Download the entire proceedings as an Adobe PDF - Eastern Snow ...

Proceedings of the 

63rd ANNUAL 

EASTERN SNOW CONFERENCE 

7–9 June 2006 

Newark, Delaware, USA

ISBN 0-920081-28-2 

ISSN: 0424-1932 

Proceedings of the Eastern Snow Conference 

Printed and Bound in the United States of America 

ii

FOREWORD 

T 

his proceedings volume contains papers presented at the 63rd Eastern Snow Conference (ESC) held 

7–9 June 2006 at the University of Delaware in Newark, Delaware, USA. Sessions covering snow– 

climate interactions, snow processes, ground ice and periglacial processes, and remote sensing of snow 

were convened. 

The ESC is a joint United States and Canadian forum for discussing recent work on operational, applied, and 

scientific issues related to snow and ice. It also retains an increasing interest as a symposium where novel 

approaches to cryospheric science of international significance are presented. The ESC has published an annual 

series of proceedings since 1952. Typical topics include studies of snow and ice as materials, snow removal, 

meteorological forecasting, river ice control, snow hydrology, snow chemistry, glaciology, remote sensing of snow 

and ice, and snow ecology. Membership in the ESC is open to all interested individuals and corporations. Additional 

copies of the current proceedings and all back issues can be obtained from the Secretary. More information about the 

Eastern Snow Conference may be found at http://www.easternsnow.org/. 

We continue with an optional review process and authors may submit extended abstracts or full papers to the 

proceedings. Participants in the Western Snow Conference (WSC) (http://www.westernsnowconference.org/) have 

been invited to join the ESC participants in the option of submitting papers relevant to winter hydrology to the 

international journal Hydrological Processes. These papers have gone through a formal journal review, revision, and 

referee process, and many appear both in these proceedings and in the December 2006 issue of the journal. John 

Pomeroy (University of Saskatchewan), Kelly Elder (USDA Forest Service, Fort Collins, Colorado), and Andrew 

Klein (Texas A&M University) edited this special issue of Hydrological Processes dedicated to ESC/WSC papers. 

We thank the editors at ERDC–CRREL. Over the years this group has contributed considerable time to this 

publication and has enhanced the quality of these proceedings. 

The 2006 meeting of the Eastern Snow Conference and these proceedings were made possible by sponsorship 

and corporate memberships by the following: 

Campbell Scientific (Canada) Corporation ERDC–CRREL 

Edmonton, Alberta, Canada Hanover, New Hampshire, USA 

http://www.campbellsci.com/offices/csc.html http://www.crrel.usace.army.mil/ 

University of Delaware GEONOR 

Newark, Delaware, USA Milford, Pennsylvania, USA 

http://www.udel.edu http://www.geonor.com 

Cryospheric Specialty Group of the Association of American Geographers (AAG) 

Washington, DC, USA 

http://www.geo.hunter.cuny.edu/AAG_CrSG 

We look forward to seeing many of you at the 64th Eastern Snow Conference meeting to be held jointly with 

the Canadian Geophysical Union, Canadian Meteorological Society, and the American Meteorological Society, 

28 May–1 June 2007 at Memorial University, St. John’s Newfoundland, Canada. The 2007 meeting theme is “Air, 

Ocean, Earth, and Ice on the Rock.” It should be exciting, especially with the addition of the other groups. 

Robert Hellström and Susan Frankenstein 

ESC Proceedings Co-Editors 

Bridgewater State College and ERDC–CRREL 

iii

This page is intentionally blank. 

iv

CONTENTS 

Foreword.......................................................................................................................................................................iii 

Statement of Purpose...................................................................................................................................................vii 

Executives for the 63rd Eastern Snow Conference.......................................................................................................ix 

President’s Page............................................................................................................................................................xi 

Weisnet Medal for Best Student Paper 

The Influence of Snow–Soil Moisture Flux on Snowpack Metamorphism in Late Winter and Early Spring 

Y.C. Chung and A.W. England ...............................................................................................................................3 

Campbell Scientific Award for Best Canadian Student Paper 

Potential of a Water Balance Model with High Temporal Resolution for the Distributed Modelling of Ice- and 

Snowmelt Processes at High Elevated Sites 

G. Koboltschnig, H. Holzmann, W. Schoener, and M. Zappa ..............................................................................19 

Snow and Climate 

20th Century North American Snow Extent Trends: Climate Change or Natural Climate Variability? 

A. Frei, G. Gong, D.A. Robinson, G. Choi, D. Ghatak, and Y. Ge.......................................................................39 

Snow and Climate Posters 

Observed Differences Between Snow Extent and Snow Variability at Continental Scales 

Y. Ge and G. Gong ...............................................................................................................................................45 

Climate Variability, Snowmelt Distribution, and Effects on Streamflow in a Cascades Watershed 

A. Jefferson, A. Nolin, S. Lewis, M. Payne, G. Grant, and C. Tague ...................................................................51 

Synoptic Patterns Associated with the Record Snowfall of 1960 in the Southern Appalachians 

L.B. Perry and C.E. Konrad II .............................................................................................................................55 

Influence of Snowfall Anomalies on Summer Precipitation in the Northern Great Plains of North America 

S.M. Quiring and D.B. Kluver..............................................................................................................................65 

Snow Remote Sensing 

Estimating Sublimation of Intercepted and Sub-Canopy Snow Using Eddy Covariance Systems 

N. Molotch, P. Blanken, M. Williams, A. Turnipseed, R. Monson, and S. Margulis ............................................75 

The Retrievals of Snow Cover Extent and Snow Water Equivalent from a Blended Passive Microwave—Interactive 

Multi-Sensor Snow Product 

C. Kongoli, C.A. Dean, S.R. Helfrich, and R.R. Ferraro......................................................................................89 

Time Series Analysis and Algorithm Development for Estimating SWE in Great Lakes Area Using Microwave Data 

A.E. Azar, H. Ghedira, P. Romanov, S. Mahani, and R. Khanbilvardi ..............................................................105 

On the Evaluation of Snow Water Equivalent Estimates Over the Terrestrial Arctic Drainage Basin 

M.A. Rawlins, M. Fahnestock, S. Frolking, and C.J. Vörösmarty......................................................................121 

v

Snow Remote Sensing Posters 

AMSR-E Algorithm for Snowmelt Onset Detection in Subarctic Heterogeneous Terrain 

J.D. Apgar, J.M. Ramage, R.A. McKenney, and P. Maltais...............................................................................137 

Combination of Active and Passive Microwave to Estimate Snowpack Properties in Great Lakes Area 

A.E. Azar, T. Lakhankar, N. Shahroudi, and R. Khanbilvardi ...........................................................................153 

Enhancements and Forthcoming Developments to the Interactive Multisensor Snow and Ice Mapping System (IMS) 

S.R. Helfrich, D. McNamara, B.H. Ramsay, T. Baldwin, and T. Kasheta..........................................................165 

Retreat of Tropical Glaciers in Colombia and Venezuela from 1984 to 2004 as Measured from ASTER and Landsat 

Images 

J.N. Morris, A.J. Poole, and A.G. Klein .............................................................................................................181 

Snowpack Processes 

Snow Cover Patterns and Evolution at Basin Scale: GEOtop Model Simulations and Remote Sensing Observations 

S. Endrizzi, G. Bertoldi, M. Neteler, and R. Rigon.............................................................................................195 

Microstructural Characterization of Firn 

I. Baker, R. Obbard, D. Iliescu, and D. Meese...................................................................................................211 

Computational Time Steps of Winter Water Balance for Snow Losses at United States Meteorological Stations 

S.R. Fassnacht ....................................................................................................................................................219 

Shaped Solution Domains for Snow Properties 

R.A. Melloh, S.A. Shoop, and B.A. Coutermarsh................................................................................................231 

Glacial and Periglacial Processes 

Quantifying the Effect of Anisotropic Properties in Snow for Modelling Meltwater Retention 

C.E. Bøggild .......................................................................................................................................................247 

The Equilibrium Flow and Mass Balance of the Taku Glacier, Alaska, 1950–2005 

M.S. Pelto, G.W. Adema, M.J. Beedle, S.R. McGee, M.M. Miller, K.F. Sprenke, and M. Lang.........................251 

An Embedded Sensor Network for Measuring Hydrometeorological Variability Within an Alpine Valley 

R.Å. Hellström and B.G. Mark ...........................................................................................................................263 

Snow and Periglacial Processes Poster 

A Treatise on the Preponderance of Designs Over Historic and Measured Snowfalls, or No Two Snowflakes Are 

Alike: Considerations About the Formation of Snowflakes and the Possible Numbers and Shapes of Snowflakes 

M. Pilipski and J.D. Pilipski...............................................................................................................................283 

62nd ESC Papers 

Simulations of North American Snow Cover by AGCMs and AOGCMs 

A. Frei, G. Gong, and R. Brown.........................................................................................................................293 

The Impact of Patchy Snow Cover on Snow Water Equivalent Estimates Derived from Passive Microwave 

Brightness Temperatures Over a Prairie Environment 

K.R. Turchenek, J.M. Piwowari, and C. Derksen...............................................................................................297 

� � � 

Sno-Foo Award .........................................................................................................................................................311 

vi

STATEMENT OF PURPOSE 

The Eastern Snow Conference (ESC) is a joint Canadian/U.S. organization founded in the 1940s, originally 

with members primarily from eastern North America. Our current members are scientists, snow surveyors, 

engineers, technicians, professors, students, and operational and maintenance professionals from North America, the 

United Kingdom, Japan, and Germany. There is a western counterpart to the ESC, the Western Snow Conference 

(WSC), which also is a joint Canadian/U.S. organization. Every fifth year the ESC and the WSC hold joint 

meetings. 

The Eastern Snow Conference is a forum that brings the research and operations communities together to 

discuss recent work on scientific, applied, and operational issues related to snow and ice. The location of the 

conference alternates yearly between the United States and Canada, and attendees present their work by either 

giving a talk or presenting a poster. Most resulting papers are reviewed, edited, and published in our yearly 

Proceedings of the Eastern Snow Conference. In recent years, the ESC meetings have included sessions on snow 

physics, winter survival of animals, snow and ice loads on structures, river ice, remote sensing of snow and ice, and 

glacier processes. Volumes of the Proceedings can be found in libraries throughout North America and Europe, and 

the papers are also available through the National Technical Information Service (NTIS) in the United States and 

CISTI in Canada. 

� � � � � � � � � 

Le Colloque sur la neige-région est (ESC) est une organisation americain-canadienne fondée dans les années 

’40 et dont les membres provenaient a l’origine surtout de l’est de l’Amérique-du-Nord. Actuellement, les membres, 

qu’ils soient chercheurs, techniciens en enneigement, ingenieurs, techniciens, professeurs, étudiants, et spécialistes 

des services d’éxploitation et d’entretien, viennent non seulement d’Amérique-du-Nord, mais aussi du Royaume 

Uni, du Japon, et d’Allemagne. Le Colloque sur la neige-region ouest (WSC), aussi une organisation americaincanadienne, 

est l’homologue de l’ESC pour l’ouest nord-americain. Tous les cinq ans, l’ESC et la WSC organisent 

des réunions en commun. 

Le Colloque sur la neige-region est un forum qui rassemble chercheurs et responsables des services 

d’exploitation pour discuter des travaux récents sur les problemes scientifiques, operationnels, ou autres dus à la 

neige et à la glace. Le site de cette réunion annuelle alterne entre les États Unis et le Canada. Les participants y 

présentent les résultats de leurs travaux par des communications orales ou au moyen d’affiches. Ces 

communications, une fois revues et éditées, sont publiées dans les Annales de l’ESC. Dans les années récentes, les 

réunions de l’ESC ont inclus des sessions sur la physique de la neige, la survie hivernale de la faune, les forces 

exercées par la neige et la glace sur les structures et les batiments, la glace de rivière, la télédétection de la neige et 

de la glace, et les processus glaciaires. Les Annales de l’ESC sont accessibles dans la plupart des bibliothèques 

scientifiques d’Amerique-du-Nord et d’Europe. Des copies d’articles peuvent être obtenues du National Technical 

Information Service (NTIS) aux États Unis et son équivalent au Canada, le CISTI. 

vii


viii

EXECUTIVES FOR THE 63rd EASTERN SNOW CONFERENCE 

PRESIDENT PAST PRESIDENT 

Claude Duguay Susan Taylor 

University of Waterloo ERDC-CRREL 

Waterloo, Ontario, Canada Hanover, New Hampshire, USA 

VICE-PRESIDENT AND PROGRAM CHAIR 

Andrew Klein 

Texas A&M University 

College Station, Texas, USA 

SECRETARY–TREASURER (Canada) SECRETARY–TREASURER (United States) 

Miles Ecclestone Derrill Cowing 

Trent University U.S. Geological Service (Retired) 

Peterborough, Ontario, Canada Monmouth, Maine, USA 

ESC PROCEEDINGS EDITORS 

Robert Hellström, Bridgewater State College, Bridgewater, Massachusetts, USA 

Susan Frankenstein, ERDC-CRREL, Hanover, New Hampshire, USA 

HYDROLOGICAL PROCESSES SPECIAL ISSUE EDITORS 

John Pomeroy 

University of Saskatchewan 

Saskatoon, Saskatchewan, Canada 

Andrew Klein 



STEERING COMMITTEE 

Chris Derksen (Chair) 

Meteorological Service of Canada 

Downsville, Ontario, Canada 

Janet Hardy 

ERDC-CRREL 

Hanover, New Hampshire, USA 

Steven Fassnacht 

Colorado State University 

Fort Collins, Colorado, USA 

Mauri Pelto 

Nichols College 

Dudley, Massachusetts, USA 

RESEARCH COMMITTEE 

Steven Fassnacht (Chair) 

Colorado State University 

Fort Collins, Colorado, USA 

Steven Déry 

Princeton University 

Princeton, New Jersey, USA 

ix

RESEARCH COMMITTEE (continued from page ix) 

Allan Frei 

Hunter College of the CUNY 

New York, New York, USA 

Richard Kelly 

Maryland, USA 

Natasha Neumann 

Meteorological Service of Canada 

Saskatoon, Saskatchewan, Canada 

WEB MASTER 

Andrew Klein 



LOCAL ARRANGEMENTS 

Delphis Levia (Chair) 

University of Delaware 

Newark, Delaware, USA 

Andrew Klein 



Miles Ecclestone 

Trent University 

Peterborough, Ontario, Canada 

LIFE MEMBERS 

Art Eschner 

Austin Hogan 

H. Gerry Jones 

John Metcalfe 

Hilda Snelling 

Don Wiesnet 

x

THE PRESIDENT’S PAGE 

The 63rd annual meeting of the Eastern Snow Conference (ESC) was held at the University of Delaware in 

Newark, Delaware (7–9 June 2006). There were sessions on snow and climate, remote sensing of snow processes, 

and glacial and periglacial processes. The quality of the presentations was exceptional and I thank all who shared 

their data at the meeting. 

The success of the 2006 meeting rests with the Executive. I thank Delphis Levia, who, along with Andrew 

Klein and Miles Ecclestone, made all the local arrangements; our treasurers, Derrill Cowing and Miles Ecclestone, 

for keeping us in the black; and Claude Duguay (VP and Program Chair) for organizing and running the 63rd 

meeting. Andrew Klein, our Webmaster, has helped the ESC join the digital age. I also thank our Corporate 

members (listed in the Foreword) for their continued support of the ESC. 

The winner of this year’s Weisnet Medal for the best student paper was Y.C. Chung for his work on the 

influence of the snow–soil moisture flux on snowpack metamorphism. The Campbell Scientific Medal, which is 

awarded to the best student paper in which instrumentation plays a key role, was awarded to Gernot Koboltschnig 

for his investigation into the potential of using a water balance model with high temporal resolution for distributed 

modeling of melting at high elevation. 

The ESC Proceedings have been published for 63 years. I find this a remarkable feat for a small organization 

run by a handful of researchers. Editing the proceedings is one of the most time-consuming jobs and I thank our 

Editors, Dr. Robert Hellström and Susan Frankenstein, for their hard work. This is also the eleventh anniversary of 

ESC’s partnership with Hydrological Processes, in which a number of the papers presented at the conference are 

published. I thank Hydrological Processes for providing ESC participants the opportunity to publish in the journal, 

and Dr. John Pomeroy and Dr. Andrew Klein for handling the Hydrological Processes submissions. 

I invite you to attend the 64th meeting of the Eastern Snow Conference, to be held in conjunction with the 

Canadian Meteorological and Oceanographic Society, the Canadian Geophysical Society, and the American 

Meteorological Society at the end of May 2007 in St. John’s, Newfoundland. Please visit our Web site 

(www.easternsnow.org) for up-to-date details of this meeting. 

xi 

Best Regards, 

Claude Duguay 

63rd President 

Eastern Snow Conference


xii

Weisnet Medal 

for Best Student Paper 

1


2

3 

63 rd EASTERN SNOW CONFERENCE 

Newark, Delaware USA 2006 

The Influence of Snow–Soil Moisture Flux 

on Snowpack Metamorphism in Late Winter and Early Spring 

ABSTRACT 

Y. C. CHUNG 1 AND A. W. ENGLAND 2 

The importance of land–atmosphere snow cover feedback in the climate system, water storage 

and release from snowpacks, and the large land areas covered by snow in the northern hemisphere, 

are reasons to accurately model late winter and early spring snow metamorphic processes. 

Snowpack models like SNTHERM predict snow behavior very well during the cold periods but do 

not adequately capture liquid and vapor transport between soil and snow conditions during the 

snowmelt period. To get a more realistic description of the dynamics between snowpack and soil, 

we developed a snow–soil–vegetation–atmosphere transfer (ssvat) model, which combines soil 

processes from our land surface processes (lsp) with the snowpack processes of SNTHERM. 

We validated the model with data from the NASA cold land processes field experiment (CLPX) 

field experiment conducted in late winter and early spring near Fraser, Colorado, in 2003. We 

show an example where soil/snow fluxes increase the energy stored within the snowpack by as 

much as 17%. Thermal conduction was the predominant soil/snow energy flux. Other energy 

fluxes, those associated with air convection and vapor diffusion, were 10 –7 smaller but do 

contribute to the formation of depth hoar and to grain size growth. Grain sizes in the upper 

snowpack are increased by soil/snow fluxes. Because radiobrightness scatter darkening at 

frequencies above 19 GHz increases with grain size, static snow water equivalent (swe) algorithms 

that use scatter darkening to estimate swe become less reliable when the soil/snow vapor fluxes 

are large as they would be when the soil is wet and unfrozen. Comparisons of SNTHERM and 

SSVAT with the CLPX data show that SSVAT provided better estimates of soil temperature by 

1.4 K in the late winter and of snow temperature by 3.2 K in early spring. SSVAT also provides 

better moisture profiles in both snow and soil. 

The combination of SSVAT and a new radiobrightness module will enable monitoring of water 

stored in snow over frozen or unfrozen soil during periods when the snowpack is experiencing 

thawing and refreezing. It will also contribute to the design of the NASA cold lands processes 

pathfinder (CLPP) field experiment planned for the arctic and help prepare the cold lands 

community to interpret the 1.4 GHz brightness observations from SMOS. 

Keywords: snow processes; soil processes; vapor diffusion; water flow; free convection; depth 

hoar 

INTRODUCTION 

Snowmelt runoff in spring is a major source of hydroelectric and irrigation water at middle and 

high latitudes. Snow and ice also significantly increase the Earth's albedo, promoting cooling. 

1 

Geosciences and Remote Sensing Graduate Program, University of Michigan, Ann Arbor, 

Michigan 48109-2143, USA. 

2 

College of Engineering, University of Michigan, Ann Arbor, Michigan 48109-2102, USA.

Even small changes in climate influence these snow processes. The removal of the snow cover 

also initiates the melting of river and lake ice, the thawing of the active layer, and marks the 

beginning of the evaporation season. Meltwater increases soil moisture and initiates or increases 

streamflow. The decaying snowpack causes a major change in the surface energy balance of arctic 

regions. 

A comprehensive search of the literature by the Snow Models Intercomparison Project 

(SNOWMIP) in 2001 identified more than 40 snow models motivated either by snow hazard 

investigation or as the lower boundary of an atmospheric model (Yang, 2004). Vapor transfer 

processes are treated in only 10% of the snow models. Most snow models are validated with field 

data but 10% are validated with remote sensing data. Effects of sub-grid-scale topography on 

distribution of precipitation, air temperature and snow depth are considered in 25% of the snow 

models (Jordan, 1991; Yang, 2004). 

SNTHERM was chosen as the basis for our snow cover-soil model because it offers the high 

physical fidelity (Jordan, 1991; Jordan, 1999). SNTHERM considers retention/percolation, 

refreezing, vapor transfer, and heat transport by rainfall or snowfall for the snowpack. Snow 

parameters, such as density and heat capacity/conductivity evolve with the snowpack. SNTHERM 

predicts the temperature profiles within the frozen soil and the stratified snow with unlimited 

layers. SNTHERM has been verified with field data for several years from sites in Grayling, 

Michigan, and Hanover, New Hampshire. SNTHERM’s weaknesses are that it ignores the liquid 

water and vapor transport in the soil and allows water to be artificially drained at the snow/soil 

interface. 

Although soil processes have been combined in some models, these models were simplified in 

other ways that compromise overall performance. Among these models are SHAW, 

SNOWPACK, and ISBA-ES (Flerchinger and Hanson. 1989; Lehning et al., 1998; Boone and 

Etchevers, 2001). ISBA-ES and SNOWPACK only consider the soil processes on the surface or 

subsurface. In ISBA-ES, the subsurface soil column is divided into two subsurface soil moisture 

reservoirs consisting of a root-zone layer and a subroot-zone layer (Boone and Etchevers, 2001). 

The purpose of the ISBA scheme is to calculate the surface radiative heat, momentum, and 

moisture exchanges with the atmosphere and the components of the near-surface hydrological 

budget. SHAW considers similar soil processes to those of our LSP model but it has less detailed 

snow processes than SNTHERM. Neither precipitation nor litter fall are allowed and the 

interaction between snow grain and vapor diffusion are ignored. SHAW was developed for a 

continuous canopy cover. 

Our Prairie Land Surface Processes (LSP) model is a one-dimensional, high physical fidelity 

model with coupled heat and moisture transport for prairie soils (Liou, 1996; Judge, 1999). Figure 

1 illustrates the processes in the Prairie LSP model, including interactions between the 

atmosphere, the canopy and the soil, and the effect of transpiration on moisture and energy fluxes 

within the root zone. To get a more realistic description of the snowpack–soil dynamics, we 

developed the Snow–Soil–Vegetation–Atmosphere Transfer (SSVAT) model, which combines 

soil processes from our Land Surface Processes (LSP) model with the snowpack processes of 

SNTHERM (Chung and England, 2005; Chung et al., 2006). 

MODEL MODIFICATIONS 

The coupled SSVAT model employs an index of layers in snow and soil in descending order 

from the top down as seen in Fig 1. 

4

Figure 1. Layer 1 represents the soil base; nsoil represents the soil layer under the snow/soil interface; and n 

represents the top snow layer in the model structure (modified from Liou, 1996). 

Water Flow 

SNTHERM includes only gravity driven liquid water flow but liquid water is driven by both 

capillarity and gravity (Tao and Gray, 1994). Ignoring the capillary effect limits water flow to the 

downward direction. Particularly, the capillary effect should not be neglected under non-steady 

state conditions (Su et al, 1999; Waldner et al, 2004). Capillary forces can be neglected only for a 

stationary unsaturated water flow downwards. Water flow processes in SNTHERM have been 

modified in SSVAT to incorporate the capillary effect. 

Heat convection due to liquid water flow is described by Eqn (1) in SNTHERM where cw is 

specific heat of water per unit volume, Uw is the downward flow rate of liquid water, T i is the 

temperature of the i th layer, the superscripts (I ±1/2) refer to the interfaces at the top and bottom of 

i+ 

1 

layer i. The heat flux due to water flow across upper boundary into layer i is H and the heat 

w 

i 

flux across lower boundary out of layer i is H : w 

1 

i+ 

1 i+ 

1 i+ 

2 i+ 

1 

H w = cw 

U w T 

(1) 

1 

i i i− 

2 i 

H = c U T 

w 

w 

w 

Including water flow due to capillary requires consideration of the direction of water flow. For 

example, Eqn (2) describes flux out of the upper boundary of layer i if the water flow is upward 

1 

i+ 

2 

due to capillary forces ( U w 0). 

1 

1 

i+ 

1 i i+ 

2 i i+ 

2 

H w = cwU 

w T , if ( U w < 0) 

(2) 

H 

i+ 

1 

w 

= c 

i+ 

1 

w 

U 

1 

i+ 

2 

w 

T 

i+ 

1 

, if 

1 

i+ 

2 ( U w > 0) 

T i+1 

T i 

Figure 2. Red line shows the heat transport across upper boundary of the layer i. 

Depth Hoar 

Depth hoar, which is unique to SSVAT, can change the insulating effect of the seasonal snow 

cover significantly during ablation periods because of the lower density and thermal conductivity 

5 

snow layer i+1 

snow layer i 

(3)

of depth hoar. The growth of depth hoar is favored when the difference between soil and air 

temperatures is greater and when vegetation buried within the snow results in a high porosity of 

snow cover near the snow–soil interface (Zhang et al., 1997). Natural convection of air in snow 

can occur to form depth hoar if a critical temperature gradient is exceeded. The convective flux 

can be determined from the relationship between the thickness of the snow cover and the 

temperature gradient (Fig 44 of Akitaya, 1974). 

Snow/Soil Interface 

Mass and energy transfer at the snow/soil interface in SSVAT are based on the algorithms from 

LSP (Liou, 1996). The infiltration model for soil can be estimated using a quasi-analytic solution 

to Richard’s equation for vertical infiltration in a homogeneous soil (Liou, 1996; Judge, 1999). To 

precisely capture mass and energy transfer at the soil/snow interface, snow and soil layers at the 

soil/snow interface should be thin. 

Figure 3. Simulations of temperature and liquid water content for a thatch layer(modified from Chung et al., 

2006) 

Heat transfer at the snow/soil interface can be affected by vegetation. A thatch layer in LSP was 

defined as a 2cm layer of organic matter at the base of the grass canopy that is subject to the heat 

exchange with the atmosphere, the canopy and the underlying soil (Liou, 1996). To see if this 

thatch layer buried within the snow is important for heat exchanges in SSVAT, a simulation was 

run for two weeks using a forcing data from Feburary 23, 2003 in CLPX (Chung et al., 2006). Our 

results showed that a thatch layer did not affect moisture and temperature profiles of either snow 

or soil when the thatch fraction varied (Fig 3). SSVAT thus ignores thatch at the snow/soil 

interface. 

RESULTS 

Study Area 

Fig 4 shows the study area at the Local Scale Observation Site (LSOS) of the NASA Cold Land 

Processes Field Experiment (CLPX). SSVAT requires temperature, moisture and grain size 

profiles for initialization, which were taken from the Micrometeorological Data and Snow 

Measurements (Cline et al., 2002; England, 2003). SSVAT also requires precipitation data, which 

were taken from Ground Based Passive Microwave Radiometer Data (Graf et al, 2003). Data for 

model validation came from snow pit measurements of density and grain size profiles and subcanopy 

meteorological observations for late winter and early spring (Cline et al., 2002; Hardy et 

al., 2002). 

6

Figure 4. The general layout of the LSOS during IOP3 and IOP4 (England, 2003). 

The soil type in the study area is sandy loam, whose dry substance consists of 52% quartz, 7% 

clay, and residual for silt (Cline et al., 2001). Trees in the study area are of mixed species 

(predominantly lodgepole pine with some Englemann Spruce and Subalpine Fir) with an average 

tree height of 7.8 m (standard deviation = 4.8 m; n = 88) and heterogeneous spacing between trees 

(Hardy et al., 2002). 

Energy Stored within the Snowpack 

SSVAT and SNTHERM were forced by downwelling radiance and observed meteorology for 

the periods of IOP3 (2/19~2/24) and IOP4 (3/25~3/29). The results were compared with 

observations. Figure 5 shows the model-derived heat contents stored in the snowpack over the 

study period. The simulations show an average increase for SSVAT relative to SNTHERM of 0.54 

W/m 2 for late winter and 1.87 W/m 2 for early spring. That is, SSVAT stored heat contents were 

increased by soil/snow fluxes. Exceptions occurred on some afternoons (e.g. DOY 86~87 and 

DOY 51~52) when the snowpack became warmer in the late afternoon transporting heat to its 

underlying frozen soil. In general, SSVAT predicted a warmer snowpack. Soil processes increased 

energy stored in the snowpack by as much as 17%. 

Figure 5. Heat content stored in the snowpack predicted by two models and the model differences over the 

study period. 

7

Thermal Conduction 

Heat fluxes at the snow/soil interface include thermal conduction, vapor diffusion, and 

convection from free air and flow water (Chung et al., 2006). Figure 6 shows the thermal 

conduction fluxes, with an average difference between two models of 1.43 W/m 2 for late winter 

(IOP3) and 0.06 W/m 2 for early spring (IOP4). The daily oscillation was less dramatic in early 

spring than in late winter since the temperature differences at the interface was less significant in 

early spring (soil and snow both thawed). This suggests that soil conduction is predominant in the 

soil heat contribution to the snowpack. 

Figure 6. Simulated soil heat fluxes at the snow/soil interface over the study period. 

Diffusion and Convection 

SNTHERM only considers thermal conduction from the soil. Vapor diffusion from the soil and 

natural convection of free air in snow are unique to SSVAT. On average, vapor diffusion and 

natural convection were 10 –7 times smaller than thermal conduction at the interface (Fig 7). Vapor 

diffusion displayed a diurnal cycle in late winter because it was affected by the diurnal 

temperature cycle. Natural convection of free air in snow was significant in early spring, 

especially on DOY 87–88. Comparing Fig 5 with Fig 7, air convection may be responsible for the 

increasing variability of the heat content in snow on DOY 87–89. This suggests the importance of 

convection in snow may exceed that of vapor diffusion or the thermal conduction. 

Figure 7. Simulated heat fluxes at the snow/soil interface over the study period. 

Air Convection 

Fig 8 shows the occurrence of air convection in a snowpack in late winter (IOP3) and early 

spring (IOP4). Red color and black color represented the occurrence of the air convection whereas 

8

lue color represented no convection. Air convection was predicted to occur more frequently in 

the upper snowpack, especially in early spring. This may reflect on the snow and soil 

characteristics, such as temperature, moisture and grain size. 

Figure 8. Air convection simulated over the study period. 

Temperature Profiles 

Figure 9 compares the temperature profiles predicted by the two models with observations. The 

maximum difference between SSVAT and observation of 0.3 K was observed in soil temperature 

in late winter scenario, better then that of SNTHERM (1.7 K). The maximum difference of 2.8 K 

was observed by SSVAT in snow temperature, better than 6 K by SNTHERM in early spring 

scenario. As expected, the bigger improvement was displayed in early spring once the snowpack 

started melting. It also reflected on the simulated flux in the figures above. The temperature in the 

upper snowpack can be affected by the soil processes even when the soil and lower snowpack 

temperature predictions by two models are similar (Fig 7 and 8). This suggests that SSVAT better 

estimates temperature during the variable climate of late winter and early spring, and that the soil 

processes cannot be ignored for predicting the evolution of the snowpack. 

9

Figure 9(a). Simulated temperature profiles in late winter. 

Figure 9(b). Simulated temperature profiles in early spring. 

10

Total Density of Water Profiles 

The total density profile includes both liquid and solid (ice) phases. Regression coefficients of 

the density predicted by two models with observations in the soil can be seen in Table 1. It shows 

that SSVAT predicted soil moisture well whereas SNTHERM were poor for soil moisture profiles. 

SNTHERM artificially draining the water at the snow/soil interface so cannot accurately predict 

the soil moisture in late winter. 

Table 1. Regression coefficients of the model predictions and observations. 

SSVAT SNTHERM 

1.5 cm* 0.99 0.69 

4.5 cm* 0.89 0.68 

10 cm* 0.96 ** 

27 cm* 0.98 ** 

45 cm* 0.91 ** 

*means the depth beneath the ground (snow/soil interface). 

** SNTHERM failed to capture the variation of the water density. 

Fig 10 represents that simulated density of water profiles in late winter and early spring. It also 

shows that SSVAT predicted density profiles well. The differences between two model predictions 

were not significant in early spring since the model initialization in early spring was more realistic 

than that in late winter. Others may consider liquid water fractions to be significantly different. 

However, we do not have observations of liquid water content to compare. This suggests that the 

SSVAT provides a realistic representation of moisture profiles in snow and soil even under an less 

realistic initialization. 

Figure 10(a). Simulated density of water profiles in late winter. 

11

Figure 10(b). Simulated density of water profiles in early spring. 

Snow Grain Size Profiles 

Comparisons show that snow grain size was predicted larger by SSVAT in response to vapor 

diffusion contribution from the soil that can grow the grain size more (Fig 11). This contribution 

accumulates when transporting through the snowpack, making the model difference more obvious 

near the snow surface. These differences might significantly affect the radiobrightness. 

12

Figure 11(a). Simulated snow grain size profiles in late winter 

Figure 11(b). Simulated snow grain size profiles in early spring. 

13

CONCLUSIONS 

The SSVAT model was validated with data from the CLPX field experiment conducted in late 

winter and early spring near Fraser, Colorado, in 2003. We show an example where soil/snow 

fluxes increase the energy stored within the snowpack by as much as 17%. Thermal conduction 

was the predominant soil/snow energy flux. Other energy fluxes, those associated with air 

convection and vapor diffusion, were 10 –7 smaller but do contribute to the formation of depth hoar 

and to grain size growth. Grain sizes in the upper snowpack are increased by soil/snow fluxes. 

Because radiobrightness scatter darkening at frequencies above 19 GHz increases with grain size, 

static Snow Water Equivalent (SWE) algorithms that use scatter darkening to estimate SWE 

become less reliable when the soil/snow vapor fluxes are large as they would be when the soil is 

wet and unfrozen. 

The maximum difference between soil temperatures of SSVAT and observations was 0.3 K in 

late winter. The maximum difference between soil temperatures for SNTHERM in that period was 

1.7 K. The maximum difference between snow temperatures of SSVAT and observations was 2.8 K 

in early spring. The maximum difference between snow temperatures of SNTHERM and 

observations in that period was 6 K. That is, comparisons of SNTHERM and SSVAT with the 

CLPX data show that SSVAT provided better estimates of soil temperature by 1.4K in late winter 

and of snow temperature by 3.2 K in early spring. SSVAT also provides better moisture profiles in 

both snow and soil. 

The combination of SSVAT and a new radiobrightness module will enable monitoring of water 

stored in snow over frozen or unfrozen soil during periods when the snowpack is experiencing 

thawing and refreezing. It will also contribute to the design of the NASA Cold Lands Processes 

Pathfinder (CLPP) field experiment planned for the Arctic and help prepare the cold lands 

community to interpret the 1.4 GHz brightness observations from SMOS. 

REFERENCES 

Akitaya, E, 1974. Studies on depth hoar. Contributions from the Institute of Low Temperature 

Science, Series A, 26:1–67. 

Boone A, Etchevers P, 2001. An intercomparison of three snow schemes of varying complexity 

coupled to the same land surface model: Local scale evaluation at an alpine site. J. 

Hydrometeor., 2: 374–394. 

Chung YC, England, AW, 2005. A coupled soil–snow–atmosphere transfer model. Proc. of the 

IEEE International Geoscience and Remote Sensing Symposium, IGARSS'05, Seoul. ` 

Chung YC, England AW, De Roo RD, Weininger E, 2006(submitted). Effects of vegetation and of 

heat and vapor fluxes from soil on snowpack evolution and radiobrightness. Proc. of the IEEE 

International Geoscience and Remote Sensing Symposium, IGARSS'06, Denver, CO. 

Cline D (Chair, Cold Land Processes Working Group), Armstrong R, Davis R, Elder K, Liston G., 

2001. NASA Cold Land Processes Field Experiment Plan Home 

Page. 

Cline D, Armstrong R, Davis R, Elder K, and Liston G., 2002, Updated July 2004. CLPX-Ground: 

ISA Snow Pit Measurements. Edited by M. Parsons and M.J. Brodzik., Boulder, CO: National 

Snow and Ice Data Center. Digital Media, 

England AW, 2003. CLPX-Ground: Micrometeorological Data at the Local Scale Observation 

Site (LSOS). Boulder, CO: National Snow and Ice Data Center. Digital Media. 

Graf T, Koike T, Fujii H, Brodzik M, Armstrong R, 2003. CLPX-Ground: Ground Based Passive 

Microwave Radiometer (GBMR-7) Data. Boulder, CO: National Snow and Ice Data Center. 

Digital Media. 

Flerchinger GN, Hanson CL, 1989. Modeling soil freezing and thawing on a rangeland watershed. 

Trans. Amer. Soc. of Agric. Engr., 32(5): 1551–1554. 

Hardy J, Melloh R, Koenig G, Pomeroy J, Rowlands A, Cline D, Elder K, Davis R, 2002, updated 

2004. Sub-canopy energetics at the CLPX Local Scale Observation Site (LSOS). Boulder, CO: 

National Snow and Ice Data Center. Digital Media. 

14

Jordan R, 1991. A one-dimensional temperature model for a snow cover. Technical documentation 

for SNTHERM.89, Special Technical Report 91-16, US Army CRREL. 

Jordan R, Andreas E, 1999. Heat budget of snow-covered sea ice at North Pole 4. J. Geophys. 

Res., 104(C4): 7785–7806. 

Judge J, 1999. Land surface process and radiobrightness modeling of the Great Plains, Ph.D. 

thesis, University of Michigan. 

Lehning et al., 1998. A network of automatic weather and snow stations and supplementary model 

calculations providing SNOWPACK information for avalanche warning, ISSW 98 International 

Snow Science Workshop, Sunriver, Oregon. 

Liou, YA, 1996. Land surface process/ radiobrightness models for northern prairie, Ph.D. thesis, 

University of Michigan. 

Su, GW, Geller, JT, Pruess, K, Wen, F, 1999. Experimental studies of water seepage and 

intermittent flow in unsaturated, rough-walled fractures,” Water Resources Research, 35: 

1019–1037. 

Tao, Y.-X. and Gray, D.M.. 1994. Prediction of Snow-Melt Infiltration into Frozen Soils. Numer. 

Heat Transfer Part A, 26: 643–645. 

Waldner, PA, Schneebeli, M, Zimmermann, US, Fluhler, H, 2004. Effect of snow structure on 

water flow and solute transport. Hydrological Processes, 18: 1271–1290. 

Yang, ZL, 2004(in press). Description of recent snow models. Book Chapter, in Snow and 

Climate, E. Martin and R. Armstrong (editors), International Committee on Snow and Ice. 

Zhang, T, Osterkamp, TE, Stamnes K, 1997. Effects of Climate on the Active Layer and 

Permafrost on the North Slope of Alaska, U.S.A. Permafrost and Periglacial Processes, 8: 45– 

67. 

15


16

Campbell Scientific Award 

for 

Best Canadian Student Paper 

17


18

19 



Potential of a Water Balance Model with High Temporal 

Resolution for the Distributed Modelling of Ice- and Snowmelt 

Processes at High Elevated Sites 

ABSTRACT 

GERNOT KOBOLTSCHNIG 1 , HUBERT HOLZMANN 1 , 

WOLFGANG SCHOENER 2 , AND MASSIMILIANO ZAPPA 3 

The potential of the distributed hydrological water balance model PREVAH at the small, highly 

glacierized catchment area of Goldbergkees in the Austrian Alps has been investigated. The model 

is driven by meteorological data from the observatory at Hoher Sonnblick, situated at the highest 

point of the catchment area. A dense network of field observations as additional input and 

validation data has been applied. In the final setting PREVAH has been run in an hourly time step 

based on 722 hydrological response units covering the catchment area. Both snow- and icemelt 

have been simulated by means of an advanced air temperature-index based approach taking 

potential direct solar radiation into account. A multi-validation approach using discharge 

hydrographs, measured snow water equivalent data (SWE), snow cover patterns derived from 

satellite data, and glacier mass balance investigations have been applied to validate the water 

balance of the hydrological year of 2004/2005. The comparison of modelled SWE with spatially 

dense SWE measurements at four different dates within the period May to July 2005 shows quite 

good accordance for both individual elevation bands and the entire catchment. The period of 

2003/2004 has been used for cross-validating the model for discharge-hydrograph and ice melt. 

Icemelt and maximum snow accumulation have been validated against glacier mass balance 

measurements. The individual components contributing to runoff such as rainfall, snow- and 

icemelt have been separated for the hydrological year 2004/05 to estimate their fraction to total 

discharge which is 3.8% for rain, 86.8% for snow, and 9.4% for ice respectively. Finally 

recommendations are given for a possible improvement of hydrologic models considering snow- 

and icemelt at high elevated sites. 

