecondition (AMC I, II, III).According to this method, the depth of surfacerunoff is given by the following equation:Q =( P − I a ) 2( P − I a ) + S(5.56)where Q is the depth of surface runoff, P is theaccumulated depth of rainfall, I ais an initialabstraction: no runoff occurs until accumulatedrainfall exceeds I a, and S is the potential storage inthe soil.All units are in mm, and for values of P > I a. Usingobserved data, the Natural Resources ConservationService found that I ais related to S, and on averageis assumed to be I a= 0.2S; thus the equationbecomes:Q =( P − 0.2 S )2P + 0.8S(5.57)for P > 0.2S, and Q = 0 when P ≤ 0.2S. Since initialabstraction consists of interception, depressionstorage and infiltration prior to the onset of directrunoff, the value of I acan be modified to accountfor local conditions.Soils are classified as A, B, C, and D according to thefollowing criteria:(a) Group A soils have low runoff potential andhigh infiltration rates, greater than 7.6 mm/hr,and consist primarily of deep well-drainedsands and gravel;(b) Group B soils have moderate infiltration rates(3.8–7.6 mm/hr) and consist primarily ofmoderately fine to moderately coarse texturedsoils, such as loess and sandy loam;(c) Group C soils have low infiltration rates(1.27–3.8 mm/hr) and consist of clay loam,shallow sandy loam and clays;(d) Group D soils have high runoff potencial andlow infiltration rates (less than 1.27 mm/hr)and consist primarily of clays with high swellingpotential, soils with a permamnt high watertable or shallow soils over nearly imperviousmaterial.CN values for urban and composite areas should bedetermined.The runoff from a particular rainfall event dependson the moisture already in the soil from previousrainfall. The three antacedent moisture conditionsare as follows:(a) AMC I – Soils are dry but not to wilting point;(b) AMC II – Average conditions;(c) AMC III – Heavy rainfall or light rainfall withlow temperature have occured within the lastfive days saturating the soil.Table II.5.8 provides CN(II) values for average condtionsAMC II. CN(I) and CN(III) corresponding toAMC(I) and AMC(III) can be estimated from:CN(I) = 4.2CN(II)/(10 – 0.058CN(II) (5.58)andCN(III) = 23CN(II)/(10 + 0.13CN(II) (5.59)

II.5-50GUIDE TO HYDROLOGICAL PRACTICESTable II.5.8. Runoff curve numbers for selected agricultural, suburban and urban land use(AMCII, and I a= 0.25) (after Bedient and Huber, 2002)Hydrological soil groupLand-use description A B C DCultivated land aWithout conservation treatment 72 81 88 91With conservation treatment 62 71 78 81Pasture or rangelandPoor condition 68 79 86 89Good condition 39 61 74 80MeadowGood condition 30 58 71 78Wood or forest landThin stand, poor cover, no mulch 45 66 77 83Good cover b 25 55 70 77Open spaces: lawns, parks, golf courses and so forthGood condition: grass cover = 75% or more 39 61 74 80Fair condition: grass cover = 50–75% 49 69 79 84Commercial and business areas (85% impervious) 89 92 94 95Industrial districts (72% impervious) 81 88 91 93Residential cAverage lot sizeAverage % impervious d1/8 acre e or less 65 77 85 90 921/4 acre 38 61 75 83 871/3 acre 30 57 72 81 861/2 acre 25 54 70 80 851 acre 20 51 68 79 84Paved parking lots, roofs, driveways and so forth f 98 98 98 98Streets and roadsPaved with curbs and storm sewers f 98 98 98 98Gravel 76 85 89 91Dirt 72 82 87 89abcdefFor a more detailed description of agricultural land-use curve numbers, please refer to National Engineering Handbook (NaturalResources Conservation Service, 1972).Good cover is protected from grazing and litter and brush cover soil.Curve numbers are computed assuming that the runoff from the house and driveway is directed toward the street with a minimum ofroof water directed to lawns where additional infiltration could occur.The remaining pervious areas (lawns) are considered to be in good condition for these curve numbers.1 ha = 0.404687 acreIn some warmer climates of the country a curve number of 95 may be used.5.10.4.2.3 Soil Conservation Service unithydrographThe earliest Soil Conservation Service methodassumed that a hydrograph is a simple triangle, asshown in Figure II.5.12, with rainfall duration D(hours), time to peak TR (hours), time of fall B(hours) and the peak discharge Q p(m 3 s –1 ) given bythe following equation (Bedient and Huber, 2002):calculations (mm). Figure II.5.12 shows that thetime to peak (hours) is as follows:T R= D/2 + t p(5.61)Where D is the rainfall duration (in hours) and t pisthe lag time (in hours) from centroid of rainfall toQ p(m 3 s –1 ). Lag time t pis estimated from any one ofseveral empirical equations used by the SCS, such as:Q p =0.208 AQ RT R(5.60)t p = l 0.8 (S + 1) 0.71900y 0.5 (5.62)where A is the watershed area (km 2 ) and Q Rindicatesthe runoff depth for unit hydrographwhere l is the distance to the watershed divide (infeet), y is the average watershed slope (per cent) and

CHAPTER 5. EXTREME VALUE ANALYSISII.5-51allow for parameter verification in space and time,use of remotely sensed data and application ofgeographical information systems. Advancedcomputer-based technologies such as spreadsheets,databases and graphical capabilities facilitate theflexibility of data entry procedures.Figure II.5.12. SCS traiangular unit hydrograhS and CN are obtained from Table II.5.7. The basinlag (t p) is applicable to CN values between 50 and95, and watershed areas less then 8 km 2 . For urbanareas, t pshould be adjusted for imperviousness. Thecoefficient 0.208 in equation 5.60 is an averagevalue for many watersheds. It may be reduced byabout 30 per cent for flat or swampy watersheds, orincreased by about 20 per cent for steep basins.When such a change is introduced, then the unithydrograph must also be adjusted accordingly.Once Q pand t pare estimated, the unit hydrographcan be graphed and/or tabulated using the dimensionlessunit hydrograph shown in Table II.5.9.Muzik and Chang (2003) developed a regionaldimensionless hydrograph.The SCS method is widely used (Byczkowski, 1996;Maidment, 1993) because of its simplicity, readilyavailable watershed information, ease of application,and because it gives reasonable results.However, the results of studies comparingprediction with measured data have been mixed(Dingman, 2002) and so the method should be usedwith caution.5.10.5 Flood hydrograph conceptualmodelsRecent advances in computer technology and theoreticalhydrological developments haverevolutionized the manner in which computationsare now routinely performed. Hydrologic modelsSome of the more widely used models that havealso been developed include:(a) HEC-1 which was developed and is maintainedby the US Army Corps of Engineers HydrologicEngineering Center (www.hec.usace.army.mil).This model simulates the watershed as a seriesof hydraulic and hydrological components andcalculates runoff from single storms. The usercan select from a variety of sub-models thatsimulate precipitation, infiltration and runoff,as well as a variety of techniques to performthe flow routing. The model also includesdam safety and failure analysis, flood damageanalysis and parameter optimization. Morerecent improvements include considerationof radar rainfall as input and the use ofgeographical information system and mappingtools (HEC-GeoRAS) for handling output anddata manipulation;(b) SCS-TR 20 (for agricultural watersheds) and SCS-TR 55 (for urban watersheds) were developedand are maintained by the Natural ResourcesConservation Service, US Department of Agriculture.This combined model uses a curvenumber (CN) method to calculate the runoffhydrograph resulting from a single storm fromsub areas and routed through drainage systemsand reservoirs;(c) SWMM was developed and is maintained bythe US Environmental Agency (www.epa.gov/cdnnrmrl/models/swmm). This model consistsof a runoff module, a transport module and astorage/treatment module. It simulates runoffquantity and quality, routes sewer flows,computes hydraulic head and simulates theeffects of detentions basins and overflows. It isthe most comprehensive model for handlingurban runoff.There are certainly many other good models thatcan perform the same tasks. Model capabilitieschange rapidly and therefore it is advisable to seekTable II.5.9. Ordinates of the Natural Resources Conservation Service Dimensionless Unit Hydrographt/T R0 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.4 4.6 5.0Q/Q p0 0.310 0.930 0.930 0.560 0.280 0.147 0.077 0.029 0.003 0.000

II.5-52GUIDE TO HYDROLOGICAL PRACTICEScurrent information through the Websites of variousmodel developers. Links to other popularmodels are www.wallingfordsoftware.com, www.dhi.dk, http://water.usgs.gov/software/lists/surface_water and www.haested.com.All of the above models can be run on microcomputersand some are proprietary. Bedient and Huber(2002) provided a more comprehensive list of manyinternet sources to operational computer models,but many more have certainly become available inthe intervening years.5.10.6 Snowmelt contribution to floodIn some regions of the world, floods are caused by acombination of snowmelt and rainfall runoff orsnowmelt alone. Factors affecting the contributionof snowmelt to floods include accumulated snowpack depth at time of melt, ice jamming, basin storageand the return period of the event in question.Synthesis of runoff hydrographs associated withsnowmelt requires empirical equations, since snowmeltis not measured directly.After the depth of melt has been estimated, it canbe treated like rainfall and converted into streamflowby application of the unit hydrograph orrouting technique. Such a procedure does notprovide the probability of occurrence of a flood.Comparison of several snowmelt runoff models isdescribed by WMO (1986b). There are several operationalmodels that have a snowmelt routine,including HEC-1 (USACE, 1985).evaluation of the risks associated with the occurrenceof floods higher than the design flood.Knowledge of these risks is important because oftheir social, environmental and economic implications,for example in the determination offlood-insurance rates, flood-zoning policies or waterquality conservation. As floods are stochasticphenomena, their magnitude and the time of theirfuture occurrence cannot be predicted. The onlypossibility is to assess them on a probabilistic basis,that is, to assign a probability to the possibility thata flood of a given magnitude will be exceededwithin a specific period of time. A variable that hasa probability of exceedance p has a return periodT = 1/p.Guidance for general frequency analysis is providedin 5.3, and in 5.9 for flood frequency analysis. Acomprehensive risk assessment procedure for naturalhazards is provided in Comprehensive RiskAssessment for Natural Hazards (WMO/TD-No. 955).The probability of exceedance of a given magnitudeof event, as derived from a probability distributionmodel, pertains to each future event. Thus, if anannual flood series is considered, the exceedanceprobability p defines the risk that the given magnitudewill be exceeded in any one year. However, itis often necessary to calculate a probability p nthat agiven event, for example the exceedance of a particularflood peak, will occur at least once in n years,for example, during the design life of a project. Ifthe assumption of independence of floods in individualyears is satisfied, this probability is:5.10.7 Calculating discharges from urbandrainage systemsp n = 1 − (1 − p ) n = 1 −⎛1 − 1 ⎝ Tn⎞⎠(5.63)Urban hydrology is concerned mainly with theprediction of runoff peaks, volumes and completehydrographs anywhere in the system. The solutionto the above problems requires various analyticalmethods. Peak volumes can be obtained fromsimplified techniques such as the rational method(see 5.10.4.2.1), while hydrographs usually requiremore comprehensive analysis including the NaturalResources Conservation Service method (see5.10.4.2.2), or computer models (see 5.10.5). Urbandrainage is discussed in more detail in 4.7.5.10.8 RiskThe probability that the design flood will beexceeded at least once is known as the risk of failure,and the probability that the design flood willnot be exceeded is referred to as the reliability. Oneof the main concerns in design-flood synthesis is anwhere T is the return period. This measure of riskprovides a more probabilistic indication of thepotential failure of the design than that encapsulatedin the concept of return period. Note that therisk of an event occurring at least once during itsreturn period follows from equation 5.63 for n equalto T. When T is large, this risk approaches theasymptotic value:1 – e –1 = 0.63 (5.64)From equation 5.63, it is possible to express T as afunction of n and p n, that is, to calculate a returnperiod such that the risk of occurrence of the eventduring a period of n years will have a specified valuep n. This return period is called the design returnperiod T dand is as follows:T d= 1/[1 – (1 – p n) 1/n ] (5.65)

CHAPTER 5. EXTREME VALUE ANALYSISII.5-53Table II.5.10. Required design return period T dof an event whose risk of occurrence inn years is equal to p np n0.010.100.500.75n year2 10 50 100199.0 995.0 4975.0 9950.019.5 95.4 475.0 950.03.4 14.9 72.6 145.02.0 7.7 36.6 72.6Some values of the variables T d, n, and p nare shownin Table II.5.10. In order to illustrate its use,assume that the design life of a dam is 50 years andthat the designer wishes to take only a 10 per centrisk that the dam will be overtopped during itsdesign life. Thus n equals 50, p nequals 0.10, and thedam must be designed to withstand a flood that hasa return period T dof 475 years, which gives a probabilityof exceedance p = 1/Td ≈ 0.2 per cent.References and further readingAdamowski, K., 1981: Plotting formula for floodfrequency, Water Research Bulletin, Vol. 17, No. 2.Adamowski, K., 1985: Nonparametric kernel estimationof flood frequencies, Water Resources Research,21(11): 1585–1590.Adamowski, K., 1989: A Monte Carlo comparison ofparametric and nonparametric estimation of floodfrequencies, Journal of Hydrology, 108: 295–308.Adamowski, K., 1996: Nonparametric Estimation ofLow-Flow Frequencies, ASCE Journal of HydraulicEngineering, 122: 46–49.Adamowski, K., Y. Alila and P.J. Pilon, 1996: Regionalrainfall distribution for Canada, AtmosphericResearch, 42, 75–88.Akaike, H. 1974: A New Look at the Statistical ModelIdentification. IEEE transactions on automaticcontrol, AC-19(6): 716–723.American Society of Civil Engineers,1996: HydrologyHandbook, Second Edition, ASCE Manual andReports on Engineering Practice No. 28, New York,784 pp.Arihood, L.D. and D.R. Glatfelter, 1986: Method forestimating low-flow characteristics for ungaged streamsin Indiana, US Geological Survey Open-File Report86–323.Arihood, L.D. and D.R. Glatfelter, 1991: Method forestimating low-flow characteristics for ungaged streamsin Indiana, US Geological Survey Water-Supply Paper2372, 18 pp.Arnell, N.W., 2003: Relative effects of multi-decadalclimate variability and change in the mean andvariability of climate due to global warming: futurestreamflows in Britain, Journal of Hydrology 270,195–213.Arnell, N.W., C. Liy, R. Compagnucci, L. da Cunha,K. Hanaki, C. Howe, G. Mailu, I. Shikomanov,E. Stakhiv, Hydrology and Water Resources, In: J.J.McCarthy, O. Canziani, N.A. Leary, D.J. Dokkenand K.S. White (eds.), 2001: Climate Change2001: Impacts, Adaptations, and Vulnerability.Contribution of the Working Group II to the ThirdAssessment Report of the Intergovernmental Panelon Climate Change, Cambridge University Press,Cambridge, pp. 191–233.Arnell, V., P. Harremoes, M. Jensen, N.B. Johansen andJ. Niemczynowicz, 1984: Review of rainfall data applicationfor design and analysis,Water Science and Techn.,16, pp. 1–45.Asselin, J., T.B.M.J. Ouarda, V. Fortin and B. Bobée,1999: Une procedure Bayésienne bivariée pour détecterun décalage de la moyenne. INRS-Eau, Rapport deRecherche R-528, 35 pp.Bedient, P. B. and W. C. Huber, 2002: Hydrology andFloodplain Analysis, Prentice Hall, Inc. New Jersey,USA.Benson, M.A., 1962: Factors affecting the occoranceof floods in the south-west. US Geological SurveyWater-Supply Paper 1580-D, Reston, Virginia.Beran, M., J.R.M. Hosking and N. Arnell, 1986:Comment on “Two-Compent Extrreme ValueDistribution for Flood Frequency Analysis, WaterResources Research 220(2), 263–266.Bingham, R.H., 1982: Low-flow characteristics of Alabamastreams, US Geological Survey Water-Supply Paper2083, 27 pp.Bobée, B. and F. Ashkar, 1991: The gamma family andderived distributions applied in Hydrology. WaterResources Publications, Littleton, CO, 203 pp.Box, G.E.P. and G.C. Tiao, 1973: Bayesian inference instatistical analysis, Addison-Wesley, Reading, MA.Bras, R. and I. Rodriguez-Iturbe, 1985: Random functionsin hydrology. Addison-Wesley, Reading,Massachusetts.Burn, D.H., 1990: Evaluation of regional flood frequencyanalysis with a region of influence approach. WaterResources Research, 26(10): 2257–2265.Byczkowski, A, 1996: Hydrology, (in Polish),Wydawnictwo SGGW (Poland.)Carpenter, D.H. and D.C. Hayes, 1996: Low-flow characteristicsof streams in Maryland and Delaware, USGeological Survey Water-Resources InvestigationsReport 94–4020, 113.Carswell, W.J. and S.V. Bond, 1980: Multiyear low flowof streams in northeastern Kansas, US GeologicalSurvey Open-File Report 80–734, 26 pp.Cavadias, G.S., 1990: The canonical correlation approachto regional flood estimation. In M. A. Beran,M. Brilly, A. Becker and O. Bonacci (eds.), Proceedings

II.5-54GUIDE TO HYDROLOGICAL PRACTICESof the Ljubljana Symposium on Regionalizationin Hydrology, IAHS Publ. No. 191: 171–178.Wallingford, England.Chow, V.T., 1964: Handbook of Applied Hydrology.McGraw-Hill, New York.Chowdhury, J.U., J.R. Stedinger and L-H. Lu, 1991:Goodness-of-fit tests for regional GEV flooddistributions, Water Resources Research 27(7),1765–1776.Clarke, R.T., 1994: Statistical Modelling in Hydrology, JohnWiley and Sons, Chichester, England.Cluckie, I.D., P.P. Ede, M.D. Owens and A.C. Bailey,1987: Some Hydrological Aspects of Weather RadarResearch in the United Kingdom, HydrologicalSciences Journal, 32,3,9, pp. 329–346.Cluckie, I.D. and M.L. Pessoa 1990: Dam Safety: AnEvaluation of Some Procedures for Design FloodEstimation, Hydrological Sciences Journal, 35(5),pp. 547–569.Cohn, C.A., W.L. Lane and J.R. Stedinger, 2001:Confidence Intervals for EMA Flood QuantileEstimates, Water Resources Research, 37(6),1695–1706.Collier, C.G., 1993: Estimating Probable MaximumPrecipitation (PMP) Using a Storm Model andRemotely-sensed Data, British HydrologicalSociety 4th National Hydrology Symposium,Cardiff, September 13–16, Institute of Hydrology,pp. 5.33–5.38.Condie, R. and L.C. Cheng, 1982: Low Flow FrequencyAnalysis Program LOFLOW, Inland WatersDirectorate, Water Resources Branch, EnvironmentCanada.Condie, R. and G.A. Nix, 1975: Modelling of lowflow frequency distributions and parameter estimation,presented at the International Water ResourcesSymposium, Water for Arid Lands, Tehran, Iran,Dececmber 8–9.Court, A.,1961: Area-depth rainfall formulas. Journalof Geophysical Research, Vol. 65, No. 6, June,pp. 1823–1831.Cunnane, C., 1978: Unbiased plotting positions-Areview, Journal of Hydrology, 37 (3/4).Cunnane, C. 1988: Methods and merits of regionalflood frequency analysis. Journal of Hydrology, 100:269–290.Cunnane, C. 1989: Statistical Distributions for FloodFrequency Analysis, Operational Hydrology ReportNo. 33, World Meteorological OrganizationWMO-No. No. 718, Geneva, Switzerland.Dingman, S. L., 2002: Physical Hydrology, Prentice Hall,Inc. New Jersey, USA.Draper, N. and Smith, H., 1981: Applied RegressionAnalysis, Second Edition, John Wiley & Sons,Toronto.Durrans, S.R. and Tomic, S., 2001: Comparison ofParametric Tail Estimators for Low-Flow FrequencyAnalysis, Journal of the American Water ResourcesAssociation, 37(5), pp. 1203–1214.England, J.F. Jr., J.D. Salas and R.D. Jarrett, 2003:Comparisons of two moments-based estimatorsthat utilize historical and paleoflood data for thelog Pearson type III distribution, Water ResourcesResearch 39(9).Fennessey, N.M. and R.M. Vogel, 1990: Regional FlowDuration Curves for Ungaged Sites in Massachusetts,ASCE Journal of Water Resources Planning andManagement, 116(4), pp. 530–549.Fernandez Mills, G.,1995: Principal component analysisof precipitation and rainfall regionalization inSpain, Theoretical and Applied Climatology, 50,169–183.Fill, H., and J. Stedinger, 1995: L–moment and PPCCGoodness-of-Fit tests for the Gumbel Distributionand Effect of Autocorrelation, Water ResourcesResearch, 31(1), 225–29.Fill, H. and J. Stedinger, 1998: Using regional regressionwithin index flood procedures and an empiricalBayesian estimator, Journal of Hydrology, 210(1–4),128–145.Fortin, V., H. Perron, J. Bernier and B. Bobée, 1998:Determination des crues de conception, Rapport derecherché R-532, INRS-Eau, 103 pp.Francés, F., 1998: Using the TCEV distribution functionwith systematic and non-systematic datain a regional flood frequency analysis, StochasticHydrology and Hydraulics 12 (4), 267–283.Frederick, R.H., V.A. Meyers and E.P. Auciello, 1977:Five-to-60-minute precipitation frequency for theeastern and central United States, NOAA TechnicalMemorandum NWS HYDRO-35, National WeatherService, US Dept.of Commerce, Washington, D.C.Gabriele, S. and N. Arnell, 1991: A Hierachical Approachto Regional Flood Frequency Analysis, WaterResources Research, 27(6), 1281–1289.Gabriele, S. and P. Villani, 2002: The estimation of rainfallfrequency in a Mediterranean environment dueto extraordinary combinations of hydrological andclimatic conditions, The Extremes of the Extremes:Extraordinary Floods (Proceedings of a symposiumheld at Reykjavik, Iceland, July 2000). IAHS Publ.No. 271, pp. 45–50.Gingras, D., M. Alvo and K. Adamowski, 1995: Regionalflood relationship by nonparametric regression,Nordic Hydrology, 26, 73–90.Gray, D.M., ed., 1973: Handbook on the Principles ofHydrology, National Research Council ofCanada, Water Information Center Inc.,Syosset, N.Y..Greenwood, J.A., J.M. Landwehr, N.C., Matalas andJ.R. Wallis, 1979: Probability weighted moments:definition and relation to parameters of severaldistributions expressible in inverse form. WaterResources Research, Vol. 15, No. 5, pp. 1049–1054.

CHAPTER 5. EXTREME VALUE ANALYSISII.5-55GREHYS, 1996a: Presentation and review of some methodsfor regional flood frequency analysis. Journal ofHydrology, 186(1–4): 63–84.GREHYS, 1996b: Intercomparison of flood frequencyprocedures for Canadian rivers. Journal of Hydrology,186(1–4): 85–103. F.E. Grubbs and G. Beck (1972).Extension of sample sizes and percentage pointsfor significance tests of outlying observations.Technometrics, 14(4): 847–854.Griffis, V.W., J.R. Stedinger and T.A. Cohn, 2004: LP3Quantile Estimators with Regional Skew Informationand Low Outlier Adjustments, accepted to WaterResources Research.Grygier, J.C., J.R. Stedinger and H.B. Yin, 1989: Ageneralized move procedure for extendingcorrelated series, Water Resources Research, 25(3),345–350.Guillot, P., 1993: The Arguments of the Gradex Method:a Logical Support to Assess Extreme Floods.Extreme Hydrological Events: Precipitation, Floodsand Droughts. Proceedings of the YokohamaSymposium, IAHS Publ. No. 213.Guo, S.L., 1991: Nonparametric variable kernel estimationwith historical floods and paleofloodinformation, Water Resources Research, 27(1): 91–98.Guo, S.L., R.K. Kachroo and R.J. Mngodo, 1996:Nonparametric Kernal Estimation of Low FlowQuantiles, Journal of Hydrology, 185, pp. 335–348.Gumbel, E.J., 1958: Statistics of extremes, ColumbiaUniversity Press, NY, 375 pp.Gustard, A. and K.M. Irving, 1994: Classification of theLow Flow Response of European Soils, FRIEND:Flow Regimes from International Experimentaland Network Data, IAHS Publication No. 221,pp. 113–117.Hansen, E.M., L.C. Schreiner J.F. and Miller, 1982:Application of probable maximum precipitationestimates: United States east of the 105th meridian.Hydrometeorological Report, No. 52, US NationalWeather Service.Helsel, D.R., and R.M. Hirsch, 1992: Statistical Methods inWater Resources, Elsevier, NY.Herbst, P. H., K.B. Bredenkamp and H.M.G. Barker, 1966:A technique for the evaluation of drought fromrainfall data. Journal of Hydrology, Vol. 4, No. 3,pp. 264–272.Hershfield, D.M., 1961: Rainfall frequency atlas ofthe United States for durations from 30 minutesto 24-hours and return periods from 2 to 100years, Technical Paper 40, US Weather Bureau,Washington, D.C.Hershfield, D.M., 1965: Method for estimating probablemaximum rainfall. Journal of the AmericanWaterworks Association, Vol. 57, August, pp. 965–972.Hirsch, R.M., 1982: A comparison of four record extensiontechniques, Water Resources Research, 18(4),1081–1088.Hirsch, R. M, and J. R. Stedinger,1987: Plotting positionsfor historical floods and their precision, WaterResources Research, 23(4), 715–727.Hirschboeck, K.K., L.L. Ely and R.A. Maddox, 2000:Hydroclimatology of Meteorologic Floods, Chapter2, Inland Flood Hazards: Human, Riparian andAquatic Communities, E.E. Wohl (ed.), CambridgeUniversity Press, Stanford, United Kingdom.Hogg, W.D. and D.A. Carr, 1985: Rainfall frequency atlasfor Canada, Canadian Government PublishingCentre, Ottawa, Canada.Hosking, J.R.M., 1990: L–moments: Analysis and estimationof distributions using linear combinations oforder statistics, Journal of Royal Statistical Society, B,52(2), 105–124.Hosking, J.R.M. and J.R. Wallis, 1987: Parameter andquantile estimation for the generalized Pareto distribution,Technometrics, 29(3):339–349.Hosking, J.R.M. and Wallis, J.R., 1988, The effect ofintersite dependence on regional flood frequencyanalysis. Water Resources Research, Vol. 24, No. 4,pp. 588–600.Hosking, J.R.M. and J.R. Wallis, 1995: A comparison ofunbiased and plotting-position estimators ofL moments, Water Resources Research, 31(8),2019–25.Hosking, J.R.M. and J.R. Wallis, 1997: RegionalFrequency Analysis: An Approach Based on L–moments, Cambridge University Press.Hosking, J.R.M., J.R. Wallis and E.F. Wood, 1985:Estimation of the generalized extreme-valuedistribution by the method of probability weightedmoments, Technometrics 27(3), 251–261.Huff, F.A. and S.A., Jr. Changnon, 1964: A model10-inch rainstorm, Journal of Applied Meteorology,pp. 587–599.Institute of Hydrology, 1980: Low Flow Studies, ResearchReport, Wallingford, Oxon, UK.Institute of Hydrology, 1999: Flood Estimation Handbook,CEH, Wallingford, UK.Interagency Advisory Committee on Water Data, 1982:Guidelines for Determining Flood Flow Frequency,Bulletin #17B, US Department of the Interior,US Geological Survey, Office of Water DataCoordination, Reston Virginia.Jain, S. and U. Lall, 2001: Floods in a changing climate:Does the past represent the future?, Water ResourcesResearch 37(12), 3193–3205.Javelle P., 2001: Caractérisation du régime des crues: lemodèle débit-durée-fréquence convergent. Approchelocale et régionale, PhD thesis.Cemagref-Lyon.Institut National Polytechnique de Grenoble.268 pp. can be downloaded at: http://www.lyon.cemagref.fr/doc/these/javelle.Javelle, P., T.B.M.J. Ouarda and B. Bobée, 2003: Floodregime definition using the flood-durationfrequencyapproach: Application to the Provinces of

II.5-56GUIDE TO HYDROLOGICAL PRACTICESQuebec and Ontario, Canada. Hydrological Processes.17(18): 3717–3736.Javelle, P., T.B.M.J. Ouarda, M. Lang, B. Bobée, G. Galeaand J.-M. Gresillon, 2002: Development of regionalflow-duration-frequency curves based on theindex-flood method, Journal of Hydrology. 258:249–259.Kite, G.W., 1988: Frequency and Risk Analysis inHydrology, Water Resources Publication, Littleton,CO.Kottegoda, M. and R. Rosso, 1997: Statistics Probability,and Reliability for Civil and Environmental Engineers,McGraw-Hill, New York.Kroll, C.N. and J.R. Stedinger, 1996: Estimation ofMoments and Quantiles Using Censored Data, WaterResources Research, 32(4), pp. 1005–1012.Kuczera, G., 1982: Combining site-specific and regionalinformation: an empirical Bayes Approach. WaterResources Research, 18(2):306–314.Kundziewicz Z.W., D. Rosbjerg, S.P. Simonowicz andK. Takeuchi, 1993: Extreme Hydrological Events:Precipitation, Floods and Droughts, IAHS Press,Institute of Hydrology, UK, Publication No. 213.Landwehr, J.M., N.C. Matalas and J.R. Wallis, 1979:Probability weighted moments compared with sometraditional techniques in estimating Gumbel parametersand quantiles, Water Resources Research, 15(5),1055–-1064.Lang, M., Ouarda, T.B.M.J. and B. Bobée, 1999: Towardsoperational guidelines for over-threshold modeling.Journal of Hydrology, 225: 103–117.Lawal, S.A. and W.E. Watt, 1996: Non-Zero Lower Limitin Low Flow Frequency Analysis?, Water ResourcesBulletin, 32(6), pp. 1159–1166.Lettenmaier, D.P. and K.W. Potter, 1985: Testing floodfrequency estimation methods using a regionalflood generation model. Water Resources Research,Vol. 21, No. 12, pp. 1903–1914.Lettenmaier, D.P., J.R. Wallis and E.F. Wood, 1987: Effectof regional hereogeneity on flood frequency estimation,Water Resources Research, 23(2), 313–324.Madsen, H., and D. Rosbjerg, 1997, Generalized leastsquares and empirical Bayes estimation in regionalpartial duration series index-flood modeling, WaterResources Research, 33(4), 771–782.Maidment, D. R. (editor in chief), 1993: Handbook ofHydrology, McGraw-Hill, Inc., New York, USA.Mann, H.B. and D.R. Whitney, 1947: On the test ofwhether one of two random variables is stochasticallylarger than the other. Annual MathematicalStatistics, 18: 50–60.Martins, E.S. and J.R. Stedinger, 2000: GeneralizedMaximum Likelihood Pareto-Poisson Flood RiskAnalysis for Partial Duration Series, Water ResourcesResearch 37(10), 2559-2567, 2001.Martins, E.S. and J.R. Stedinger, 2001: Historical informationin a GMLE-GEV Framework with PartialDuration and Annual Maximum Series, WaterResources Research, 37(10), 2551–2557.Matalas, N.C. and B. Jacobs, 1964: A correlation procedurefor augmenting hydrologic data, US Geol. Prof. Paper,434-E, E1-E7.Matalas, N., 1963: Probability Distributions of Low Flows,Professional Paper 434-A, USGS, Washington D.C.Miller, J.F. 1972: Physiographically Adjusted PrecipitationfrequencyMaps: Distribution of Precipitation inMountainous Areas. WMO-No. 326 (11):264–277.Miller, J.R., R.H. Frederick and R.J. Tracey, 1973:Precipitation frequency atlas of the western UnitedStates, NOAA Atlas 2, National Weather Service,NOAA, US Department of Commerce, Silver Spring,MD.Moore, R., 1993: Applications of Weather Radar Datato Hydrology and Water Resources, WMO, RegionalAssociation VI, (Europe), Working Group onHydrology, 26 pp.Morlat G., 1956: Les lois de probabilité de Halphen. Revuede statistique appliqué, 3: 21-43. (http://archive.numdam.org/ARCHIVE/RSA/RSA_1956__4_3/RSA_1956__4_3_21_0/RSA_1956__4_3_21_0.pdf).Muirhead, R.J., 1982: Aspect of Multivariate StatisticalTheory. J. Wiley, 673 pp.Muzik,I. and C. Chang, 2003: Regional dimensionlesshydrograph for Alberta foothills, HydrologicalProcesses, 17, 3737–3747.Nathan, R.J. and T.A. McMahon, 1990: Practical Aspectsof Low-Flow Frequency Analysis, Water ResourcesResearch, 26(9), pp. 2135–2141.Nathan, R.J. and T.A. McMahon, 1992: Estimating LowFlow Characteristics in Ungauged Catchments,Water Resources Management, 6, pp. 85–100.National Research Council, 1988: Estimating probabilitiesof extreme floods, National Academy Press,Washington, D.C., 141 pp.National Research Council of Canada (NRCC), 1989:Hydrology of floods in Canada: a guide to planning anddesign, W.E.Watt editor, Ottawa, 245 pp.Natural Environment Research Council, 1975: FloodStudies Report, Volume II: Meteorological Studies,London, England, 81 pp.NERC, 1975: Flood Studies Report, Vol. 1, HydrologicalStudies, Natural Environment Research Council, 27Charing Cross Road, London.Nguyen,V-T-V., N. In-na and B. Bobee, 1989: A newplotting position formula for Pearson type III distribution,ASCE Journal of Hydraulic Engineering, 115(6),pp. 709–730.Nguyen, V-T-V., T-D. Nguyen and F. Aaskar, 2002:Regional Frequency Analysis of Extreme Rainfalls,Water Science and Technology, Vol. 45, No. 2,pp. 75–81.Nguyen,V-T-V., J. Rousselle and M.B. McPherson, 1981:Evaluation of areal versus point rainfall with sparse

CHAPTER 5. EXTREME VALUE ANALYSISII.5-57data, Canadian Journal of Civil Engineering, 8(2),pp. 173–178.Niemczynowicz, J. 1982: Areal intensity-durationfrequencycurves for short term rainfall events,Nordic Hydrology, 13(4), pp. 193–204.NRC, 1998: Decade-to-Century-Scale Climate Variabilityand Change: A Science Strategy. National ResourcesCouncil, Panel on Climate Variability on Decadeto-CenturyTime Scales, Washington, DC: NationalAcademy Press.NRCS, 1972: National Engineering Handbook, Section 4,“Hydrology,” Chapter 9.Olsen, J.R., J.R. Stedinger, N.C. Matalas and E.Z. Stakhiv,1999. Climate variability and flood frequencyestimation for the Upper Mississippi and LowerMissouri Rivers, Journal of the American WaterResources Association, 35(6), 1509–1524.Ouarda, T.B.M.J., G. Boucher, P.F. Rasmussen andB. Bobée, 1997: Regionalization of floods by canonicalanalysis. In “Operational Water Management”, J.C.Refsgaard and E.A Karalis Eds., A.A. Balkema Pub.Rotterdam, 297–302.Ouarda, T.B.M.J., M. Haché and B. Bobée, 1998:Régionalisation des événements hydrologiques extrêmes,Research Report No. R-534, INRS-Eau, University ofQuebec, Ste-Foy, Canada, 181 pp.Ouarda, T.B.M.J., M. Lang, B. Bobée, J. Bernier, P. Bois,1999: Analysis of regional flood models utilizedin France and Quebec, Review Science Eau, 12(1):155–182.Ouarda, T.B.M.J., M. Haché, P. Bruneau and B. Bobée,2000: Regional flood peak and volume estimationin northern Canadian basin. Journal of Cold RegionsEngineer, ASCE. 14(4): 176–191.Ouarda, T.B.M.J., C. Girard, G.S. Cavadias and B. Bobée.2001: Regional flood frequency estimation withcanonical correlation analysis. Journal of Hydrology254: 157–173.Ozga-Zielinska, M.J. Brzezinski and B. Ozga-Zielinski,1999: Calculation of Design Flood for HydraulicStructures, Institute of Meteorology and WaterResources, No. 27 (in Polish).Ozga-Zielinski B., 2002: The Gradex-KC and Gradex-ZN Methods for Estimation of Maximum Floodswith T-year Return Period for Cases with Shortor Nonexisting Flow Observations. Monographsof Water Management Committee of the PolishAcademy of Sciences, No.21 (in Polish).Pandey,G. R. and V.T.V. Nguyen. 1999: A comparativestudy of regression based methods in regional floodfrequency analysis. Journal of Hydrology. 225: 92–101.Paulson, R.W., E.B. Chase, R.S. Roberts and D.W.Moddy, 1991: National Water Summary 1988–89– Hydrologic Events and Floods and Droughts, USGeological Survey Water.Pekarova, P., P. Miklanek and J. Peklar, 2003: Spatial andtemporal runoff oscillation analysis of the mainrivers of the world during the 19th and 20th centuries,Journal of Hydrology, 274, 62–79.Perreault L., B. Bobée and P.F. Rasmussen, 1999a:Halphen distribution system, I: Mathematical andstatistical properties. Journal of Hydrologic Engineering,ASCE, 4(3): 189–199.Perreault L., B. Bobée and P.F. Rasmussen, 1999b:Halphen distribution system, II: Parameter andquantile estimation. Journal of Hydrologic Engineering,ASCE, 4(3): 200–208.Perreault, L., M. Haché, M. Slivitzky and B. Bobée,1999c: Detection of changes in precipitation andrunoff over eastern Canada and US using a Bayesianapproach, Stoch. Env. Resear. And Risk Assess. 13:201–216.Perreault, L, J. Bernier, B. Bobée and E. Parent, 2000:Bayesian change-point analysis in hydrometeorologicaltime series, Part 1, The normal model revisited,Journal of Hydrology. 235: 221–241.Piechota, T.C. and J. A. Dracup, 1999: Long-rangestreamflow forecasting using El Nino-SouthernOscillation indicators, Journal of HydrologicEngineering, ASCE, 4(2), 144–151.Pilgrim D.M. and I. Cordery, 1993: Flood Runoff, inHandbook of Hydrology, editor Maidment D. R.,McGraw-Hill, Inc., New York, USA.Pilgrim, D.H. and D.G. Doran, 1993: Practical Criteriafor the Choice of Method for Estimating ExtremeDesign Floods, IAHS Publication No. 213, Instituteof Hydrology, Wallingford, UK.Pilgrim, D. and Cordery, I., 1975: Rainfall temporalpatterns for design floods, ASCE Journal of HydraulicEngineering, 101 (HY1), pp. 81–95.Pilgrim, D.H., ed., 1998: Australian Rainfall and Runoff:A Guide to Flood Estimation, Volumes I and II,Institution of Engineers Australia, Canberra.Pilon, P.J. and K.D. Harvey, 1992: The ConsolidatedFrequency Analysis (CFA) Package. (HOMS componentI.81.2.02), Environment Canada, WaterResources Branch, Ottawa.Pilon, P.J. and S.Yue, 2002: Detecting climate-relatedtrends in streamflow data, Water Science andTechnology, 45(8), 89–104.Porparto, A. and L. Ridolfi, 1998: Influence of weaktends on exceedance probability, Stochastic Hydrologyand Hydraulics 12(1), 1–15.Reilly, C.F. and C.N. Kroll, 2003: Estimation of 7-day,10-year low-streamflow statistics using baseflowcorrelation, Water Resources Research, 39(9),1236.Ries, K.G., 1994: Development and application ofGeneralized least squares regression models to estimatelow-flow duration discharges in Massachusetts,US Geological Survey Water-Resources InvestigationsReport 94–4155, 33 pp.Robson, A. and D. Reed, 1999: Flood EstimationHandbook, Volume 3, Statistical procedures for

II.5-58GUIDE TO HYDROLOGICAL PRACTICESflood frequency estimation, Institute of Hydrology,Wallingford, Oxfordshire, United Kingdom.Rossi, F., M. Fiorentino and P. Versace, 1984: Two-Component Extreme Value Distribution for FloodFrequency Analysis, Water Resources Research 20(7),847–856.Salas, J.D., E.E. Wold and R.D. Jarrett,1994:Determination of flood characteristics using systematic,historical and paleoflood data, in G. Rossi et al.,Coping with Floods, pp. 111–134, Kluwer AcademicPublishers, Netherlands.Schaefer, M.G., 1990: Regional analysis of precipitationannual maxima in Washington State, WaterResources Research, 26(1), pp. 119–131.Schwarz, G., 1978: Estimating the dimension of a model.The Annals of Statistics, 6: 461–464.Searcy, J.K., 1959: Flow-Duration Curves, US GeologicalSurvey Water Supply Paper 1542–A.Sevruk, B., 1985: Correction of precipitation measurements:summary report, Workshop on thecorrection of precipitation measurements, April 1–3,1985, Zurich, Switzerland, pp. 13–23Sherwood, J.M., 1994: Estimation of volume-durationfrequencyrelations of ungauged small urbanstreams in Ohio. Water Resources Bulletin 302:261–269.Singh, V. P., 1992: Elementary Hydrology, Prentice-HallInc., Englewood Cliffs, NJ 07632.Smakhtin,V.U., 2001: Low Flow Hydrology: A Review,Journal of Hydrology, 240, pp. 147–186.Smith, J.A. 1993: Precipitation, in Handbook of Hydrology,D.R. Maidment, ed., McGraw-Hill, New York,3.1–3.47.Stedinger, J.R. and V.R. Baker, 1987: Surface waterhydrology: historical and paleoflood information,US National report to international Union ofGeodesy and Geophysics, 1983–1987, Reviews ofGyophsysical and Space Physics, 25(2), 119–124.Stedinger J.R., R.M. Vogel and E. Foufoula-Georgiou,1993: Frequency analysis of extreme events.Chapter 18 (1–66) in: Handbook of Applied Hydrology;D.R. Maidment (Ed.) McGraw-Hill, NY.Stedinger, J.R. and L. Lu, 1995: Appraisal of Regional andIndex Flood Quantile Estimators, Stochastic Hydrologyand Hydraulics, 9(1), 49–75.Stedinger,J.R. and T.A. Cohn, 1986, Flood frequencyanalysis with historical and paleoflood information,Water Resources Research, 22(5):785–793.Stedinger, J.R. and G.D. Tasker, 1985: RegionalHydrologic Analysis 1. Ordinary, Weighted, andGeneralized Least Squares Compared, WaterResources Research, 21(9), pp. 1421–1432.Stedinger, J.R., 1980: Fitting Log Normal Distributionsto Hydrologic Data, Water Resources Research, 16(3),481–490.Stedinger, J.R., G. Tasker, 1986: Regional hydrologicalanalysis, 2, model-error estimators, estimation ofsigma and log Pearson type 3 distributions. WaterResources Research, 22(10):1487–1499.Stedinger, J.R. and W.O. Thomas, Jr., 1985: Low-flow estimationusing base-flow measurements, US GeologicalSurvey Open-File Report 85–95, 22 pp.Task Committee of the Hydraulics Division, 1980:Characteristics of Low Flows, ASCE Journal of theHydraulics Division, 106 (HY5), pp. 717–737.Tasker, G.D., 1980: Hydrologic Regression with WeightedLeast Squares, Water Resources Research, 16(6),pp. 1107–1113.Tasker, G.D., 1987: A Comparison of Methods forEstimating Low Flow Characteristics of Streams,Water Resources Bulletin, 23(6), pp. 1077–1083.Tasker, G.D., 1991: Estimating low-flow characteristicsof streams at ungaged sites, US GeologicalSurvey Open-File Report 91–244, Proceedings ofthe United States – People’s Republic of ChinaBilateral Symposium on Droughts and Arid-RegionHydrology, September 16–20, 1991, Tucson, Arizona,pp. 215–219.Terry, M.E. 1952: Some rank order tests which are mostpowerful against specific parametric alternatives.Ann. Math. Statist., 23 : 346–366.Thomas, W.O., 1985: A uniform technique for floodfrequency analysis, Journal of Water Resources.Planning and Management, 111(3), 321–337.Thomas, D.M. and M.A. Benson, 1970: Generalizationof Streamflow Characteristics from Drainage-BasinCharacteristics, US Geological Survey Water-SupplyPaper 1975, Reston, Virginia.Thomas, W.O., Jr. and J.R. Stedinger, 1991: Estimatinglow-flow characteristics at gaging stations andthrough the use of base-flow measurements,US Geological Survey Open-File Report 91-244,pp. 197–206.US Army Corps of Engineers, 1985: HEC-1 FloodHydrograph Package, Hydrologic Engineering Center,Davis, California.Viessman, W. and G.L.Lewis, 2003: Introduction toHydrology, Prentice Hall, Pearson Education, Inc.,Upper Saddle River, NJ 07458, USA.Vogel, R. and J.R. Stedinger, 1984: Flood plain delineationin ice jam prone regions, Journal of theWater Resources Planning and Management, ASCE,110(WR2), 206–219.Vogel, R.M. and N.M. Fennessey, 1994: Flow DurationCurves I: A New Interpretation and ConfidenceIntervals, ASCE Journal of Water Resources Planningand Management, 120(4), pp. 485–504.Vogel, R.M. and N.M. Fennessey, 1995: Flow DurationCurves II: A Review of Applications in WaterResources Planning, Water Resources Bulletin, 31(6),pp. 1029–1039.Vogel, R.M. and C.N. Kroll, 1989: Low-FlowFrequency Analysis Using Probability-PlotCorrelation Coefficients, ASCE Journal of Water

CHAPTER 5. EXTREME VALUE ANALYSISII.5-59Resources Planning and Management, 115(3),pp. 338–357.Vogel, R.M. and C.N. Kroll, 1990: Generalized Low-Flow Frequency Relationships for Ungaged Sitesin Massachusetts, Water Resources Bulletin, 26(2),pp. 241–253.Vogel, R.M. and C.N. Kroll, 1992: RegionalGeohydrologic-Geomorphic Relationships for theEstimation of Low-Flow Statistics, Water ResourcesResearch, 28(9), pp. 2451–2458.Vogel, R.M., and J.R. Stedinger, 1985: Minimum variancestreamflow record augmentation procedures, WaterResources Research, 21(5), 715–723.Vogel, R.M., and N.M. Fennessey, 1993. L–momentdiagrams should replace product moment diagrams,Water Resources Research, 29(6), 1745–52.Wald, A. and J. Wolfowitz, 1943: An exact test forrandomness in the non-parametric case based onserial correlation. Annual Mathematical Statistics, 14:378–388.Waltemeyer, S.D., 2002: Analysis of the Magnitudeand Frequency of the 4-Day Annual Low Flow andRegression equations for Estimating the 4-Day,3-Year Low-Flow Frequency at Ungaged Sites onUnregulated Streams in New Mexico, US GeologicalSurvey, Water-Resources Investigations Report01–4271, Albuquerque, New Mexico.Wang, Q.J., 1991: The POT model described by thegeneralized Pareto distribution with Poisson arrivalrate, Journal of Hydrology, 129, pp. 263–280.Wang, Q.J., 1996: Direct sample estimators of L–moments, Water Resources Research, 32(12),3617–19.Waylen, P. and M-K. Woo, 1982: Prediction of annualfloods generated by mixed processes, Water ResourcesResearch, 18(4), 1283–86.Waylen, P.R. and M.-K. Woo, 1987: Annual Low FlowsGenerated by Mixed Processes, Hydrological SciencesJournal, 32(3), pp. 371–383.Weisberg, S., 1980: Applied Linear Regression, John Wiley& Sons, TorontoWeisner, C.J., 1970: Hydrometeorology, Chapman & Hall,London.Wenzel, H.G., 1982: Rainfall for urban stormwaterdesign, in Urban Storm Water Hydrology, D.F. Kibler,ed., Water Resources Monograph No. 7, AmericanGeophysical Union, Washington, D.C.Wilks, D.S. 1993: Comparison of the three-parameterprobability distributions for representing annualextreme and partial duration precipitation series,Water Resources Research, 29(10), pp. 3543–549.Wohl, E.E. (ed.), 2000: Inland flood hazards: human, riparianand aquatic communities, Cambridge UniversityPress, Stanford, UK.World Meteorological Organization, 1969a: Estimation ofMaximum Floods, WMO No. 233, TP 126, TechnicalNote No. 98, Geneva, Switzerland.———, 1969b: Manual for Depth-area-duration Analysis ofStorm Precipitation, WMO-No. 237, Geneva.———, 1986a: Manual for Estimation of ProbableMaximum precipitation, Second Edition, OperationalHydrology Report No. 1, WMO-No. 332, Geneva,269 pp.———, 1986b: Intercomparison of models of snowmeltrunoff, WMO-No.646, Operational Hydrology Rep.No. 23, Geneva.———, 1981: Selection of Distribution Types forExtremesof Precipitation (B. Sevruk and H. Geiger).Operational Hydrology Report No. 15, WMO-No. 560,Geneva.———, 1999: Comprehensive Risk Assessment for NaturalHazards, WMO/TD-No. 955.WRC, 1967: Guidelines for Determining flood flowfrequency. US Water Resources Council HydrologyCommittee, Washington, D.C. Revised versions:Bulletin 15(1975) – 17(1976) – 17 A (1977) – 17B(1981).Yue, S., P. Pilon and G. Cavadias, 2002a: Power ofthe Mann-Kendall and Spearman’s rho tests fordetecting monotonic trends in hydrological series,Journal of Hydrology, 259, 254–271.Yue, S., P. Pilon, B. Phinney and G.Cavadias, 2002b: Theinfluence of autocorrelation on the ability to detecttrend in hydrological series, Hydrological Processes,16, 1807–1829.Yue, S. and P. Pilon, 2004: A comparison of the power ofthe t test, Mann-Kendall and bootstraptests for trend detection, Hydrological Sciences, 49(1),21–37.Zaidman, M.D., V. Keller, A.R. Young and D. Cadman,2003: Flow-Duration Frequency Behaviour of BritishRivers Based on Annual Minima Data, Journal ofHydrology, 277, pp. 195–213.Zalina, M.D., Desa, M.N.M., Nguyen, V-T-V., and Amir,K. 2002, Statistical Analysis of Extreme RainfallProcesses in Malaysia, Water Science and Technology,Vol. 45, No. 2, pp. 63–68.