Keywords: glacier melt; snowmelt; alpine hydrology; water balance of high elevated sites; SWE 

investigation 

INTRODUCTION 

According to the location, elevation and topography the contribution to runoff from glacier melt, 

snowmelt and rainfall at glacierized, alpine watersheds varies strongly depending on climate 

1 

University of Natural Resources and Applied Life Sciences (BOKU), Muthgasse 18, A-1190 

Vienna, Austria. E-mail: gernot.koboltschnig@boku.ac.at 

2 

Central Institute of Meteorology and Geodynamics (ZAMG), Hohewarte 38, A-1190 Vienna, 

Austria 

3 

Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Zürcherstrasse 111, 

CH-8903 Birmensdorf, Switzerland

20 



conditions (Singh and Bengtsson, 2005). For calculating the water balance of glacierized 

watersheds, meteorological data have to be obtained for the entire hydrological year to fully cover 

the process of snow accumulation during winter period as well as the snow- and icemelt during 

summer period. Icemelt depends on snow depletion of the glacier and the area of bare ice exposed 

to melt. Consequently, the knowledge of the distributed snow accumulation is of high importance 

to simulate the snow line retreat at the catchment area (Blöschl et al., 1991). The numerical 

modelling of all of the components contributing to runoff and their superposition requires tools for 

the assessment of spatially interpolated meteorological variables, as well as for the treatment of 

distributed geographical variables (Kirnbauer et al., 1994; Gurtz et al., 2003). Due to the 

complexity of different hydrological processes at high elevated sites a multi-validation of 

simulation results is needed (e.g. Günter et al., 1999). Verbunt et al. (2003) recommended using 

discharge hydrographs, water balance elements, snow water equivalent data or soil moisture 

values. High elevated sites with sparse hydro-meteorological observations make model 

assumptions e.g. temperature-index models using wide available and easy to spatially interpolate 

air temperature data necessary. The use of temperature-index models already has a long history. 

Such a relation was first applied for an Alpine glacier by Finsterwalder and Schunk (1887). Hock 

(1999) and Pellicciotti et al. (2005) adapted the classical degree-day method for glacierized areas 

linking air temperature to global radiation respectively potential clear-sky direct solar radiation to 

improve the diurnal cycles of melt. 

The physical basis of temperature-based melt-index methods is demonstrated by Ohmura 

(2002). The paper concluded that the longwave atmospheric radiation is the most dominant heat 

source, and the majority of the atmospheric radiation received at the surface comes from the nearsurface 

layer of the atmosphere. Hence, there is a good, physical based relation between air 

temperature and melt. 

Many studies have been presented for the simulation of snowmelt, icemelt or both for mesoscale 

basins with physically based (Cline et al., 1998; Marks et al, 1999; Lehning et al., 2006) and 

conceptual approaches (Schaefli et al., 2005; Klok et al. 2001; Verbunt et al. 2003), but only few 

studies tested the applicability of such models for very small basins (Arnold et al., 1996 & 1998). 

Zappa et al. (2003) introduced a study comparing different temperature-index models and an 

energy balance model for the snowmelt modelling of an alpine catchment. They showed that the 

improved temperature-index model (according to Hock, 1999) performed better than the energy 

balance approach implemented in PREVAH. 

Thus, the main goal of this study has been to assess the potential of a distributed hydrological 

modelling in a highly glacierized, small, and high elevated basin. For this we simulated the water 

balance for the hydrological year 2004/2005 of the Goldbergkees watershed (Austria) with the 

conceptual distributed hydrological model PREVAH (Precipitation–Runoff–Evapotranspiration– 

HRU model, Gurtz et al., 1999). For validation we compared the computed results with maps of 

the snow water equivalent derived from detailed field campaigns (similar to Elder et al. 1998), 

with a remotely sensed map of the snow distribution (as e.g. Cline et al., 1998; Blöschl et al., 

2002), with observed data of the glacier mass balance (as e.g. Schaefli et al., 2005) and with runoff 

data gauged during the summer and fall at the basin outlet. 

STUDY REGION 

Glacier Goldbergkees is situated directly beneath the Hoher Sonnblick observatory (3106 

m a.s.l., 47°03’16” N, 12°57’25” E) in the central part of the Austrian Alps (Figure 1). Hoher 

Sonnblick is the oldest and highest, permanently staffed observatory in the Alps above 3000 m 

a.s.l.. Meteorological observations at the observatory are available back to 1886, detailed mass 

balance measurements at the nearby glaciers started in 1983 (Auer et al., 2002), and detailed 

hydrological investigations are carried out since 2002. Hence a wide range of hydrological, 

meteorological and glaciological observations are available for detailed melt runoff and water 

balance modelling. The Goldbergkees watershed has an area of about 2.72 km² with a about 52 % 

glacierized (1.43 km² computed for 2003). Elevations ranges between 2350 and 3106 m a.s.l.. The

21 



catchment area is well defined by an automatic discharge gauging station at about 250 meters 

distance downstream the glacier tongue situated at the outlet of a lake (Figure 1). The solid rock 

river bed and nearly laminar flow guarantees reliable runoff gauging. The area is structured into 

three major topographical sections: The upper part comprises south and southeast facing slopes, 

the middle part faces east and northeast, and the lower part comprises the tongue of the glacier 

which faces north and north east. All parts of the catchment are above the timber line. The 

dominant land cover is rock (central alpine gneiss), gravel and ice (unpublished data from 

Koboltschnig et al., 2006). The mean air temperature at Sonnblick observatory is about –5.7°C. 

The annual precipitation at Sonnblick observatory averages about 2680 mm, with 89% as snow 

(climate normals 1961–1990, Auer et al., 2002). 

Figure 1. Catchment area of Goldbergkees. Numbers from 1 to 5 indicate hydro-meteorological stations: 1 

observatory at the top Hoher Sonnblick; 2 air temperature station; 3 air temperature station; 4 discharge 

gauge, air temperature station, and temporary tipping bucket; 5 automatic ultra sonic snow depth 

measurement 

METHODS 

Field investigations and meteorological network 

Hourly data of precipitation, air temperature, moisture, wind speed, sunshine duration and 

global radiation were taken from the observatory at the top of Hoher Sonnblick (Figure 1, Nr. 1).

22 



Precipitation and air temperature data observed at the observatory are shown in Figure 8 (e and f). 

Three additional air temperature stations were installed during 2005 melt season in and next to the 

catchment area (Figure 1, Nr. 2, 3 and 4). A temporary tipping bucket, for liquid precipitation 

measurements only, was installed at the catchment outlet (Figure 1, Nr. 4). The tipping bucket was 

fixed together with the temperature station and the data logger of the discharge gauge. Water 

levels could only be recorded between July and October at a natural cross section of a lake outlet. 

To prevent damage the instrumentation has to be removed during winter period. The rating curves 

were calculated for every melt season separately due to the changes of the hydraulic conditions. 

Snow depths have been measured automatically in a daily interval at about hundred meters of 

elevation beneath the observatory (see Figure 1, Nr. 5). Starting in May 2005 four field campaigns 

for mapping the SWE (snow water equivalent) of the catchment area were performed in monthly 

steps. Aluminium probes were used to measure the snow depth at about 60 to 140 points 

irregularly distributed over the entire catchment area. The snow density has been measured at two 

snow pits following the instructions of Kaser et al. (2003). During the earliest campaign in May 

eight pits have been dug with regard to the higher variability of the snow layers at that time 

(Figure 1, grey triangles). From this very detailed data set we interpolated distributed maps of 

SWE using a spline method to generate 10 m grids of snow depth. Additional sampling points at 

areas with no snow cover have been set to ensure better interpolation results at border areas. The 

measured snow density data have been interpolated using inverse distance weighting technique. 

Finally, SWE were computed from spatialized snow depth and snow density. Snow free areas 

were mapped in field using GPS. The maps of snow free areas have been overlaid to generate the 

final depletion maps. This unique set of SWE grid has been available for visual and quantitative 

comparison with the computation of the hydrological model (Figure 2). 

As a standard program of the mass balance measurements following the glaciological method 

(Hoinkes, 1970; Østrem and Brugman, 1991; Kaser et al., 2003) the ice ablation have been 

measured using ablation stakes, which have been drilled into the bare ice of the glacier. Using 17 

ablation stakes distributed over the entire ablation area of glacier Goldbergkees the net ablation of 

the glacier has been calculated (Hynek and Schöner, 2004).

23 



Figure 2. Simulated vs. observed distributed SWE data for different dates in 2004 and 2005. Black areas 

indicate snow free areas. Mentioned values of SWE mean are distributed SWE values averaged over the 

catchment area (mm) 

The hydrological model PREVAH 

The spatially distributed hydrological model PREVAH (Precipitation–Runoff– 

Evapotranspiration–HRU model, Gurtz et al., 1999) has been used to simulate the processes 

contributing to runoff. PREVAH has already been used at glacierized sites at different spatial 

resolution (Badoux, 1999; Gurtz et al., 2003; Zappa et al., 2003; Zappa et al., 2000). The 

catchment area is subdivided into HRU’s (hydrological response units, Ross et al., 1979) based on 

DEM and land use data. For every HRU the hydrological response to the meteorological input is 

simulated using a typical storage cascade approach. The runoff contributions of all units are added 

to provide the total discharge at the outlet of the entire catchment. Due to the small size of the 

catchment and steep slopes no routing between the spatial units and the river outlet is carried out.

24 



The runoff delay caused by the different glacial reservoirs is managed by the main model storages 

for the melt simulations. This are the snow-, ice- and firn storage, parameterized using the storage 

coefficient and the translation time (Badoux, 1999). PREVAH uses different approaches for the 

snow and icemelt simulations. For this study the radiation based temperature-index approach of 

Hock (1999) has been applied: 

⎧1 

⎪ (MF 

M = ⎨n 

⎪ 

⎩ 

snow / ice 

+ a 

0 

snow / ice 

⋅I 

) ⋅T 

: 

: 

T > 0 

T ≤ 0 

where M is the calculated melt rate (mm h –1 ), MFsnow / ice is the melt factor for snow respectively 

ice (mm d –1 K –1 ), asnow / ice is the radiation melt factor for snow respectively ice, I is the potential 

clear-sky direct solar radiation at ice or snow surface (W m –2 ), T is the air temperature (°C) and n 

is the number of time-steps per day, in this case 24 hours. The melt factors for snow and ice are 

empirical coefficients and I is a calculated value following Hock (1999). 

Only air temperature is needed as input to the model, radiation is calculated as the site adjusted 

potential direct radiation for every HRU, considering exposition and slope. This approach has 

shown the best results in the comparative study of Zappa et al. (2003), where the performance of 

an energy balance model and three degree-day-factor models have been investigated for a mesoscale 

basin. 

For the calculation of the snow accumulation, the simulation of the surface runoff, the 

calculation of snow- and icemelt, and the calculation of the evapotranspiration the following input 

to the PREVAH model is required: air temperature, precipitation, water vapour pressure, global 

radiation, wind speed and sunshine duration. 

Model application at Goldbergkees watershed 

As a major source for the description of the topography to the modelling system a DEM of 10 m 

resolution (Auer et al., 2002), covering the entire catchment area, has been applied. Such high 

resolution is required to account for the small scale variability of the investigated processes for 

such a small basin. The separation into HRU’s has been realized using two land use classes 

(glacier and rock), 50 m elevation bands (16 classes), nine aspect classes and six slope classes. 

Thus 722 HRU’s and 197 meteorological units (MU) were generated. MU are the spatial units 

covering the watershed for which the meteorological data have been interpolated based on hourly 

data from available stations (Figure 1, stations 1, 2, 3, and 4). Because elevation and exposition 

are the main factors governing climatological and meteorological variability in such areas, raster 

elements in the same elevation band and showing similar aspect belong to the same MU. The 

PREVAH model has been run in an hourly time step. For the interpolation of the meteorological 

input an inverse distance weighting and altitude-dependent regression approach has been used 

(Klok et al. 2001). Precipitation has been permanently observed at the observatory and temporary 

at the catchment outlet (Figure 1, stations 1 and 4). For a better weighting two additional virtual 

stations at the sites of the stations 2 and 3 (Figure 1) have been applied, where monthly measured 

precipitation sums have been available. The laps rate for the air temperature has been set to 

0.65°K/100 m. This value has been estimated using long term air temperature measurements of 

two stations in the area. The air temperature input for the melt modelling is taken from 3 stations 

outside the glacier, which is proposed by Lang and Braun (1990). 

The simulation of one entire hydrological year with PREVAH, producing tables of hourly 

runoff and storage output, and daily maps of SWE takes less than 3 minutes computation time 

using a standard PC. We proceeded therefore with manual tuning of the parameters governing the 

processes of snow accumulation, snowmelt, icemelt and runoff generation (Zappa et al., 2003; 

Gurtz et al., 2003) for the hydrological year 2004/2005, where we have additional observations on 

snow cover and SWE for multi-criteria verification. The verification for the discharge simulation 

(1)

25 



has been carried out by the analysis of the model performance for the hydrological year 

2003/2004. 

Calibration procedure 

First step: Due to the extreme climate conditions during 2003 melt season the catchment area of 

Goldbergkees glacier was nearly snow free. Hence it has been possible to calibrate the melt 

parameters for icemelt separated from other processes. Therefore the observed and simulated 

discharge hydrographs have been compared; 

Second step: The model has been initialized using spatially distributed SWE data at the time of 

the maximum snow accumulation (begin of May). Thus the degree day factor for snowmelt has 

been calibrated comparing the daily results of the internal snow storage with the in field observed 

spatially distributed SWE data at different points in time; 

Third step: The model has been initialized using spatially distributed SWE data at the beginning 

of the accumulation period (equal to the beginning of the hydrological year in the northern 

hemisphere at the beginning of October). For this reason the solid precipitation has been corrected 

by plus 18 % more precipitation following Sevruk (1986). The snow accumulation of the 

PREVAH model has been compared with the in field observed spatially distributed SWE data at 

the time of the maximum accumulation; 

Fourth step: Fine tuning of the melt parameters and adaptation of the storage parameters 

comparing observed and simulated hydrographs. The storage time for snow- and icemelt are 

calibrated for the Goldbergkees catchment adducting the recession curves of the observed 

hydrograph, which result when summer snowfalls reduce ablation by raising albedo (Collins, 

1982). Three events of this kind have been observed at 2005 melt season (see Figure 4). For the 

determination of the translation time of snow- and icemelt the diurnal maximums of the simulated 

runoff have been fitted to the observed hydrograph. 

Figure 4. Simulated vs. observed snow cover pattern at 29 July 2005. The observed image is generated by a 

classification of an ASTER image. Black coloured areas indicate snow free areas, white areas are still snow 

covered. 

Verification procedure 

First step: the distributed SWE simulation has been verified at 4 dates of the 2004/2005 period; 

Second step: the simulation of the snow patterns has been verified with satellite data from 

ASTER; 

Third step: the skill of the runoff simulation has been independently verified without further 

adjustment of the calibrated parameters by comparisons with the hydrological year 2003/2004; 

Fourth step: observed data on glacier ablation for both 2003/2004 and 2004/2005 have been 

compared with the simulated ice ablation.

26 



Main model settings 

For simulating the hydrological year of 2004/2005 the calibrated model has been initialized at 

the beginning of October 2004 by in field observed SWE data. The main calibrated parameters of 

the model are shown in Table 1 in comparison to similar studies (Pellicciotti et al., 2005; Hock, 

1999; Zappa et al., 2003). The melt factors presented in this paper are of the same size as the 

compared values. The degree-day-factor for ice is slightly higher, which could be explained by the 

dark ice surface and therefore quite low albedo of Goldbergkees glacier. The snowmelt model is 

parameterized by the radiation melt factor and the maximum and the minimum temperature 

dependent melt factors, which define the maximum at 21 st of June and the minimum at 21 st of 

December of a sinus shaped function. The temperature dependent melt factor for ice is constant in 

time. 

Parameter 

Table 1. Comparison of the model parameters calibrated at Goldbergkees watershed and the 

parameters of the studies of Pellicciotti et al. (2005), Hock (1999) and Zappa et al. (2003) 

RESULTS AND DISCUSSION 

Cal. 

Value 

Zappa Hock Pellicciotti 

Threshold temperature for snowmelt 0 0 0 1 [°C] 

Max. degree day factor 3.2 – – – 

[mm d –1 K – 

1 

] 

Min. degree day factor 1 – – – 

[mm d –1 K – 

1 

] 

Degree day factor – 0.8 1.8 1.97 

[mm d –1 K – 

1 

] 

Radiation melt factor for snow 0.00015 0.00027 0.0008 0.00052 

[mm W –1 

m² K –1 h –1 ] 

Temperature melt factor for ice 2.15 – 1.8 1.97 

[mm d –1 K – 

1 

] 

Radiation melt factor for ice 0.0003 – 0.0006 0.00106 

[mm W –1 

m² K –1 h –1 ] 

Storage time for snowmelt 25 – – – [h] 

Storage time for icemelt 2 – – – [h] 

Translation time for snowmelt 3 – – – [h] 

Translation time for icemelt 2 – – – [h] 

Distributed snow-water-equivalent (SWE) 

The PREVAH model makes a daily output of the distributed SWE storage possible. Thus the 

validation using observed SWE data is an additional alternative. Figure 2 shows the validation of 

the SWE for four different points in time starting with the initial SWE model settings of 6 October 

2004. The simulated plot for this date shows the mean value of the SWE storage at the end of the 

first simulated day described for the internal distribution of the meteorological units. Through this 

semi-distributed approach some information on spatial distribution is lost. The simulated and 

observed maximum snow accumulation at 6 May 2005 are in a very good agreement. At 2 June 

2005 the simulation shows a little smaller SWE storage than the month before but the observed 

SWE seems to show an overestimation through the field measurement. There is no possibility to 

explain the high accumulation by observed precipitation data. At 7 July 2005 still a slightly 

overestimation is seen. The last observation at 28 July 2005 shows a good accordance of the 

simulated SWE, averaged over the catchment area. The simulated distribution of snow left in the 

Unit

27 



catchment strongly depends on the distribution of precipitation, whereas the in field observed 

distribution of snow depends on precipitation gradients, snow redistribution by wind and vertical 

drift of snow induced by avalanches (Hartmann et al., 1999; Blöschl et al., 1991). 

The spatial analysis of the distributed observed and simulated SWE maps is shown in Figure 3. 

The catchment area has been divided into 8 100 m elevation bands starting at 2300 m a.s.l.. There 

is a very good agreement between observation and simulation at 6 May 2005 at the elevation 

bands between 2600 and 2900 m (Figure 3a). The upper bands show an overestimation and the 

lower bands an underestimation of the simulation. This result is due to the snow redistribution 

from the upper to the lower parts, induced by wind or avalanches. The problem of too high snow 

accumulation measured at 2 June 2005 can be seen in Figure 3b, where the simulated SWE is 

constantly about 10% lower than the observed, despite the two uppermost elevation bands, which 

are fine modelled. At Figure 3c and Figure 3d it can be seen, that the elevation band at 2700 to 

2800 m is in a good agreement and again the upper ones are overestimated and the lower ones are 

underestimated by the simulation. This result is again due to the problem of vertical snow 

redistribution. 

Figure 3. Simulated SWE (solid line, dots) vs. observed SWE (dashed line, x) averaged over 100 m elevation 

bands at four different points in time. 

Figure 8b shows the observed daily snow depth at station 5 (Figure 1) and Figure 8c shows the 

simulated SWE averaged over the entire catchment area. The correlation coefficient calculated for 

the ascending phase is about r²=0.91, for the descending r²=0.99, and for the entire period r²=0.92. 

We imply that the good correlation at the descending phase is due to a quite homogeneous snow 

density and quite similar snow melt processes over the entire catchment area. The ascending phase 

of the observed snow depth measurement (Figure 8b) shows typical settling effects of the 

snowpack, which can not be simulated by models without physically based assumptions for the 

snowpack modelling (Lehning et al., 2006). 

Snow cover patterns 

An ASTER (L1B) image of good quality has been available for 29 July 2005. The image has 

been classified (unpublished data from Vollmann, 2006) to generate a map of snow cover patterns. 

Figure 4 shows the comparison of the observed and simulated snow cover map. The lower parts of 

the catchment area are simulated as nearly snow free. The observation shows a more complex 

structure of the snow line retreat, but the simulated snow cover pattern is in a good agreement with

28 



the observed pattern. Inadequacies arise from processes not included in the model such as 

redistribution of snow by wind and avalanches (Blöschl et al., 1991; Lehning et al., 2000; 

Doorschot et al., 2001). At such small scale and with the adoption of such a high spatial resolution 

we can see the role of snow redistribution and understand how distributed models have to be 

improved to account for such processes. 

Runoff simulations 

For the 2005 melt season (calibration period) discharge data have been available for the period 

of 9 July–30 September (Figure 5). 

Figure 5. Simulated icemelt averaged over the catchment area, M (mm h –1 ), hourly data of air temperature, T 

(°C), and precipitation P (mm h –1 ) at Hoher Sonnblick observatory, and observed and simulated discharge Q 

(m³ s –1 ) at the catchment outlet of Goldbergkees. The increasing lines at the bottom plot indicate the 

cumulated discharge Q cum (10 6 m³) over the period of the discharge observations from 10 July to 30 

September 2005. 

The beginning of the observations shows rain-induced discharge peaks. Typical diurnal 

variations of the observed runoff, induced by icemelt are seen in the period from end of August 

until mid of September. The performance of the simulation has been employed according to the 

efficiency criterion R NS² (Nash and Sutcliffe, 1970), defined as:

R 

∑i 

∑ 

n 

2 

= 1 

NS = − n 

i = 1 

29 



2 

( Qobs 

− Qsim 

) 

1 (2) 

2 

( Q − Q ) 

obs 

obs 

where Q is the hourly value of the catchment runoff (m³ s –1 ) and the subscripts obs and sim refer 

to the observed and simulated runoff. The bar refers to the mean of the observed runoff and n is 

the number of time-steps for which R NS² is calculated. The final model performance accounts for 

R NS² = 0.77. The scatter plot in Figure 6 shows the comparison of observed and simulated runoffs. 

It is seen, that low flow and mean flow conditions are good represented but that there are some 

higher values of discharges below the 45° line which have not been simulated. Higher discharges 

are cushioned through the quite high value of the snow melt storage time (25 h), since rainfall on 

the snow surface has to go through this storage. The lowering of the snow melt storage time would 

effect into a steep recession curve (e.g. recession between 4th and 5 th of August in Figure 2) and 

on the other hand the lifting of peak discharges is quite low. Hence, the simulated runoff 

hydrograph shown in Figure 5 is the result of an optimization. The main nature of the observed 

hydrograph with respect to the diurnal and seasonal fluctuations through the superposition of 

melting processes are very good represented by the simulation. 

Figure 6. Scatter plot of the observed vs. simulated hourly discharge for the period from 10 July to 30 

September 2005. 

Looking at the entire simulation period of 2004/2005 (Figure 8, d) it is seen that there has been a 

time span of about five month with no flow. The earliest runoff processes of the melt season 

started at the beginning of May. During the period from May to July total discharge is mainly 

formed by snowmelt (see Figure 8, d and Figure 9). Peak discharges occur during June and July, 

where all components contributing to runoff, despite icemelt, reach their maximum. 

For the verification season 2003/2004 the model has been run using the same set of parameters 

as shown in Table 1. Solid precipitation has been corrected by plus 18% like in the calibration 

period 2004/2005. The model performance, employed according to the efficiency criterion 

following Nash and Sutcliffe (1970), has been calculated for RNS² = 0.60. Figure 7 presents the 

model results in detail. The hydrologic response of the Goldbergkees catchment has been different 

from the year 2004/2005 presented before, but the processes of snow- and icemelt are well 

simulated. A slightly overestimation of the simulated cumulated runoff can be seen. During

30 



2003/2004 period only the meteorological input data of the observatory at Hoher Sonnblick 

(Figure 1, Nr. 1) has been available for simulations. Hence, there is a lack of input data for the 

interpolation and regionalisation of meteorological variables, which results in a lower model 

performance. Nevertheless this shows that the calibrated model is able to simulate different 

climatic conditions independently from additional validation data and re-calibration of model 

parameters. 

Figure 7. Simulated icemelt averaged over the catchment area, M (mm h –1 ), hourly data of air temperature, T 

(°C), and precipitation P (mm h –1 ) at Hoher Sonnblick observatory, and observed and simulated discharge Q 

(m³ s –1 ) at the catchment outlet of Goldbergkees. The increasing lines at the bottom plot indicate the 

cumulated discharge Q cum (10 6 m³) over the period of the discharge observations from 27 July to 30 

September 2004.

31 



Figure 8. Main hydro-meteorological observations vs. simulated hourly discharge and daily output of the 

SWE storage. a: daily observed depths of fresh snow (cm) and cumulated depths of fresh snow (maximum at 

1787 cm) over the period 1 October 2004 to 30 September 2005 at the automated ultrasonic station; b: daily 

measurements of the snow depth (cm) at the automated ultrasonic station; c: simulated SWE storage (mm); d: 

simulated hourly discharge (m³ s –1 ); e: hourly air temperature measurements (°C) at Hoher Sonnblick 

observatory; f: hourly precipitation (mm) at Hoher Sonnblick observatory.

32 



Water balance 

The water balance, as a final model result, states an additional possibility for the model 

validation. As the model generates output for all water balance components using easy balance 

equations the results of the hydrological year 2004/2005 (see Table 2) can be verified. 

dS = (SWEend 

- SWEbegin 

) - ICEmelt 

dS = (309 -152) 

- 277 = −120 

mm 

Table 2. Components of the annual water balance of Goldbergkees catchment for the hydrological year 

of 2004/2005 

WB component Values (mm) Description 

SWE begin 152 SWE for model initialization at begin of October 2004 

P adj 2991 Adjusted precipitation over the catchment area 

ETA 128 Real evapotranspiration 

Q tot 2956 Total runoff 

SNOWmelt 2566 Total snowmelt 

ICE melt 277 Total icemelt 

SWEend 309 SWE left at the end of the period at end of September 2005 

Equation 3 calculates the change of the main storages, where dS is the change of the storages over 

the entire simulation period. (definitions of symbols can be seen at Table 2) 

WB = Padj 

- Qtot 

WB = 

- ETA - dS 

2991 - 2956 -128 

+ 120 = 27 mm 

Equation 4 evaluates the water balance, whereas WB is the annual water balance. The result of 

WB = 27 mm is explained by the difference of the base flow storage component at the simulation 

end minus the base flow storage value for the initialization at the beginning. Equation 5 shows the 

share of the components snow melt, icemelt and rainfall contributing to runoff. 

Q = SNOWmelt 

+ ICEmelt 

+ RAINdirect 

=> RAIN = 2956 - 277 - 2566 = 113 mm 

RAINdirect indicates the share of the total runoff which has been induced by effective rainfall, 

directly routed to discharge. Liquid precipitation (RAINdirect) accounts for 3.8%, icemelt (ICE) 

accounts for 9.4%, and snowmelt (SNOW) accounts for 86.8%. The value for liquid precipitation 

seems to be underestimated, because snowmelt has a very high value. Snow melt is defined as the 

snow which is melted by the energy affecting the snow surface, whereas melted snow does not 

have to be routed directly to discharge. 

The simulated snow accumulation reached its maximum at 20 May 2005 at 1507 mm (Figure 8, 

c), compared to SWE field measurements at 6 May 2005 showing a value of about 1391 mm. The 

monthly water balance is presented in Figure 9. Icemelt starts in July and typically has its 

maximum in August. Glacier melt was also simulated for November 2004, because of quite high 

temperatures at that time. Using field measurements of the ice ablation stakes during the 

hydrological year of 2004/2005, 510 mm of ice loss are calculated over the glacierized area 

(unpublished data from Schöner et al., 2006). The simulated ice loss accounts for 277 mm over the 

(3) 

(4) 

(5)

33 



catchment area, which would be 502 mm assigned to the glacierized area. This very good 

agreement between simulated and observed icemelt is due to a reliable estimation of all modelled 

components. Evaporation accounted for 128 mm for the entire simulated year and is mainly 

affected by low temperatures at this alpine site. 

Figure 9. Monthly simulated balance of runoff (mm), precipitation (mm), icemelt (mm), snow melt (mm), 

and real evaporation (mm) averaged over the catchment area over the period October 2004 to September 

2005. 

The observed ice loss for the verification period of 2003/2004 accounted for 197 mm averaged 

over the glacierized area and the snow accumulation measured at the beginning of May 2004 

accounted for 1744 mm (unpublished data from Schöner et al., 2006). The simulation results for 

the verification period accounted for 275 mm ice loss averaged over the glacier and 1690 mm of 

SWE calculated for the beginning of May. This means a slightly overestimation of the ice loss and 

a slightly underestimation of the snow accumulation . Once more we can show the quality of the 

stabile multi-validation approach used for the 2004/2005 period. 

CONCLUSIONS AND OUTLOOK 

The PREVAH model shows the ability to model the waterbalance as well as the discharge 

hydrograph of the high elevated Goldbergkees watershed. Melt processes have been simulated 

accurately in different temporal scales: the simulated diurnal variations of runoff matched the 

runoff observations during melt season and the seasonal balance is in a good agreement with the 

observed mass balance data. This paper has shown the need of various observed data like 

discharge hydrographs, distributed snow water equivalent and ice ablation data to guarantee a 

stabile cross-validation of simulation results (Verbunt et al. 2003). The model results concerning 

runoff relied on the input of precipitation and air temperature, all the other input has been 

necessary for the evaporation simulations. Hence, there should be a quite small operating expense 

for data acquisition for melt modelling, despite all the problems and uncertainties occurring during 

precipitation and air temperature measurements (Sevruk, 1986, 1989). The watershed of 

Goldbergkees has demonstrated its great advantage of having a meteorological observatory on site 

and long term and detailed mass balance measurements (Auer et al., 2002). The availability of 

spatial dense meteorological data of good quality at this isolated location is a unique feature and 

an advantage of the study region. Main efforts for the preparation of simulations have been due to

34 



the acquisitions of validating data, with respect to logistic manner, mountainous risks and strictly 

weather depending planning of field investigations. 

As PREVAH does not include the redistribution of snow by wind or avalanches the spatial 

distribution of the SWE cannot be matched perfectly, but the mean values of the modelled SWE 

are identical to measured catchment means of the SWE. Hence, it seems that the main part of the 

drifted snow originates from the simulated watershed and remains inside. An additional 

conceptual module for the snow redistribution by wind and avalanches (Hartmann et al., 1999) 

should be applied to the model to satisfy these claims. It seems to be obvious, that only in a very 

good observed catchment area reliable simulation results of high quality can be obtained. 

ACKNOWLEDGEMENTS 

This ongoing research was supported by a grant from the Austrian Academy of Sciences under 

the project SNOWTRANS HOE29, part of the IHP PUB (International Hydrological Program, 

Prediction in Ungauged Basins). Many thanks to Markus Vollmann for the classification of the 

ASTER image. The authors are grateful to all the students, colleagues and friends who helped to 

carry out the exhausting field work. 

REFERENCES 

Arnold N., Richards K., Willis I., Sharp M. 1998. Initial results from a semi-distributed, physically 

based model of glacier hydrology. Hydrological Processes 12: 191–219. 

Arnold N.S., Willis I. C., Sharp M., Richards K.S., Lawson W. 1996. À distributed surface 

energy balance model for a small valley glacier. I. Development and testing for the Haut 

Glacier d'Arolla, Valais, Switzerland', Journal of Glaciology, 42: 77–89. 

Auer I., Böhm R., Leymüller M., Schöner W., Kaiser A., Scheifinger H., Langer M., Scheider St., 

Häberli Ch. 2002. Das Klima des Sonnblicks. Klimaatlas und Klimatographie der GAW 

Station Sonnblick einschließlich der umgebenden Gebirgsregion. Österreichische Beiträge zu 

Meteorologie und Geophysik. Heft 28. Zentralanstalt für Meteorologie und Geodynamik, 

Wien. ISSN 1016–6254. 

Badoux A. 1999. Untersuchung zur flächendifferenzierten Modellierung von Abfluss und 

Schmelze in teilvergletscherten Einzugsgebieten. Diploma thesis at the Geographical Institute 

of ETH, Zürich. 

Blöschl G., Kirnbauer R., Jansa J., Kraus K., Kuschnig D., Gutknecht D., Reszler C. 2002. Using 

remote sensing methods for calibrating and verifying a spatially distributed snow model. 

Österreichische Wasser- und Abfallwirtschaft (ÖWW) 54: 16 pp. 

Blöschl G., Kirnbauer R., Gutknecht D. 1991. Distributed Snowmelt Simulations in an Alpine 

Catchment, 1, Model Evaluation on the Basis of Snow Cover Patterns. Water Resources 

Research, 27(12), 3171–3179. 

Collins D.N., 1982. Water storage in Alpine glaciers. Hydrological Aspects of Alpine and High 

Mountain Areas (Proceedings of the Exeter Symposium, July 1982) IAHS Publ, no 138: 113– 

122. 

Doorschot J., Raderschall N., Lehning M. 2001. Measurements and one-dimensional model 

calculations of snow transport over a mountain ridge. Annals of Glaciology, 32: 153–158. 

Elder K., Rosenthal W., Davis R.E. 1998. Estimating the spatial distribution of snow water 

equivalence in a montane watershed. Hydrological Processes 12: 1793–1808. 

Finsterwalder S., Schunk H. 1887. Der Suldenferner. Zeitschrift des Deutschen und 

Oesterreichischen Alpenvereins 18: 72–89. 

Gurtz J., Zappa M., Jasper K., Lang H., Verbunt M., Badoux A., Vitvar T. 2003. A Comparative 

Study in Modelling Runoff and its Components in Two Mountainous Catchments. 

Hydrological Processes 17: 297–311. DOI: 10.1002/hyp.1125

35 



Güntner A., Uhlenbrook S., Seibert J., Leibundgut C. 1999. Multi-criterial validation of 

TOPMODEL in a mountainous catchment. Hydrological Processes 13: 1603–1620. 

Hartman M.D., Baron J.S., Lammers R.B., Cline D.W., Band L.E., Liston G.E. Tague C. 1999. 

Simulations of snow distribution and hydrology in a mountain basin. Water Resources 

Research, 35(5): 1587–1603. 

Hock R. 1999. A distributed temperature-index ice- and snowmelt model including potential direct 

solar radiation. Journal of Glaciology 45(149): 101–111. 

Hock R., Holmgren B. 2005. A distributed surface energy-balance model for complex topography 

and its application to Storglaciären, Sweden. Journal of Glaciology 51(172): 25–36. 

Hoinkes H. 1970. Methoden und Möglichkeiten von Massenhaushaltsstudien auf Gletschern. 

Ergebnisse der Meßreihe Hintereisferner (Ötztaler Alpen) 1953–1968. Zeitschrift für 

Gletscherkunde und Glazialgeologie, 6, 37 – 90. 

Hynek B., Schöner W. 2004. Massenhaushalt 2002/2003 der Gletscher in der Goldberggruppe. 

101–102. Jahresbericht des Sonnblick-Vereines: 1–15. Eigenverlag der Zentralanstalt für 

Meteorologie und Geodynamik, Wien. 

Kaser G., Fountain A., Jansson P. 2003. A manual for monitoring the mass balance of mountain 

glaciers. IHP-VI, Technical Documents in Hydrology No. 59, UNESCO, Paris, 2003. 

Kirnbauer R, Blöschl G, Gutknecht D. 1994. Entering the Era of Distributed Snow Models. Nordic 

Hydrology 25: 1–24. 

Klok E.J., Jasper K., Roelofsma K.P., Gurtz J., Badoux A. 2001. Distributed hydrological 

modelling of a heavily glaciated Alpine river basin. Hydrological Sciences Journal—Journal 

Des Sciences Hydrologiques 46: 553–570. 

Lang H., Braun L. 1990. On the Information content of air Temperature in the Context of Snow 

Melt Estimation. In: Hydrology of Mountainous Areas, L. Molar (Ed.). IAHS Publ. no. 190: 

347–354. 

Lehning M., Voelksch I., Gustafsson D., Nguyen T.A., Stähli M., Zappa M. 2006. Influence of 

snow, forest, soil and local meteorological forcing on mountain catchment hydrology. 

Hydrological Processes, in press: DOI: 10.1002/hyp.6204. 

Lehning M., Doorschot J., Bartelt P. 2000. A snowdrift index based on SNOWPACK model 

calculations. In Annals of Glaciology, 31: 382–386. 

Marks D., Domingo J., Susong D., Link T., Garen D. 1999. A spatially distributed energy balance 

snowmelt model for application in mountain basins. Hydrological Processes 13:1935–1959. 

Nash J.E., Sutcliffe J.V. 1970. River flow forecasting through conceptual models (1), a discussion 

of principles. Journal of Hydrology 10 (3): 282–290. 

Ohmura A. 2001. Physical Basis for the Temperature-Based Melt-Index Method. Journal of 

Applied Meteorology 40, No. 4: 753–761. DOI: 10.1175/1520- 

0450(2001)0402.0.CO;2 

Østrem G., Brugman M. 1991. Glacier mass-balance measurements—A manual for field and 

office work: Environment Canada, National Hydrology Research Institute Science Report No. 

4, and Norwegian Water Resources and Energy Administration, 224 p. 

Pellicciotti F., Brock B., Strasser U., Burlando P., Funk M. Corripio J. 2005. An enhanced 

temperature-index glacier melt model including the shortwave radiation balance: development 

and testing for Haut Glacier d’Arolla, Switzerland. Journal of Glaciology 51(175): 573–587. 

Ross B.B., Contractor D.N., Shanholtz V.O. 1979. A Finite Element Model Of Overland And 

Channel Flow For Assessing The Hydrologic Impact Of Landuse Change. Journal of 

Hydrology 41: 1–30. 

Schaefli B., Hingray B., Niggli M., Musy A. 2005. A conceptual glacio-hydrological model for 

high mountainous catchments, Hydrology and Earth System Sciences Discussions 2: 73–117. 

Sevruk B. (ed.) 1986. Correction of Precipitation Measurements. ETH/IASH/WMO Workshop on 

the Correction of Precipitation Measurements. Zürich, April 1–3, 1985, Zuercher 

Geographische Schriften, Heft 23.

36 



Sevruk B. (ed.) 1989. Precipitation Measurement. Proc. of the WMO/IASH/ETH Workshop on 

Precipitation Measurement. St. Moritz, December 3–7, 1989, WMO Instruments and 

Observing Methods Rep., No. 48, WMO/TD-No. 328, 199–204, 1989, Geneva. 

Singh P., Bengtsson L. 2005. Impact of warmer climate on melt and evaporation for the rainfed, 

snowfed and glacierfed basins in the Himalayan region. Journal of Hydrology 300: 140–154. 

DOI: 10.1016/j.hydrol.2004.06.005 

Verbunt M., Gurtz J., Jasper K., Lang H., Warmerdam P., Zappa M. 2003. The hydrological role 

of snow and glaciers in alpine river basins and their distributed modelling. Journal of 

Hydrology 282: 36–55. DOI: 10.1016/S0022-1694(03)00251-8 

Zappa M., Pos F., Strasser U., Gurtz J. 2003. Seasonal water balance of an alpine catchment as 

evaluated by different methods for spatially distributed snow melt modelling. Nordic 

Hydrology 34(3): 179–202 

Zappa M, Badoux A., Gurtz J. 2000. The application of a complex distributed hydrological model 

in a highly glaciated alpine river catchment, in Limnological Reports, Horvatic J. (Editor), 33 rd 

Conference of International Association for Danube Research, Osijek, Croatia, 3–9 Sept. 2000, 

Vol. 33: 23–28.

Snow and Climate 

37


38

39 



20 th Century North American Snow Extent Trends: 

Climate Change or Natural Climate Variability? 

EXTENDED ABSTRACT 

ALLAN FREI, 1 GAVIN GONG, 2 DAVID A. ROBINSON, 3 

GWANGYONG CHOI, 4 DEBJANI GHATAK, 5 AND YAN GE 6 

The purpose of this analysis is to test the null hypothesis that continental scale variations in 

North American snow cover extent (NA SCE) can be explained by atmospheric circulation alone, 

without need to invoke additional explanatory factors such as climate change. We test the null 

hypothesis by (1) presenting what is known about decadal scale variations in twentieth century 

continental scale NA SCE, and (2) examining historical variations in surface climate, tropospheric 

and stratospheric circulation, as well as corollary evidence from arctic sea ice variations, to 

determine whether the available evidence supports or refutes the null hypothesis. In this 

presentation, preliminary results are presented focusing on snow extent during spring (i.e., March). 

The full report is being prepared for submission for publication elsewhere. 

METHODS AND DATA 

In order to test the null hypothesis, we utilize data sets that extend back to, and beyond, the mid 

twentieth century. Variations in snow depth, surface temperature and precipitation rate, as well as 

upper tropospheric and mid stratospheric geopotential heights and wind speeds are examined 

using time series, composite, and correlation analyses. Time series analyses are used to identify 

climatic features in these fields that covary over interannual and decadal time scales. To evaluate 

decadal changes, time series were smoothed using 9-year running means as a low pass filter in 

order to focus on decadal and longer scale variations. The results and conclusions are robust with 

regards to changes in the smoothing window size. Observations of snow extent are taken from the 

NOAA visible based satellite product (Ramsay 1998; Robinson et al. 1999; Helfrich et al. 2006); 

reconstructed snow extent is from two sources (Brown 2000; Frei et al. 1999); snow depth is from 

a new gridded product (described in Dyer and Mote 2006); climatological fields are from the 

NCEP/NCAR Reanalysis project (Kalnay and co-authors 1996); and teleconnection indices are 

from the NOAA Climate Diagnostic Center, the Climate Research Unit of the University of East 

Anglia, and the National Center for Atmospheric Research. 

1 

Department of Geography, Hunter College, Graduate Program in Earth and Environmental 

Sciences, City University of New York, NY. 

2 

Department of Earth and Environmental Engineering, Columbia University, NY. 

3 

Department of Geography, Rutgers University, NJ. 

4 

Department of Geography, Rutgers University, NJ. 

5 

Department of Geography, Hunter College, Graduate Program in Earth and Environmental 

Sciences, City University of New York, NY. 

6 

Department of Earth and Environmental Engineering, Columbia University, NY.