CHAPTER 6MODELLING OF HYDROLOGICAL SYSTEMSThe general term modelling means replacing anobject under consideration by a quasi-object, ormodel, in order to draw information about theobject from the model. The model imitates ormimics selected aspects of the object of interest,which are deemed important in the study at hand.A model can be seen as a working analogy of thereal object, allowing similarity, though not identity,of properties, which are important in the particularstatement of the problem. The basic rationale ofmodelling is the possibility to simulate and predictthe behaviour of a complex object or system, withthe help of a simpler, and/or more tractable model.Details of the real object can be ignored becausethey are not important in a particular case orbecause they are too complex, hence not tractable(see Dooge, 1973).Many ways of classifying models have beenproposed, starting from the early distinction ofintuitive and formalized models. Formalized modelscan be divided into material models and symbolicmodels. The class of material models, the representationof a real system by another real system, canbe divided into physical models, also called iconic,or look-alike models, such as hydraulic laboratorymodels of a dam or a channel built to an appropriatescale, and analogue models, such as electricalanalogs. Material models have similar properties tothe object under consideration and are easier andcheaper to study. Experiments on material modelscan be made under more favourable and observableconditions (Singh, 1988), while experiments on theobject may be difficult or even impossible. Symbolicmodels can be classified into verbal, graphical andmathematical models. Nowadays, mathematicalmodels are by far the most commonly used, mainlybecause of the computational capabilities offeredby affordable computers.The notion of the mathematical modelling ofhydrological systems can be understood in a verybroad sense as the use of mathematics to describefeatures of hydrological systems or processes.Hence, every use of a mathematical equation torepresent links between hydrological variables, orto mimic a temporal or spatial structure of a singlevariable, can be called mathematical modelling.Under such a broad definition of the term, thereare multiple links to many chapters of the Guide,as every hydrological process can be described viamathematical formalisms. The term mathematicalmodelling of hydrological systems includes timeseriesanalysis and stochastic modelling, where theemphasis is on reproducing the statistical characteristicsof a hydrological times series of ahydrological variable.Developments in the modelling of hydrologicalsystems have been linked closely with the emergenceand progress of electronic computers,user-friendly operation systems, application softwareand data-acquisition techniques. Theubiquitous availability of computers and the developmentof associated numerical methods haveenabled hydrologists to carry out complex, repetitivecalculations that use large quantities of data.Streamflow modelling has become an importantelement in the planning and management of watersupplyand control systems and in providingriver-forecast and warning services. The nature ofmodelling and the forced reliance on computerprogramming makes it impractical to includecomputational aids in this Guide. References arecited for further guidance on specific aspects ofmodelling, but no attempt is made to provide readilyusable programs for the innumerable modelsthat exist.6.1 MATHEMATICAL DETERMINISTICMODELS[HOMS J04, J80, K22, K35, K55, L20]There are many ways to classify mathematicalmodels. For example, a model can be static ordynamic. A relationship between values of two variables,for example, between the river stage anddischarge in a cross-section, in the same time instantcan be interpreted as a static, or steady-state, modeland described with the help of an algebraic equation.An example of a dynamic model is aquantitative relationship between the river flow ina cross-section of interest in a given time instantand a set of earlier values of rainfall over the basinterminated by this cross-section: rainfall-runoffmodels. Dynamic models are typically formulatedin terms of differential equations, ordinary orpartial. There are a number of dichotomy-type categorizationsof dynamic models. For a discussion ofthese, see Singh (1988).

II.6-2GUIDE TO HYDROLOGICAL PRACTICESThe category of dynamic hydrological models isvery general and covers an entire spectrum ofapproaches. On one extreme are the purely empirical,black box techniques: those that make noattempt to model the internal structure but onlymatch the input and output of the catchmentsystem. A special category of black box models areartificial neural networks. On the other extreme aretechniques involving complex systems of equationsbased on physical laws and theoretical conceptsthat govern hydrological processes: the so-calledhydrodynamic models (see Hydrological Model forWater-Resources System Design and Operation,Operational Hydrology Reoort No. 34). Betweenthese two extremes there are various conceptualmodels. These models represent a structure built ofsimple conceptual elements, such as, linear or nonlinearreservoirs and channels that simulate, in anapproximate way, processes occurring within thebasin. Whether the models are black box, conceptualor hydrodynamic, they yield outputs withoutthe possibility of evaluating associated probabilitiesof occurrence. For this reason, they are often referredto as deterministic models.Lumped models have constant parameters, whichdo not change in space and are typically describedby ordinary differential equations, while parametersof distributed models, whose physics is describedby partial differential equations, may vary in space.Distributed and semi-distributed models havebecome common as they make use of the distributeddata fields which are available fromremote-sensing. Linear models are convenient touse because they may have closed-form solutionsand obey the superposition principle, which nonlinearmodels do not. Models can be stationary, inother words, time-invariant, if the input–outputrelationship and model parameters do not changewith time. Otherwise, models are non-stationary:time-variant. Models can be continuous and hencedescribed by differential equations and integrals, ordiscrete and described by difference equations andsums.Purely empirical and black box relationships haveproven and will continue to prove beneficial undercertain circumstances, but they are subject to seriouserror when it becomes necessary to rely onthem under conditions not previously experienced.Models that treat through theoretical conceptsvaried and interacting hydrological processes,namely, physically based models, are expected to bemore trustworthy under such conditions, andexperimentation with them holds great promise forscience. Any attempt to classify deterministicmodels as hydrodynamic, conceptual or black boxmodels admittedly forces a decision as to the degreeof empiricism. The division of dynamic hydrologicalmodels is, to some extent, arbitrary in the sensethat one person’s empiricism may be anotherperson’s theory (Singh, 1988). Nevertheless, it hasbeen deemed appropriate to follow such a divisionin the treatment of deterministic models.6.1.1 Black box modelsA river basin can be regarded as a dynamic systemin which lumped parameters, which are invariantover the basin, transform the input factors, precipitationand snowmelt, into a hydrograph of outflowfrom the basin. The same is true for a river reach,except that the inflow at the upstream point orpoints must be treated as an additional input factor.Diagrammatically, such systems can be representedas shown in Figure II.6.1, where P(t) is the input andQ(t) is the output, both functions of time t. Fromthe standpoint of dynamic systems theory, hydrologicalsystems behave as linear systems if theysatisfy the principle of superposition, namely, thatthe reaction of the system to a combination ofinputs is equal to the sum of its responses to theseparate inputs, and that the system parameters areindependent of the system’s inputs or outputs. Thepremise that the outflow hydrograph of a basin canbe predicted from a sequence of precipitation andsnowmelt only involves the assumption that thevariability of other natural inputs, such as evapotranspiration,is small or negligible, or follows a knownfunction of time.P(t)Figure II.6.1. Black box systemThe general expression for the relationshipbetween input P(t) and output Q(t) of a lumpedparameter,linear dynamic system may be writtenas follows:a n (t ) d n Qdt n+ a n−1 (t ) d n−1 Qdt n−1Systemmodel= b n (t ) d n Pdt n + b n−1 (t ) d n−1 Pdt n−1+ ... + a dQ1 (t )dtQ(t)+ a 0 (t ) Q+ ... + bdP1 (t ) + b 0 (t )P (6.1)dtwhere the coefficients a iand b iare the parameterscharacterizing the properties of the system. Thesolution to equation 6.1 for zero initial conditionsgives:tQ (t ) = h (t,τ )P (τ )dτ(6.2)∫0

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-3where the function h(t,τ) represents the response ofthe system at a time t to a single input impulse attime τ. There are numerous approaches to the representationof hydrological systems by formulationsinvolving the influence function h(t,τ), also calledimpulse response. These can be expressed in termsof the coefficients a iand b iof equation 6.1. If thecoefficients are constant in time, the system is timeinvariant and equation 6.2 becomes the Duhamelintegral:tQ (t ) = ∫ h (t −τ)P (τ )dτ(6.3)0It can be shown that the unit hydrograph conceptand the routing techniques discussed in 6.3.2.2.5and 6.3.4.3 are all examples of linear dynamicsystems involving the principle of superposition.Non-linear systems are those for which the superpositionprinciple is not observed. In general, theresponse of a non-linear, lumped-parameter systemto an input can be expressed either by an ordinarynon-linear differential equation or by the integralequation:tQ (t ) = ∫ h (τ )P (t −τ)dτ0t+ ∫ h (τ 1 ,τ 2 )P (t −τ 1 )P (t −τ 2 )dτ 1 dτ 20t t+ ∫ ... ∫ h (τ 1 ,τ 2 ,...,τ n )P (t −τ 1 )0 0P (t −τ 2 )... P(t−τ n )dτ 1 dτ 2 ...dτ n(6.4)where h(τ 1, τ 2, ... τ n) is a function expressing thetime-invariant characteristics of the physicalsystem. It is analogous to the influence function inequation 6.2. The first term on the right-hand sideof equation 6.4 defines the linear properties of thesystem, while the second defines the quadraticproperties; the third defines the cubic properties,and so on.6.1.2 Artificial neural networksA particular class of mathematical model is the artificialneural network, which is being increasinglyused as an alternative way to solve a range of hydrologicalproblems. The approach can be regarded asa modelling tool composed of a number of interconnectedsignal-processing units called artificialneurons. Artificial neural networks, which cancapture and represent complex input–output relationships,resemble the parallel architecture of thehuman brain; however, the orders of magnitude arenot as great. The idea behind the development ofartificial neural networks was the desire to simulatethe basic functions of the natural brain and developan artificial system that could perform intelligenttasks similar to those performed by the brain.Artificial neural networks acquire knowledgethrough learning and store the acquired knowledgewithin inter-neuron connection or synapticweights.Artificial neural networks are a simple clustering ofthe primitive artificial neurons. Each neuron isconnected to a number of its neighbours. Thisclustering occurs by creating layers, which areconnected to one another. The connections determinewhether it is possible for one unit to influenceanother. Some of the neurons in input and outputlayers interface with the real world: neurons in theinput layer receive input from the external environment,while those in the output layercommunicate the artificial neural network outputto the external environment (Figure II.6.2). Thereare usually a number of hidden layers between theinput and output layers.When the input layer receives the input, itsneurons produce an output which becomes aninput to the next layer of the system. The processcontinues until the output layer fires the output tothe external environment. An input–output function,or transfer function, should be specified forthe artificial neural network units. For instance,the transfer function can follow a linear, thresholdor sigmoid law. To construct a neural network thatperforms a specific task, the structure of thenetwork and the scheme of connections betweenunits must be chosen and the weights on theInput layerInputsHidden layerOutput layerFigure II.6.2. Structure of an artificial neuralnetworkOutputs

II.6-4GUIDE TO HYDROLOGICAL PRACTICESconnections specifying the strength of the connectionsmust be set.The learning ability of a neural network is determinedby its architecture and the algorithm chosenfor training. There are a variety of learning lawswhich are used in artificial neural networks. Theselaws are mathematical algorithms used to updatethe connection weights. Changing connectionweights of an artificial neural network, known astraining, causes the network to learn the solution toa problem. Gathering new knowledge is accomplishedby adjusting connection weights in such away that the overall network produces appropriateresults. An artificial neural network developer mustdecide on the arrangement of neurons in variouslayers, on inter-layer and intra-layer connections,on the way a neuron receives input and producesoutput, and on the principle of the learning process.Determination of a number of hidden neuronsin the network can be seen as an optimization taskwhich is often done by means of trial and error.Excessive increase of the hidden number of neuronsleads to an overfit, when generalization is difficult.A range of artificial neural network architecturesand training algorithms ranging from feed-forwardartificial neural networks trained using back propagationto self-organizing maps for pattern discoveryhave been devised. Artificial neural networks are aquick and flexible approach which has been foundsuitable for hydrological modelling in a wide varietyof circumstances. There are several applicationsof neural networks of interest to hydrology in areassuch as rainfall-runoff modelling (see Minns andHall, 1996), flow routing (Cigizoglu, 2003) and sedimenttransport (Tayfur, 2002). As neural networksare best at identifying patterns or trends in data,they are well suited to prediction or forecasting.The principal advantage of neural networks lies intheir ability to represent both linear and non-linearrelationships and to learn these relationshipsdirectly from the data being modelled. Traditionallinear models are simply inadequate when it comesto modelling data that contains non-linear characteristics,as is the case for most hydrological systems.At the start of the twenty-first century, a great dealof research is being conducted in neural networksand their application to solve a variety of problemsworldwide. However, hydrological practices havenot yet accommodated these methods on a routinebasis. Well-established technologies are stillpreferred to novelties whose advantages are yet tobe proved. Also, the black box nature of artificialneural networks has caused reluctance on the partof some hydrologists.6.1.3 Conceptual modelsThe approaches discussed in the previous sectionsmake use of only very general concepts of the transformationof input data into the outflow hydrograph,while more structural information about a systemor a process may be available. Such an approach isinadequate for solving catchment modelling problemsin which it is necessary to evaluate the effectsof climate variability and change, changes in landuse and other human activities. As a result, a modellingapproach has been developed that involvesstructures based on various simplified concepts ofthe physical processes of flow formation. These arecommonly referred to as conceptual models.One of the most difficult aspects of applying conceptualmodels is the calibration of a chosen model toa particular catchment. Most of the parameters aredetermined by iterative processes, either automaticor manual, that use historical input-output data.Owing to data limitations, model imperfectionsand the interrelationships among the model parameters,a small increase in the number of parametersis likely to have a major, and negative, effect on thedifficulty experienced in attempting calibration. Itis necessary, therefore, that the number of parametersbe compatible with the reliability of the inputdata and the required accuracy. In other words,modern concepts of theoretical merit must generallybe simplified in favour of utility.A wide variety of conceptual models are describedin the literature (Intercomparison of ConceptualModels Used in Operational Hydrological Forecasting(WMO-No. 429)). Under the circumstances, it seemsappropriate to limit the discussion to a brief descriptionof three models, representing a reasonablecross-section of those suitable for treatment in thisGuide. Several conceptual models are included inthe Hydrological Operational Multipurpose System(HOMS) of WMO.6.1.3.1 Sacramento modelThe Sacramento model was developed by the staffof the National Weather Service River ForecastCenter in Sacramento, California. This modelembodies a complex moisture-accounting algorithmto derive volumes of several runoffcomponents, while a rather simple and highlyempirical method is used to convert these inputs tothe outflow hydrograph. The soil mantle is treatedin two parts, an upper zone and a lower zone, witheach part having a capacity for tension water andfree water. Tension water is water that is closelybound to the soil particles and depleted only by

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-5evapotranspiration. Provision is made for free waterto drain downward and horizontally. The storagecapacities for tension water and free water in eachzone are specified as model parameters. Water enteringa zone is added to tension storage, as long as itscapacity is not exceeded, and any excess is added tofree water storage.A portion of any precipitation is diverted immediatelyto the channel system as direct runoff. This isthe portion that falls on the channel system and onimpervious areas adjacent thereto. The extent ofthis area is time variant in the model. All rainfalland snowmelt, other than rainfall and snowmeltthat are diverted to direct runoff, enter the upperzone. Free water in the upper zone is depleted eitheras interflow or percolation to the lower zone. If therate of moisture supply to the upper zone is greaterthan the rate of depletion, the excess becomessurface runoff. Free water in the lower zone isdivided between primary, slow drainage storage andsecondary storage. Figure II.6.3 illustrates the principalfeatures of the model.Percolation from the upper to the lower zone isdefined as:PRATE = PBASE ⎡⎣ 1 + ZPERC * RDC REXP ⎤⎦ UZFWCUZFWM (6.5)where PRATE is the percolation rate, and PBASE isthe rate at which percolation would take place ifthe lower zone were full and if there were an unlimitedsupply of water available in the upper zone. Itis numerically equal to the maximum lower zoneoutflow rate and is computed as the sum of thelower-zone primary and secondary free-watercapacities, each multiplied by its depletion coefficient.RDC is the ratio of lower zone deficiency tocapacity. That is, RDC is zero when the lower zoneUpperzoneLowerzonePrimaryfreewaterstorageTensionwaterstorageTension water storageFree water storageTensionSupplementarywaterstoragefreewaterstorageInterflowBaseflowSurface runoffSubstratecutflowDirect runoffFigure II.6.3. Structure of the Sacramento modelis full and is in unity when it is empty. ZPERC is amodel parameter that defines the range of percolationrates. Given an unlimited supply of upper zonefree water, the rate will vary from PBASE (lowerzone full) to PBASE(1 + ZPERC) when the lowerzone is empty. REXP is a model parameter thatdefines the shape of the curve between the minimumand maximum values described above.UZFWC is the upper zone freewater content.UZFWM is the upper-zone free capacity. The ratio,UZFWC/UZFWM, represents the upper zone drivingforce. With the upper zone empty, there will be nopercolation. With it full, the rate will be governedby the deficiency in the lower zone.This equation is the core of the model. It interactswith other model components in such a way that itcontrols the movement of water in all parts of thesoil profile, both above and below the percolationinterface, and, in turn, is controlled by the movementin all parts of the profile. Evapotranspirationrates are estimated from meteorological variables orfrom pan observations. Either day-by-day or longtermmean values can be used. The catchmentpotential is the product of the meteorologicalevapotranspiration and a multiplier that is a functionof the calendar date, which reflects the state ofthe vegetation. The moisture accounting within themodel extracts the evapotranspiration loss directlyor indirectly from the contents in the various storageelements and/or from the channel system. Theloss is distributed according to a hierarchy of prioritiesand is limited by the availability of moisture aswell as by the computed demand.The movement of moisture through the soil mantleis a continuous process. The rate of flow at anypoint varies with the rate of moisture supply andthe contents of relevant storage elements. Thisprocess is simulated by a quasi-linear computation.A single time-step computation of the drainage andpercolation process involves the implicit assumptionthat the movement of moisture during thetime step is defined by the conditions existing atthe beginning of the step. This approximation isacceptable only if the time step is relatively short.In the model, the length of the step is volumedependent. That is, the step is selected in such away that no more than five millimetres of watermay be involved in any single execution of thecomputational loop.Five components of runoff are derived in the model.The three upper components – direct, surface andinterflow – are summed and transformed by a unithydrograph (see 6.3.2.2.5). The two componentsfrom the lower zone, and primary and secondary

II.6-6GUIDE TO HYDROLOGICAL PRACTICESbase flow, are added directly to the outflowhydrograph derived from the other three components.Provision is also made for routing theresultant hydrograph with variable routingcoefficients.The Sacramento model is a HOMS component withthe following identification code: J04.3.01.6.1.3.2 Tank modelThis model was developed at the National ResearchInstitutes for Earth Science and Disaster Preventionin Tokyo, Japan (Sugawara and others, 1974). Asthe name implies, the soil mantle is simulated bya series of tanks arranged one above the other, asshown in Figure II.6.4 (a). All rainfall and snowmeltis assumed to enter the uppermost tank. Eachtank has one outlet in the bottom and one or twoon a side at some distance above the bottom. Waterthat leaves any tank through the bottom entersthe next lower tank, except for the lowermosttank, in which case the downflow is a loss to thesystem. Water leaving any tank through a sideoutlet, known as sideflow, becomes input to thechannel system. The number of tanks and the sizeand position of the outlets are the modelparameters.The configuration is a suitable representation of therainfall-runoff process in humid regions, but amore complex arrangement is required for catchmentsin arid and semi-arid areas, as shown inFigure II.6.4 (b). If extended dry periods are typical,xS 1two or more series of tanks, as described above, areplaced in a parallel arrangement. The downflows ineach series are the same as for the simple tankmodel. Each tank in each series contributes sideflowto the corresponding tank of the next series, exceptthat all sideflow from the last series feeds directlyinto the channel system. Additionally, provision ismade for sideflow from the uppermost tank in allother series to feed directly into the channel system.Each series is considered to represent a zone of thecatchment, the lowest corresponding to the zonenearest the channels. As hydrological conditionsmake their seasonal progression from wet to dry,the zone nearest the channels can continue to berelatively wet after the one furthest removed hasbecome rather dry. The originators of the model donot claim that the representation of storageelements is entirely realistic, but rather that theconfiguration of tanks is an approximation somewhatresembling the finite-element method.Furthermore, the mathematical formulations definingthe flow of water through the tanks resembleclassical hydrological concepts.Two types of water are recognized in the model,confined water, namely, soil moisture, and freewater that can drain both downward and horizontally.Provision is also made for free water toreplenish soil moisture by capillarity action. Themodel computes evapotranspiration loss from thecatchment based on measured or estimated dailyevaporation, on the availability of water in storage,and on a hierarchy of priorities from the differentstorage elements.The basic numerical calculation within a tankinvolves a withdrawal function defined by:xS 2xS 3dxdt=αx(6.6)xS 4where x is the contents of the tank and t is time.The outflow in a finite unit of time, Δt, is therefore[1 – e –αΔt ]x. The [1 – e –αΔt ] quantity is computed foreach outlet, based on the value of α and the specifiedtime interval.(a)Figure II.6.4. Tank model(b)xS 4xS 4xS 4The computation for each time interval proceeds inthe following order:(a) For the uppermost tank:(i) Extraction of evapotranspiration;(ii) Transfer of free water to soil moisture;(iii) Addition of rainfall and snowmelt;(iv) Calculation and extraction of channelsystem input (sideflow) and percolation(downflow) from free water contents;

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-7(b) For a lower tank:(i) Extraction of evapotranspiration, dependingon hierarchy of priorities;(ii) Transfer of free water to soil moisture;(iii) Addition of percolation water from tankimmediately above;(iv) Calculation and extraction of channelsystem input (sideflow) and percolation(downflow) from free water contents.The input to the channel system is the output fromthe moisture-accounting phase of the model. Theoutflow hydrograph is derived from the channelsystem input by routing, assuming that:Q = KS 2 (6.7)where Q is outflow, S is channel storage and theconstant K is an additional parameter of the model.A limit of unity is imposed on dQ/dS to prevent theoutflow from exceeding channel storage. An interestingfeature of the tank model is that changes inthe values of model parameters can actually changethe structure of the model.The tank model is a HOMS component with identificationcode J04.1.01.6.1.3.3 HBV modelThe HBV model, developed by Bergstrom (1992,1995) at the Swedish Meteorological andHydrological Institute, is a conceptual watershedmodel which converts precipitation, air temperatureand potential evapotranspiration intosnowmelt, if applicable, and streamflow or reservoirinflow. The model has been modified many timesand exists in different versions in a number ofcountries.The model describes the general water balance inthe following way:P − E − Q =ddt[ SP + SM + UZ + LZ + VL] (6.8)where P is precipitation, E is evapotranspiration, Qis runoff, SP is snowpack, SM is soil moisture, UZ isupper groundwater zone, LZ is lower groundwaterzone and VL is the volume of lakes.The HBV model can be regarded as a semi-distributedmodel by dividing the catchment intosub-basins and using elevation zoning. The modelcontains subroutines for meteorological interpolation,snow accumulation and melt,evapotranspiration, soil moisture accounting,runoff generation and, finally, routing throughrivers and lakes. For basins of considerable elevationrange, a subdivision into elevation zones ismade. Each elevation zone can be divided furtherinto vegetation zones, such as forested and nonforestedareas.The standard snowmelt routine of the HBVmodel is a degree-day approach, based on airtemperature. Melt is further distributed accordingto the elevation zoning and the temperaturelapse rate and is modelled differently in forestsand open areas. The snowpack is assumed toretain melt water as long as the amount doesnot exceed a certain water holding capacity ofsnow. When temperature decreases below thethreshold temperature, this water refreezesgradually.The soil moisture accounting of the HBV model isbased on a modification of the tank approach inthat it assumes a statistical distribution of storagecapacities in a basin. This is the main control ofrunoff formation. Potential evapotranspiration isreduced to actual values along with a growing soilmoisture deficit in the model and occurs from lakesonly when there is no ice. Ice conditions aremodelled with a simple weighting subroutine onair temperature, which results in a lag between airtemperature and lake temperature.The runoff generation routine is the response functionwhich transforms excess water from the soilmoisture zone to runoff. It also includes the effectof direct precipitation and evaporation on lakes,rivers and other wet areas. The function consists ofone upper, non-linear, and one lower, linear, reservoir,producing the quick and slow, or base-flow,runoff components of the hydrograph. Lakes canalso be modelled explicitly so that level pool routingis performed in lakes located at the outlet of asub-basin. The division into submodels, defined bythe outlets of major lakes, is thus of great importancefor determining the dynamics of the generatedrunoff. River routing between sub-basins can bedescribed by the Muskingum method, (see, forexample, Shaw, 1994) or simple time lags.A comprehensive re-evaluation of the model wascarried out during the 1990s and resulted in themodel version called HBV-96 (Lindström andothers, 1997). The objectives were to improve thepotential for accommodating spatially distributeddata in the model, make the model more physicallysound and improve the model performance. Themodel revision led to changes in the process descriptions,automatic calibration and optimalinterpolation of precipitation and temperature, via

II.6-8GUIDE TO HYDROLOGICAL PRACTICESa geostatistical method. When combined, the modificationsled to significant improvements in modelperformance. The option of higher resolution inspace is also necessary for future integration ofspatially distributed field data in the model. Theimprovements in model performance were duemore to the changes in the processing of input dataand the new calibration routine than to the changesin the process descriptions of the model.Required input to the HBV model are precipitation(daily sums), air temperature (daily means) andestimates of potential evapotranspiration. Thestandard model is run with monthly data of longtermmean potential evapotranspiration, usuallybased on the Penman formula, adjusted for temperatureanomalies (Lindström and Bergström, 1992).As an alternative, daily values can be calculated asbeing proportional to air temperature, but withmonthly coefficients of proportionality. Laterversions of the HBV model can be run with data ofhigher resolution in time, that is, hourly data.Although the automatic calibration routine is not apart of the model itself, it is an essential componentin practice. The automatic calibration method forthe HBV model developed by Lindström (1997) hasoptions for use of different criteria for differentparameters or for use of combined criteria. Thisprocess generally requires three to five years ofsimultaneous streamflow and meteorologicalrecords. If no streamflow records are available, theparameters can, in some cases, be estimated fromknown basin characteristics.The area of applicability of HBV is broad andembraces spillway design (Bergström and others,1992; Lindström and Harlin, 1992), water resourcesevaluation, nutrient load estimates (WMO, 2003),and climate change studies (Bergström and others,2001). A recent trend is the use of the model fornation-wide hydrological mapping, as in Norway(Beldring and others, 2003) and Sweden (SNA,1995). The HBV model is a HOMS component withidentification code J04.2.02. For further information,see: http://www.smhi.se/sgn0106/if/hydrologi/hbv.htm.6.1.4 Distributed modelsThe field of mathematical modelling in hydrologyhas been traditionally dominated by lumped modelswith constant parameters, representing a wholedrainage basin. However, several semi-distributedand distributed models have been developed morerecently. Their formulation aims at following thehydrological processes more closely and thus mayincorporate several meteorological variables andwatershed parameters. Their product can be, forexample, a series of synthetic streamflow data,water quality characteristics and rates of groundwaterrecharge. The basic input is a dataseries of rainfalldata; however, provisions may be made for factorssuch as snowfall, temperature, radiation and potentialevapotranspiration. Models for urbancatchments may contain a description of theirdrainage network. Models for rural catchments maycontain unit hydrographs, time-area curves or routingsubroutines.However, the distributed, physically based modelsare still being used at a fraction of their potential(Refsgaard and Abbott, 1996). There are severalreasons for this. Distributed models require a largeamount of data that often do not exist or are notavailable. Operational remote-sensing is still not acommon practice, except for snow cover and landuse/vegetation.Numerous parameters of a physically baseddistributed model cannot be measured in the fieldand calibration of such a model is a difficult optimizationtask. Moreover, more complex andjustified descriptions are rarely implemented indistributed models because they would requiremore parameters, which need to be identified.This simplification may undermine the rigour ofthe physical basis.As noted by Beven (1996), physically based distributedmodels use small-scale equations with anassumption that the change of scales can be accommodatedby the use of effective parameter values.However, physically based equations of small scaledo not scale up easily in a heterogeneous system.Beven (1996) saw a possible solution in an approachthat recognizes the limitations of the modellingprocess, such as within an uncertainty framework.Scale-dependent parameters could be based on astatistical model of heterogeneity. In general, efficientaggregate parameterization is not a trivialmatter.Distributed models provide a basis for full use ofdistributed information relevant to the physicalprocesses in the catchment. The EuropeanHydrological System (DHI, 1985) is an example of adistributed hydrodynamically based model and isillustrated in Figure II.6.5. The EuropeanHydrological System is a model with distributedparameters that has been developed from partialdifferential equations describing the physical processesin the basin: interception, evapotranspiration,overland and channel flow, movement of water

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-9through unsaturated and saturated zones, andsnowmelt.The interception process is represented by a variantof the Rutter model that gives the rate of change inthe amount of water stored on the canopy:∂c∂t= Q − Ke b (C − S )where: Q =⎧⎪⎨⎩⎪P 1 P 2 ( P − E p C / S )P 1 P 2 ( P − E p )when C < Swhen C ≥ S(6.9)C is the actual depth of water on the canopy, S isthe canopy storage capacity, P is the rainfall rate, P 1is the proportion of ground in plan view hidden byvegetation, P 2is the ratio of total leaf area to area ofground covered by vegetation, E pis the potentialevaporation rate, K and b are drainage parametersand t is time.For the prediction of actual evapotranspirationrates, the Penman–Monteith equation is used:E a =ϕ C p v eΔ R nr aλ[ Δ + γ (17γ s / r a ](6.10)where ϕ is the density of air, λ is the latent heat ofvaporization of water, E ais the actual evapotranspirationrate, R nis the net radiation minus the energyflux into the ground, Δ xis the slope of the specifichumidity/temperature curve, C pis the specific heatof air at constant air pressure, v eis vapour pressuredeficit of the air, r ais the aerodynamic resistance towater vapour transport, γ sis the canopy resistanceto water transport, and γ is the psychometricconstant.EvapotranspirationRain andsnow inputRoot modelzoneCanopy interceptionmodelLayered snowmeltmodelInterflowOverland and channel flow model (rectangular grid)Saturated flow model(rectangular grid)1-dimensional unsaturatedzone model for each grid elementLess permeable (e.g. bedrock)Figure II.6.5. Structure of the EuropeanHydrological SystemThe interception process is modelled as an interceptionstorage, which must be filled before stemflow to the ground surface takes place. The size ofthe interception storage capacity, I max, depends onthe vegetation type and its stage of development,which is characterized by the leaf area index, LAI.Thus:I max= C int× LAI (6.11)where C intis an interception coefficient whichdefines the interception storage capacity of thevegetation. A typical value is about 0.05 mm, but amore exact value may be determined through calibration.The area of leaves above a unit area of theground surface is defined by the leaf area index.Generalized time-varying functions of the leaf areaindex for different crops have been established.Thus, when employing the modelling tool MIKESHE, the user must specify the temporal variationof the leaf area index for each crop type during thegrowing seasons to be simulated. Climatic conditionsdiffer from year to year and may require ashift of the leaf area index curves in time but willgenerally not change the shape of the curve.Typically, the leaf area index varies between 0 and7. Evaporation from the canopy storage is equal tothe potential evapotranspiration, if sufficient waterhas been intercepted on the leaves, that is:E can= min I maxE pΔt (6.12)where E canis the canopy evaporation, E pis thepotential evapotranspiration rate and Δt is the timestep length for the simulation.Water accumulated on the soil surface responds togravity by flowing downgradient over the landsurface to the channel system, where it subsequentlydischarges through the stream channels tothe catchment outlet. Both phenomena aredescribed by equations of unsteady free-surfaceflow, based on physical principles of conservationof mass and momentum (DHI, 1985).In the most comprehensive mode, the flow in theunsaturated zone can be computed by using theRichards equation:C = ∂Ψ∂t= ∂∂Z⎛K⎝∂Ψ∂Z⎞⎠ + ∂K∂Z + S (6.13)where Ψ is the pressure head, t is the time variable,Z is the vertical coordinate (positive upwards),C = ∂Θ / ∂Ψ is the soil-water capacity, Θ is the volumetricmoisture content, K is hydraulic conductivityand S is a source/sink term.

II.6-10GUIDE TO HYDROLOGICAL PRACTICESThe infiltration rate into the soil is determined bythe upper boundary condition, which may shiftfrom flux-controlled conditions to soil-controlledor saturated conditions and vice versa. The lowerboundary is usually the phreatic-surface level. Thegoverning equation describing the flow in thesaturated zone is the non-linear Boussinesqequation:S = ∂h∂t= ∂∂x⎛K x H ∂h ⎞⎝ ∂x ⎠ +∂∂y⎛K y H ∂h ⎞⎝ ∂y ⎠ + R (6.14)where S is the specific yield; h is the phreatic surfacelevel; K x, K yare the saturated hydraulic conductivitiesin the x and y directions, respectively; H is thesaturated thickness; t is the time variable; x, y arethe horizontal space coordinates and R is an instantaneousrecharge/discharge term.Equation 6.14 is solved by approximating it to aset of finite difference equations, that is, by applyingDarcy’s law in combination with the massbalance equation for each computational node.Considering a node i inside the model area, thetotal inflow R from neighbouring nodes andsources/sinks between time n and time n+1 isexpressed as follows:R = Σq zn+1+ Σq xn+1+ RH iΔx 2 (6.15)where the first term on the right-hand side is thevolumetric flow in the vertical direction, the secondterm is the volumetric flow in horizontal directions,R is the volumetric flow rate per unit volume fromall sources and sinks, Δx is the spatial resolution inthe horizontal direction and H iis either the saturateddepth for unconfined layers or the layerthickness for confined layers.The snowmelt component of SHE represents anattempt to model both energy and mass flux withina snowpack by taking into account changes in thestructure of the pack. Two semi-empirical equationsare used to complete the set of relationships requiredto define temperature and water-content distributions.Empirical equations are also used to definethe hydraulic and thermal properties of the snow interms of the structure, water content andtemperature.There have been several products linked to SHE,including MIKE SHE, SHETRAN, or SHESED developedrecently. Basic description of processes fromthe original SHE remains in MIKE SHE. This latterpackage (Storm and Refsgaard, 1996) extended incomparison to SHE, has been used in a number ofpractical applications, including flow simulation,solute transport, applications in irrigation andsalinity planning, and management models.6.1.5 Parameter evaluationGeneral methods of parameter evaluation or identificationsometimes referred to as model calibration,have been developed for a wide range of dynamicsystems. Experience has shown that the success ofsuch methods depends on the availability ofadequate information concerning the system characteristicsand the form of the influence function,or impulse response. There are two basic approachesto calibration.In the first approach, the mathematical model iscombined with the data to solve for the unknowncoefficients, the system parameters. Such a taskbelongs to the category of ill-posed mathematicalinverse problems and is usually difficult to solve. Inthe linear case, matrix inversion may be needed.The solutions can be very sensitive to inaccuracy inthe data. They tend to be unstable and multiplesolutions may exist. The optimum found by theoptimization software can be a local, rather than aglobal optimum.The second approach involves experimentationwith various combinations of parameter values inan effort to minimize or maximize an adoptedcriterion of optimization. A number of strategieshave been developed by applied mathematicianswith a view to minimizing the number of calculationsrequired in the optimization of parametervalues. Among the strategies used in hydrology arethe gradient and non-gradient methods. Theadequacy of the solution can be highly dependenton the criterion used in the analysis. A number ofcriteria have been developed and introducedthrough WMO projects (WMO, 1986, 1987,1991a). These can be recommended for generaluse.To determine the parameters of complex, conceptualhydrological models which have severalcomponents, the following principles arerecommended:(a) Separate testing of the model components usingall available experimental and scientific information.It is well known that the global determinationof all parameters of a model throughoptimization may result in unrealistic values ofthe parameters, some even falling outside theirphysical range. This is the case when certainmodel components contain systematic errorsthat are subsequently compensated within themodel. In order to avoid such situations, it is

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-11recommended that the parameters of complexconceptual models be determined separatelyfor each of the basic components and notglobally;(b) Data from a minimum three-year time intervalshould be used for the calibration of models,and another time interval of similar lengthshould be used for verification. The calibrationand verification intervals in this split-sampleapproach should be selected so that they representdifferent conditions favouring runoffformation, for example: floods generated byrainfall, floods resulting from snowmelt processesand low flows;(c) In the case of basins with a hydrological regimeunder anthropogenic influences, it is recommendedthat the model be calibrated for thenatural runoff regime. The values of certainparameters may be modified subsequently toaccount for human influences. The validationof the model parameters should be done for arepresentative period that is not influenced byhuman activities.The parameters of hydrodynamic models representbasin characteristics, such as roughness of theslopes and the river bed, soil hydraulic conductivityand soil porosity. In principle, all of theseparameters are physically based and determinedby field measurements and not through optimization.However, this is not always possible inpractice.6.1.6 Selection of modelsIn addition to software packages developed inEurope and North America, several products fromother countries are being increasingly used in theinternational context. For example, two modelsdeveloped in South Africa have gained internationalrecognition. The ACRU (AgriculturalCatchments Research Unit) agrohydrologicalmodelling system developed by Schulze at theUniversity of Natal since the early 1970s is a multipurposeintegrated physical conceptual modelsimulating streamflow, sediment and crop yields.The Pitman rainfall-runoff model for monthly timesteps has been widely used in southern Africa forbroad strategic water resource planning purposes(see Hughes and Metzler, 1998). Recently, Hughes(2004a) extended the Pitman model by adding twonew components, recharge and groundwaterdischarge, thus responding to the urgent need inpractice for an integrated surface water and groundwatermodelling tool which can be applied atvarious basin scales in southern Africanconditions.The choice of models is not restricted to the modelsdescribed above. Many models produced by researchinstitutions and commercial software companiesare available. It is often difficult to ascertain therelative advantages and disadvantages of modelsproposed for operational use. The selection of amodel suitable for a specific hydrological situationhas implications in water resources planning, developmentand management; in hydrologicalforecasting activities and in setting directions offurther research in modelling. Some of the factorsand criteria involved in the selection of a modelinclude the following:(a) The general modelling objective: hydrologicalforecasting, assessing human influences on thenatural hydrological regime or climate changeimpact assessment;(b) The type of system to be modelled: small watershed,aquifer, river reach, reservoir or largecatchment;(c) The hydrological element to be modelled:floods, daily average discharges, monthlyaverage discharges, groundwater levels, waterquality and so forth;(d) The climatic and physiographical characteristicsof the watershed;(e) Data availability with regard to type, lengthand quality of data versus data requirementsfor model calibration and operation;(f) Model simplicity, as far as hydrological complexityand ease of application are concerned;(g) The possible need for transposing modelparameters from smaller catchments to largercatchments;(h) The ability of the model to be updatedconveniently on the basis of current hydrometeorologicalconditions.Useful information and guidance on the selectionand application of conceptual models in varioushydrological situations can be found in documentationof several WMO projects carried out sincethe 1970s, such as the following:(a) Intercomparison of conceptual models used inoperational hydrological forecasting (WMO,1987);(b) Intercomparison of models of snowmelt runoff(WMO, 1986);(c) Simulated real-time intercomparison of hydrologicalmodels (WMO, 1991a).Many hydrological software packages have beendeveloped by scientific research institutes andcommercial companies for PCs and work stationsusing MS Windows, UNIX and LINUX platforms.Many models are equipped with a geographicalinformation system interface.

II.6-12GUIDE TO HYDROLOGICAL PRACTICESHydrological models within HOMS are groupedin a number of sections. Section J, HydrologicalForecasting Models, includes models whosemain purpose is the operational forecasting ofvarious hydrological elements. SubsectionJ04, Forecasting Streamflow fromHydrometeorological Data, includes the threemodels, Sacramento, tank and HBV, introducedin 6.1.3.1 to 6.1.3.3.At the time of writing, further components in thissubsection of HOMS include the following: J04.1.04,Snowmelt-runoff model (SRM); J04.1.05, Inflow–storage–outflow (ISO) function models; J04.2.01,Conceptual watershed model for flood forecasting;J04.3.03, Snow accumulation and ablation model(NWSRFS-SNOW-17); and J04.3.07, Synthetizedconstrained linear system (SCLS).Subsection J15, Combined Streamflow Forecastingand Routing Models, includes components J15.2.01,Streamflow synthesis and reservoir regulation(SSARR) model, and J15.3.01, Manual calibrationprogram (NWSRFS-MCP3).Further models are grouped under section K,Hydrological Analysis for the Planning and Designof Engineering Structures and Water ResourceSystems), for example, K15, Site-Specific FloodStudies, and K15.3.02, Dam-break flood model(DAMBRK). Subsection K22, Rainfall-RunoffSimulation Models, contains K22.2.02, Floodhydrograph package (HEC-1); K22.2.10,Hydrological rainfall runoff model (HYRROM);K22.2.11, Unit hydrographs and component flowsfrom rainfall, evaporation and streamflow data (PCIHACRES); K22.2.12, Non-linear rainfall-runoffmodel (URBS) and K22.3.01, Urban rainfall-runoffmodel (SWMM). Subsection K35, StreamflowSimulation and Routing, includes the followingcomponents: K35.1.05, Numerical solutions of thenon-linear Muskingum method; K35.2.09, Interiorflood hydrology (HEC-IFH); K35.3.06, River analysissystem (HEC-RAS); K35.2.06, Water-surfaceprofile computation model (WSPRO); K35.3.13,Branch-network dynamic flow model (BRANCH);and K35.3.14, Flow model for a one-dimensionalsystem of open channels based on the diffusionanalogy (DAFLOW). Subsection K 55, Water QualityStudies, includes the following components:K55.2.04, Transport model for a one-dimensionalsystem of open channels (BLTM); K55.2.06,Modelling faecal coliform concentrations instreams; K55.3.04, Mathematical model for twodimensionalsalinity distribution in estuaries; andK55.3.07, PC-QUASAR – Quality simulation alongrivers.Section L, Groundwater, includes subsection L20,Aquifer Simulation Models, with the followingcomponents: L20.2.04, Modular finite-differencegroundwater flow model (MODFLOW); L20.3.05,A model for unsaturated flow above a shallowwater table (MUST); L20.3.13, Complete programpackage for modelling groundwater flow(TRIWACO); L20.3.07, Pathlines and travel timesbased on analytical solutions (AQ-AS); L20.3.10,Groundwater head drawdowns based on analyticalsolutions (AQ-AP); L20.3.11, Aquifer simulationmodel; L20.3.12 , SGMP – Simulation of watertablebehavior in groundwater systems; and L20.3.14,MicroFEM – Finite-element multiple-aquifersteady-state and transient groundwater flowmodelling.6.2 TIME SERIES AND SPATIAL ANALYSISMany hydrological data consist of time series ofobservations of a hydrological variable in one pointin space. Studying a single time series of hydrologicaldata allows the identification of the temporalcorrelation structure of this variable in one point inspace. If more than one variable is being observedat the station, cross-correlations between time seriesof several variables in the same spatial point can bestudied. By considering time series of the same variableat a number of points in space, a spatial-temporalfield of this hydrological variable can be exploredand a cross-correlation between the time series ofthe same variable in different spatial points can beexamined. This makes it possible to interpret thetemporal and/or spatial-temporal structure of thehydrological processes and to use this insynthetic streamflow generation and extendingdata, such as for filling in gaps in data andextrapolation.Hydrological time series can be continuous in thatthey are derived from a continuously recordingdevice, discrete because they are sampled at discretetime instants at regular or irregular time intervals,or quantized, if each value of the time series is anintegral of a variable over a defined time interval.Continuous time series can be analysed in thetemporal domain or in the operational domain, forexample, via Fourier or Laplace integral transforms,which can be convenient in specific cases.When studying hydrological time series, it is importantto use appropriate time intervals. Data may behourly, daily, monthly or annual, but in a particularapplication it may be necessary to use either thetime interval dictated by the data acquisition, or a

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-13longer period, requiring aggregation, or a shorterperiod, requiring disaggregation. This has effects onthe characteristics of the series. Series of hourlystreamflow are very likely to contain highly correlatedvalues, while the correlation coefficient mayfall to zero in a series of annual values.There is extensive literature related to time seriesanalysis, including the monumental monographby Box and Jenkins (1970). Hydrological applicationsof time series analysis can be studied in Salas(1992). Elements of time-series analysis are widelyincluded in general-purpose statistical softwarepackages. The present section briefly describes practicalproblems in the field, dealing with stochasticsimulation and change detection in hydrologicalrecords.6.2.1 Stochastic simulation ofhydrological times seriesStochastic models are black box models, the parametersof which are estimated from the statisticalproperties of the observed times series. Stochasticmethods were first introduced into hydrology inconnection with the design of storage reservoirs.Annual or monthly flow volumes provide adequatedetail for such purposes, but the capacity of thereservoir must reflect the probability of occurrenceof critical sequences of flow that can best be evaluatedfrom a set of flow sequences. Each must span aperiod of many years and should be indistinguishablefrom the historic record in so far as its relevantstatistical characteristics are concerned. The statisticalproperties of the historical record that are to bepreserved are of primary concern in the selection ofan appropriate stochastic model. Modelling is muchmore difficult when it becomes necessary to generatesimultaneous flow sequences for two or morereservoir sites in a basin because of the requirementthat intercorrelations be preserved. Stochasticmodelling has also been used in the establishmentof confidence limits of real-time flow forecasts. Suchapplications are not given further treatment here. Adiscussion of the design and operation of storagereservoirs is provided in 4.2.6.2.1.1 Markovian lag-1 modelsMany models for simulating monthly, seasonal orannual flow volumes assume a first-order Markovstructure which assumes that the flow in any periodis determined by the flow in the preceding period,plus a random impulse. One such model for annualflows can be expressed as:in which Q iis the flow of the i th member of theseries numbering consecutively from 1 regardlessof month or year, j is the month in which the ithmember of the series falls, Q – jis the mean flow forthe jth month, σ jis the standard deviation for thejth month, ρ jis the serial correlation between Q – jand Q j–1and ε iis a random variate from an appropriatedistribution, with mean zero, unit varianceand serial independance.Equation 6.16 is also suitable for seasonal flows (j =1, 2, 3, 4) and annual flows (j = 1). In the latter caseit becomes:Q i = Q j + ρ (Q i−1 − Q j−1 ) + ε i σ 1 −ρ 2 (6.17)Values of Q – σ and ρ, derived from the historicalrecord, are assumed to be applicable for the purposesto be served, and an initial value of Q i–lneed onlybe selected to simulate a series of any length. MonteCarlo techniques are generally used with sequentialvalues of the random variate derived by computer.In principle, the development and application ofthe models depicted in equation 6.16 are relativelystraightforward and simple. Nevertheless, severalquestions requiring careful consideration and decisionsmay be critical to the particular problemunder study:(a) What is the distribution of the randomvariate?(b) Should the variance be corrected for serialcorrelation, if present?(c) How accurate is the calculated value of theserial correlation?6.2.1.2 Autoregressive moving averagemodelsAn important extension of the univariate stochasticmodels is represented by the group developed byBox and Jenkins (Box and Jenkins, 1970; Hipel andothers, 1977): the autoregressive moving averagemodels (ARMA).There are three types: autoregressive (AR), movingaverage (MA) and mixed (ARMA) models. The mostgeneral type (ARMA), of order p and q, and themoving average (MA), of order q, are,respectively:x i= φ 1x i–1+ φ 2x i–2+ ... + φ px i–p+ ε i– θ 1ε i–1– ... – θ qε i–q(6.18)Q i = Q j + ρ jσ jσ j−1(Q i−1 − Q j−1 ) + ε i σ j 1 −ρ j2(6.16)x i= ε 1– θ iε i–1– ... – θ qε i–q(6.19)

II.6-14GUIDE TO HYDROLOGICAL PRACTICESwhere x iis the deviation of the i th observation fromthe series mean, φ 1and θ 1are parameters to be estimated,and ε iis a random variate as defined above(see 6.2.2.1).A systematic approach has been developed forfitting ARMA models (Box and Jenkins, 1970):(a) Identification: The correlogram of the seriesunder study is compared with the autocorrelationfunctions of various ARMA models asa basis for selecting the appropriate type andorder;(b) Estimation: The parameters of the model areestimated (Salas, 1992) by using the method ofmoments, the method of maximum likelihoodor the method of least squares, where estimatesminimizing the sum of squared residuals areselected;(c) Diagnostic checking: The randomness of theresiduals is checked to verify the adequacy ofthe selected model.Autoregressive moving average models are used togenerate synthetic flow sequences by Monte Carlotechniques in the manner previously described. It isimportant to bear in mind that methods of stochasticgeneration should be used with caution andwith critical consideration of the characteristics ofthe record that are important for the water resourceproject under study.climatic change, anthropogenic factors or simplynon-homogeneity of the data series.6.2.2 Change detection in hydrologicalrecords6.2.2.1 IntroductionDetection of changes in long time series of hydrologicaldata is an issue of considerable scientific andpractical importance. It is fundamental for planningof future water resources and flood protection.If changes are occurring within hydrologicalsystems, existing procedures for designing structuressuch as reservoirs, dams and dykes will haveto be revised; otherwise, systems will be over- orunder-designed and will either not serve theirpurpose adequately or will be more costly thannecessary.Activities undertaken within the World ClimateProgramme – Water (WMO, 1988) have led to theestablishment of general recommendations onmethodology for use in detection of change inhydrological data, presented by Cavadias (WMO,1992) and Kundzewicz and Robson (WMO, 2000,2004). The present section is based on the latter tworeferences, which may be consulted for moredetailed recommendations regarding differentcomponents of the process of testing for changes.6.2.1.3 Fractional Gaussian noise andbroken-line process modelsHurst discovered (Hurst, 1951) that very longgeophysical records displaying characteristics atodds with stationary Markovian processes led tothe development of two stochastic models thatcan accommodate long-term persistence or lowfrequencyelements. The first of these, thefractional Gaussian-noise (FGN) model(Mandelbrot and Wallis, 1968) is a self-similar,random process characterized by a spectral densityfunction that emphasizes very low frequenciestypifying the Hurst phenomenon. It also has beenshown that a long-memory model of the brokenlineprocess will preserve the Hurst phenomenon(Rodriguez-Iturbe and others, 1972; Mejia andothers, 1972).The findings of Hurst do not necessarily indicatevery long-term persistence and, moreover, someversions of ARMA models are capable of simulatingsubstantial low-frequency effects. The non-stationarityof the process’s mean value could also result inthe characteristics that Hurst found when analysinglong records, whether these be the result of6.2.2.2 Basics of statistical testing for changedetectionChange in a time series can occur in numerousways: gradually (a trend), abruptly (a step change)or in a more complex form. It may affect the mean,median, variance, autocorrelation or other aspectsof the data.In order to carry out a statistical test, it is necessaryto define the null and alternative hypotheses whichdescribe what the test is investigating. For example,to test for trend in the mean of a series, the nullhypothesis would be that there is no change in themean of a series, and the alternative hypothesiswould be that the mean is either increasing ordecreasing over time. To perform a test, it is necessaryto begin by assuming that the null hypothesisis true. The next step is to check whether theobserved data are consistent with this hypothesis. Ifnot, the null hypothesis is rejected.To compare between the null and alternativehypotheses, a test statistic is selected and its significanceis evaluated, based on the available evidence.The test statistic is simply a numerical value that is

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-15calculated from the data series that is being tested.A good test statistic should highlight the differencebetween the two hypotheses. A simple example of atest statistic is the linear regression gradient, whichcan be used to test for a trend in the mean. If thereis no trend (null hypothesis), the regression gradientshould have a value near to zero. If there is alarge trend in the mean (alternative hypothesis),the value of the regression gradient would be verydifferent from zero: positive for an increasing trendand negative for a decreasing trend.The significance level measures whether the teststatistic differs significantly from the range of valuesthat would typically occur under the null hypothesis.It is the probability that a test erroneouslydetects trend when none is present; this is referredto as a type I error. A type II error occurs when thenull hypothesis is accepted – no trend is present –when in fact the alternative hypothesis is true (atrend exists). The power of a test is the probabilityof correctly detecting a trend when one is present;powerful tests that have a low type II error probabilityare preferred.In carrying out a statistical test it is always necessaryto consider assumptions. Standard tests requiresome or all of the following assumptions: a specifiedform of distribution, for example, assumingthat the data are normally distributed; constancy ofthe distribution in that all data points have an identicaldistribution so that there are no seasonalvariations or any other cycles in the data; and independence.This last assumption is violated if thereis either autocorrelation, namely, correlation fromone time value to the next. This is also referred to asserial correlation or temporal correlation or, in thecase of a multi-site study, spatial correlation, inparticular, correlation between sites.If the assumptions made in a statistical test are notfulfilled by the data, then test results can be meaningless,in the sense that estimates of the significancelevel would be grossly incorrect. Hydrological dataare often clearly non-normal; this means that teststhat assume an underlying normal distribution willbe inaccurate. Hydrological data may also showautocorrelation and/or spatial correlation; therefore,data values are not independent. This can havea negative impact on the ability to detect trend in atime series (Yue and others, 2003). The data mayalso display seasonality, which violates assumptionsof constancy of distribution. The power of tests candepend on the sample size, the variability of thetime series, the magnitude of the characteristic thatis being tested, such as a trend, and the distributionand skewness of the time series. Results of the powerof the Mann–Kendall and Spearman’s rho tests areprovided by Yue and others (2002) and Yue andPilon (2004).The main stages in statistical testing are as follows:(a) Decide what type of series/variable to test,depending on the issues of interest, for example,monthly averages, annual maxima ordeseasonalized data;(b) Decide what types of change are of interest(trend/step change);(c) Check out data assumptions, for instance byusing exploratory data analysis;(d) Select one or more tests/test statistics that areappropriate for each type of change; more thanone is good practice;(e) Select a suitable method for evaluating significancelevels;(f) Evaluate significance levels;(g) Investigate and interpret results.The process of selecting a statistical test can beconsidered to be composed of two parts: selectingthe test statistic and selecting a method for determiningthe significance level of the test statistic. Byviewing the process in manner, it becomes possibledistinguish between how to select a test statisticand how to evaluate the significance level.6.2.2.3 Distribution-free testingThere are many ways of testing for trend or otherchanges in hydrological data. In a particular groupof methods, referred to as distribution-free methods,there is no need for assumptions as to the formof distribution from which the data were derived,for example, it is unnecessary to assume data arenormally distributed. The following approaches aredistribution-free:(a) Rank-based tests: These tests use the ranks ofthe data values but not the actual data values. Adata point has rank r if it is the rth largest valuein a dataset. Most rank-based tests assume thatdata are independent and identically distributed.Rank-based tests have the advantage thatthey are robust and usually simple to use. Theyare generally less powerful than a parametricapproach.(b) Tests using a normal-scores transformation:Many tests for change rely on the assumptionof normality. They are generally not suitablefor direct use with hydrological data, which aretypically far from being normally distributed.However, such tests can be used if the data arefirst transformed. The normal scores transformationresults in a dataset that has a normaldistribution. It is similar to using the ranks of

II.6-16GUIDE TO HYDROLOGICAL PRACTICESa data series, but instead of replacing the datavalue by its rank, r, the data value is replaced bythe typical value that the rth largest value froma sample of normal data would have (the rthnormal score). The advantages of using normalscores are that the original data need not followa normal distribution, and the test is relativelyrobust to extreme values. The disadvantage isthat statistics measuring change, such as theregression gradient, cannot be easily interpreted.Normal-scores tests are likely to beslightly more powerful than equivalent rankbasedtests.(c) Tests using resampling approaches: Resamplingmethods, introduced below, are methods thatuse the data to determine the significance of atest statistic.6.2.2.4 Introducing resampling methodsResampling methods, permutation testing and thebootstrap method are a robust set of techniques forestimating the significance level of a test statistic.They are flexible and can be adapted to a wide rangeof types of data, including autocorrelated or seasonaldata, and are relatively powerful. Resampling methodsare very useful for testing hydrological databecause they require relatively few assumptions tobe made about the data, yet they are also powerfultests. They provide a flexible methodology thatallows significance levels to be estimated for anysensible choice of test statistic. They enable traditionalstatistical tests to be adapted for applicationto hydrological series by using a robust method todetermine significance.The basic idea behind re-sampling methods is verystraightforward. Consider testing a series for trend:a possible test is the regression gradient. If there isno trend in the data (the null hypothesis) then theorder of the data values should make little difference.Thus shuffling, or permuting, the elements ofthe data series should not change the gradientsignificantly. Under a permutation approach thedata are shuffled very many times. The test statisticis recalculated after each shuffle or permutation.After many permutations, the original test statisticis compared with the generated test statistic values.If the original test statistic differs substantially frommost of the generated values, this suggests that theordering of the data affects the gradient and thatthere was trend. If the original test statistic liessomewhere in the middle of the generated values,then it seems reasonable that the null hypothesiswas correct in that the order of the values does notmatter; hence there is no evidence of trend. In otherwords, if an observer or, in this case, the statisticaltest can distinguish between the original data andthe resampled or permuted data, the observed dataare considered not to satisfy the null hypothesis.The bootstrap and permutation methods are twodifferent approaches to resampling the data. Inpermutation methods, sampling with no replacement,the data are reordered, that is each of thedata points in the original data series appears onlyonce in each resampled or generated data series. Inbootstrap methods, the original data series issampled with replacement to give a new series withthe same number of values as the original data. Theseries generated with this method may containmore than one of some values from the originalseries and none of other values. In both cases, thegenerated series has the same distribution as theempirical, observed distribution of the data. Ingeneral, bootstrap methods are more flexible thanpermutation methods and can be used in a widerrange of circumstances.The simplest resampling strategy is to permute orbootstrap individual data points, as describedabove. This technique is applicable only when itcan be assumed that the data are independent andnon-seasonal. If data show autocorrelation, or additionalstructure such as seasonality, the seriesgenerated by resampling should replicate this structure.A straightforward means of achieving this is topermute or bootstrap the data in blocks. For example,for a 40-year series of monthly values, it wouldbe sensible to treat the data as consisting of40 blocks of one year. Each year’s worth of data isleft intact and is moved around together as a block,thus maintaining the seasonal and temporaldependencies within each year. The 40 blocks arethen reordered many times. In this way, the resampledseries will preserve the original seasonality.Similarly, blocks can be forced to replicate the autocorrelationin the data. It is important that the sizeof the blocks should be sensibly selected.Many distribution-free tests, such as the rank-basedtests, depend on assumptions of independence. Ifthis assumption does not hold, as is common forhydrological data, the recommended approach is toextract the test statistics from these tests and toevaluate significance using block-bootstrap andblock-permutation methods, rather than using classicalformulae for significance, which may lead togross errors. Such methods can be useful when thereis spatial dependency in a set of multi-site data thatis to be tested as a group. In this case, the usualchoice of blocks would be to group data across allsites that occurred in the same time interval (forexample, Robson and others, 1998).

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-176.2.2.5 Commonly used tests and teststatisticsTable II.6.1 presents a summary of standard parametricand non-parametric tests for changedetection, their essential properties and necessaryassumptions. The tests are described in their standardform, namely, in a non-resampling framework.Each of these tests can easily be adapted to be aresampling test. For this, the test statistic for a testis calculated, but the significance level isobtained using the resampling approach describedabove. Guidelines for test selection are shown inTable II.6.2.Note that if resampling techniques are to be used, itis possible to construct new test statistics to test fora particular type of change – it is not necessary toselect test statistics from known tests. Having theflexibility to construct custom test statistics allowsgreat flexibility in what can be tested and for what.When interpreting test results, it is necessary toremember that no statistical test is perfect, even ifall test assumptions are met. Hence, it is recommendedto use more than one test. If several testsprovide significant results, this provides strongerevidence of change, unless they are very similar, inwhich case multiple significance is not an additionalproof of change.It is important to examine the test results alongsidegraphs of the data and with as much historicalknowledge about the data as possible. For example,if both step-change and trend results aresignificant, further information will be needed todetermine which of these provides the best descriptionof the change. If historical investigationsreveal that a dam was built during the period, andthis is consistent with the time series plot, it wouldbe reasonable to conclude that the dam caused astep change.If test results suggest that there is a significantchange in a data series, it is important to try tounderstand the cause. Although the investigatormay be interested in detecting climate change,there may be many other possible explanations,which need to be examined (Kundzewicz andRobson, 2004). It can be helpful to look out forpatterns in the results that may indicate furtherstructure, such as regional patterns in trends.6.2.3 Spatial analysis in hydrologyHydrological variables may form a spatial-temporalrandom field, for example, a set of time series ofvalues of a variable for a number of gauges. A spatialfield describes discrete observations of a variable inthe same time instant in a number of spatial pointsor remotely sensed data covering the whole area.The spatial aspects of random fields such as rainfall,groundwater level or concentrations of chemicalsin groundwater, are important issues in hydrology.Geostatistics is a set of statistical estimation techniquesfor quantities varying in space. It lends itselfwell to applications to random spatial fields, such asprecipitation or groundwater quality, thus beingapplicable to a range of hydrological problems (seeKitanidis, 1992). Geostatistics offers solutions toseveral practical problems of considerable importancein hydrology. It can be used in interpolation,such as estimating a value for an ungauged location,based on observations from several neighbouringgauges, or plotting a contour map based on scarceinformation in irregularly spaced locations. It cansolve aggregation problems: finding areal estimatesbased on point observations such as determiningareal precipitation from point values. It can aid inmonitoring network design, for example, in optimalnetwork extension or, unfortunately more common,optimal network reduction. These applicationsanswer the following question: how to reduce thenetwork while minimizing information loss. By usinggeostatistics with groundwater flow or transportmodels, the inverse problem of parameter identificationcan be solved by determining transmissivityfrom observed hydraulic head, for example.A statement of the principal problem of the geostatisticalkriging technique can be formulated aslooking for a best linear unbiased estimator (BLUE)of a quantity at some unmeasured location x 0fromobservations z(x 1), z(x 2), …, z(x n) in a number oflocations x 1, x 2, …, x n:nZ ( x 0 ) = ∑ λ i z ( x i )(6.20)i=1where Z(x 0) is the estimator of z(x 0) and λ iareweights.Under the so-called intrinsic hypothesis, the estimationvariance can be expressed with the help ofa mathematical equation containing weights fromequation 6.20 and values of a semivariogram. Setsof weights are sought which provide an optimalestimate in the sense that the estimation variance isa minimum. An important advantage of kriging isthat it provides, not only the estimated value, butalso evaluates the estimation variance. This usefultechnique originates from mining, where observationsare costly, and hence scarce, and optimal

II.6-18GUIDE TO HYDROLOGICAL PRACTICESorganization of the available knowledge is ofprimary importance.Today geostatistics has become an importantelement of distributed modelling in the geograpicalinformation system environment and an option ininterpolation packages.6.3 MODELLING HYDROLOGICALSYSTEMS AND PROCESSES[HOMS J04, K22, K35, K55, L20, L30]6.3.1 IntroductionThe hydrological cycle refers to the circulation ofwater in the world. It is composed of a number ofwater fluxes between different water storages.Examples of water fluxes within hydrological processesare liquid or solid precipitation, infiltration,runoff, snowmelt, river flow and evapotranspiration.Examples of the corresponding water storagesare atmosphere; land surface such as depressions,ponds, lakes and rivers; vegetation; soil; aquifersand snow cover.All hydrological processes and hydrological systemshave been described by mathematical equations,some of which were derived from rigorous physicallaws of conservation of mass and momentum.Others are either of conceptual nature, or a blackbox type. A comprehensive review of mathematicalequations of use in dynamic hydrology can befound in Eagleson (1970). The present sectioncontains a few illustrative examples related toTable II.6.1. Comparison of parametric and non-parametric tests for change detection, their propertiesand the assumptions made (after Kundzewicz and Robson, 2004)Test name What it does Properties and assumptions madeTests for step changeMedian change-point test/Pettitt’stest for changeMann-Whitney test/rank-sum testDistribution-free CUSUM (maximumcumulative sum) testTest that looks for a change in themedian of a series with the exact timeof change unknownTest that looks for differences betweentwo independent sample groups,based on the Mann-Kendall teststatisticTest in which successive observationsare compared with the median of theseries with the maximum cumulativesum of the signs of the difference fromthe median as the test statisticPowerful rank-based test, robust tochanges in distributional formRank-based testRank-based testKruskal-Wallis test Tests equality of sub-period means Rank-based testCumulative deviations and otherCUSUM testsStudent’s t-testWorsley likelihood ratio testTests for trendSpearman’s rhoKendall’s tau/Mann-Kendall testLinear regressionTest works on rescaled cumulativesums of the deviations from the meanTests whether two samples havedifferent means – assumes normallydistributed data and a known changepointtimeSuitable for use when the change-pointtime is unknownTest for correlation between time andthe rank seriesSimilar to Spearman’s rho, but uses adifferent measure of correlation withno parametric analogueUses the regression gradient as a teststatisticNote: All the tests make the assumption that the data are identically distributed and independent.Parametric test, assumption of normaldistributionStandard parametric test, assumptionof normal distributionSimilar to Student’s t-test, assumptionof normal distributionRank-based testRank-based test – extended testsallowing for seasonality exist, forexample, Hirsch and Slack (1984)– and autocorrelationOne of the most common testsfor trend, assumption of normaldistribution

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-19rainfall runoff, flow routing, groundwater, waterquality, and snow and ice phenomena.Hydrological modelling is contributing more andmore to integrated models. Beyond simulatinghydrological runoff, integrated models include soilerosion, river sediment, ecohydrology, crop yield,and interfaces with other disciplines, such as ecohydrology,climate impact assessment and watermanagement.6.3.2 Rainfall–runoff relationships6.3.2.1 GeneralRainfall–runoff relationships are used primarilyfor design, forecasting and evaluation. If streamflowdata are unavailable or are too limited forreliable interpretation, rainfall–runoff relationshipscan be very helpful because of their ability toextract streamflow information from the precipitationrecords. Because of the relative simplicity andinexpensive nature of the collection of rainfalldata, they are generally more abundant than arestreamflow data. If a strong relationship can beestablished between the rainfall and runoff for acatchment of interest, the combination of therainfall–runoff relationship and the rainfall datamay, for example, give more reliable estimates ofthe frequency of high streamflows than either aregional flood relationship (see Chapter 5) or anextrapolation of meagre streamflow data from thecatchment.In general, rainfall–runoff relationships are developedin two distinct steps: the determination of thevolume of runoff that results from a given volumeof rainfall during a given time period and the distributionof the volume of runoff in time. The firststep is necessary because of the partitioning of rainfallamong evapotranspiration, infiltration andrunoff (see Volume I, Chapter 4). The second step isrequired to account for the travel time and theattenuation of the wave of runoff that is generatedby the rainfall. Discussion of these two steps constitutesthe remainder of this chapter.6.3.2.2 Runoff volumes6.3.2.2.1 Antecedent precipitation indexThe antecedent precipitation index has been developedprimarily for river forecasting and is appliedover a wide range of drainage areas and conditions.Its derivation for a particular drainage area requiresobserved rainfall and runoff data over some timeinterval. It is defined as:I t= I ok t + ΣP ik t(i) (6.21)where I ois the initial value of the index, k is a recessionfactor, t is the time interval for the computation,P iis the number of daily rainfalls that have occurredduring the time interval and t(i) is the number ofdays since each day with precipitation.It is often convenient to use simplified forms of theantecedent precipitation index. One or more of thevariables may have a negligible influence in certaincatchments and it is then possible to reduce thenumber of these variables. However, in all cases thegeneral method is the same.The effects of vegetative cover, soil type and otherimportant catchment characteristics, as well as thetime of year, are reflected in the recession factor.Time of year is expressed as a family of curves representingthe seasonal trend of solar energy, vegetativecondition and other factors that influence the evaporationand transpiration of moisture in thecatchment. The antecedent precipitation index isan expression of the moisture in the catchment andthe moisture retention in the soil.Figure II.6.6 illustrates an example of behaviour ofthe antecedent precipitation index for a dailyTable II.6.2. Test selection guidelinesCase(a) Data are normally distributed andindependent.(b) Data are non-normal, but areindependent and non-seasonal.(c) Data are non-normal, and are notindependent or are seasonal.Which test to selectThis is an unlikely scenario for hydrological data. If applicable, any of the testslisted in Table II.6.1 should be suitable.Any of the distribution-free tests are suitable. Tests that are based on normalityassumption can also be applied by first applying a normal scores or rankstransformation, or by using a relevant test statistic and evaluating significanceusing resampling techniques.The data do not meet the assumptions for any of the basic tests given above. Itis necessary to extract the test statistic and to evaluate significance levels usingblock-permutation or block-bootstrap methods.

II.6-20GUIDE TO HYDROLOGICAL PRACTICESPrecipitation or antecedent index (mm)605040302010Antecedent precipitationindex (K = 0.90)Dailyprecipitation0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20DaysFigure II.6.6. Antecedent precipitation indexrecession factor of 0.9. The antecedent precipitationindex can be computed from averageprecipitation for several stations or individually foreach station in a drainage area. The latter is oftenpreferable.Figure II.6.7 illustrates the method of estimatingrunoff volume from rainfall and the antecedentprecipitation index. The dashed lines and arrowsdemonstrate the use of this diagram. For example,the diagram is entered with a value of 22 mm forthe antecedent precipitation index. The long dashesand arrows lead to the month of July and down tostorm duration of 24 hours. The example thenproceeds to the right to the assumed storm rainfallof 40 mm and up to a runoff of 16-mm averagedepth over the drainage area.If the hypothetical storm in the foregoing examplehad occurred in February, with other conditionsbeing the same, the effect of 22-mm antecedentprecipitation would be different. Ordinarily, inFebruary as contrasted with July, the same amountof antecedent precipitation would have left thesoil nearly saturated because of dormant vegetationand less evapotranspiration in winter. Theshort dashes and arrows in Figure II.6.7 show thatthe runoff from the 40-mm rain in the secondexample would be 30 mm, hence nearly twice ashigh as in July.Frozen ground and accumulations of snow requirespecial consideration in estimating antecedentmoisture conditions. With frozen ground, thetime-of-year curve that gives the maximum runoffis commonly used. The influence of snow on theground is properly expressed in terms of theamount and rate of melting, instead of the totalaccumulation. Snowmelt is discussed in 6.3.5.Time of yearJulyAugustMayJuneSeptemberAprilOctoberJanuaryNovemberDecemberMarchFebruary605040302010Antecedent precipitation index (mm)Storm runoff (mm)10 20 30 40 50 60Storm duration (hours)102030405060Storm rainfall (mm)024487296120Figure II.6.7. Using the antecedent precipitation index to estimate rainfall runoff

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-216.3.2.2.2 Initial base flow as an index to runoffvolumeIn humid areas, where streams do not often go dry,base flow in the form of groundwater discharge, atthe beginning of a storm is often used as an indexof the initial basin conditions. An example of sucha relationship is shown in Figure II.6.8. Base-flowdischarge reflects conditions throughout the entirearea. In some areas, it is found necessary to varythis relationship with season. A common method isto develop one relationship for summer and one forwinter, which leads to the inevitable problem ofstorm events occurring between seasons. The usualsolution is to make an estimate of runoff on thebasis of each curve and then to interpolate.The use of initial groundwater discharge as an indexto runoff conditions is usually limited to smallbasins with short times of concentration. In largerareas during a rainy season, one rise of thehydrograph tends to be superimposed on the last,which makes a determination of initial groundwaterdischarge difficult. The usual approach is todetermine initial groundwater discharge for smallindex basins and to apply them to other nearbyareas having similar hydrological characteristics.6.3.2.2.3 Moisture-accounting techniquesSoil moisture deficiency is probably the mostimportant factor involved in the relationshipbetween rainfall and runoff. A practical means ofestimating initial soil moisture deficiencies for anarea would provide a very useful variable for inclusionin a procedure for correlating storm rainfall toresultant runoff. Instruments for measuring soilmoisture for a specific soil profile have becomereasonably practical, but the wide variety of soilprofiles and moisture conditions that exist in evena small basin makes point measurements of soilStorm runoff (cm)321011975Initial groundwater discharge incubic metres per second32 4 6 8 10 12 14 16Storm precipitation (cm)Figure II.6.8. Base flow as an index to rainfall–runoff relationshipmoisture of questionable value in a rainfall–runoffrelationship.A more promising approach is the use of an arealaccounting technique that results in soil moisturevalues related to the entire area. In such an approach,precipitation is the inflow and outflow consists ofrunoff leaving the area by the stream channels plusevapotranspiration into the atmosphere from soiland plant surfaces. The means of estimating theprecipitation over the area is the usual problem ofderiving spatial averages from point values. Runofffrom the area can be determined from streamflowrecords. The problem becomes one of matchingflow to the particular storm that caused it (see6.3.2). The difference, rainfall minus runoff, is thewater that remains in the area and is referred to asrecharge, R c.The third element, evapotranspiration, is the mostdifficult to evaluate because its direct measurementis extremely difficult. Most soil moisture accountingtechniques are based on the premise that actualevapotranspiration bears a simple relationship topotential evapotranspiration, ET p, and soil moisturedeficiency.A simple form of soil moisture accounting is one inwhich the soil profile is considered to have onecapacity, S, over the entire area. Soil moisture deficiency,DU s, is then determined by the followingequation:DU s(t + 1) =0 if DU s(t) – R c+ ET ≤ 0DU s(t) – R c+ ET if 0 < DU s(t) – R c+ ET < SSif DU s(t) – R c+ ET ≥ S(6.22)where DU s(t) is the soil moisture deficiency at timet, DU s(t + 1) is the value one time period later, R cisthe recharge resulting from precipitation and/orsnowmelt) and ET is the evapotranspiration thatoccurs between times t and t+1. The deficiencyvaries between the limits of zero and S.This approach can be made more realistic by multiplyingthe evapotranspiration by the ratio(S – DU s(t))/S, which acknowledges that actualevapotranspiration decreases along with the supplyof available moisture in the soil profile.Another possible modification would divide thesoil profile into layers. In this approach, it is assumedthat the upper-layer moisture must first be depleted

Stream discharge (m 3 s –1 )II.6-22GUIDE TO HYDROLOGICAL PRACTICESbefore any depletion of the lower layer, and,conversely, recharge to the lower layer is limited tooverflow from the upper layer.The application of soil moisture accounting valuesin a rainfall–runoff relationship can be made byrelating runoff, Q, to discharge computed in theaccounting:Q = cQ U+ (1 – c)Q L(6.23)where c is a constant, Q Uis the computed runofffrom the upper layer, and Q Lis the computed runofffrom the lower layer.6.3.2.2.4 Temporal distribution of runoffTo account for the travel time and the attenuationof a volume of water imposed on the catchment bya rainfall event, an accounting through time at thecatchment outlet must be performed. This step isusually accomplished by the use of a unithydrograph, which describes the temporal distributionof runoff leaving the catchment. The unithydrograph is constrained by the principle of continuityof mass in the following manner:V = ∫Q(t)dt (6.24)where Q(t) is the instantaneous discharge rate, t istime, and V is the runoff volume. The function Q(t)defines a curve whose shape correctly representsthe catchment characteristics. To comparehydrographs of different catchments and to assistin the preparation of synthetic hydrographs, deterministicmodels have been developed that relatethe hydrograph characteristics to hydrological andmeteorological data. These models are discussedbelow.the direct or storm runoff associated with a particularstorm. Another major component is thestreamflow persisting from previous contributionsto flow. The third major component is the flow fromthe earlier storms that is delayed by passing throughthe ground. A portion of that component is knownas interflow, that is, water passing through the soilwith little delay, and is often included as part ofdirect runoff. Some of the more recent conceptualmodels for the continuous simulation of streamflowhave provisions for computing each of the abovecomponents separately.This type of analysis does not allow identificationof each component by inspecting the observedhydrograph. In less complex methods of analysis inwhich only two components are recognized, it ispossible to separate the observed hydrograph andevaluate the magnitude of the two components. Inthe following illustration, direct runoff includesboth surface runoff and interflow.One of the simplest of many methods for separatinga hydrograph into its major components isillustrated in Figure II.6.9. The trace of base flow isextrapolated (see line segment AB) to the time ofpeak flow by extending its trend prior to the streamrise. From point B, a straight line is drawn to intersectthe hydrograph at point C a fixed time later.The time in days from B to C is determined largelyby the size of the drainage area. It is generally about(A/2) 0.2 , where A is drainage area in squarekilometres.Several methods of hydrograph separation arecommonly used. The same technique should beused in both application and development, arequirement that is more important than themethod, however.6.3.2.2.5 Unit hydrographThe unit hydrograph for a catchment is defined asthe discharge hydrograph resulting from a unit ofeffective rainfall generated uniformly over thecatchment at a uniform rate during a specifiedperiod of time. In application, the unit hydrographis assumed to be time invariant. It is further assumedthat events with runoff volumes other than oneunit produce hydrographs that are proportional tothe unit hydrograph.6.3.2.2.6 Derivation from streamflow recordsTo determine the volume of runoff from a particularrainstorm, it is necessary to separate the hydrographinto its pertinent components. One component is40302010012-houreffective rain+ E ABObserved hydrographHydrograph ofdirect runoffUnit hydrographFigure II.6.9. Hydrograph analysisCBase flow1 2 3 4 5 6 7 8 9 10 11Time (days)D+

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-23The total runoff associated with a particular stormor stormy period can be determined by the followingmethod. In Figure II.6.9, the area under thehydrograph between times A and D is the stormrunoff because the beginning and ending pointsrepresent the same groundwater recession conditionsand, therefore, the same storage.Figure II.6.9 illustrates the essential steps for derivinga direct runoff unit hydrograph from observeddata. These steps may be performed either graphicallyor numerically. The hydrograph of directrunoff is the flow in excess of the trace ABC. Thevolume of direct runoff is obtained by integratingthe area under the hydrograph. If a planimeter isnot available, a convenient method is the countingof squares. In this hypothetical example, the volumeof direct runoff is found to be 4 320 000 m 3 . Overan assumed drainage area of 200 km 2 , this volumerepresents an average depth of 2.16 cm. To obtainthe unit hydrograph, it is necessary to divide eachordinate of the direct runoff hydrograph by 2.16.The hydrograph thus determined shows the shapeof the hydrograph that would result from one centimetreaverage depth of direct runoff over thedrainage area, that is, the unit hydrograph.In the records of some catchments, it is difficult tofind unit or single storms that produce stream risesuncomplicated by other events. In such cases, thederivation of a unit hydrograph becomes morecomplex. One method of deriving a unit hydrographunder these circumstances is to assume an initialunit hydrograph, and to reconstruct the hydrographsof direct runoff for several storms using estimatedrunoff increments and to refine the unit hydrographby successive approximations as indicated by theresults. This reconstruction method is shown inFigure II.6.10 and by:hydrograph shape and allow a forecast to be madebefore too large a time increment has elapsed. Fordrainage areas larger than about 2 000 km 2 , unithydrographs of larger time increments may beused, but as a rule, unit hydrographs should beapplied to tributary areas and may be combinedby routing.As might be expected from considerations of channelhydraulics, there is a tendency for the peakednessof unit hydrographs to increase with the magnitudeof runoff. Accordingly, in practical applications, afamily of unit hydrographs may be used for a particularcatchment area, with higher peaked unithydrographs for the cases with large amounts ofrunoff and flatter peaks for the lesser amounts ofrunoff. Often only two categories comprise thefamily.Skill in the use of unit hydrograph is acquired fromstudy and practice. For other methods than thosedescribed in this section, and for refinements, referencemay be made to textbooks and handbooks ofagencies that routinely use unit hydrographs intheir regular operations.6.3.2.2.7 Derivation by synthetic methodsIt is often necessary to plan constructions or operationsfor ungauged streams. In such cases, it ishelpful to develop synthetic unit hydrographs(Dooge, 1973). A commonly used derivation of aunit hydrograph is the procedure derived by Snyderin which a large number of basins and unithydrographs were analysed to derive relationshipsbetween the shape of the unit hydrograph and theq n+ Q nU 1+ Q n–1U 2+ Q n–2U 3+ ...(6.25)300+ Q n–i+1U i+ ...+ Q 1U nwhere q nis the rate of discharge from direct runoffat time n, U iis the i-unit hydrograph ordinate andQ n–i+1is the direct runoff for the i th interval. Thisequation can also be used as the regression modelfor unit hydrograph derivation by least squares.For drainage areas of 200 to 2 000 km 2 , timeincrements of six hours are commonly used forunit hydrograph development, but for higheraccuracy, shorter time intervals may be employed.Smaller drainage areas may also require shortertime increments. The time increments should besmall enough to give good definition of theDirect runoff (m 3 s –1 )2001000U 4U U 5U 1QU 31 U 3U 2UU 4 3U 4 U 5U 1 Q 2Q 3U 2U 2U 112 24536 48Time (hours)Figure II.6.10. Reconstruction of direct runoffhydrograph

II.6-24GUIDE TO HYDROLOGICAL PRACTICESobjective physical characteristics of the drainagebasin.The important parameters in the shape of a unithydrograph are its peakedness, the length of itsbase and the basin lag, which may be defined invarious ways; here, however, it is the time from thecentroid of rainfall to the peak of the hydrograph.In Snyder’s method, the basin lag, t p, is given inhours as:t p= C 1(ll c) n (6.26)where C 1converts units and is an empirical coefficient,l is the length of the main stream inkilometres, l cis the distance in kilometres from thecentroid of the drainage area to the outlet and n isan empirical exponent.For peakedness of the unit hydrograph, this methoduses a standard duration of rain, t p/C 2, with C 2beingderived empirically. For rains of this duration:Q p= C 3A/t p(6.27)where Q pis peak rate of runoff in m 3 s –l , C 3is anempirical constant, A is drainage area in km 2 andlag t pis in hours. The time base in days T bis asfollows:T b= d + C 4t p(6.28)The constants d and C 4are fixed by the procedureused to separate base flow from direct runoff.For durations T Rother than the standard durationof rain, the corresponding lag, t c, is the following:t c= t p+ f (T R) (6.29)here f(T R) is a function of duration.Snyder’s coefficients were derived for streams in theAppalachian Mountains of the United States. Thegeneral method has been found applicable in otherregions, but different coefficients are to be expectedfor different types of topography, geology andclimate.Rodriguez-Iturbe and Valdes (1979) developed aphysically based methodology for synthesizing aninstantaneous unit hydrograph with the help ofempirical laws of geomorphology and climaticcharacteristics. They proposed the geomorphologicinstantaneous unit hydrograph, later knownas the geomorphoclimatic instantaneous unithydrograph. They also developed equations forthe value of peak and time to peak of the geomorphologicinstantaneous unit hydrograph asfunctions of the bifurcation ratio, length ratio,area ratio, length of highest order stream and flowvelocity.6.3.2.2.8 Conversion of unit hydrographdurationsA suitable rainfall of unit duration is rarelyobserved. Variations of rainfall in time and spaceproduce different hydrographs, though the totalamount and duration of the rain may be exactlythe same. Thus, the derivation of a general unithydrograph requires an averaging of several unithydrographs.One technique for generalizing unit hydrographs isby comparison of unit hydrographs of differentdurations. If a unit hydrograph of duration t hoursis added to itself, lagged t hours, and the ordinatesdivided by two, the result is a unit hydrograph for2t hours. Similar conversions are evident.A broader application of this basic idea for manipulatingunit hydrographs is known as the summationor S-curve method. The S-curve is the hydrographthat would result from an infinite series of runoffincrements of one centimetre in t hours. It isconstructed by adding a series of unit hydrographs,each lagged T hours with respect to the precedingone. With a time base of T hours for the unithydrograph, a continuous rain producing onecentimetre of direct runoff per t hours woulddevelop a constant outflow at the end of T hours.Thus, T/t hours would be required to produce anS-curve of equilibrium flow.Construction of an S-curve can be accomplished bya numerical, rather than a graphical, procedure. Aunit hydrograph for any duration t can be obtainedby lagging the S-curve t hours and obtaining ordinatesof lagged and unlagged S-curves. To obtainunit volume, these ordinates must be multiplied bythe ratio of the duration of the original unithydrograph to t hours.The instantaneous unit hydrograph is the unithydrograph whose time unit, t, is infinitely small.Construction of a t-hour unit hydrograph from aninstantaneous one is performed by means of anS-curve.6.3.2.2.9 Isochrone methodThe isochrone method is an expression of one ofthe first concepts of runoff from a basin. The runoff