Time series of NA SCE and AO 

During spring, the AO appears to contribute significantly to the variance in snow extent at both 

interannual and decadal scales (table 1). Neither the PNA nor the PDO are significantly correlated 

to snow extent individually, but at interannual time scales they do appear to contribute in the 

multiple correlation analysis. The smoothed (i.e. decadal scale) correlation between AO and snow 

extent is very high (rsmooth = –0.93), explaining ~86% of the variance, while neither of the other 

two indices contribute to the explanatory power. Note that a correlation of this magnitude (rsmooth 

=–0.93) is not significant (p=0.05) in this analysis because of the diminished number of effective 

degrees of freedom due to autocorrelation in the time series. Figure 1 shows smoothed time series 

that have been normalized so that units are comparable. March snow extent (inverted), including 

values from satellite observations and an historical reconstruction, are shown along with AO 

variations and two historical NAO time series derived from station observations. March snow 

extent and the AO appear to vary inversely, and the reconstructed snow and NAO time series 

indicate that this relationship holds as far back as the mid 1940s; prior to that time the relationship 

appears to weaken. It is not clear whether the pre-1940s deterioration of the snow-circulation 

relationship is real, or due to (1) the difference between the EOF-based AO and the station-based 

NAO time series, or (2) to possible inaccuracies in one or more of the historical data sets. 

Table 1. Pearson correlation coefficients, and multiple correlation coefficients, between North 

American snow extent based on satellite observations and atmospheric circulation indices for annual 

(r annual) and 9-year running mean (r smooth) time series. Values marked with an asterisk are statistically 

significant at p=0.05. Sample sizes for annual and smoothed time series are n annual = 39 and n smooth = 31, 

respectively. However, effective sample sizes are smaller due to temporal autocorrelation. 

SPRING 

January–February–March Indices 

versus 

March Snow Extent 

rannual rsmooth AO –0.45* –0.93 

PNA –0.26 –0.25 

PDO 0.00 –0.25 

AO/PNA 0.54* 0.93 

AO/PDO 0.46* 0.93 

PNA/PDO 0.38 0.25 

AO/PNA/PDO 0.60* 0.93 

Figure 1. 9-yr running mean values of JFM AO (solid), extended NAO from station observations (from the 

Climate Research Unit at the University of East Anglia) (dashed), extended NAO from station observations 

(from the National Center for Atmospheric Research) (dots), March snow extent from satellite observations 

and reconstructed March snow extent from (Brown 2000) (asterisks) which have been inverted for easier 

comparison. All time series have been normalized to fit on the same scale. 

40

Spatial distributions of changes in snow extent, snow depth, temperature, precipitation, and 

atmospheric circulation 

The differences in snow extent between periods of more extensive and less extensive snow 

cover occur at the southern boundary of the snow pack. This is in contrast to the differences in 

snow depth, which are shallower across wide swaths of the continent. These variations in the 

March snow pack are associated with changes in surface climate during March and/or preceding 

months. Mean temperature differences on the order of 1–2 C across wide swaths of the continent 

are observed, and moderate precipitation changes on the order of 0.5 mm/day primarily are found 

near the southern boundary of the snow pack. The spatial distribution of these surface changes 

suggests that they might be caused by a southward displacement of the mean polar front and jet 

stream. Composite difference maps of the same atmospheric fields for positive versus negative AO 

years show that the upper tropospheric and stratospheric changes associated with the AO are 

similar in character, and greater in magnitude, than changes associated with snow cover. 

Links to sea ice variability 

In light of the well-studied impact of the AO on Arctic sea ice extent, and in conjunction with 

the results of the analyses presented in the previous sections, one would expect to find a signal in 

the sea ice record that is correlated to both the winter–spring AO and to march NA SCE. However, 

recent reports of decreasing sea ice suggest a possible discrepancy. Results of an EOF analysis of 

arctic sea ice extent shows that the record of sea ice extent appears to be consistent with our null 

hypothesis. Winter – spring AO variations do appear to be influencing both NA SCE and arctic 

sea ice. The arctic sea ice signal is regionally dependent, exerting opposite effects in different 

regions, and therefore dampened in the spatially integrated signal for the entire arctic. 

DISCUSSION AND CONCLUSIONS 

Evidence presented here suggests that these variations are manifestations of decadal scale 

variations in the position of the polar front and jet stream, partially influenced by some of the 

primary modes of atmospheric variability. Statistically significant relationships are found between 

March North American snow extent / snow depth / surface climate and the winter AO since the 

mid-20 th century. These relationships are stronger at decadal than at interannual time scales, and 

appear to weaken prior to mid century. Further supporting evidence is provided by the dominant 

EOF of Northern Hemisphere winter sea ice extent, which displays decadal scale variations that 

are also correlated to historical variations in both the AO and snow extent, as would be expected 

under the null hypothesis. Results from climate models are consistent with the empirical results 

presented here. Current-generation coupled atmosphere–ocean climate model results suggest that 

decadal scale variability of the magnitude discussed here is likely due to internal climate 

variability and not to external forcing. Thus, evidence examined in this analysis leads to the 

conclusion that we can not presently reject our null hypothesis, and that changes observed thus far 

in the continental scale snow extent record over North America can not be attributed to 

anthropogenic forcing. 

REFERENCES 

Brown, R. D. (2000). Northern hemisphere snow cover variability and change, 1915–1997. 

Journal of Climate 13(13): 2339–2355. 

Dyer, J. L. and T. L. Mote (2006). Spatial variability and trends in snow depth over North 

America. Geophysical Research Letters: in review. 

Frei, A., D. A. Robinson and M. G. Hughes (1999). North American snow extent: 1900–1994. 

International Journal of Climatology 19: 1517–1534. 

Helfrich, S. R., D. McNamara, B. H. Ramsay and T. Kasheta (2006). Enhancements and 

forthcoming developments to the Interactive Multisensor Snow and Ice Mapping System 

(IMS). 63rd Eastern Snow Conference. University of Delaware, Newark Delaware. 

41

Kalnay, E. and co-authors (1996). The NCEP/NCAR 40-year reanalysis project. Bulletin of the 

American Meteorological Society 77: 437–471. 

Ramsay, B. H. (1998). The interactive multisensor snow and ice mapping system. Hydrological 

Processes 12: 1537–1546. 

Robinson, D. A., J. D. Tarpley and B. H. Ramsay (1999). Transition from NOAA weekly to daily 

hemispheric snow charts. 10th Symposium on Global Change Studies, Dallas, TX, American 

Meteorological Society. 

42

Snow and Climate Posters 

43


44

45 



Observed Differences Between Snow Extent and Snow Depth 

Variability at Continental Scales 


YAN GE 1 AND GAVIN GONG 1 

Snow extent and snow depth are two related characteristics of a snowpack, but they need not be 

mutually consistent. Different behaviors of snow depth and snow extent are readily apparent at 

local scales; for example, during conditions of uninterrupted snow cover, snow depth at a point 

can vary substantially due to snowfall events, metamorphosis and ablation. However, the behavior 

of snow extent vs. snow depth at regional to continental scales is understudied. 

Regional/continental-scale gridded datasets of snow depth and snow water equivalent (SWE), 

obtained from field observations and satellite data, were utilized to quantitatively evaluate the 

relationship between snow extent and snow depth over North America. Statistical methods were 

applied to assess the mutual consistency of monthly snow depth vs. snow extent. Results from all 

the datasets indicate the significance of snow depth variations, and that snow depth anomalies and 

snow extent anomalies are not necessarily consistent, especially at higher latitude. This observed 

lack of mutual consistency at continental scales suggests that average snowpack depth variations 

may be of sufficiently large magnitude, spatial scope and temporal duration to influence regionalhemispheric 

climate, in a manner unrelated to the more extensively studied snow extent variations. 

Keywords: Snow depth, Snow extent, Continental-Scale, Consistency 

Datasets 

In order to study the temporal differences between snow depth and snow extent at regional to 

continental scales, datasets characterized by spatial coverage across North America and/or 20th 

century temporal duration have been acquired and utilized. 

• Gridded snow depth dataset spanning North America interpolated from 349 Stations 

located in the United States and southern Canada for the 1915–1997 period. (Brown, 

2000) 

• Daily 1 o x1 o grid dataset covering all of North America interpolated from a 

comprehensive set of station snow depth observations from 1956 through 2000. (Dyer 

and Mote, 2006) 

• SWE Maps for the Canadian Prairies derived using passive microwave data from the 

NASA Scanning Multichannel Microwave Radiometer (SMMR) and Special Sensor 

Microwave Imager (SSM/I) instruments from 1978 to 1997. (Derksen et al., 2004) 

Results 

The North America snow extent (i.e., spatial snow covered area) was computed by Ross Brown 

(Brown, 2000). The corresponding mean snow depth variable is computed as the area-weighted 

average snow depth over this snow covered extent. Monthly statistics are applied to assess the 

1 Department of Earth and Environmental Engineering, Columbia University, 500 W. 120 St., 

Room 918, MC4711, New York, NY. Corresponding author: Yan Ge, yg2124@columbia.edu.

mutual consistency of snow depth vs. snow extent over North America. Results from all three of 

the datasets described above consistently indicate that at continental/regional scales, snow depth 

and snow extent anomalies are not mutually consistent. 

Figure 1a shows that the monthly evolution of both mean and variability of snow depth and 

snow extent for North America follows different temporal patterns. Snow extent increases in the 

early snow season and reaches its maximum value in January, but interannual variability of 

monthly snow extent is small in January. On the contrary, both mean value and range of 

variability of monthly snow depth increases steadily through February/March. Another indication 

of different behavior between these two snow parameters is the monthly correlation between snow 

depth and snow extent, presented in Figure 1b. The interannual snow depth and snow extent 

timeseries are often poorly correlated, especially in winter. Figure 1c shows the coefficient of 

variation (i.e., ration of standard deviation to mean) for both extent and depth, and indicates that 

the interannual variability of snow depth comprises a larger percentage of its climatological mean 

state than that for snow extent, consistently across all months. Hence the magnitude of snow depth 

variability can be considered more substantial than the magnitude of snow extent variability. 

Monthly scatterplots (Figure 1d) support the correlation results, i.e., wide scatter and a lack of an 

apparent snow extent vs. snow depth relationship during winter, and a more noticeable but still 

modest relationship during autumn and spring. 

sne (10 6 km 2 ) 

snd(cm) 

corr(snd,sne) 

20 

18 

16 

14 

12 

10 

50 

40 

30 

20 

10 

0 

1 

0.8 

0.6 

0.4 

0.2 

Nov Dec Jan Feb Mar Apr 

month 


month 

0 


month 

a) b) 

c) d) 

Figure 1. Nov.–Apr. timeseries of North America snow extent “sne” (Brown, 2000) and grid domain average 

snow-covered depth “snd” (Dyer and Mote, 2006) from 1956 to 1997. a) Monthly box plots of interannual 

sne and snd; b) Monthly correlations between interannual sne and snd; c) Monthly cv (std/mean) of 

interannual sne and snd time series; d) Monthly scatterplots of interannual sne and snd. 

CV 

snd (cm) 

snd (cm) 

46 

0.2 

0.15 

0.1 

0.05 

0 


month 

15 

snd=0.81*sne-1.12 

R2 Nov. 

=0.2220 

10 

5 

10 15 20 

45 

snd=1.61*sne+4.36 

40 R2 Feb. 

=0.1011 

35 

30 

25 

10 15 20 

30 

snd=0.78*sne+6.09 

25 R2 Dec. 

=0.1052 

20 

15 

50 

snd=1.86*sne+1.76 

45 R 

40 

35 

30 

25 

20 

10 15 20 

2 =0.1530 

sne (10 6 km 2 10 

10 15 20 

sne (10 

Mar. 

) 

6 km 2 ) 

40 

snd=0.80*sne+13.05 

35 R2 Jan. 

=0.0269 

30 

25 

20 

10 15 20 

35 

snd=2.25*sne-9.23 

30 R2 Apr. 

=0.2508 

25 

20 

15 

snd 

sne 

10 

10 15 20

Figure 2 presents correlation maps between gridded snow depth (Dyer and Mote, 2006) and 

North America snow extent (Brown, 2000) from November to April. All the months show strong 

correlation near the snowline and weak correlation over virtually all points north of the snowline. 

Correlations drop off dramatically north of the snowline, resulting in a very large high-latitude 

region in which snow depth and snow extent are poorly correlated. During winter months there are 

even regions of negative correlation north of the snow line, which may be due to the interannual 

variation of northeastward mid-latitude cyclone tracks. 

30 ° N 

30 ° N 

30 ° N 

45 ° N 

45 ° N 

45 ° N 

60 ° N 

60 ° N 

60 ° N 

Nov. 

120 ° W 90 ° W 60 ° W 

Dec. 

120 ° W 90 ° W 60 ° W 

Jan. 

120 ° W 90 ° W 60 ° W 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

Figure 2. Correlation maps between gridded snow depth (Dyer and Mote, 2006) and North America snow 

extent (Brown, 2000). 

47 

30 ° N 

30 ° N 

30 ° N 

45 ° N 

45 ° N 

45 ° N 

60 ° N 

60 ° N 

60 ° N 

Feb. 

120 ° W 90 ° W 60 ° W 

Mar. 

120 ° W 90 ° W 60 ° W 

Apr. 

120 ° W 90 ° W 60 ° W 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4

Figure 3 shows the correlation maps between gridded SWE (Derksen et al., 2004) and North 

America snow extent (Brown, 2000). The satellite data confirms the observed inconsistency 

between snow depth and North America snow extent from station datasets seen in Figure 2. 

Correlations decline from south to north in central North America region (38N–63N) for three 

consecutive months (December, January and February). Regions of negative correlation also exist 

north of snowline during January and February. 

30 ° N 

30 ° N 

30 ° N 

45 ° N 

45 ° N 

45 ° N 

60 ° N 

60 ° N 

60 ° N 

Dec. 

120 ° W 90 ° W 60 ° W 

Jan. 

120 ° W 90 ° W 60 ° W 

Feb. 

120 ° W 90 ° W 60 ° W 

Figure 3. Correlation maps between gridded SWE (Derksen et al., 2004) and North America snow extent 

(Brown, 2000). 

48 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4

CONCLUSIONS 

Three regional-continental scale snow depth datasets have been evaluated, derived from station 

observational data or satellite remote sensing data. All of the datasets consistently exhibit differing 

behavior between interannual timeseries anomalies of average snow-covered depth and North 

America snow extent at regional/continental scales, generally consistent with what is expected at 

local scales. Correlation maps of snow depth vs. snow extent show a broad, continental-scale 

region is maintained throughout much of the snow season in which snow depth variations are not 

mutually consistent with North America snow extent variations. Research also recognized that 

relative magnitude of snow depth anomalies is considerably larger than that for snow extent. 

The observed differing behavior of the two snowpack parameters at regional/continental scales 

suggest that snow depth variations may be of sufficiently large magnitude, spatial scope and 

temporal duration to influence regional-hemispheric climate, in a manner unrelated to the more 

extensively studied snow extent variations. Such analysis represents the first step towards the goal 

of identifying explicit snow depth – climate relationships and integrating them with snow extent– 

climate relationships. 

REFERENCES 

Brown, R.D., 2000. Northern Hemisphere snow cover variability and change, 1915–1997. Journal 

of Climate 13: 2339–2355. 

Derksen, C., Brown, R. D. and Walker, A., 2004. Merging conventional (1915–1992) and passive 

microwave (1978–2002) estimates of snow extent and water equivalent over central North 

America. Journal of Hydrometeorology 5(5): 850–861. 

Dyer, L.J. and Mote,T.L., 2006. Spatial Variability And Trends In Snow Depth Over North 

America. Geophysical Research Letters, in press. 

49


50

51 



Climate Variability, Snowmelt Distribution, 

and Effects on Streamflow in a Cascades Watershed 

ANNE JEFFERSON 1 , ANNE NOLIN 1 , SARAH LEWIS 1 , MEREDITH PAYNE 2 , 

GORDON GRANT 3 , AND CHRISTINA TAGUE 4 


Cascades Range rivers provide critical water supply for agriculture, ecosystems, and 

municipalities in the Pacific Northwest, and they derive much of their water from snowmelt 

filtered through groundwater aquifers. Recent analyses show that this region is particularly 

sensitive to current and projected climate warming trends, specifically reduced snow accumulation 

and earlier spring melt (Mote, 2003; Stewart et al., 2005). By 2050, Cascades snowpacks are 

projected to be less than half of what they are today (Leung et al., 2004), potentially leading to 

major water shortages. Broad regional-scale characterizations identify climatic gradients as the 

most important controls on spatial variability in streamflow regimes, but the potential for other 

hydrological factors, particularly groundwater, to influence this response has received much less 

attention. Our objective is to develop an understanding of how discharge from a groundwaterdominated 

watershed is controlled at the event, seasonal, and interannual scales by snowpack 

dynamics, antecedent conditions, and global climate signals. 

The study watershed is that of the McKenzie River at Clear Lake (Figure 1), in the central 

Oregon Cascades, includes extensive areas of high permeability Quaternary (High Cascade) 

basalts that result in a substantial groundwater system, as well as runoff-dominated Tertiary 

(Western Cascades) landscapes (Sherrod and Smith, 2000; Tague and Grant, 2004). This 239 km 2 

watershed has long-term records of streamflow from United States Geological Survey gage 

#14158500. It also has record of precipitation, snow, and temperature from three Natural 

Resources Conservation Service SNOTEL sites: Hogg Pass (1451 m, 21E05S), Santiam Junction 

(1143 m, 21E06S) and Jump Off Joe (1067 m, 22E07S). Annual precipitation in the watershed 

ranges from ~1.8 to 3 m, and 70% falls between November and March. 47% of the watershed lies 

between 918 and 1200 m, in the transient snow zone. From 1200 m to the peak elevation (2051 

m), seasonal snowpacks occur from November through June. Peak snow water equivalent (SWE) 

occurs around April 1 st at Hogg Pass, and around March 1 st at Santiam Junction and Jump Off Joe. 

In order to examine relationships between hydrological variables in space and time, we 

performed Pearson’s correlations, autocorrelations, and cross-correlations using 42 parameters 

derived from discharge, precipitation, SWE, and temperature from the stations listed above. We 

also correlated discharge and SWE with monthly values of the Niño 3.4 index of sea surface 

1 Department of Geosciences, Oregon State University, 104 Wilkinson Hall, Corvallis OR 

97331. 

2 College of Oceanic and Atmospheric Sciences, Oregon State University, 104 COAS 

Administration Bldg., Corvallis, OR 97331. 

3 USDA Forest Service, Pacific Northwest Research Station, 3200 SW Jefferson Way, 

Corvallis, OR 97331. 

4 Bren School of Environmental Science and Management, University of California, Santa 

Barbara, 93106.

temperature (Trenberth and Stepaniak, 2001). A simple water balance was constructed for the 

2001–2004 water years, using values for the average basin elevation (1215 m) interpolated from 

precipitation and SWE values at Jump Off Joe and Hogg Pass. Basin-averaged evapotranspiration 

was calulcated in RHESSys (Tague and Band, 2004). Predictive models of September–November 

minimum discharge were developed using stepwise regression in SAS 9.1. 

Fluctuations in discharge are muted relative to daily variability in the recharge (rain plus 

snowmelt) signal (Figure 2). Summer streamflows are sustained by groundwater, not snowmelt. 

There is a high degree of discharge auto-correlation for ~2.5 months, and there is a strong crosscorrelation 

between the previous year’s precipitation and the current year’s discharge at a 1 year 

lag. The El Niño-Southern Oscillation is a reasonably good predictor of SWE and a moderate 

predictor of annual discharge. Internanual variability in the 26-year SNOTEL record masks any 

long-term trends in precipitation or SWE, but the longer discharge records suggests that climate 

warming is altering the streamflow regime at Clear Lake. The hydrograph temporal center of mass 

(Stewart et al., 2005) has moved earlier by a statistically significant 15 days since 1950, which is 

probably a function of relatively more winter rain and earlier snowmelt. This finding is in line 

with other watersheds throughout the mountainous west (Stewart et al., 2005). September– 

November minimum flows have declined since 1947, probably as a result of longer summer 

recession periods resulting from earlier snowmelt. 

Clear Lake 

watershed 

Figure 1. Location map of the study watershed, relative to the topography of Oregon and extent of High 

Cascades geology. 

52

Discharge (mm) 

20 

18 

16 

14 

12 

10 

8 

6 

4 

2 

0 

10/1/03 2/1/04 6/1/04 10/1/04 2/1/05 6/1/05 

Figure 2. Discharge at Clear Lake for water years 2004 and 2005 (bottom, black line), compared to the 

interpolated rain + snowmelt time series for the median basin elevation (top, blue line). 

The main conclusions of this study are: 

1. In the study watershed, pronounced seasonal variability of water inputs is damped by 

extensive groundwater systems. The study area may serve as a model for other 

groundwater-dominated watersheds in the mountainous west. 

2. Delays between precipitation and discharge are a function of snowpack storage and 

slow release of groundwater. From July to October, streamflow is sustained by 

groundwater. 

3. Groundwater-dominated watersheds are somewhat buffered from


This work was funded by the Institute for Water and Watersheds at Oregon State University, 

through the USGS water resources research institute program, and by the Eugene Water and 

Electric Board. This material is based on work supported under a National Science Foundation 

Fellowship. 

REFERENCES 

Leung LR, Qian Y, Bian X, Washington W, Han J, Roads JO. 2004. Mid-century ensemble 

regional climate change scenarios for the western United States. Climatic Change 62: 75–113. 

Mote PW. 2003. Trends in snow water equivalent in the Pacific Northwest and their climatic 

causes. Geophysical Research Letters 30: 1601, doi:10.1029/2003GL017258. 

Sherrod DR, Smith JG. 2000. Geologic Map of Upper Eocene to Holocene Volcanic and Related 

Rocks of the Cascade Range, Oregon. U.S. Geological Survey: Washington; 

Stewart IT, Cayan DR, Dettinger MD. 2005. Changes toward earlier streamflow timing across 

western North America. Journal of Climate 18: 1136–1155. 

Tague CL, Band LE. 2004. RHESSys: Regional Hydro-Ecologic Simulation System—An objectoriented 

approach to spatially distributed modeling of carbon, water, and nutrient cycling. 

Earth Interactions 8: Paper No. 19, p. 1–42. 

Tague C, Grant GE. 2004. A geological framework for interpreting the low flow regimes of 

Cascade streams, Willamette River Basin, Oregon. Water Resources Research 40: W04303 

10.1029/2003WR002629. 

Trenberth KE, Stepaniak DP. 2001. Indices of El Niño evolution. Journal of Climate 14: 1697– 

1701. 

54

ABSTRACT 

55 



Synoptic Patterns Associated with the Record Snowfall 

of 1960 in the Southern Appalachians 

L. BAKER PERRY 1 AND CHARLES E. KONRAD II 2 

Record snowfall totals of up to 211 cm blanketed the Southern Appalachians between mid- 

February and late-March 1960. Snow was reported on average every other day in the higher 

elevations, and mean temperatures for the period were nearly 6 C below normal. Snow piled up to 

great depths, with Boone, NC, reporting a maximum depth of 112 cm. Snow drifts buried roads 

and made travel impossible, requiring food, fuel, and hay to be airlifted into the region. This paper 

analyzes the synoptic patterns associated with the record snowfall of February and March 1960. 

Snowfall events are identified using a combination of first order hourly observations and 

cooperative observer daily snowfall totals. The spatial patterns of snowfall are mapped using a 

GIS, while mean values for various synoptic fields (e.g. 850 hPa temperature, 500 hPa height) are 

calculated and compared to 50-year climatological means. Snowfall events during this period are 

then classified according to the pattern of cyclogenesis and prevailing flow direction. 

Keywords: Synoptic patterns, snowfall, 1960, Southern Appalachians 

INTRODUCTION 

February and March 1960 continue to be remembered as the snowiest period on record in the 

Southern Appalachians. Snow was nearly a daily occurrence between 13 February and 26 March 

at higher elevations, with Boone, NC, reporting 211 cm and many other locations in excess of 175 

cm. These snowfall totals are considerably greater than the current 30-year mean annual snowfall 

of 102 cm for Boone and approach the mean annual values of nearly 250 cm for the 2,000 m peaks 

(e.g Mt. Leconte and Mt. Mitchell) of the Southern Appalachians. Mean temperatures for the 

period were nearly 6 ºC below normal (Hardie 1960). The combination of frequent snowfall and 

low temperatures allowed the snow to pile up to great depths, with Boone, NC, reporting a 

maximum depth of 112 cm on 13–14 March (NCDC 2002). Considerable blowing and drifting of 

snow compounded problems, closing roads and requiring emergency distribution of food, fuel, and 

hay (Figs. 1 & 2). Although the impacts were greatest in the Southern Appalachians, the entire 

Eastern U.S. was adversely affected. March alone broke more records for cold and snow across the 

Eastern U.S. than any other March on record until then (Ludlum 1960a, Ludlum 1960b). It 

remains the coldest March on record for many locations in the Southern Appalachians (NCDC 

2002. 

1 

Department of Geography and Planning, Box 32066, Appalachian State University, Boone, 

NC 28608 (perrylb@appstate.edu) 

2 

Department of Geography, University of North Carolina, Chapel Hill, NC 27599

Figure 1. Headlines from the Watauga Democrat (Boone, NC) 10 Mar 1960 (Watauga 1960). 

Figure 2. Snow depth on 17 Mar 1960 in Ashe County, NC (Photo courtesy of the NC DOT). 

56

This paper analyzes the synoptic patterns associated with the record snowfall of February and 

March 1960 across the Southern Appalachians (Fig. 3). Three major questions serve to guide the 

research. 1) What were the spatial patterns of snowfall during the record-breaking period between 

13 Feb and 26 Mar 1960? 2) What were the synoptic field values and map patterns for 1000 hPa 

height, 500 hPa height, and 850 hPa temperature during this period and how anomalous were these 

values? 3) What synoptic patterns were primarily responsible for the snowfall, and how did their 

relative importance vary across the region? 

DATA AND METHODS 

Figure 3. Location of study area. 

Daily snowfall records for February and March 1960 served as the source of snowfall data. We 

extracted these data from the National Climatic Data Center’s Cooperative Summary of the Day 

CD-Rom (NCDC 2002) for 121 cooperative observer (COOP) stations in the Southern 

Appalachian Region (Fig. 4). We restricted our analyses to 43 COOP stations that had complete 

coverage during the study period. 

We defined a snowfall event as having occurred if at least one COOP station in the region 

reported snow accumulation on a given date. To improve the temporal resolution of each event, 

we referenced hourly surface observation summaries from four nearby first order stations (Fig. 4). 

From interpretation of these data, we were able to approximate the onset, maturation, and ending 

time as well as the duration of reported snowfall across the region. Assessment of the maturation 

time involved determining the hour in which the spatial extent of snowfall was greatest across the 

network of first-order stations. An event remained active if precipitation was reported during a sixhour 

period. When precipitation was no longer reported at any of the four first-order stations for 

more than six hours, we defined the event as having ended at the hour precipitation was last 

reported. Using this approach, we identified 17 snowfall events during the 43-day period between 

13 February and 26 March 1960. 

57

Figure 4. Topography of the region and weather stations used in this study. 

We used gridded (2.5 by 2.5 degree latitude/longitude mesh), twice-daily synoptic fields that 

were extracted from CDs containing the National Center for Environmental Prediction (NCEP) 

Reanalysis dataset (Kalnay et al. 1996) to analyze the synoptic patterns during the period. These 

fields were spatially interpolated to the center of each snowfall event. Using the 0000 and 1200 

UTC gridded synoptic fields, we undertook a temporal interpolation to estimate field values 

during the event maturation time. We employed an inverse distance technique to carry out all 

spatial and temporal interpolations. Synoptic Climatology Suite (Konrad and Meaux 2003) was 

used for compositing of synoptic fields and calculation of anomalies. 

Each of the 17 snowfall events identified during the period was classified manually into one of 

five synoptic types (Table 1) according to the synoptic patterns identified from twice-daily surface 

analyses (NOAA 1960) and 850 hPa wind components (Kalnay et al. 1996). Miller Type A and 

Miller Type B cyclones are responsible for most of the big snowstorms across the Southern 

Appalachians, mid-Atlantic, and into the northeastern U.S. (Miller 1946, Kocin and Ucellini 1990, 

Keeter et al. 1995, Mote et al. 1997). In fact, all of the 20 major snowstorms in the northeastern 

U.S. analyzed by Kocin and Ucellini (1990) were of the Miller Type A or B variety. Miller Type 

A cyclones are characterized by the development of a surface cyclone along a frontal boundary in 

the Gulf of Mexico separating cold continental air from maritime tropical Gulf or Atlantic air. The 

surface low tracks northeastward out of the Gulf of Mexico, paralleling the Atlantic coastline and 

in some cases intensifying further. Miller Type B cyclones, however, initially track west of the 

Appalachians. As the primary low dissipates in the Ohio Valley, a secondary low develops along 

the Atlantic coast. 

Table 1. Major synoptic types observed during February and March 1960. 

Class Synoptic Type 

1 Miller Type A 

2 Miller Type B 

3 Ohio Valley 

4 Northwest Flow 

5 Other 

58

Cyclones that track west of the Appalachians through the Ohio Valley and do not undergo 

secondary cyclogenesis along the Atlantic coast constitute a third synoptic type responsible for 

producing snowfall during the period. These Ohio Valley cyclones (Knappenberger and Michaels 

1993) typically result in minor snowfall accumulations as most of the region remains in the warm 

sector until the cold front passes. Northwest Flow Snow (NWFS) events result from the 

orographic ascent of a cold and moist low-level northwest flow, often in the absence of synopticscale 

lifting (Perry and Konrad 2005, Perry and Konrad 2006, Perry et al. 2006). In this study, 

NWFS were defined on the basis of a northwest (270 to 360 degrees) 850 hPa wind at event 

maturation. Other types of events (e.g. Alberta clippers, prolonged isentropic lift) were also 

classified during the period, but contributed minimally to snowfall totals. 


The greatest snowfall totals between 13 February and 26 March 1960 occurred in the North 

Carolina High Country, where Boone recorded 211 cm (Fig. 5). The 265 cm of snowfall recorded 

during the 1959–60 snow season in Boone still stands as the annual record, and only one other 

year (1967–68 at 215 cm) received more than the 211 cm that fell during the 43-day period in 

1960. All of the higher elevations in the Southern Appalachians received considerable snowfall 

during the period, with many locations measuring two or three times as much snow during this 

period than they average in an entire year. As previously reported, the greatest snow depth 

measured was 112 cm, also in Boone, NC. The sustained cold, coupled with significant blowing 

and drifting snow, greatly increased the societal impacts. Residents would awake to find freshly 

plowed roads buried once again in newly drifted snow (Minor 1960). 

Figure 5. Total snowfall between 13 Feb and 26 Mar 1960. 

59

Figure 6. 500 hPa height (m) and anomaly for 13 Feb to 26 Mar 1960 (reference period 1951–2000). 

Figure 7. 850-hPa temperature anomaly (ºC) for 13 Feb to 26 Mar 1960 (reference period 1951–2000). 

60

The synoptic-scale atmospheric circulation was highly anomalous during this period. The 500hPa 

height, for example, was considerably below the 50-year mean for the Southern Appalachians 

(–2.29 z) and throughout eastern North America. A 500 hPa trough dominated eastern North 

America throughout this period, providing an active, southeasterly displaced storm track that 

allowed frequent intrusions of cold continental air (Fig. 6). The 850 hPa temperature anomaly was 

even more extreme, registering nearly 8.0 C cooler (–3.46 z) than the 50-year mean over northern 

portions of the study area (Fig. 7). Again, the highly anomalous values were not restricted to the 

Southern Appalachians, but covered the entire eastern U.S. As a result, March 1960 was the 

coldest March on record across much of the eastern U.S. (Ludlum 1960b). 

Three primary synoptic types contributed the bulk of the snowfall across the Southern 

Appalachians (Fig. 8). The majority of the snowfall across the region occurred in association with 

Miller Type A and Miller Type B cyclones (Fig. 9). The two Miller Type A cyclones occurred in 

February, whereas the three Miller Type B storms moved by in March. Interestingly, the greatest 

accumulations in association with the Miller Type A cyclones were in the western and northern 

portions of the study area, while areas in the southeast of the study area received mostly rain 

(NCDC 2002). Miller Type B cyclones contributed the greatest amount to snowfall totals during 

the period at all but two of the regions across the study area. In the south and east, these three 

storms contributed nearly all of the snowfall, occurring during a two-week period in early March. 

In the higher elevations and along northwestern slopes, NWFS contributed substantially to 

snowfall totals, constituting as much as 35 percent of the total. Interestingly, none of the NWFS 

events were tied to 500 hPa cutoff lows, which were responsible for the extremely heavy spring 

snow events in April 1987 and May 1992 (NWS 1987, Fishel and Businger 1993). The NWFS 

events were more numerous than the Miller Type A and B cyclones, but the snowfall was limited 

in spatial extent. Mainly as a result of the NWFS events, the cumulative percentage of total 

snowfall during the period contributed by each event displayed considerable variability across the 

Southern Appalachians (Fig. 10). At lower elevations, particularly in the southeast, only two or 

three exceptionally heavy events occurred during the period. Elsewhere, and particularly at higher 

elevations and along northwestern slopes, lighter NWFS events were interspersed with the big 

Miller Type A and B cyclones, reflecting a greater diversity of synoptic types responsible for the 

snowfall totals. 

Figure 8. Three primary synoptic patterns responsible for the record snowfall and associated dates. 

61

Figure 9. Total snowfall between 13 Feb and 26 Mar 1960 by synoptic pattern. 

Figure 10. Cumulative percentage of total snowfall contributed by each event 

(from heaviest to lightest) and number of events needed to reach 90%. 

62

CONCLUSIONS 

The record snowfall in February and March 1960 across the Southern Appalachians occurred in 

association with highly anomalous 500 hPa circulation and 850 hPa temperatures. The persistence 

of a 500 hPa trough over the eastern U.S. was associated with a southeasterly displaced storm 

track that kept temperatures well below normal. As a result, the 1959–60 snow season still holds 

the annual snowfall record (265 cm) and maximum snow depth record (112 cm) for Boone, NC, 

and countless other locations across the region. Miller Type A cyclones, Miller Type B cyclones, 

and NWFS events were responsible for nearly all of the total snowfall across the region. However, 

NWFS made the greatest contribution to snowfall totals across the higher elevations and along 

northwestern slopes, where snow fell on average every other day during the period. A small 

number of exceptionally heavy events produced the record snowfall in the Georgia and Carolina 

foothills in contrast with the elevated areas to the north and west where numerous, relatively light 

NWFS events contributed substantially to the total snowfall. Unanswered questions include the 

causes of the anomalous and persistent 500 hPa circulation pattern. In particular, the role of 

hemispheric teleconnections versus persistent snow cover across the region in forcing this pattern 

needs to be explored. The influence of favorable air trajectories extending downwind from the 

Great Lakes in periods of NWFS also deserves further scrutiny, particularly since these have been 

tied to enhanced snowfall in the Southern Appalachians (Perry and Konrad 2005). Lastly, given 

the deep snowpack and reports of flooding during rapid melting at the end of March (Ludlum 

1960b), an investigation of the hydrological significance (e.g. SWE estimations and flooding 

reports) of February and March 1960 is also warranted. 

ACKNOWLEDGMENTS 

Peter Robinson, Tom Whitmore, Walt Martin, Tom Schmidlin, Laurence Lee, and Michael 

Mayfield provided valuable feedback on the design of the study. Julie Whichard from the North 

Carolina Department of Transportation kindly provided archived photos. Addie Mae Love, Louiva 

Ward, and Howell Cooke shared their many memories. 

REFERENCES 

Fishel, G.B., and S. Businger. 1993. Heavy orographic snowfall in the southern Appalachians: a 

late season case study. Postprints, Third National Heavy Precipitation Workshop: 275 – 284. 

Pittsburgh, PA: NWS/NOAA. 

Hardie, A. V. 1960. Special Weather Summary, in Climatological Data, North Carolina: March 

1960. Asheville, NC: National Climatic Data Center. 

Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, and 18 others. 1996. The NCEP / NCAR 40year 

reanalysis project. Bulletin of the American Meteorological Society 77: 437–471. 

Keeter, K.K., S. Businger, L.G. Lee, J.S. Waldstreicher. 1995. Winter weather forecasting through 

the eastern United States. Part III: The effects of topography and the variability of winter 

weather in the Carolinas and Virginia. Weather and Forecasting 10: 42–60. 

Knappenberger, P.C., and P.J. Michaels. 1993. Cyclone tracks and wintertime climate in the mid- 

Atlantic region of the USA. International Journal of Climatology 13: 509–531. 

Kocin, P.J, and L.W. Uccelini. 1990. Snowstorms along the Northeastern Coast of the United 

States: 1955 – 1985. Meteorological Monograph No. 44, American Meteorological Society. 

Konrad, C.E., and D.Meaux. 2003. Synoptic Suite Software Package. Chapel Hill, NC: University 

of North Carolina. 

Ludlum, D. 1960a. The great early March snowstorm of 1960. Weatherwise 13: 59–62. 

Ludlum, D. 1960b. The great temperature reversal: March to April. Weatherwise 13: 120–128. 

Miller, J.E. 1946. Cyclogenesis in the Atlantic Coastal region of the United States. Journal of 

Meteorology 3: 31–44. 

63

Minor, J. 1960. Watauga in Disaster Area: Red Cross, Guard Units Act to Aid Snow Victims. 

Watauga Democrat, March 17, 1960, p. 1. 

Mote, T.L., D.W. Gamble, S. J. Underwood, M.L. Bentley. 1997. Synoptic-scale features common 

to heavy snowstorms in the Southeast United States. Weather and Forecasting 12: 5–23. 

NWS. 1987. Snowstorm in the Appalachian Region on April 2–5, 1987. Storm Data 29: 6–14. 

NCDC. 2002. Cooperative Summary of the Day: Eastern U.S. Asheville, NC: National Climatic 

Data Center. 

NOAA. 1960. Daily Weather Maps, Weekly Series. Washington, D.C.: Department of Commerce, 

Environmental Data Service. 

Perry, L. Baker, Charles E. Konrad. 2005. The Influence of the Great Lakes on Snowfall Patterns 

in the Southern Appalachians. Proceedings of the 62 nd Eastern Snow Conference: 279–289. 

Perry, L. Baker, and Charles E. Konrad. 2006. Relationships between NW flow snowfall and 

topography in the Southern Appalachians, USA. Climate Research 32: 35–47. 

Perry, L. Baker, Charles E. Konrad, Thomas W. Schmidlin. 2006. Antecedent upstream air 

trajectories associated with northwest flow snowfall in the Southern Appalachians, USA. 

Weather and Forecasting, in press. 

Schmidlin, T.W. 1992. Does lake-effect snow extend to the mountains of West Virginia? 

Proceedings of the 49 th Eastern Snow Conference: 145–148. 

Watauga Democrat. 1960. Watauga Digging Out Of Mountainous Snowfall: Crews Work Round 

Clock to Open Roads. Watauga Democrat, March 10, 1960. 

64

65 



Influence of Snowfall Anomalies on Summer Precipitation 

in the Northern Great Plains of North America 

ABSTRACT 

STEVEN M. QUIRING 1 , AND DARIA B. KLUVER 2 

Using observations from 1929 to 1999, we examine the relationship between winter/spring 

snowfall anomalies and summer precipitation over the northern Great Plains of North America. 

Both composite and correlation analysis indicate that anomalously dry (wet) summers are 

associated with negative (positive) snowfall anomalies during the preceding winter and spring. It 

is posited that below (above) normal snowfall is associated with decreases (increases) in 

spring/early summer soil moisture and associated decreases (increases) in local moisture recycling 

during summer. It appears that the snowfall anomalies must exceed some minimum threshold 

before they have a significant impact on atmospheric circulation and precipitation during the 

following summer. There is also significant temporal variability in the strength of the correlations 

between snowfall and summer moisture. Relationships between April–May snowfall and summer 

moisture were generally quite strong between 1929 and 1954 and 1970 to 1987, but were 

relatively weak during 1955 to 1969 and after 1987. This suggests that other factors may be 

modulating the importance of land surface processes. 

Keywords: snowfall, precipitation, drought, Great Plains 

INTRODUCTION 

Land surface conditions (e.g., snow cover, soil moisture) are important sources of seasonal 

climate predictability (Koster and Suarez, 2001; Koster et al., 2003; Koster et al., 2004). 

Numerous studies have demonstrated that Eurasian/Tibetan snow cover influences Indian/Asian 

monsoonal circulation and precipitation (Kripalani et al., 2002; Robock et al., 2003; Wu and Qian, 

2003; Fasullo, 2004; Zhang et al., 2004). Snow cover and snow water equivalent have also been 

linked to variability in the North American Monsoon (Gutzler, 2000; Ellis and Hawkins, 2001; 

Hawkins et al., 2002; Lo and Clark, 2002; Matsui et al., 2003). Other studies have identified 

connections between Eurasian snow cover extent and summer air temperature in the United 

Kingdom (Qian and Saunders, 2003), and Canadian river discharge (Déry et al., 2005). Although 

both the presence and amount of snow can have a significant impact on the climate of local and 

remote regions, our understanding of the relationship between snow cover and climate is still 

incomplete. This is especially the case in the northern Great Plains of North America where there 

are a paucity of studies examining the relationship between snowfall/snow cover and summer 

precipitation. 

While Koster et al. (2004) found a strong coupling between soil moisture and precipitation in 

the Great Plains, to date no observational studies have examined the relationship between 

winter/spring snowfall (which contributes to soil moisture recharge) and summer drought 

1 Department of Geography, Texas A&M University, College Station, Texas, USA 

2 Department of Geography, University of Delaware, Newark, Delaware, USA

conditions. Therefore the utility of land surface conditions for making seasonal climate forecasts 

merits further attention in this region. This paper utilizes observational data (1929 to 1999) to 

examine the relationship between winter/spring snowfall anomalies and summer moisture 

(precipitation) anomalies in the northern Great Plains of North America. 