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-25from different portions of a drainage basin arrivesat a point in the stream at different times. The firstwater to leave the basin during a stream rise generallycomes from the area nearest the catchmentoutlet. Later, water comes from larger areas in thecentral portion of the basin, and finally, watercomes from remote portions of the drainage area.Thus, the drainage basin may be divided into zonesfrom which the water arrives sequentially at themeasurement point. The lines dividing these zonesin Figure II.6.11(a) are called isochrones. The distributionof the isochronal areas, the time-areadistribution, is considered to be constant for a givenbasin for all flood hydrographs.To compute this distribution, it is necessary first tocompute or assume an average travel time or averagevelocity of streamflow. The isochrones aredrawn on a map of the basin according to the averagevelocity of flow in the channel or average traveltime. The area of each zone is then determined byusing a planimeter and the values are plotted againstthe corresponding time lag (Figure II.6.11(b)).The time-area distribution is indicative of thehydrograph for uniform rainfall of unit duration,Δt, the time difference between isochrones. If thereare several periods of rainfall, each resulting invarying quantities of runoff over the differentzones:delineate the rainfall pattern reliably. This is anadvantage over the unit hydrograph describedpreviously.6.3.3 Groundwater modelling6.3.3.1 General modelling considerationsGroundwater is increasingly becoming an importantsource of water for mankind as surface waterresources become more depleted through theincreased effects of abstraction and pollution.Because groundwater is a concealed and obscureasset, its conservation and management is expensiveand scientifically challenging owing to the lackof evidence, knowledge and understanding of itslocation, quantity and character. Consequently, inorder to assess its extent, volume and quality and inturn develop, manage and protect it, it is necessaryto construct representative modelled scenarios toexamine its potential and reliable volume and quality.This section provides a summary of the scienceand methods employed in hydrogeological modellingwithin a practical context of which the principalQ tΔt = A 1V t+ A 2V t–1+ A 3V t–2+ ... + A cV t–c+1(6.30)where Q tis the average discharge during the period,Δt, ending at time t, A is the time-area histogramordinate at that period and V tis the zonal runoffduring the same period. Care must be taken toensure consistent units. Figure II.6.11(c) illustratesthe computation of the resultant hydrograph withthree periods of uniform runoff from thecatchment.T = OA rAA 2 A 3 AT 1= A 4 A 5 = A cT = 5 = cT = 1c = time of concentrationT = 2 = 1 + 4t T = 3T = 4(a) Basin map with isochronesThe resultant hydrograph reflects the lag characteristicsof the catchment. Since the actual hydrographwould be affected by channel storage, thehydrograph computed from equation 6.30 shouldbe routed through storage. Any of the several routingtechniques described in the literature can beused. Two such techniques are described in 6.3.5. Itis usually found to be advantageous to adjust theisochrones and routing parameters by trial anderror to obtain the best combination for simulationof observed hydrographs.The isochrone method allows non-uniform distributionsof rainfall to be taken into account whenthere are enough raingauges in the basin toA 4A 3AA 2 A 51T1 2 3 4 5(b) Time-area distributionA V 2 3Q r A V 3A V 2 3 3A V 1 3A V 2 3A V 4 3 AA V 4 34 VA V 1 1 2A V A V A 3 V 12 1A V 5 11A V A V 5 315 21 2 3 4 5 6 7(c) Resultant hydrographFigure II.6.11. Isochrone method

II.6-26GUIDE TO HYDROLOGICAL PRACTICESelements are process, development, control andconservation.6.3.3.2 Development of a conceptual modelTo adequately represent a hydrogeological regime,there are a significant number of characteristics thatmust be replicated by using a model. These modelelements comprise several representations forconsideration (Bear, 1980, 1988).The type and detail of the conceptual model willdepend on the scale, amount of time and resources– availability of data, technical expertise, staffresources, computing facilities – assigned to thetask, as well as the quality of the decision-makingprocess, professional risks and legal and statutoryframework.Conceptual modelling is continuous and cyclical;therefore, a tiered approach from basic to intermediateto detailed is appropriate. The assumptionsincluded in the conceptual model should relate tothe issues described below.6.3.3.3 Development of a mathematicalmodelThe principal elements of the model include thefollowing components:(a) A definition of the geometry of the surfacesthat bound the domain;(b) Equations that express the balances of thecomponents, for example, mass of fluids, massof chemical species and energy;(c) Flux equations that relate the fluxes of thecomponents to the relevant variables of theproblem;(d) Constitutive equations, which define thebehaviour of the particular phases and chemicalspecies involved, for example, dependenceof density and viscosity on pressure, temperatureand solute concentration;(e) Sources and sinks, often referred to as forcingfunctions, of the component quantities.In terms of the modelling runs, the settings comprisethe following states:(a) Initial conditions that describe the known stateof the system at some initial time;(b) Boundary conditions that describe theinteraction of the considered domain with itsenvironment, that is, outside the delineateddomain, across their common boundaries.If a new numerical model and its associated codemust be used to solve the mathematical model thatis being employed, a strict verification procedureshould be undertaken to check that it is fit forpurpose through previous proven applications. Ifpractical, comparative scenarios should be runusing different codes.The groundwater regime is controlled by geologicaland climatic conditions and is exploited by man tomeet the needs of water development while theenvironmental requirements are met through theresidual balance. To assess the presence, extent andvariability of available groundwater resources, arange of investigation and testing mechanisms haveto be undertaken. These draw on a wide skills baseencompassing a range of Earth sciences, includinghydrometeorology, hydrology, pedology, geomorphology,petrology, geology and water chemistry.Groundwater constitutes a portion of the Earth’swater circulatory system known as the hydrologicalcycle with water-bearing formations of the Earth’scrust acting as conduits for the transmission and asreservoirs for the storage of water. Water entersthese formations from the ground surface or frombodies of surface water, after which it travels slowlyfor varying distances until it returns to the surfaceby the action of natural flow, plants or man. Thestorage capacity of groundwater reservoirs combinedwith slow flow rates can provide large and extensivelydistributed sources. Groundwater emerginginto surface water stream channels aids in sustainingstreamflows when surface runoff is low ornon-existent. Similarly, water pumped from wellsin many regions represents the sole water source inmany arid areas during much of the year.Water within the ground moves downwards throughthe unsaturated zone under the action of gravity,whereas in the saturated zone it moves in a directiondetermined by the hydraulic situation. Theprincipal sources of natural recharge include precipitation,streamflows, lakes and reservoirs. Dischargeof groundwater occurs when water emerges fromunderground. Most natural discharge occurs as flowinto surface water bodies, such as streams, lakes andoceans, and flow to the surface appears as a spring.Groundwater near the ground surface may returndirectly to the atmosphere by evaporation fromwithin the soil and by transpiration from vegetation.Pumpage from wells constitutes the majorartificial discharge of groundwater.Groundwater occurs in permeable geologicalformations known as aquifers that have a structurefacilitating the flow of water to take place undernatural conditions with aquicludes beingimpermeable formations that preclude the

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-27transmission of water movement. The portion of arock or soil that is not occupied by solid mineralmatter may be occupied by groundwater (Todd,2005). These spaces are known as voids, interstices,pores or pore spaces and are characterized by theirsize, shape, irregularity and distribution. Originalinterstices were created by geologic processesgoverning the origin of the geologic formation andare found in sedimentary and igneous rocks.Secondary interstices developed after the rock wasformed and include joints, fractures and solutionopenings. The porosity of a rock or soil is a measureof the contained interstices and is expressed as thepercentage of the void space to the total volume ofthe mass. If a is the porosity, then:a = 100w/V (6.31)where w is the volume of the water to fill, or saturateall of the pore space, and V is the total volumeof the rock or soil.Porosities can range from zero to 50 per cent,depending on the shape and arrangement of theindividual particles, size distribution and degree ofcompaction and cementation.There are a number of models that are used torepresent groundwater movement and transportphenomena. They include the following:(a) A physical representation using a scaled modelcomprising a medium through which a fluidis introduced and monitored by pressure andhead instrumentation;(b) An electrical representation in which head,flow and conductivity are represented by voltage,current and resistance;(c) A mathematical representation using a setof algorithms to represent the principalprocesses;(d) A stochastic analysis to characterize subsurfaceflow and transport modelling.In practice, most hydrogeological modellingcurrently used falls under (c) and (d) above. Withregard to transport phenomena relating to groundwatercontamination in which two- and three-phaseflow conditions occur, the use of mathematicalmodels is regarded as essential because of thecomplex scenarios that have to represented andanalysed.The movement of groundwater in its natural state isgoverned by established hydraulic principles. Theflow of water through aquifers can be expressed bya law derived by Darcy in 1856 that states that theflow rate through a porous media is proportional tothe head loss and inversely proportional to thelength of the flow path. Darcy’s law may expressedin general terms as:Q = KA dh/dL (6.32)where Q is the flow rate, K is the coefficient ofpermeability (sometimes referred to as hydraulicconductivity) and dh/dL is the hydraulicgradient. This relationship is shown inFigure II.6.12.Figure II.6.12. A representation of Darcy’s experiment

II.6-28GUIDE TO HYDROLOGICAL PRACTICESWith the nomenclature of this figure, Darcy’s lawtakes the following form:Q = KA h 1 – h 2(6.33)Lwhere, h (dimension: [L]) is the piezometric head:h = z +pρg(6.34)where z is the elevation of the point at which thepiezometric head is being considered above somedatum level, p and ρ are the fluid’s pressure and massdensity, respectively, and g is the gravity acceleration.The hydraulic conductivity, K, can then be expressedas follows:K = k ρgμ= kgv(6.35)where g is the gravity acceleration, and k (dimension:[L 2 ]) is the permeability or intrinsicpermeability of the porous medium. It is a coefficientthat depends solely on the properties of theconfiguration of the void space.Groundwater flow is an important aspect of hydrogeologyand is based on the principles that govern the flowof fluids through porous media. It requires a broadknowledge of fluid mechanics but cannot be adequatelydescribed within this brief outline of hydrogeology.However, the principal subdivisions of groundwaterflow can be summarized according to the dimensionalcharacter of the flow, the time dependency of the flow,the boundaries of the flow region or domain and theproperties of the medium and the fluid.6.3.3.4 Model operational optionsDifferent options can be selected when designing amodel:(a) The dimensionality of the model (one, two, orthree dimensions);(b) Steady state or time-dependent behaviour;(c) The number and kinds of fluid phases and therelevant chemical species involved;(d) The possibility of phase change and exchangeof chemical species between adjacent phases;(e) The flow regimes of the fluids involved, forexample, laminar or non-laminar;(f) The existence of non-isothermal conditionsand their influence on fluid and solid propertiesand on chemical-biological processes;(g) The relevant state variables and the areas orvolumes over which averages of such variablesshould be taken.All groundwater flow in nature is to a certain extentthree dimensional but the difficulty in solvinggroundwater flow problems depends on the degreeto which the flow is three dimensional. It is practicallyimpossible, however, to analyse a naturalthree-dimensional flow problem unless it can beexpressed in terms of a two-dimensional problemthat assumes that a degree of symmetry exists.Consequently, most solutions are based on assumingthat the problems being analysed are twodimensional or have special symmetry features.In general, groundwater flow is evaluated quantitativelybased on a knowledge of the velocity,pressure, density, temperature and velocity ofwater percolating through a geological formation.These water characteristics are often unknownvariables and may vary in space and time. If theunknown or dependent variables are functions ofonly the space variables, the flow is assumed to besteady; if the unknowns are also functions of time,the flow is considered to be unsteady or timedependent.The flow of groundwater in the space made up bythe water-filled pores – the aquifer – is dependenton the bounding surfaces of the medium, boundaries.If these boundaries are fixed in time and spacefor different states of flow, the aquifer is confined.However, if it possesses a free surface that varieswith the state of the flow, it is unconfined.Groundwater flow in the aquifer is controlled bythe nature, properties and isotropy of the medium.If the medium’s properties at any one point are thesame in all directions from that point, it is consideredto be isotropic; and if not, it is considered tobe anisotropic. The medium is considered to be ofheterogeneous composition if its nature, propertiesor conditions of isotropy or anisotropy varyfrom point to point in the medium, and homogeneousif its nature, properties and isotropic oranisotropic conditions are constant over themedium.Another subdivision is that of saturated and unsaturatedflow. Flow is saturated if the voids of themedium are completely filled with fluid in thephase of the main flow. The flow is unsaturated ifthis is not the case. Deep percolating groundwaterflow is always saturated, whereas above the saturatedmedium in the absence of overlyingimpermeable strata there is a zone of unsaturation.The boundary between these two zones is called thewater table or the phreatic surface. The latter zoneis occupied partially by air and partially by waterand is referred to as the unsaturated zone or the

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-29zone of aeration. It is comprised of an upper zone,which is referred to as the soil water zone, an intermediatezone and a lower zone, which is known asthe capillary zone.Water in the soil water zone exists at less thansaturation with a soil moisture deficit occurring,except when excessive water enters it as aresult of prolonged rainfall. The zone extendsfrom the ground surface through the major rootzone; its thickness depends on soil type andvegetation.The three principal hydraulic properties of an aquiferare porosity, which determines the volumestored, specific yield that controls the volume thatit yields either through natural drainage or when itis pumped, and permeability, which governs therate that water flows through it.A range of approaches to groundwater modeling arecontained in the literature. The most common arephysically based methods originating from rigorousphysics of flow in porous media. However, conceptual,and even black box methods are also used.The mathematics of flow and transport in unsaturatedand saturated porous media are relativelycomplex. There are a variety of physically basedequations, developed for different assumptions,and simplifications for a variety of configurations,including confined, leaky, and unconfined aquifers.A classical review of physically based approachescan be found in Eagleson (1970). See also Maidment(1992).The process of groundwater flow is governed bynon-linear partial differential equations in threedimensions, expressing conservation of mass, orcontinuity, conservation of momentum and a stateequation. Under the assumption of negligiblecompressibility of water and the porous medium,the isothermal unsteady laminar flow, with nosources or sinks of water, can be described by thefollowing partial differential equation (see Eagleson,1970):W + S s∂h∂t+∂∂y⎛⎝K y= ∂∂x∂h∂y⎞⎠ +⎛⎝K x∂∂z∂h∂x⎛⎝K z⎞⎠∂h∂z⎞⎠(6.36)where h is the piezometric head, K x, K y, and K zarethe hydraulic conductivities (for isotropic mediums,K x= K y= K z= K), W is a general term for sources andsinks of water, and S sis the specific storage. In thevariably saturated flow formulation, the unsaturatedhydraulic conductivity and specific moisturecapacity are functions of moisture content.An equation describing transport processes ingroundwater can be developed by usingequation 6.51. The advection-dispersion equationof transport is used to simulate quality aspects ofgroundwater flow, such as transport of solutes, inconservative and reactive cases.Physically sound partial differential equations ofgroundwater flow and transport form a core ofdistributed groundwater models in which thesolution is commonly achieved by finite-differenceor finite-element techniques. A review ofcomputer software for solving subsurface waterproblems has been carried out by Anderson andothers (1992). Among the codes included inHOMS component L20.2.04 is the modular finitedifferencegroundwater flow, or MODFLOW,model, a versatile software package developed bythe United States Geological Survey, which hasbeen commonly used worldwide in numerousapplications.The MODFLOW model (McDonald and Harbaugh,1988) simulates groundwater flow in a porousmedium in three dimensions, as well as modellingflow in two dimensions. A modular structure wasused for the program and documentation in orderto make the model easier to understand and modifywhen necessary. The code consists of a series ofpackages or modules that can be selected for a problemat hand. Modules include those for equationssolvers, stream, recharge, pumping and evapotranspiration.The application area of MODFLOWincludes steady-state and transient groundwaterflow, groundwater flow in confined, leaky andunconfined aquifers and a number of special flowproblems such as spring flow and flow to a well.Wells, rivers, drains, evapotranspiration andrecharge can be simulated and are represented ashead-dependent sources or sink terms in which thehead outside the model is user specified. MODFLOWcan be used in studies of interactions betweengroundwater and surface water interactions, such asflow to partially penetrating rivers and lakes. Aquiferhydraulic parameters, boundary conditions, initialconditions and stresses are required model input.The input is from text files with the data laid out ina prescribed order and format. The input data mustcorrespond to the specified grid structure. Theprimary model output is the head at each modelnode. In addition, a water budget is calculated, andthe flow through each model cell can be stored in adisk file. MODFLOW is probably the most widelyused groundwater model in the world.

II.6-30GUIDE TO HYDROLOGICAL PRACTICESThe MODFLOW package is designed for use byexperienced groundwater hydrologists. Useful preprocessorsand post-processors are available thatreduce the user’s efforts.6.3.3.5 PlanningTo undertake a groundwater modelling project, thefirst stage – and a very essential one – is to definethe purpose of the work. For larger projects, thismay require an initial scoping study to define therequirements and carry out an analysis that willidentify the aims of the project, and review previousstudies and available data. As well as definingthe objectives of the project and the principal tasksto be undertaken, scoping should also define themain outputs anticipated for the work.Planning also involves the identification of the typeof information the model is expected to provide tomake management decisions and of the data that isavailable or will have to be derived by establishinga monitoring programme. Additionally, it is essentialto determine the available resources, includingexpertise, skilled personnel, monitoring equipment,field data and computers, that are required toconstruct and utilize the model within the budgetconstraints that can be identified. This includes theability to understand and describe processes thattake place and the data required for validating themodel and determining the numerical values of itscoefficients. Consideration should also be given tothe local legal and regulatory framework whichpertains to the case under consideration to ensurethat the model results will be sufficiently robust,extensive and detailed to satisfy future scrutiny.It is good practice to set up a management team,which should include relevant stakeholders to guidethe project, review interim outputs, resolve differencesof opinion and reach an agreement on theacceptability of each stage of the model developmentprocess.Having determined the objectives of the modellingproject, a phased approach is required because ofthe range of uncertainties and the relatively highcosts and long work programme that is generallyassociated with groundwater modelling. Recognizingthese issues, the Environment Agency of Englandand Wales produced a guideline (EnvironmentAgency, 2002) setting out the sequential phasesthat should be considered in the groundwatermodelling process, shown in Figure II.6.13.The approach illustrated above comprises a decisionsupport system that progresses from a scopingstudy, to a conceptual model and a historical model,ending with a predictive model that can subsequentlybe refined with operational data. Thesemeans are designed to meet the objectives, assessthe options, derive a response to options, evaluatethe results, select a preferred solution and set up amonitoring system to assess the outcomes.6.3.4 Snowmelt modelsSnowmelt is analogous to rainfall with respect tothe supply of water for infiltration and runoff,except for the lag of the melted snow in thesnow cover. In some areas, snowmelt water isthe principal contributor to reservoirs, rivers,lakes and aquifers. In mountainous snow regions,snowmelt becomes an important component ofrunoff, usually making up more than 50 per centof the total streamflow. In some mountainbasins, snowmelt makes up 95 per cent of therunoff.Ordinary measurements of incremental changes inwater equivalent of the snow cover are not satisfactorymeasurements of snowmelt, largely because ofthe inherent observational and sampling errors.Two additional and compelling reasons exist forestimating, rather than observing, snowmelt. Oneis in forecasting streamflow, where it is advantageousto forecast the causes of melt instead ofmerely waiting for the resulting melt. The otherreason, particularly for design and planning, is theneed to extrapolate extreme melting rates on thebasis of physical processes. Snowmelt has beenincorporated in a number of hydrological models asindicated in the short review of HOMS componentsin 6.1.6.In principle, a conceptual snowmelt runoff modelis the coupling of a routine for snow accumulationand ablation with a rainfall-runoff model. The jointmodel can be used in all climatic conditions foryear-round forecasting. Snowmelt runoff modelshave also been developed specifically for use for thespring snowmelt period. In all cases, melting of asnowpack is driven by the energy balance.Conservation of energy dictates that the change insnow temperature is balanced with the energyfluxes entering or leaving the pack. The conservationof mass within a snowpack can be described bythe following simple continuity equation:I − O = dSdt(6.37)The inputs are precipitation, condensation andfreezing surface water, while the outputs are

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-31Define purposeDefinition of purposeand project briefField dataScoping studyReportscoping studyField dataDevelopconceptual modelCollation of data andformulation ofconceptual modelReport conceptualmodelPhase 1Project reviewConstruct historicalnumerical modelDevelopment ofhistorical modelField dataCompare results withfield data andrefine modelPhase 2Report historicalnumerical modelField dataPrediction andoptions appraisalPrediction andoptions appraisalReport predictionsimutationPhase 3Final reportFurther operational useField dataUpdate numericalmodelMaintain modelas anoperational toolReport modelevaluation andupdateFigure II.6.13. Groundwater modelling process (Environment Agency, 2002)

II.6-32GUIDE TO HYDROLOGICAL PRACTICESsublimation and runoff, with all units expressedtypically in millimetres of water. Mass changes tothe snowpack can also occur through blowing snow,sublimation and accumulation (Pomeroy and Brun,2001) and snowfall on the ground can be influencedby canopy interception, which can also represent asignificant loss of snow mass. Some of the interceptedsnow may also blow off the trees and ultimatelyback to the ground, but much of it may sublimateand be lost to the snowpack (for further details onsnow canopy interception, please refer to Hedstromand Pomeroy, 1998).Because of these complexities, conceptual methodsfor describing physical properties of snow and subsequentmelt at the catchment scale have beendeveloped. These include snowcover depletioncurves and temperature index melt models, both ofwhich are used in many operational hydrologicalmodelling systems to describe and predict thehydrological response of snow-covered catchments.6.3.4.1 Index methods for estimating basinrunoffMany medium- and long-term snowmelt streamflowforecast models are based on statistical indexmethods. Available data on precipitation andsnowcover in the mountains do not, as a rule,make it possible to determine the amount of snowpackon the ground and may serve only as anindex of this value. For this reason, relationshipsbetween seasonal flow and a snow-accumulationindex are of a statistical nature. Although suitablefor forecasting purposes, they cannot be used forwater-balance analyses in most cases.The success of a long-term forecast depends verymuch on how well the snow-accumulation indexrepresents the actual conditions. There are at leastfive additional factors that may have someinfluence on runoff, and consequently on thecorrelation between the runoff and the snowaccumulationindex:(a) Antecedent groundwater storage;(b) Amount of precipitation occurring between thelast snow survey and the issuing date of theforecast;(c) Amount of precipitation during the snowmeltperiod or the period for which the forecast isissued;(d) Amount of sublimation of the snowpackbetween the last survey and the issuing date ofthe forecast;(e) Amount of sublimation of the snowpack duringthe snowmelt period or the period for whichthe forecast is issued.In those river basins where base flow from aquifersrepresents a substantial proportion of the totalrunoff and varies considerably from year to year,the accuracy of the correlation can be increased bytaking antecedent groundwater conditions intoaccount.Precipitation can be taken into account in twoways:(a) By combining a precipitation index with thesnow-accumulation index or using the sum ofthese indices as a single variable;(b) By using a precipitation index as a supplementaryvariable.Subsequent precipitation should be included in therunoff relationship during procedure development.This ensures that the precipitation effects areincluded in deriving the statistical snowmelt forecastingrelationships.If financial budgets allow, snow surveys should beconducted in the mountains several times duringthe winter so that snow-accumulation trends canbe derived. The final snow survey is generally carriedout at the end of the snow-accumulation periodjust before the beginning of the spring snowmelt.Snow-survey data at the end of the snow-accumulationperiod are used for calculating thesnow-accumulation index.Snow courses located at various altitudes are usedto obtain data to establish a relationship betweenthe snow water equivalent and the altitude,w = ƒ(z). A different relationship is obtained foreach year. When the observation data are insufficientfor plotting graphs of w = ƒ(z), the multiplecorrelation between runoff and the snow waterequivalent at each point of observation can be used.Snow course data may still be used as input to statisticalmodels to forecast runoff.In most cases, the best index of the water availablefor runoff from mountainous areas can be developedfrom a combination of precipitation andsnow-survey data. This can be accomplished bystatistical approaches.6.3.4.2 Conceptual snowmelt runoff modelsCatchment runoff can be estimated using a numberof possible algorithms that represent the physics ofa melting snowpack. In many ways, melted snow istreated the same as rainfall and infiltrated into thesoil matrix using a number of possible infiltrationalgorithms. Snowmelt runoff simulation modelsgenerally consist of a snowmelt model and a

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-33transformation function. The snowmelt modelgenerates liquid water from the snowpack that isavailable for runoff and the transformation modelis an algorithm that converts the liquid output atthe ground surface to runoff at the basin outlet(Donald and others, 1995). The snowmelt andtransformation models can be lumped or distributedin nature. Lumped models use one set of parametervalues to define the physical and hydrologicalcharacteristics of a watershed. Distributed modelsattempt to account for spatial variability by dividingthe basin into sub-areas and computing snowmeltrunoff for each sub-area independently using a setof parameters corresponding to each of the subareas.Snowmelt models generally include asnowcover representation which can range from asimple single-layered snowpack (see, for example,Anderson, 1973) to a multi-layer conceptualsnowpack, as illustrated by Brun and others (1992).Snowpack representation has implications for thetiming of the snowmelt runoff because of its abilityto store water.Many operational snowmelt runoff models usesome form of a temperature index or a degree-daymethod to determine when snowmelt occurs andhow much snowmelt may occur in a specific periodof time. Snow-accumulation and ablation modelsuse temperature and precipitation to accumulatethe snowcover and air temperature as the soleindex to the energy exchange across the snow-airinterface. The latter aspect is usually modelledusing the degree-day method, which uses airtemperature as the index of snowcover outflow.The degree-day method does not explicitly accountfor those processes that cause snowcover outflowto differ from snowmelt, that is, refreezing snowmeltcaused by a heat deficit and retention andtransmission of liquid water. A diagram of themodel developed by Anderson (Anderson, 1973) isshown in Figure II.6.14. Actual measurements ofsnowcover from snow surveys or point measurementsmay be used as an additional source ofinformation to improve the seasonal volume forecastsfrom conceptual models that use onlytemperature and precipitation as input (Todiniand others, 1978).6.3.4.3 Extended streamflow modellingConceptual models can only simulate snowmeltrunoff for the period for which input data are available.Forecasts for the future can be made by usingforecast values of precipitation and temperaturederived from statistical or stochastic analysis orfrom extended predictions using numerical weathermodels. The pattern of the seasonal runoff cannotbe forecast satisfactorily unless the effects of futureweather conditions are taken into account.For index and statistical forecast procedures, thiscan be accomplished by using indices for the rest ofthe season based on past records of precipitationand temperature. For conceptual models, climatologicaldata for many years, generally 20 or more,should be used to develop hypothetical runoffsequences for each year’s conditions. Probabilitydistributions may be developed from these simulationsfor any specific period of time in the futureand for a specific hydrological characteristic, suchas peak flow, volume or discharge per unit area(Twedt and others, 1977). This pre-supposes thatRainand nosnow ongroundPrecipitation,air temperatureRain or snowSnowpackHeat exchange,snow-air interfaceAreal extent ofsnow coverSnowpack heatstorageNegativeheatstorageMelt from excess heatLiquid-waterstorageTransmission ofexcess waterthrough the packGround meltSnowpackoutflowT a > P • temperature = rainT a < P • temperature = snowLegendInputFunctionStorageOutputFigure II.6.14. Snow accumulation and ablationmodel flow chart

II.6-34GUIDE TO HYDROLOGICAL PRACTICESthe historic sequences are representative of whatcan be expected in future years.6.3.4.4 Input dataInput data for use in physically based or indextypes of conceptual models may be either precipitationmeasurements and/or measurements of thewater equivalent of the snow cover. With physicallybased conceptual models, corrections shouldbe made for systematic errors (see Volume I, 3.3.6)in the precipitation measurements, so that theinput data are as representative as possible of theaverage precipitation and/or snowcover. In mountainousregions, where the snowcover is highlydependent on altitude, the observations frommeteorological stations are often affected by localexposure including wind, local slope and aspectand must be adjusted to better represent the averagemeteorological conditions if they are used tosimulate the snowcover conditions. In practice,snowcover and precipitation measurementscomplement each other.The spatial distribution of snowcover is often bestdescribed by snowcover depletion curves whichsummarize the per-cent areal coverage of the snowpackas it increases in average depth. Watershed-widesnowcover depletion curve relationships arecurrently used in lumped hydrological models suchas the National Weather Service River ForecastSystem, or NWSRFS (Anderson, 1973), to describethe snowcover distribution as the snowcover melts.These relationships are difficult to obtain andrequire calibration for each specific watershed. Thesimplest representation of snowcover is uniformsnowcover, which is of constant depth and completeareal coverage. Knowledge of the areal distributionof the snowcover within and between land units isrequired to make reasonable estimates of the totalwater available in the snowcover of a watershed.The areal distribution of the snowcover within aland-unit type can then be summarized in the formof an areal distribution curve. An areal distributioncurve is a summary of the state of the snowcover ata given time within a basin. Intense samplingprogrammes are required to develop datasets toquantify the snow distribution in the form of arealdistribution curves.Since it is not practical to physically model thedistribution of snowcover, the development ofstatistical or empirical distribution relationshipsbased on landcover and physiographicconsiderations is a sensible approach to the problem.This is accomplished by the use of snowcoverdepletion curves. Rango and his colleagues presenta depletion curve where the percentage of snowcoveredarea is on the y-axis and time is on thex-axis, providing a conceptually based approach tounderstanding basin snowmelt (Rango and others,1983).6.3.4.5 Theory of snowmelt at a pointA rational approach to estimating the rate of snowmeltis based on an energy budget, which accountsfor the significant modes of heat exchange. Heat istransmitted to snow by absorbing solar radiation,net long-wave radiation, convective heat transferfrom the air, latent heat of vaporization by condensationfrom the air, relatively small amounts ofheat from rain and generally negligible amountsof heat from the underlying ground.The equation for energy balance can be used todetermine the amount of energy available forsnowmelt, Q m, which can be directly transformedinto the amount of snowmelt for a unit cube ofsnow:Q m= Q n+ Q h+ Q e+ Q g+ Q a– dS i/dt (6.38)where energy fluxes (per unit area) are respectively:Q nnet longwave radiation, Q hsensible heat transferdue to temperature difference between thesurface and the air, Q elatent energy flux caused bywater vapour change (release of heat by condensationor its removal by sublimation or evaporation),Q gconduction of heat from the underlying ground,Q aadvection of heat (rain), and S isnowpack heatstorage.A melting snow cover typically contains from twoto five per cent by weight of liquid water, but occasionallyas much as 10 per cent is held for briefperiods when melting rates exceed transmissioncapacity. Thus, for short periods of time, the totalrelease of water from a snow cover may slightlyexceed the amount of snow actually melted by theprevailing meteorological conditions. For practicalpurposes, this release of previously melted water isimplicitly incorporated into the empiricalconstants, which are therefore burdened withuncertainties.Absorbed solar radiation varies with latitude,season, time of day, atmospheric conditions, forestcover, slope, orientation of surface and the reflectivityof the snow. The effects of latitude, season,time of day and atmospheric conditions are includedin solar radiation observations, which must generallybe interpolated because of the sparse networkof such stations. These effects may also be computed

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-35on a daily total basis by means of formulae ordiagrams that express solar radiation as a functionof degree of cloudiness, time of year and latitude.The effect of forest cover on the transmission ofsolar radiation is important, and in experimentalareas, it has been expressed as an empirical factorthat relates the transmission coefficient to canopydensity. Usually direction and steepness of slopeand forest cover are represented by constantfactors, derived empirically for a given drainagearea.Reflectivity of a snow surface ranges from about90 per cent for newly fallen snow to about 40 percent for old snow that is coarse grained and whichis ordinarily covered late in the season by a thinlayer of dark debris such as wind-blown organic ormineral dust. In middle latitudes during late spring,an unforested snow cover with low reflectivitycommonly absorbs sufficient solar radiation to melt50 millimetres of water equivalent per day.Long-wave radiational exchange is the differencebetween outgoing radiation from the snow surfaceand downward radiation from clouds, trees and theatmosphere. With dense low clouds or heavy forestcover warmer than 0°C, the exchange is a gain tothe snow. Long-wave radiation from the atmospherein the absence of clouds or forest cover islargely a function of air temperature and is nearlyalways less than the loss from the snow. Long-waveradiational exchange commonly ranges from again of heat equivalent to as much as 20 mm ofmelt water per day to a loss equivalent to 20 mmper day.The main factors in the convectional exchange ofsensible heat are the temperature gradient in the airimmediately above the snow and the intensity ofturbulent mixing expressed by horizontal windspeed.The principal factors in heat from condensation arethe vapour-pressure gradient and intensity of turbulentmixing, which may be indicated by wind speed.The combined exchange of sensible and latent heatby turbulent exchange may range from a gain ofheat that is equivalent to more than 100 mm ofmelt per day to a loss corresponding to two or threemillimetres. The potential gain greatly exceeds thepotential loss because the temperature and vapourpressuregradients for heat gain can be very greatwith the snow temperature limited to 0°C, whereaswith very low air temperatures and vapour pressuresaccompanying the loss of heat, thesnow-surface temperature generally fallscorrespondingly. Thus, the gradients are reduced.Heat gain from warm rain can be computed fromthe latent heat of fusion of the ice (80 calg –l ) whichcomprises the snow, and the temperature of therain, which can usually be taken as the wet-bulbtemperature of the air. Computations show that anunusually heavy rain – at least 120 mm of rain witha temperature of 16°C – is required to produce asmuch as 25 mm of snowmelt in a day.The conduction rate of heat from the soil to a newlyformed snow cover may be rapid for a short time,but the usual geological gradient of temperatureand the gradient of temperature after steady-statehas been established produce less than about onemillimetre of snowmelt per day.The foregoing rates of snowmelt from variousmodes of heat exchange are not additive. For example,the conditions for maximum turbulentexchange would occur during stormy weather andnot with maximum solar radiation. Numerousequations have been published expressing themodes of heat exchange in terms of observableelements. For further information, please refer toWMO-No. 749, Operational Hydrology ReportNo. 35 – Snow Cover Measurements and ArealAssessment of Precipitation and Soil Moisture (WMO,1992) and WMO-No. 646, Operational HydrologyReport No. 23 (WMO, 1986) – Intercomparison ofModels of Snowmelt Runoff.The integration of a rational snowmelt functionover a heterogeneous drainage area of significantsize is extremely difficult at best and practicallyfutile without elaborate instrumentation. Estimatingthe quantity or rate of melt is based on water budgetaccounting in addition to heat budget accounting.In the absence of rain, radiational exchange isrelatively important, and consequently the effectsof snow reflectivity and forest canopy density areimportant; however, these are rarely measured.During periods of heavy rain, the rate and amountof snowmelt may be no greater than the error inestimating the amount and effects of the rain.During storms accompanied by considerableturbulent mixing and heavy, low clouds, there isrelatively little short-wave solar radiation, and longwaveradiation, convection and condensation arethe major sources of heat. The difficulty of separatingthe contribution of rain from that of snowmelt hasleft the question of snowmelt during rain largely inthe realm of theory with very little empiricalevaluation (US Army Corps of Engineers, 1960).Daily solar radiation for a given latitude and time ofyear is influenced by local cloudiness, which in turnis observed subjectively and sparsely – rarely with

II.6-36GUIDE TO HYDROLOGICAL PRACTICESrespect to its radiative transmissivity. Further, thereis the problem of determining the active orcontributing area of the snow.The active or contributing area may be defined asthe area over which snow is melting or over whichsnowmelt reaches the soil. This area, howeverdefined, varies diurnally. If the diurnal cycleincludes nocturnal freezing, some account must betaken of the heat and moisture storage involved.Early in the melting period, some heat is necessaryto raise the temperature of the snow to 0°C and tomelt sufficient snow to meet the water-holdingcapacity of the snow cover. This heat is relativelysmall with respect to the total heat required to meltthe snow cover.The most widely applied method for estimatingbasin-wide snowmelt is the use of degree-dayfactors. Temperature data are usually available, andthe variation of temperature over a drainage areacan generally be determined for deriving and applyingdegree-day functions. The rationale for thedegree-day method is twofold. First, air temperaturenear the snow is largely a physical integration ofthe same modes of heat exchange that melt snow.Second, each mode of heat exchange can be relatedto air temperature except during abnormal winds.For example, minimum daily air temperature ishighly correlated with dewpoint temperature,which determines the vapour-pressure gradient forcondensation melting. Maximum daily temperatureor temperature range is an index of solarradiation. Within its usual range, long-wave radiationcan be expressed as a linear function of airtemperature.Efforts have been made to give the maximum andminimum daily temperatures various weights andto use degree-day bases other than 0°C. Efforts havealso been made to divide the day into smaller timeunits and to use degree-hour factors. However, thediurnal cycle of heat exchange and snowmelt makesthe day a logical and convenient unit for snowmelt,and the usual degree-day base is 0°C, which is generallytaken as the mean of the daily maximum andminimum air temperatures. Point snowmelt degreedayfactors for several mountainous regions in themiddle latitudes of North America have been averagedin Table II.6.3, in millimetres of melt, and themean of daily maximum and minimum temperatureabove a base of 0°C. Individual values maydepart widely from these averages.Similar degree–day factors are given in Table II.6.4for lowlands in moderate latitudes of the formerUnion of Soviet Socialist Republics.With a shallow snow cover, the storage and delayof melt water passing through the cover aregenerally inconsequential, compared with storageand delay in the soil mantle and uncertaintiesin the amount of snowmelt itself. The timerequired for liquid water to drain from a snowcover is about one hour, plus an hour for each50 cm of depth.Areal variations in melting rate and in the distributionand diminishing size of the area covered bysnow during a melting period are related to fairlypermanent characteristics of the catchment area,such as its topography and distribution of vegetativecover. Consequently, the melting rate over acatchment reflects a fairly consistent trend incontributing area and snow condition during amelting period. This trend influences the shape ofempirically defined S-shaped curves such as thosein Figure II.6.15. Because of the areal dispersion ofthe snow and of its local melting rates, some of thesnow starts to melt before the rest. Thus, the averagemelt rate per unit area is low early in themelting period and increases as more of the areacontributes. Toward the end of the melting period,the slopes of the curves of Figure II.6.15 diminishbecause of the diminishing area of snowmeltcontribution. The steepest portions of the curvesoccur after melting conditions have become establishedover a large contributing area. Theproportionality of melting rates to the initialTable II.6.3. Degree–day factors (mm °C –1 ) formountainous regions in North AmericaMonthModeratelyforestedPartlyforestedNon-forestedApril 2 3 4May 3 4 6June 4 6 7Table II.6.4. Degree–day factors for lowlandregions in the former Union of Soviet SocialistRepublicsAreaDegree–day factors(mm °C –1 )Non-forested areas 5Sparse coniferous andaverage density of harwoods 3–4Average density of coniferouswoods and dense mixed woods 1.7–1.8Dense coniferous woods 1.4–1.5

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-37120120more consideration must be given to the retentionof melted snow in the cover.Accumulated snowmelt (mm)100806040200Figure II.6.15. Typical degree–day snowmeltrelationship for a catchment for different values ofinitial water equivalentquantities of snow comes largely from the factthat, with more snow, the contributing area islarger. The steepest portions of the curves ofFigure II.6.15 have a slope that corresponds to thevalues of Tables II.6.3 and II.6.4.Evaporation loss from the snow cover is negligiblysmall during brief melting periods and may bemore than balanced by condensation on the snowsurface. Equations for condensation on a snowcover may also be used to estimate evaporationfrom the snow. Measuring evaporation from asnow or ice surface is difficult and probably aboutas accurate as the computation of evaporation ingeneral. It is estimated that during winter periods,evaporation occurs from a typical snow surface atrates ranging from zero to as much as 20 mm permonth. During melting periods, condensationtends to prevail and occurs at rates spanning fromzero to possibly as much as 10 mm of condensateper day.In mountainous regions, where great quantities ofsnow accumulate, where the melting season maycover several months and where melting conditionsvary greatly with a large range of elevation, the reliabilityof curves such as those of Figure II.6.15 arelimited. Evaporation during long warm periodsmay be significant. During the melting season,successive aerial or other surveys show the changingsnow-covered area and meteorologicalobservations are interpreted to express the variationof melting rate with elevation. The contributionof snowmelt should be determined by elevationzones. In addition, with deep mountain snow cover,6010 20 30 40 50 60 70 °CAccumulated degree–days above 0°C80100Higher than average degree–day factors should beused when unusually high wind speeds or humidityoccur.6.3.4.6 Estimating snowmelt inflow ratesTo determine the total snowmelt runoff in lowlandbasins, water balance studies can be adopted. Fromthese, the expected total snowmelt runoff can beestimated at the beginning of the snowmelt period.However, values of the daily snowmelt inflow areoften required for hydrograph calculations. Thefollowing factors should be taken into accountwhen estimating these values:(a) Heat inflow to the snow cover;(b) Water-retention capacity of the snow cover;(c) Area covered with snow;(d) Water-retention capacity of the basin.6.3.4.7 Probable maximum precipitation andsnowmeltIn the case of very large basins at high latitudes,snowmelt, rather than rainfall, may be theprimary cause of the probable maximum flood.Flood-runoff volume and temporal distributionare then based on the estimation of snowmeltresulting from the estimated maximum values oftemperature, wind, dewpoint and insolation in amanner analogous to maximization of stormrainfall.A more common situation in lower latitudes is forrainfall to be the primary factor producing the probablemaximum flood with snowmelt adding anincrement to the maximum hydrograph. Snowmelt,compatible with estimated synoptic conditionsaccompanying the maximized storm, is then addedto the maximized rainfall depth.For some basins, only a detailed analysis will revealwhether the probable maximum flood will resultfrom a cool-season rainstorm combined with snowmeltor from a summer rainfall that may be moreintense but cannot logically be expected to occur incombination with snowmelt.6.3.4.7.1 Probable maximum snow accumulationThe snowmelt contribution to the probable maximumflood will depend on the maximum rate ofmelting and the water equivalent of the snow coveravailable for melting. Water equivalent of a snowcover is the depth of water that would result from