The northern Great Plains of North America, as defined in this study, include portions of three 

Canadian provinces (Manitoba, Saskatchewan, and Alberta) and 12 US states (Colorado, Iowa, 

Idaho, Minnesota, Missouri, Montana, Nebraska, North Dakota, Nevada, South Dakota, Utah, and 

Wyoming). The analysis is based on monthly drought index and snowfall data (1929–1999) that 

have been interpolated to a one-degree grid spanning 40° to 54° N; 95° to 113° W (18 grid cells in 

the northeastern corner of the study region were omitted due to inhomogenities in the data). 

Drought (moisture) indices are commonly used to quantify moisture conditions within a region, 

to detect the onset, and to measure the severity and spatial extent of drought events (Alley, 1984). 

The Moisture Anomaly Index (subsequently referred to as the Z-index) was developed by Palmer 

(1965) and it is calculated using a soil moisture/water balance algorithm. The Z-index represents 

the departure from normal (or climatically appropriate) moisture conditions in a given month, 

when the Z-index is positive (negative) conditions are wetter (drier) than normal (Palmer, 1965). 

The Z-index was selected to represent summer moisture anomalies in the northern Great Plains 

since previous research has shown that this index is well-suited for monitoring moisture (drought) 

conditions in this region (Quiring and Papakyriakou, 2003). Drought data for the US were 

provided by the National Climatic Data Center at the climate division level (available at: 

http://www.ncdc.noaa.gov) and then interpolated to a one-degree grid. Details on how the 

Canadian Z-index data were generated can be found in Quiring and Papkyriakou (2003, 2005). 

The snowfall data were developed by T. Mote and collaborators at the University of Georgia (T. 

Mote, 2004, personal communication). It is based on daily observations from the National 

Weather Service cooperative station network and the Canadian Daily surface observations. The 

data were interpolated to a one-degree latitude by one-degree longitude grid using Spheremap, a 

spatial interpolation program. 

RESULTS 

Only a weak linear relationship exists between snowfall and summer moisture when conditions 

are averaged over the study area (248 grid cells). Correlations are weakest between fall (SON) 

snowfall and summer moisture anomalies (0.05), they increase to 0.17 during winter (DJF) and 

spring (MAM). During the final two months of spring (April and May) the correlation improves to 

0.22. Averaging over space and time (1929 to 1999) masks significant spatial and temporal 

variability present in the snowfall-summer moisture relationship. 

Figure 1 shows the relationship between April–May snowfall and summer moisture anomalies 

in the northern Great Plains (1929–1999). It is clear that the nature of this relationship has varied 

over time. Linear correlations were calculated using a sliding 15 yr window (e.g., first correlation 

was calculated using 1929 to 1944, the second correlation was calculated using 1930 to 1945, and 

the last correlation was calculated using 1984 to 1999) (Figure 2). Correlations between April– 

May snowfall and summer moisture anomalies varied from 0.82 (1971–1985) to –0.01 (1955– 

1969). The relationship was quite strong in the 1930s/1940s and 1970s/1980s, but was relatively 

weak in the 1950s/1960s, and since the late 1980s. Correlations between winter snowfall and 

summer moisture anomalies varied from 0.58 (1954–1968) to –0.54 (1942–1956). During the 

early part of the record winter snowfall was negatively correlated with summer moisture. A 

significant shift in the relationship occurred during the 1950s and since the 1950s there has 

generally been a positive relationship between winter snowfall and summer moisture. The 

relationship between winter snowfall and summer moisture is different than the relationship 

between April–May snowfall and summer moisture. Typically when there are relatively strong 

66

correlations between April–May snowfall and summer moisture, there are weak (or negative) 

correlations between winter snowfall and summer moisture. 

A more detailed examination revealed that there is a stronger relationship between April–May 

snowfall and summer moisture during years that have large snowfall anomalies (snowfall 

anomalies that are more than one standard deviation above/below the mean). There are 11 yrs with 

April–May snowfall anomalies more than one standard deviation below the mean. Eight of these 

11 yrs were followed by drier than normal summers in the northern Great Plains and the mean Zindex 

for the 11 yrs is –0.44. There were also 11 yrs with April–May snowfall anomalies more 

than one standard deviation above the mean. Nine of these 11 yrs were followed by wetter than 

normal summers and the mean Z-index for the 11 yrs is 0.46. Based on all 22 yrs, the linear 

correlation between April–May snowfall anomalies and summer moisture is 0.49 (statistically 

significant at the 95% confidence level). 

Figure 1. April–May snowfall anomalies (blue line) and summer moisture anomalies (Z-index) (red line) 

averaged over the northern Great Plains study region (1929–1999). The two 15 yr periods with the highest 

(0.82) and lowest (–0.01) linear correlations are indicated. 

67

Figure 2. Linear correlations between winter (DJF) (green line) and April–May (blue line) snowfall 

anomalies and summer moisture anomalies (Z-index) calculated for all 15 yr time periods between 1929 and 

1999. Dashed lines indicate the 95% significance level. 

The relationship between April–May snowfall and summer moisture anomalies is further 

strengthened if the candidate years are restricted to those occurring during the two periods (1929– 

1954 and 1970–1987) when land surface conditions appear to be the dominant source of seasonal 

climate predictability. During these two periods there were 8 yrs with April–May snowfall 

anomalies that were more than one standard deviation below the mean. Seven of these 8 yrs were 

associated with drier than normal moisture conditions during the summer (mean Z-index = –0.73). 

There were also 8 yrs with April–May snowfall anomalies that were more than one standard 

deviation above the mean. Seven of the 8 yrs with large positive April–May snowfall anomalies 

were followed by wetter than normal moisture conditions during the summer (mean Z-index = 

0.63). Based on these 16 yrs the linear correlation between April–May snowfall and summer 

moisture is 0.70 (statistically significant at the 95% confidence level). 

Significant spatial variability is also evident in the relationship between April–May snowfall 

and summer moisture anomalies (Figure 3). Linear correlations across the study region are 

generally positive with the highest correlations (up to 0.63) in southern Manitoba and South 

Dakota. There is only one grid cell in Wyoming where there is a statistically significant negative 

correlation (–0.23). Statistically significant correlations between April–May snowfall and summer 

moisture anomalies are found in approximately 56% of the study area. These grid cells tend to be 

concentrated on the eastern side of the study region and are notably absent from most of 

Wyoming, the eastern part of Montana, and southern Saskatchewan. 

68

Figure 3. Linear correlations between April–May snowfall and summer moisture anomalies (Z-index). 

Colored grid cells are those with statistically significant correlations (95% significance level). 

Based on a composite analysis, the five driest summers between 1929 and 1999 (Table 1) are 

associated with a mean winter (spring) snowfall anomaly of –66.7 mm (–62.4 mm) (Figure 4). 

Approximately 85% of the study region received below normal snowfall during the winter and 

spring seasons prior to the five driest summers. About 28% (25%) of the study region had winter 

(spring) snowfall anomalies that are more than 100 mm below average and only 2% (1%) received 

winter (spring) snowfall that was more than 100 mm above normal. The five wettest summers 

between 1929 and 1999 were associated with a mean winter (spring) snowfall anomaly of 6.2 mm 

(21.6 mm). Approximately 53% (46%) of the study region received above normal snowfall during 

the prior winter (spring). About 15% (12.1%) of the study region had winter (spring) snowfall that 

was greater than 100 mm above normal and only 7% (1%) received winter (spring) snowfall that 

was more than 100 mm below normal. Results of the composite analysis indicate that anomalously 

dry (wet) summers are associated with significant negative (positive) snowfall anomalies during 

the preceding winter and spring, which supports the results of the correlation analysis. However, 

the composite analysis demonstrated that the winter/spring snowfall anomalies associated with the 

driest summers are typically greater in magnitude and more spatially extensive than the snowfall 

anomalies associated with wettest summers. 

69

Table 1. Ten driest and wettest summers (mean Z-index) between 1929 and 1999 (ranked by severity). 

Driest Years Wettest Years 

Year Z-Index Year Z-Index 

1961 –2.52 1993 4.00 

1936 –2.51 1944 1.95 

1988 –2.48 1951 1.63 

1934 –2.13 1965 1.62 

1931 –1.73 1995 1.54 

1933 –1.40 1947 1.52 

1929 –1.24 1999 1.36 

1940 –1.23 1942 1.33 

1959 –1.09 1975 1.30 

1937 –0.99 1968 1.28 

Figure 4. Composite snowfall anomalies (mm) in winter, and spring associated with the five wettest summers 

(1993, 1944, 1951, 1965, 1995) and the five driest summers (1961, 1936, 1988, 1934, 1931). 

DISCUSSION 

The observational data demonstrate that below (above) normal snowfall in winter/spring is 

generally associated with anomalously dry (wet) summers in the northern Great Plains. It is 

hypothesized that below normal snowfall is linked to summer drought via negative soil moisture 

anomalies in spring and early summer that reduce local moisture recycling. Our findings appear to 

support those of Namias (1991), who suggested that reduced soil moisture during the late 

70

winter/early spring could contribute to a warm, dry summer in the region by reducing the amount 

of local moisture recycling and by modifying the large-scale atmospheric circulation. 

The strength of the relationship between winter/spring snowfall and summer moisture anomalies 

has varied significantly over space and time. Linear correlations between April–May snowfall 

anomalies and summer moisture conditions ranged from approximately zero (1955–1969) to 0.82 

(1971–1985). Other empirical studies have also found that the strength of the relationship between 

land surface conditions (e.g., snow cover and soil moisture) and precipitation has varied 

significantly during the 20 th century (Gutzler, 2000; Hu and Feng, 2002; Zhu et al., 2005). Hu and 

Feng (2004) suggest that the relationship between land surface conditions and precipitation 

patterns over the North American Monsoon region is modulated by sea surface temperature (SST) 

anomalies in the Pacific Ocean. They found that when SST anomalies were strong (weak), land 

surface conditions tend to have less (more) influence. Therefore it is hypothesized that 

atmospheric and/or oceanic forcings are modulating the relationship between snowfall and 

summer moisture conditions in the northern Great Plains. However, the reasons for the differential 

influence of winter versus spring (April–May) snowfall (Figure 2) are unknown and merit future 

study. 

Relationships between April–May snowfall and summer moisture anomalies also varied 

spatially. The strongest relationships were found in southern Manitoba and South Dakota and 

statistically significant correlations were present across approximately 56% of the study region. 

Previous research has also demonstrated that the coupling between land surface conditions (e.g., 

soil moisture and snow) and precipitation can be highly spatially variable (Lo and Clark, 2002; 

Koster et al., 2004; Dominguez et al., 2006). Our results demonstrate that even within a relatively 

small area there can be substantial differences in the strength of the relationship between spring 

snowfall and summer moisture anomalies. 

The relationship between spring snowfall and summer moisture may be non-linear since it 

appears that snowfall anomalies must exceed some minimum threshold before they have a 

significant (and consistent) influence on summer moisture conditions. The mean correlation 

between April–May snowfall and summer moisture anomalies increased from 0.22 (all years) to 

0.49 when only the years with snowfall anomalies more than one standard deviation above/below 

the mean were considered. 

The lack of spatial and temporal stability in the relationship between snowfall and summer 

moisture anomalies has significant implications for understanding and forecasting the occurrence 

of severe hydrologic events (e.g., floods and droughts). Additional study is needed to identify the 

factors that are responsible for modulating the strength of the snowfall-summer moisture 

relationship over space and time. Although spring snowfall conditions can, in some cases, explain 

more than half of the variance in summer moisture, the lack of spatial and temporal stability in this 

relationship limits its utility for producing accurate forecasts of summer droughts in the northern 

Great Plains. 


A version of this paper has been submitted to Geophysical Research Letters. The authors would 

like to thank Tom Mote for providing the snowfall data and Dan Leathers for reviewing an earlier 

version of this manuscript. 

REFERENCES 

Alley WM. 1984. The Palmer Drought Severity Index: Limitation and assumptions. J. Clim. Appl. 

Meteorol. 23: 1100–1109. 

Déry SJ, Sheffield J, Wood EF. 2005. Connectivity between Eurasian snow cover extent and 

Canadian snow water equivalent and river discharge. J. Geophys. Res. 110: D23106, 

doi:10.1029/2005JD006173. 

71

Dominguez F, Kumar P, Liang XZ, Ting MF. 2006. Impact of atmospheric moisture storage on 

precipitation recycling. J. Clim. 19: 1513–1530. 

Ellis AW, Hawkins TW. 2001. An apparent atmospheric teleconnection between snow cover and 

the North American monsoon. Geophys. Res. Lett. 28: 2653–2656. 

Fasullo J. 2004. A stratified diagnosis of the Indian monsoon–Eurasian snow cover relationship. J. 

Clim. 17: 1110–1122. 

Gutzler DS. 2000. Covariability of spring snowpack and summer rainfall across the southwest 

United States. J. Clim. 13: 4018–4027. 

Hawkins TW, Ellis AW, Skindlov JA, Reigle D. 2002. Intra-annual analysis of the North 

American snow cover-monsoon teleconnection: Seasonal forecasting utility. J. Clim. 15: 

1743–1753. 

Hu Q, Feng S. 2002. Interannual rainfall variations in the North American summer monsoon 

region: 1900–98. J. Clim. 15: 1189–1202. 

Hu Q, Feng S. 2004. Why has the land memory changed? J. Clim. 17: 3236–3243. 

Koster RD, Dirmeyer PA, Guo Z, Bonan GB, Chan E, Cox P, Gordon CT, Kanae S, Kowalczyk E, 

Lawrence D, Liu P, Lu CH, Malyshev S, McAvaney B, Oleson K, Pitman AJ, Sud YC, Taylor 

CM, Verseghy D, Vasic R, Xue Y, Yamada T. 2004. Regions of strong coupling between soil 

moisture and precipitation. Science 305: 1138–1140. 

Koster RD, Suarez MJ. 2001. Soil moisture memory in climate models. J. Hydrometeorol. 2: 558– 

570. 

Koster RD, Suarez MJ, Higgins RW, Van den Dool HM. 2003. Observational evidence that soil 

moisture variations affect precipitation. Geophys. Res. Lett. 30(5): 1241, 

doi:10.1029/2002GL016571. 

Kripalani RH, Kim B-J, Oh J-H, Moon S-E. 2002. Relationship between Soviet snow and Korean 

rainfall. Int. J. Climatol. 22: 1313–1325. 

Lo F, Clark MP. 2002. Relationships between spring snow mass and summer precipitation in the 

southwestern United States associated with the North American monsoon system. J. Clim. 15: 

1378–1385. 

Matsui T, Lakshmi V, Small E. 2003. Links between snow cover, surface skin temperature, and 

rainfall variability in the North American monsoon system. J. Clim. 16: 1821–1829. 

Namias J. 1991. Spring and summer 1988 drought over the contiguous United States—Causes and 

prediction. J. Clim. 4: 54–65. 

Palmer WC. 1965. Meteorological drought. Research Paper No. 45, U.S. Weather Bureau: 

Washington, D.C. 

Qian B, Saunders MA. 2003. Summer U.K. temperature and its links to preceding Eurasian snow 

cover, North Atlantic SSTs, and the NAO. J. Clim. 16: 4108–4120. 

Quiring SM, Papakyriakou TN. 2003. An evaluation of agricultural drought indices for the 

Canadian prairies. Ag. For. Meteorol. 118: 49–62. 

Quiring SM, Papakyriakou TN. 2005. Characterizing the spatial and temporal variability of June– 

July moisture conditions in the Canadian prairies. Int. J. Climatol. 25: 117–138. 

Robock A, Mu M, Vinnikov K, Robinson D. 2003. Land surface conditions over Eurasia and 

Indian summer monsoon rainfall. J. Geophys. Res. 108(D4): 4131, 

doi:10.1029/2002JD002286. 

Wu T, Qian Z. 2003. The relation between the Tibetan winter snow and the Asian summer 

monsoon and rainfall: An observational investigation. J. Clim. 16: 2038–2051. 

Zhang Y, Li T, Wang B. 2004. Decadal change of the spring snow depth over the Tibetan plateau: 

The associated circulation and influence on the east Asian summer monsoon. J. Clim. 17: 

2780–2793. 

Zhu CM, Lettenmaier DP, Cavazos T. 2005. Role of antecedent land surface conditions on North 

American monsoon rainfall variability. J. Clim. 18: 3104–3121. 

72

Snow Remote Sensing 

73


74

75 



Estimating Sublimation of Intercepted and Sub-Canopy Snow 

Using Eddy Covariance Systems 

NOAH MOLOTCH 1 , PETER BLANKEN 2 , MARK WILLIAMS 2,3 , ANDREW TURNIPSEED 4 , 

RUSSELL MONSON 5 , AND STEVEN MARGULIS 1 

ABSTRACT: 

Direct measurements of winter water loss due to sublimation were made in a sub-alpine forest in 

the Rocky Mountains of Colorado. Sub-canopy and over-story eddy covariance systems indicated 

substantial losses of winter-season snow accumulation in the form of snowpack (0.41 mm d –1 ) and 

intercepted snow (0.71 mm d –1 ) sublimation. The partitioning between these over and under story 

components of water loss was highly dependent on atmospheric conditions and near-surface 

conditions at and below the snow / atmosphere interface. High over-story sensible heat fluxes lead 

to strong temperature gradients between vegetation and the snow-surface, driving substantial 

specific humidity gradients at the snow surface and high sublimation rates. Intercepted snowfall 

resulted in rapid response of over-story latent heat fluxes, high within-canopy sublimation rates, 

and diminished sub-canopy snowpack sublimation. These results indicate that sublimation losses 

from the under-story snowpack are strongly dependent on the partitioning of sensible and latent 

heat fluxes in the canopy. This compels comprehensive studies of snow sublimation in forested 

regions that integrate sub-canopy and over-story processes. 

Keywords: vegetation canopy; snow interception; sublimation; Rocky Mountains; eddy covariance 

INTRODUCTION 

Sublimation of intercepted snow constitutes a significant component of the overall water 

balance in many seasonally snow-covered coniferous forests [Essery, et al., 2003; Lundberg and 

Halldin, 1994; Pomeroy and Gray, 1995; Schmidt and Troendle, 1992]; sublimation losses are 

capable of exceeding 30% of total winter snowfall [Montesi, et al., 2004]. For a given canopy 

structure and snowfall history the distribution of radiant and turbulent fluxes dictates sublimation 

rates and therefore strongly influences the magnitude of spring snowmelt and subsequent growingseason 

water availability. Interactions between these fluxes and the sublimation of intercepted 

snow and the sub-canopy snowpack are poorly understood in forested mountainous regions [Bales, 

et al., 2006]. This knowledge gap and the complexity of interactions between the snowpack and 

vegetation have motivated detailed analyses of mass and energy fluxes between the snowpack, 

vegetation, and the atmosphere [Davis, et al., 1997; Sicart, et al., 2004]. 

1 

Department of Civil and Environmental Engineering, University of California, Los Angeles, 

California, 90095. 

2 

Department of Geography, University of Colorado, Boulder, Colorado, 80309. 

3 

Institute for Arctic and Alpine Research, University of Colorado, Boulder, Colorado 80309. 

4 

National Center for Atmospheric Research, Boulder, Colorado, 80305. 

5 

Department of Ecology and Evolutionary Biology; Cooperative Institute for Research in 

Environmental Sciences, University of Colorado, Boulder, Colorado 80309

Various techniques have been used to estimate sublimation rates from intercepted snow. 

Measurement of the components of snow sublimation is particularly challenging in forested terrain 

as winter-time above-canopy water vapor flux measurements integrate mass loss from intercepted 

snow and from the sub-canopy snowpack. In this regard, numerous studies have focused on 

estimating sublimation losses from snowpacks in unforested areas. Similarly, much work has been 

devoted toward estimating sublimation losses from intercepted snow [Montesi, et al., 2004; 

Pomeroy and Schmidt, 1993; Schmidt and Troendle, 1992]. Lacking is a thorough analysis of the 

proportion of these two different components of snow sublimation at an individual site. 

Measurement of sublimation from intercepted snow has primarily focused on tree-weighting 

techniques [Montesi, et al., 2004; Nakai, et al., 1994; Schmidt, 1991; Schmidt, et al., 1988]. 

Several factors lead to uncertainty in this approach and toward limiting applicability at the stand 

scale. First, a somewhat subjective analysis must be used to separate unloading from sublimation. 

Second, sublimation of unloaded snow is not considered and thus sublimation losses may be 

underestimated [Montesi, et al., 2004]. Third, tree-instability can cause false readings. Finally, 

intermittent snowfall events and small trace events can introduce uncertainty, effectively 

countering sublimation losses and leading to underestimates in sublimation losses if not 

considered. In terms of scaling from individual trees to the stand scale, challenges are encountered 

with regard to the lack of detailed canopy information. This lack of detailed canopy information 

also complicates the use of models for estimating sublimation losses [Pomeroy, et al., 1998; 

Pomeroy and Schmidt, 1993]. All of these limitations could be accounted for in techniques that 

integrate all of these processes by measuring above and below canopy water vapor flux. 

Advances in process-level knowledge have been limited as sublimation can occur either from 

snow intercepted by the canopy, and/or from the snow that reaches the ground. Coniferous forests 

can intercept large quantities of snow, much of which sublimates from the canopy and does not 

reach the ground. Sublimation from the below-canopy snowpack is thought to be insignificant due 

to the low exposed surface area of the snowpack and low below-canopy wind speeds. However, 

there are potentially large longwave radiation fluxes if the canopy above is warm and snow-free, 

thus promoting sublimation and/or melting [Woo and Giesbrecht, 2000]. Understanding the 

balance between sublimation from the canopy and snowpack is crucial to assist water and forest 

managers, especially in regions where forest thinning treatments are being considered to increase 

water yield. 

Direct measurements of winter water loss by sublimation of snow from a subalpine forest in the 

Rocky Mountains of Colorado are presented here. Eddy covariance instruments were placed both 

above and beneath the canopy during March and early April 2002; the time before melting begins 

when winter sublimation is thought to be large due to the heavy late-winter snows. The above and 

below-canopy measurements allowed sublimation of intercepted snow to be separated from that of 

the snowpack, and estimates obtained over a much larger sample area than individual trees. 

Simultaneous measurements of the physical properties of the snow pack, soil moisture, as well as 

carbon dioxide flux measurements ensured that sublimation and not evaporation of melting snow 

or transpiration were being measured. The specific objectives of this research were to: a) 

determine snow sublimation rates in a sub-alpine forest; b) partition snow sublimation into above 

and below canopy components; and c) explore relationships between atmospheric and snowpack 

conditions, and snow sublimation rates. 

STUDY SITE 

This work was conducted at the Niwot Ridge, Colorado Ameriflux site (40º 1’ 58”N; 105º 32’ 

47” W), located at an elevation of 3050 m approximately 8 km east of the Continental Divide 

(Figure 1). The area 1 km 2 east of the tower is dominated by Engelmann spruce (7 trees ha –1 ) and 

lodgepole pine (27 trees ha –1 ). Rising at a slope of about 6 – 7º, the 1 km 2 area west of the tower 

contains subalpine fir (16 trees ha –1 ), Engelman Spruce (10 trees ha –1 ) and lodgepole pine (9 trees 

ha –1 ). Maximum leaf area index during the growing season is approximately 4.2 m 2 m –2 . Canopy 

76

height averaged 11.4 m with an average gap fraction of 17%. The site is in a state of aggradation, 

recovering from logging activities in the early part of the 20 th century. The hydrology of the site is 

dominated by moderate snowpacks that account for approximately 80% of total annual water input 

to the system [Caine, 1995]. The prevailing wind direction is from the west, particularly in the 

winter when periods of high wind speeds and neutral atmospheric stability conditions are frequent 

[Turnipseed, et al., 2002]. A detailed description of the site characteristics can be found in 

Turnipseed et al. [2002]. 

Figure 1. Composite image of Niwot Ridge, LTER site and the CU-Ameriflux tower located at C-1. 

METHODS 

Flux measurements 

Water vapor fluxes, (latent heat flux; λE) were calculated as 30-min means of 10-Hz 

measurements over a 40 d mid-winter period (DOY 60 – 100, 2002) using the eddy covariance 

(EC) method described by Turnipseed et al. [2002]: 

v ' v ' w L λ E = ρ 

where Lv is the latent heat of sublimation, w' is the deviations of vertical wind velocity (m s –1 ) 

from the ½-hr mean, ρv’ is the deviations of the water vapor density from the ½-hr mean. The 

over-story and under-story EC systems were mounted at a height of 21.5 m and 1.7 m aboveground, 

respectively, from towers separated by a distance of approximately 20-m. The over- and 

under-story EC systems and other meteorological instruments are summarized in Table 1. 

Components of snow sublimation were computed as: 

λEc,t = λEc,s + λEc,i 

where λEc,t is the total sublimation from the system measured using the over-story EC instruments 

(21.5 m above ground) and λEc,s is snowpack sublimation determined from the sub-canopy EC 

instruments (1.7 m above ground). Water vapor fluxes associated with sublimation of intercepted 

snow, λEc,i were determined as the difference of measured over-story and sub-canopy fluxes. 

Measurements of the above-canopy CO2 flux were used to confirm that photosynthesis from the 

forest canopy was negligible (i.e. values were positive indicating canopy respiration but no carbon 

uptake) and therefore over-story water flux observations could be inferred to be entirely associated 

with snow sublimation since transpiration was insignificant. 

77 

(1) 

(2)

Atmospheric stability was calculated by dividing the Monin-Obukhov length, L [Monin and 

Obukhov, 1954] into the measurement height (z): 

u * × z × ( T ) × c p ( e, 

p) 

× T 

L = 

k × z × g × H 

3 

ρ 

where u* is the friction velocity (m s –1 ), ρ(T) is the air density as a function of air temperature (T) 

(Kelvin), cp is the specific heat of dry air (kJ kg –1 K –1 ) as a function of vapor pressure, e (kPa), and 

barometric pressure, p (kPa), k is von Karman’s constant (0.41), g is acceleration due to gravity 

9.81 (m s –2 ), and H is the sensible heat flux (W m –2 (3) 

). Negative z/L values correspond to unstable 

atmospheric conditions, positive values represent stable conditions, and values near 0 are neutral. 

Table 1. Observations and instruments on the above and below canopy towers at the University of 

Colorado, Ameriflux site. 

observation 

measurement height, 

meters 

instrument 

relative humidity, % 21.5 HMP-35D, Vaisala, Inc. 

air temperature, ºC 21.5 | 1.7 CSAT-3, Cambell Scientific 

pressure, kpa 18 PT101B, Vaisala, Inc. 

net radiation, W m -2 26 4-component CNR-1, Kipp & Zonen 

H2O flux, mg m -2 s -1 21.5 | 1.7 IRGA-6260, Li-Cor 

CO 2 flux, mg m -2 s -1 21.5 IRGA-6260, Li-Cor 

wind speed, m s -1 21.5 | 1.7 propvane-09101, RM Young Inc. 

wind direction, degrees 21.5 | 1.7 propvane-09101, RM Young Inc. 

precipitation, mm 12 385-L, Met One 

soil heat flux, W m -2 

-0.07 - -0.1 HFT-1, REBS 

soil moisture, % by volume 0 - -.15 CS-615, Campbell Scientific 

soil temperature, ºC 0- - 0.1 STP-1, REBS 

Note: Above and below canopy eddy covariance systems were located 21.5 and 1.7 m above ground, 

respectively. 

Turbulent flux estimates were evaluated by exploring total energy balance closure; turbulent 

fluxes should be equal to the available energy. A linear regression between the summation of the 

sensible (H) and latent heat fluxes and the difference between the net radiation (Rn) and ground 

(G) heat flux was developed [Blanken, et al., 1998; Blanken, et al., 1997]. The relationship 

between the 30-min above canopy (λE +H) and (Rn-G) was y = 0.77x + 13 (R 2 = 0.89; p < 0.01) 

indicating adequate energy balance closure. 

The sampling area, or flux footprint, was calculated using the method described by Schuepp et 

al. [Schuepp, et al., 1990]. The upwind distance that the understory flux measurements were most 

sensitive to occurred at a distance of 23, 27, and 29-m during typical daytime, neutral, and 

nighttime atmospheric stability conditions, respectively (Figure 2a). The cumulative flux footprint, 

indicative of the upwind sampling area where 80% of the flux originated from, was 207, 243, and 

263 m (daytime, neutral, and nighttime atmospheric stability conditions, respectively) (Figure 2b). 

Supporting Understory Measurements 

Observations of soil, snow, and air temperature from three thermistor strings were used to 

develop relationships between snowpack temperature and rates of snowpack sublimation. In this 

regard, we investigated relationships between snowpack temperature gradients and diurnal 

variability in snow temperature, and rates of snowpack sublimation; snowpack temperature 

gradients control vapor pressure gradients in the snowpack and therefore the movement of water 

vapor from deeper in the snowpack toward the snowpack / atmosphere interface [McClung and 

Schaerer, 1993]. The three thermistor strings were placed along a transect through a small clearing 

(~ 6 m in diameter) in the forest adjacent to the understory flux tower (Figure 3). The thermistor 

78

strings were buried 20 – 30 cm below the soil before snow accumulation began and extended to 

80, 180, and 200 cm above the ground surface; a guy wire tied to two trees at opposite ends of the 

clearing was used to tether the tops of the thermistor strings. During the study period the 

thermistor strings provided observations of soil, snow, and air temperature. 

16 

12 

8 

4 

a 

day 

neutral 

night 

0 

0 

0 100 200 300 400 500 0 100 200 300 400 500 

upwind distance, m 

Figure 2. (a) Normalized change in the understory turbulent flux (Q) with upwind distance (x) during typical 

daytime (green), neutral (blue) and nighttime (red) atmospheric stability conditions. (b) Cumulative change in 

the understory turbulent flux (Q) with upwind distance (x) for typical daytime (green), neutral (blue) and 

nighttime (red) atmospheric stability conditions. 

Eight water content reflectometers (Campbell Scientific model CS-615) were used to monitor 

soil moisture conditions surrounding the towers (Figure 3). These observations were used to 

ensure that latent heat fluxes were primarily allocated to sublimation as opposed to snowmelt and 

to confirm that water from snowmelt had not entered the soil horizon which might trigger the 

onset of transpiration. 

Snowpack properties 

Ground observations of snow depth and snow density were derived from snow pits excavated 

weekly at two different locations (sub-canopy and within a small clearing adjacent to the flux 

towers). Within each snowpit, samples were taken at 10 cm vertical intervals over the entire 

snowpit depth using a 1000 cc stainless steel cutter. Snow density stratigraphy and bulk density 

and snow water equivalent were calculated from weighted-average density values and total 

snowpack depth. 

Observations of precipitation were used to determine the mass input between the weekly 

snowpit observations, allowing us to approximate sublimation losses; changes in snow water 

equivalent between the weekly snowpit observations result from input of mass due to snowfall and 

reduction in mass due to sublimation. This provides a field based technique for evaluating 

sublimation estimates from the sub-canopy EC system. Precipitation observations were obtained at 

a height of 12 m from the above-canopy EC tower; an Alter gauge shield was used to improve 

precipitation gauge catch efficiency. 

79 

1 

0.8 

0.6 

0.4 

0.2 

b

Figure 3. Location of above and below canopy flux towers and supporting ground based instruments. Water 

content reflectometers (M), thermistor strings, ground thermistors (T), and soil heat flux plates (H). 

Additional H, M and T measurements were made along east / west (T ew) and north / south (T NS) transects. 

RESULTS 

Soil temperature, moisture and ground heat flux were consistent with mid-winter conditions 

throughout the study period (Figure 4). Temporal variability in soil temperature (coefficient of 

variation = .77) and ground heat flux (coefficient of variation = 2.6) was considerably greater than 

that of soil moisture (coefficient of variation = 0.11). Spring onset of snowmelt percolation 

occurred on DOY 100; soil moisture increased by threefold over the subsequent 20-d period. 

1 

0 

-1 

-2 

-3 

east / west distance, meters 

Snow temp. 

thermistor strings 

Lodgepole Pine 

Engelmann Spruce 

Subalpine Fir 

sub-canopy tower 

25 

20 

15 

10 

MT HMT 

HM 

5 

60 80 100 120 60 80 100 120 

day of year 

80 

1 

.5 

0 

-.5 

-1 

60 80 100 120 

Figure 4. Time series of soil temperature, moisture, and ground heat flux from day of year 60 – 120, 2002. 

The diurnal energy fluxes, Rn, λE and H above and below the canopy are shown in Figure 5, 

together with precipitation. The CO2 flux above the canopy is included to show that the forests had 

not yet transitioned from losing to gaining carbon, and therefore transpiration at this time was 

negligible. The majority of the above-canopy net radiation was partitioned as H above the canopy, 

and as λE beneath the canopy; above-canopy ratios of the daytime mean H/Rn and λE/Rn were 

0.67 and 0.16, respectively. Beneath the canopy, these ratios were 0.02 (H/Rn) and 0.06 (λE/Rn). 

HMT 

north / south distance, meters 

T EW 

above-canopy tower 

T NS 

HM 

N

Although the λE/Rn fraction was on average relatively small, large increases in λE with a 

subsequent decrease in H occurred several times in response to snowfall events. 

Average sublimation rates over the study period were 0.7 and 0.41 mm d –1 for intercepted snow 

and the sub-canopy snowpack, respectively. Both fluxes exhibited considerable variability 

(coefficient of variation = 0.66 for both total sublimation and snowpack sublimation), with 

intercepted snow sublimation rising after snowfall events (Figure 6). The ratio between subcanopy 

snowpack sublimation and total sublimation averaged 0.45 during the study period, 

increasing with time after snowfall and approaching 1 during consecutive days without snowfall; 

e.g. DOY 63 – 65 and DOY 87 – 93 (Figure 6). On average snowpack to total sublimation ratios 

peaked 3 days after snowfall; timing to peak varied considerably with snowfall magnitude. 

day of year 2002 (MST) 

Figure 5. Diurnal variability in net radiation (R n), sensible (H) and latent (λE) heat fluxes, carbon flux (F c), 

and precipitation (P) measured above the canopy (blue lines) and beneath the canopy (green lines). Period 

shown is from DOY 60 – 100 2002. Positive values represent fluxes toward the surface. 

A total of 34.8 mm of snow fell during the measurement period (Fig 7). 38.5 mm of sublimation 

was measured above the canopy over the same time period, and 14.8 mm sublimated from the 

snowpack at the forest floor. These correspond to sublimation to precipitation ratios of 1.11 (total) 

and 0.43 (snowpack), with the total ratio exceeding one due to sublimation of snow that fell prior 

to the start of the measurements. Subtracting the above-canopy λE measurements from that below 

the canopy (Figure 7) reveals that 23.7 mm of intercepted snow was sublimated from the canopy 

itself. This corresponds to a sublimation to precipitation ratio of 0.68. 

Diurnal fluctuations in snowpack and near surface air temperatures were notably different for 

time periods with high snowpack sublimation rates. For example, only 0.1 mm of water 

81

sublimated from the snowpack on DOY 60 whereas over 0.6 mm sublimated on DOY 64. At 60 

cm above the ground surface snow temperatures fluctuated by less than 5˚ during DOY 60 and by 

more than 10˚ during DOY 64 (Figure 8). Similarly, diurnal variability in snow temperature was 

significantly lower on DOY 74 relative to DOY 93; sublimation rates were 0.1 versus 0.6 mm d –1 

for these two days, respectively. Temperature fluctuations in the surface layers, associated with 

cool nights and warm dry days potentially drive significant water vapor movement in the surface 

layers of the snowpack and enhance sublimation rates. 

precipitation 

total sublimation 

snowpack / total sublimation ratio 

day of year 

82 

snowpack sublimation 

Figure 6. Time series of daily average sublimation measured above the canopy (blue line) and beneath the 

canopy (red line). Precipitation and the ratio of snowpack sublimation to total sublimation are also shown. 

40 

30 

20 

10 

precipitation 

total sublimation 

canopy sublimation 

snowpack sublimation 

0 

60 70 80 90 100 

day of year 

Figure 7. Cumulative sublimation from the snowpack, intercepted snow (canopy), and total sublimation 

throughout the study period. Cumulative precipitation is also shown. 

Estimates of snow depth on snow temp profile plots were derived from coincident pit 

observations when available. In the case of DOY 60, the majority of precipitation was recorded on 

DOY 59 and early hours of DOY 60 and therefore we assume snow depth equivalent to that 

measured in the snowpit on DOY 64. In the case of DOY 74, snow depth was difficult to estimate 

as there was a large (12.45 mm) snowfall event on DOY 73. Thus, we assumed a snow depth of 80 

cm, corresponding to the observed snow depth from the snowpit on DOY 84. For DOY 93, we 

mean

estimated snow depth based on the 2:00 temperature curve which showed a distinct inflection 

point at the snow–atmosphere interface. 

Above and below canopy friction velocities were considerably greater for DOY 64 and 93 

relative to that on DOY 60 and 74 (Figure 9). The combination of the relatively high air 

temperatures on these days with sufficient turbulence lead to enhanced near-surface gradients in 

specific humidity and sublimation. 

80 

60 

40 

20 

a. DOY = 60 

atmosphere 

b. DOY = 64 c. DOY = 74 d. DOY = 93 

snow 

0 

soil 

-20 

-25 -20 -15 -10 -5 0 -12 -8 -4 0 -20 -15 -10 -5 0 -10 -5 0 5 10 15 

temperature, C 

Figure 8. Snowpack, soil, and air temperature profiles from 80 cm above the ground surface to 20 cm below. 

Profiles are shown for 4 different days for hours 2, 8, 14, and 20 MST. Dotted horizontal lines represent the 

snowpack / atmosphere interface. Solid horizontal lines indicate the soil / snowpack interface. 

below canopy 

above canopy 

83 

hour 

X 20:00 

14:00 

8:00 

2:00 

day of year 

Figure 9. Above (blue lines) and below canopy (green lines) diurnal variability in friction velocity and air 

temperature for the same 4 days shown in Figure 8.

Unstable atmospheric conditions resulted in considerable sublimation of intercepted snow. For 

example, on DOY 62 measurement-height / Monin-Obukhov ratios dropped below –200 (Figure 

10) and daily sublimation was 2.09 mm (Figure 6). Conversely, measurement-height / Monin- 

Obukhov ratios on DOY 76 were less than 0 but greater than –3, suggesting only slight 

atmospheric instability. Sublimation of intercepted snow on DOY 76 was 1.74 mm; only 17% 

lower than that of DOY 62. Precipitation magnitude is likely responsible for these differences with 

12 mm of precipitation falling on DOY 73 and a combined 6 mm of precipitation falling over the 

course of DOY 60 and 61 (Figure 5). 

0 

-100 

-200 

5 

0 

-300 

60 65 70 75 80 85 90 95 100 

day of year 

84 

24-hour 

moving average 

-5 

60 70 80 90 100 

Figure 10. Measurement-height / Monin-Obukhov-length ratios calculated from 30-minute, above-canopy 

observations. See equation (3) for derivation of Monin-Obukhov length. Negative and positive values 

represent unstable and stable conditions, respectively. 

DISCUSSION 

A variety of techniques have been developed to estimate sublimation from snowpacks and 

intercepted snow [Montesi, et al., 2004; Pomeroy, et al., 1998]. It is especially challenging to 

capture the impact of vegetation on variability in turbulence and subsequent vapor fluxes. Results 

of previous work performed at the individual tree scale provide useful values to evaluate results of 

our new technique. Comparisons, however, must be made with caution as our technique integrates 

fluxes over the stand scale from two systems with different flux footprints (despite reasonably 

homogenous stand characteristics); tree scale studies provide limited information at the stand scale 

due to introduction of uncertainty associated with vegetation properties. Further, quantitative 

comparison with previous studies is difficult given that meteorological conditions and site specific 

attributes can have dramatic impacts on the energy balance of forested environments – in 

particular, variability in vegetation structure [Sicart, et al., 2004]. Here we compare general 

observations of both snowpack and intercepted sublimation rates. Average mid-winter snowpack 

sublimation rates observed here (0.41 mm d –1 ) were low relative to the highest of values found 

within the literature (1.2 – 1.8 mm d –1 [Pomeroy and Essery, 1999]) and are within 14% of values 

observed at the nearby Fraser Experimental Forest (e.g. 0.36 mm d –1 ) [Schmidt, et al., 1998]. 

Fassnacht [Fassnacht, 2004] estimated winter sublimation rates at 0.75 mm d –1 at an open site in 

Leadville, Colorado; open sites are known to exhibit substantially greater sublimation rates [West, 

1962]. The weekly snowpits excavated in a clearing adjacent to the flux towers used in this 

research indicated a total sublimation rate of 0.8 mm d –1 . While these estimates have inherent 

uncertainties, these on-site observations and comparisons with previous studies indicate that 

sublimation rates are not being overestimated using the sub-canopy EC system. In this regard, it is

important to note that the average snowpack to total sublimation ratio of 0.45 (Figure 6) represents 

the low-end of the contribution of sub-canopy sublimation to overall water loss; a significant 

finding given previous assumptions that sublimation losses in forested systems are primarily the 

result of intercepted snow sublimation [Montesi, et al., 2004]. 

The assessment of sub-canopy sublimation estimates mentioned above must be considered when 

evaluating the EC estimates of intercepted snow sublimation as they are calculated from the 

residual of total sublimation and sub-canopy sublimation (equation (2)). Sublimation rates of 

intercepted snow estimated using our EC approach (0.71 mm d –1 ) compared favorably with 

previous works. For example, Parviainen [Parviainen and Pomeroy, 2000] estimated intercepted 

snow sublimation from a boreal forest at 0.5 mm d –1 ; at higher latitudes available energy is 

diminished due to higher solar zenith angles. 