II.6-38GUIDE TO HYDROLOGICAL PRACTICESmelting and depends on the snow density as wellas its depth. Various methods have been used toestimate probable maximum snow accumulation;the three most common are as follows:(a) Partial-season method – The highest observedsnow accumulations in each month or twoweekperiod, according to the frequency ofobservations, are combined, regardless of theyear of occurrence of each observation, to givea synthetic year of very high snowfall. Themethod can be applied to shorter time intervals,such as a week or four-day period, if suitablerecords are available;(b) Snowstorm maximization – The ratio ofmaximum atmospheric moisture content inthe project area at the time of year at whicha snowstorm occurs to the actual moisturecontent of the snowstorm is determined. Theobserved snowfall produced by the snowstormis multiplied by this ratio to give maximizedsnowfall for the snowstorm. Maximization ofmoisture content must be restricted to a valuethat will produce snow and not rain;(c) Statistical methods – A frequency analysis ofprecipitation and snow-depth records is madeto determine the values for various return periods.Analyses are made of three types of data:station precipitation depth, basin snowfalldepth and water equivalent of snow on theground.6.3.4.7.2 Snowmelt estimationOwing to the complex spatial and temporal variabilityof snowmelt over most catchments caused bydifferences in slope, aspect, forest cover and depthof snow cover, the degree–day method is oftenadopted as a practical solution to the problem ofestimating snowmelt over a catchment. Maximumdegree–day conditions may be estimated fromtemperature records for the project basin or aneighbouring area and may be applied to the estimateof probable maximum snow accumulation toprovide an estimate of probable maximum floodrunoff.For probable maximum conditions, the air temperatureand wind speed are made consistent withthe assumed synoptic conditions accompanyingthe storm-producing probable maximum rainfall.It also is assumed that an optimum snow coverexists. Optimum in this situation means thefollowing:(a) The snow cover has only sufficient water equivalentto melt completely during the storm;(b) The snow cover has been melting and containsa maximum amount of liquid water;(c) The water equivalent of the snow cover isdistributed so as to be at a maximum where themelting is maximum, which is different fromthe usual situation of increasing the snow-coverwater equivalent with increasing elevation.6.3.4.8 Runoff from short-period snowmeltIn a plains region, where increments of runoff arerelatively small and the melting period is brief,runoff may be estimated by incorporating estimatedsnowmelt obtained by methods such as describedabove, into a rainfall–runoff relationship (see6.3.2). It may be necessary to use the relationship ina way that reflects a high percentage of runoffbecause the snow cover or cold weather inhibitsevapotranspiration losses antecedent to the meltingperiod. In mountain catchment areas, where deepsnow covers prevail and the melting season lastsseveral months, methods commonly used forestimating runoff from brief rainstorms do notnecessarily apply. Runoff from the melt that occurson a particular day is ordinarily spread over a longperiod, overlapping the melting increments ofmany other days. Also, evapotranspiration losses,which may be neglected during a period of rainfall,become important during a long melting season.One way to estimate runoff from day-to-daysnowmelt is first to estimate the seasonal volume ofrunoff and then to distribute it in accordance withobserved or estimated local daily melting rates(6.3.4.6 and 6.3.4.7), basin-storage characteristics,contributing area and seasonal evapotranspiration.Basin storage and lag may be accommodated byrouting though an analogous system of reservoirswith constants determined empirically fromhistorical basin data. Where the catchment is sosmall that the diurnal increments of snowmelt arenot damped out by storage, six-hour – rather thandaily melting increments should be used, or acharacteristic diurnal distribution can be introducedinto the routing method.6.3.4.9 Snowmelt runoff analysis usingremote-sensingSnowmelt runoff procedures have followed twodistinct paths: an empirical approach and a deterministicmodelling approach. The choice ofapproach depends on both the availability of datato quantify the snowpack and the extent of detailrequired of the output. In order to make accurateestimates of snowmelt runoff, hydrologists need toquantify the snowmelt in the following terms: theareal extent of the snow, S; the snow water equivalent,SWE; and the condition or properties of thesnow such as depth, density, grain size and

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-39presence of liquid water (Engman and Gurney,1991). The gradually decreasing areal extent is acharacteristic feature of the seasonal snow cover.Regardless of the approach used to conduct day-todaysimulations of snowmelt runoff, whether it bean empirical approach based on historical data or adeterministic approach, it is sufficient to know thedaily snow-covered area in the basin without knowingthe initial accumulation of snow in terms ofwater equivalent (WMO, 1994).For many basins, there is a very good relationshipbetween runoff and snow cover area (Engman andGurney, 1991). For operational runoff forecasts,however, the water equivalent must also be determined(WMO, 1994). Remote-sensing offers a newvaluable tool for obtaining snow data in order topredict snowmelt runoff (Engman and Gurney,1991). Ostrem and others (1991) developed amethod using data from the National Oceanic andAtmospheric Administration (NOAA) and the televisioninfrared observation satellite (TIROS) tomeasure the remaining snow and predict the correspondingsnowmelt runoff volume for a number ofNorwegian high mountain basins. Many largehydropower companies use snow cover extent mapsfrom NOAA AVHRR, advanced very high-resolutionradiometer, on an operational basis as input to theirhydropower production planning (Andersen,1991).Remote-sensing techniques using appropriatewavelength bands allow to a certain degree theestimation of snowcover features such as grainsize, albedo, layering, surface temperature andsnowpack temperature. This, in turn, allows a goodestimate of the time when the snowpack is readyto transmit melt water from the surface to lowerlayers, known as a ripe condition, and to eventuallyproduce runoff at the base of the snowpack(Rango, 1993). The first empirical approach tosnowmelt runoff estimation using remote-sensingwas developed by Rango and others (1977); theyused satellite-observed snow cover data in empiricalregression models developed for the Indus andKabul Rivers in the Himalayas. Martinec and Rango(1987) and Rango and van Katwijk (1990) laterused remotely sensed snow-water-equivalent andtemperature data to construct modified snow coverdepletion curves for use in the snowmelt-runoffmodel for snowmelt forecasts in the Rio Grandebasin.Overall, remote-sensing is very successful in mountainregions, especially when the aim is to mapsnow cover. This is hindered only in regions withvery dense forest cover.New models developed to use remote-sensing datawill also improve snow hydrology predictions.Further, the merging of remote-sensing data withdigital elevation modelling and geographical informationsystems enables different types of datato be combined objectively and systematically(Engman and Gurney, 1991). Digital elevationmodelling is used to normalize imagery by usingthe elevation of the sun and the slope, aspect andelevation of the terrain (Baumgartner, 1988; Millerand others, 1982). Geographical informationsystems are helpful in combining vegetation maskswith satellite imagery (Keller, 1987).6.3.5 Streamflow routingRunoff from a headwater area moves downstreamas a wave whose changing configuration at variousstations can be computed by a technique known asflood routing. Storage and other effects tend toattenuate the wave. Irregularities in channel conditionsand tributary inflows are inherent complexitiesof the problem. The routing of flood waves throughreservoirs and channels is accomplished by manymethods.6.3.5.1 Hydrodynamic methodsHydrological research has gained much knowledgeof the physical processes that comprise the watercycle in nature. Similarly, the high technologyemployed in continuous data acquisition and integrationin time and space, combined with moderncomputers, permit rapid processing of hydrologicaland meteorological data of all types. All this hashelped improve the third type of modelling, hydrodynamicmodelling.Hydrodynamic models are based on numerical integrationof the equations of momentum and massconservation that describe the physical processes inthe basin. Since hydrodynamic models are based onthe physical laws governing the processes, extrapolationbeyond the range of calibration may be performedmore confidently than with conceptual models.Complete dynamic routing, which accounts for flowaccelerationeffects and the water-surface slope, candetermine flows and water-surface elevations accuratelyin the following unsteady flow situations:(a) Upstream movement of waves, such as thoseproduced by tidal action or sea-storm surges;(b) Backwater effects produced by downstreamreservoirs or tributary inflows;(c) Flood waves occurring in rivers having flatbottom slopes: less than 0.05 per cent;(d) Abrupt waves caused by controlled reservoirreleases or by the catastrophic failure of a dam.

II.6-40GUIDE TO HYDROLOGICAL PRACTICESDynamic routing is generally based on the onedimensionalhydrodynamic equations of unsteadyflow, known as the Saint Venant equations. Theseequations are generally expressed in their conservativeform below.Continuity:∂Q∂x + ∂s c ( A + A 0 )∂tMomentum:− q = 0 (6.39)∂ ( s m Q )+ ∂ (β Q 2 /A)∂t ∂x(6.40)+ ⎛gA∂h ⎞+ S∂t f + S⎝ec ⎠ − qv x + W f B = 0in which:S f =n e QA 2 R 4/3 (6.41)where Q is discharge, A is the active cross-sectionalarea, A 0is the inactive or dead-storage crosssectionalarea, s mis a depth-weighed sinuositycoefficient, S ecis the expansion-contraction slope, βis the momentum coefficient for non-uniformvelocity distribution within the cross-section, W fBis the resistance effect of wind on the water surface,h is the water-surface elevation, v xis the velocity oflateral inflow in the x-direction of the river, B is thetop width of the active cross-sectional area, n is theManning roughness coefficient, R is the hydraulicradius; other symbols are as previously defined,except that:S ecKecΔ (Q / A )2=2 g Δ x(6.42)where K ecis the expansion and contraction coefficient,Δ(Q/A) 2 represents the difference in the term(Q/A) 2 at two adjacent cross-sections separated by adistance Δx.No analytical solutions of the complete non-linearset of equations 6.39 to 6.41 exist. The numericaltechniques for solving the aforementioned equationsfor natural rivers may be classified into twobroad categories: the method of characteristics,which is not widely used nowadays, and finitedifferencemethods in explicit and implicit schemes,which are very common. Finite-difference methodstransform partial differential equations 6.39 and6.40 into a set of algebraic equations. The explicitmethods solve these algebraic equations sequentially,at each cross-section, computational reach,and at a given time, while the implicit methodssolve algebraic equations simultaneously for allcomputational reaches at a given time.There are advantages and disadvantages to the varioussolution techniques. Factors such as numericalstability and convergence, required computationaltime and computer storage, and degree of programmingand mathematical complexity must beconsidered. Some solution techniques requiremodifications to the form of equations 6.39 and6.40 before they can be applied.In general, implicit finite difference techniques aremore complex but more efficient than explicitmethods when calculating unsteady flows of severaldays duration. Much larger time steps can be usedwith the implicit techniques. Explicit techniquesare simple; they are confronted, however, withnumerical stability problems unless the time step isproperly selected. These and other limitationsshould be thoroughly understood before selecting aparticular solution technique to develop a dynamicroutingforecasting method or select an existingdynamic-routing technique for a particularapplication.A critical task in applying dynamic routing to anactual forecast situation is the determination of theroughness parameter in the S f, friction-slope termgiven by equation 6.40. The roughness parameteroften varies with flow or elevation, as well as withdistance along the river. A prior determination ofthe roughness parameter relationship with flowand distance by trial and error is very time consuming.Techniques for automatically determining therelationship greatly facilitate the operational utilizationof dynamic routing in a forecastingenvironment. A proper evaluation of the boundaryand the initial conditions for solution of the SaintVenant equations in an operational mode is anothercritical task in the implementation of dynamicrouting techniques.Another critical task is the establishment of an efficientdata-acquisition and management programmelinked integrally with the computational element.Cross-section geometry should be processed as efficientlyas possible for use by the dynamic-routingprogram. Anticipated flow conditions shouldrequire as little data entry as possible for a dynamicroutingtechnique to be feasible for use as anoperational forecasting tool.By slightly rewriting the momentum equation andignoring momentum from lateral inflows, avery clear picture can be obtained showing the

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-41fundamental differences between dynamic, diffusion,and kinematic routing.Consider:1g∂h∂t+ v g∂v∂x+ ∂h∂x– s o+ s f= 0kinematic modeldiffusion modeldynamic model(6.43)At the first level of approximation, the terms representingthe accelerations related to the timevariation of inflow and the spatial variation invelocity are neglected. The resulting model isreferred to as the diffusion model. In some flowsituations, it is also possible to neglect the pressureforceterm and treat the momentum equation as abalance between the forces of gravity and friction.This approximation is known as the kinematicmodel.Both the kinematic and diffusion approximationshave been used successfully to describe overlandflows and flows in streams where slopes are greaterthan approximately 0.1 per cent. The diffusionmodel can be used on rivers with smaller slopes,but with caution because the inertia terms maybecome important. The kinematic model hasbecome popular in applications where the irregulargeometry and topography of natural catchmentscan be replaced by a series of simple elements, suchas flow planes and regular channel segments. Thekinematic equations are also used in water qualitymodels that predict the transport of pollutants. Akinematic model does not consider backwatereffects from lateral inflows or downstream reservoiroperations, nor can it be used to predict waveprogressions in the upstream direction.6.3.5.2 Hydrological methodsHydrological flood-routing methods use only thecontinuity equation, or mass conservation law. Inthese techniques, only the wave propagation isstudied by considering the increases and decreasesof storage in a reach lying between two measuringpoints. However, because the relationship betweenstorage and flow is determined empirically bythese methods, they cannot be used directly whenflow data or levels are required for designpurposes.When hydrological routing methods areemployed, the flow at an upstream point is givenor assumed, and routing is used to compute theflow and stage at a downstream point. Routingconsists of the solution to the following continuityequation by using a relationship betweenstorage and flow:I – Q = dS/dt (6.44)where I and Q are the discharges at upstream anddownstream points, respectively, S is storage in theriver reach between the upstream and the downstreamcross-sections, and t is time. Solution of thisequation involves approximations concerning thestorage–flow relationship, the main difficulty inhydrological streamflow routing. However, withsufficient hydrometric data, this relationship canbe derived empirically.The simplest routing methods are based on linearstorage–flow relationships, which make it possibleto obtain analytical solutions. Two such methodsare applicable in short-range forecasting practice, asindicated below.(a) The Muskingum method, which is based on thefollowing storage–flow relationship:S = K [xQ 1+ (1 – x)Q 2] (6.45)The constants K and x are derived empirically for agiven reach from discharge data. They can be determinedby plotting S versus xI+ (1 – x)Q for variousvalues of x. The best value of x is that which resultsin the data plotting most closely to a single valuecurve.The Muskingum method is often used in the followingdiscrete form:Q j+1= C 1I j+1+ C 2I j+ C 3Q j(6.46)where C 1, C 2and C 3, being functions of Muskingumparameters K and x and the time step Δt, sum tounity ensuring that the sum of the constants isequal to unity;(b) The specific reach method, proposed by Kalininand Miljukov (1958), is based on the followinglinear storage–flow relationship:Q = K S (6.47)where K is the storage constant equal to the traveltime through the reach. The above equation isapplicable to transit reaches of specific length, L,which is roughly equal to:L =QZ ∂Q∂h(6.48)

II.6-42GUIDE TO HYDROLOGICAL PRACTICESwhere Z is the slope of the water surface, and ∂Q/∂his the tangent of a stage–discharge relationship. If ariver segment consists of several specific reaches,routing is carried out in succession from one specificreach to the next downstream. The computeddischarge for the downstream point of the firstreach is taken as the inflow for the second reach,and so on.The following formula expressing the transformationof flow by a system of identical linear reservoirscan be used for long river reaches that lack the dataneeded to determine the number of specificreaches:Q (t ) = I 0Δ tK N ( N − 1)! t N−1 e −t /K (6.49)where N is the number of characteristic reaches orreservoirs, K is the travel time for one characteristicreach, I 0is the inflow into the first characteristicreach and t is the time. The K and N parameters aredetermined by trial and error or optimization.6.3.5.3 Reservoir routingA reservoir leads to a decrease in the peak discharge,compared with that which would have occurredhad the reservoir not been in place because thepassage of a flood through a reservoir differs somewhatfrom its passage through a channel.Because the velocity of the flood wave in a reservoiris higher than in channels, the delay in the peakoutflow with respect to the peak inflow does notnecessarily mean a delay with respect to the peakthat would have occurred under the conditionsprevailing prior to the construction of the reservoir.Furthermore, the construction of a reservoir maysometimes worsen downstream flood conditions,despite its effect in decreasing peak discharges. Theattenuated peak may occur in phase with peaks oftributaries that are usually out of phase. Thus, itshould not be taken for granted that reservoirconstruction will improve downstream flood conditions.The hydrology and the hydraulics that wouldprevail under the design conditions should be studiedcarefully.6.3.5.4 Dam breaksCatastrophic flash flooding results when a damfails, and the outflow, through the breach in thedam, inundates the downstream valley. A damthat fails can be man-made or, for example, icejam or flow debris. Often the dam-break outflow isseveral times greater than any previous flood onthe river concerned. Little is known of failuremodes of artificial or natural dams. Hence, realtimeforecasting of dam-break floods is almostalways limited to occasions when failure of thedam has actually been observed. Different failuremodes may be assumed for planning calculationswhen the implications to downstream developmentare investigated with regard to zoning orevacuation contingency plans.Earlier classical studies of this problem haveassumed instantaneous dam failure and idealizeddownstream conditions. More recently, engineershave sought to approach the problem by assuminga triangular-shaped outflow hydrograph based onthe Schocklitsch or similar maximum-flowequation:Q m = 827 W d gY 03 (6.50)where g is the acceleration due to gravity, W dis thewidth of the breach and Y 0is the height of the waterbehind the dam. By using equation 6.49 and anempirical recession coefficient, the synthesizedhydrograph is routed through the downstreamvalley via a hydrological-routing technique.Alternatively, a more realistic approach can befound in dynamic-routing techniques (see 6.3.4.2)to route the rapidly changing and relatively largedam-break flood wave. Explicit account is taken ofdownstream dams, overbank storage, downstreamhighway embankments, and expansion andcontraction losses.As time is essential in real-time forecasting of adam-break flood, operational techniques must becomputationally efficient. However, an even moreimportant consideration is the data requirementof the forecast technique. If dynamic routing is tobe used, every effort should be made to minimizethe amount of cross-sectional data needed in therouting phase of the forecast, and all data andprogram files must be immediately available foruse.6.3.6 Modelling other processes6.3.6.1 Sediment transport modellingSediment transport models predict sediment transportrate and direction based on water surfaceelevations or velocities determined by using ahydrodynamic model (see 6.3.4.2), an essential partof the sediment transport model based on thenumerical solution of the Saint Venant equationsof continuity and momentum.

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-43Basic processes in sediment transport can bebroken down into erosion, entrainment, transportationand deposition. Sediment on thestreambed will remain immobile only as long asthe energy forces in the flow field remain lessthan the critical shear stress threshold for erosion.After the critical shear stress is reached, the sedimentsbegin moving by jumping or bouncing,rolling and sliding. This movement is known asbed load. Various researchers have developedrelationships describing this bed load as a functionof the bed shear stress and the grain sizediameter. These are known as sediment transportfunctions and are mostly applicable to noncohesivematerial (see 4.8.6).Computation of the particle-settling velocity isnecessary for several non-cohesive sediment transportfunctions.ω f = F dg (G – 1) (6.51)where:F =⎡⎣23 + 36v 2gd 3 (G − 1)1/2⎤ −⎦⎡⎣36v 2gd 3 (G − 1)1/2⎤⎦(6.52)for particles with diameter d between 0.0625 mmand 1 mm. For particles greater than 1 mm,F = 0.79. In the above equations, ω fis the fall velocityof sediments, g is the acceleration due to gravity,G is the specific gravity of the sediments and v is thekinematic viscosity of water.Most of the sediment transport models allow useof more than one function, since there is nouniversal function that can be applied accuratelyto all sediment and flow conditions. Most ofthese transport functions were developed tocompute total bed load without breaking downthe load by size fraction. Some transport modelsapply these functions for different size fractionsto account for variation in the bed load grainsize distribution and can simulate bed-materialmixing processes, and therefore armouringeffects.The bulk of sediment in transport can be characterizedas being transported in suspension.Suspended load calculations include the timespacelag in the sediment transport response tochanges in local hydraulic conditions. Cohesivesediments in transport will remain in suspensionas long as the bed shear stress exceeds the criticalvalue for deposition. Cohesive sediments tend tosegregate to low density units, a process that isstrongly dependent on the type of sediment, theconcentration of ions in water and flow conditions,and the settling velocity that is no longer afunction of particle size. This aggregation isaccounted for in the models by assigning settlingvelocities. In general, simultaneous deposition anderosion of cohesive sediments do not occur, butthe structure of cohesive sediment beds doeschange with time and with overburden.There may be no net change in the elevation of thebed unless the erosion rate is different from thedeposition rate; these are two processes that go oncontinuously and independently. The change ofbed level may be determined by a sediment continuityequation. The equation is derived based onthe assumption that the changes in volume ofsuspended sediment are much smaller than thechanges in bed sediment volume, which is generallytrue for long-term steady-flow simulations. Themass conservation equation for sediment reducesto:∂Q s∂x+ ε ∂A d∂t− q s= 0(6.53)where ε is the volume of sediment in a unit bedlayer volume (one minus porosity), A dis the bedsediment volume per unit length, Q sis the volumetricsediment discharge and q sis the lateralsediment inflow per unit length.Certain sediment transport and morphologicalmodels, such as MIKE 21C, consider helical flows inconnection with sediment transport simulations inorder to simulate the development of bend scour,confluence scour and the formation of point bars aswell as alternating bars. These models do providecurvilinear computational grids which are moresuitable for river morphology modelling. Bankerosion is included at each computational timestep. The eroded bank material is included in thesolution of the sediment continuity equation. Bankerosion will produce a retreating bank line, which ismodelled by the movement of the adaptive curvilineargrid.Additional information on sediment modelling isto be found in 4.8.6.6.3.6.2 Water quality modellingThe management of water quality in natural andartificial water bodies is a complex task that requiresmonitoring of the water quality characteristics,interpretation of the monitored data in relation tocausative factors and prediction of future changesof these characteristics in terms of the various

II.6-44GUIDE TO HYDROLOGICAL PRACTICESmanagement alternatives under consideration. Thesolution to these problems can be greatly aided bythe use of water quality models. These enableprediction on the basis of the following factors:(a) A series of input data on pollution inflow;(b) Meteorological-environmental initial conditions;(c) Hydraulic-hydrological and land-use characteristicsof the water body and its watershed;(d) The evolution in time and/or space of certainwater quality characteristics of the water bodyconsidered for various water managementalternatives.Water quality models are frequently linked tohydraulic and hydrological models.Mathematical water quality models may be classifiedaccording to the general taxonomy of models(see 6.1) and according to the following criteria:(a) Water quality constituents: into single- ormulti-constituent models;(b) Type of constituent modelled: into conservative,for example, salt; non-conservative physical,for example, temperature; non-conservativechemical, for example, dissolved oxygen; ornon-conservative biological, for example,coliform bacteria.For the description of pollutant transport in rivers,the most commonly used model in practical applicationsis the one-dimensional model based on theadvection-dispersion equation:∂c∂t+ u ∂c∂x = D L∂ 2 c∂x 2 (6.54)where c is the pollutant concentration, u is themean water velocity, D is the longitudinal dispersioncoefficient, t is time and x is distance.The longitudinal dispersion coefficient is calculatedon the basis of the Fisher equation:D L = 0.0 7 u'2 l 2ε z(6.55)in which u ’2 is the deviation from the cross-sectionalmean, l is the distance from the thread of the maximumvelocity to the most distant bank and ε zis thetransverse mixing coefficient.To apply this model to pollutant transport in a river,the river is divided into reaches, each several kilometresin length, within which the water velocity isconsidered to be constant. The water velocity withineach sector is calculated by means of a hydraulic orhydrological model (see 6.3.4).Water quality models can be used in water qualitymanagement for several purposes, including thedesign of water quality monitoring networks inspace and time, the interpretation of data obtainedin relation to factors determining water quality,interfacing with other environmental (air, soil)pollution models and with ecological models, theassessment of trends in water quality with or withoutvarious alternative pollution-correctionmeasures and the forecast of the arrival time of apollutant and of a concentration profile along theriver.Water quality models have been applied with variousdegrees of success to the solution of waterquality management problems in many countries(Biswas, 1981). For example, a relatively simplemodel was used for investigating the effect onwater quality of large-scale water transfers from theriver Severn into the river Thames in the UnitedKingdom of Great Britain and Northern Ireland.The model was used to assess the effect of suchtransfers on the concentration of a number ofconservative and nearly conservative substancescontained in the water. The model was based onriver-flow separation according to source, for example,surface, interface and base flow, and ondeveloping relationships between the concentrationsof the determinants considered and the waterinflow and inflow variation for each source. Thesimulation results matched the recorded datareasonably well.Another example of the practical application of awater quality model for water managementpurposes is the study of the effect of removal ofbiochemical demand loads by waste-treatmentplants on the dissolved oxygen concentration inthe water of the Thames river in Ontario, Canada.The results indicate that obtaining dissolvedoxygen concentrations above the criterionaccepted for good water quality by removingbiochemical demand loads is feasible at one point,while at another point, this would be very difficult.Hydrological Aspects of Accidental Pollution ofWater Bodies (WMO-No. 754) provides a detailedreview of a number of water quality models appliedin Canada, France, Germany, Poland, the UnitedKingdom and the United States to a variety ofrivers having significant pollution problems.Water quality models are also used for computingthe propagation of accidental pollution events.Such models have been operational on the Rhineriver since 1989. While most of the modelsmentioned above primarily consider pollutantsoriginating from industrial and municipal wastes,

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-45some also consider pollution originating fromdiffuse sources such as forestry and agricultureactivities or non-sewered residences.Among the most widely used models is SWAT,which stands for soil and water assessment tool. Itallows the simulation of the fate of nutrients andpesticides migrating to water from diffused sourcessuch as agriculture. SWAT is a watershed-scalemodel developed by Arnold and collaborators forthe Agricultural Research Service of the UnitedStates Department of Agriculture (USDA) to predictthe impact of land management practices on water,sediment and agricultural chemical yields (Arnoldand others, 1993). The model combines significantelements of both a physical and semi-empiricalnature and can be called a process-based model.Hence, it requires specific information aboutweather, soil properties, topography, vegetationand land management practices occurring in thewatershed. The physical processes associated withwater movement, sediment movement, cropgrowth, nutrient cycling and the like are directlymodelled by SWAT using this input data. SWAT is acontinuous time model and is not designed tosimulate in detail single-events such as floods withhourly time steps.The objective of SWAT is to predict the effect ofmanagement decisions on water, sediment, nutrientand pesticide yields with reasonable accuracyon large, ungauged river basins. The model containsthe following components: weather, surface runoff,return flow, percolation, evapotranspiration, transmissionlosses, pond and reservoir storage, cropgrowth and irrigation, groundwater flow, reachrouting, nutrient and pesticide loading, and watertransfer. Interfaces for the model have been developedin Windows (Visual Basic), GRASS andArcView. SWAT has also undergone extensive validation.For further information, see http://www.brc.tamus.edu/swat.A number of follow-up models are based on SWAT.For instance, SWIM, which stands for soil andwater integrated model, was developed byKrysanova and others (1998, 2000) specifically forclimate and land-use change impact assessment inmesoscale and large river basins and at the regionalscale. It includes a three-level disaggregationscheme down to hydrotopes and several modifiedroutines, for example, river routing and forestmodules, and new routines for impact studies suchas a crop generator, climate data interpolation,adjustment of photosynthesis and transpiration tohigher CO 2, nutrient retention and a carbon cyclemodule.There are numerous models which replicatechemical movement in aquifer systems. Some arebespoke for a particular situation and others arelinked to flow models, such as the MT3D link toMODFLOW.In the case of groundwater, modelling water qualityis dependant on understanding the flowregime of the aquifer. Thus, unless the groundwaterflow rates and direction and their variabilityare known, there is little point in attempting tomodel complex chemical changes in the aquifer.However, by understanding of the chemical processesat work within the aquifer and thedistribution of chemical constituents, both naturaland anthropogenic, significant insights can begained on the flow process within the aquifer.Thus the two processes can be used in conjunctionto aid overall calibration.6.3.6.3 Modelling ice formationThe formation of ice in a river begins when thesurface layer of water cools down to 0°C. Below thesurface of the stream, the water temperature at thattime generally remains above 0°C. Thus, forecastingthe date of appearance of ice consists of computingthe heat exchange at the surface of the water so thatthe surface layer of the water will cool to 0°C.Forecasting water temperature should be performedby the stepwise solution of the heat-budget equation,while taking the variables affecting the heatloss into consideration. The heat loss from the watersurface is a function of air temperature, wind speedand turbulence of the water. In its most generalform, the equation of the heat balance at the airwaterinterface for a certain interval of time is asfollows:α (θ – w – θ sw ) + Q = 0 (6.56)where θ – wis the mean temperature of the water massof the stream θ sw, is the water-surface temperature(in °K), α is the coefficient of heat transfer (Watt/m 2 °K) from the water mass to the air-water interfaceand Q is the heat loss from the water surface inWatt/m 2 .The basis of modern short-term forecasts of the dateof initial occurrence of ice on rivers is the methoddeveloped by Hydrometeoizdat (1989). This methodis based on the inequality between the two heatfluxes:α n T wn ≤−Q * Q*mm or T wn ≤ −α n(6.57)

II.6-46GUIDE TO HYDROLOGICAL PRACTICESwhere T wis the mean temperature of water flow, α nis the heat-yield coefficient of the water body, Q m*is the heat loss through the air-water interface andn refers to the time when this inequality appears.The calculation of α n, T w, and Q m* requires knowledgeof several meteorological and hydrologicalvariables. The method can be used if air temperatureforecasts are available several days ahead. Itsaccuracy is affected mostly by errors in the anticipatedair temperature.The necessary condition for the beginning of freezeupis the accumulation of sufficient amounts offloating ice with intensive heat loss so that themerging of ice floes resists the force exerted by theflowing water. This condition is expressed by thefollowing empirical formula:(Q a ) c = − 6.5 v 2 b⎝ ∑ Q a⎛0.8⎞⎠(6.58)where (Q a) cis the critical, or highest possible,mean daily air temperature on the day of freezing,v is the mean velocity of flow in the reach, bis the river width, and ΣQ ais the sum of meandaily temperatures from the first day of ice appearance(Buzin and others, 1989). Calculations aremade with forecast mean daily temperatures foreach day successively until the mean daily airtemperature falls below the critical point (Q a) c, ascalculated in equation 6.55. When the criticalpoint is reached, the formation of a frozen sectionis forecasted.In operational practice, the full version of themodel of ice cover formation, including some formof simplified updating with respect to the specificlocation and hydrometeorological data, is generallynot used. As a rule, model development andapplications are tailored to meet user requirements.Thus, the operation of water-managementschemes under winter conditions should be basedon appropriate reports and forecasts. An iceorientedhydrological network, which can operateaccording to forecasting requirements, needs to beimplemented according to those principles.Regular feedback from the water managers to theforecasting centre is also necessary. For thehydropower production it is important to haveforecasts of the beginning of intensive frazil iceand slush production. Empirical formulae basedon a simplified variant of the theoretical methodare generally developed for this purpose. Ingeneral, the empirical ratios are represented asnomograms, an example of which is provided inFigure II.6.16.Short-term forecasting of ice phenomena isbased on knowledge of the physical or statisticalrelationships that exist as necessary conditions forthe formation of ice (Hydrometeoizdat, 1989). Thephysical interpretation of these relationships isbased on theories concerning the processes thatgovern the cooling of a mass of water in naturallakes. These equations are used to determine a criticalor threshold air temperature, or the sum ofnegative air temperatures, which, when exceeded,results in the occurrence of ice cover on a waterbody. As terms of freezing of a reservoir depend onthe heat content of the mass of water, the criticalair temperature is determined by using an empiricalratio connecting this air temperature with watersupply parameters derived from the water level orstreamflow.6.3.6.4 Modelling ice thicknessIn addition to forecasts of the date of ice occurrenceand the formation of ice cover, other types of forecastsof the autumn ice phenomena are also issued.Ice thickness forecasts are based on calculations ofheat loss. Increases in ice thickness mainly occur onits underside and are determined by the energystate in the water column. Sometimes the ice coverthickness grows from its top surface because of thefreezing of water from the melting of the snowpackon the ice surface. Melt water is often accompaniedwith rainfall. This can also result in an additionalamount of water occurring on the ice cover causedby the increased pressure on the ice. Forecasting icethickness is based on estimates of the differenceAir temperature °C–15–10–50(b)5 10 15Figure II.6.16. Prediction of frazzle-ice formation:(a) slush possible, (b) no slush(a)Wind speed, m s –1

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-47between the inflow of heat from the water columnto the bottom surface of the ice and the outflow ofheat through the ice surface to the atmosphere.Loss of heat will lead to an increase in thickness ofice of depth:Δ h ice =t tB i ∑ C i1 1(6.59)L ice ρ ice∑where Δh iceis the growth in ice thickness intcentimetres, ∑ B iis the flow of heat to the atmospherefrom the top surface of a snow-ice cover, ∑ C i1t1is the inflow of heat to the bottom surface of ice, L iceis the specific heat of ice formation; and ρ iceis thedensity of ice.Formulae for calculating and forecasting ice thicknessunder various conditions of ice cover formationare present in the literature.ice flows of various sizes, which then start to moveas a general drift of ice. The condition for thecommencement of a drift of ice is expressed by aninequality of the following form:ϕd g1/2≤ CU 2 (6.60)where ϕ is the compactness of the melting ice(relative bending stress), d gis the thickness of theice in centimetres, U is the maximum wind speedover a 24-hour period in m s –1 and C is an empiricalcoefficient that depends on wind speed and isa constant for a given reservoir. For a number ofreservoirs in the Commonwealth of IndependentStates, the value of C was found to be 0.018. Thecompactness of the ice ϕ and the thickness d gwhen the ice starts to drift are calculated frommeteorological elements using heat balance equations.Specific information on applying thismethod was provided in Hydrometeoizdat(1989).6.3.6.5 Modelling ice break-upOne method for forecasting the date of icebreak-up is based on estimating the critical sum ofdegree-days of positive air temperature required forbreak-up on the river reach in question. To determinethis sum, the relationship between break-upand the negative degree-day sum for the winterperiod is used. To forecast the date of break-up ofice by this method, it is necessary to have an airtemperature forecast for a few days in advance. Thedate of break-up is obtained by calculating the sumof degree-days and comparing it with the criticalvalue, using expected air temperatures for a fewdays in advance.Forecasts of the reduction of ice thickness andtensile strength of the ice cover and forecasts of icebreak-up for rivers and clarifications from ice ofreservoirs are made with ice cover destructionmodels, such as the models that can be found inHydrometeoizdat (1989) or through the Bulatovmodel. The latter is a method of forecasting icebreak-up dates on rivers that was developed using ageneralized equation, allowing the issuance ofmedium-term forecasts with a lead-time of ten days.It makes it possible to develop forecasts of ice breakupanywhere, including for rivers with only sparsedata (Borsch and others, 1987).6.3.6.5.1 Ice break-up on reservoirsBreak-up of an ice cover on a reservoir results frommelt and a gradual decrease in compactness. Underthe action of wind, the ice may break into separate6.3.6.5.2 Ice break-up on riversForecasting the break-up of ice on rivers can be basedon models in which the condition for the break-upof the ice cover is determined from the thickness andcompactness of the ice and the tractive force of thecurrent. When the forces of resistance become equalto or less than the tractive force, the ice cover breaksup and an ice run begins.The condition for break-up is expressed by thefollowing relationship:ϕd g≤ f(H,ΔH) (6.61)where ϕd g, the product of relative stress of themelting ice and its thickness, is a measure of thecompactness of the ice cover at the time of breakup,and H and ΔH are parameters representing thetractive force of the current. H is the height of thewater level at the time of break-up and reflectsdischarge and speed of flow, and ΔH is the rise, upto the time of break-up, in the water level abovethe minimum winter level H 3, numerically equalto ΔH = H – H 3. As H and ΔH are interrelated inmost cases, it is sufficient to consider just one ofthese quantities in the relationship described inequation 6.61. The quantities are based on forecastand actual data for a few days before break-up. Anapproximation of the relationship may beexpressed as follows:ϕd g≤ a+ b (ΔH) 2 (6.62)where a and b are empirical coefficients.

II.6-48GUIDE TO HYDROLOGICAL PRACTICESFor the forecast of ice break-up dates on ungaugedrivers, or where there are only short periods ofobservations available, a forecasting methodologyhas been developed based on a generalizedequation:(ϕd g) b–i/(ϕd g) N≤ [1–e– (i+1)(Q b–i )/(Q b ) N](Q b–i)/(Q b) N+ 0.005i + 0.25(6.63)where (ϕd g)N is the average relative durability ofthe ice on the day of ice break-up, (ϕd g) b–iis the relativedurability of ice for i days before the icebreak-up, Q b–iis the water discharge for i days beforethe ice break-up, (Q b) Nis the average discharge onthe day of the ice break-up. Calculation and forecastof (ϕd g)N, (Q b) N, and d gare made using speciallydeveloped maps, nomograms and tables (Borschand Silantjeva, 1987).The model of ice cover break-up allows the developmentof some additional special forecasts, such asthe forecast of maximum permissible loading forice and forecasts tailored for the deployment oficebreakers.6.4 MODELLING CHALLENGES6.4.1 Accuracy and availability of inputdataA modelling challenge, related to ungauged basins,is the need to improve the availability and accuracyof the data used in models. This may include theinput time series of data such as rainfall and evaporation,and the time-series data used to calibrate orvalidate model results such as streamflow, groundwaterlevels and water quality data, as well as theinformation that is used to estimate model parametervalues. If hydrological models are to realize theirtrue potential as operational water resourcesmanagement tools, it is essential that the informationrequired to apply them successfully be available.The use of processed satellite imagery withinmodelling research projects has been reported for anumber of years and there are examples of suchtechnology being used for operational purposes.However, there is a tremendous potential for themore widespread use of these techniques by waterresources management agencies, especially in thedeveloping world where ground-based observationsare not being sustained.Global, or near-global, datasets of a wide range ofterrestrial information derived from satelliteimagery are becoming increasingly available andaccessible. The information available includes relativelystatic characteristics such as topography, landcover (d’Herbès and Valentin, 1997), as well as timeseriesvariations of parameters such as temperature(Xiang and Smith, 1997), evapotranspiration (Kiteand Droogers, 2000), soil moisture (Valentijn andothers, 2001) and precipitation (WMO-WCRP,1986). Many of these have the potential to fill someof the information gaps and provide input data forwater resources estimation models. However, severalpractical considerations need to be addressed ifsuch products are to be used successfully and withconfidence:(a) Hydrological models, calibrated against historicalgauged data, may already be in use;(b) Satellite data have relatively short periods ofrecord;(c) Ideally, gauged and satellite data need to be usedtogether; therefore, the relationships betweenthe two data sources need to be quantified andclearly understood;(d) Data should be accessible to water resourcespractitioners in the developing countries;(e) The techniques required to make effective useof the data should not be excessively complexor difficult to understand, as the resourcesavailable for data analysis and processing arefrequently limited in developing countries.One of the future challenges in hydrologicalmodelling will be to expand the operational use ofthese techniques. The implication of this is theneed to ensure that the results are as accurate andrepresentative as possible. At the very least, it isessential to understand the limitations and errorbounds of the model results so that water resourcesdevelopment decisions can be made on the basisof adequate information. The following subsectionrefers to one of the objectives of theInternational Association of HydrologicalSciences’s initiative for the Decade on Predictionin Ungauged Basins: a reduction in predictiveuncertainty. For operational uses of models, it isnot only important to achieve a reduction inuncertainty, but also to be able to quantify thatuncertainty by having a thorough understandingof the accuracy of the input data.6.4.2 Ungauged basinsDrainage basins in many parts of the world areeither ungauged or inadequately gauged and thesituation is worsening because the existing observationnetworks are in decline. At the same time,water resources are under growing threat in a worldthat is becoming more populated and where the

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-49demand for water per capita is constantly increasing.Therefore, as the supply of data declines, theneed for such data increases. This poses a considerablechallenge to the hydrological and waterresources communities: that of finding the meansto assess and manage water resources with an inadequatesupply of data.Recognition of the need for techniques applicableto ungauged basins is not new. The nineteenthcenturyrational formula, based on the concept of arunoff coefficient, can be regarded as a precursor ofregionalization. Extrapolation from gauged toungauged sites to solve hydrological problems hasbeen a standard technique in practice. This chaptercontains several examples. The synthetic unithydrographs and geomorphoclimatic unithydrographs mentioned in 6.3.2.2.5 allow a modellerto estimate a runoff hydrograph for areas withfew, if any rainfall and runoff data. An estimationproblem is presented in 6.2.3, where an introducionto geostatistics is provided to estimate a valueof the variable in an ungauged location, based on anumber of values of this variable measured in otherlocations.An example of regionalization conducted on anational scale is set out for the United Kingdom inthe Flood Estimation Handbook, published in 1999,which superseded the Flood Studies Report and itssupplements. The Handbook explains regionalizationof model parameters and extrapolation fromgauged to ungauged catchments. It is recommendedto conduct flood frequency estimation by statisticalanalysis of peak flow, annual maxima or peak-overthreshold,or by using a rainfall-runoff approach, ifsufficiently long data records are available. Whileflood data at the subject site are of greatest value,data may be transferred from a nearby site, a donorcatchment, a similar catchment or an analogouscatchment if there is no donor catchment nearby.Estimation of the index flood – the median annualflood – in the absence of flood peak data can bedetermined from catchment descriptors. Pooledanalysis may be needed for growth curve estimation,dependent on the length of gauged record orthe target period such as a 100-year or 10-year flood.The last choice is to estimate parameters for a rainfall-runoffmodel using only catchmentdescriptors.The International Association of HydrologicalSciences, which launched the Decade onPrediction in Ungauged Basins, 2003–2012, aimsto achieve major advances in the capacity to makepredictions in ungauged basins (Sivapalan andothers, 2003). It is hoped that the Decade willbring a reduction in predictive uncertainty andcontribute to the development of new theoriesbased on scaling and multi-scaling, complexsystems approach, non-linear dynamics and ecohydrologicalrelationships. This cannot be donewithout extending the range and scale of observationsused in estimating hydrological variables.The initiative is of considerable interest to operationalhydrology, and it is hoped that, by the endof the decade, the toolkit of operational techniquesfor dealing with ungauged basins will havegrown considerably.6.4.3 Coupling of modelsWith an increasing emphasis on integrated waterresources management, it is often necessary tomake use of several models to solve practicalwater resources problems. Examples mightinclude the combined use of water quantity andquality models with systems models and economicimpact models. A further example is the use ofclimate models to generate meteorological inputsto models of basin hydrology. In the past this hasbeen achieved by modelling the different processesseparately in series and using the outputs ofone model as inputs to the next. This approachhas the potential of ignoring many of the feedbacksthat exist in complex natural systems. Abetter approach is to run the models in parallel,whereby the links between processes are coupledat each time step of the simulation and feedbackmechanisms are included. Using traditionalmethods, this involves combining all the algorithmsof the separate models into a single model,a substantial development task that precludes theflexibility of selecting different modellingapproaches for specific applications. The couplingof models can be facilitated by the developmentof modelling frameworks that integrate themanagement of data, geograpical informationsystem visualization tools and model links into asingle software package that includes severalmodels. There are a number of such systems availableworldwide, all of which have been developedfor different purposes. Examples can be found athttp://www.epa.gov/waterscience/basins/bsnsdocs.html and Hughes (2004b).A recent innovation, the Open ModellingInterface (OpenMI – see http://www.harmonit.org) represents an attempt to allow models simulatingdifferent water-related processes to belinked on a temporal and spatial basis and thuspermit the simulation of process interactions.The objective is to simplify the linking of modelsrunning in parallel, and operating at different

II.6-50GUIDE TO HYDROLOGICAL PRACTICEStemporal and spatial scales, through the directtransfer of data between the models. Many existingmodels are expected to become OpenMIcompliant in the near future.References and further readingAndersen, T., 1991: AVHRR data for snow mapping inNorway, Proceedings of the 5th AVHRR Data UsersMeeting, Tromsoe, Norway.Anderson, E.A., 1973: National Weather Service RiverForecast System: Snow Accumulation and AblationModel, Programs and Test Data. NOAA NWSHYDROTechnical Memorandum 17.Anderson, M. P., D.S. Ward, E.G Lappala and T.A.Prickett, 1992: Computer models for subsurface water.Chapter 22 in: Maidment, D. R. (ed.) Handbook ofHydrology, McGraw-Hill, 22–1 to 22–34.Arnold, J.G., P.M. Allen and G. Bernhardt, 1993: Acomprehensive surface-groundwater flow model.Journal of Hydrology. 142:47–69.Baumgartner, M. F., 1988: Snowmelt runoff simulationbased on snow cover mapping using digital Landsat-MSS and NOAA/AVHRR data, USDA-ARS, HydrologyLab. Tech. Rep.Bear, J., 1980: Hydraulics of Groundwater. New York, NY:McGraw-Hill College, 1980. ISBN: 0070041709.Bear, Jacob. 1988: Dynamics of Fluids in Porous Media.New York, NY: Dover Publications, 1988. ISBN:0486656756.Beldring, S., K. Engeland, L.A. Roald, N.R. Saelthunand A. Voks, 2003: Estimation of parameters in adistributed precipitation – runoff model for Norway.Hydrology and Earth System Sciences, 7(3):304–316.Bergstörm, S., 1992: The HBV model – its structure andapplications. SMHI Reports RH, No. 4, Norrkping,Sweden.———, 1995: The HBV model. In Singh, V.P. (ed):Computer Models of Watershed Hydrology, WaterResources Publications. Colorado, United States, 443,476.Bergstörm, S., J. Harlin and G. Lindstrm, 1992: Spillwaydesign floods in Sweden. I: New guidelines. HydrologicalSciences Journal, 37(5):505–519.Bergström, S., B. Carlsson, M. Gardelin, G. Lindstrm, A.Pettersson and M. Rummukainen, 2001: Climatechange impacts on runoff in Sweden – assessmentsby global climate models, dynamical downscalingand hydrological modelling. Climate Research,16(2):101–112.Beven, K. J., 1996: A discussion of distributed hydrologicalmodelling. In: M.B. Abbott and J. Ch. Refsgaard,(eds) Distributed Hydrological Modelling, WaterScience and Technology Library, Vol. 22, Kluwer,Dordrecht, 255–278.Biswas, A.K. (ed.), 1981: Models for Water QualityManagement. McGraw-Hill, New York.Borsch S.V. and T.P. Silantjeva, 1987: Method of theshort-term and medium-term forecast of openingof the rivers on the basis of the generalizeddependence. The methodical instructions. Moscow,Hydrometeorological Center of the Union of SovietSocialist Republics. pp. 28.Box, G.E.P. and G.M. Jenkins, 1970: Time Series Analysis,Forecasting and Control, San Francisco, Holden-Day.Brun, E., P. David, M. Sudak and G. Brunot, 1992, Anumerical Model to Simulate snow-cover stratigraphyfor operational avalanche forecasting, Journal ofGlaciology, 38(128), 13–22.Carroll, S.S., G.N. Day and T.R. Carroll, 1993:Incorporating airborne data into spatial model usedto estimate snow water equivalent, GeographicInformation Systems and Water Resources, Journal ofthe American Water Resources Association, March,pp. 259–264.Cigizoglu, H. K., 2003: Estimation, forecastingand extrapolation of river flows by artificialneural networks. Hydrololgoical Science Journal,48(3):349–361.d’Herbès, J.M. and C. Valentin, 1997: Land surfaceconditions of the Niamey region: ecological andhydrological implications. Journal of Hydrology,188–189, 18–42.DeWeist, J.M., 1965: Geohydrology. John Wiley & Sons,366 pp.DHI (Danish Hydraulic Institute), 1985: Introduction tothe SHE-European Hydrologic System, Horsholm.Domenico, Patrick A. and W. Scgwartz Franklin, 1998:Physical and Chemical Hydrogeology. New York, NY:John Wiley & Sons Inc., 1998. ISBN: 0471597627.Donald J.R., E.D. Soulis, N. Kouwen, A. Pietroniro, 1995:A land cover-based snow cover representation fordistributed hydrologic models. Water ResourcesResearch. 31, No. 4: pp. 995–1009.Dooge, J.C.I., 1973: Linear theory of hydrologic systems.Technical Bulletin No. 1468, Agricultural ResearchService, United States Department of Agriculture,Washington, DC.Eagleson, P.S., 1970: Dynamic Hydrology, New YorkMcGraw-Hill.Engman, E.T. and R.J. Gurney, 1991: Remote-sensing inhydrology, London, Chapman and Hill, 225 pp.Environment Agency, 2002: Groundwater ResourcesModelling, Guidance Notes and Template ProjectBrief, R & D Guidance Notes W213, June 2002.Fetter, C. W., 1998: Contaminant Hydrogeology.2nd ed. Upper Saddle, NJ, Prentice Hall,ISBN: 0137512155.Freeze, Alan R. and A. Cherry John, 1979: Groundwater.Englewood Cliffs, NJ: Prentice Hall. ISBN:0133653129.