Montesi et al., [2004] explored the impact of elevation on sublimation rates and found that 

increased wind speeds, lower relative humidity and warmer air temperatures contributed to a 23% 

increase in sublimation rates at lower elevation. On average Montesi’s results indicate 

considerable differences between our estimates, with sublimation losses equivalent to 20 – 30% of 

total snow water equivalent during the 21 storms considered. These differences may be due to an 

underestimate in sub-canopy sublimation from the sub-canopy EC system used here. Differences 

may also be due to previously mentioned sources of underestimates in sublimation using the treeweighting 

method of Montesi [2004]. 

As previously found by Niu and Yang [2004], the relatively high over-story sensible heat fluxes 

lead to strong temperature differences between vegetation and the snow-surface, driving strong 

specific humidity gradients at the snow / atmosphere interface and elevated snowpack sublimation 

rates (e.g. DOY 78 – 82 & 88 – 95, Figure 6). When snowfall occurred, over-story available 

energy was partitioned into latent heat fluxes (e.g. Figure 5, DOY 74), leading to high withincanopy 

sublimation rates but diminished diurnal variability in temperatures at the snow – 

atmosphere interface (DOY 74, Figure 8). These results indicate that sublimation losses from the 

under-story snowpack is strongly dependent on the partitioning of sensible and latent heat fluxes 

in the canopy. 

CONCLUSIONS 

Sub-canopy and over-story eddy covariance systems indicated substantial losses of winterseason 

snow accumulation in the form of snowpack (0.41 mm d –1 ) and intercepted snow (0.71 mm 

d –1 ) sublimation. The partitioning between these over and under story components of water loss 

was highly dependent on atmospheric conditions and near-surface conditions at and below the 

snow – atmosphere interface. High over-story sensible heat fluxes lead to strong temperature 

gradients between vegetation and the snow-surface, driving substantial specific humidity gradients 

at the snow surface and high sublimation rates. Intercepted snowfall resulted in rapid response of 

over-story latent heat fluxes, high within-canopy sublimation rates, and diminished sub-canopy 

snowpack sublimation. These results indicate that sublimation losses from the under-story 

snowpack is strongly dependent on the partitioning of sensible and latent heat fluxes in the 

canopy. This compels comprehensive studies of snow sublimation in forested regions that 

integrate sub-canopy and over-story processes. 


This research was supported by NASA grant #NNG04GO74G and a visiting research fellowship 

awarded to the primary author at the Cooperative Institute for Research in Environmental 

Sciences, University of Colorado at Boulder. Instrumentation infrastructure was developed in part 

through the Ameriflux network and through support of the National Science Foundation, Niwot 

Ridge LTER Project. Phillip Jacobson provided technical support. Andrew Rossi is acknowledged 

for collecting snow pit data. 

85

REFERENCES 

Bales, R., N. Molotch, T. Painter, M. Dettinger, R. Rice, and J. Dozier (2006), Mountain 

hydrology of the western United States., Water Resources Research, 42, W08432, 

doi:08410.01029/02005WR004387. 

Blanken, P. D., T. A. Black, H. H. Neumann, G. Den Hartog, P. C. Yang, Z. Nesic, R. Staebler, 

W. Chen, and M. D. Novak (1998), Turbulent flux measurements above and below the 

overstory of a boreal aspen forest, Boundary-Layer Meteorology, 89, 109–140. 

Blanken, P. D., T. A. Black, P. C. Yang, H. H. Neumann, Z. Nesic, R. Staebler, G. den Hartog, M. 

D. Novak, and X. Lee (1997), Energy balance and canopy conductance of a boreal aspen forest: 

Partitioning overstory and understory components, Journal Of Geophysical Research- 

Atmospheres, 102, 28915–28927. 

Caine, N. (1995), Temporal trends in the quality of streamwater in an Alpine environment: Green 

Lakes valley, Colorado Front Rand, USA, Geografiska Annaler Series A-Physical Geography, 

77A, 207–220. 

Davis, R. E., J. P. Hardy, W. Ni, C. E. Woodcock, J. C. Mckenzie, R. Jordan, and X. Li (1997), 

Variation of snow cover ablation in the boreal forest: A sensitivity study on the effects of 

conifer canopy, Journal of Geophysical Research, 102, 29389–29395. 

Essery, R., J. Pomeroy, J. Parviainen, and P. Storck (2003), Sublimation of snow from coniferous 

forests in a climate model, Journal Of Climate, 16, 1855–1864. 

Fassnacht, S. R. (2004), Estimating Alter-shielded gauge snowfall undercatch, snowpack 

sublimation, and blowing snow transport at six sites in the coterminous USA, Hydrological 

Processes, 18, 3481–3492. 

Lundberg, A., and S. Halldin (1994), Evaporation of intercepted snow: Anlaysis of governing 

factors., Water Resources Research, 30, 2587–2598. 

McClung, D., and P. Schaerer (1993), The Avalanche Handbook, 271 pp., The Mountaineers, 

Seattle, Washington. 

Monin, A. S., and A. M. Obukhov (1954), Basic laws of turbulent mixing in the ground layer of 

the atmosphere., Tr. Inst. Teor. Geofiz. Akad. Nauk SSSR, 1, 95–115. 

Montesi, J., K. Elder, R. A. Schmidt, and R. E. Davis (2004), Sublimation of intercepted snow 

within a subalpine forest canopy at two elevations, Journal of Hydrometeorology, 5, 763–773. 

Nakai, Y. T., T. Sakamoto, T. Terajima, H. Kiahara, and T. Saito (1994), Snow interception by 

forest canopies: Weighing a conifer tree, meteorological observation and analysis by hte 

Penman-Monteith formula, IAHS Publ., 223, 227–236. 

Niu, G. Y., and Z. L. Yang (2004), Effects of vegetation canopy processes on snow surface energy 

and mass balances, Journal Of Geophysical Research-Atmospheres, 109. 

Parviainen, J., and J. W. Pomeroy (2000), Multiple-scale modelling of forest snow sublimation: 

initial findings, Hydrological Processes, 14, 2669–2681. 

Pomeroy, J. W., and R. L. H. Essery (1999), Turbulent fluxes during blowing snow: field tests of 

model sublimation predictions, Hydrological Processes, 13, 2963–2975. 

Pomeroy, J. W., and D. M. Gray (1995), Snowcover accumulation, relocation and management, 

NHRI Science Report 7, National Hydrology Research Institute, Environment Canada, 134 pp. 

Pomeroy, J. W., N. Parviainen, N. Hedstrom, and D. M. Gray (1998), Coupled modeling of forest 

snow interception and sublimation., Hydrological Processes, 12, 2317–2337. 

Pomeroy, J. W., and R. A. Schmidt (1993), The use of fractal geometry in modeling intercepted 

snow accumulation and sublimation., Proceedings of the 50th Eastern Snow Conference., 

Quebec City, QC, Canada., 50, 1–10. 

Schmidt, R. A. (1991), Sublimation of snow intercepted by an artificial conifer, Agricultural and 

Forest Meteorology, 54, 1–27. 

Schmidt, R. A., R. L. Jairell, and J. W. Pomeroy (1988), Measuring snow interception and loss 

from an artificial conifer., Proceedings of the 56th Western Snow Conference., Kalispell, MT., 

166–169. 

Schmidt, R. A., and C. A. Troendle (1992), Sublimation of intercepted snow as a global source of 

water vapor., Proceedings of the 60th Western Snow Conference., Jackson, WY., 60, 1–9. 

86

Schmidt, R. A., C. A. Troendle, and J. R. Meiman (1998), Sublimation of snowpacks in subalpine 

conifer forests, Canadian Journal Of Forest Research-Revue Canadienne De Recherche 

Forestiere, 28, 501–513. 

Schuepp, P. H., M. Y. Leclerc, J. I. Macpherson, and R. L. Desjardins (1990), Footprint Prediction 

Of Scalar Fluxes From Analytical Solutions Of The Diffusion Equation, Boundary-Layer 

Meteorology, 50, 353–373. 

Sicart, J. E., J. W. Pomeroy, R. E. Essery, J. Hardy, T. Link, and D. Marks (2004), A sensitivity 

study of daytime net radiation during snowmelt to forest canopy and atmospheric conditions, 

Journal of Hydrometeorology, 5, 774–784. 

Turnipseed, A. A., P. D. Blanken, D. E. Anderson, and R. K. Monson (2002), Energy budget 

above a high-elevation subalpine forest in complex topography, Agricultural And Forest 

Meteorology, 110, 177–201. 

West, A. (1962), Snow evaporation from a forested watershed in the central Sierra Nevada, 

Journal of Forestry, 60, 481–484. 

Woo, M. K., and M. A. Giesbrecht (2000), Simulation of snowmelt in a subarctic spruce 

woodland: Scale considerations, Nordic Hydrology, 31, 301–316. 

87


88

89 



The Retrievals of Snow Cover Extent and Snow Water Equivalent 

from a Blended Passive Microwave–Interactive Multi-Sensor 

Snow Product 

ABSTRACT 

CEZAR KONGOLI 1 , CHARLES A. DEAN 1 , 

SEAN R. HELFRICH 2 , AND RALPH R. FERRARO 3 

The retrieval of Snow Water Equivalent (SWE) from remote sensing satellites continues to be a 

very challenging problem. In this paper, we evaluate a new SWE product derived from the 

blending of a passive microwave Snow Water Equivalent product based on the Advanced 

Microwave Sounding Unit (AMSU) with the Interactive Multi-sensor Snow and Ice Mapping 

System (IMS). The microwave measurements have the ability to penetrate the snow pack, and thus 

the retrieval of SWE is best accomplished using the AMSU measurements. On the other hand, the 

IMS maps snow cover more reliably due to the use of multiple satellite and ground observations. 

The evolution of global snow cover extent from the blended, the AMSU and the IMS products was 

examined during the 2006 snow season. Despite the overall good inter-product agreement, it was 

shown that the retrievals of snow cover extent are improved using IMS, with implications for 

improved microwave retrievals of SWE. In a separate investigation, the microwave retrievals of 

SWE were examined globally and in Central Europe. Qualitative evaluation of global SWE 

patterns showed dependence on land surface temperature: the lower the temperature, the higher the 

SWE retrieved. This temperature bias was attributed in part to temperature effects on those snow 

properties that impact microwave response. Therefore, algorithm modifications are needed with 

more dynamical adjustments for changing snow cover. Quantitative evaluation over Slovakia for a 

limited period in 2006 showed reasonably good performance for SWE less than 100 mm. 

Sensitivity to deeper snow decreased significantly. 

Keywords: Snow cover, Snow Water Equivalent, Advanced Microwave Sounding Unit (AMSU), 

Interactive Multisensor Snow and Ice Mapping Unit (IMS) 

INTRODUCTION 

The retrievals of Snow Water Equivalent (SWE) from satellites continue to be a very difficult 

and challenging problem. While mapping of global snow cover has been accomplished using 

visible or passive microwave measurements, the mapping of SWE from space has long been an 

exclusive domain of passive microwave sensors. Visible measurements are typically more 

sensitive to snow cover surfaces than passive microwave measurements due to the high visible 

1 

QSS Group Inc. – NOAA Science Center, 5200 Auth Rd., Camp Springs, MD, 20746, 

cezar.kongoli@noaa.gov 

2 

NOAA/NESDIS/OSDPD - NOAA Science Center, 5200 Auth Rd., Camp Springs, MD, 20746, 

sean.helfrich@noaa.gov 

3 

NOAA/NESDIS/OSDPD - NOAA Science Center, 5200 Auth Rd., Camp Springs, MD, 20746, 

sean.helfrich@noaa.gov

eflectance of snow cover as compared to other surface features, and therefore, they typically 

provide more accurate mapping of snow cover than the microwave imagery. On the other hand, 

passive microwaves at specific window frequencies penetrate much deeper into the snow pack due 

to longer wavelengths, and have shown to provide information on the snow cover properties 

including SWE. Another advantage of passive microwaves is their ability to penetrate cloudy 

atmospheres at specific window frequencies and thus to provide retrievals of land surface 

parameters in near all-weather conditions. Despite the near all-weather capability, the current 

microwave sensors provide measurements and thus retrievals at much coarser resolution (25 km) 

than the visible satellite sensors (1 km). Another disadvantage of passive microwave imagery is its 

limited capability to penetrate wet snow cover. 

The need for continuous regional and global snow cover mapping for climate, hydrological and 

weather applications has led, especially in recent years, to the development of snow cover 

mapping products based on multi-sensor data sources. For example, the National Oceanic and 

Atmospheric Administration (NOAA), the National Environmental Satellite Data and Information 

Systems (NESDIS) uses the Interactive Multi-sensor Snow and Ice Mapping System (IMS) 

operationally to provide Northern Hemispheric (NH) snow and ice mapping as input to 

environmental prediction models. The IMS system utilizes in an interactive fashion, through the 

interpretation of an analyst, visible and microwave satellite imagery as well as ground 

observations. Over the years, as more satellite sensors have become available, the IMS has 

incorporated more satellite data sources and has evolved into a more efficient snow mapping 

system (Ramsay, 1997, Helfrich et al., this issue). Another multi-sensor snow and ice mapping 

product is the NOAA’s AUTOSNOW product (Romanov et al., 2000). It utilizes visible, infrared 

and microwave satellite data to automatically generate high-resolution and continuous NH snow 

and ice mapping. 

Main research objective of this study is to evaluate the accuracy of a new SWE product derived 

from the blending of a passive microwave SWE product based on the Advanced Microwave 

Sounding Unit (AMSU) with the IMS snow cover extent product. The blended SWE product is 

needed (in addition to snow cover extent) as input to environmental prediction models. Besides 

being delivered in compatible format with the IMS snow cover extent, the blended product would 

be value-added in that SWE would be improved due to more reliable IMS snow mapping. Similar 

approaches have been reported in literature, e.g., the blending of a microwave SWE product based 

on the Special Sensor Microwave Imager (SSM/I) or the Advanced Microwave Scanning 

Radiometer-Earth Observing System (AMSR-E) with the snow cover extent product derived from 

the Moderate Resolution Imaging Spectrometer (MODIS) (Armstrong et al. (2003). Compared to 

SSM/I and AMSR-E, the AMSU has some unique characteristics: Despite its coarse resolution, it 

has a wider spatial coverage than AMSR-E and SSM/I, and has additional microwave channels at 

frequencies in the window, oxygen and water vapor absorption regions. These additional features 

could potentially be utilized for the development of more robust SWE retrievals. 

In this paper, we evaluate the blended SWE product for its accuracy in mapping snow cover and 

SWE. In terms of snow cover mapping, we inter-compare the AMSU snow cover extent product 

with that of IMS. Inter-comparability of snow cover extent retrieved from the AMSU and from 

IMS is important to investigate as it has implications for the accuracy of the blended SWE 

product. Next, the microwave-derived SWE estimates are also evaluated globally and regionally 

over Central Europe (Slovakia). 

90

METHODS 

The IMS Product 

The IMS was designed to allow mapping of snow cover interactively on a daily basis using a 

variety of data sources within a common geographic system. Such data sources include NOAA 

Polar Orbiting Environmental Satellites (POES) and Geostationary Environmental Satellites 

(GOES) data, Japanese Geostationary Meteorological Satellites (GMS), European geostationary 

meteorological satellites (METEOSAT) and US Department of Defense (DOD) polar orbiters 

(DMSP). Since its inception in 1997, the IMS system has evolved significantly in terms of input 

data sources, production technology and output format. For more detailed description of evolution 

and capabilities of the IMS system, the reader is referred to Helfrich et al., (this issue) and Ramsay 

(1998; 2000). The IMS snow and ice mapping is currently accomplished once daily over the 

Northern Hemisphere. The product is generated at 4 km and 24 km resolution in Polar 

Stereographic (PS) Projection. 

The AMSU Instrument 

The AMSU instrument contains two modules: AMSU-A and AMSU-B. The A module has 15 

channels in the 23-89 GHz frequency range (1-15, Table 1). The B module has five channels in the 

89-183 GHz frequency range (16–20, Table 1). The AMSU-A has an instantaneous field of view 

(FOV) of 48 km at nadir for all frequency channels, and scans ±48 0 from nadir with a total of 30 

measurements across the scan. For AMSU-B, 90 measurements are made across the scan with a 

nadir resolution of 16 km for all channels. The nadir resolution degrades at larger scan angles. The 

swath width of both AMSU-A and –B is 2343 km. 

The AMSU-A and -B is flown on board the NOAA 15, NOAA 16, and NOAA 17 POES. This 

three-satellite suite offers a near-real time global sampling nearly every 4 hours. The unique 

combination of channels in the microwave window (23, 31, 89, 150 GHz), opaque water vapor 

(183±1, ±3, ±7 GHz) and oxygen absorption (50-60 GHz) regions has led to the development of a 

variety of surface and atmospheric products, generated in near-real time within a system called the 

Microwave Surface and Precipitation Product System (MSPPS). The suite of operational MSPPS 

products (Ferraro et al., 2002) includes rain rate, total precipitable water, cloud liquid water, ice 

water path, snow cover, sea ice concentration and land surface temperature and emissivities at 23, 

31 and 50 GHz. Recent additions include a snowfall detection (Ferraro et al., 2004), and the 

extension of the snow cover extent product to include snow water equivalent (SWE) retrievals 

(Kongoli and Ferraro, 2004). In May 2005, a new POES satellite, NOAA 18, was launched and 

was also incorporated into MSPPS. This new satellite contains a new instrument, the Microwave 

Humidity Sensor (MHS) which replaced the AMSU-B. Similar to AMSU-B, the MHS is a fivechannel 

radiometer. However, it differs from AMSU-B in that channel 2 and channel 5 (channels 

17 and 20 in Table 1 ) are 157 GHz and 190 GHz. Calibration and validation of the new MHS 

instrument within MSPPS has been completed and is described in Meng et al. (2006). As a result, 

similar products that are generated from the three satellites are now also generated from NOAA- 

18. 

91

Table 1. AMSU-A and –B channel characteristics. Channels 1-15 are AMSU-A channels and channels 

16-20 are AMSU-B channels 

Channel 

number 

Center 

frequency 

(GHz) 

Number 

of pass 

bands 

92 

Band 

width 

(MHz) 

Center 

frequency 

stability 

(MHz) 

FOV 

at 

Nadir 

(km) 

1 23.80 1 251 10 48 

2 31.40 1 161 10 48 

3 50.30 1 161 10 48 

4 52.80 1 380 5 48 

5 53.59±0.115 2 168 5 48 

6 54.40 1 380 5 48 

7 54.94 1 380 10 48 

8 55.50 1 310 0.5 48 

9 57.29 = fo 1 310 0.5 48 

10 fo±0.217 2 76 0.5 48 

11 fo±0.322±0.048 4 34 0.5 48 

12 fo±0.322±0.022 4 15 0.5 48 

13 fo±0.322±0.010 4 8 0.5 48 

14 fo±0.322±0.004 4 3 0.5 48 

15 89.00 1 2000 50 48 

16 89.00 1 5000 50 16 

17 150 1 4000 50 16 

18 183±1 1 1000 50 16 

19 183±3 2 2000 50 16 

20 183±7 2 4000 50 16 

The AMSU Snow Cover Extent Product 

The identification of snow cover over land is based on the algorithm of Grody (1991) and Grody 

and Basist (1996). Snow is identified in a series of steps that discriminate snow from nonscattering 

surfaces such as wet land and vegetation (Figure 1) and from other scatterers such as 

deserts and rain. This is accomplished using a number of scattering indices that utilize a 

combination of the AMSU window frequency channels at 23, 31, 50 and 89 GHz (Table 1, 

AMSU-A channels 1, 2, 3 and AMSU-B channel 16, respectively). As shown in Figure 1, snow 

exhibits a unique spectral signature in the 10–100 GHz microwave frequency region: The 

brightness temperature, and hence, the surface emissivity decreases with increasing frequency. In 

contrast, other surfaces such as wet land and vegetation exhibit a rather flat or reverse response. 

More recently, additional filters have been incorporated that utilize a combination of AMSU 

channels at 150 GHz (Table 1, channel 17) at 53.6 GHz (Table 1, channel 5) and at 183 ± 3 GHz 

(Table 1, channel 19) for improved snow–rain discrimination (Kongoli et al. 2005).

Figure 1. Microwave spectral characteristics of vegetation and snow cover (Matzler, 1994). 

The AMSU SWE Product 

The retrieval of SWE is based on the AMSU brightness temperature measurements at 23 

(TB23), 31 (TB31) and 89 (TB89) GHz. Based on these channel measurements, two scattering 

indices are computed: 

SI31 = TB23 – TB89 (1) 

SI89 = TB23 – TB31 (2) 

The SWE is computed for the snow-covered pixels by the following empirical relationships: 

SWE = K1 + K2* SI89 (3) 

SWE = K3 + K4 * SI31 (4) 

where K1, K2, K3, and K4 are empirically derived coefficients. SWE is computed only over 

snow-covered land retrieved by the AMSU snow cover extent algorithm. The SWE algorithm 

implicitly differentiates between two snow-cover types: finer-grained fresh snow (Eq. 3) and 

coarser-grained older (Eq. 4) snow-cover via the estimation of a switch (mean grain size) 

parameter, which is also estimated from a linear relationship with TB23, TB31 and TB89. The 

coefficient values for K1, K2, K3 and K4 and the switch parameter are fixed. They were derived 

from regional studies in the U.S Great Plains areas (Kongoli et al., 2004). It is also important to 

note that atmospheric or land surface vegetation effects are not accounted for in the current 

version. The concept behind the algorithm using the scattering indices (1) and (2) and Equations 

(3) and (4), respectively, is illustrated in Figure 2, which depicts examples of AMSU 

measurements at 23, 31, 89 and 150 GHz window frequency channels. Note that microwave 

sensitivity to snow cover shifts from lower (20–30 GHz) to higher (89–150 GHz) frequency 

regions as the snow ages. For freshly fallen snow (6 hr old), the steepest microwave gradient 

occurs in the 89–150 GHz region (about 25 K), due to strong scattering at 150 GHz by finergrained 

snow cover. Note the flat response in the 20–30 GHz region, suggesting low microwave 

sensitivity in this lower frequency range. As the snow becomes older and the grain size increases, 

sensitivity shifts to the 30–90 GHz region which attains the steepest gradient. Only for coarsegrained, 

metamorphosed snow cover, referred to in Figure 2 as “old snow” does the microwave 

sensitivity increase significantly at 23 and 31 GHz. In the current algorithm, 150 GHz is not 

utilized for the retrievals of SWE. It is important to note, however, that this dual snow type 

discrimination is a crude approximation and representation of the much greater snow cover type 

and hence grain size and (micro)structure variability, e.g., as illustrated in Figure 1. 

93

Figure 2. Example of AMSU measurements over snow covered land 

Blending Procedure 

Primary satellite data are not used in the blended product. Instead, the data inputs include 

AMSU SWE product files and IMS snow cover product files. The SWE data are obtained from the 

MSPPS Level 2 swath geographical products available in the Hierarchical Data Format-Earth 

Observing System (HDF-EOS) at NOAA/NESDIS. These swath files are generated on an orbital 

basis from NOAA-15, 16, 17 and more recently the NOAA-18 satellites. Spatial resolution of the 

AMSU SWE swath data is that of AMSU-B: 16 km at nadir viewing angle. Resolution degrades as 

viewing angle increases. The IMS snow cover data are obtained daily as ASCII files in 1/16 th 

mesh PS Projection, or approximately 24 km in resolution. The data flow and rules for generating 

merged product are as follows: 

- Obtain daily IMS snow cover product file, 

- Obtain the AMSU SWE swath product files. The NOAA-18 instrument was selected as 

the most recent instrument, and the descending (night-time) was selected to minimize 

SWE retrieval errors due to surface melting during day-time, 

- Convert the AMSU swath SWE data into gridded data compatible with the IMS snow 

cover product file. 

- Inter-compare the AMSU SWE and IMS snow cover values on a grid cell-by-cell basis. 

If SWE value is missing or zero over snow cover as identified by IMS, retrieve previous 

day SWE value (two-day compositing). If previous-day SWE value is positive assign that 

SWE value to grid cell. Otherwise, flag as “indeterminate” the grid cell that corresponds 

to missing or zero AMSU SWE and IMS snow-covered land. Also, label as 

“indeterminate” instances of positive AMSU SWE value that correspond to IMS snowfree 

land. At this point, no revision of SWE values over these “indeterminate” grid cells 

is made. This is reserved for future development, 

- Generate blended SWE output file, a statistics file with grid cell confusion data, and 

image input and blended output grid files for product visualization and monitoring 

Ground SWE measurements over Slovakia 

The SWE retrievals were quantitatively evaluated against SWE measurements over Slovakia 

during February–March 2006. The SWE data and maps were provided by the Slovak National 

Hydro-meteorological Institute which maintains a dense network of ground stations that record 

94

snow depth and SWE on a weekly basis (Figure 3). Dotted green points denote station locations, 

whereas in red are depicted grid points spaced 0.25 by 0.25 degrees Latitude and Longitude. The 

areas depicted in brown in Central and North Slovakia denotes higher elevation terrain. To 

statistically evaluate SWE retrievals, station SWE data were matched up with the AMSU 

observations and retrieved SWE. Station SWE data were averaged over a cell of 0.5 X 0.5 degree 

in latitude and longitude centered at the centroid of AMSU FOV. This size represents the average 

resolution of the AMSU instrument. Along with the average SWE, average values of snow depth 

and elevation, the number of stations per cell, as well as standard deviations of SWE, depth and 

elevation were computed. Only cells with over 5 stations were included in the analysis. Due to the 

higher density of stations, cells with over five stations were predominant. 


Figure 3. Map of Slovakia and locations of ground SWE stations 

Inter-comparison of blended snow cover with that retrieved from the AMSU and IMS 

products 

The blended product system has been automated and running experimentally on a UNIX 

machine since November 2005. Examples of blended SWE product are shown in Figure 4. Figure 

4 displays merged SWE (left) and inter-comparison of AMSU and IMS snow cover extent (right) 

on November 7, 2005 (top panels) and on February 15, 2006 (bottom panels). The intercomparison 

map depicts the blended SWE area in blue, the underestimated microwave SWE area 

in green (AMSU-SWE values are zero but IMS snow cover is positive), and the overestimated 

microwave SWE area in brown (AMSU-SWE is positive but IMS snow cover is zero). The red 

color depicts areas when AMSU-SWE is labeled as “undetermined” due to confusion with rain, 

and is displayed for diagnostic purposes. Note the green-coded area over Eastern Canada on 

November 7, 2005 (top right), denoting underestimation of snow cover and hence SWE by the 

AMSU snow cover extent product. In the blended SWE product, these undetermined SWE areas 

are denoted as “white”. Examination of ground observations from several Canadian 

meteorological stations indicated continuous snowfall occurrences prior to November 7. The 

deposited new snow was not captured by the AMSU snow cover extent product, but was identified 

by the IMS snow cover product. Interestingly, these snowfall occurrences were retrieved by the 

AMSU snowfall product. An example of the AMSU snowfall product retrievals on November 2, 

2005 is given in Figure 5. The AMSU snowfall algorithm utilizes frequencies at 89 GHz and 

above (AMSU-B channels 16-20 in Table 1) to identify precipitation in the form of snowfall 

95

(Kongoli et al., 2003). Examination of the AMSU measurements on November 7, 2005 indicated 

that the 89 GHz frequency channel was not sufficiently sensitive to this newly deposited snow. 

However, the 150 GHz window frequency channel exhibited scattering response, e.g., depressed 

brightness temperatures relative to that of 89 GHz (see also Figure 2). Note that frequency 

channels above 89 GHz are not available for the SSM/I or AMSR-E sensors. This unique 

capability of the AMSU instrument to retrieve snowfall and to detect newly deposited snow cover 

could be potentially incorporated and utilized for improved SWE retrievals for new snow. On the 

other hand, in the blended SWE product, SWE over part of Mongolia as determined by the AMSU 

product but labeled as “snow-free” land by the IMS is correctly left out. This overestimation of 

snow cover over Mongolia by the AMSU persisted during much of the winter season of 2006, and 

is also shown on February 15, 2006 (bottom panels). It is shown that overall, there is better 

agreement between the microwave-derived SWE and the IMS snow cover, e.g., over Eastern 

Canada, than in early winter. Exception is areas over Mongolia where overestimation of the 

AMSU-SWE persists throughout the winter. Examination of MODIS snow cover maps indicated 

that indeed these areas did not have snow cover, and therefore, they represent “false” snow areas 

by the AMSU SWE product. Examination of the AMSU measurements over Mongolia indicated 

relatively small values of the scattering index at 89 GHz (SI89), which is used in the AMSU snow 

cover extent product to identify snow cover and in the AMSU-SWE product to compute fresh 

snow SWE (Eq. 2 and 4). This low scattering signal at 89 GHz was present during night-time (low 

temperatures) and day-time (above freezing temperatures). The presence of scattering in a wide 

range of atmospheric conditions over Mongolia would be indicative of ground rather than 

atmospheric effects, e.g., soil grain scattering (Basist et al., 1996). Figure 6 is a plot of the 

percentage of the retrieved AMSU-SWE area relative to the IMS snow coverage from December 

1, 2005 through April 15, 2006 (left-hand vertical axis) and of the overestimated AMSU-SWE 

area as a percentage of the IMS snow coverage (right-hand vertical axis). It is shown that the net 

SWE area mapped by the AMSU is at about 75% as compared to that of IMS, which is relatively 

high for a microwave instrument. The overestimation of snow cover by the AMSU is about 6 % 

relative to IMS snow coverage. 

96

Figure 4. Example of blended SWE product retrievals over Northern Hemisphere on November 7, 2005 and 

February 15, 2006. Left-hand images depict the blended SWE and the right-hand images depict blended and 

unblended areas. 

97

Figure 5. The AMSU snowfall retrievals over North American on November 2, 2005. 

Figure 6. Inter-comparison of global SWE and IMS snow cover extent 

98

Evaluation of the global distribution of retrieved SWE 

Figure 7 depicts example of retrievals of AMSU SWE (left) and Land Surface Temperature 

(LST) (right) in March 2002 and 2003. Note the predominance of high SWE values (depicted in 

red) over North-central and Eastern Siberia, over Northern Canada and Alaska. Interestingly, these 

patterns are similar to the ones retrieved in 2006 (Figure 4). Although global SWE data are not 

available to quantitatively evaluate retrieved SWE, some general observations can be made. For 

example, the increase in SWE from Western to Eastern Siberia may be unrealistic given the dry, 

low-temperature climate predominant in Eastern Siberia. These trends appear to be strikingly 

similar in 2002, 2003 and 2006. Despite large-scale similarity, a closer visual inspection revealed 

some inter-annual differences over specific regions. For instance, retrieved SWE over some 

Western US regions, e.g., Colorado, Wyoming and Idaho, and over Canada, e.g, Alberta, were 

higher in 2002 than in 2003. Another observed difference was the larger extent of high SWE over 

Siberia in 2003 than in 2002. This inter-annual variability in SWE appears to follow that of LST: 

The lower the LST the higher the retrieved SWE. For instance, lower LST in 2002 than in 2003 

over Western US was associated with retrieved SWE higher in 2002 than in 2003. LST also 

appears to influence to a large extent the global distribution of retrieved SWE. A possible 

explanation of this temperature bias could be the influence of temperature on the evolution of 

snow cover properties that impact the microwave response. As explained earlier, the AMSU SWE 

algorithm coefficients are static. Foster et al., 2005 report a new dynamical approach where 

algorithm coefficients are adjusted based on a geo-referenced snow classification system (Sturm et 

al., 1995). The authors report improved performance compared to static retrievals. 

Figure 7. AMSU retrieval of SWE (top) and Land Surface Temperature (bottom) in March 2002 and 

2003. 

99

Evaluation of SWE retrievals over Slovakia 

Figure 8 plots the histogram of station averaged SWE over Slovakia (top panel) and the average 

and standard deviation of SWE matched up with the AMSU measurements as a function of 

average elevation (bottom panel). As shown, SWE is most frequent around 50 mm. There is, 

however, a good distribution of SWE in the 50- to 150-mm range. Note also that the standard 

deviation of SWE increases with elevation and the SWE amount. Figure 9 displays plots of 

retrieved SWE versus measured SWE for elevations less than 400 m (top panel left) and for all 

elevation ranges (bottom panel left). As shown, there is good agreement of retrieved SWE versus 

measured SWE for elevations less than 400 m. Elevations less than 400 mm had an average SWE 

of 40 mm and maximum SWE and snow depth of 100 mm and 30 cm, respectively. For higher 

elevations, retrieved SWE is significantly underestimated. The right hand panels in Figure 9 depict 

plots of the bias (retrieved SWE–measured SWE) as a function of elevation (top) and as a function 

of measured SWE (bottom). As shown, the bias exhibits stronger dependence on SWE amounts 

than on elevation. Microwave signal saturation at SWE values above 100 mm is also reported in 

literature (Dong et al., 2005). However, given the high standard deviation of measured SWE 

within the AMSU FOV, a more rigorous examination of the possible error sources would have 

required additional information on snow cover and ground station distribution. Figure 10 displays 

AMSU SWE retrievals over Central Europe including Slovakia (left panel) and the map of station 

derived SWE over Slovakia (right panel) on February 27, 2006. As shown, the highest AMSU 

SWE retrieved over Slovakia is in the 100- to 120-mm range in Northeastern Slovakia, compared 

to station derived SWE over 180 mm. Note, however, the highly variable station-derived SWE 

over Northeastern Slovakia. Large-scale SWE patterns are retrieved reasonably well, e.g., the 

increase in SWE towards the Northeast. Small-scale patterns, however, are not well captured due 

to the coarse AMSU resolution. 

Figure 8. Average measured SWE distribution matched up with AMSU data 

100

Figure 9. Inter-comparison plots of AMSU versus measured SWE (left panels) and of the bias of retrieved 

SWE as a function of elevation and measured SWE (right panels) 

Figure 10. AMSU retrieved SWE over central Europe (left) and station SWE over Slovakia on February 27, 

2006. 

101

SUMMARY 

Main research objective of this study was to evaluate the accuracy of a new SWE product 

derived from the blending of a passive microwave SWE product based on AMSU instrument with 

the IMS snow cover extent product. The blended SWE product is needed (in addition to snow 

cover extent) as input to environmental prediction models. The paper described data and 

evaluation methodology. The blending procedure between the AMSU SWE and IMS snow cover 

extent products was also described. Next, snow cover extent retrievals by the AMSU, IMS and the 

blended product were evaluated for the 2005–2006 snow season. Despite good inter-product 

agreement, it was shown that the IMS had better skills than the AMSU snow cover extent product 

in retrieving new snow cover in early winter and in correctly identifying snow-free land, e.g., over 

Mongolia, that was otherwise reported as “snow” by the AMSU product. These problem areas 

were also evaluated using other observation sources, e.g., the MODIS snow cover product and the 

AMSU snowfall product. In a separate investigation, the paper also described the evaluation of the 

microwave SWE product globally and over central Europe (Slovakia). Qualitative evaluation of 

the large-scale, global SWE patterns showed dependence of AMSU retrieved SWE on land surface 

temperature: the lower the land surface temperature, the higher the SWE retrieved. This 

temperature bias was attributed in part to temperature effects on snow properties that impact 

microwave response. Therefore, AMSU SWE algorithm modifications are needed, e.g., 

adjustment of algorithm coefficients for changing snow properties. Quantitative evaluation over 

Slovakia for a limited period in 2006 showed reasonably good agreement for SWE less than 100 

mm and snow depth less than 30 cm, associated with low elevation terrain (less than 400 m). 

Sensitivity to snow deeper than 100 mm in SWE decreased significantly. 

ACKNOWLEDGEMENT 

The authors wish to thank Dr. Yan Kanak and Joseph Pecho from the Slovak National Institute 

of Hydrometeorology for their cooperation and for graciously providing valuable station data over 

Slovakia. 

REFERENCES 

Armstrong, R., M.J. Brodzik, and M. Savoie, 2003: Multisensor approach to mapping snow cover 

using data from NASA’s Terra and Aqua Spacecraft (AMSR-E and MODIS), Eos Trans. 

AGU, 84(46). 

Basist, A., D. Garrett, R. Ferraro, N. Grody and K. Mitchell, 1996: A comparison between snow 

cover products derived from visible and microwave observations, J. Appl. Meteorol., 35:163– 

177. 

Dong, J., J.P. Walker, and P. R. Houser, 2005: Factors affecting remotely sensed snow water 

equivalent uncertainty, Remote Sensing of Environment 97: 68–82. 

Ferraro, R., C. Kongoli, and P. Pellegrino, 2004: An Evaluation of a New AMSU-Derived Falling 

Snow Retrieval Algorithm (Preprint), 13 th Conference on Satellite Meteorology and 

Oceanography, American Meteorological Society, 20–23 September, 2004, Norfolk, VA. 

Ferraro, R. R., F. Weng, N.C. Grody, I. Guch, C. Dean, C. Kongoli, H. Meng, P. Pellegrino, and 

L. Zhao, 2002: NOAA satellite-derived hydrological products prove their worth, Eos Trans. 

AGU, 83: 429–437. 

Foster J.L., C. Sun, J.P. Walker, R. Kelly, A. Chang, J. Dong, and H. Powell, 2005: Quantifying 

the uncertainty in passive microwave snow water equivalent observations, Remote Sensing of 

Environment 94: 187–203. 

Grody, N.C., 1991: Classification of snow cover and precipitation using the special sensor 

microwave imager, J. Geophys. Res., 96(D4): 7423–7435. 

Grody, N.C., and A.N. Basist, 1996: Global identification of snow cover using SSM/I 

Measurements, IEEE Trans. Geosci. Remote Sens. 34(1): 237–249. 

102

Helfrich, S.R., D. McNamara, B.H.Ramsay, T. Baldwin, and T. Kasheta (2006), Enhancements 

and forthcoming developments to the Interactive Multisensor Snow and Ice Mapping System 

(IMS), Hydrological Processes (this issue). 

Kongoli, C., N.C. Grody, and R. Ferraro, 2004: Interpretation of AMSU measurementsfor the 

retrievals of snow water equivalent and snow depth, J. Geophys. Res, 109: 

do:10.1029/2004JD004836. 

Kongoli, C., and R. Ferraro, 2004: Development and evaluation of the AMSU-based SWE product 

(Preprint), 13 th Conference on Satellite Meteorology and Oceanography, American 

Meteorological Society, 20–23 September, 2004, Norfolk, VA. 

Kongoli, C., R. Ferraro, P. Pellegrino, and H. Meng, 2005 (Preprint): Snow Microwave Products 

from the NOAA’s Advanced Microwave Sounding Unit, 19 th Conference on Hydrology, 

American Meteorological Society, 8–14 January, 2005, San Diego, CA. 

Kongoli, C., P. Pellegrino, R.R. Ferraro, C. Grody, and H. Meng, 2003: A new snowfall detection 

algorithm over land using measurements from the Advanced Microwave Sounding Unit 

(AMSU), Geophys. Res. Let. 30(14): 1756–1759. 

Matzler, C., 1994: Passive microwave signatures of landscapes in winter, Meteorol. Atmos. Phys. 

54: 241–260. 

Meng H., L. Zhao, R. Ferraro, F. Weng, Q. Liu, 2006 (Preprint): Calibration andvalidation of N- 

18 AMSU-A and MHS, 14 th Conference on Satellite Meteorology and Oceanography, 

American Meteorological Society, 28 January–2 February, 2006, Atlanta, GA. 

Ramsay, B.H., 1998: The interactive multisensor snow and ice mapping system, Hydrological 

Processes, 12:1537–1546. 

Ramsay, B.H., 2000: Prospects for the Interactive Multisensor Snow and Ice Mapping System 

(IMS), Proceedings of the 57 th Eastern Snow Conference, 18–19 May 2000, Syracuse,NY, 

161–170. 

Romanov, P. G. Goodman and I. Csiszar, 2000: Automated monitoring of snow cover over North 

America with multispectral satellite data, J. Appl. Meteorol., 39: 1866–1880. 

Sturm, M., J. Holmgren, and G.E. Liston, 1995: A seasonal snow cover classification system for 

local to global applications, Journal of Climate, 8(5): 1261–1283. 

103


104

105 



Time Series Analysis and Algorithm Development 

for Estimating SWE in Great Lakes Area Using Microwave Data 

ABSTRACT 

AMIR E AZAR, 1 HOSNI GHEDIRA 1 , PETER ROMANOV 2 , 

SHAYESTEH MAHANI 1 , AND REZA KHANBILVARDI 1 

The goal of this study is to develop an algorithm to estimate Snow Water Equivalent (SWE) in 

Great Lakes area based on a three-year of SSM/I dataset along with corresponding ground truth 

data. The study area is located between latitudes 41N and 49N and longitudes 87W and 98W. The 

area is covered by 28*35 SSM/I EASE-Grid pixels with spatial resolution of 25km. Nineteen test 

sites were selected based on seasonal average snow depth, land cover type. Each of the sites 

covers an area of 25km*25km with minimum of one snow reporting station inside. Two types of 

ground truth data were used: 1) point-based snow depth observations from NCDC; 2) grid based 

SNODAS-SWE dataset, produced by NOHRSC. To account for land cover variation in a 

quantitative way a NDVI was used. To do the analysis, three scattering signatures of GTVN 

(19V–37V), GTH (19H–37H), and SSI (22V–85V) were derived. The analysis shows that at lower 

latitudes of the study area there is no correlation between GTH and GTVN versus snow depth. On 

the other hand SSI shows an average correlation of 75 percent with snow depth in lower latitudes 

which makes it suitable for shallow snow identification. In the model development a non-linear 

algorithm was defined to estimate SWE using SSM/I signatures along with the NDVI values of the 

pixels. The results show up to 60 percent correlation between the estimated SWE and ground truth 

SWE. The results showed that the new algorithm improved the SWE estimation by more than 20 

percent for specific test sites. 