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-51Hedstrom N.R. and J.W. Pomeroy, 1998: Measurementsand modelling of snow interception in the borealforest. Hydrological Processes 12(10–11):1611–1625.4.Hillel, Daniel, 1980: Fundamentals of Soil Physics. NewYork, NY, Academic Press, 1980. ISBN: 0123485606.Hipel, K.W., A.L. McLeod and W.C. Lennox, 1977:Advances in Box-Jenkins modelling. Part I:Model construction. Water Resources Research,13:567–575.Hirsch, R.M. and J.R. Slack, 1984: A nonparametric trendtest for seasonal data with serial dependence: WaterResources Research, pp. 727–782.Hughes, D.A., 2004a: Incorporating groundwaterrecharge and discharge functions into an existingmonthly rainfall-runoff model. Hydrololgoical ScienceJournal, 49(2):297–311.Hughes, D.A. (ed.), 2004b: SPATSIM, an integratingframework for ecological reserve determinationand implementation. Water Research CommissionReport No. TT 245/04, United States.Hughes, D.A. and W. Metzler, 1998: Assessment ofthree monthly rainfall-runoff models for estimatingthe water resource yield of semiarid catchmentsin Namibia, Hydrololgoical Science Journal,43(2):283–297.Hurst, H.E., 1951: Long-term storage capacity ofreservoirs. Transactions, American Society of CivilEngineers, 16:770.Hydrometeoizdat, 1989: Guide on the hydrologicalforecasts. The ice forecasts. Vol. 3. Leningrad,Hydrometeoizdat, pp. 168 (in Russia).Institute of Hydrology, 1999: Flood Estimation Handbook(FEH): (in five volumes), Wallingford, Oxfordshire,United Kingdom.Kalinin, G. P. and P.I. Miljukov, 1958: Approximate methodsfor computing unsteady movement of water masses.Transactions, Central Forecasting Institute, Issue 66(in Russian)Keller, M., 1987: Ausaperungskartierung mitLandsat-MSS Daten zur Erfassung oekologischeEinflussgnessen in Gebirge, Remote-sensing Series,No. 10, PhD Thesis, University of Zurich.Kitanidis, P.K., 1992: Analysis of Macrodispersionthrough Volume–Averaging: Moment Equations,Stochastic Hydrology and Hydraulics, 6:5–25.Kite, G.W. and P. Droogers, 2000: Comparingevapotranspiration estimates from satellites, hydrologicalmodels and field data. Journal of Hydrology,229(1–2):3–18.Krysanova, V., Müller-Wohlfeil D.I., and Becker A., 1998:Development and test of a spatially distributedhydrological/water quality model for mesoscalewatersheds. Ecological Modelling 106:261–289.Krysanova, V., E. Wechsung, J. Arnold, R. Srinivasan andJ. Williams, 2000: SWIM (Soil and Water IntegratedModel). User Manual, PIK Report 69. PotsdamInstitute for Climate Impact Research.Kundzewicz, Z. W. and A. Robson, 2004: Changedetection in hydrological records – a review ofthe methodology. Hydrololgoical Science Journal,49(1):7–19.Leaf, C.F., and G.E. Brink, 1973: Hydrologic simulationmodel of Colorado subalpine forest, US Forest Service,Res.Pap. RM-107, Fort Collins, CO.Leavesley, G.H. and L.G. Stannard, 1990: Applicationof remotely sensed dataon a distributed parameterwatershedmodel, Proceedings of the Workshop onApplications of Remote-sensing in Hydrology, Editors:Kite, G.W. and Wankiewicz, A., National HydrologyResearch Institute, Saskatoon, Saskatchewan,Canada.Lindström, G., 1997: A simple automatic calibrationroutine for the HBV model. Nordic Hydrology,28(3):153–168.Lindström, G. and J. Harlin, 1992: Spillway design floodsin Sweden. II: Applications and sensitivity analysis.Hydrololgoical Science Journal, 37(5):521–539.Lindström, G. and S. Bergström, 1992: Improving theHBV and PULSE-models by use of temperatureanomalies. Vanneti Norden, 25(1):16–23.Lindström, G., Johansson, B., Persson, M., Gardelin,M. and Bergström, S., 1997: Development and test ofthe distributed HBV-96 model. Journal of Hydrology.201:272–288.Maidment, D. R. 1992: Handbook of Hydrology. New York,McGraw-Hill.Mandelbrot, B.B. and J.R. Wallis, 1968: Noah, Joseph,and operational hydrology. Water ResourcesResearch, 4(5):909–918. Available at the addresshttp://www.math.yale.edu/mandelbrot/web_pdfs/051noahJoseph.pdfMarks, D., 1988: Climate, energy exchange and snowmelton Emerald Lake watershed, Sierra Nevada,Unpublished PhD Dissertation, University ofCalifornia, Santa Barbara, CA, 149 pp.Marsily, Ghislain De., 1986: Quantitative Hydrogeology:Groundwater Hydrology for Engineers. Orlando, FL,Academic Press, 1986. ISBN: 0122089162.Martinec, J. and A. Rango, 1987: Interpretation andutilization of areal snow cover data from satellites,Annual Glaciology, 9:166–169.McDonald, M.G. and A.W. Harbaugh, 1988: A modularthree-dimensional finite-difference groundwaterflow model: US Geological Survey, Techniques ofWater Resources Investigations, Book 6, Chapter Al,586 pp.Mejia, J.M., I. Rodriguez-Iturbe and D.R. Dawdy, 1972:Streamflow simulation, (2) The broken line processas a potential model for hydrologic simulation.Water Resources Research, 8(4):931–941.Miller, W.A., M.B. Shasby, W.G. Rhode andG.R. Johnson, 1982: Developing inplace data basesby incorporating digital terrain data into the Landsatclassification process. Place Resource Inventories:

II.6-52GUIDE TO HYDROLOGICAL PRACTICESprinciples and practices, Proc. Workshop, August4–14, 1981, University of Maine, Orono, Sponsoredby the American Society of Photogrammetry.Minns, A.W. and M.J. Hall, 1996: Artificial neuralnetworks as rainfall-runoff models, HydrololgoicalScience Journal, 41(3):399–417.Ostrem, G., T. Anderson and H. Odegaard, 1991:Operational use of satellite data for snow inventoryand runoff forecast, Satellite Hydrology, AmericanWater Resources Association, Minneapolis, MN,pp. 230–234.Pomeroy J.W. and E. Brun, 2001: Physical Propertiesof Snow. In Snow Ecology: An InterdisciplinaryExamination of Snow-covered Ecosystems. H.G.Jones, J.W. Pomeroy, D.A. Walker and R.W. Hoham(eds). Cambridge University Press: Cambridge;pp. 45–126.Price, Michael, 1985: Introducing Groundwater. GeorgeAllen & Unwin: London, United Kingdom, 1985.195 pp.Rango, A., 1993: Snow hydrology processes and remotesensing,Hydrological Processes, 7:121–138.Rango, A., and van Katwijk. 1990: Development andtesting of a snowmelt-runoff forecasting technique,Water Resources Bulletin, 26:135–144.Rango, A., V.V. Salomonson and J.L. Foster, 1977:Seasonal streamflow estimation in the Himalayanregion employing meteorological satellite snowcover observations, Water Resources Research,13:109–112.Refsgaard, J.C. and M.B. Abbott, 1996: The role ofdistributed modelling in water resources management.In: M.B. Abbott and J. Ch. Refsgaard, (eds),1996: Distributed Hydrological Modelling, WaterScience and Technology Library, Vol. 22, Kluwer,Dordrecht.Robson, A.J., T.A. Jones and D.W. Reed, 1998 Astudy of national trend and variation in UnitedKingdom floods. International Journal of Climatology,18:165–182.Rodriguez-Iturbe, I. and B. Valdes, 1979: The geomorphologicstructure of hydrologic response, WaterResources Research, 15(6):1409–1420.Rodriguez-Iturbe, I., J.M. Mejia and D.R. Dawdy, 1972:Streamflow simulation, (1) A new look at Markovianmodels, fractional Gaussian noise, and crossingtheory. Water Resources Research, 8(4):921–930.Sahimi, Muhammad, 1995: Flow and Transport in PorousMedia and Fractured Rock. Weinheim, Germany:Wiley-VCH Verlag GmbH, 1995. ISBN: 3527292608.Salas, J.D., 1992 Analysis and modelling of hydrologicaltime series. Chapter 19 in: Maidment, D. R. (ed.)Handbook of Hydrology, Mc Graw-Hill, 19–1 to 19–72.Scheidegger, Adrian, 1960: The Physics of Flow throughPorous Media, University of Toronto Press.Shaw, E.M., 1994: Hydrology in Practice, third edition.London, Chapman and Hall.Singh, V.P., 1988: Hydrologic Systems. Rainfall-RunoffModeling, Vol. I, Englewood Cliffs, Prentice Hall.Sivapalan, M., K. Takeuchi, S. Franks, D. Schertzer, P.E.O’Connell, V.K. Gupta, J.J. McDonnell,J. W. Pomeroy, S. Uhlenbrook, E. Zehe and V.Lakshmi, 2003: IAHS Science Decade on Predictionin Ungauged Basins (PUB), 2003–2012: Shapingan exciting future for the hydrological sciences.Hydrololgoical Science Journal, 48(6):857–880.SNA, 1995: Climate, Lakes and Rivers, The National Atlasof Sweden, Almqvist and Wiksell International,Stockholm.Storm, B. and A. Refsgaard, 1996: Distributed physicallybased modelling of the entire land phase ofthe hydrological cycle. In: M.B. Abbott and J. Ch.Refsgaard, (eds): Distributed Hydrological Modelling,Water Science and Technology Library, Vol. 22,Kluwer, Dordrecht.Sugawara, M., E. Ozaki, L. Watanabe and S. Katsuyama,1974: Tank Model and its Application to Bird Creek,Wollombi Brook, Bikin River, Kitsu River,Sanaga River and Nam Mune. Research Noteof the National Research Center for DisasterPrevention, Science and Technology Agency,Tokyo, Japan.Tayfur, G., 2002: Artificial neural networks for sheetsediment transport. Hydrololgoical Science Journal,47(6):879–892.Todd, David Keith, 1980: Groundwater Hydrology.New York, NY. John Wiley & Sons Inc. ISBN:047187616X.Todini, E. and J.R. Wallis, 1978: A Real-time RainfallRunoff Model for an On-line Flood Warning System.AGU Chapman conference on applications ofkalman filtering theory and techniques to hydrology,hydraulics and water resources, Pittsburgh,Pennsylvania, 22–24 May 1978.Twedt, T. M., J.C. Schaake and E.L. Peck, 1977: NationalWeather Service extended streamflow prediction.Proceedings of the western snow conference,Albuquerque, New Mexico, 19–21 April 1977.US Army Corps of Engineers, 1960: Runoff from Snowmelt.Engineer Manual 1110-2-1406, US Department ofthe Army, Washington, DC.US Environmental Protection Agency, 1989: EPAGroundwater Handbook. Rockville, Maryland,Government Institutes, Inc. 212 pp.Valentijn R.N.P., R. Hoeben, N.E.C. Verhoest and E.P. DeTroch, 2001: The importance of the spatial patternsof remotely sensed soil moisture in the improvementof discharge predictions for small-scale basinsthrough data assimilation. Journal of Hydrology,251(1–2):88–102.Walton, W., 1970: Groundwater Resource Evaluation,McGraw-Hill, 664 pp.World Meteorological Organization, 1986:Intercomparison of Models of Snowmelt Runoff.

CHAPTER 6. MODELLING OF HYDROLOGICAL SYSTEMSII.6-53Operational Hydrology Report No. 23,WMO-No. 646, Geneva.———, 1987: Intercomparison of Conceptual Models Usedin Operational Hydrological Forecasting. OperationalHydrology Report No. 7, WMO-No. 429, Geneva.———, 1988: Analysing Long Time Series of HydrologicalData with respect to climate variability and change,WCAP-3, WMO/TD-No. 224, Geneva.———, 1990: Hydrological Models for Water-ResourcesSystem Design and Operation. Operational HydrologyReport No. 34, Geneva.———, 1991a: Simulated Real-time Intercomparison ofHydrological Models. Operational Hydrology ReportNo. 38, WMO-No. 779, Geneva.———, 1991b: Hydrological Aspects of AccidentalPollution of Water Bodies. Operational HydrologyReport No. 37, WMO-No. 754, Geneva.———, 1992a: A study of current approaches to modellingof hydrological times series with respect to climaticvariability and change (G. Cavadias). World ClimateProgramme – Water, Project A2. WCASP–23 WMO/TD-No. 534, Geneva.———, 1992b: Snow Cover Measurements and ArealAssessment of Precipitation and Soil Moisture.Operational Hydrology Report No. 35, WMO-No. 749, Geneva.———, 1994: Applications of remote-sensing bysatellite, radar and other methods to hydrology,Operational Hydrology Report No. 39,WMO-No. 804, Geneva.——— , 2000: Detecting Trend and Other Changes inHydrological Data. World Climate Programme– Water (Z.W. Kundzewicz and A. Robson, eds)World Climate Programme Data and Monitoring,WCDMP-45, WMO/TD-No. 1013, Geneva.———, 2003: Integration and Coupling of HydrologicalModels with Water Quality Models: Applications inEurope (B. Arheimer and J. Olsson). WMO TechnicalReports in Hydrology and Water Resources, No. 75.WMO/TD-No. 1174. Geneva.———/WCRP, 1986. Report of the Workshop on GlobalLarge Scale Precipitation Datasets for the World ClimateResearch Programme. WCP-111, WMO/TD-No. 94,WMO, Geneva, 45pp.Xiang, X. and E.A. Smith, 1997: Feasibility of simultaneoussurface temperature-emissivity retrieval usingSSM/I measurements from HAPEX-Sahel. Journal ofHydrology, 188–189, 330–360.Yue, S. and P. Pilon, 2004: A comparison of the powerof the t test, Mann-Kendall and bootstrap testsfor trend detection. Hydrololgoical Science Journal,49(1):21–37.Yue, S., P. Pilon and B. Phimney, 2003: Canadianstreamflow trend detection: impacts of serial andcross-correlation. Hydrololgoical Science Journal,48(1):51–63.Yue, S., P. Pilon and G. Cavadias, 2002: Power of theMann-Kendall and Spearman’s rho tests for detectingmonotonic trends in hydrological series. Journalof Hydrology, 259:254–271.

CHAPTER 7HYDROLOGICAL FORECASTING7.1 INTRODUCTION TO HYDROLOGICALFORECASTING[HOMS J]7.1.1 ScopeA hydrological forecast is the estimation of futurestates of hydrological phenomena. They are essentialfor the efficient operation of water infrastructureand the mitigation of natural disasters such asfloods and droughts. In addition, they are becomingincreasingly important in supporting integratedwater resources management and reducing floodinducedlosses.Describing and predicting future water states can becategorized on the basis of how far into the futurethe event is forecast to occur. For instance, forecastsfor various hydrological elements such as discharges,stages and velocities can be made from the start ofthe forecast up to different times in the future. TheTechnical Regulations provide the followingclassification:(a) Short-term hydrological forecasts, which covera period of up to two days;(b) Medium-range hydrological forecasts, whichapply to a period ranging from 2 to 10 days;(c) Long-range hydrological forecasts, which referto a period exceeding 10 days.This section discusses the importance and necessityof establishing an end-to-end hydrological forecastingprogramme, while 7.1.5 provides an introductionto the communications technology used to collectdata and distribute critical forecasts and warnings toits users, and 7.2 describes the data requirements forhydrological forecasting. An overview is provided in7.3 of the various forecasting techniques available,from simple index models to robust hydrologicalforecasting systems. Forecasting of flash floods (see7.4) and snowmelt (see 7.6) have been dealt with ingreater detail because there is a need for guidancematerial on those issues. Finally, water supply forecastsare covered briefly in 7.5. The discussion ofhydrological forecasts in this chapter will be limitedto predicting quantities of water.7.1.2 Hydrological forecast operationsA hydrological forecasting service is composed oftrained hydrological forecasters working with acombination of real-time and historical datainputs, which can include use of radar and satelliteas well as in situ data, communications hardwareand software, hydrological models or modellingsystems, meteorological models or model productor inputs and computer hardware. There are manyways to configure a hydrological forecastingservice. There are, however, a critical number offactors that are necessary to ensure reliable deliveryof a service meeting the needs of a diverse usercommunity.The operations concept of a hydrological forecastingservice defines how the operational forecastservice will operate on a day-to-day basis, as well asduring fooding conditions. It covers the followingpoints:(a) The mission and legal mandate of theorganization;(b) The users and the required products orservices;(c) Deadlines for dissemination;(d) How the hydrological forecasting service isorganized;(e) The hydrometeorological data network andhow it operates;(f) How the hydrologist will interact with themeteorological forecasting office;(g) Communications hardware and software usedto receive data and information as well asdisseminate forecasts;(h) How forecast products are produced;(i) What policies and standard operating procedureswill be followed to ensure best practicesduring routine and emergency conditions;(j) The outreach of the hydrological forecastingservice through the education and training ofpolicymakers, emergency operations staff andthe general public.Sample products should be readily available forpotential customers.The mission and legal mandate of the hydrologicalforecasting service needs to be clearlydefined. It is important that only one officialsource of forecast and warnings be authorized bylaw. Multiple sources of forecasts can result inconflicting information that produces confusionand reduces the possibility of effectiveresponse.

II.7-2GUIDE TO HYDROLOGICAL PRACTICESThe principal users of warning products are national,regional and local emergency management or civildefence organizations, the media, agriculture,industry, hydropower organizations, flood controlmanagers, water transportation and municipalwater supply organizations and the public. Therequirements of hydrological data, forecast productsand warnings vary according to the targeteduser community. It is essential for the hydrologistto understand user requirements so that data andforecast products can be tailored to meet theirneeds. There are many segments of a national economy,such as transportation, emergencymanagement, agriculture, energy and water supply,that have unique needs for such information.Recognizing these needs and providing data, forecastsand products to meet them ensures that thehydrological forecasting service is of greatest benefitto the community. Sophisticated users, such ashydropower organizations, require hydrometeorologicaldata, forecasts, inflow hydrographs andanalyses to support the generation of electricity,while emergency management operations requiresimpler but more urgent forecasts and warnings.The network, including stream gauges, precipitationgauges and the associated meteorologicalnetwork, should be defined, taking into accountthe availability of data from all sources, such as theradar network and satellite downlink products.However, the continuous availability of such productsmust be established before they are used on aregular basis in national hydrological forecastingservices. Close cooperation between meteorologicalforecasting services and hydrological forecastingservices is essential. The procedure, or system definition,for the acquisition of data and forecasts, aswell as analysis, are needed as input to hydrologicalforecasts and should be defined in the operationsconcept. Communications hardware and softwareused in flood forecasting systems depend on theinfrastructure available in the country concerned.However, modern data communication systems,such as satellite and the Internet, provide a varietyof choices and should be utilized appropriately.It is important to assess staff requirements such asthe number of technicians or professionals neededto run the centre during routine and emergencyoperations. Their roles and responsibilities, workinghours and the continuous training needs offorecasters should also be addressed.Hydrological forecasting programmes must be reliableand designed to operate during the most severefloods. The greatest benefits for an effective hydrologicalforecasting programme occur when floodingis severe, widespread and/or sudden. Normally,there is a greater strain on resources during extremeevents such as floods. The operation of the centreduring extreme events must be well defined. Insuch instances, there is generally an increase in dataflow and staffing needs, as more products must bedelivered to more users with short deadlines.Frequently the hours of operation must be expandedto meet higher demands for service.During routine conditions, the staff of a hydrologicalforecasting service collect data andquality-control information, receive and analysemeteorological forecasts, run hydrological modelsand forecasting systems, assess present and futurehydrological conditions and produce forecast productsfor distribution to users. During non-forecastingportions of the day, hydrologists update data suchas rating curves, evaluate operational performance,re-calibrate models and seek further means ofimproving the accuracy and timeliness of futureforecasts.It is never possible to achieve continuous,100 percent reliability of hardware, software and/or powerfor operations even with dependable maintenanceprogrammes. Therefore, a hydrological forecastingservice must establish backup procedures to safeguardfuture operations of all components: datacollection; forecasting system operations, includingbackup of hardware, software and data; forecastdissemination and other communications systems;power, uninterruptible power supply and backupgenerators; and provision of an alternate site foroperations if the location of forecast centre itself isdamaged.The key to making a forecast centre operationallyreliable is to establish a robust maintenanceprogramme. Unfortunately this can be an expensiveundertaking, especially if the network is spread outand difficult to access. All hardware and softwaremust be routinely maintained, otherwise the systemmay not function when most needed. In somecountries, the hydrological forecasting serviceincludes a system administrator, who is responsiblefor maintaining the communications andforecasting system.7.1.3 End-to-end hydrological forecastingsystemsToday’s hydrological forecasting systems areaffordable and powerful. The degree of success inusing these systems generally depends on theamount of training received by the hydrologistsemploying them. These systems are capable of

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-3producing forecasts for floods that occur in a fewhours to seasonal probabilistic outlooks manymonths in advance for larger river basins.Establishing a viable hydrological forecasting andwarning programme for communities at riskrequires the combination of meteorological andhydrological data, forecast tools and trained forecasters.Such a programme must provide sufficientlead time for individual communities in the floodplainto respond. In case of flood forecasts, leadtime can be critical in reducing damage and loss oflife. Forecasts must be sufficiently accurate topromote confidence so that communities and userswill take effective action when warned. If forecastsare inaccurate, credibility is reduced and an adequateresponse is not made.Experience and lessons from the past have demonstratedthat an end-to-end hydrological forecastingand response system (see Figure II.7.1) consists ofthe following steps, which must be linked to achievereduction in flood losses:(a) Data collection and communication;(b) Hydrological forecasting and forecast productgeneration;(c) Dissemination of forecasts to users;(d) Decision-making and support;(e) Action taken by users.The interaction of the technological components ofthe integrated end-to-end hydrological forecastingsystem can be represented as a chain composed ofmany links. Each link must be fully functional tobenefit the user community or population at risk. Aswith links in a chain, should one link not be functioningproperly, the entire system breaks down. Inother words, if a perfect flood forecast is generatedbut does not reach the population at risk, or no capabilitiesto take preventive action exist, then theforecast system does not serve its desired purpose.7.1.4 Uncertainity and probabilisticforecastsIn general, the primary objective of hydrologicalforecasting is to provide maximum lead time withsufficient accuracy so that users may take appropriateaction to mitigate losses or optimize watermanagement decisions. All forecasts containuncertainty and one of the most successful ways ofdealing with this is the use of ensembles. Theuncertainty associated with a hydrological forecaststarts with the meteorology. Given that all mesoscaleatmospheric models attempt to model an essentiallychaotic atmosphere, meteorology has been seen asthe primary source of uncertainty for some years. Inaddition, hydrological model parameters and themodel mechanics also contribute to the associateduncertainty or error in forecasts. Adequacy of datais generally the main limiting factor. If only observedhydrological data are used to generate forecasts,lead times may be so short that the utility offorecasts to users is of little value. By couplinghydrological models with meteorological forecaststhat are the result of meteorologists implementingglobal and regional numerical weather predictionmodels and accounting for local climatologicalconditions, streamflow forecasts can be extendedfrom many days to weeks in the future. Althoughcoupling of models can indeed extend the lead timefor users, it also increases forecast uncertainty.Climatological or seasonal forecasting has nowbecome a useful tool for managing water and reducingthe risk of flooding. Extreme events arecorrelated with major changes in atmospheric andocean circulation patterns; once such patterns canbe identified, the potential for a lesser or greaterdegree of storm activity can be forecast. This informationcan then be used to improve emergencyresponse and increase the degree of readiness offorecasting agencies.GIS toolsMathematical modelsData Communication ForecastDecisionsupportNotification Coordination ActionsSense wateravailabilityGet datawhereneededFuturewateravailabilityWatermanagementand floodcontroldecisionsAppropriateindividualsandgroupsTasks fromresponseplansOrganizationsCivil societyRelocationSand baggingOther responsesFigure II.7.1. Integrated flood forecasting, warning and response system in integrated water resourcesmanagement: a critical chain of events and actions

II.7-4GUIDE TO HYDROLOGICAL PRACTICESWhen the probability of an extreme floodingevent is forecast to be greater than normal, certainmeasures can be taken in anticipation of theevents, for example, stockpiling sandbags, emergencyfood and water supplies, and movinghigh-value stored crops or goods from floodproneareas. This is also a good time to createawareness among the public as to the potentialfor flooding, highlighting the actions that thepublic and others should take, and to carry outemergency-response exercises to test the degreeof readiness. In some cases, emergency measuressuch as the temporary raising of flood protectionbarriers may be warranted. Recent developmentsin computing power have allowed global andregional atmospheric models to increase theirspatial resolution. Local area non-hydrostaticmodels, for instance, have been successfullyreduced to a spatial scale of approximately onekilometre. In addition, smaller-scale processes,such as convection and orographic enhancement,have been modelled more effectively.Probability forecasting should not be confusedwith forecast error. The latter is internal to themodel and data, and represents the error causedby model inadequacy and data error. Perhapsthe best way to distinguish between them is toview probability forecasting as an expression ofthe range of outcomes that are possible in lightof the conditions that may arise before the forecastdate, whereas forecast error is a totallyundesirable feature of the shortcomings of thestate of forecasting science and of the availabledata.The primary mechanism used to incorporateuncertainty directly has been to perturb the initialconditions of the non-linear partial differentialequations describing the atmosphere using mesoscaleconvective system approaches. However,most methods currently in use are suboptimaland still rely on a judicious choice being exercisedby the forecaster. The ensemble Kalmanfilter is widely used to propagate and describeforecast uncertainty. The European Centre forMedium-Range Weather Forecasting (http://www.ecmwf.int) and other international agencies havebeen investigating the use of mesoscale convectivesystem-based ensembles in recent years and alarge-scale intercomparison hydrological ensembleprediction experiment, which was launchedin 2005. While this approach is indeed promising,it has yet to be proven, and a considerableamount of work will be required to develop proceduresfor the propagation of uncertainty throughcomplex model systems.7.1.5 Dissemination of forecasts andwarningsForecasts lose value with time. The faster data andforecasts can be sent to users, the more time can beapplied to response, thereby saving lives, reducingproperty damage and enhancing the operation ofwater resources structures. Dissemination of forecastsand warnings to communities and villages atrisk of flooding is frequently a weak link in theend-to-end chain. Significant progress in communicationstechnology allows for the rapidtransmission of data, forecasts and informationover large distances and to remote locations.The delivery of hydrological products from a forecastservice can be categorized as normal dailyforecasts and non-routine urgent forecasts. Manyusers require the routine transmission of data andforecasts on a daily basis, in the form of a hydrologicalbulletin. Information is generally given for keyrivers, reservoirs and other water bodies of interestto the region. Daily bulletins vary in their compositionand frequently include information aboutcurrent values and trends of stages, discharges,tendency of stages and discharges, water temperature,reservoir data such as pool and discharge,precipitation, hydrological forecasts and ice conditions,if prevalent. Figure II.7.2 provides an exampleof such a bulletin.Many routine hydrological products can beproduced based on user needs. Water supply forecastsand flow summaries can be issued on a weeklyand monthly basis. These summaries frequentlyprovide figures and data for key locations in riverbasins that may include medium- and long-rangeforecasts with lead times of weeks, months, orseasons. Distribution of routine hydrological productsshould be as widespread as possible, sincemany types of users can benefit from data and forecasts.Opening up data and forecasts to many usersenhances the value of forecasting services andbuilds a constituency for such services, which isnecessary if they are to sustain operations in thefuture.The Internet is the best means of disseminatinginformation. Although communication ishindered by bandwidth limitations in manydeveloping countries, its use and accessibility isimproving. Hydrological forecasting systems canmake use of this channel of communication intheir dissemination strategies for hydrologicalproducts. Other means of distributing productsare the public media, use of continuous radiobroadcasts and fax.

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-5Figure II.7.2. Example of a hydrological bulletin (Finnish Environment Institute)

II.7-6GUIDE TO HYDROLOGICAL PRACTICESThe composition and distribution of forecast andwarning products for medium-term extreme eventsrequire that input data be assembled rapidly andthat forecasts be made and reach the population atrisk in enough time to trigger response measuresthat will minimize the impacts. Hydrological forecastingcentres should explore all availablecommunication channels to reach the specificpopulation at risk. Communication mediacommonly used are direct transmission lines, satellite,radio and landline to emergency operationcentres and radio and television stations.Under emergency conditions such as flooding,warning products should clearly identify the typeof hydrological threat, the location of the predictedevent, namely the rivers and streams involved, themagnitude of the event expected, such as the peakflood level at critical locations, the forecast time ofoccurrence of the peak and, if possible, when theriver is expected to fall below warning or dangerlevel. Further details, such as what portion of theinfrastructure will be affected by the event, shouldbe provided, if possible. Such information providesemergency response units with locations whereaction needs to be taken to conduct evacuationsand road closures. As more data and informationbecome available, flood-warning products shouldbe updated and disseminated to the media andemergency-response officials.Advances in coupling hydrological models withexpanding geographical information datasets haveresulted in the development and implementationof high-visual hydrological forecasting products.This new class of hydrological products shows floodinundation produced by models linked to highresolutiondigital elevation model data. By linkingthat data with hydrological model forecast elevationscomputed for river channels, the area of floodinundation for the flood plain can be overlaid ontop of detailed digital maps of human infrastructureshowing how forecast flooding will impact agiven location. An illustration of a flood map productis provided in Figure II.7.3.7.1.6 Decision supportOrganizations responsible for water resourcesmanagement use decision-support tools to provideguidance for the operation of infrastructure.Forecasts for water management are needed to planeffective use of water, ranging from hydroelectricpower generation to water supply and irrigation.Measures taken by managers can have significantnegative consequences if future water availability isnot considered. If hydrological forecasts are available,water resources managers can operate watersupply systems to better meet water demand andminimize the potential for conflict.Flood losses can be reduced if communities andcountries invest in flood preparedness and responseplanning prior to the occurrence of the event.Emergency service organizations are responsible forestablishing flood-response plans that outline therole of various national, provincial and local organizationsin protecting life and property. This link inthe end-to-end chain includes setting up evacuationroutes, educating the population at risk of theFigure II.7.3. Flood map inundation forecast for Hurricane Mitch flood in Tegucigalpa, Honduras

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-7flood hazards and establishing procedures andtraining personnel to respond well in advance ofthe occurrence of a flood.The perfect flood forecast has no value unless stepsare taken to reduce losses. In the end-to-end process,data and forecasts must be produced as quicklyas possible to give users time to take action. In thecase of flooding, especially flash flooding, time iscritical. Understanding users’ needs and how forecastsare used is important. There are a wide varietyof users, ranging from federal response agencies tolocal governments, that have different roles andneeds in responding to and mitigating losses. Usersmust understand the forecast or warning messagefor appropriate response to occur.7.1.7 Cooperation with the NationalMeteorological ServiceAlthough a few countries have a combined meteorologicaland hydrological service, in most cases themeteorological and water management authoritiesare separate. Indeed, they are seldom within thesame government department or ministry. Theprovision of good weather forecast information,particularly in relation to severe precipitationevents, is a vital part of a flood forecasting andwarning operation. It is therefore important thatclose cooperation be developed between theNational Meteorological Service and the flood forecastingservice.Generalized meteorological forecasts are of littleuse to hydrologists; therefore, an initial step indeveloping cooperation should be taken to decidewhere value can be added to the meteorologicalinformation and how it can be structured to meethydrological requirements. This will be largely amatter of improving the information on rainfallforecasts in terms of quantity – quantitative precipitationforecasts – timing and geographicdistribution. Typical forecast products are asfollows:(a) Routine forecasts made on a daily basis, givinginformation on rainfall, temperature andweather for 24–48 hours, and an outlook periodof some 3–5 days;(b) Event forecasts, particularly forecasts andwarnings of severe events, such as heavy rainfall,snow and gales, which have hydrologicalimpacts. These should provide good quantitativeand areal information over a lead time of12–36 hours;(c) Outlook forecasts for periods of weeks ormonths, or particular seasons. These are usefulfor planning purposes, especially with regardto drought or cessation of drought conditions.Some national meteorological services, such asthose of South Africa, Australia and Papua NewGuinea, provide forecasts of El Niño activitywhere this is known to have direct impacts onweather patterns.The national meteorological service and the hydrologicalagency must agree on the structure andcontent of forecast and warning products. This isusually achieved by an evolutionary or iterativeprocess over time.It is also useful for other weather service productsto be provided to the hydrological agency. Themost common products are satellite imagery andrainfall radar information. The information istransferred by dedicated data feeds, which willhave a higher degree of reliability than informationtransferred through public service networksor available from Websites. A direct arrangementfor data transfer by the national meteorologicalservice also should ensure that updates are automaticallyprovided, for example, every 3–6 hoursfor satellite imagery and every 5–10 minutes forradar scans. By using satellite and radar information,hydrological agency staff can make their ownassessment and judgement of the current andimmediate future weather situation. It is importantthat staff be adequately trained to do this; thenational meteorological service has an importantrole in providing the necessary training throughintroductory and updating courses.The aforementioned arrangements and facilitiesshould be covered by a formal service agreementdefining levels of service to be achieved in timelinessof delivery and accuracy of forecasts. Theservice agreement should also include the costs ofservice provision. In this manner, both parties candefine the economic costs and benefits in providingthe service, and the value that the service may havein impact management.7.2 DATA REQUIREMENTS FORHYDROLOGICAL FORECASTS7.2.1 GeneralData requirements for hydrological forecastingdepend on many factors:(a) Purpose and type of forecast;(d) Basin characteristics;(b) Forecasting model;(c) Desired degree of accuracy of forecast;

II.7-8GUIDE TO HYDROLOGICAL PRACTICES(e) Economic constraints of the forecastingsystem.Data requirements vary considerably according tothe purpose of the forecast. Operating a reservoirrequires reservoir inflow forecasts relating to shortintervals of time and the volume of water likely toenter the reservoir flow as a result of a particularstorm flood. Water-level forecasts relating to large,slow-rising rivers can be estimated easily by measuringthe water level of upstream stations. Therefore,the data input in such cases will be the water levelat two or more stations on the main river or its tributaries.However, for small, flashy rivers, apart fromthe observations of water level and discharge atrelatively small intervals, the use of rainfall datapractically becomes unavoidable.Various considerations, discussed in subsequentsections, go into deciding what type of forecastingmodel should be used. Input data requirements forcalibration and operational forecasts vary significantlyfrom model to model. For example, in caseof a simple gauge-to-gauge co-relation model whichmay be suitable for large rivers, the only datarequirement may be water level at two or morestations. However, the use of a suitable comprehensivecatchment model requires a number of otherdata.Although the accuracy of the forecasts is of primaryconcern, the constraints in respect of economy andthe relative importance and purpose of forecastingmay permit a lesser degree of accuracy. In such situationsa model may be selected where datarequirement may be less rigorous. However, forforecasts at critical sites, such as those located neardensely populated areas or otherwise highly sensitiveareas, greater accuracy is essential.Apart from the type of data to be used for forecastingpurposes, the information regarding datafrequency, the length of record of data and dataquality are equally important, and should be dulyaccounted for in any flood forecasting system planning.Care must be taken to ensure that there is nobias between the data used to develop the forecastprocedure or to calibrate models and data used foroperational forecasting.On the whole, the availability and quality of dataneeded to produce a forecast is improving. Thenumber of automated gages and radars is increasing,while the quality of new satellites and rainfallestimation algorithms is producing enhancedinputs to hydrological forecasting procedures andforecast systems. A key issue in achieving datareliability in hydrological forecasts is themaintenance of the data platforms and thecommunications system.7.2.2 Data required to establish aforecasting systemRealistic hydrological forecasts cannot be producedwithout data. Data required for hydrological forecasting,as discussed in the previous sections, canbe broadly categorized as:(a) Physiographic;(b) Hydrological;(c) Hydrometeorological.Data relating to Geographical Information Systems(GISs) are required for both calibration and for visualizationof model states and outputs. The dataconsist of many types of land cover informationsuch as, soils, geology, vegetation and digital elevationmodel elevations. Hydrological forecastingsystem performance will depend on the quality andquantity of the historical data and GIS data used toestablish parameters.Hydrological data relating to river water levels, suchas discharges, ground water level, water quality andsediment load, and hydrometeorological data dealingwith evaporation, temperature, humidity,rainfall and other forms of precipitation, such assnow and hail, are key to hydrological forecasting.Some or all of the above data may be needed eitherfor model development or for operational use,depending on the model. Over the past ten years,databases and database-processing software havebeen coupled with hydrological models to producehydrological forecasting systems which utilizehydrological and or meteorological data and processthe data to be used by hydrological models. Thelatter then produce outputs used by the hydrologistto forecast river flow conditions, including floodsand droughts.An adequate hydrometeorological network is themain requirement for flood forecasting. In mostcases, the operational performance of the datanetwork is the weakest link within an integratedsystem. In particular, for the forecasting of floodsand droughts, there needs to be at least adequateprecipitation and streamflow/stream-gauge data. Ifsnowmelt is a factor, measurements of snow-waterequivalent, extent of snow cover and air temperatureare also important. Therefore, when establishinga hydrological forecasting system, it is important toask the following questions:(a) Are the rainfall and stream-gauge networkssatisfactory for sampling the intensity and

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-9spatial distribution of rainfall and the streamflowresponse for the river basin?(b) Are the stream gauges operating properly, andare they providing accurate data on the waterlevel and streamflow?(c) Are the data communicated reliably betweenthe gauge sites and the forecast centre?(d) How often are observations taken, and howlong does it take for them to be transmitted tothe forecast centre?(e) Are data available to users who need the informationfor decision-making?(f) Are the data archived for future use?(g) Are the data collected according to knownstandards, and is the equipment properlymaintained and calibrated and the data qualitycontrolled?Analysing the existing network is the first step. Aninventory of available monitoring locations, parameters,sensors, recorders, telemetry equipment andother related data should be made and presented ingraphical form. In low relief basins, monitoringsites from adjacent basins should also be listed, asdata from those sites can be very useful. An evaluationshould be performed to identify sub-basinsthat are hydrologically or meteorologically similar.The main objective is to take advantage of existinghydrometeorological networks operated by variousgovernment agencies and the private sector that arerelevant for the basin. In some respects, it is preferablethat the network serve many purposes, as thismay lead to broader financial support of thenetwork.The sufficiency of networks can be determinedaccording to forecasting needs, and required modificationsshould be noted. These could includenew stream gauges, raingauges and other sensorsin the headwaters, or additional telemetry equipment.In some cases, network sites may not be wellsuited for obtaining flow measurements or otherdata under extreme conditions. Structural alterationsmay be required. Interagency agreementsmay be needed for the maintenance and operationof the network.There are many sources of such data, ranging fromin situ manual observations to automated datacollection platforms and remote-sensing systems.Automated data systems consist of meteorologicaland hydrological sensors, a radio transmitter orcomputer – data logger – and a downlink or receiversite that receives and processes data for applications.There are many types of automatedhydrometeorological data systems that utilize lineof-sight,satellite or meteor-burst communicationstechnology. The rapid transmission of hydrometeorologicalinformation is extremely useful to waterstakeholders and users because it can be accessedinstantaneously by many users by downlinks and/or the Internet. Radar is a very popular and powerful,yet expensive, tool that can be used to estimateprecipitation over large areas. The use of geostationaryand polar orbiting satellites to derive largevolumes of meteorological and hydrological productsis advancing rapidly. Remotely sensed data cannow be used to provide estimates of precipitation,snow pack extent, vegetation type, land use,evapotranspiration and soil moisture, and to delineateinundated areas.For information on the instruments to be used incollecting, processing, storing and distributinghydrological and related data, see Volume I of thisGuide.7.2.3 Data required for operationalpurposesThe basic parameters that control hydrologicalprocesses and runoff are initial conditions andfuture factors. Initial conditions are conditionsexisting at the time the forecast is made andwhich can be computed or estimated on the basisof current and past hydrometeorological data.Future factors are those which influence thehydrological forecast after the current time. Itcan be claimed that the most severe resourcemanagement issues exist under extreme conditions:in time, during floods and droughts; and inspace, in arid, semi-arid and tropical areas and incoastal areas.A key variable to be established is the time stepneeded to adequately forecast a flood for a givenlocation. If the time step is six hours, for example,the data must be collected every three hours oreven more frequently. In many cases, supplementinga manual observer network with someautomated gauges may provide an adequate operationalnetwork. The use of more and more datamay become necessary to improve the model efficiency,which will most likely increase the costs.This is a major factor governing the choice forobservation, collection and analysis of data to beused for development of a suitable model and foroperational flood forecasting. More data entailmore expenditure and more time in collection andanalysis and man power, for example. Cost-effectivenessof the model vis-à-vis relative accuracyand consequences resulting therefrom should beduly considered when determining the datarequirements.