Keywords: Microwave SSM/I, NDVI, SWE. 

INTRODUCTION 

Knowing the seasonal variation of snowcover and snowpack properties is of critical importance 

for an effective management of water resources. Satellites operating in the optical wavelength 

have monitored snowcover throughout the Northern Hemisphere for more than thirty years. These 

sensors can detect snowcover during daylight and cloud-free conditions. In contrast to visible 

bands, remote measurements operation in microwave region offers the potential of monitoring the 

snowpack water equivalent and wetness due to penetrating capability of the radiation at these 

frequencies. Hallikainen et al. (1984) introduced an algorithm for estimating SWE using passive 

microwave Scanning Multi-channel Microwave Radiometer (SMMR) data. The process involved 

the subtraction of a fall image from a winter image in vertical polarization of 18 and 37 GHz 

frequencies. The difference, ΔT, was used to define linear relationships between ΔT and SWE. 

Aschbacher (1989) proposed an SPT algorithm for estimating snow depth and snow water 

equivalent that was based on a combination of SSM/I channels. Further studies revealed that since 

1 NOAA-CREST, City University of NY, 137 th St & Convent Ave. New York, NY. 

2 NOAA World Weather Building, 5200 Auth Rd, Camp Springs, MD.

land cover is not considered as part of equation in the algorithm, the model is not very accurate. 

Chang et al. (1987) related the difference between the SMMR brightness temperatures in 37 GHz 

and 18 GHz channels to derive snow depth – brightness temperature relationship for a uniform 

snow field, SD=1.59 [Tb 18H–Tb37H]. Goodison and Walker (1995) introduced another 

algorithm to estimate SWE using SSM/I channels. They used vertical gradient (GTV) between 

brightness temperatures at 37 GHz and 19 GHz and defined a linear relationship between SWE and 

GTV. This gradient value is obtained by subtracting the brightness temperature, Tb at frequencies 

of 37 and 19 GHz and dividing it by a constant (Goodison, Walker 1995). Grody (1996) 

developed an image classification algorithm to generate global snow map from Special Sensor 

Microwave Imager (SSM/I) data. The algorithm employs a decision tree technique and uses 

thresholds to filter out precipitation, warm desert, cold desert and frozen surfaces. De Seve et al. 

(1997) applied two previously developed models by Hallikainen and Goodison-Walker to James 

Bay area in La Grande River watershed, Quebec, Canada to estimate SWE using SSM/I images. 

The investigations revealed that both models tend to underestimate SWE especially when SWE 

was more than 200mm. A modified version of Goodison-Walker algorithm was suggested. Foster 

et al. (1999) have modeled various snow crystals shapes in different sizes and concluded that the 

shape of the crystal has little effect on the scattering in microwave. A physically based snow 

emission model was introduced by Pulliainen et al. (1999) of Helsinki University of technology 

(HUT snow emission model). The model assumes that scattering of the microwave radiation 

inside the medium is mostly in forward direction. The scattering coefficient is weighted by an 

empirical factor. The brightness temperature is computed by solving the radiative transfer 

equation. A boreal forest canopy model proposed by Kurvonen et al. (1994) was used to account 

for the influence of vegetation on the brightness temperature. Atmospheric effects were neglected 

and the snow grain size was allowed to vary in the model. Derksen (2004) carried out a detailed 

evaluation of SWE and SCE derived using SMMR and SSM/I data over the south Central part of 

Canada. The new technique to infer SWE from satellite data incorporated different algorithms, 

open environments, deciduous, coniferous, and spars forest cover and calculated SWE as weighted 

average of all four estimates. SWE = FDSWED + FC SWEC + FS SWES + FOSWEO, where (F) is the 

fraction of each land cover type within a pixel, D, C, S, and O correspondingly represent 

deciduous forest, coniferous forest, S sparse forest, and O open prairie environments. Passive 

microwave dataset and in situ SWE observation were compared and showed that the SMMR 

brightness temperature adjustments are required to produce SWE that would fit SWE inferred 

from SSM/I. SWE and SCE time series for December through March for a period of 88 years were 

analyzed to examined the variability of SWE and SCE (Derksen, 2004). Tedesco et al. (2004) 

proposed an Artificial Neural Network (ANN) technique for the retrieval of SWE from SSM/I 

data. They have used a multilayer perceptron (MLP) with various inputs to estimate SWE. First, 

brightness temperatures simulated by means of HUT snow estimation model were employed. The 

second approach made use of a subset of measured values. The input layer consists of four 

neurons, made up of 19 and 37 GHz vertical and horizontal brightness temperatures and the output 

was snow parameters. The results showed higher performance of ANN model compare to other 

methods. In 2005 Derksen conducted a study to assess the accuracy of an inter-annually consistent 

zone of high passive microwave derived SWE retrievals co-located with the Canadian northern 

boreal forest, using extended transects of in situ snow cover measurements (Derksen, 2005). The 

research conducted by Kelly et al. (2001) was focused at the development of a global snow 

monitoring for The Advanced Microwave Scanning Radiometer – EOS (AMSR-E) onboard Aqua 

satellite. The proposed algorithm had the following form: SWE (mm) = B*(TbH18–TbH37), 

where TbH18 and TbH37 are horizontal polarized brightness temperature at 18 and 37 GHz and B 

coefficient has been calibrated as 4.8 mm K –1 for SMMR data. Latter Kelly et al. (2003) described 

the development and testing of an algorithm to estimate global snow cover volume from 

spaceborne passive microwave, AMSR-E. 

The above algorithms used the spectral difference between microwave channels from various 

sensors to estimate SWE or snow depth. However other snow or land parameters such as snow 

grain size and land cover type and conditions have effects on scattering in microwave. Although, 

some researchers introduced land cover type to their models but their algorithms were developed 

106

and validated regionally so they can not be used for other study areas. In addition, these 

algorithms use a multi-regression approach to account for the land cover type variation. Then, 

development of an algorithm which considers variation of land cover quantitatively and can be 

used and validated for different areas is necessitated. Normalized Difference Vegetation Index 

(NDVI) has been widely used to represent the health and greenness of the vegetation. In this study 

a non-linear algorithm is proposed, which estimates SWE using spectral difference between SSM/I 

channels along with NDVI data. 

DATA USED 

SSM/I Data 

SSM/I passive microwave radiometer with seven channels is operating at five frequencies (19, 

35, 22, 37.0, and 85.5 GHz) and dual-polarization (except at 22GHz which is V-polarization only). 

The sensor spatial resolution varies for different channels frequencies. In this study Scalable Equal 

Area Earth Grid EASE-Grid SSM/I products distributed by National Snow and Ice Data Center 

(NSIDC) were used. EASE-Grid spatial resolution is slightly more than 25km (25.06km) for all 

the channels (NSIDC) although the recorded resolution of the microwave spectrum with longer 

wavelengths is more than 50km. The three EASE-Grid projections comprise two azimuthal equalarea 

projections for the Northern or Southern hemisphere, respectively and a global cylindrical 

equal area projection. In this we study have used a Northern hemisphere azimuthal equal-area. 

Normalized Difference Vegetation Index (NDVI) 

NDVI is typically used to represent the vegetation cover properties and it defines as a difference 

between reflectance in visible and near infrared spectral bands divided by their sum (NDVI = 

(NIR – VIS)/(NIR + VIS)). The NDVI data for this study were obtained from the NOAA/NASA 

Pathfinder Advanced Very High Resolution Radiometer (AVHRR) which is distributed at 

Goddard Space Flight Center (GSFC). The spatial resolution is 8km * 8km obtained within a 10 

day period that has the fewest cloud. To facilitate the comparison and matching of the two datasets 

(NDVI and SSM/I) NDVI data resampled and were brought to the same EASE-Grid projection at 

25km spatial resolution. 

Ground Truth Snow Data 

Point Gauge Measurements 

Surface observations of snow depth data for the study were obtained from point the National 

Climate Data Center (NCDC). The point measurements were averaged and gridded to 25km 

spatial resolution to match EASE-Grid SSM/I spatial resolution. To increase the reliability and 

avoid errors due to interpolation, we have used only those pixels where station data were 

available. If more than one station data were available in a given SSM/I pixel, the station data 

were averaged. 

SNODAS SWE 

Snow products generated by the Snow Data Assimilation System (SNODAS) of NOAA 

National Weather Service's National Operational Hydrologic Remote Sensing Center (NOHRSC) 

are available since October 2003. SNODAS includes a procedure to assimilate airborne gamma 

radiation and ground-based observations of snow covered area and snow water equivalent, 

downscaled output from Numerical Weather Prediction (NWP) models combined in a physically 

based, spatially distributed energy and mass balance model. The output products have 1km spatial 

and hourly temporal resolution. In order to match the EASE-Grid pixels the SNODAS SWE data 

were averaged to 25km. 

107

ANALYSIS OF THE DATA 

The study area is located in Great Lakes area between latitudes 41N and 49N and longitudes 

87W and 98W. It covers parts of Minnesota, Wisconsin and Michigan states. The area has various 

land covers (Fig 1). The area is covered by 980 (28 by 35) EASE-Grid pixels. To do the time 

series analysis 19 test sites were selected. Each site, 25km*25km, represents an SSM/I pixel. The 

sites were selected based on their latitude and their land cover type along with the annual snow 

accumulation. In order to avoid wet snow conditions we used the data starting December 1 of each 

year to the February 28 of the year after. Three 90 day sets of data were derived for each winter. 

Table 1 shows geographical location of the selected sites and their NDVI characteristics 

including the mean value and variance. High variance of NDVI indicates substantial changes. 

Table 1. Coordinates of selected pixels along with NDVI values 

Test Site 

SSM/I Pixels 

Latitude Longitude 

NDVI= (P–128) × 0.008 

P value P mean P variance 

1 42.33 –93.62 123.00 123.86 1.07 

2 42.89 –91.97 123.00 123.20 0.45 

3 43.63 –91.43 124.00 124.86 2.27 

4 44.14 –90.57 145.00 143.11 6.19 

5 44.39 –89.12 128.00 132.00 4.24 

6 45.12 –89.11 130.00 132.22 5.14 

7 46.07 –88.19 162.00 155.33 13.53 

8 45.59 –88.21 152.00 157.67 9.06 

9 46.09 –88.79 160.00 161.44 9.10 

10 46.80 –88.46 166.00 152.56 10.63 

11 46.80 –88.16 150.00 157.22 7.95 

12 46.83 –89.69 159.00 158.89 10.53 

13 45.36 –91.18 132.00 132.78 7.50 

14 45.56 –92.68 125.00 130.89 3.66 

15 47.26 –92.78 149.00 143.33 5.10 

16 48.01 –91.88 148.00 154.56 9.82 

17 47.92 –94.08 135.00 137.89 6.55 

18 48.40 –95.99 135.00 132.33 4.36 

19 47.47 –97.47 125.00 124.78 0.67 

108 

Dry land and 

Cropland 

Cropland/Grassland 

Cropland 

/Woodland 

Mixed Forest 

Deciduous Broadleaf 

Evergreen 

Needleleaf 

Urban and Built-up 

Land 

Water Body 

Figure 1: Land Cover Image According to USGS National Atlas of Land Cover Characteristics

109 

Test Site 

Figure 2. Variation of SWE for winter season 2003–2004 and the corresponding Box plot 

In this study we verify the correlation between SSM/I channels and snow depth and SWE for 

different types of land cover located in different latitudes. Three series (2002–2004) of SSM/I 

channels versus snow depth and SWE for each of the selected sites were derived. The SSM/I data 

were obtained from the descending pass of Defense Meteorological Satellite Program (DMSP) 

satellites. There are three SSM/I scattering signatures used in this analysis. The first scattering 

signature named GTH (19H–37H) is the difference between 19 and 37 GHz in horizontal 

polarization. The second signature, GTVN (19V–37V) shows the discrepancy between vertically 

polarized 19 and 37 GHz. Finally, SSI (22V–85V) presents the difference between 22 and 85 GHz 

in vertical polarization. SSI can be used to identify shallow snowcover. The Box plot of the 

signatures GTH and GTVN for winter season 2003–2004 is shown in figure 3. The outliers in the 

Box plots range from –20 to 20 are due to either sensor or data processing errors which need to be 

eliminated. GTVN mean ranges from 5 to 15 for all the pixels except for the test site 12 which is 

very close to the lake. GTH Box plot and mean has the same pattern as GTVN. The test site 12 has 

a mean around –5 for GTVN (19V–37V) indicating of the fact that part of SSM/I pixel is water. 

Since the SSM/I sensor have different spatial resolution for various channel, the scattering from 

pixel 12 is disturbed in channel 19GHz (69km resolution) by the lake however it is not disturbed 

in channel 37GHz (37km resolution). The low scattering of the water and high scattering of the 

land make the difference of 19V–37V a negative number. 

GTVN (19H-37H) (k) 

Box Plot of GTVN for Points 1-19 Box Plot of GTH for Points 1-19 

Test Site 

GTH (19H-37H) (k) 

Test site 

Figure 3. Box-Whiskers plot of GTVN and GTH for winter season 2003–2004

After elimination of the outliers and negative signatures, a three year time series of GTVN and 

snow depth for each of the test sites was produced. Figure 4 illustrates the trend of SSM/I 

signature of GTVN (19V–37V) versus snow depth at site 9. The plot shows that the discrepancy 

GTVN (19V–37V) increases with increasing snow during the winter seasons. This is due to the 

high sensitivity of channel 37GHz to snow. Contradictory to the high latitudes, low latitude pixels 

do not show a consistent seasonal pattern for snow depth and GTVN (Figure 5). 

GTVN (K) 

GTVN (K) 

20 

10 

0 

0 100 200 300 400 500 

Days 

600 700 800 900 1000 

0 

20 

10 

110 

GTVN 

Snow Depth 

Figure 4. Three year time series of GTVN (19V–37V) vs. Snow Depth for point 9 

GTVN 

Snow Depth 

0 

0 100 200 300 400 500 

Days 

600 700 800 900 1000 

0 

Figure 5. Three year time series of GTVN (19V–37V) vs. Snow Depth for point 2 

The above figures indicate high correlations between snow depth and GTVN for high latitudes 

and lack of correlation for sites located in low latitudes. To quantify the relationships, correlation 

coefficient of various SSM/I scattering signature versus snow depth and SWE are presented in 

figure (6). SSM/I signatures at sites 11, 12, and 13 do have correlation with snow depth since the 

scattering is disturbed by water bodies. A graphical representation of correlation coefficients for 

GTVN, GTH, and SSI for all the test sites is shown in figure 6. 

800 

600 

400 

200 

800 

600 

400 

200 

Snow Depth (mm) 

Snow Depth (mm)

Correlation Coe. 



GTVN vs. Snow Depth SSI vs. Snow Depth GTH VS Snow Depth 

0.90 

0.80 

0.70 

0.60 

0.50 

0.40 

0.30 

0.20 

0.10 

0.00 

Winter 01-02 

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 

0.90 

0.80 

0.70 

0.60 

0.50 

0.40 

0.30 

0.20 

0.10 

0.00 

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 

0.90 

0.80 

0.70 

0.60 

0.50 

0.40 

0.30 

0.20 

0.10 

0.00 

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 

111 

Test Site 

Test Site 

Test Site 

Winter 02-03 

Winter 03-04 

Figure 6. Correlations of snow depth vs. SSM/I signatures GTVN (19v–37v), GTH (19h-37h), and SSI (22v- 

85v) for various test sites (TS) for winter seasons 01–02, 02–03, 03–04 

The correlation coefficients between snow depth and scattering signatures follow a consistent 

pattern for all the winter seasons. For all the sites GTVN and GTH show the same correlation with 

snow depth. In other words, the difference between vertically and horizontally polarized signatures 

is negligible in terms of correlations with snow depth. Contrary to GTVN and GTH, SSI has a 

different pattern. It has the dominant correlation for test sites 1 to 5 but for sites located in high 

latitudes GTVN becomes the dominant. This is because of the saturation of the 85GHZ channel 

over a deep snow pack. SSI can be used to identify and to estimate SWE over shallow snow. In 

case of SWE and SSM/I signatures, Figure (7) illustrates the correlations between SWE and 

different SSM/I spectral signatures. For test sites 1 to 5 SSI has the higher correlation but for the 

other sites GTVN and GTH show better correlations with SWE. Figure 7 also shows that the 

correlations between SWE and scattering signatures are higher than those for snow depth. This 

indicates that SSM/I signatures can be a better estimator of SWE than of the snow depth.


1 

0.9 

0.8 

0.7 

0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

0 

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 

Points 

GTVN vs SWE SSI vs SWE GTH VS SWE 

112 

Winter 03-04 

Figure 7. Variation of correlations of SWE vs SSM/I scattering signatures for various points for winter 03–04 

The scatter plots of SWE versus the three SSM/I signatures (GTVN, GTH, SSI) have been 

produced for the all the test sites. Figure 8 illustrates the variation of SWE versus scattering 

signatures for selected test sites (2, 9, and 18). 

RESULTS OF THE ANALYSIS 

The presented correlation coefficients in figures 6, 7 represent different winter seasons from 

2001–2004. The analysis of the results indicates the following: 

1) For test sites located in low latitudes, below 45N, (1, 2, 3, and 4) only SSI exhibits some 

correlation with the snow depth. There is no noticeable correlation of GTH and GTVN vs. the 

snow depth. This is due to the saturation saturation effect in of channel 85GHz which makes SSI 

only suitable for estimating properties of a shallow snow pack. 

2) Sites located in midlatitudes, 45N–46N, (sites 6, 7, 8, 9) there is some correlation between 

GTH and GTVN vs. snow depth but SSI shows no correlation with the snow depth. 

3) No correlation is observed at sites that are very close to the lake (10, 11, 12, 13, and 14). This 

is due to the different spatial resolution of SSM/I in various spectral bands. The sensors field of 

view increases from 37km at 37GHz to 69km for 19GHz. Therefore if a measurement is made 

close to the lake, the effect of the open water may be different in two channels. 

4) Test sites located in forested areas away from the lake show moderate correlations of snow 

with GTH and GTVN. In addition, scatter plots show an attenuation of brightness temperature due 

to forested land cover. 

5) Both GTH and GTVN show high correlations with physical characteristics of the snow pack 

which makes them good potential estimators for snow depth and SWE. The highest correlations 

are observed in north of the US which is due to larger amount of seasonal snow, colder weather 

and less number of freeze/thaw events during a winter season. 

6) Table 3 presents the correlation between SSM/I spectral signatures and SWE obtained from 

SNODAS. The results show higher and more consistent correlation coefficients for SWE than for 

snow depth. 

7) Figure (8) shows the scatter plots of SSM/I Signatures versus SWE (SNODAS) and SSM/I 

Signatures versus Snow Depth (stations) for winter 2003–2004 in different test sites. Lines fitted 

to each graph have various slopes and intercepts. This demonstrates that having one linear 

algorithm (e.g. Chang or Goodison-Walker) may not be enough for snow depth or SWE in a 

variety of environmental and geographical conditions.

Figure 8. SSM/I scattering signatures signature (GTH [19H–37H], GTVN [19H–37V], and SSI [22H-85H]) 

vs. SWE (SNODAS) for winter 2003–2004 

113

ALGORITHM DEVELOPMENT 

The study area, Great Lakes area, is located in the transitional zone for snow which experiences 

snow melting during the winter season. In addition to complexity of snow characteristics, 

variability of the land cover types makes it more difficult to accurately estimate SWE from passive 

microwave observations with a single linear model such as the one of Chang or Goodison-Walker 

(Chang et al, 1987, Goodison-Walker 1995). Figure (8) shows that the slope of the best fitted 

equations varies for different test sites (2, 9, and 18). These test sites have different land cover 

type and different NDNI values. In order to quantitatively account for scattering attenuation 

originating from forested areas, NDVI values are suggested to use. Figure (9) illustrates the 

variation of slope and NDVI for winter 2003–2004 for all the sites. Both NDVI and the slope have 

the same trend. This indicates that the increase of NDVI makes the slope of the relationship 

steeper. 

16 

14 

12 

10 

Slope (m) 

8 

6 

4 

2 

Winter 03-04, Variation of NDVI with the slope of SWE vs GTVN 

Slope 

NDVI 

0 

-0.1 

0 5 10 15 20 

Test site 

0.3 

0.25 

0.2 

0.15 

NDVI 

0.1 

0.05 

0 

-0.05 

114 

6 

5 

4 

Slope 

3 

2 

1 

0 

03-04 Slopes 

SD Slopes and NDVI 

0 5 10 15 20 

Test site 

0.15 

NDVI 

Figure 9. Variations of the slope of the bets fitted line to scatter plots with NDVI for the test sites for winter 

03–04, SWE vs. GTVN (left), SD vs. GTVN (right) 

The scatter plots of slope versus NDVI for SWE and snow depth are illustrated in figure 10. The 

modified scatter plot for 2002–2003 is for the test sites that snow depth vs scattering signature 

GTVN (19v–37v) showed correlation coefficient more than 50 percent. 

Slope 

slope 

16 

14 

12 

10 

8 

6 

4 

2 

0 

4 

3.5 

3 

2.5 

2 

1.5 

1 

0.5 

0 

Winter 03–04 , SWE (mm) Winter 03–04, Snow depth (cm) 

0 0.05 0.1 0.15 0.2 0.25 0.3 

NDVI 

slope 

6 

5 

4 

3 

2 

1 

0 

NDVI 

0.3 

0.25 

0.2 

0.1 

0.05 

0 

-0.05 

0 0.05 0.1 0.15 0.2 0.25 0.3 

Winter 02–03, Snow depth (cm) Winter 02–03, Snow depth (cm), Modified 

0 0.05 0.1 0.15 0.2 0.25 0.3 

NDVI 

4 

3.5 

3 

2.5 

2 

1.5 

1 

0.5 

NDVI 

0 

0 0.05 0.1 0.15 0.2 0.25 0.3 

Figure 10. Variation of the derived slope from the scatter plots vs. NDVI 

slope 

NDVI 

-0.1

Considering the facts mentioned above we propose a new algorithm which relates SWE and the 

SSM/I scattering signature GTVN (19v–37v) and accounts for possible variation of NDVI: 

SWE=F* (A* NDVI* + B)*GTVN, While NDVI >= 0 

SWE=C*SSI+D While NDVI


Snow Depth (cm) 



100 

90 

80 

70 

60 

50 

40 

30 

20 

10 

Goodison-Walker Alg. Chang Alg. New Alg. 

Winter 03–04 (SWE) Winter 03–04 (SWE) Winter 03–04 (SWE) 

GOODISON-WALKER Linear Algorithm 

R=0.7734 

RMSE=63mm 

Bias= -47mm 

MAX(E)=32mm 

MAX(S)=173mm 

0 

0 10 20 30 40 50 60 70 80 90 100 

Estimated (mm) 


CHANG Linear Algorithm 

100 

90 

R=0.80306 

80 

RMSE=58mm 

70 


60 

MAX(e)=43mm 

50 

40 

30 

20 

10 

MAX(s)=173mm 

0 

0 10 20 30 40 50 60 70 80 90 100 

Estimated SWE (mm) 

116 


AZAR Non-linear Algorithm 

100 

90 

R=0.7734 

80 

RMSE=30mm 

70 

Bias= 3mm 

60 

MAX(E)=127mm 

50 

40 

30 

20 

10 

MAX(S)=173mm 

0 

0 10 20 30 40 50 60 70 80 90 100 


Winter 03–04 (SD) Winter 03–04 (SD) Winter 03–04 (SD) 


100 

90 

R=0.50634 

80 

RMSE=18mm 

70 


60 

MAX(E)=24mm 

50 

40 

30 

20 

10 

MAX(S)=50mm 

0 

0 10 20 30 40 50 60 70 80 90 100 




100 

90 

R=0.4396 

80 

RMSE=13mm 

70 


60 

MAX(e)=43mm 

50 

40 

30 

20 

10 

MAX(s)=36mm 

0 

0 10 20 30 40 50 60 70 80 90 100 

Estimated (cm) 



100 

90 

R=0.50634 

80 

RMSE=14mm 

70 


60 

MAX(E)=33mm 

50 

40 

30 

20 

10 

MAX(S)=50mm 

0 

0 10 20 30 40 50 60 70 80 90 100 




100 

90 

R=0.69708 

80 

RMSE=9mm 

70 


60 

MAX(E)=27mm 

50 

40 

30 

20 

10 

MAX(S)=36mm 

0 

0 10 20 30 40 50 60 70 80 90 100 




100 

90 

R=0.4396 

80 

RMSE=13mm 

70 


60 

MAX(e)=43mm 

50 

40 

30 

20 

10 

MAX(s)=36mm 

0 

0 10 20 30 40 50 60 70 80 90 100 




100 

90 

R=0.69708 

80 

RMSE=8mm 

70 


60 

MAX(E)=38mm 

50 

40 

30 

20 

10 

MAX(S)=36mm 

0 

0 10 20 30 40 50 60 70 80 90 100 




100 

90 

R=0.62132 

80 

RMSE=7mm 

70 


60 

MAX(E)=36mm 

50 

40 

30 

20 

10 

MAX(S)=38mm 

0 

0 10 20 30 40 50 60 70 80 90 100 




100 

90 

R=0.53409 

80 

RMSE=16mm 

70 


60 

MAX(e)=51mm 

50 

40 

30 

20 

10 

MAX(s)=38mm 

0 

0 10 20 30 40 50 60 70 80 90 100 




100 

90 

R=0.62132 

80 

RMSE=7mm 

70 


60 

MAX(E)=36mm 

50 

40 

30 

20 

10 

MAX(S)=38mm 

0 

0 10 20 30 40 50 60 70 80 90 10 


Figure 11. Comparison of the results for different algorithms for test site 10 (Lat = 48.6N, Lon = –88.46W, 

and NDVI = 0.2)

Besides the temporal validation, the new algorithm was spatially validated for the whole study 

area (Latitudes: 41N to 49N & Longitudes: –87W to –98W). There were eleven days (3 days 

December, 4 days January, and 4 days February) in winter 2003–2004 selected. For those days the 

full coverage of the study area from SSM/I data was available. The ground truth data was obtained 

by averaging NOHRC SNODAS dataset. Figure 12 shows the ground truth and estimated SWE for 

January 25, 2004. 

30 

25 

20 

15 

10 

5 

Chang Estimated SWE (mm) Ground Truth NOHRSC SWE (mm) 

0 

0 5 10 15 20 25 30 35 

30 

25 

20 

15 

10 

5 

Estimated SWE CHANG Algorithm (mm),Jan 25,04 

180 

160 

140 

120 

100 

80 

60 

40 

20 

0 

-20 

30 

25 

20 

15 

10 

117 

5 

SNODAS , Jan25,04 

0 

0 5 10 15 20 25 30 35 

Goodison-Walker Estimated SWE (mm) New Algorithm Estimated SWE (mm) 

Estimated SWE AZAR Algorithm (mm),Jan 25,04 

0 

0 5 10 15 20 25 30 35 

180 

160 

140 

120 

100 

80 

60 

40 

20 

0 

30 

25 

20 

15 

10 

5 

Estimated SWE Goodison Walker Algorithm (mm),Jan 25,04 

0 

0 5 10 15 20 25 30 35 

Figure 12. Comparison of estimated SWE by various algorithms with ground truth data for January 25, 2004 

for the study area (Lat: 41N to 49N & Lon: –87W to –98W) 

The NDVI image of the study area (Fig 13) shows higher values of NDVI around the lake. This 

is the area that both Chang and Goodison-Walker algorithms highly underestimate the SWE (Fig 

12). In contrast, the new non-linear algorithm can estimate SWE in the area in the vicinity of the 

lake with much higher accuracy (Fig 13, 14). The calculated RMSE and correlation coefficient 

(R 2 ) are shown for all the three algorithms. The use of NDVI in the new algorithm results in a 

decrease of the RMSE and the increase of the correlation coefficient. It also increases the range for 

the estimated SWE. Figure 15 demonstrates a consistent improvement in the accuracy of the 

estimated SWE for the winter season of 2003–2004. For all days, application of the new 

developed algorithm results in the highest correlation coefficient between SSM/I and SWE. At the 

same time, the RMSE of SWE derived with the new algorithm is lower for all days but one. There 

80 

60 

40 

20 

0 

180 

160 

140 

120 

100 

80 

60 

40 

20 

0 

180 

160 

140 

120 

100

is a decreasing trend of in correlations and increasing trend in SWE in February. The most 

probable reason for this trend is snow melt. In February, the study area and especially its southern 

part experienced several melts and refreeze of snow. Estimates of snow depth and SWE with 

satellite observations in microwave become practically impossible when snow is wet. 

SNODAS SWE (mm) 

180 

160 

140 

120 

100 

80 

60 

40 

20 

NDVI Image New Algorithm Estimated SWE vs Ground Truth 

0.4 

0.3 

0.2 

0.1 

0 

-0.1 

-0.2 

-0.3 

-0.4 

-0.5 

118 


180 

160 

140 

120 

100 

80 

60 

40 

20 


R=0.69786 

RMSE=19mm 


Ave(E)=14mm 

Ave(S)=15mm 

MAX(E)=181mm 

MAX(S)=177mm 

0 

0 20 40 60 80 100 120 140 160 180 


Figure 13. NDVI image and results of estimated SWE vs. ground truth for January 25, 2004 

Change Alg. Goodison-Walker Alg. 


R=0.53679 

RMSE=24mm 

Ave(E)=22mm 

Ave(S)=15mm 

MAX(e)=72mm 

MAX(s)=177mm 

0 

0 20 40 60 80 100 120 140 160 180 



180 

160 

140 

120 

100 

80 

60 

40 

20 


R=0.51162 

RMSE=23mm 

Ave(E)=18mm 

Ave(S)=15mm 

MAX(E)=51mm 

MAX(S)=177mm 

0 

0 20 40 60 80 100 120 140 160 180 


Figure 14. Results of estimated SWE using Chang and Goodison-Walker algorithm vs ground truth for 

January 25, 2004

RMSE 

Non linear Azar 

Chang 

Goodison-Walker 

Non linear Azar 

Chang 

Goodison-Walker 

CONCLUSIONS 

60 

50 

40 

30 

20 

10 

0 

0.80 

0.70 

0.60 

0.50 

0.40 

0.30 

0.20 

0.10 

0.00 

6-Dec-03 

6-Dec-03 

Days 

Days 

13-Dec-03 

13-Dec-03 

20-Dec-03 

20-Dec-03 

27-Dec-03 

27-Dec-03 

3-Jan-04 


Figure 15. Correlation and RMSE variations for selected days in winter 2003–2004 

A non-linear method was developed to estimate SWE using SSM/I scattering Signatures and 

NDVI. The study has shown that current linear algorithms such as Goodison-Walker and Chang 

algorithms are not sufficient for accurate estimations of SWE. In order to resolve this problem 

three winter seasons were studied. SSM/I data with corresponding snow depth, and snow water 

equivalent (SWE) were used to examine the sensors response to the changes in snow pack 

properties. SSM/I response in GTVN (19V–37V), GTH (19H–37H), and SSI (22V–85V) to snow 

depth or water equivalent changes were analyzed. The analysis has revealed that in low latitudes 

with shallow snow SSI has the highest correlation with SWE. In higher latitudes GTVN and GTH 

are better estimators of SWE however the slope of the relationship between the spectral signature 

and SWE varies with location. It is found that the variation of the slope of this relationship is 

correlated with NDVI. This fact was used to propose the new algorithm to estimate SWE using 

SSM/I data and NDVI. Validation of the new algorithm has shown that it allows reducing of the 

error of SWE estimates by more than 20 percent as compared to earlier linear algorithms. The 

analysis of derived SWE distributions over the study area has revealed a consistent improvement 

of retrieval accuracy of SWE with the new algorithm. 



119 







7-Feb-04 

7-Feb-04 

14-Feb-04 

14-Feb-04 

21-Feb-04 

21-Feb-04

ACKNOWLEDMENTS 

The authors express their gratitude to Dr C. Derksen of Meteorological Service of Canada 

(MSC) and Dr A. Frei of the City University of New York. The SNODAS datasets produced by 

NOHRSC, were obtained through NSIDC. Thanks to Dr T. Carrol at NOHRSC and L. Ballagh at 

NOAA at NSIDC, University of Colorado. 

REFRENCES 

Aschbacher, J. (1989). Land surface studies and atmospheric effects by satellite microwave 

radiometry, university of Innsbruk. 

Chang A.T.C., J. L. F., and D.K.Hall. (1987). "Nimbus-7 SMMR derived global snow cover 

parameters." Annals Glaciology 9: 39–44. 

Derksen, R. B., A. Walker (2004). "Merging Conventional (1915–92) and Passive Microwave 

(1978–2002) Estimates of Snow Extent and Water Equivalent over Central North America." 

Journal of Hydrometeorology 5: 850–861. 

Derksen C., A. W., B.Goodison (2005). "Evaluation of passive microwave snow water equivalent 

retrievals across the boreal forest/tundra transition of western Canada." Remote Sensing of 


De Seve D., B. M., Fortin J. P. , and Walker A., (1997). "Preliminary analysis of snow microwave 

radiometry using the SSM/I passive-microwave data: The case of La Grande River watershed 

(Quebec)." Annals of Geology 25. 

Foster J. L., H. D. K., Chang A.T., Rango A., Wergin W., and Erbe E., (1999). "Effect of Snow 

Crystal Shape on the Scattering of Passive Microwave Radiation." IEEE Transaction on 

Geosciences and Remote sensing 37(2). 

Grody N., B. N. (1996). "Global Identification of Snowcover Using SSM/I Measurements." IEEE 

Transaction on Geosciences and Remote sensing 34(1). 

Hallikainen, M. T. (1984). "Retrieval of snow water equivalent from Nimbus-7 SSMR data: effect 

of land cover categories and weather conditions." IEEE Oceanic Eng 9(5): 372–376. 

Kelly R.E., A. T. C., L.Tsang, J.L.Foster (2003). "A prototype AMSR-E Global Snow Area and 

Snow depth Algorithm." IEEE Transaction on Geosciences and Remote Sensing, 41(2): 230– 

242. 

Kelly R.E.J. , A. T. C. C., J.L. Foster and D.K. Hall (2001). Development of a passive microwave 

global snow monitoring algorithm for the Advanced Microwave Scanning Radiometer-EOS. 

Kurvonen, L. (1994). Radiometer measurements of snow in Sodankyla, Helsinki University of 

Technology, Laboratory of Space Technology. 

Pullianen, J. T., Grandell J., and Hallikainen M (1999). "HUT snow emission model and its 

applicability to snow water equivalent retrieval." IEEE Transaction on Geosciences and 

Remote sensing 37(3): 1378–1390. 

Tedesco M., P. J., Takala M., Hallikainen M., and Pampaloni P (2004). "Artificial neural networkbased 

techniques for the retrieval of SWE and snow depth from SSM/I data." Remote sensing 

of Environment 90. 

Walker, G. B. E. a. A. E. (1995). Canadian development and use of snow cover information from 

passive microwave satellite data. Passive microwave remote sensing of land–atmosphere 

interactions. B. J. Choudhury, Y. H. Kerr, E. G. Njoku and P. Pampaloni. The Netherlands, 

VSP BV 245–262. 

120

63 nd EASTERN SNOW CONFERENCE 


On the Evaluation of Snow Water Equivalent Estimates over the 

Terrestrial Arctic Drainage Basin 

ABSTRACT 

Michael A. Rawlins 1 , Mark Fahnestock 1,2 ,SteveFrolking 1,2 , 

Charles J. Vörösmarty 1,2 

Comparisons between snow water equivalent (SWE) and river discharge estimates are important 

in evaluating the SWE fields and to our understanding of linkages in the freshwater cycle. In 

this study we compared SWE drawn from land surface models and remote sensing observations 

with measured river discharge (Q) across 179 arctic river basins. Over the period 1988-2000, 

basin-averaged SWE prior to snowmelt explains a relatively small (yet statistically significant) 

fraction of interannual variability in spring (April–June) Q, as assessed using the coefficient of 

determination (R 2 ). Over all river basins, mean R 2 s vary from 0.20 to 0.28, with the best agreement 

noted for SWE drawn from simulations of the Pan-Arctic Water Balance Model (PWBM) that 

are forced with data from the National Center for Environmental Prediction / National Center 

for Atmospheric Research (NCEP-NCAR) Reanalysis. Variability and magnitude in SWE derived 

from Special Sensor Microwave Imager (SSM/I) data are considerably lower than the variability 

and magnitude in SWE drawn from the land surface models, and generally poor agreement is noted 

between SSM/I SWE and spring Q. We find that the SWE vs. Q comparisons are no better when 

alternate temporal integrations—using an estimate of the timing in basin thaw—are used to define 

pre-melt SWE and spring Q. Thus, a majority of the variability in spring discharge must arise 

from factors other than basin snowpack water storage. This study suggests that SWE estimated 

from remote sensing observations or general circulation models (GCMs) can be evaluated effectively 

using monthly discharge data or SWE from a hydrological model. The relatively small fraction of Q 

variability explained by basin SWE warrants further investigation using daily discharge observations 

to more accurately define the snowmelt contribution to river runoff. 

Keywords: SWE; River Discharge; Remote Sensing; SSM/I 

1 

Water Systems Analysis Group, Institute for the Study of Earth, Oceans, and Space, University of New 

Hampshire, Durham, NH 03824 (USA) 

2 Department of Earth Sciences, University of New Hampshire, Durham, NH 03824 (USA) 

121

INTRODUCTION 

Winter snow storage and its subsequent melt are integral components of the climate system. 

Much remains unknown regarding the magnitudes and interannual variations of this key feature of the 

arctic water and energy cycles. Across large parts of the terrestrial Arctic direct snow observations 

are unavailable, and this lack of information limits our ability to monitor a region which is exhibiting 

signs of change (Peterson et al., 2002; Vörösmarty et al., 2001). Yet, amid declines in Pan-Arctic 

station observations (Shiklomanov et al., 2002), a growing number of models and remote sensing 

data are being brought to bear for studying the arctic hydrological cycle. Retrospective analysis 

or “reanalysis” of the atmospheric state such as the National Centers for Environmental Prediction 

(NCEP) and the National Center for Atmospheric Research (NCAR) Reanalysis Project (Kalnay 

et al., 1996) provide benchmark, temporally-consistent data sets for water cycle studies. Remote 

sensing techniques offer the potential for more complete coverage at regional scales (Derksen et al., 

2003; McDonald et al., 2004). 

High quality estimates of snow storage and melt can be used to validate the behavior of hydrological 

models and GCMs, which have difficulty reproducing solid precipitation dynamics (Waliser et al., 

2005). Snow cover and snow water equivalent (SWE) estimates are also needed for climate change 

analysis and flood prediction studies. Approximately 8000 (daily) snow depth observations were 

analyzed to create monthly snow depth and SWE climatologies for North America (Brown et al., 

2003) for use in evaluating Atmospheric Model Intercomparison Project II (AMIP II) snow cover 

simulations. Comparisons of continental-scale snow parameters with river discharge time series are 

useful to improving our understanding of the role of snow accumulation and melt in runoff generation 

processes. Yang et al. (2002), examining the snow-discharge relationship, noted a weak correlation 

(R = 0.14 to 0.27) between winter precipitation (a proxy for snow thickness) and streamflow between 

May and July across the Lena river basin in Siberia. Across the Ob basin, winter snow depth derived 

from Special Sensor Microwave Imager (SSM/I) agrees well with runoff in June (R=0.61), with lower 

correlations for comparisons using May or July discharge (Grippa et al., 2005). Strong links have 

been reported between end-of-winter SWE and spring/early summer river discharge in the Churchill 

River and Chesterfield Inlet Basins of Northern Canada (Déry et al., 2005). Frappart et al. (2006) 

recently compared snow mass derived from SSM/I data and three land surface models with snow 

solutions derived from GRACE geoid data. GRACE (Gravity Recovery and Climate Experiment) 

is a geodesy mission to quantify the terrestrial hydrological cycle through measurements of Earth’s 

gravity field. They found that GRACE solutions correlate well with the high-latitude zones of strong 

accumulation of snow at the seasonal scale. 

To better understand the agreement between SWE and observed river discharge, we examine 

comparisons of their year-to-year changes across 179 river basins over the period 1988–2000. Gridded 

SWE estimates across the Pan-Arctic drainage basin are taken from both satellite microwave data 

and land surface model estimates. The objective of our study is to evaluate several common SWE 

data sets using monthly discharge for watersheds across the terrestrial arctic basin. 