II.7-10GUIDE TO HYDROLOGICAL PRACTICESRemote-sensing plays an important role in collectingup-to-date information and data in both thespatial and temporal domains. The use of remotesensingtechniques is vital in areal estimation, inparticular of precipitation and soil moisture. Suchtechniques enhance seasonal forecasting capabilities;contribute to the development of storm-surgeforecasting, drought and low-flow forecasting; andhelp improve risk management.The rapid spread of the Internet throughout theworld has not only produced an excellent mechanismto distribute hydrological data and forecaststo a diverse user community, but has also produceda rich source of data, forecasts and information ofuse to National Meteorological and HydrologicalServices. The Internet provides a source of valuableinformation for hydrological forecast services. Thismay include meteorological and hydrologicalmodels, hydrological forecasting documentation,geographical information system data, real-timeglobal meteorological forecasting products, hydrometeorologicaldata and hydrological forecastinginformation. A vast amount of data, software anddocuments are available for use, and these sourcesare growing daily. Some sample URLs, or uniformresource locators, are provided at the end of thechapter for reference.7.3 FORECASTING TECHNIQUES[HOMS J04, J10, J15, J80]7.3.1 Requirements for flood forecastingmodelsGiven the recognized variability of climate and itsexpected influence on the severity, frequency andimpact of floods and droughts, the importance offorecasting has increased in recent years. Thissection descibes the basic mathematical and hydrologicaltechniques forming the component parts ofany forecasting system. A brief discussion of thecriteria for selecting the methods and determiningthe parameters is also provided. Examples of theuse of these components for particular applicationsare given in 7.4 to 7.6.Flood forecasting operations are centred aroundtime and the degree of accuracy of the forecast. Infact, a professional assigned with formulating aforecast has to race against time. Clearly, themodels to be used by forecasting organizationsmust be reliable, simple and capable of providingsufficient warning time and a desired degree ofaccuracy. Model selection depends on thefollowing factors: amount of data available;complexity of the hydrological processes to bemodelled; reliability, accuracy and lead timerequired; type and frequency of floods that occur;and user requirements.A comprehensive model involving very detailedfunctions which may provide increased warningtime and greater degree of accuracy may have veryelaborate input data requirements. All input datafor a specific model may not be available on a realtimebasis. Therefore, from a practical point of view,a flood forecasting model should satisfy the followingcriteria:(a) Provide reliable forecasts with sufficient warningtime;(b) Have a reasonable degree of accuracy;(c) Meet data requirements within available dataand financial means, both for calibration andfor operational use;(d) Feature easy-to-understand functions;(e) Be simple enough to be operated by operationalstaff with moderate training.Indeed, the choice should never be restricted to aspecific model. It is always desirable to select andcalibrate as many models as possible with a detailednote on suitability of each of the models underdifferent conditions. These models should beapplied according to the conditions under whichthey are to be operated.Comprehensive models, which are rather complicated,generally require computational facilitiessuch as computers of a suitable size. At many places,however, such facilities are not available. Sometimessuitably trained staff are not available; what is more,these machines cannot be operated because ofrecurring problems such as electricity failures.Therefore, both computer-based comprehensivemodels and simple types of model can be developed.A computer-based technique can be used ingeneral, and in case of emergency, conventionaltechniques, which are generally of a simple type,may be adopted.Apart from the selection of different models, it isdesirable to have a calibration of the models underdifferent conditions. For example, a model may becalibrated with a suitably large data network;however, at the same time, a model must be calibratedfor a smaller network and give dueconsideration to the possible failures in observationand real-time transmission of some of the data. Thiswill be helpful in utilizing the model even in emergencysituations in which the data are not availablefrom all the stations. This will require different sets

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-11of parameters to be adopted under differentconditions.7.3.2 Flood forecasting methodsOn the basis of the analytical approach used todevelop a forecasting model, flood forecastingmethods can be classified as follows:(a) Methods based on a statistical approach;(b) Methods based on a mechanism of flood formationand propagation.Forecasting methods in the form of mathematicalrelationships produced with the help of historicaldata and statistical analysis have been widely usedin the past. These include simple gauge-to-gaugerelationships, gauge-to-gauge relationships withsome additional parameters and rainfall–peak stagerelationships. These relationships can be easilydeveloped and are most commonly used as a startingpoint while establishing a flood forecastingsystem. The use of artificial neural networks to forecastflood flows is another modelling approach thathas recently gained popularity.Increasingly, forecast procedures are based on morecomplete physical descriptions of fundamentalhydrological and hydraulic processes. In manyinstances when forecast flows and stages are neededalong rivers, hydrologists use rainfall-runoff modelscoupled with river-routing models. If precipitationis in the form of snow, snowmelt models are applied.These models vary in accuracy and complexity,ranging from simple antecedent index models tomulti-parameter conceptual or process models.With advances in computing and telemetry developments,forecasting models are now more flexiblein providing information and allowing new dataand experience to be incorporated in real time.There are many varieties of these basic categories ofmodels, and most differ according to how hydrologicalprocesses are parameterized. Modelscan range from simple ones featuring a statisticalrainfall-runoff relationship combined with a routingequation to others characterized by a muchhigher degree of complexity.Hydrological models can be classified as lumped,semi-distributed or distributed. Models are eitherevent driven or continuous. If a model is capable ofestimating only a particular event, for example thepeak flood resulting from a storm, it is known as asevent driven. A continuous model is capable ofpredicting the complete flood hydrograph at a specifiedtime interval. Model selection requirementsinclude the following factors:(a) Forecast objectives and requirements;(b) Degree of accuracy required;(c) Data availability;(d) Availability of operational facilities;(e) Availability of trained personnel for developmentof the model and its operational use;(f) Upgradeability of the model.Significant progress over the past two decades hasbeen made in improving the science and performanceof such models. However, performance usuallyvaries according to the type of river basin characteristicsbeing modelled, the availability of data tocalibrate models and the experience and understandingof the model by the operationalhydrologist. There are a large number of publicdomain and proprietary models available for use inflood forecasting. Chapter 6 of this Guide reviews awide range of currently available hydrologicalmodels.7.3.2.1 Statistical methodThe correlation coefficient measures the linear associationbetween two variables and is a widely usedmathematical tool at the root of many hydrologicalanalyses. Regression is an extension of the correlationconcept that provides formulae for deriving avariable of interest, for example, seasonal low flow,from one or more currently available observations,such as maximum winter groundwater level (seeDraper and Smith, 1966).The formula for calculating the correlation coefficientr between n pairs of values of x and y is asfollows:n∑ ( x i − x )(y i − y )i=1r =(7.1)n∑ ( x i − x ) 2 n( yi=1 ∑ i − y ) 2i=1wherex = 1 n∑nx i and y = 1 i=1n∑ni=1 y iA lack of correlation does not imply lack of association,because r measures only linear associationand, for example, a strict curvilinear relationshipwould not necessarily be reflected in a high value ofr. Conversely, correlation between two variablesdoes not mean that they are causatively connected.A simple scatter diagram between two variables ofinterest amounts to a graphical correlation and isthe basis of the crest-stage forecast technique (see7.3.4 for verification of forecasts).If either x or y has a time-series structure, especiallya trend, steps should be taken to remove this

Discharge (m 3 s –1 )II.7-12GUIDE TO HYDROLOGICAL PRACTICESstructure before correlating, and caution should beexercised in interpreting its significance. Time-seriestechniques may be applied (see 7.5.3) when previousvalues of a variable such as river discharge areused to forecast the value of the same variable atsome future time.Likewise, regression equations have many applicationsin hydrology. Their general form is asfollows:Y = b o+ b 1X 1+ b 2X 2+ b 3X 3+ ... (7.2)where X refers to currently observed variables and Yis a future value of the variable to be forecast.Regression coefficients estimated from observed Yand X values are indicated by b. The X variablesmay include upstream stage or discharge, rainfall,catchment conditions, temperature or seasonalrainfall. The Y variable may refer to maximum orminimum stage. The multiple-correlation coefficientmeasures the degree of explanation in therelationship. Another measure of fit, the standarderror of estimate, measures the standard deviationof departures from the regression line in the calibrationset. The theory is explained in all generalstatistical texts.Linear combinations of the variables are sometimesunsatisfactory, and it is necessary to normalizeeither the X or Y. A powerful transformation methodcan be used to transform Y to Y Tby the followingequations:Y T= (Y T – 1)/T T ≠ 0ifY T= ln(Y)T = 0(7.3)which encompasses power, logarithmic andharmonic transformations on a continuous T scale.A suitable T value can be found by trial and error asthat which reduces skewness or graphically by usingdiagrams such as Figure II.7.4.Non-linearity can also be accommodated in aregression by using polynomials, for example, byusing X i, X i2or X i 3 . Alternatively, non-linear regressionusing function-minimization routines offers asimply applied route to fitting parameters ofstrongly non-linear equations. The selection of auseful subset from a large potential set of explanatoryvariables calls for considerable judgement and,in particular, careful scrutiny of the residuals, thedifferences between observed and estimated valuesin the calibration dataset. The circumstances givingrise to large residuals are often indicative of adjustmentsthat need to be made. Advantage should betaken of computer facilities and graphical displaysof residuals to explore a number of alternativecombinations. The exclusive use of wholly automaticsearch and selection procedures, such asstepwise, stagewise, backward and forward selectionand optimal subsets should be avoided.100806050403020108643210.5 1 2 5 10 20 50 80 90 95 98 99 99.5 99.8 99.9Per cent of time daily discharge exceededFigure II.7.4. Flow-duration curve of daily discharge

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-13Examples of the application of regression to forecastingproblems are given in 7.3.2.3 and 7.4.7.7.3.2.2 Soil moisture index modelsThe antecedent precipitation index is described in6.3.2.2. This method has been a primary tool foroperational forecasting in many countries. As ameasure of the effect of precipitation occurringprior to the time of the forecast, it provides an indexto the moisture in the upper level of the soil. Themost frequently encountered indices are the antecedentprecipitation index and the antecedentmoisture condition. The moisture index methodshave two main features with respect to their applicationto hydrological forecasting. First, becausethe index is updated daily, it is suited to an eventtype of analysis rather than continuous modelling.Thus, to apply this method to most forecasting, it isnecessary to divide a precipitation period intoevents or to divide an event into separate precipitationperiods. For example, during extended periodsof precipitation interrupted by brief periods of littleor no rainfall, the decision as to whether one orseveral storms are involved may be difficult.The second feature is that the computed surfacerunoffvolume, when applied to a unit hydrograph,produces a hydrograph of surface runoff only. Inorder to synthesize the total runoff hydrograph, thebase flow must be determined by some othermethod. The technique is of operational use only ifevent runoff is of importance and a simple approachis all that can be justified.7.3.2.3 Simplified stage-forecasting methodsA very common requirement in an event is to forecastthe maximum stage or crest. A proven practicaltechnique used in relation to moderate-sized riversis to construct a simple graphical correlation withan upstream stage hydrograph, thus providing aforecast with a lead time equal to the travel time ofthe flood wave. Figure II.7.5 illustrates thisprocedure.It is common to chain such crest-to-crest forecastsso that the output from an upstream forecastprovides the input to a downstream one. Suchgraphs can often be used to forecast the hydrographsif account is taken of the difference in lag timeduring the periods of rise and fall. The followingcorrelation relationship is useful when simplestation-to-station relationships (Figure II.7.5) arenot successful:(h 2) t+Δt= f((h 1) t,I loc) (7.4)where h 1and h 2denote maximum stages at anupstream and downstream station, respectively, I locis the local inflow between the stations, and Δt islag time. Figure II.7.6 gives an example of the relationshipof this type. Sums of discharges at two ormore upstream stations at appropriate times, as acombined variable instead of individual tributarystage heights, may reduce the number of variablesin the correlation. Variations on these basicapproaches can be devised to suit differing circumstancesof travel time and tributary inflow. Thegraphical approach can be replaced by an entirelynumerical one by making use of multiple regression(see 7.3.2.1). The regression equation may take thefollowing form:h max= b 0 + b 1 Q 1 + b 2 Q 2+ ... (7.5)where Q 1Q 2... are discharges at upstream stations ata given time. Other explanatory variables, such asrainfall and antecedent catchment conditions(7.3.2.2), may supplement or be substituted fordischarge.7.3.2.4 Conceptual streamflow modelsThere are many basic categories of models, andmost vary according to how hydrological processesare conceptualized. Hydrological models and/orforecast procedures use real-time precipitation andCrest stage at downstream station (cm)Time (days)1 6001 4001 2001 000800600510800 1 000 1 200 1 400 1 600Crest stage at upstream station (cm)600 800 1 000 1 200 1 400 1 600Figure II.7.5. Crest stage and travel time for theVolga river

II.7-14GUIDE TO HYDROLOGICAL PRACTICESCrest stage at downstream station (cm)700600500400300200100200 300 400 500 600Crest stage at upstream station (cm)Note: Average travel time is 24 hours.25Figure II.7.6. Typical gauge relationship withvariables for local inflowstreamflow data and translate observed conditionsinto future stream conditions. Hydrological modelsor procedures vary in complexity, accuracy and easeof use. Simple hydrological models consist of tables,graphs or empirically derived relationships. Moresophisticated hydrological modelling systems usein situ and remotely sensed data, and multiplehydrological models integrated to produce veryaccurate hydrological forecasts. New developmentsin personal computer technology have made itpossible for complex modelling systems to be runon such computers. These systems are easier to useand sustain than their predecessors.Large strides have been made over the past twodecades in improving the science and performanceof models. Model performance varies according tothe type of river basin characteristics beingmodelled, the availability of data to calibrate modelsand the experience and understanding of the modelmechanics of the hydrologist applying the model.Data are usually the limiting factor in attainingacceptable accuracy in operational application.However, with the advances in GIS data availability,hydrological model parameters can be estimatedwithout relying exclusively on historical hydrologicaldata to calibrate the models.The availability of operational precipitation estimateswith high spatial and temporal resolution2015105Runoff from local area (mm)from weather radars and the substantial increasesin computer power have made possible the use ofdistributed hydrological models. There is a wealthof distributed models, owing to the advent ofdistributed databases of land-surface and soil characteristics.Carpenter and others (2001), Ogden andothers (2001), Beven (2002), and Smith and others(2004a) provide recent overviews of distributedhydrological modelling and the issues surroundingits possible use for operational forecasting.The significant influence of rainfall input uncertaintiesand model structure and parameter errors onthe small scales of flash flood occurrence havehindered the early utilization of distributed modelsfor operational forecasting. Nevertheless, distributedmodels promise to provide additional informationand insight regarding hydrological conditions atlocations without existing streamflow observations.In the United States of America, the NOAA-sponsoredDistributed Model Intercom parison Projectprovided a forum to explore the applicability ofdistributed models using operational quality dataand to highlight issues surrounding their use (Smithand others, 2004b). To account for uncertainty inrainfall estimates on small scales (see Collier andKrzysztofowicz, 2000) and for hydrological modelerrors, it is advisable to produce probabilistic, ratherthan deterministic forecasts in flash-flood-proneareas when using distributed hydrological models.This area of probabilistic flow prediction remains anactive research area in hydrology (see Carpenter andGeorgakakos, 2004).7.3.3 Model updating techniquesForecast adjustments are usually based on modeloutput and direct measurements of the statevariables. There are many techniques for updatingforecasts. If an observation is made of the forecastoutput Y i, there is an opportunity to adjustsubsequent forecasts with the benefit of the knownforecast error e i= Y i– Y^i, where Y^iis the forecastestimate. Most adjustments are the result of thesubjective judgement of the forecaster, but variousmathematical techniques allow this process to beformalized. The underlying principles of the formalapproach are described below.At their simplest, adjustments to forecasts may bemade by subtracting the current error from the newforecast. In order to avoid discontinuities, theadjustment is generally blended into the computedhydrograph over several time periods. A morecomplicated procedure is to subject the error seriese i, e 2,..., e ito a time-series analysis to extract possibletrends or periodicities that can be extrapolated

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-15to estimate the potential new error e^i+1, which canbe used to modify the new forecast Y^i+1 .There are two major types of real-time modelupdating:(a) Parameter updating, where the estimates ofsome, and possibly all, of the model parametersare updated regularly on the basis of incomingdata such as rainfall and flow. These data areobtained from conventional telemetry or themore modern supervisory control and dataacquisition systems, also known by their acronym,SCADA;(b) State updating, where estimates of the state variablesin the model, such as flow or water level,are updated regularly on the basis of incomingdata.Sometimes these updating operations are carriedout in a fully integrated manner by using some formof parameter-state estimation algorithm such as theextended Kalman filter. Alternatively, they arecarried out concurrently but in separate algorithms.These algorithms are normally known as recursiveestimation algorithms because they process data ina recursive manner whereby new estimates are functionsof previous estimates, plus a function of theestimated error. Examples of these algorithms arethe recursive least squares algorithms, widely usedin operational hydrology (see Cluckie and Han,2000) and the recursive instrumental variables algorithm,as described in Young (1993).The Kalman filter and the extended Kalman filterare recursive estimation techniques that have beenapplied to hydrological forecasting, but they requireconsiderable mathematical and hydrological skillsto ensure that the forecast model is in a suitableform for analysis.The generic form of the recursive parameter estimationalgorithm is as follows:Innovations process (one-step-ahead prediction)a^t = a^t–1 + G t {y t– y^t|t–1}; y^t|t–1= f {a^t–1, y^t–1} (7.6)While the generic form of the state estimation algorithmis:Model equationPrediction: x^t|t–1= f {x^t–1, a^t–1}Innovations processCorrection: x^t = x^t|t–1 + G t {y t– y^t|t–1} (7.7)where y = g{x t} is the observed data that is related tothe state variables of the model in some definedmanner and G tis a time variable matrix, oftencalled the system gain, that is also computed recursivelyand is a function of the uncertainty in theparameter or state estimates. An algorithm thatcombines these two recursive estimation operationsis often called a data assimilation algorithm (seeYoung, 1993).However, a more conceptual technique for adjustingthe output of a hydrological model may also beused. The method does not require any changes inthe model structure or in the algorithms used in themodel. Rather, this approach adjusts the input dataand, consequently, the state variables in such a wayas to reproduce more closely the current and previousflows. These adjusted values are then used toforecast the hydrograph.Forecast adjustments need not be based solely onthe output of the model. It may also be accomplishedby using measurements of state variablesfor comparison with the values generated by themodel. For example, one such technique usesobserved measurements of the water equivalent ofthe snow cover as a means of improving the seasonalwater supply forecasts derived from a conceptualmodel. Direct substitution of field measurementsfor numerically generated values of the state variablesof the model would be incorrect because, inpractice, model simplifications can result in suchstate variables loosing their direct physicalidentity.7.3.4 Forecast verificationForecast verification characterizes the correspondencebetween a set of forecasts and a correspondingset of observations. No forecast system is completewithout verification procedures in place to conductadministrative, scientific and user oriented verificationof the forecasts.A variety of statistics can be computed to evaluateforecast skill. The statistics to be used will dependon the type of forecast and the purpose of the forecastand of the verification. A study of the utility ofproposed metrics to effectively characterize theforecast skill should be conducted prior to implementinga verification programme.To be effective, a verification system must include aforecast archive and the observations against whichthe forecasts are to be measured. In addition, abaseline forecast must be included to assist with theinterpretation of the computed verification

II.7-16GUIDE TO HYDROLOGICAL PRACTICESmeasures. Selection of the baseline forecast willdepend on the forecast type to be verified and theforecast process used to develop the forecasts. Forshort-term deterministic forecasts of less than twodays, persistence is a useful baseline.For longer-term forecasts and for probabilisticforecasts, climatological distributions or laggedclimatology are more appropriate baselines. If theforecast process consists of several steps, additionalintermediate data must also be archived to enablevalidation of each step in the forecast process. Ifpossible, the input data used to compute the forecastsshould be archived to enable hindcast studiesof possible forecast process updates. The data to bearchived should include the observations, theinput forecasts, such as precipitation and temperature,and the model parameters, including ratingcurves. Joliffe and Stephenson (2003) are an excellentreference, providing more detailedinformation. In 1995, WMO developed MOFFS,the management overview of flood forecastingsystems, to seek an internationally applicable basisfor providing fast, focused information on theperformance of flood forecasting systems based onexceedence of specified trigger levels on rivers. Theobjective of MOFFS is to swiftly identify and highlightdeficiencies in the facilities and performanceof individual flood forecasting systems in orderthat appropriate management action may be takento remedy the defects before the next flood eventoccurs.7.4 FORECASTING FLASH FLOODS[HOMS J04, J10, J15]Flash floods are rapidly rising flood waters that arethe result of excessive rainfall or dam break events.Rain-induced flash floods are excessive water flowevents that develop within a few hours – typicallyless than six hours – of the causative rainfall event,usually in mountainous areas or in areas with extensiveimpervious surfaces such as urban areas.Although most of the flash floods observed are raininduced, breaks of natural or human-made damscan also cause the release of excessive volumes ofstored water in a short period of time with catastrophicconsequences downstream. Examples arethe break of ice jams or temporary debris dams.7.4.1 National flash flood programmesPrior to the advent and availability of highresolutionspatially extensive digital data fromweather radars and from satellite platforms, and ofhigh-resolution digital terrain elevation data,forecasting of flash floods, as well as with therequired spatio-temporal resolution, was notpossible on a national scale. In recent years,however, high-resolution data have becomeavailable in most countries, and expanded computercapabilities have made it possible to developnational flash flood forecasting programmes.7.4.1.1 Cooperation between hydrologistsand meteorologistsOwing to the short concentration times of flashfloods, the timely and accurate detection and shorttermprediction of rainfall and streamflow and/orwater levels are important ingredients of a successfulflash flood forecast and warning system. Thisrenders flash flood forecasting a truly hydrometeorologicalendeavour, which benefits much from closecollaboration between meteorologists and hydrologistsin national and regional forecasting centres. Inaddition, the local nature of rain-induced flashfloods requires detailed regional and local observations,understanding and modelling of the heavyrainfall and the runoff-production/channel-routingprocesses in the flash-flood-prone areas, supportedby databases of high resolution in both space andtime.7.4.1.2 Cooperation between national andregional or local agenciesEven when national flash flood forecastingprogrammes are in place, regional and local involvementis necessary for the operation of the systemsto succeed. Individual regional and local physicalsettings significantly affect flash flood genesis anddevelopment. The meteorological and hydrologicalsituation may change from the time of data inputat the national level to the time when regional andlocal response to forecasts is required. The errorlevels in the measurements by weather radar andsatellite data vary considerably from place to place.Finally, individual end-users at the local level – thepublic at large, individual industries, water resourcesmanagement agencies and so forth – are likely tohave different requirements for flash flood warningsthat may not all be addressed by the nationalflash flood forecast programme. This national andregional or local collaboration ideally involvesregional forecast offices, local response agenciesand end-users.It may be necessary for end-users to develop additionalproducts that utilize the national flash floodforecasts and other ancillary information producedby the national forecast centres to address their

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-17individual needs at the local level. For instance,this may include procedures for further refinementof the forecasts for certain flood-stage levels notaddressed by the national products, or installationand operation of local automated networks ofraingauges and special-purpose radars in areaswhere national weather radars and satellites donot provide reliable data. In such cases, thenational flash flood programme provides flashflood guidance.7.4.1.3 Cooperation with end-usersFor flash flood forecasts that are highly resolvedin space and time, it is desirable to establish asignificant forecaster-user collaborationprogramme that will serve several purposes:inform the users – the regional weather serviceoffices, the local response agencies, the public atlarge or other end-users – as to what the nationalflash flood forecasts mean; provide informationabout forecast validation and the limitation ofthe national systems implemented; supportdecision-making at the local level; develop guidelinesfor appropriate user action when warningsare issued; identify ways to receive feedback fromthe end-users as to the performance of the operationalsystem; and other purposes. Thiscollaborative programme will in the long termhelp improve the local effectiveness of nationalflash flood forecast products.In several countries, flash flood forecasts are disseminatedby means of watches and warnings. Ifmeteorological conditions conducive to heavy rainfallare observed or foreseen for an area, a watch isissued on radio and/or television. This alerts residentsin the area to the potential occurrence ofrainfall that could produce flooding. When floodproducingrainfall is reported, the watch is followedby a warning advising the residents in the area totake necessary precautions against flooding.7.4.2.1 Self-help forecast programmesSelf-help flash flood warning systems are operatedby the local community to minimize delays in thecollection of data and dissemination of forecasts. Alocal flood warning coordinator is trained to prepareflash flood warnings based on pre-planned proceduresor models prepared by qualified forecastauthorities. The procedures are employed whenreal-time data and/or forecast rainfall indicate apotential for flooding. Multiple regression equationsprovide an operationally simple flash floodforecasting technique that is summarized in asimple flood advisory table. The procedure is suitablefor a range of different flood-producingconditions of rainfall, soil moisture andtemperature.The growing availability of microprocessors has ledto an increased tendency to automate much of thedata collection and processing that are required toproduce flash flood warnings. Automatic rainfalland stage sensors can be telemetered directly to thecomputer that will monitor the data-collectionsystem, compute flood potential or a flood forecast,and even raise an alarm. The most critical componentin the self-help system is maintenance ofactive community participation in the planningand operation of the system.7.4.2.2 Alarm systemsA flash-flood alarm system is an automated versionof the self-help type of warning programme. A stagesensor is installed upstream of a forecast area and islinked by land or radio telemetry to a receptionpoint in the community such as a fire or policestation that is staffed around the clock. This receptionpoint contains an audible and visual internalalarm and relay contacts for operating an externalalarm. The alarm is activated when the water levelat the sensor reaches a pre-set critical height.7.4.2 Local flash flood systemsThere is a wide variety of flash flood forecasting andwarning approaches implemented for specificgauged sites. They range from self-help proceduresbased on local networks of automated stream gaugesto more sophisticated procedures that include localshort-term rainfall and flow forecasting. Theseprocedures are designed to provide early warningfor local communities, utility companies and otherregional or local organizations so that they can actimmediately on receiving the warning. A few representativesite-specific approaches are discussedbelow.7.4.2.3 Integrated hydrometeorologicalsystemsThese systems are more sophisticated state-of-theartsystems and are generally implemented byutilities and other regional or local organizationsthat maintain in-house hydrometeorological expertise.In most cases these systems provide the mostreliable flash flood forecasts for specific locations.Typical implementations involve integratedhydrometeorological models, either conceptual orprocess based (see Georgakakos, 2002). The componentsof these models consist of a regionalinterpolator of operational numerical weather

II.7-18GUIDE TO HYDROLOGICAL PRACTICESprediction information to the scale of analysis,100 km 2 or less, a soil-water accounting model anda channel-routing model. To account for uncertaintiesin real-time numerical weather predictions andsensor-data configurations, state estimators orassimilators provide feedback to model states fromavailable real-time observations. Various forms ofthe extended Kalman filter and non-linear filtershave been used in these systems.An example of the implementation and use of integratedhydrometeorological systems is the PanamaCanal watershed flash flood forecast system: moreinformation may be found in Georgakakos andSperfslage (2004). The 3 300-km 2 Canal watershedhas been subdivided into 11 sub-catchments basedon topography, stream gauge availability, reservoirlocation and local hydrometeorology (seeFigure II.7.7). Short-term forecasts covering one- tosix-hour periods are necessary to mitigate damageto Canal equipment and operations. A meteorologistand a hydrologist operate the system andinterpret the rainfall and flow forecasts.There is a 10-cm weather radar and more than35 automated ALERT-type raingauges in the region.The computational grid of the US National WeatherService operational numerical weather predictionmodel ETA covers the region with an 80-km resolutionand provides large-area forecasts of atmosphericstate twice daily with six-hourly resolution and amaximum lead time of several days.The rainfall forecast component uses informationfrom the 80-km ETA model and upper-air radiosondeand surface meteorological data. Theprecipitation model produces sub-catchment rainfallforecasts that are compared to the merged radargauge estimates to produce a forecast error. Theserainfall forecasts are fed into the soil water accountingmodel of each sub-catchment that generatesrunoff and feeds the channel-routing model. Aseparate state estimator is used to update the soilwater model states from real-time dischargeobservations.An important aspect of local hydrometeorologicalsystems is forecast validation for significant flashflood events. This activity provides useful informationto forecasters to assist them with theinterpretation of the system forecasts and the translationof these forecasts into warnings and watches.Typical least squares performance measures may beused, such as residual mean, residual variance,mean square error and coefficient of efficiency,together with other measures of performance9.59009.48009.37009.2600Latitude °N9.195004008.98.88.730020010080.1 80 79.9 79.8 79.7 79.6 79.5 79.4 79.3Longitude °W0Figure II.7.7. The Panama Canal watershed showing terrain elevation (1 km digital terrain model) andsub-catchments (Georgakakos and Sperfslage (2004))

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-19produced in collaboration with the forecast users,including errors in forecast water volume over agiven duration, peak hourly flow timing and magnitude.Figure II.7.8 is an example of a flash floodwarning.7.4.3 Wide-area flash flood forecastsThe ability to measure precipitation routinely withhigh spatial and temporal resolution and the availabilityof high-resolution spatial databases for theland surface and subsurface have led to the emergenceof flash-flood-scale, operational, wide-areaforecasts produced by national agencies. Two mainapproaches may be identified for the production ofwide-area flash flood forecasts with high resolution:(a) those that use the concept of flash flood guidanceand (b) those that are based on spatiallydistributed hydrological models. In either case,rainfall observations and forecasts highly resolvedin space and time are necessary.To obtain rainfall estimates on the scales requiredfor flash flood forecasting, dense raingauge networksare needed. For national flash flood forecasting overlarge areas with high resolution, rainfall estimationon such small scales includes data from automatedraingauges complemented by data from regionalweather radars and/or satellite sensors. Differentsensors measure different attributes of rainfall and amerged product is often computed as a best estimatethat combines all available data. It is oftenuseful to produce measures of uncertainty in therainfall estimates because measurement errors varyfrom sensor to sensor and region to region.Many studies discuss operational quantitative rainfallestimation achieved by merging raingauge andweather radar data, ranging from the early results ofCollinge and Kirby (1987) in the United Kingdomof Great Britain and Northern Ireland to morerecent results reported in the United States byFulton and others (1998) and Seo and Breidenbach(2002). In such cases, the spatial variability of therainfall field on flash flood occurrence scales isobtained mainly from weather radar data, while useof the automated raingauges is made to correctfield-mean or range-dependent bias of the weatherradar estimates using a variety of procedures, asdescribed, for example, by Cluckie and Collier(1991), Braga and Massambani (1997) and Tachikawaand others (2003).Satellite rainfall data are often calibrated withweather radar data existing in similar hydroclimaticregions and/or any sparse local or regional automatedraingauge networks. Combinations ofpolar-orbiting and geostationary satellite productsare also under development (see Bellerby andothers, 2001).7.4.4 Flash flood guidanceThe concept of flash flood guidance has been usedfor wide area forecasts of flash flood occurrencesince the mid 1970s in the United States (Mogil andothers, 1978). Flash flood guidance is the volume ofrainfall of a given duration, for example, one to sixhours, over a given small catchment that suffices tocause minor flooding at the outlet of the drainingstream. The volume estimate is updated frequentlyand is used to assess the potential for flooding whencompared with volumes of observed or forecastrainfall of the same duration and over the samesmall catchment.Determination of flash flood guidance in an operationalenvironment requires the development ofthe following tools:(a) Estimates of threshold runoff volume of variousdurations, done offline;(b) A soil moisture accounting model to establishthe curves that relate threshold runoff to flashflood guidance for various estimated soil moisturedeficits (Sweeney, 1992).Early flash flood guidance operations used statisticalrelationships to develop the required thresholdrunoff estimates from a variety of regional and localdata, such as topographic and climate data. Usingexisting digital spatial databases of land-surfaceproperties such as terrain, streams and land use orland cover, together with GIS, Carpenter and others(1999) set the threshold runoff estimation problemon a physical basis and provide methods for developingobjective threshold runoff estimates on anational scale with high resolution. For a givensmall catchment, the basic threshold runoff relationshipis as follows:Q flood= Q p(R,t r) (7.8)where Q floodis the flow that is considered likely tocause minor flooding at the catchment outlet, andQ pis the peak of the surface runoff over the catchmentcaused by the effective rainfall volume R, thethreshold runoff, of the given duration, t r. The Q floodmay be estimated by the flow of a given returnperiod, for example, two or four years, or by hydraulicformulae for uniform steady flow at streambankfull and at the catchment outlet. Synthetic orgeomorphologic unit hydrograph formulationsmay be used to estimate Q pfrom R and t r. The use ofbankfull flow and geomorphologic unit hydrograph

II.7-20GUIDE TO HYDROLOGICAL PRACTICESFigure II.7.8. Flash flood warning sign

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-21formulations requires no calibration and producesthreshold runoff estimates that are conservative interms of flood damage.Channel cross-sectional properties at the catchmentoutlet are required in order to estimatebankfull flow and the geomorphologic unithydrograph runoff peak. Such estimates are typicallyobtained from regional regressions of channelcross-sectional properties, that is, bankfull width,or hydraulic depth that use GIS-estimated catchmentproperties – area, stream length and averagestream slope – as predictors. These regional regressionsare based on data obtained from surveys ofnatural streams in the region of interest and areused to provide estimates of channel cross-sectionalparameters in all of the small catchments of theregion.The resolution of the digital terrain elevation datalimits the size of the smallest catchments for whichthreshold runoff analysis may be made. For instance,a 90-m resolution terrain database may be used toproduce catchment properties, such as area andstream location, length and slope, with relativeerrors between +/–10 per cent and +/–25 per centfor catchments greater than 5 km 2 . For such catchments,up to a maximum size of 50 km 2 , typicalerrors of threshold runoff estimates based on GISanalysis of digital terrain elevation data can reach+/–30 per cent of the value estimated at sites with afull complement of hydrometeorological data.Threshold runoff is the volume of effective rainfallof a given duration generated over a small catchmentthat is sufficient to cause minor flooding atthe outlet of the draining stream. Once the thresholdrunoff estimates have been obtained for theregions of interest or for the entire country, they areused in conjunction with real-time estimates of soilwater deficit to produce threshold runoff. Theprocedure is outlined below (see Georgakakos,2004).Typically, the national forecast services run a hydrologicalmodel operationally over basins of area inthe order of 1 000 km 2 to produce streamflow estimatesand forecasts at each of several forecastpreparation times. Upon completion of these operationalforecast runs, the current estimates of thesoil water indices, valid at the forecast preparationtime, are stored. To support flash flood computationunder these initial conditions, the hydrologicalmodel is run offline in “what if” scenario runs withincreasing amounts of rainfall volumes of the samegiven duration. These “what if” runs use the sameinitial soil water estimates produced by the modelduring a normal operational run. The surface runoffvolume produced by these runs is plotted againstthe volume of required rainfall of a given duration.This plot may then be interpreted as the relationshipof threshold runoff (effective rainfall or surfacerunoff volume) to flash flood guidance (actual rainfallvolume). It is used with the estimated value ofthreshold runoff for the catchment to obtain therequired flash flood guidance volume, both of thesame given duration.Estimates of most recent catchment rainfall volumeof duration equal to the flash flood guidance durationmay then be used to determine whether a flashflood is imminent in a certain catchment. Likelyflash flooding occurrence is indicated when theobserved rainfall volume is greater than the flashflood guidance estimate. Following this procedure,maps of entire regions highlighting catchmentswith a high potential for flash flooding may beproduced on regional and national scales. Similarmaps may be produced showing the future potentialof catchments for flash flooding using forecast,rather than observed, catchment rainfall volumesof a given duration. The US National WeatherService uses a national operational implementationof flash flood guidance for forecasts of wide areaflash floods. A regional implementation of flashflood guidance is operated for the countries ofCentral America. National programmes for collectingthe necessary flash flood occurrence data tovalidate the flash flood forecasts produced on thebasis of flash flood guidance are a necessary complementto the operational forecast programmes.7.4.5 Dam-break flash flood forecastingCatastrophic flash flooding results when a dam failsand the outflow passes through the breach in thedam and inundates the downstream valley. Methodsused to predict the floods that result from such failuresare described in 6.3.5.4.In recent years, the development of GIS and digitalterrain elevation data of high resolution has led tothe production of risk maps for specific areas downstreamof existing dams. These inundation mapsindexed with flood wave travel time informationare useful when distributed to local officials downstreamof a dam site for the development ofcontingency evacuation plans.7.4.6 Storm surges in riversStorm surges in the open sea are produced by windand atmospheric pressure and can generate gravitywaves that propagate upstream into rivers. Suitable

II.7-22GUIDE TO HYDROLOGICAL PRACTICEStechniques to forecast the development and propagationof the storm surge in the open sea, such asthe US National Weather Service SPLASH model(Jelesnianski, 1974) and its propagation into bays –as presented by Overland (1975) – are required todefine the surge at the river mouth, where it is thenrouted upstream via a suitable dynamic-routingtechnique. As the upstream movement of the gravitywave is opposed by the downstream flow, routingof the storm surge upstream may best be accomplishedby dynamic-routing techniques (see 6.3.5).Hydrological routing techniques or kinematichydraulicrouting techniques are not suited toprediction of wave motions that propagateupstream. Also, the inertial components of thegravity wave that are ignored in the diffusionhydraulicrouting techniques are too important tobe neglected in the case of a storm surge. A numberof papers on tidal rivers have been published by theUnited Nations Educational, Scientific and CulturalOrganization (1991). More recent applicationsinvolve the use of GIS to produce risk maps for areasprone to flooding by combined storm surge andflood waves (see publications of the WMO TropicalCyclone Programme).7.4.7 Urban floodingContinued urbanization of natural flood plains hascontributed to a sharp increase in loss of life anddamage to property. Rapid demographic and socialchanges, coupled with increasing land prices andenvironmental concerns relating to water pollutionand potential climate change characterized byincreased variability and extreme magnitude, makeadvances in urban water management worldwideall the more urgent (Pielke and Downton, 2000;Dabberdt and others, 2000).There are two types of urban flooding. First, urbanareas can be inundated by rivers overflowing theirbanks – this is fluvial flooding. Areas of inundationare forecast from the specific river-stage forecasts.Second, urban flooding can occur in local drainageas a special case of flash flooding.A considerable volume of literature has beenpublished on urban hydrology and water management,for example, reviews in Urbonas and Roesner,1993; Kovar and Nachtnebel, 1996; and Dabberdtand others, 2000. Unique characteristics of urbanhydrology are large areas of impervious or nearimperviousareas and the co-existence of bothnatural and technological drainage systems: sewers,levees, pumps, detention basins and the like. As aresult, surface runoff production from rainfall ishighly variable and non-homogeneous, and theflow of water and contaminants is acceleratedtoward higher peaks of outlet hydrographs. Highspatial-temporal variability in rainfall translatesinto high spatial-temporal variability in runoff, asthe urban catchments do not significantly dampensuch fluctuations. The technological drainage andimprovements to the natural drainage make forearlier and higher peak flows. With respect tohydrological impacts, the flood prediction andcontrol problem becomes severe for events with 5to 100 years’ return period, while the water qualityproblem can be acute with storms occurring withshort return periods of even less than two years.Owing to the characteristics of the urban responseto rainfall and contaminant forcing, very highspatial and temporal resolutions are required indata, models and controls over large urban areas inorder to ensure effective water resources management(Dabberdt and others, 2000). Thus, digitalterrain elevation data, distributed hydrologicalmodels and weather radar data – combined with insitu automated raingauge data and GIS – can beused to develop urban runoff forecast and managementsystems. (Cluckie and Collier, 1991; Bragaand Massambani, 1997; Georgakakos and Krajewski,2000; Kovar and Nachtnebel, 1996; Riccardi andothers 1997). In areas where significant urbangrowth is combined with mountainous terrain andconvective weather regimes (Kuo, 1993), there is agreat need to develop urban water resourcesmanagement systems capable of very high resolutionover large urban areas.7.4.8 Flooding from local drainageIn this case, intense rainfall over the urban area maycause flash flooding of streets and property in lowlyingareas or built-up areas in old waterways,underpasses and depressions in highways. Suchfloods arise primarily from inadequate storm-drainagefacilities, and are invariably aggravated by debrisclogging inlets to pipes and channels or outlets ofretention basins. Flood warning schemes similar tothose outlined for flash floods can be employed.These usually consist of local automated flash floodwarning systems or generalized warnings that arebased on national flash flood guidance operations. Itis also possible to target the flash flood guidance estimatesfor the urban environment on the basis ofvery high resolution digital spatial databases ofterrain, drainage network, both natural and technological,and existing hydraulic works.On causeways subject to flooding, traffic can bealerted by using lights activated in the same manneras the flash flood alarm system. Urban flooding

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-23usually affects sewer systems, even when wastewater and storm sewerage are piped separately.Forecasts of urban runoff can be helpful in the treatmentof sewage and the handling of polluted floodwater in combined systems. The opposite problemis the high level of pollution accompanying urbanrunoff. Since this is ultimately discharged into naturalwatercourses, it leads to increased pollution withproblems for downstream water users. The forecastingof pollution loads depends on forecasting urbanflood runoff.7.5 LONG-TERM FORECASTING7.5.1 Water supply forecastingWater supply forecasts are essential for the operationof domestic, industrial, irrigation andhydroelectric water supply systems. Forecastscommonly take the form of flow volumes overspecific durations: annual, seasonal or monthly.The duration depends on the nature of the demandand the amount of storage in the system. Sincewater supply forecasts cover a wider time span thanmeteorological forecasts, errors will always be inherentbecause of climatic events during the forecastingperiod. Therefore, it is recommended that severalforecast values with probabilities of exceedance beissued (see 7.3.4).The choice of the forecasting technique is governedby the character of the drainage basin, availabledata and user requirements. Water supply forecastsmay be made by using three basic techniques:(a) Snowmelt forecasts;(b) Conceptual models;(c) Time-series analysis.Snowmelt methods are used in basins wheresnowmelt runoff dominates the flow regime.Forecasting of snowmelt is described in 7.6.Normally, some measures of the snow-water equivalentand the basin losses can be related empiricallyto total seasonal runoff by regression techniques.Satellite measurements of snow cover have beenrelated to the discharge of the Indus river, for example,and reasonable results have been obtained inthis basin, where conventional ground data arevery scarce. These methods are primarily suited toforecasts of total runoff volume and do not describethe time distribution of the runoff.Conceptual models can be used for water supplyforecasting by running the model repeatedly andusing a number of historical climate time series asinputs. The output becomes a range of forecastedvalues to which probabilities of exceedance can beassigned. Models used for water supply forecastsshould be calibrated so that deviations betweenobserved and simulated runoff volumes are minimized.Since short-term variations are of minorimportance, simple model structures may yieldsatisfactory results.Time-series methods may be useful for water supplyforecasts, where discharge is a valid measure of thestate of the basin. The forecast relationships aregenerally very simple to apply. Regression modelsin which seasonal runoff is forecast from previoushydrological and climatic variables may be regardedas a special case of time-series methods.Long-term forecasts, especially of seasonal runoff,are often expressed in probabilistic terms: a statisticaldistribution of possible runoff volumes iscontingent on rainfall subsequent to the date whenthe forecast is made. One source of uncertainty isthe future weather between the date of preparingthe forecast and the operative date of the forecast.For example, if a regression-based forecast gives thefollowing equation:Q summer= b 0+ b 1R autumn+ b 2R winter+ b 3R spring+ b 4R summer(7.9)a less informative, probabilistic forecast can beissued after only receiving the rainfall data for theprevious autumn and winter. The probabilisticcomponent must take into account the distributionof possible spring and summer rainfalls that mightoccur.Unless the forecast model is very simple, it is almostcertain that it will be necessary to simulate possibleQ summervalues either by repeated sampling from thedistribution of R springand R summervalues or by repetitivelyapplying the model to the historical traces ofR springand R summer. If the sampling approach isadopted, it will be necessary to incorporate anycorrelation that might be present between the independentvariables. If the historical approach is used,at least 30 years of record is desirable to obtain arepresentative range of combinations. The applicationof this technique is not limited to regressionmodels. Any hydrological forecasting model can beperturbed retrospectively by real or synthetic datato construct a distribution of possible outcomes. Amore realistic description of the distribution ofactual values is obtained if a noise term is includedin the model. This can be accomplished by addingto each forecast a random number whose standard