Spatial, gridded estimates of monthly SWE and discharge for river basins across the Pan-Arctic 

were analyzed for the period 1988–2000. Monthly SWE is drawn from the analysis scheme described 

by Brown et al. (2003) and archived at the Canadian Cryospheric Information Network (CCIN, 

122

http://www.ccin.ca); from simulations using the Pan-Arctic Water Balance Model (PWBM) (Rawlins 

et al., 2003); from snowpack water storage in the Land Dynamics Model (LaD) (Milly and 

Shmakin, 2002); and from SSM/I brightness temperatures (Armstrong and Brodzik, 1995; Armstrong 

et al., 2006). PWBM uses gridded fields of plant rooting depth, soil characteristics (texture, 

organic content), vegetation, and is driven with daily time series of climate (precipitation (P )andair 

temperature (T )) variables. Monthly PWBM SWE is obtained from model runs using P and T from 

3 different sources (i) ERA-40 (ECMWF, 2002), (ii) NCEP-NCAR Reanalysis (NNR) (Kalnay et al., 

1996), (iii) Willmott-Matsuura (WM) (Willmott and Matsuura, 2001). We refer to these SWE estimates 

as PWBM/ERA-40, PWBM/NNR, and PWBM/WM, respectively. The NNR P data have 

been adjusted based on a statistical downscaling approach (Serreze et al., 2003). Implemented in an 

effort to minimize biases through the use of observed P data, this method involved (1) interpolation 

of observed monthly totals from available station records with bias adjustments and (2) disaggregation 

of the monthly totals to daily totals, making use of daily P forecasts from the NCEP/NCAR 

Reanalysis. PWBM simulations were performed on the 25×25 km Equal Area Scalable Earth Grid 

(EASE-Grid) (Brodzik and Knowles, 2002). The LaD has previously been found to explain half of 

the interannual variance of the runoff/precipitation ratio of 44 major river basins (Shmakin et al., 

2002). In a study of SWE derived from GRACE and from three LSMs, LaD estimates most closely 

matched those from GRACE, with a good correspondence at seasonal time scales (Frappart et al., 

2006). Our analysis also includes SWE (0.25 ◦ resolution for years 1988-1997) from the analysis 

scheme described by Brown et al. (2003) and archived at the Canadian Cryospheric Information 

Network (CCIN, http://www.ccin.ca). LaD SWE at 1 ◦ resolution was mapped to the EASE-Grid 

using inverse-distance weighted interpolation, while CCIN SWE was aggregated to the EASE-Grid 

using the average of all 0.25 ◦ grids falling within each EASE-Grid cell. The PWBM-, LaD-, and 

SSM/I-derived SWE estimates are Pan-Arctic in nature, ie. defined at all 39,926 EASE-Grid cells 

encompassing the Pan-Arctic drainage basin (Figure 1). CCIN SWE are available across EASE-Grid 

cells over North America only. 

Passive microwave radiances from SSM/I—aboard the Defense Meteorological Satellite Program 

satellite series since 1987—have been used to produce maps of SWE across large regions (Armstrong 

and Brodzik, 2001; Armstrong and Brodzik, 2002; Derksen et al., 2003; Goita et al., 2003). Monthly 

SWE estimated from SSM/I radiances for the period 1988–1999 (on the EASE-Grid) were acquired 

from the US National Snow and Ice Data Center (M. J. Brodzik, personal communication, March 

2, 2004) and are archived under the ArcticRIMS project (http://RIMS.unh.edu). The snow depth 

algorithm (Armstrong and Brodzik, 2001) is: snow depth (cm) = 1.59 * [(T19H − 6) − (T37H − 1)], 

where T19H is the brightness temperature at 19 GHz and T19H is the brightness temperature at 37 

GHz. Water equivalent is obtained from the product of snow depth and density. 

Our analysis involves the use of what we term “pre-melt” SWE (the average of February and 

March monthly SWE) and spring total Q (the total discharge flow over the months April–June). We 

chose an average of two months of SWE over one month (or maximal monthly) to better represent 

mid-winter conditions. The SWE and Q time series are prewhitened to remove any trends prior 

to the covariance analysis. The Q records are drawn from an updated version of R-ArcticNET 

(Lammers et al., 2001). Although SWE estimates are available for more recent years, our analysis 

here ends in 2000 due to a lack of more recent river discharge data for river basins across Eurasia. 

Alternate comparisons using SWE and Q which vary depending thaw timing derived from SSM/I 

data, and by simulated snowmelt, are described in the Results section. In this study, all SWE 

data sets have valid data at each grid defining the Pan-Arctic drainage basin. For each of the 179 

river basins, pre-melt SWE is then determined as an average over all EASE-Grid cells defining the 

123

Figure 1: Pan-Arctic land mass (north of 45 ◦ N, dark gray), the arctic drainage basin (light gray), 

and locations of 179 river basins used in the study. Dot sizes are scaled by basin area. A total of 

39,926 EASE-Grid cells comprise the approximately 25 million km 2 drainage basin. Areas for the 

179 river basins range from 20,000 km 2 to 486,000 km 2 . 

respective the basin. 

Satellite-borne remote sensing at microwave wavelengths can be used to monitor landscape 

freeze/thaw state (Ulaby et al., 1986; Way et al., 1997; Frolking et al., 1999; Kimball et al., 2001). 

A step edge detection scheme applied to SSM/I brightness temperatures (McDonald et al., 2004) 

was used to identify the predominant springtime thaw transition event for each EASE-Grid cell. 

As with SWE, we derived a basin average date of thaw by averaging thaw event dates across the 

basin grid cells. Snow thaw across arctic basins often can occur over a period of weeks or months. 

Therefore, for large watersheds, our timing estimates derived from SSM/I brightness temperatrures 

must be interpreted with caution. Nonetheless they provide a general approximation of the timing 

in landscape thaw for use in estimating pre-melt SWE and spring Q. As an illustration, monthly 

river discharge, SWE, and thaw date for the Yukon basin are shown in Figure 2a–d). 

A simulated topological network (Vörösmarty et al., 2000), recently implemented at 6 minute 

resolution, defines river basins over the approximately 25 million km 2 of the Pan-Arctic basin. The 

degree to which SWE and Q covary over the period 1988-2000 is evaluated using the coefficient of 

determination, R 2 (squared correlation). Throughout our analysis we assume a significance level 

of 0.05 (5%) as the cutoff to determine whether a given SWE vs. Q comparison is statistically 

significant, and not due to chance. For a sample size of 13 years this correspondes to R 2 ≥ 0.22. 

RESULTS 

Interannual variability in basin averaged, pre-melt SWE is compared with spring Q for 179 basins 

124

over the period 1988–2000. With the exception of SWE derived from SSM/I data, interannual 

variability in pre-melt SWE agrees well with spring Q variability across the Yukon basin in Alaska 

(Figure 2). Variability in basin SWE from the CCIN analysis scheme explains nearly 75% of the 

variability in spring Q. When reanalysis data drives the PWBM (PWBM/ERA-40 or PWBM/NNR), 

pre-melt SWE explains well over 50% of the variability in spring Q. Across the Yukon basin, the 

greater SWE variability and magnitude (among all SWE products) is noted for LaD SWE, along 

with a lower R 2 (Figure 2c–d). Basin averaged SWE derived from SSM/I, however, shows little 

interannual variability and relatively low magnitude. For snow packs above 100 mm, the bias in 

SWE estimated from Scanning Multichannel Microwave Radiometer (SMMR) data was shown to be 

linearly related to the snow pack mass, with root-mean-square errors approaching 150 mm (Dong 

et al., 2005). 

In contrast to the result over the Yukon basin, strong agreements in pre-melt SWE and spring Q 

variability are not noted across many of the study basins. Although R 2 s for a majority of the North 

American basins are significant (mean values between 0.26 and 0.36, Table 1), agreements across 

eastern Eurasia are generally low (Figure 3, Table 1). Of the comparisons involving SWE from 

PWBM, more than half (130 of 231) are significant. Mean R 2 s from comparisons using the PWBM 

are comparable to those involving CCIN SWE estimates, which were developed using observed snow 

depth observations (Brown et al., 2003). Across all basins analyzed, the highest proportion of 

negative correlations (very poor agreement) and lowest overall R 2 are associated with SSM/I SWE. 

The algorithm used to produce these estimates, like many of the early passive-microwave SWE 

algorithms, tends to underestimate SWE in forested regions. Models which account for the differing 

influences on the microwave signature have shown promise in reducing errors in forested regions 

(Goita et al., 2003). The best agreements involving SSM/I SWE are found across the prairies of 

south-central Canada. This is expected, as the SSM/I SWE algorithm was developed for application 

across the non-forested prairie provinces of Canada. 

Comparisons using SWE from the PWBM simulations (PWBM/ERA-40, PWBM/NNR, and 

PWBM/WM), produce similar R 2 values across each region, with mean value by region ranging 

from 0.15 to 0.36. Given that water budget models like PWBM are most sensitive to time-varying 

climatic inputs (Rawlins et al., 2003), small differences in R 2 among these SWE estimates suggest 

similar spatial and temporal variability among the underlying precipitation data. Basin R 2 s obtained 

from comparisons using LaD SWE are comparable with those from the comparisons using PWBM 

SWE across North America, while lower correlations are noted for Eurasia. Mean R 2 sarehigher 

across eastern Eurasia (east of longitude 90 ◦ E) as compared with western Eurasia. The better 

agreement across Siberia is likely attributable to the higher fraction of precipitation which falls as 

snow and the higher discharge/precipitation ratios across the colder east. When the PWBM is 

driven with precipitation data from a new gauge-corrected archive for the former USSR (“Daily and 

Sub-daily Precipitation for the Former USSR”) (National Climatic Data Center, 2005), basin R 2 s 

are generally no higher (figure not shown). This suggests that precipitation-gauge undercatch is not 

a significant influence on the computed SWE vs. Q agreements. 

Snowmelt and subsequent rises in river Q begins in southerly regions of the terrestrial arctic and 

progresses northward each spring. Comparisons of winter SWE storage and Q over a fixed interval 

(e.g. April–June) are complicated when inputs from rainfall are significant, or a large fraction of the 

snowmelt occurs outside of the April–June period. A more meaningful comparison of SWE and river 

Q wouldberestrictedtothatfractionofQ which is attributable to the melting of snow. For example, 

simulated spring Q from PWBM—driven by ERA-40 data—explains a much higher proportion of 

observed spring Q than does the pre-melt SWE across the study basins (Figure 4, Table 1). The 

125

Figure 2: Monthly total SWE (mm) and mean discharge (Q, mmday −1 ) across the Yukon basin for 

(a) 1992, a year with relatively high SWE and Q, and (b) 1998, a year with low SWE and Q totals. 

Vertical bars show SWE— in this case from PWBM driven with ERA-40 data (PWBM/ERA-40). 

February and March SWE values (gray bars here) are averaged to give “pre-melt” SWE in this 

study. April–June SWE are depicted by white bars. Spring Q (monthly values indicated by dots at 

middle of month) is the integration of the monthly Q for April through June (hatched area), and is 

used for comparisons with the pre-melt SWE. A “thaw date” (marked Thaw) estimated from SSM/I 

data are used in alternate Q integrations. (c) Scatterplot of pre-melt SWE from each data set, for 

years 1988-2000. The best fit line based on linear least squares regression is shown. (d) Time series 

of SWE and Q. Statistics (R 2 and associated p-value) for each covariance comparison are shown in 

parenthesis. 

126

Figure 3: Explained variance (R 2 ) for pre-melt SWE and spring Q comparisons (1988–2000) at the 

179 river basins and for the 6 SWE products. SWE is taken from the CCIN SWE analysis; PWBM 

simulations driven by ERA-40, NNR and WM; LaD model; and SSM/I data. The ’X’s mark basins 

with a negative correlation. Average R 2 values across all basins, North America (NA), western 

Eurasia (WE), and eastern Eurasia (EE) are shown in Table 1. The vertical line in colorbar is level 

(R 2 = 0.22) at which R 2 is significant at 5% level. P-values associated with each R 2 interval are 

shown above the colorbar. Note that p < 0.01 for all R 2 ≥ 0.40. 

127

Feb–Mar Min, Max, and Mean Coefficient of Variation, R2 SWE Data SWE (mm) %neg. All North Am. W. Eurasia E. Eurasia 

CCIN N/A 15.6 N/A 0.00, 0.87, 0.35 N/A N/A 

PWBM/ERA-40 103 8.5 0.00, 0.91, 0.28 0.00, 0.91, 0.36 0.00, 0.45, 0.15 0.00, 0.66, 0.27 

PWBM/NNR 109 12.6 0.00, 0.87, 0.25 0.00, 0.87, 0.33 0.00, 0.56, 0.15 0.00, 0.71, 0.23 

PWBM/WM 109 11.9 0.00, 0.91, 0.26 0.00, 0.91, 0.33 0.00, 0.75, 0.22 0.00, 0.53, 0.17 

LaD 144 20.1 0.00, 0.83, 0.24 0.00, 0.83, 0.33 0.00, 0.49, 0.12 0.00, 0.69, 0.16 

SSM/I 80 72.1 0.00, 0.76, 0.20 0.00, 0.76, 0.26 0.00, 0.40, 0.10 0.00, 0.57, 0.14 

SimRO 103 5.3 0.00, 0.92, 0.46 0.01, 0.91, 0.44 0.00, 0.79, 0.35 0.05, 0.92, 0.57 

PWBM/ERA-40a 103 10.0 0.00, 0.80, 0.27 0.00, 0.80, 0.33 0.00, 0.61, 0.22 0.00, 0.64, 0.22 

PWBM/ERA-40b 103 18.7 0.00, 0.93, 0.34 0.00, 0.93, 0.37 0.00, 0.86, 0.38 0.00, 0.70, 0.26 

PWBM/ERA-40c 103 10.0 0.00, 0.80, 0.27 0.80, 0.00, 0.33 0.61, 0.00, 0.22 0.00, 0.64, 0.22 

PWBM/ERA-40d 103 10.0 0.00, 0.80, 0.27 0.80, 0.00, 0.33 0.61, 0.00, 0.22 0.00, 0.64, 0.22 

PWBM/ERA-40e 103 22.8 0.00, 0.76, 0.25 0.00, 0.76, 0.35 0.00, 0.28, 0.12 0.00, 0.58, 0.19 

128 

Table 1: Mean February-March SWE (mm), percent negative correlations, and minimum, maximum, and mean coefficient 

of determination (R2 ) from the pre-melt SWE and spring Q comparisons. SWE is taken from data sets described in Data 

and Methods and shown in Figure 3. Percentage of negative correlations, and mean explained variance is also tabulated for 

simulated spring Q vs. observed spring Q (row SimRO), where simulated Q is from PWBM/ERA-40 model simulation. Mean 

February-March SWE is averaged across the terrestrial arctic basin, excluding Greenland. Individual R2 values for each study 

basin (shown in Figure 3) are averaged (excluding negative correlations) over all 179 river basins (All), and the basins across 

North America, western Eurasia, and eastern Eurasia, with the latter two separated by the 90◦E meridian. Mean R2 for CCIN 

SWE are determined for North American sector only. PWBM/ERA-40a−e represent the alternate comparisons, defined in 

Results section.

correspondence between simulated and observed spring Q suggests the model—to some degree—is 

accounting for processes connecting the snowpack and spring river flow, e.g. sublimation, rainfall, 

and soil infiltration. 

To better understand the covariance between SWE and Q, alternate comparisons were made 

using PWBM/ERA-40 monthly SWE and an estimate of when thaw is assumed to have occurred. 

The month of thaw (TM,withTM− 1, and TM+ 1 indicating the month preceding and postceding 

the thaw month, respectively) was determined with a step edge detection scheme applied to SSM/I 

brightness temperatures (McDonald et al., 2004). Then, SWETM becomes monthly basin SWE 

during TM (or TM− 1), and QTM is discharge in month TM. These alternate comparisons (across 

all 179 basins) are defined (a) SWETM vs. spring Q, (b)SWETM-1 vs. spring Q, (c)SWETM-1 vs. 

QTM+1, (d)SWETM-1 vs. QTM+1,2. R 2 s are highest for alternate comparison (b), which compared 

SWE in the month before thaw (TM − 1) with spring (April–June) Q (Table 1). Yet, despite the 

fact that the mean R 2 across western Eurasia improves from 0.15 (using default PWBM/ERA-40) 

to 0.38 (alternate comparison b), little difference is noted with the remianing alternate comparisons 

and other regions. 

Lastly, we scaled spring Q using a factor S, whereS = PWBM monthly snow melt–runoff ratio, 

with 0 < S < 1. Then, snowmelt Q each month is Qs =Q·S. Each occurrence of QS was then 

summed resulting in a total QS each spring, for each basin. Using QS in place of the default Q (and 

PWBM/ERA-40 SWE), we note a decrease in agreement across eastern Eurasia, with no change 

across most of the domain. And although SWE from simulations with ERA-40, in general, explains 

more than a third of the variation in Q, a large proportion of the interannual variability is not 

due to SWE variability. When considering these results, it is interesting to note that Lammers et 

al. (2006) recently found that annual simulated discharge across Alaska (drawn from three separate 

models) was in poor agreement with observed discharge data between 1980–2001. Better agreements 

across northwestern North America, eastern Eurasia (EE in Figure 3), and parts of western Eurasia 

(WE) in this study are attributable to relatively higher snowfall rates and a greater interannual 

variability in spring discharge (Figure 4b). Conversely, the region of eastern Eurasia with numerous 

negative correlations is characterized by low spring discharge variability. Delays in snowmelt water 

reaching river systems, which can be significant (Hinzman and Kane, 1991), are likely an additional 

influence on these reported correlations. For large arctic basins, comparisons between snow storage 

and discharge volume are complicated by the large temporal variation in basin thaw and the delays 

in snowmelt water reaching the gauge. More meaningful comparisons between spatial SWE and 

river discharge are possible through the use of hydrograph separation to partition discharge into 

overland and baseflow components. This, however, requires the use of daily discharge data which 

are more limited for the Pan-Arctic region. 

CONCLUSIONS 

In our comparisons of interannual variations in pre-melt SWE and spring Q, R 2 values are 

highest (mean of 0.25 to 0.28 over all basins) when PWBM is driven by ERA-40, NNR or WM 

climate data. Similar agreements are noted when SWE from the observed data analysis scheme are 

used, which suggests that the hydrological model is capturing as much variability in the spring flow 

as does the observed SWE scheme. Average R 2 determined from the SSM/I SWE and spring Q 

comparisons are generally low, and a sizable majority (over 72%) of these correlations are negative. 

The low variability and magnitude is likely related to saturation of the SSM/I algorithm at high SWE 

129

Figure 4: (a) R 2 for PWBM simulated spring total Q vs. observed spring total Q. The vertical 

line in colorbar is level at which R 2 is significant. Minimum, maximum, and mean R 2 sacrossall 

basins, North America (NA), western Eurasia (WE), and eastern Eurasia (EE) are shown in Table 

1. The ’X’s mark basins with a negative correlation. (b) Standard deviation of spring (April–June) 

discharge for the period 1988-2000. 

130

values. Continued development of new regional schemes which account for microwave emmission 

from forests should improve large-scale SWE estimates. Poor agreements among all SWE products 

—particularly across parts of western Eurasia—are noted in areas with low discharge variability. 

Pre-screening to eliminate basins with low flow or insufficient variability would likely improve the 

SWE vs. Q agreements. 

Results of the covariance analysis using alternate temporal integrations to derive pre-melt SWE 

(or Q) suggest that our choice of a fixed interval for spring, ie. April–June, is not the primary cause 

of the relatively low R 2 s. Furthermore, we conclude that much of the interannual variability in river 

discharge must be influenced by factors other than basin SWE storage variations. The unexplained 

variability is likely due to a combination of effects from physical processes (sublimation, infiltration) 

and errors in spatial SWE. Relatively good agreement between simulated and observed spring Q 

suggests that hydrological models can be useful in understanding the SWE-to-Q linkages. Our 

results provide a benchmark of the relationship between the solid precipitation input and spring 

discharge flux, and demonstrate that hydrological models driven with reanalysis data can provide 

SWE estimates sufficient for use in validation of remote-sensing and GCM SWE fields. Additional 

studies using daily discharge data to better quantify snowmelt runoff will further facilitate SWE 

product evaluations and the understanding of linkages in arctic hydrological system. 


The authors thank Richard Lammers, Ernst Linder, and Dominik Wisser (University of New 

Hampshire) for frequent useful discussions, and Kyle McDonald (Jet Propulsion Laboratory) for the 

thaw timing data. We also thank Chris Milly (NOAA/GFDL) for providing the LaD SWE estimates, 

and Ross Brown (Environnement Canada) for the CCIN data. This study was supported by the 

NSF ARCSS program and NSF grants OPP-9910264, OPP-0230243, OPP-0094532, and NASA grant 

NAG5-9617. 

REFERENCES 

Armstrong, R. L. and Brodzik, M. J. 1995. An earth-gridded SSM/I data set for cryospheric studies 

and global change monitoring. Advances in Space Research, 16(10):155–163. 

Armstrong, R. L. and Brodzik, M. J. 2001. Recent Northern Hemisphere snow extent: a comparison 

of data derived from visible and microwave sensors. Geophys. Res. Lett., 28(19):3673–3676. 

Armstrong, R. L. and Brodzik, M. J. 2002. Hemispheric-scale comparison and evaluation of passivemicrowave 

snow algorithms. Annals of Glaciology, 34(1):38–44. 

Armstrong, R. L., Knowles K. W., Brodzik M. J., Hardman M. A. 2006. DMSP SSM/I Pathfinder 

daily EASE-Grid brightness temperatures. Technical report, National Snow and Ice Center. 

CDROM, Boulder, CO, USA. http://nsidc.org/data/nsidc-0032.html. 

Brodzik, M. J. (2004). personal communication, 2 March, 2004. 

Brodzik, M. J. and Knowles K. 2002. EASE-Grid: A versatile set of equal-area projections and 

grids. in M. Goodchild (Ed.) Discrete Global Grids. Santa Barbara, CA, USA: National Center 

for Geographic Information and Analysis. 

Brown, R. D., Brasnett, B., Robinson, D. (2003). Gridded North American monthly snow depth 

and snow water equivalent for GCM evaluation. Atmosphere–Ocean, 41(1):1–14. 

131

Derksen, C., Walker, A., LeDrew, E., Goodison, B. (2003). Combining SMMR and SSM/I data 

for time series analysis of central North American snow water equivalent. J. Hydrometeorol., 

4(2):304–316. 

Déry, S. J., Sheffield, J., Wood, E. F. (2005). Connectivity between Eurasian snow cover extent and 

Canadian snow water equivalent and river discharge. J. Geophys. Res., 110(D23). D23106, 

doi:10.1029/2005JD006173. 

Dong, J., Walker, J. P., Houser, P. R. (2005). Factors affecting remotely sensed snow water equivalent 

uncertainty. Remote Sens. Environ., 97:68–82. 

European Centre for Medium Range Weather Forecasts (ECMWF) (2002). ERA-40 Project Report 

Series. 3. Workshop om Re-analysis, 5-9 november 2001. Technical report, European Centre for 

Medium Range Weather Forecasts. 443 pp. 

Frappart, F., Ramillien, G., Biancamaria, S., Mognard, N. M., Cazenave, A. (2006). Evolution of 

high-latitude snow mass derived from the GRACE gravimetry mission (2002–2004). Geophys. 

Res. Lett., 33. L02501, doi:10.10292005GL024778. 

Frolking, S., McDonald, K. C., Kimball, J., Way, J. B., Zimmermann, R., Running, S. W. (1999). 

Using the space-borne NASA scatterometer (NSCAT) to determine the frozen and thawed 

seasons of a boreal landscape. J. Geophys. Res., 104(D22):27,895–27,907. 

Goita, K., Walker, A. E., Goodison, B. E. (2003). Algorithm development for the estimation of 

snow water equivalent in the boreal forest using passive microwave data. Int. J. Remote Sens., 

24(5):1097–1102. 

Grippa, M., Mognard, N., Toana, T. L. (2005). Comparison between the interannual variability 

of snow parameters derived from SSM/I and the Ob river discharge. Remote Sens. Environ., 

98:35–44. 

Hinzman, L. D. and Kane, D. L. (1991). Snow hydrology of a headwater Arctic basin 2. Conceptual 

analysis and computer modeling. Water Resour. Res., 27(6):1111–1121. 

Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., Iredell, M., Saha, 

S., White, G., Woolen, J., Zhu, Y., Chelliah, M., Ebisuzaki, W., Higgens, W., Janowiak, J., 

Ropelewski, K. C., Wang, J., Leetma, A., Reynolds, R., Jenne, R., Joseph, D. (1996). The 

NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteorol. Soc., 77:437–471. 

Kimball, J. S., McDonald, K. C., Keyser, A. R., Frolking, S., Running, S. W. (2001). Application 

of the NASA scatterometer (NSCAT) for determining the daily frozen and nonfrozen landscape 

of Alaska. Remote Sens. Environ., 75:113–126. 

Lammers, R. B., Rawlins, M., McGuire, D., Clein, J., Kimball, J., Wu, W. (2006). Water budget 

closure over the western arctic and Yukon river basin - a model inter-comparison. Earth 

Interactions. in preparation. 

Lammers, R. B., Shiklomanov, A. I., Vörösmarty, C. J., Fekete, B. M., Peterson, B. J. (2001). 

Assessment of contemporary Arctic river runoff based on observational discharge records. J. 

Geophys. Res., 106(D4):3321–3334. 

McDonald, K. C., Kimball, J. S., Njoku, E., Zimmermann, R., Zhao, M. (2004). Variability in 

springtime thaw in the terrestrial high latitudes: Monitoring a major control on biospheric 

assimilation of atmospheric CO2 with spaceborne microwave remote sensing. Earth Interactions, 

7:1–23. 

Milly, P. C. D. and Shmakin, A. B. (2002). Global modeling of land water and energy balances: I. 

The Land Dynamics (LaD) model. J. Hydrometeorol., 3:283–299. 

132

National Climatic Data Center (2005). Daily and Sub-daily Precipitation for the Former 

USSR. Technical report. Available from National Geophysical Data Center, 

http://www.ncdc.noaa.gov/oa/documentlibrary/surface-doc.html9813. 

Peterson, B. J., Holmes, R. M., McClelland, J. W., Vörösmarty, C. J., Lammers, R. B., Shiklomanov, 

A. I., Shiklomanov, I. A., Rahmstorf, S. (2002). Increasing river discharge to the Arctic Ocean. 

Science, 298:2171–2173. 

Rawlins, M. A., Lammers, R. B., Frolking, S., Fekete, B. M., Vörösmarty, C. J. (2003). Simulating 

pan-Arctic runoff with a macro-scale terrestrial water balance model. Hydrol. Processes, 

17:2521–2539. 

Serreze, M. C., Clark, M. P., Bromwich, D. H., Etringer, A. J., Zhang, T., Lammers, R. (2003). 

Monitoring precipitation over the Arctic terrestrial drainage system: Data requirements, shortcomings, 

and applications of atmospheric reanalysis. J. Hydrometeorol., 4(2):387–407. 

Shiklomanov, A. I., Lammers, R. B., Vörösmarty, C. J. (2002). Widespread decline in hydrological 

monitoring threatens Pan-Arctic research. Eos Trans. AGU, 83(2):13–17. 

Shmakin, A. B., Milly, P. C. D., Dunne, K. (2002). Global modeling of land water and energy 

balances. part III: Interannual variability. J. Hydrometeorol., 3(3):311–321. 

Ulaby, F. T., Moore, R. K., Fung, A. K. (1986). Microwave Remote Sensing: Active and Passive, 

Vol. III – Volume Scattering and Emission Theory, Advanced Systems and Applications. Artec 

House Inc., Dedham, MA. 

Vörösmarty, C. J., Fekete, B. M., Maybeck, M., Lammers, R. B. (2000). Geomorphometric attributes 

of the gloabl system of rivers at 30-min spatial resolution. J. Hydrol., 237:17–39. 

Vörösmarty, C. J., Hinzman, L. D., Peterson, B. J., Bromwich, D. H., Hamilton, L. C., Morrison, 

J., Romanovsky, V. E., Sturm, M., Webb, R. S. (2001). The hydrologic cycle and its role in 

arctic and global environmental change: A rationale and strategy for synthesis study. Technical 

report, Fairbanks, AK: Arctic Research Consortium of the U.S. 

Waliser, D. E., Waliser, S. E., Seo, K., Enjoku, E. (2005). Evaluation and climate change projections 

of the global hydrological cycle in IPCC AR4 model simulations. In Eos Trans. AGU,volume86. 

Fall Meet. Suppl., Abstract H51I-07. 

Way, J. B., Zimmermann, R., Rignot, E., McDonald, K., Oren, R. (1997). Winter and spring thaw 

as observed with imaging radar at BOREAS. J. Geophys. Res., 102(D24):29673–29684. 

Willmott, C. J. and Matsuura, K. (2001). Arctic terrestrial air temperature and precipitation: 

Monthly and annual time series (1930–2000) version 1. available online at: 

http://climate.geog.udel.edu/ climate/. 

Yang, D., Kane, D. L., Hinzman, L. D. (2002). Siberian Lena River hydrologic regime and recent 

change. J. Geophys. Res., 107(D23). 4694, doi:10.1029/2002JD002542. 

133


134

Snow Remote Sensing Posters 

135


136

ABSTRACT 



AMSR-E Algorithm for Snowmelt Onset Detection 

in Subarctic Heterogeneous Terrain 

J.D. APGAR 1 , J.M. RAMAGE 1 , R.A. MCKENNEY 2 , AND P. MALTAIS 3 

Snowmelt onset in the upper Yukon River basin, Canada, can be derived from brightness 

temperatures (Tb) obtained by the Advanced Microwave Scanning Radiometer for EOS (AMSR- 

E) on NASA’s Aqua satellite. This sensor, with a resolution of 14 x 8 km 2 for the 36.5 GHz 

frequency and two to four observations per day, improves upon the twice-daily coverage and 37 x 

28 km 2 spatial resolution of the Special Sensor Microwave Imager (SSM/I). The onset of melt 

within a snowpack causes an increase in the average daily 36.5 GHz vertically polarized Tb as well 

as a shift to high diurnal amplitude variations (DAV) as the snow melts during the day and 

refreezes at night. The higher temporal and spatial resolution makes AMSR-E more sensitive to 

sub-daily Tb oscillations, resulting in DAV that often show a greater daily range compared to 

SSM/I. Therefore, thresholds of Tb > 246 K and DAV > ±10 K developed for use with SSM/I 

have been adjusted for detecting melt onset with AMSR-E using ground-based surface 

temperature and snowpack wetness relationships. Using newly developed thresholds of Tb > 252 

K and DAV > ±18 K, AMSR-E determined snowmelt onset correlates well with SSM/I 

observations in the small subarctic Wheaton River basin through the 2004 and 2005 winter/spring 

transition. In addition, snowmelt onset derived from AMSR-E data gridded at a higher resolution 

than the SSM/I data indicates that finer-scale differences in elevation and land cover affect 

snowmelt onset and are detectable with the AMSR-E sensor. Based on these observations, the 

enhanced resolution of AMSR-E is more effective than SSM/I at delineating spatial and temporal 

snowmelt dynamics in the heterogeneous terrain of the upper Yukon River basin. 

Keywords: snowmelt, passive microwave, SSM/I, AMSR-E, Yukon River, Wheaton River 

INTRODUCTION 

This Hydrologic and atmospheric processes are closely tied with snow characteristics 

throughout the arctic and subarctic regions of the Northern Hemisphere. The extent of snow 

cover, snow water equivalent (SWE), and the timing of spring snowmelt influence local hydrology 

as well as regional and global water budgets, and the high albedo of snow has an important 

influence on the global energy budget (Vorosmarty et al., 2001). The spring snowmelt is an 

annual event in snow-covered, high-latitude drainage basins that sends significant quantities of 

meltwater downstream during a relatively short period of time. The timing and magnitude of 

these events are variable, but they can create a large impact on the basin hydrology and influence 

geomorphic change within the basin. The use of passive microwave data to examine the 

1 Earth and Environmental Sciences, Lehigh University, Bethlehem, PA 18015, USA 

2 Geosciences/Environmental Studies, Pacific Lutheran University, Tacoma, WA 98447, USA 

3 Water Survey of Canada, Environment Canada, 91782 Alaska Highway, Whitehorse, Yukon 

Territory, Canada Y1A 5B7 

137

characteristics associated with this seasonal snowmelt may provide a better understanding of the 

sensitivity of snowmelt timing to climatic conditions as well as the effects of the snowmelt upon 

basin hydrology. 

The Yukon River extends from the Juneau Icefield, British Columbia, to the Bering Sea, passing 

through the Yukon Territory and Alaska. It receives drainage from a basin more than 850,000 km 2 

in area. The majority of this drainage flows from the Bering Sea northward to provide 8% of the 

total fluvial freshwater input to the Arctic Ocean (Aagaard and Carmack, 1989). The basin 

includes a variety of ecosystems as well as a range of elevations and local relief. This region has a 

heterogeneous terrain marked by boreal forests, mountain ranges, low-lying valleys, and a variety 

of lakes. Much of the upper basin is seasonally snow covered aside from some glaciated regions 

near the headwaters. 

This paper will focus on the timing of the spring snowmelt transition in the Wheaton River subbasin 

of the Yukon River (Fig. 1), from which the findings may then be applied to larger subbasins 

and eventually the upper Yukon River basin as a whole. The Wheaton River basin is a 

small (875 km 2 ) upland basin that has heterogeneous topography and terrain with only a small 

glacial input. 

Two passive microwave sensors, the Special Sensor Microwave Imager (SSM/I) and Advanced 

Microwave Scanning Radiometer for EOS (AMSR-E), provide at least twice daily observations in 

the high latitude regions, and the record of SSM/I observations extends back nearly two decades. 

Due to differences in brightness temperature emitted from dry versus wet snow, these passive 

microwave sensors are effective at differentiating between snow that is melting and snow that is 

not melting, or frozen. In addition, the ability of these sensors to observe snow characteristics 

through clouds and most other atmospheric conditions, as well as during the night, proves useful 

for examining snowmelt in high latitude areas that may have increased cloud cover and darkness 

during the winter and spring months. 

The purpose of this study is to derive the timing of spring snowmelt in the Wheaton River basin 

with recently acquired AMSR-E observations and to test the sensitivity of this sensor to the 

dynamic and varied regional snowpack characteristics. A snowmelt onset algorithm developed by 

Ramage et al. (2006) for use with SSM/I data is modified for use with AMSR-E data to allow for 

a direct comparison of snowmelt onset timing from both of these passive microwave sensors. In 

addition, higher-resolution AMSR-E data and elevation data are used to show improvements of 

this sensor over the SSM/I sensor in detecting snowmelt within heterogeneous terrain. The 

algorithm is fine-tuned using hourly near-surface air temperature data and ground-based snow 

wetness data from within the Wheaton River basin. Due to the greater spatial and temporal 

resolution of the AMSR-E sensor compared to SSM/I, snowmelt dynamics can be more accurately 

described with AMSR-E data in the mixed terrain of the upper Yukon River basin. 

DATA 

Passive microwave satellites provide a means by which to differentiate between snow that is 

melting and snow that is frozen. Both AMSR-E and SSM/I brightness temperature data were used 

for this analysis. In addition, 90 meter digital elevation models of the Yukon Territory were 

utilized (Environment Yukon, 2000), as well as hourly temperature measurements and snow 

wetness measurements from within the Wheaton River basin. 

AMSR-E swath data, in the form of Level-2A Global Swath Spatially-Resampled Brightness 

Temperatures (Tb) V001, are supplied by the National Snow and Ice Data Center (NSIDC) 

through the EOS Data Gateway with a record of observation that extends back to 2002 (Ashcroft 

and Wentz, 2004). The AMSR-E sensor provides two to four observations of the study area per 

138

day and has a mean pixel resolution at 36.5 GHz of 12 x 12 km 2 (actual footprint size is 14 x 8 

km 2 ). Twice-daily SSM/I data extending back to 1987 come from the Defense Meteorological 

Satellite Program (DMSP) SSM/I and have a nominal resolution for the 37 GHz channel of 37 x 

28 km 2 . SSM/I images gridded to the Equal Area Scalable Earth Grid (EASE-Grid) 25 x 25 km 2 

resolution are provided by the NSIDC, and this product separates the ascending and descending 

passes. The AMSR-E data were gridded first at the EASE-Grid 25 x 25 km 2 resolution to compare 

directly with the SSM/I data and establish to what degree they are similar with a minimum of 

complicating factors. The AMSR-E data then were gridded at the finer EASE-Grid 12.5 x 12.5 

km 2 resolution to examine improvements in snowmelt detection of AMSR-E upon the SSM/I 

sensor. 

Passive microwave sensors have been used to examine characteristics of snow such as snow 

extent (e.g. Abdalati and Steffen, 1997; Walker and Goodison, 1993; Wang et al., 2005), snow 

depth (e.g. Josberger and Mognard, 2002; Kelly et al., 2003), and snow water equivalent (e.g. 

Derksen et al., 2005; Foster et al., 2005; Goita et al., 2003). The SSM/I sensor has been used in 

previous studies to establish the timing of the spring melt transition in the upper Yukon River 

basin (Ramage et al., 2006) as well as in the Juneau Icefield (Ramage and Isacks, 2002, 2003). 

Even though the SSM/I sensor provides twice-daily observations and has been shown to correlate 

well with ground-based brightness temperature measurements over fairly homogeneous terrain 

such as the Alaskan North Slope (Kim and England, 2003), the pixel resolution of greater than 25 

x 25 km 2 that results from the passive nature of the sensor is a problematic issue in monitoring 

dynamic changes over heterogeneous terrain. The AMSR-E sensor, recently launched aboard 

NASA’s Aqua satellite in 2002, provides more observations of the study area per day and, with an 

improved pixel resolution over SSM/I, can help to provide a more accurate examination of snow 

characteristics over mixed terrain. 

A significant difference in brightness temperature (Tb) between dry and wet snow occurs at 

frequencies greater than 10 GHz. The Tb of a material is related to its surface temperature (Ts) and 

emissivity (E), 

Tb = ETs. (1) 

A rapid increase in emissivity occurs as a result of a small amount (~ 1-2%) of liquid water within 

the snowpack, causing the Tb to increase for wet snow (Ulaby et al., 1986). The Tb in the 19 and 

37 GHz frequencies associated with the SSM/I sensor are useful in detecting melt on glaciers 

(Ramage and Isacks, 2002, 2003) and on heterogeneous terrain (Ramage et al., 2006) since the Tb 

transition from dry to wet snow occurs as surface temperatures approach 0°C. From the AMSR-E 

sensor, the Tb from the vertically polarized 36.5 GHz frequency (wavelength of 0.82 cm) is 

comparable to the SSM/I sensor for the detection of snowmelt. Here, we focus on comparing and 

contrasting the Tb from the two sensors and comparing the ability to detect the onset of snowmelt. 

During the spring snowmelt, the snowpack experiences cyclical daytime melt and nighttime 

freeze, changes that are manifested as diurnal differences in Tb. As a result of the at least twicedaily 

observations by the sensors, these high diurnal amplitude variations of Tb (referred to as 

DAV) can be detected and are useful in identification of these dynamic transition periods. 

The Wheaton River basin is covered by six EASE-Grid 25 x 25 km 2 pixels (Fig. 1a and Table 

1). Due to the coarse pixel size in relation to the size of the basin, some pixels only cover a small 

percentage of the basin (see Table 1). The Wheaton basin has a range of elevations and land cover 

that include bare mountain slopes, dense boreal forest, and slopes of varying aspect (Brabets et al., 

2000; Ramage et al., 2006). Pixels A02 and A03 cover the upland headwaters of the Wheaton 

River and have high mean elevations. The rest of the basin is mostly covered by pixels B03 and 

B02, representing the middle and lower parts of the basin respectively. Pixels C02 and C03 cover 

a small fraction of the basin but can be used to examine the lowest elevations of the basin. In 

139

elation to the AMSR-E data gridded to the EASE-Grid 12.5 x 12.5 km 2 resolution, the Wheaton 

River basin is covered by parts of 14 pixels, improving the spatial scale at which differences in 

snowmelt characteristics can be investigated (Fig. 1b). Adjacent pixels in this area appear to have 

similar Tb signatures that differ only slightly due to differences in land cover as well as elevation 

and topography. Seasonal variations related to frost, vegetation, and lakes are also factors in this 

complex landscape, but snowmelt has the largest influence on Tb. 

Figure 1. Wheaton River basin EASE-Grid (a) 25 x 25 km 2 pixels and (b) 12.5 x 12.5 km 2 pixels 

superimposed on a 1:250,000 (90 m) digital elevation model (Environment Yukon, 2000). While the actual 

sensor footprints are more elliptical in shape, these square pixels provide a good representation for the land 

area covered. The inset map shows the location of the basin within the Yukon Territory. The basin (black 

outline) is represented by parts of six 25 x 25 km 2 pixels and fourteen 12.5 x 12.5 km 2 pixels. The Wheaton 

River near Carcross stream gauge is located near the river’s mouth at an elevation of 668 m (white dot in 

pixels B02 and B02B). 

140

Table 1. Wheaton River basin EASE-Grid 25 x 25 km 2 pixel characteristics (from Ramage et al., 2006) 

Near-surface air temperatures for the Wheaton River basin were acquired in 2004 and 2005 

using a HOBO automatic temperature logger located within pixel B02 at 668 m, approximately 10 

m from Environment Canada’s “Wheaton River near Carcross” stream gauge (60°08′05″N, 

134°53′45″W). The HOBO recorded hourly air temperature and relative humidity and was 

situated about 1.5 m above the forest floor on the north facing side of a spruce tree (shaded most 

of the time). It was installed under a protective platform in August 2004, and data were 

downloaded periodically by Pat Maltais of the Water Survey of Canada. 

A field campaign was carried out during the spring of 2005 to collect a variety of in situ 

snowpack measurements from snowpits, including snow density and the dielectric properties of 

the snow at various depths, at times that coincided with AMSR-E satellite overpasses. The 

dielectric properties of the snow, and ultimately the snow liquid water content as a percentage of 

the volume, were derived from these measurements to relate to AMSR-E brightness temperature 

data. Other observations of the snowpack in the upper Yukon River basin provided support to 

show that the response in Tb could be attributed to snowmelt instead of liquid precipitation events. 