II.7-24GUIDE TO HYDROLOGICAL PRACTICESdeviation is equal to the standard error of the modelestimate. A more detailed discussion is provided inLong-range Water-supply Forecasting (WMO-No. 587,Operational Hydrology Report No. 20).7.5.2 Flow recession forecastingLong periods without rain are a feature in manyparts of the world, particularly where continentaland highly seasonal tropical and subtropicalclimates prevail. The occurrence of prolonged dryperiods is significant for agriculture, which can beadapted to suit conditions by using particular practices,growing crops adapted to the conditions orproviding irrigation. Drought takes place when theperiod without rain extends beyond the normalduration, placing stress on plants and furtherdepleting water resources. It is therefore importantform an operational point of view to forecastdrought or to provide projections on how longdrought conditions will last.There is no single, simple definition of drought, asits nature will vary with the climate type, and theimpact of the drought, for example, water supply,irrigation and stock rearing. Where drought is aregular occurrence, its severity, which is a factor ofduration and temperature, becomes important.Drought extremes may result from an early startrelative to the normal dry season or a delay in thereturn of wet conditions, or a combination of both.A simple means of recording drought duration canbe defined as follows:(a) Drought begins after a period of 14 days withoutrain;(b) Drought ends after a period of 20 days duringwhich rainfall is recorded on 11 or more days;(c) As well as duration, intensity of drought can berecorded by cumulative temperature, that is, indegree days.The Palmer Drought Severity Index (Palmer, 1965)is widely used in the United States as a means ofdefining drought conditions. The method usescurrent and recent measurements of temperatureand rainfall, which are mathematically relatedthrough local mean values to provide an index ofseverity between –4, very dry, to +4, very wet. Themethod lends itself very well to mapping and GISdisplays and is routinely published on the Web bythe Drought Information Center of the NationalOceanic and Atmospheric Association (www.drought.gov).The characteristic behaviour of a river and, catchmentto drought can be expressed as a flow-durationcurve and a recession curve. A flow-duration curveis clearly a probability relationship taken over thewhole of the historic record, and it is thereforepossible to equate a current flow to a probabilitylevel, and thus project the situation for moreextreme flows. A flow-recession curve is constructedby plotting the relationship between flows at setinterval separations, for example daily, 5 days or10 days; the size of interval is influenced by thetotal length of the dry season and size of catchment.Thus plotting Q at t 0and t –5throughout theperiod of declining flow, a curve of the form:Q(t) = Q 0e –C(t–t 0 ) (7.10)is produced. Successive years of recession curvescan be combined by eye to give a full recession relationship.This allows the current situation to beassessed in terms of the overall catchment recessionand to provide an estimate of possible future durationand severity of projected conditions, forexample, one or two months ahead.Meteorological forecasts can be of value in droughtmanagement. Most major forecast services nowgive a long-range forecast for durations of two to sixmonths. These are broad in their approach, and areusually expressed in terms related to average orextreme conditions.Analysis of the falling limbs of hydrographs, or riverrecessions, is an important component of flood andlow-flow analysis; in forecasting, however, its use islargely confined to low-flow forecasting. Some lowflowforecasting is achieved by analysing masterrecession curves on the large river basins, thusmaking it possible to forecast weeks or even monthsahead. This type of forecasting is of value to hydropowerand irrigation schemes where the long-termsupply of water is vital to optimal managementpractice. In addition, there is a highly specializedarea where the principle long-term forecasts areproduced by meteorologists using sophisticatedglobal climate models. The subsequent hydrologicalwork then focuses principally on the developmentof forecasts of flow and aquifer levels for use withreservoir control rules and water allocationstrategies.The most direct method is probably to perform agraphical correlation between the current flow orstage and flow or stage n days ago wheren = 1, 2, 4, . . . (see 7.3.2.1). The defined relationshipcan be used to extrapolate forward in time ifthere are no disturbing influences, such as precipitationevents. Departures from the mostcharacteristic line can often be associated withnatural or man-made phenomena, and this

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-25information can also be brought to bear on anyparticular forecast.7.5.3 Time-series analysisA set of observations that measures the variation intime of a particular phenomenon such as the rate offlow in a river or the water level in a well or lake isdescribed as a time series. A time series can be specifiedin continuous or discrete time, depending onwhether observations of a system state variablesuch as flow are made continuously or quantizedinto a discrete set of measurements which approximatethe variation of the state variable over time(see Kottegoda, 1980).Since runoff is an indicator of the state of the drainagebasin, univariate time-series analysis may beused to establish forecast relationships. One suchapproach is to use autoregressive moving averagemodels, (Box and Jenkins, 1976) that are well suitedfor use in basins with limited precipitation data,because only antecedent discharge is needed tomake a forecast of this type:Q t+1= a 0Q t+ a 1Q t–1+ a iQ t–2+ ... + b (7.11)where Q t+1is the forecast with unit lead time andQ t–iare the measured values earlier than i timeincrements. Coefficients a iand b are estimated inthe time-series analysis. In addition to the forecastvalue Q t+1, a time-series model can yield the distributionof possible deviations from the forecastvalue so that an estimate of forecast error is readilyavailable. If a time-series forecast of monthly flowsis to be reliable, then autocorrelation in the monthlytime series must be large. This is the case in largerivers and in streams draining large aquifers andlakes. As a rule, however, forecasts are feasible onlyone to four months ahead. It is possible to includemeteorological variables in a time-series model but,if such data are available, it is often preferable tomake forecasts by using regression or a conceptualmodel. Time-series models may also be fitted to theerror series as discussed in the next section.7.6 SNOWMELT FORECASTS7.6.1 GeneralMany countries use forecast methods based onconceptual models of snowmelt runoff (see 6.3.3).Such methods make it possible to forecastsnowmelt from observational and forecast meteorologicaldata. Short- and medium-term forecasts arepossible for rivers and lowlands, and medium- andlong-term forecasts, for streams in mountainousareas. Seasonal volume forecasts may be preparedfor lowland and mountain basins, where snowmeltrunoff produces a significant portion of the totalstreamflow.Snowmelt runoff is a characteristic feature of theregime of lowland rivers in temperate and coldclimates and of some of the world’s largest rivers,even in tropical zones. Snowmelt runoff of manyrivers accounts for 50–70 per cent of the annualrunoff, and in dry regions the corresponding figuremay reach 80–90 per cent. Runoff estimates areused in reservoir management and planning forconsumptive use, power generation, public worksand land development. As a result, a number ofsnowmelt hydrological models have been developedto predict snowmelt runoff with a focus oncapturing or predicting peak flows and volumes forengineering design and reservoir managementpurposes.7.6.2 Snowmelt runoff processes inlowland and mountain riversDuring snowmelt, many of the processes thatgovern runoff in lowland and mountain river basinsare similar, for example, snowmelt, water retentionof snow, snowmelt inflow to a basin, snowmeltrunoff losses, water yield of a basin and time lag ofsnowmelt runoff to the outlet. However, some ofthe processes differ in two cases. For example, theyear-to-year variation in the snowmelt runoff lossesfrom snow and free water are significantly greaterin the plain regions than in mountainous riverbasins. More importantly, higher relief regions willhave very different snow-covered distributions,with elevation playing an important role in theamount, redistribution and sublimation of thesnowcover.The total snowmelt runoff from lowland basinsdepends on the water equivalent of the snow coverat the time the snow begins to melt – the volume ofprecipitation occurs after the snow has begun tomelt – and on the amount of water lost by infiltrationand evaporation over the river basin. The firstfactor can be determined to some degree by measurement;however, these measurements are highlylandscape dependent (see Volume I, Chapter 3, ofthis Guide). The second factor, the subsequentamount of precipitation and the water losses duringthe runoff period, must be handled by a forecastprocedure, either probabilistically or by assumingclimatological average values. The possibility ofusing numerical weather prediction for short-term

II.7-26GUIDE TO HYDROLOGICAL PRACTICESforecasts is becoming a viable option in forecastingmeteorological forcing. The third factor, snowmeltrunoff loss from the basin, is controlled by the infiltrationcapacity of the soil and surface-depressionstorage, including large non-capillary pores in theupper soil layer. Evaporation losses are relativelysmall and vary little from year to year. Snowpackaccumulation and ablation, especially during thespring thaw, are significant inputs into daily hydrologicalforecasting systems, which in turn, areextremely useful for flood prevention and hydroelectricgeneration. The measurement andcharacterization of the distribution of snow withina catchment are critical to the prediction of subsequentmelt.Volume I, Chapter 3, of this Guide states that catchment-basedassessments of snow are typicallyderived from snow surveys and snow courses and,as a rule, recommends that in high relief regions,snow courses be at elevations and exposures wherethere is little or no melting until peak accumulationis achieved. In mild to low relief regions, thesesurveys need to represent the average snow conditionswithin a given catchment, and should becarried out on a variety of landscapes in order toproperly depict the natural variability of thelandscape.Infiltration of water into the soil during the snowmeltperiod is a factor that varies greatly fromyear to year, depending on the soil conditions.The rate of infiltration into frozen soil and thetotal amount of water absorbed depend on thesoil moisture content prior to freezing, thetemperature, the depth of freezing and the soil’sphysical properties. The size of the area coveredby depression storage can be expressed mathematicallyas distribution functions of the depthof water required to fill these depressions. Suchfunctions are relatively stable characteristics foreach river basin.difference in altitude between the head andthe mouth of the river system, as well as thevariability of hydrometeorological conditionswith the zone. In the experience of somehydrologists, the optimum altitude rangefor such zones is 200 to 400 metres with thenumber of zones around 20;(c) The models are calibrated with hydrometeorologicaldata from preceding years;(d) The forecast flows for the partial basins, or altitudezones for mountain areas, are routed to adownstream forecast point (see 6.3.5).7.6.4 Long-term snowmelt forecastsTo devise a method for long-term forecasting ofsnowmelt runoff, it is necessary to establish waterbalancerelationships. This is preceded by thefollowing tasks:(a) Determination of the relevant characteristics ofthe river basin, such as topography, land-coverdistribution and the nature of the soils;(b) Determination of any factors governingthe way in which water is absorbed by thesoil and retained on the surface of the drainagearea;(c) Definition of the basic factors governingthe loss of water in the river basin and theextent to which such factors vary from year toyear;(d) Determination of the role of precipitationoccurring after the snowmelt has begun, inrelationship to runoff, and of the variability ofsuch precipitation;(e) Evaluation of the accuracy of data for runoff,snow-water equivalent and precipitation.Snowmelt runoff forecasts may be improved andextended by including probabilisticallyrepresentative data and/or quantitative meteorologicalforecasts for the subsequent snowmeltperiod.7.6.3 Short- and medium-term snowmeltrunoff forecastsShort- and medium-term snowmelt runoff forecastsfor large river basins may be developed asfollows:(a) Lowland river basins are divided into small,partial basins, which are assumed to behydrometeorologically homogeneous, eachwith an area of up to 15 000 km 2 , and the riversystem is divided into sections beginning withthe upper reaches;(b) Mountainous basins are divided into altitudezones. The number of zones depends on the7.6.4.1 Seasonal snowmelt forecasts for theplains regionsThe relationship between total snowmelt runoffQ nand the snow-water equivalent for plains areasmay be expressed theoretically as:Q n = (w n − f )∫w n−fow n−ff ( y d ) dy d− ∫ y d f ( y d )dy d(7.12)where w nis the snow-water equivalent and f is thetotal infiltration during the snowmelt period, bothexpressed in millimetres. The function ƒ(y d) is thearea distribution function in relation to the deptho

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-27of water, (y d), that is necessary to fill depressions onthe river basin surface.In the absence of infiltration or when its intensity ispotentially greater than the rate of snowmelt, equation7.12 can be simplified as follows:w nw nQ n = w n ∫ f ( y d ) dy d − f ( yo ∫ y d )dyo d d(7.13)In this case, the runoff becomes a function of thesnow-water equivalent and the infiltration capacityof the basin.The amount of water contributing to the seasonalsnowmelt runoff is calculated for each year as thesum:W = w – n + P– (7.14)where w – nis the mean snow-water equivalent for thebasin at the end of winter and P – is the mean precipitationduring the runoff period, both expressed inmillimetres.The mean snow-water equivalent for the basin maybe calculated as either an arithmetic mean or aweighted mean. The arithmetic mean method isused when the number of snow-measuring stationsin the basin is sufficiently large and when the spatialdistribution of these stations is good. The weightedmean method is used when observation points areunevenly distributed over the area and/or when thedistribution of the snow cover is irregular. To calculatethe weighted mean of the snow-waterequivalent, a map showing snowcover averagedistribution in the area is drawn.In regions where a thaw may take place in winter,an ice crust often forms on the ground. If measurementsare available, the amount of water containedin such crusts should be added to the snow-waterequivalent. Very often, direct determination of soilmoistureconditions throughout the river basin,particularly in winter, is not feasible because of thelack of adequate data. This is the main reason whyindirect indices are so common.In dry steppe regions, the difference betweenprecipitation and evapotranspiration characterizesthe potential rate of infiltration. In the humid forestzone where every year the autumn soil-moisturecontent is equal to, or greater than, field capacity,this difference represents changes in the storage ofthe basin as a whole. The runoff caused by lateautumn precipitation can also be used as an indexof the retention capacity of river basins in theseregions.7.6.4.2 Seasonal snowmelt forecasts formountainous regionsIn mountainous areas, there tend to be considerabledifferences in climate, soil and vegitationbecause of the range of altitudes. These featuresdetermine the nature of the snowmelt runoff andflow regime of the streams. Therefore, the mostimportant characteristic of a mountain basin isits area–elevation distribution. The main sourcesof runoff are seasonal snow, which accumulatesin the mountains during the cold season, andprecipitation that occurs during the warm seasonof the year.Owing to the long period between the beginningand end of the snowmelt period, long-term forecastsof the seasonal flow of mountain rivers arefeasible. The most favourable conditions for suchforecasts exist where seasonal snow is the mainsource of runoff and the amount of summer precipitationis relatively small.Steep slopes, rocks and an extensive, highly permeabledeposit of rough rubble in mountainous basinscreate conditions in which the water finds its wayinto channels, mainly through layers of rubble andclefts in the rocks. Under such conditions, waterlosses do not vary greatly from year to year, andthere should be a good relationship betweenseasonal runoff and the amount of snow in thebasin. This relationship can be established empiricallyif measurements are available for a number ofyears. However, in practice, determining such relationshipsis often difficult.7.7 FORECASTS OF ICE FORMATION ANDBREAK-UP[HOMS J45]7.7.1 GeneralMany rivers and lakes in middle latitudes freezeover in winter. The most important ice regimephases for which forecasts are made are as follows:(a) The first appearance of ice;(b) The formation of complete ice cover;(c) The break-up of the ice cover;(d) The final disappearance of all ice.The ice regime of rivers is closely related to weatherconditions. Thus, the dates of the appearance offloating ice and those of the formation and breakingof the ice cover vary over a wide range fromyear to year. Ice forecasts are of great practical value

II.7-28GUIDE TO HYDROLOGICAL PRACTICESfor navigation, but many other users apart fromthose in inland navigation are interested in theseforecasts as well.Exact relationships for calculating thermal and iceregimes are available; however, their application toice forecasting is severely limited by the stochasticnature of parameters governing the equations,which vary over the time span between the forecastand the event predicted. This subsection discussesthe different ice-regime forecasts that exist andshort-term forecasts of ice formation and icebreak-up.Modern approaches to the short-term forecasting ofice phenomena are based on thermal balance (Buzinand others, 1989). For forecasts of autumn icephenomena, formulae for the thermal balance atthe boundary between a unit of surface of waterand the adjacent atmosphere are used. Factorsinclude direct heat exchange, solar radiation, turbulentheat and moisture exchange with theatmosphere, effective radiation, inflow of heat fromthe Earth’s surface and groundwater, dissipation ofstream energy as heat and the inflow of thermalenergy from precipitation which falls on the watersurface and from the discharge of industrial andhousehold waste waters. While the role of each ofthese factors in the thermal balance is different, themost important one is the exchange of heat throughthe open water surface.Forecasts of the time when the ice cover breaks upare based on calculations of the tensile strength ofthe melting snow-ice cover, using formulae for thethermal balance and the derivation of a ratiobetween the durability of an ice cover and thedestructive force at which the ice sheet is fractured.The latter is a function of the stream discharge, itswater level and the rate at which they have changedover time.Methods of modelling the formation and break-upof ice are covered in 6.3.6.3 of the present Guide.7.7.2 Long-term ice forecastsThe development of methods for long-term forecastingof ice phenomena usually includes thefollowing tasks:(a) Consideration of the dates of ice formationand break-up on rivers across the area underconsideration, for example, average dates, variabilityof the annual dates and the delineationof regions with uniform ice phenomena, themain mathematical instrument being statisticalanalyses;(b) Synoptic analysis of conditions leading tofreeze-up or ice break-up, in which the northernhemisphere is divided into typical regions,the main mathematical instrument here beingdiscriminant analyses;(c) Analysis of the distribution of stores of heat inthe surface layers of oceans, such as the NorthernAtlantic and the northern Pacific Ocean;identifying the principles areas of interest,stores of heat within the limits of which renderthe greatest influence on processes leading tothe formation and destruction of an ice coveron the rivers, the main mathematical instrumentagain being discriminant analyses;(d) Determination of quantitative variables foratmospheric processes and ocean fields, such asexpanding meteorological and ocean fields byorthogonal functions;(e) Use of correlation analyses to determine therelationship between the time of ice occurrenceand the variables representing the appropriatemeteorological and ocean fields.7.7.3 Ice jams and methods of forecastinghigh water levelsDangerous rises in water level and resulting floodsmay occur during the formation of an ice cover orice jam, and as a river’s ice cover or ice dam breaksup. Jam floods are especially hazardous becausethey occur in a cold season and sometimes remainfor a long time. This causes sheets of water freeze toform an ice field that can cover populated areas andcan be almost impossible to remove. Often a sharpfall in water level occurs below the ice jam, starvingintakes of water and interrupting water supplies.On many rivers, maximum ice dam water levelsexceed the highest water levels of spring andsummer floods. Ice dams can form quickly andcause very rapid increases in water level, withoutany significant increase in flow.Ice jams arise more often in those reaches of riverswhere ice cover grows out from each bank towardsthe middle and in an upsteam direction. The moreslowly this process develops, the more ice is broughtby the current under the frozen ice cover, causingcontraction of the flow channel and raising thewater level at the approach of the ice field. Suchcharacteristics of freezing are typical of large riversflowing towards the poles and rivers flowing out oflarge lakes and after-bays of hydroelectric powerstations.River discharge during freezing, conditions for heatexchange, including air temperature, and the

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-29position of the ice edge in relation to the crosssectionare the primary factors to be taken intoaccount in forecasting water levels from ice dams.For a number of rivers where dangerous jams arefrequently observed, physical–statistical relationshipshave been developed to account for thesefactors. The following example relates to the Nevariver at Saint Petersburg (Buzin and others, 1989):H jam= 1.29H X1+ 0.53L + 0.24H G– 404 (7.15)where H X1is the average level of the Ladozhskoyelake in centimetres in November; L is the distanceof the ice edge in kilometres from the GorniyInstitute station; H Gis the water level at that stationin centimetres.Interestingly, by knowing only H X1, it is possible toissue good warnings of high water levels from icejams more than one month in advance and toupdate these for three- to five-day short-term forecastsusing equation 7.15.A generalized method is available for deriving shorttermforecasts of the maximum water level resultingfrom ice jams at critical locations on a river, includingthose for which there are no long time series ofhydrological observations. Initial data required arethe gradient of the given reach of the river, waterdischarge on the day that ice appears (Q 0), airtemperature during the freezing ice over the lastfew days and the curve Q = f(H) at the free ice channel.For a forecast, it is necessary to define a value ofthe critical gradient l ’ given by:l ’ = 0.0154 g/c 2 (1 – e) (7.16)where g is the acceleration due to gravity; c is theChezy coefficient; e is the porosity of ice fields estimatedaccording to air temperature, where e = 0.25at θ = –10°C and e = 0.55 at θ = –2°C.If equation 7.16 is applied, the difference between I ’and the gradient of the open surface can be usedwith a special table developed for each hydrologicalpost using the data of hydrological observations todefine a conversion factor k p, the winter dimensionlessfactor, between winter water discharge and thecorresponding open channel discharge. The valueof the discharge Q kris calculated using the followingequation:Q kr= Q 0e –k 0 T ice (7.17)where k 0depends on the weather conditions duringfreezing; T iceis the duration of the ice run in days.For example the coefficient k 0for the Amur rivercan be calculated using the following equation:k 0= 0.005 – 0.00333T XI(7.18)T XIis the average air temperature in Chabarovsk inOctober.The maximum water level for an ice jam is determinedby using the reduced discharge Q ’ (ahypothetical summer water discharge which raiseswater levels and would create an ice jam in winter)and summer curve Q = f(H). In this case it is necessaryto calculate the reductioned discharge Q ’ whichcan be estimated roughly as:Q ' = Q kr(7.19)k pSpring ice dams that build up on rivers break-updownstream under the influence of spring floodwaves from the upper part of the basin. Thisphenomena is particularly important for southnorth-flowingrivers in Canada and the northernparts of Europe and the Russian Federation. The icebreak-up on these rivers takes the shape of a chainreaction of the consecutive formation and destructionof ice dams of varying magnitude.The maximum ice dam water level on a given riverreach depends on many factors, which can bedivided into those related to the process of ice coverformation and those related to its destruction. Themost powerful dams and catastrophic floods arisewhen high water levels occur during ice coverformation. This can arise when high flows inautumn meet a channel constricted by ice slush,especially if the freezing of the river is accompaniedby some movement of the ice cover (Buzin andothers, 1989). It can also occur when a rapid riseduring a spring flood in the upper reaches coincideswith a sharp cooling at the limit of the ice cover inthe reach under consideration so that the icebecomes particularly durable and forms an ice damin a downstream reach.The factors of the first group make it possible insome river reaches, for example, on the Amur,Angara and Sukhona rivers, to predict maximumice-dam water levels over periods of one to fourmonths, using the following equation:H t,dam= 180 + 2.18 H x(7.20)where H t,damis maximum ice dam water level incentimetres; H xis the water level at the period of icecover formation in centimetres.

II.7-30GUIDE TO HYDROLOGICAL PRACTICESThe addition of ice cover formation characteristicsas parameters linked to the peculiarities of thespring processes has made it possible in the Lenariver basin to predict the probability of occurrenceof dangerous ice dam water levels during the icebreak-up period for each of the four basic sectionsof the river with forecast lead times of 20 to 40 days.Using the relationship between ice thickness overthe main part of the river and ice thickness at themain city within this part, it is possible to predictwhether an ice dam will threaten the city or willform at another location. The probability of acorrect prediction of dangerous levels is 80 percent.However, for many rivers where ice dams pose aparticular risk, the development of methods for longtermforecasting is problematic and, where suchforecasts are made, they frequently require correction.For this reason, a number of methods havebeen devised to provide short-term forecasts whereice dams occur each year. These forecasts are basedon the physical and statistical relationships whichtake into account the major factors listed above. In anumber of cases, a weather forecast for three to fivedays is taken into account to estimate the characteristicsof ice durability and probability of cooling.Some recommended methods relate to the forecastingof ice dam water levels for any part of the river,even in the absence of long observation series. Insuch cases, it is possible to use the discharge curve,Q = f(H), and local meteorological data to calculatethe relevant ice durability characteristics. The maximumwater level at an ice dam is determined fromQ = f(H) and an appropriate Q ’ is calculated fromequation 7.19. In this case, Q ’ is a conditionalsummer discharge, which could cause such a rise ofwater level, which occurs in ice dam formation. Inequation 7.19, k pis the winter factor for the periodFigure II.7.9. River reach with an ice dam

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-31of ice dam formation. This coefficient k pis derivedfrom the link with characteristics of an ice coverexpressed by the following equation:k p = 8.13⎛⎝ϕ h iceB⎞⎠0.38( k ice − 1) + 1 (7.21)where ϕ is the relationship between ice durabilityon the last day before ice cover break-up and icedurability on the day when snow disappears fromthe surface of the ice, which can be derived usingtechniques described in a number of publications(Buzin and others,1989); h iceis the thickness of icein metres before ice cover break-up; B is the widthof the river in metres; k iceis the winter factor at amaximum water level at the beginning of freezingin the autumn (for various river catchments, k ice=0.65 0.85). The winter factor can also be calculatedas follows:1k ice = 1 −1.1 − 1 log gF(7.22)where F is the catchment area in km 2 located abovean ice dam.Where there is a river reach without tributariesabove an ice dam, and there is a hydrological gaugein this part of the river (Figure II.7.9), it is possibleto use the method of equivalent on ice phasesdischarges for the forecast of discharge at the icedam (Q kr1):Q kr1= k p2Q kr2F 1/F 2(7.23)where k p2is the winter factor for the date of ice damformation in the upper section, Q kr2is the dischargein the upper section, appropriated for a maximumice dam level in the upper section according tosummer curve discharges Q = f (H), and F 1and F 2are the areas of the basin closed by the lower andupper sections of the reach, F 1= F 2+ ΔF.The following Websites provide valuable informationfor hydrological forecasting services:http://edc.usgs.gov/http://k12science.ati.stevens-tech.edu/curriculum/drainproj/reference.htmlhttp://nsidc.org/snow/http://snr.unl.edu/niwr/http://ulysses.atmos.colostate.edu/~odie/snowtxt.htmlhttp://water.usgs.gov/http://water.usgs.gov/listurl.htmlhttp://www.afws.net/http://www.cpc.ncep.noaa.gov/products/expert_assessment/threats.shtmlhttp://www.cpc.ncep.noaa.gov/products/fews/http://www.dartmouth.edu/artsci/geog/floods/http://www.epa.gov/ebtpages/water.htmlhttp://www.hpc.ncep.noaa.gov/nationalfloodoutlook/http://www.ibwc.state.gov/wad/rtdata.htmhttp://www.iwr.usace.army.mil/http://www.msc.ec.gc.ca/crysys/http://www.ncdc.noaa.gov/ol/climate/climateextremes.htmlhttp://www.nohrsc.nws.gov/http://www.nws.noaa.gov/oh/hads/http://www.nws.noaa.gov/ohd/hdsc/http://www.nwstc.noaa.gov/HYDRO/RFS/NWSRFS.htmlhttp://www.pecad.fas.usda.gov/cropexplorer/global_reservoir/http://www.sce.ait.ac.th/programs/courses/IWRM/Online_references.htmhttp://www.worldclimate.com/http://www.wri.org/watersheds/References and further readingAnderson, E. A., 1973: National Weather Service RiverForecast System: Snow Accumulation and AblationModel, Programs and Test Data. NOAA NWSHYDROTechnical Memorandum 17.Appolov, B. A., G.P. Kalinin and V.D. Komarov, 1974:Course on Hydrological Forecasting. Leningrad,Gidrometeoizdat, 418 pp.Barrett, C.B., 1999: Successful Developmentand Engineering of Global Integrated WaterManagement Systems, American Society of CivilEngineers, International Activities Committee,Roundtable Discussion paper, Washington, DC,1999.———, 2003: WMO-NOAA Hydrologic ForecastingCourse, 14 October–7 November 2003, Kansas City,Kansas.Barrett, C.B. and others, 1985: Hydrology Subcommitteeof the Interagency Advisory Committee on WaterData, Guidelines on Community Local FloodWarning and Response Systems, Springfield, 1985.Bellerby, T., M. Todd, D. Kniveton and C. Kidd, 2001:Rainfall estimation from a combination of TRMMprecipitation radar and GOES multispectral satelliteimagery through the use of an artificialneural network. Journal of Applied Meteorology,39(12):2115–2128.Bergstroem, S., 1976: Development and Applicationof a Conceptual Runoff Model for ScandinavianCatchments. SMHI Rapporter No. RH07, HydrologicalOceanography.Beven, K., 2002. Towards an alternative blueprint fora physically based digitally simulated hydrologicresponse modelling system. Hydrological Processes,16:189–206.

II.7-32GUIDE TO HYDROLOGICAL PRACTICESBorovikova, L.N., Y.M. Denisov, E.B. Trofimova andI.D. Shencis, 1972: Matematicheskoe modelirovanieprocessa stoka gornyh rek (Mathematical modellingof the runoff formation for mountain rivers). TrudySredaznigmi, Vyp. 61(76).Borsch S.V. and T.P. Silantjeva, 1987: Method of theshort-term and medium-term forecast of openingof the rivers on the basis of the generalizeddependence. The methodical instructions. Moscow,Hydrometeorological center of USSR, pp. 28.———, 1989: The improvement of the method ofaccount of an ice cover destruction at the riversand reservoirs. Report of the HydrometeorologicalCenter of USSR. v. 309, pp. 113–120. (in Russian).Box, G.E.P. and G.M. Jenkins, 1976: Time-series analysis:Forecasting and Control. San Francisco, Holden-Day.Braga, B., Jr. and O. Massambani, (eds.) 1997: WeatherRadar Technology for Water ResourcesManagement. Montevideo, UNESCO Press, 516 pp.Brier, G.W. and R.A. Allen, 1951: Verification of WeatherForecasts. Compendium of Meteorology, (T.F.Malone, ed.) American Meteorological Society,pp. 841–848.Brun, E., P. David, M. Sudak and G. Brunot, 1992; Anumerical Model to Simulate snow-cover stratigraphyfor operational avalanche forecasting, Journal ofGlaciology, 38(128):13–22.Buzin V.A., G.I. Bolotnikov and A.M. Filippov, 1989: Icejams on the rivers – methods of study, account andforecast. In book: Problems of nowadays of hydrology.Leningrad, Hydrometeoizdat, pp. 220–231 (inRussian).Carpenter, T.M. and K.P. Georgakakos, 2004: Impacts ofparametric and radar rainfall uncertaintyon the ensemble streamflow simulations of adistributed hydrologic model. Journal of Hydrology(in press).Carpenter, T.M., K.P. Georgakakos and J.A. Sperfslage,2001: On the parametric and NEXRAD-radar sensitivitiesof a distributed hydrologic model suitable foroperational use. Journal of Hydrology, 254:169–193.Carpenter, T.M., J.A. Sperfslage, K.P. Georgakakos,T. Sweeney and D.L. Fread, 1999: National thresholdrunoff estimation utilizing GIS in support ofoperational flash flood warning systems. Journal ofHydrology, 224:21–44.Carroll, T. R.,1995: Remote-sensing of snow in thecold regions. 1995. First Moderate ResolutionImaging Spectroradiometre (MODIS) Snow and IceWorkshop; NASA/Goddard Space Flight Center,Greenbelt, Maryland. 3–14.Chagnon, S.A., 1999: Record flood-producing rainstormsof 17–18 July 1996 in the Chicago metropolitanarea. Part III, Impacts and responses to the flashflooding. Journal of Applied Meteorology, 38:273–280.Cluckie, I.D. and C.G. Collier, eds, (1991): HydrologicalApplications of Weather Radar, EnvironmentalManagement, Science and Technology Series,Chichester, Ellis Horwood, 644 pp.Cluckie, I.D. and D. Han, 2000: Fluvial FloodForecasting, Journal of Water and EnvironmentalManagement. 14(8) Chartered Institution of Waterand Environmental Management (CIWEM).Colbeck, S.C. (ed), 1980: Dynamics of Snow and IceMasses. New York, Academic Press, pp. 305–395.Collier, C.G. and R. Krzysztofowicz, 2000: Special Issueon Quantitative Precipitation Forecasting. Journal ofHydrology, 239:1–337.Collinge, V.K. and C. Kirby (eds), 1987: Weather Radarand Flood Forecasting. New York, Wiley, 296 pp.Dabberdt, W.F., J. Hales, S. Zubrick, A. Crook, W.Krajewski, J. Christopher Doran, C. Mueller,C. King, R.N. Keener, R. Bornstein, D. Rodenhuis,P. Kocin, M.A. Rossetti, F. Sharrocks andE.M. Stanley Sr, 2000: Forecast issues in the urbanzone, Report of the 10th Prospectus Developmentteam of the U.S. Weather Research Program.Bulletin of the American Meteorological Society,81(9):2047–2064.Donald J.R., E.D. Soulis, N. Kouwen and A. Pietroniro,1995: A land cover-based snow cover representationfor distributed hydrologic models. Water ResourcesResearch, 31:(4):995–1009.Draper, N.R. and H. Smith, 1966: Applied RegressionAnalysis. New York, Wiley.Evensen, G., 1994: Sequential data assimilation with anon-linear quasi-geostrophic model using MonteCarlo methods to forecast error statistics, Journal ofGeophysical Research, 90(C5):10143–10162.Finnerty, B.D., M.B. Smith, D.-J. Seo, V. Koren andM.G. Moglen, 1997. Space-time sensitivity of theSacramento model to radar-gage precipitationinputs, Journal of Hydrology, 203:21–38.Fleming, G., 1975: Computer Simulation Techniques inHydrology. New York, Elsevier.Fulton, R.A., J.P. Breidenbach, D.-J. Seo and D.A. Miller,1998: The WSR-88D rainfall algorithm. Weather andForecasting, 13(2):377–395.Georgakakos, K.P., 2002: US corporate technology transferin hydrometeorology. Journal of Hydroinformatics,4(1):3–13.Georgakakos, K.P., 2004: Analytical results for operationalflash flood guidance. Journal of Hydrology (inreview).Georgakakos, K.P. and J.A. Sperfslage, 2004: Operationalrainfall and flow forecasting for the PanamaCanal watershed. Chapter 16 in Rio Chagres, AMultidisciplinary Profile of a Tropical Watershed(R.S. Harmon (ed.)). Dordrecht, Kluwer AcademicPublishers, 16.1–16.13 (in press).Georgakakos, K.P. and W.F. Krajewski (eds.), 2000:Hydrologic Applications of Weather Radar. SpecialIssue of Journal of Geophysical Research – Atmospheres,105(D2):2213–2313.

CHAPTER 7. HYDROLOGICAL FORECASTINGII.7-33Ginzburg B.M., S.V. Borsch and N.D. Efremova, 2001:Methods of the background long-range andmedium-range forecasting of dates of opening of therivers of the European territory of Russia. Journal ofMeteorology and Hydrology, 11:67–78.Ginzburg B.M., S.V. Borsch, N.D. Efremova andM.I. Silnitskaya, 1998: A method for backgraundlong-term prediction of ice appearance datesfor rivers and reservoirs in European Russia andByelorussia. Journal of Meteorology and Hydrology,6:84–91.Ginzburg B.M., S.V. Borsch, N.D. Efremova,M.I. Silnitskaya and K.N. Polyakova, 2002: Methodsof long-range and medium-range forecasting ofdates of the spring flood maximal level on rivers onEuropean territory of Russia. Journal of Meteorologyand Hydrology, 11:81–92.Ginzburg B.M. and M.I. Silnitskaya, 2000: Influenceof the oceans surface of temperature distributionon the on terms of opening of the rivers of theEuropean territory of Russia. Journal of Meteorologyand Hydrology, 6:86–91.Gray, D.M. and D.H. Male (Eds.), 1981: Handbook ofSnow: Principles, Processes, Management and Use.Toronto, Pergamon Press.Hedstrom, N.R. and J.W. Pomeroy, 1998: Measurementsand modelling of snow interception in theboreal forest. Hydrological Processes,12(10–11):1611–1625.4.Hydrometeoizdat, 1989: Guide on the hydrological forecasts.The ice forecasts. Vol. 3. Leningrad, 168 (inRussian).Jelesnianski, C.P., 1974: SPLASH II: General Trackand Variant Storm Conditions. NOAA TechnicalMemorandum, NWS, TDL-S2. US Department ofCommerce, March.Joliffe, T. Ian and B.D. Stephenson, 2003: ForecastVerification A Practitioner’s Guide in the AtmosphericSciences. Wiley, 240 pp.Jones, H.G., J. Pomeroy, D.A. Walker and R.W. Hoham,2001: Snow Ecology: an interdisciplinary examination ofsnow-covered ecosystems. Cambridge University Press,377 pp.Koren, V.I., B.D. Finnerty, J.C. Schaake, M.B. Smith,D.-J. Seo and Q.Y. Duan, 1999: Scale dependencies ofhydrology models to spatial variability of precipitation.Journal of Hydrology, 217:285–302.Kottegoda, N.T., 1980: Stochastic Water ResourcesTechnology, London, Macmillan Press.Kovar, K., and H.P. Nachtnebel (eds), 1996: Applicationof Geographic Information Systems in Hydrologyand Water Resources Management. IAHS PublicationNo. 235. Wallingford, IAHS Press, Institute ofHydrology, 711 pp.Krzysztofowicz, R., 1992: Bayesian correlation score: autilitarian measure of forecast skill. Monthly WeatherReview,120:208–219.Kummerow, C., W. Barnes, T. Kozu, J. Shiue andJ. Simpson, 1998: The Tropical Rainfall MeasuringMission (TRMM) sensor package. Journal ofAtmospheric and Oceanic Technology, 15(3):809–817.Kuo, C.Y. (ed.), 1993: Engineering Hydrology. AmericanSociety of Civil Engineers, New York, pp. 989–994.Mogil, H.M., J.C. Monro and H.S. Groper, 1978: NWSflash flood warning and disaster preparednessprograms. Bulletin of the American MeteorologicalSociety, 59:690–699.Munich Re, 2001: Topics, Annual Review: NaturalCatastrophes 2000, Munich.Murphy, A.H. and R.L. Winkler, 1987: A general frameworkfor forecast verification. Monthly WeatherReview, 115:1330–1338.Murphy, A.H., 1988: Skill scores based on the meansquare error and their relationships to the correlationcoefficient. Monthly Weather Review,116:2417–2424.Murphy, A.H., 1996: General decompositions ofMSE-based skill scores: measures of some basicaspects of forecast quality. Weather and Forecasting,124:2353–2369.National Oceanic and Atmospheric Administration andthe United Nations Department of Economic andSocial Affairs, 2004: Guidelines for Reducing FloodLosses, 87 pp.National Research Council (NRC), 1991: Opportunities inthe Hydrologic Sciences. Washington, DC, NationalAcademy Press, 348 pp.Ogden, F.L., J. Garbrecht, P.A. DeBarry and L.E. Johnson,2001: GIS and distributed watershed models, II,Modules, interfaces and models. Journal of HydrologicEngineering, 6(6):515–523.Overland, J.E., 1975: Estimation of Hurricane StormSurge in Apalachicola Bay, Florida. NOAA TechnicalMemorandum, NWS-17, US Department ofCommerce, March.Palmer, W.C., 1965: Meteorological Drought, ResearchPaper No. 45, US Weather Bureau. Washington, DC.Pielke, R.A., Jr and M.W. Downton, 2000: Precipitationand damaging floods, Trends in the United States,1932-1997. Journal of Climate, 13:3625–3637.Pomeroy, J.W. and D.M. Gray (eds.), 1995: SnowcoverAccumulation, Relocation and Management. NationalHydrology Research Centre, Saskatoon, Sask.Canada. NHRI Science Report No. 7, 144 pp.Pomeroy, J.W. and E. Brun, 2001: Physical Propertiesof Snow. In: Snow Ecology: An InterdisciplinaryExamination of Snow-covered Ecosystems.(H.G. Jones, J.W. Pomeroy, D.A. Walker, R.W.Hoham, eds). Cambridge, Cambridge UniversityPress, pp. 45–126.Rango, A., J. Martinec, J. Foster and D. Marks, 1983:Resolution in operational remote-sensing of snowcover. Hydrological Applications of Remote-sensingand Remote Data Transmission, Proceedings of the

II.7-34GUIDE TO HYDROLOGICAL PRACTICESHamburg Symposium, August 1983. IAHS Publ.No. 145.Riccardi, G.A., E.D. Zimmermann and R.-A. Navarro,1997: Zonification of areas with inundation risk bymeans of mathematical modelling in the Rosarioregion Argentina. In: Destructive Water, Water-CausedNatural Disasters, their Abatement and Control (G.H.Leavesley, H.F. Lins, F. Nobilis, R.S. Parker, V.R,Schneider, F.H.M. van de Ven, eds). IAHS Press,Institute of Hydrology, Wallingford, 123–135.Seo, D.-J. and J.P. Breidenbach, 2002: Real-time correctionof spatially nonuniform bias in radar rainfalldata using rain gauge measurements. Journal ofHydrometeorology, 3(2):93–111.Shi, J.C., J. Dozier and H. Rott, 1994: Snow Mappingin Alpine Regions with Synthetic Aperture Radar.IEEE Transcations on Geoscience and Remote-sensing,31(1):152–158.Sittner, W.T. and K.M. Krouse, 1979: Improvementof Hydrologic Simulation by Utilizing ObservedDischarge as an Indirect Input (ComputedHydrograph Adjustment Technique – CHAT). NOAATechnical Memorandum NWS HYDRO-38, February,US Department of Commerce.Sittner, W.T., C.F. Schauss and J.C. Monro, 1969:Continuous hydrograph synthesis with an API-typehydrological model. Water Resources Research, 5(5).Smith, M.B., K.P. Georgakakos and X. Liang, (eds),2004a: The Distributed Model IntercomparisonProject (DMIP). Special Issue of Journal of Hydrology(in press).Smith, M.B., D.-J. Seo, V.I. Koren and co-authors, 2004b:The Distributed Model Intercomparison Project(DMIP): An Overview. Journal of Hydrology(in press).Sweeney, T.L., 1992: Modernized areal flash flood guidance.NOAA Technical Report NWS HYDRO 44,Hydrology Laboratory, National Weather Service,NOAA, Silver Spring, MD, October, 21 pp. and anappendix.Tachikawa, Y., B.E. Vieux, K.P., Georgakakos andE. Nakakita (eds), 2003: Weather Radar Informationand Distributed Hydrological Modelling. IAHSPublications No. 282. Wallingford, IAHS Press,323 pp.Todini, E. and J.R. Wallis, 1978: A Real-time RainfallRunoff Model for an On-line Flood Warning System.AGU Chapman conference on applications ofkalman filtering theory and techniques to hydrology,hydraulics and water resources, Pittsburgh,Pennsylvania, 22–24 May 1978.Twedt, T.M., J.C. Schaake and E.L. Peck, 1977: NationalWeather Service extended streamflow prediction.Proceedings of the western snow conference,Albuquerque, New Mexico, 19–21 April 1977.US Bureau of Reclamation, 1960: Design ofSmall Dams. Appendix A, Washington, DC,pp. 413–431.US Department of Commerce, 2002: Final Reporton Hurricane Reconstruction Program, CentralAmerican and the Dominican Republic,May 2002.United Nations Educational, Scientific and CulturalOrganization (UNESCO), 1991: InternationalWorkshop on Storm Surges, River Flow andCombined Effects. 8–12 April 1991, Hamburg, Acontribution to the UNESCO-IHP project H-2-2.Urbonas, B.R. and L.A. Roesner, 1993: Hydrologic designfor urban drainage and flood control. In: Handbookof Hydrology (D.R. Maidment, ed.), New York,McGraw-Hill Inc., 28.1–28.52.Weaver, J.F., E. Gruntfest and G.M. Levy, 2000:Two floods in Fort Collins, Colorado, Learningfrom a natural disaster. Bulletin of the AmericanMeteorological Society, 81(10):2359–2366.World Bank, 2002: 2002 World Development Indicators.Washington, DC, World Bank, 405 pp.World Meteorological Organization, 1969: HydrologicalForecasting, Technical Note No. 92, WMO-No. 228,Geneva.———, 1975: Intercomparison of Conceptual Models Usedin Operational Hydrological Forecasting. OperationalHydrology Report No. 7, WMO-No. 429,Geneva.———, 1982: Long-range Water-supply Forecasting, editedby M. Dyhr-Nielsen. Operational Hydrology ReportNo. 20, WMO-No. 587, Geneva.———, 1986: Intercomparison of Models of SnowmeltRunoff, WMO-No. 646, Geneva, Switzerland.———, 1990: Hydrological Models for Water ResourcesSystem Design and Operation. Operational HydrologyReport No. 34, WMO-No. 740, Geneva.———, 1992: Simulated Real-time Intercomparison ofHydrological Models. Operational Hydrology ReportNo. 38, WMO-No. 779, Geneva.———, 1994: Guide to Hydrologic Practices, Fifth Edition,WMO-No. 168. Geneva, pp. 721–724.——— and the United Nations Educational, Scientificand Cultural Organization, 1997: The World’s Water,Is there enough?.Young, P.C., 1993: Concise Encyclopaedia of EnvironmentalSystems, Pergamon Press, pp. 22.

www.wmo.int