METHODS 

Ramage et al. (2006) developed a snowmelt onset algorithm for use with SSM/I 37V GHz data 

to detect the timing of snowmelt onset within the Wheaton River basin. This algorithm 

determines the onset of snowmelt based on the day of the year when certain thresholds for both 

brightness temperature (Tb) and diurnal amplitude variation (DAV) are met (Table 2). For the 

SSM/I data, ‘DAV Melt’ occurs when DAV is greater than ±10 K, and ‘Melt Onset’ occurs on the 

day when Tb is greater than 246 K and DAV is greater than ±10 K, indicating the snow is both 

melting periodically and experiencing strong melt-refreeze cycles (Ramage and Isacks, 2002, 

2003). In addition, ‘End DAV’ represents the end of the transition when DAV decreases below 

±10 K, and the duration of the transition period is calculated as the number of days between ‘Melt 

Onset’ and ‘End DAV’. 

Table 2. Snowmelt onset thresholds 

141

A number of assumptions must be made in detecting snowmelt onset using this passive 

microwave data. The heterogeneous and mountainous terrain within the upper Yukon River basin 

most likely creates a complex Tb signal, but based largely on field observations and local weather 

records, it is assumed that the majority of the land is covered with snow during the spring 

transition period. In addition, since the methods focus on snowmelt events during the spring 

transition from March through May, intermittent snowmelt and thaw events during the winter 

months are largely ignored. It is also assumed, based on support from field data, that the events 

identified with the passive microwave data as melting snow are representative of snowmelt and 

not liquid precipitation events. 

AMSR-E swath data were processed using the AMSR-E Swath-to-Grid Toolkit (AS2GT) in 

NSIDC’s Passive Microwave Swath Data Tools (PMSDT). The data were then gridded to the 

EASE-Grid Northern Hemisphere projection with a nominal resolution of 25 x 25 km 2 , the same 

format and pixel resolution as the SSM/I data provided by the NSIDC. The Tb were extracted 

from each channel using Interactive Data Language (IDL) code, creating a chronological time 

series of all Tb observations for 2004 and 2005. 

Since there are two to four AMSR-E observations per day, DAV cannot simply be calculated as 

the difference from one observation to the next, as is done with the twice-daily SSM/I 

observations. Extra observations in a given day are often less than 2.5 hours from the previous 

one and have fairly similar Tb (Fig. 2). These extra values cause the DAV to artificially approach 

0 K as a result of similar Tb values even if the actual daily oscillation is much greater. An 

alternative to the raw data is to average the observations that are less than 2.5 hours apart (Fig. 2). 

This creates a total of two observations for each day and eliminates extraneous DAV values that 

approach 0 K, allowing for more representative DAV values to be calculated and used for 

examining snowmelt characteristics in a similar way to SSM/I DAV. Other techniques used to 

calculate DAV, including using daily minimum and maximum values, did not provide 

representative diurnal variations as well as the above method. The absolute value of DAV is used 

since DAV can be either positive or negative. For the purposes of examining snowmelt, only the 

36.5 GHz vertically polarized observations were used. 

Figure 2. Calculation of DAV from AMSR-E 36.5V GHz Tb. In order to examine daily fluctuations, Tb less 

than 2.5 hours apart (medium gray circles) were averaged. The running differences of these observations 

with the other observations produces a more accurate DAV representation (solid black line) than if all 

observations are used to calculate daily variations (gray line). 

142

The SSM/I snowmelt onset algorithm was modified for use with AMSR-E 36.5 GHz vertically 

polarized Tb data. Initial examination of this data from the AMSR-E sensor for 2005 indicated 

that the observations were more sensitive to daily fluctuations and that the previous Tb snowmelt 

threshold for SSM/I of 246 K was too low. To develop a new threshold for the snowmelt onset 

algorithm, the AMSR-E 36.5V GHz Tb for the six Wheaton River basin pixels were plotted 

against temporally-corresponding near-surface air temperature data from the Wheaton River for 

the period of August 2004 to December 2005 (Fig. 3). When the air temperature is near 0°C, the 

Tb rapidly increases due to the presence of liquid water in the snowpack. The Tb at which this 

transition from frozen to melting snow occurs is considered the threshold. A histogram of annual 

AMSR-E 36.5V GHz Tb data was also produced to identify the new Tb threshold (Fig. 4). A 

typical year of observations in this seasonally-snow covered region has a bimodal distribution of 

Tb related to frozen, dry snow during the cold months and wet snow or no snow during the warmer 

months, and the Tb that separates these two conditions is the Tb threshold (Fig. 4, Table 2). 

Figure 3. Scatterplots of Wheaton River near-surface air temperatures compared to AMSR-E 36.5 GHz 

vertically polarized Tb for the six 25 x 25 km 2 Wheaton River basin pixels for 2004-2005. The transition in 

Tb from frozen (1,2) to melting (3) snow, which appears as an inflection in the slope near a surface 

temperature of 0°C, helps to establish the Tb threshold of snowmelt as 252 K. The two unique distributions 

that exist at temperatures below 0°C in most of the pixels (1,2) may be related to snowpack grain size growth 

or the presence of ice on vegetation. The area of high temperatures (4) represents snow-free surfaces during 

the summer months. 

143

Figure 4. Frequency histogram of all AMSR-E 36.5V GHz Tb for the six Wheaton River basin 25 x 25 km 2 

pixels during the 2004-2005 period. The Tb are grouped in bins of 2 K. The bimodal distribution suggests a 

threshold of 252 K is more suitable for distinguishing between frozen (left distribution) and wet (right 

distribution) snow using AMSR-E in the Wheaton River basin than the SSM/I threshold of 246 K. The 

individual pixel sets, while not as clearly defined, also have minima near the threshold of 252 K. 

Measurements of snowpack properties were obtained in the Wheaton River basin during the 

spring of 2005. More than fifty snowpits were dug in a variety of locations, encompassing a range 

of elevations, terrains, and forest cover in an attempt to represent the heterogeneous nature of the 

region. Characteristics of the snowpack were described and measured, including the general 

stratigraphy as well as grain size, air and snow temperature, and snow structure at various depths. 

A comprehensive analysis of these parameters is beyond the scope of this paper. A snow surface 

dielectric capacitance probe and measurements of snow density were used to derive snow wetness 

as a percentage of the volume at different depths (Denoth, 1989). Surface and near-surface 

wetness values were averaged from multiple measurements at each site, and these wetness values 

were later compared with AMSR-E 36.5V GHz Tb from overpasses obtained less than one hour 

before or after the ground observations (Figs. 5a and 5b, Table 3). In addition to supporting the 

new Tb threshold, these observations helped show that the observed Tb response is linked to 

melting snow and is not the result of liquid precipitation or other events. 

Once the new Tb threshold was determined, the AMSR-E 36.5V GHz data, along with melt 

onset and spring transition data, were compared with the melt onset results of the SSM/I 37V GHz 

data for all six of the Wheaton River basin 25 x 25 km 2 pixels (Figs. 6a and 6b, Table 4). This 

allowed for a new DAV threshold to be created for the AMSR-E data (Table 2), since the daily 

fluctuations appear to have increased with the AMSR-E sensor, and also helped to determine if the 

new thresholds are effective at detecting snowmelt as related to previous snowmelt onset detection 

studies using SSM/I Tb data. 

After determining that similar snow melt onset results could be obtained from AMSR-E and 

SSM/I Tb data gridded at the same 25 x 25 km 2 resolution, the effectiveness of the AMSR-E 

sensor’s improved spatial resolution was investigated. The AMSR-E swath data were gridded to 

the EASE-Grid with a resolution of 12.5 x 12.5 km 2 , which is more representative of the 12 km 

mean spatial resolution of the 36.5V GHz channel (Ashcroft and Wentz, 2004). This gridded 

resolution allowed for the creation of four finer resolution pixels of information for each 25 x 25 

km 2 pixel (Fig. 1b). Elevation data were used to identify the fine-resolution pixels within any 

144

coarse-resolution pixel with the highest and lowest mean elevations (Table 5). The AMSR-E 

36.5V GHz Tb for these pixels, as well as the snowmelt onset timing, were then compared with the 

coarser 25 x 25 km 2 data to examine the effects of the AMSR-E sensor’s increased spatial 

resolution on improving the ability to detect snowmelt in heterogeneous terrain (Figs. 7a and 7b, 

Table 5). 

Table 3. Spring 2005 Wheaton River basin snow wetness measurements and AMSR-E 25 km 2 Tb 

RESULTS 

The two main techniques used to determine a new snowmelt onset Tb threshold for the AMSR-E 

36.5V GHz channel both found the threshold between non-melting and melting snow to be 252 K. 

The first method, in which the Tb for the six Wheaton basin pixels were compared with air 

temperature data for 2004 and 2005 (refer to Fig. 3), illustrates that a Tb threshold of around 252 K 

is an effective boundary between (1,2) non-melting and (3) melting snow near 0°C. In these nonlinear 

plots, it can be noted that two distinct relationships occur on either side of the 0°C mark. 

Below 0°C, the Tb are below 252 K and increase very slightly with an increase in temperature. As 

the temperature approaches and exceeds 0°C, the Tb rapidly increases above 252 K (Fig. 3). This 

rapid change in the slope of Tb when the air temperature is near 0°C at Tb = 252 K is due to the 

high sensitivity of microwaves to liquid water within the melting snowpack. The transition from 

frozen to melting snow is especially clear for pixel B02, which contains the site of the near-surface 

air temperature observations, as the change in slope between these two conditions at 0°C occurs at 

the Tb threshold of 252 K (Fig. 3). 

It is also worth noting that two distinct distributions of Tb occur when compared to air 

temperatures below 0°C (Fig. 3). The upper Tb response (Fig. 3, #1) is from periods in November 

from both 2004 and 2005, while the lower response (Fig. 3, #2) is from periods in December of 

the same years. The cause of the upper response may be related to ice accumulation on trees 

during the earlier winter months, a condition that was observed in the 2006 winter season. During 

the winter, especially during unusually warm years, open lakes that have not yet frozen can create 

145

an icy fog that creeps up valleys and allows for ice to accumulate on vegetation. These distinct 

distributions may also be an effect of grain size growth during the early winter months, 

influencing the scattering properties of the snowpack. At the 36.5 GHz frequency, emissivity 

decreases proportional to an increase in grain size (Mätzler, 1994), and so as the grains within the 

snowpack grow early in the winter, their Tb decreases from the upper distribution to the lower 

distribution (Fig. 3). As these events both occur within the same time frame, further investigation 

is required to identify the source of these differing relationships. 

The second technique, examining a histogram of AMSR-E 36.5V GHz Tb distribution from the 

Wheaton basin for all 2004 and 2005 data, also produced a Tb threshold of 252 K (Fig. 4). The 

frequency histogram, with a bin size of 2 K, displays a bimodal distribution from which it is 

interpreted that brightness temperatures less than 252 K relate to frozen, non-melting snow and 

brightness temperatures greater than 252 K relate to wet, melting snow as well as no snow during 

the summer months. The sets of individual pixels, representing the upper (pixels A02 and A03), 

middle (pixels B02 and B03), and lower (pixels C02 and C03) portions of the Wheaton River 

basin, do not have as clear a bimodal distribution but do have comparable minimum values around 

252 K (Fig. 4). This, along with the plots of Tb against air temperature, suggests that the Tb 

threshold is effective in differentiating between dry and wet snow at the drainage basin and subdrainage 

basin scales in this heterogeneous terrain. 

The DAV calculation, which involves taking the difference between night and day observations, 

shows the daily contrast in brightness temperatures at this time of year. The timing of the AMSR- 

E overpasses for this region are around 03:30 and 13:30 local time (PST), which are likely near 

the times of minimum and maximum daily melt, and so the difference in the Tb from these times 

provides a good indicator of melting and refreezing snow. The DAV threshold increased due to 

the apparent increased sensitivity of the AMSR-E sensor to daily Tb fluctuations (Fig. 6a). A new 

DAV threshold of ±18 K, which was determined by comparing the AMSR-E DAV with SSM/I 

DAV for the Wheaton River pixels in 2005 (Fig. 6b), represents times when strong melt-refreeze 

cycles are occurring during the spring snowmelt transition. 

The change in Tb and DAV thresholds from SSM/I to AMSR-E (Table 2) is most likely the 

result of differences in the times of data acquisition for the two sensors. Daily AMSR-E 

overpasses for the area occur around 03:30 and 13:30 PST, whereas SSM/I overpasses occur 

around 08:30 and 18:30 PST. The AMSR-E sensor makes observations during the early morning, 

when temperatures and corresponding Tb would be near their daily minimum, and again in the 

early afternoon, when much of the mountainous terrain would be near the daily maximum air and 

brightness temperatures. This creates larger DAV in relation to the DAV from SSM/I, since the 

timing of SSM/I overpasses is further from these daily minimum and maximum, and therefore 

requires a larger DAV threshold. In addition, the DAV threshold must increase so as to not 

indicate that melting is occurring when large DAV are observed during the summer, when much 

of the land is snow-free (Fig. 6b). 

Field-observed snow liquid-water content measurements from the Wheaton River basin during 

the spring of 2005 support the Tb threshold of 252 K and also show that the assumed response in 

Tb to melting snow can be directly attributed to snowmelt instead of other effects, such as liquid 

precipitation events. Surface (0 cm) and near surface (0.5-2 cm) wetness measurements relate 

very well to AMSR-E 36.5V GHz Tb observations obtained at coincident times (Figs. 5a and 5b, 

Table 3). At wetness values below 2%, the Tb are clustered between 230 K and 245 K, but as the 

liquid-water content of the snowpack increases above 2% by volume, the Tb appear to increase 

near and above the threshold of 252 K (Fig. 5a). For the period prior to and including the 

beginning of the spring transition, the changes in the snowpack wetness through time closely 

follow that of the AMSR-E Tb, and at wetness values above 2%, the Tb are at or above the 252 K 

threshold (Fig. 5b). This relationship verifies that the Tb response is due to the presence of liquid 

water within melting snow. 

146

Figure 5. (a) Scatterplot of snowpack wetness compared to AMSR-E 36.5V GHz Tb within the Wheaton 

River basin. Below 2% wetness, Tb are clustered between 230 K and 245 K, but an increase in liquid-water 

content beyond 2% creates an abrupt rise in Tb around the threshold of 252 K. (b) The snowpack wetness 

(gray) relates well to the AMSR-E Tb (black) during the period prior to the spring transition. 

With these new AMSR-E snowmelt onset thresholds (Tb > 252 K and DAV > ±18 K), the 

AMSR-E spring transition in 2004 and 2005 was characterized for the six Wheaton River 25 x 25 

km 2 pixels and compared with SSM/I data for the same pixels and years (Figs. 6a and 6b, Table 

4). The snowmelt onset algorithm developed by Ramage et al. (2006) was used for the SSM/I 

data, using thresholds of Tb > 246 K and DAV > ±10 K (Table 2). For pixel B02, the onset of 

snowmelt compares very well between AMSR-E and SSM/I, as does the end of the spring 

transition and the total duration (Figs. 6a and 6b). It should be noted that although the AMSR-E 

thresholds were met for this pixel around day 95, they were not met for a consistent length of time, 

and so the spring transition is interpreted as beginning about 10 days later (Figs. 6a and 6b). The 

DAV of the AMSR-E observations are higher than the corresponding SSM/I observations 

throughout the spring transition as well as for much of the summer. Preliminary investigations of 

AMSR-E derived snowmelt in the larger Pelly River basin, approximately 250 km northwest of 

the Wheaton River basin, indicate the AMSR-E Tb and DAV thresholds are effective for detecting 

snowmelt on a regional scale. 

147

Figure 6. (a) Comparison of pixel B02 AMSR-E and SSM/I Tb for 2005. The AMSR-E Tb (black) match 

closely with the SSM/I Tb (gray), although they appear to show greater variability during the spring transition 

and summer months. The melt onset date and duration for the two records are also similar. (b) Comparison 

of pixel B02 AMSR-E and SSM/I DAV for 2005. The AMSR-E DAV (black) is consistently higher than the 

SSM/I DAV (gray) throughout the spring and summer. The AMSR-E DAV threshold of ±18 K appears to 

match with the SSM/I DAV threshold of ±10 K and the timing of SSM/I melt onset. 

Table 4. AMSR-E and SSM/I melt dates (day of year) for the Wheaton River Basin 

148

For both 2004 and 2005, the AMSR-E and SSM/I melt onset dates for the six Wheaton River 25 

x 25 km 2 pixels match up very well, with the dates differing by less than two days (Table 4). Pixel 

B03 in 2004 was found to have the same melt onset date for both sensors, as was pixel C02 in 

2005. These well correlated melt onset dates indicate that the new AMSR-E data and thresholds 

are effective in detecting snowmelt in the Wheaton River basin at the same resolution as SSM/I. 

In addition, this relationship may eventually allow for the observations from both sensors to be 

combined to create an even more robust satellite record of observation. 

Since the AMSR-E sensor provides enhanced resolutions over the SSM/I sensor, especially in 

the 36.5 GHz channel, the timing of snowmelt within the Wheaton River basin was also 

determined using Tb data gridded to the EASE-Grid 12.5 x 12.5 km 2 resolution (Fig. 1b). Pixel 

B02 (25 x 25 km 2 resolution) and the four finer-resolution pixels within it were examined in an 

attempt to tease out the effects of the heterogeneous terrain on snowmelt onset (Table 5). 

Differences between the plots of AMSR-E Tb in the first half of 2005 for pixels B02C and B02B, 

which represent the highest and lowest mean elevations within pixel B02 respectively, indicate 

that the spatial resolution of the AMSR-E sensor improves upon the SSM/I sensor’s ability to 

detect snowmelt in mixed terrain, especially in areas with a range of elevations (Figs. 7a and 7b, 

Table 5). 

Table 5. EASE-Grid 12.5 x 12.5 km 2 pixel characteristics and melt onset dates for 2005 

In relation to the 2005 melt onset timing and spring transition duration for pixel B02, the timing 

of these events for the finer-resolution pixels is similar (Table 5). Two notable exceptions are 

pixel B02B, the lowest in mean elevation, and pixel B02C, the highest in mean elevation. Melt 

onset occurs much earlier in B02B than in the other pixels, most likely as a result of warmer 

temperatures within its lower elevations. For B02C, the spring transition period lasts more than 20 

days longer than the other pixels, and this is due to colder temperatures in the higher elevations 

keeping the nighttime snowpack frozen and DAV high well into the spring. Examining the time 

during which snowmelt onset has begun for B02B (Fig. 7b), it can be noted that the Tb for B02B 

are consistently above the threshold from day 94 to 100, whereas the Tb for B02C never reach the 

threshold during this period. Furthermore, Tb for pixel B02 indicate melting during the day from 

days 96 to 100, and while this concurs with B02B, it disagrees with B02C, and so the higher 

resolution AMSR-E data is more effective at differentiating melting and non-melting areas in this 

heterogeneous terrain than lower resolution AMSR-E and SSM/I data. 

149

Figure 7. (a) Comparison of higher resolution AMSR-E Tb from 2005 for pixels B02B and B02C. The lower 

elevation B02B (black) has higher overall Tb than B02C (gray) that relate to warmer temperatures. This 

results in an earlier melt onset for B02B in relation to B02C and the other pixels (Table 5). (b) A detailed 

look at days 90 to 100 reveals that the Tb threshold is met consistently in the lower elevations of B02B 

(black) but not in the higher elevations of B02C (thick gray). Pixel B02 (thin gray) is more closely related to 

B02B, but the Tb threshold is not met as early as in B02B. 

CONCLUSIONS 

AMSR-E passive microwave data can be used to investigate snowmelt at the basin and regional 

scales. In addition, the finer resolution of AMSR-E over that of SSM/I allows for the snowpack 

characteristics of areas with heterogeneous terrain to be more accurately examined. For the 

Wheaton River dataset, AMSR-E is more sensitive to daily brightness temperature fluctuations 

than SSM/I at the same spatial scale. Relating the two datasets at the same scale provides the 

foundation for a longer-term study of snowmelt dynamics combining the historical record of both 

sensors. Plots of AMSR-E Tb and near-surface air temperatures, as well as a bimodal distribution 

of Tb and comparisons with snow liquid-water content, suggest a Tb threshold of 252 K can be 

used to make the distinction between melting and non-melting snow in combination with a DAV 

150

threshold of ±18 K that reflects the spring transition when large melt-refreeze cycles are occurring. 

The larger oscillations in daily Tb are most likely due to the timing of data acquisition being closer 

to the actual maximum and minimum daily Tb. The snowmelt onset algorithm will also be 

applicable in most arctic and subarctic landscapes. 

Snowmelt onset dates from AMSR-E, as determined using the new Tb and DAV thresholds, 

compare very well with SSM/I for the Wheaton River basin, differing by a maximum of two days. 

The enhanced spatial resolution of the AMSR-E data provides an improvement over SSM/I in 

detecting snowmelt onset in areas of mixed terrain. Using this AMSR-E melt onset algorithm, the 

snowpack dynamics of the Wheaton River basin, as well as other areas of the upper Yukon River 

basin, can be further investigated spatially and temporally for the record of AMSR-E observations. 

The improvements by the AMSR-E sensor upon SSM/I allow for a more effective spatial and 

temporal examination of snowmelt dynamics in the heterogeneous terrain of the upper Yukon 

River basin. It will be a significant improvement to our ability to understand the melt processes 

and timing in an important, remote, high latitude drainage basin. 


We would like to extend our thanks to Mary Jo Brodzik, Amanda Leon, and the National Snow 

and Ice Data Center for EASE-grid satellite data and continued assistance. Thanks to 

Environment Canada for ongoing support with field logistics and additional data. We thank Sarah 

Kopczynski and Shannon Haight for their assistance collecting field data. We are also thankful 

for helpful comments provided by Andrew Klein and two anonymous reviewers. Additional 

thanks to Michael Chupa. Funding was provided by Lehigh University and the National 

Aeronautics and Space Administration’s Terrestrial Hydrology Program (grant # NNG04GR31G). 

REFERENCES 

Aagaard K, Carmack EC. 1989. The role of sea ice and other freshwater in the Arctic circulation. 

Journal of Geophysical Research 94: 14485-14498. 

Abdalati W, Steffen K 1997. Snowmelt on the Greenland Ice Sheet as derived from passive 

microwave satellite data. Journal of Climate 10: 165-175. 

Ashcroft P, Wentz F. 2004 updated daily. AMSR-E/Aqua L2A Global swath spatially-resampled 

brightness temperatures (Tb) V001. January 2004 to December 2005. Boulder, CO, USA: 

National Snow and Ice Data Center. Digital media. 

Brabets TP, Wang B, Meade RH. 2000. Environmental and hydrologic overview of the Yukon 

River basin, Alaska and Canada. USGS Water Resources Investigations Report 99-4204, 

Anchorage. 

Denoth A. 1989. Snow dielectric measurements. Advances in Space Research 9: 233-243. 

Derksen C, Walker AE, Goodison BE, Strapp JW. 2005. Integrating In Situ and multiscale passive 

microwave data for estimation of subgrid scale snow water equivalent distribution and 

variability. IEEE Transactions on Geoscience and Remote Sensing 43: 960-972. 

Environment Yukon. 2000. 90 Meter Yukon Digital Elevation Model. 

http://www.environmentyukon.gov.yk.ca/geomatics/data/90m_dem.html [13 March 2005]. 

Foster JL, Chaojiao S, Walker JP, Kelly RE, Chang A, Dong J, Powell H. 2005. Quantifying the 

uncertainty in passive microwave snow water equivalent observations. Remote Sensing of 

Environment 94: 187-203. 

Goita K, Walker AE, Goodison BE. 2003. Algorithm development for the estimation of snow 

water equivalent in the boreal forest using passive microwave data. International Journal of 

Remote Sensing 24: 1097-1102. 

Josberger EG, Mognard NM. 2002. A passive microwave snow depth algorithm with a proxy for 

snow metamorphism. Hydrological Processes 16: 1557-1568. 

151

Kelly RE, Chang AT, Tsang L, Foster JL. 2003. A prototype AMSR-E global snow area and snow 

depth algorithm. IEEE Transactions on Geoscience and Remote Sensing 41: 230-242. 

Kim EJ, England AW. 2003. A yearlong comparison of plot-scale and satellite footprint-scale 19 

and 37 GHz brightness of the Alaskan North Slope. Journal of Geophysical Research 108: 

ACL 7. 

Mätzler C. 1994. Passive microwave signatures of landscapes in winter. Meteorology and 

Atmospheric Physics 54: 241-260. 

Ramage JM, Isacks BL. 2002. Determination of melt-onset and refreeze timing on southeast 

Alaskan icefields using SSM/I diurnal amplitude variations. Annals of Glaciology 34: 391-398. 

Ramage JM, Isacks BL. 2003. Interannual variations of snowmelt and refreeze timing on 

southeast-Alaskan icefields, U.S.A. Journal of Glaciology 49: 102-116. 

Ramage JM, McKenney RA, Thorson B, Maltais P, Kopczynski SE. 2006. Relationship between 

passive microwave-derived snowmelt and surface-measured discharge, Wheaton River, Yukon 

Territory, Canada. Hydrological Processes 20: 689-704. 

Ulaby FT, Moore RK, Fung AK. 1986. Microwave Remote Sensing: Active and Passive, Vol. III: 

From Theory to Applications. Artech House: Dedham. 

Vorosmarty CJ, Hinzman LD, Peterson BJ, Bromwich DH, Hamilton LC, Morison J, 

Romanovsky VE, Sturm M, Webb RS. 2001. The Hydrologic Cycle and its Role in Arctic and 

Global Environmental Change: A Rationale and Strategy for Synthesis Study. Arctic Research 

Consortium of the U.S.: Fairbanks, Alaska; 84. 

Walker AE, Goodison BE. 1993. Discrimination of a wet snowcover using passive microwave 

satellite data. Annals of Glaciology 17: 307-311. 

Wang L, Sharp M, Brown R, Derksen C, Rivard B. 2005. Evaluation of spring snow covered area 

depletion in the Canadian Arctic from NOAA snow charts. Remote Sensing of Environment 95: 

453-463. 

152

ABSTRACT 

153 



Combination of Active and Passive Microwave 

to Estimate Snowpack Properties in Great Lakes Area 

AMIR E. AZAR 1 , TARENDRA LAKHANKAR 1 , 

NARGES SHAHROUDI 1 , AND REZA KHANBILVARDI 1 

In this research we examine active and passive microwave to snow water equivalent (SWE) and 

to investigate the potential of combining active and passive microwaves to improve the estimation 

of SWE. The study area is located in Great Lakes area between the latitudes of 41N–49N and the 

longitudes of 87W–98W. Passive microwave are obtained from DMSP SSM/I sensors provided by 

NSIDC. Active microwave were obtained from different sensors: 1) RADARSAT C-Band SAR. 

2) QuikSCAT Ku-band (13.4GHz) for both vertical and horizontal polarizations. The ground truth 

data was obtained from SNODAS data set produced by NOHRSC. An Artificial Neural Network 

model was defined to model various combinations of inputs to SWE. The results indicate that 

none of the active microwave channels produce satisfactory results. However, when combined 

with passive microwave, they improve the estimated SWE. 

INTRODUCTION 

Microwave remote sensing techniques have been effective for monitoring snowpack parameters 

(snow extend, depth, water equivalent, wet/dry state). Snow parameters are extremely important 

for input to hydrological models for understanding changes in climate due to global warming. 

Snow parameters been investigated by numerous researchers using many sensors such as SMMR 

and SSMI for passive microwave and SAR and QSCAT for active microwave. Space-borne 

microwave sensors can monitor characteristics of seasonal snow cover at high latitudes regardless 

of lighting conditions, time of the day, and vegetation. 

In passive microwave radiometer, microwave energy emitted from the ground surface is 

transmitted through the snow layer into the atmosphere and recorded by the sensor. Snow 

parameters can be extracted from remote sensing data by empirical algorithms. Hallikainen (1984) 

introduced his algorithm for estimating SWE using passive microwave SMMR data. The process 

involved the subtraction vertical polarizations of 18 and 37 GHz frequencies. The subtracted 

value, dT, was used to define linear relationships between dT and SWE. Chang et al. (1987) 

proposed using the difference between the horizontally polarized channels SMMR 37 GHz and 18 

GHz to derive snow depth – brightness temperature relationship for a uniform snow field (Chang 

et al 1987). Goodison and Walker (1995) introduced the most widely used algorithm for North 

America. The algorithm was originally for Canadian prairies. It defines a linear relationship 

between GTV ([37V–19V]/18) and SWE. They also suggested using 37H and 37H polarization 

differences for identifying wet snow. Derksen et al. (2004) developed a new algorithm which 

derives SWE for open environments, deciduous, coniferous, and spars forest cover [SWE = 

FDSWED + FC SWEC + FS SWES + FOSWEO]. The algorithm represents an improvement, however 

1 NOAA-CREST, City University of NY, 137thst & Convent Ave. New York, NY.

154 



still underestimates SWE in densely forested areas. Tedesco et al. (2004) developed and tested an 

inversion technique for retrieval of SWE and dry snow depths based on artificial neural networks 

(ANN) by using 19- and 37-GHz SSM/I measured brightness temperatures. 

Hallikainen et al 2003 combined active (QuikSCAT/SeaWinds) and passive (SSMI/DMSP) data 

for monitoring key snow parameters in Finland. The results show that combined active and 

passive microwave sensors provide useful diurnal and seasonal information. These results are 

more accurate than those obtained by only passive microwave. In another research Hallikainen 

showed that using space borne scatterometer (QuikSCAT onboard SeaWinds) for dry snow 

conditions, the backscattering coefficient increases with increasing SWE. For wet snow condition 

backscattering coefficient decreases with increasing SWE. Ku-band scatterometer were used 

successfully to determine the onset and the end of snow melt, and to derive time series for the 

fraction of snow-free ground during the seasonal snow melt period (Hallikainen et al 2004). 

Synthetic Aperture Radar (SAR) particularly C-band SAR has shown the potential for 

monitoring snow and ice for more than two decades. The high spatial resolution and the 

independence of the sensors from sun illumination and cloud cover make SAR an ideal tool for 

snow studies. Launched in 1995, Radarsat-1 offers spatial resolutions between 10m to 100m and a 

swath up to 500km. To estimate SWE using C-band SAR, Bernier et al. (1998) introduced an 

approach based on the fact that snow cover characteristics influence the underlying soil. The snow 

influence on soil temperature affects the dielectric properties of the soil which has a major role on 

the backscattered signal. To recover the SWE from SAR data an algorithm made of two equations 

was used. The first equation defines a linear relationship between the snow thermal resistance and 

the backscattering ratio between a winter image and a reference (snow-free) image in DB. The 

snow-free image helps to eliminate the radiometric distortion due to topography as well as to 

minimize the effect of soil roughness on the signal. The second equation is a linear relationship 

between thermal resistance and the SWE. To estimate SWE from thermal resistance the mean 

density of the snowpack has to be derived. This approach has been applied for cold winter 

conditions and dry snow (Bernier et al. 1999). The critical variables influencing the algorithm are 

variety of land cover, specifically forest density, Snowpack properties (depth>2m), and severe 

topography. In a research on passive and active airborne microwave remote sensing of snow cover 

Sokol et al. (2003) showed that SAR sensors are highly sensitive to changes in the dielectric 

constant and have better spatial resolution than their passive counterparts. They concluded that 

passive techniques estimate SWE most accurately under dry snow conditions with minimal 

stratified snow structures (Sokol et al. 2003). 

The focus of this research is estimating Snow Water Equivalent (SWE) in Great Lakes area by 

using active and passive microwaves. Different approaches were examined for SWE estimations 

by RADARSAT SAR and also QuikSCAT-Ku along and passive SSM/I. 

STUDY AREA 

It is also located on the transitional zone for snow meaning that the northern part of the study 

area is covered by snow for the whole winter season however for the southern part there is a 

pattern of snow-fall and snow melt within the season. In addition to snow pattern, the land cover 

type varies a wide range including, Evergreen Needle leaf forest, Deciduous Broadleaf forest, 

cropland, woodland and dry land (Figure 1).

DATA USED 

SSM/I & 

QuikSCAT 

Coverage 

LANDCOVER 

Figure 1. Study area and SSM/I and RADARSAT coverage 

155 



RADARSAT 

Images 

SSM/I 

SSM/I passive microwave radiometer with seven channels is operating at five frequencies (19, 

35, 22, 37, and 85.5 GHz) and dual- polarization (except at 22 GHz which is V-polarization 

only). The sensor’s spatial resolution varies for different channel frequencies. In this study we 

used Scalable Equal Area Earth Grid EASE-Grid SSM/I products distributed by National Snow 

and Ice Data Center (NSIDC). EASE-Grid spatial resolution is slightly more than 25km (25.06) 

for all the channels (NSIDC) although the recorded resolution of the sensor for longer wavelengths 

is more than 50km. The three EASE-Grid projections comprise two azimuthal equal-area 

projections for the Northern or Southern hemispheres, respectively and a global cylindrical equal 

area projection. In our study we used a Northern hemisphere azimuthal equal-area. The study area 

is covered by 980 (28 by 35) SSM/I EASE-Grid pixels. 

Ku-Band 

The QuikSCAT/SeaWinds scatterometer provides normalized radar cross section measurements 

of the Earth’s surface at unprecedented coverage and resolution. The QuikSCAT sensor on the 

SeaWinds satellite operates at 13.4 GHz vertical and horizontal channels. The sigma (0) browse 

product of QuikSCAT has the grid size of 5 pixels per degree or about 22.5km at the equator. In 

order to match SSM/I grid size the QuikSCAT images were averaged to 25km resolution for the 

study area. 

Normalized Difference Vegetation Index (NDVI) 

NDVI is used to represent the variety of land cover in the study area. The NDVI data obtained 

from the NOAA/NASA Pathfinder AVHRR is distributed at Goddard Space Flight Center 

(GSFC). The spatial resolution is 8km by 8km obtained within a 10 day period that has the fewest 

cloud. To match the with RADARSAT images, NDVI image was resampled and projected to 

UTM.

156 



RADARSAT Images 

RADARSAT ScanSAR images were obtained for February 06, 2003 (winter image), February 

02, 2006, and May 01, 2002 (snow-free image) as show in Table 1. ScanSAR images (500km by 

500km) have the nominal spatial resolution of 100m. The ScanSAR products currently offered by 

us do not come in a map-projected format. However, images have the geo-referencing information 

contained in the CEOS format. This information is derived from the satellite orbit (ephemeris) and 

is typically accurate to 100–200 meters, depending on beam mode and the topography. To reduce 

geometric distortions caused by radar sensor viewing geometry satellite movement, earth 

curvature and rotation, both RADARSAT images were registered to a Landsat image of the study 

area. More than 15 Ground Control Points and a second order model and nearest neighbor 

resampling mode were used to register the RADARSAT images. The images were projected to a 

UTM projection and subseted for the in-common area of coverage. 

Table 1. RADARSAT images for Great Lake area 

Image Date Time(UTC) Inc Angle NW NE SW SE Center 

Winter 2/2/2006 23:47 34.26 

48 51 N 

94 13 W 

49 46 N 

86 37 W 

43 43 N 

92 27 W 

44 37 N 

85 33 W 

46 44 N 

89 42 W 

Winter 2/6/2004 23:53 34.26 

48 26 N 

94 47 W 

49 17 N 

87 46 W 

43 50 N 

93 17 W 

44 40 N 

86 50 W 

46 53 N 

90 44 W 

Snow-free 5/1/2003 23:49 34.26 

48 37N 

93 47 W 

49 27 N 

86 45 W 

44 00 N 

92 17 W 

44 51 N 

85 49 W 

46 47 N 

89 40 W 

Ground Truth Data 

NOAA National Weather Service's National Operational Hydrologic Remote Sensing Center 

(NOHRSC) started producing SNOw Data Assimilation System (SNODAS), beginning 1 October 

2003. SNODAS includes and procedures to assimilate airborne gamma radiation and groundbased 

observations of snow covered area and snow water equivalent, downscaled output from 

Numerical Weather Prediction (NWP) models combined in a physically based, spatially 

distributed energy- and mass-balance model. The output product has 1km spatial and hourly 

temporal resolution. 

METHODOLOGY AND MODEL 

The objective of this study is to estimate snow depth and SWE using active and passive 

microwave images. However, active microwave RADARSAT SAR and passive microwave SSM/I 

are totally different in their nature and applications. RADARSAT images have high spatial 

resolution and are suitable for regional studies. On the other hand, low spatial resolution and high 

temporal resolution of SSM/I make it suitable for studies on a global scale. In order to consider 

above differences, each of the active and passive data were analyzed separately to investigate their 

potential to estimate snow parameters. This section focuses on: 1) Using high resolution active 

microwave RADARSAT SAR to estimate SWE. This section focuses on how to process 

RADARSAT images and corresponding ground truth data along with suggesting modeling 

approaches to improve the estimations. 2) Using low resolution passive SSM/I and active 

QuikSCAT with statistical based models. Also, the improvement by combining various data types 

was quantified. 

An adaptive network is a network structure that consists of a number of nodes (neurons) 

connected through directional links. Each node represents a process unit, and the links specify 

casual relationship between the connected nodes. Nodes are adaptive meaning that the outputs of 

these nodes depend on modifiable parameters pertaining to these nodes. The learning rule specifies 

how these parameters should be updated to minimize error which is discrepancy between the 

networks actual output and desired one. In our study we used a feed forward backpropagation

157 



model. The network has two hidden layers with ten nodes at each layer. To train the network data 

were divided into three sets (training, validation, and test). The model testing is the process by 

which the input vectors from input/output data sets on which the network is not trained, are 

presented to the trained model, to see how well the ANN model predicts the corresponding data 

set output values. The other type of validation which is also referred as checking data set is used to 

control the potential for the model over fitting the data. In principle, the model error for the 

checking data set tends to decrease as the training takes place up to the point that overfitting 

begins, and then the model error for checking data suddenly increases. 

High Resolution Active Microwave RADARSAT SAR 

Four different approaches for ANN input data were considered: A. Input consists of the only 

backscattering ration at 25km resolution. B. Input includes NDVI data in addition to 

backscattering both in 25km resolution. C. Input includes NDVI data in addition to backscattering 

with modified ANN training. D. Input includes NDVI and Backscattering in 5km averaged 

resolution. 

A. Input consists of the only backscattering ration at 1km resolution 

In the first approach, the backscattering ratio was used as the input for the ANN model. 

Figure 5.8 shows the spatial variation of the backscattering ratio while the water bodies are filtered 

out of the image. The higher backscattering ratio is detected in high latitudes and around the lake. 

The scatter plot of backscattering versus the SWE indicates low correlation. 

1.5 

0 

-2 

Figure 2. Spatial variation of backscattering ratio and SWE in 1km resolution, February 06, 04 

R=0.22 

Figure 3. Backscattering vs. SWE, February 06, 04

158 



The ANN model was trained using backscattering ratio as input and SWE the output of the 

model. The simulated SWE based on the backscattering ratio for February 06, 2004 is shown in 

Figure 4. Both the produced image and the corresponding scatter plot show unsatisfactory results. 

The model output is highly underestimated and show very low correlation coefficient. Introducing 

land cover characteristics can be helpful to increase the accuracy of the model. 

160 

140 

120 

100 

80 

60 

40 

20 

0 

SWE(mm) 

200 

180 

160 

140 

120 

100 

80 

60 

40 

20 

Scatter Plot of ANN Output and SNODAS 

R=0.34695 

RMSE=40mm 

0 

0 20 40 60 80 100 120 140 160 180 200 

SWE Estimated by ANN (mm) 

Figure 4. Spatial variation of SWE for February 06, 04 and the scatter plot of model output vs. ground truth 

SWE 

B. Input includes NDVI data in addition to backscattering both in 1km resolution 

In order to introduce the land cover characteristics in a quantitative way the NDVI image of 

the study area was added as an input to the ANN model. The model was trained and validated 

using the data from two winter days of February 06, 2004 and February 02, 2006. The simulated 

SWE for February 06, 2004 is illustrated (Fig 5). 

160 

140 

120 

100 

80 

60 

40 

20 

0 

SWE(mm) 

200 

180 

160 

140 

120 

100 

80 

60 

40 

20 

Scatter Plot of ANN Output and SNODAS 

R=0.471 

RMSE=38mm 

0 

0 20 40 60 80 100 120 140 160 180 200 

SWE Estimated by ANN (mm) 

Figure 5. Simulated SWE based on NDVI and Backscattering for February 06, 04 and the scatter plot of 

model output vs. ground truth SWE 

The above results show the increase of correlation by adding the NDVI to the ANN model. On 

the other hand, the problem with underestimation of SWE still exists. As shown in the scatter plot, 

for SWE varying between 0 to 250mm the model output mostly varies between 50 to 150mm. This 

problem might originate from the training process considering the fact that there are more pixels 

with low values of snow than high values. This forces the training towards the low values in order 

to minimize the RMSE of the total estimation. To minimize the influence of number of the pixels 

with different values for SWE, the network was trained with same number of pixels from each 

class as the third approach.

159 



C. Input includes NDVI data in addition to backscattering with modified ANN training 

In the third approach the pixels were divided to four classes based on the SWE values 

(SWE

160 



image. On the other hand, the backscattering ratio trend is almost constant. Also, the range for the 

ratio is highly different for years 2004 and 2006. This limits any kind of modeling for estimating 

SWE from backscattering. The comparison of the scatter plots of SWE versus backscattering is 

shown in Figure 9. 

Backscattering 

SWE (mm) 

2004 2006 

Backscattering 

2004 2006 

SWE (mm) 

Figure 8. Variation of backscattering ratio and SWE for different parts of the study area 

SWE (mm) SWE (mm) 

2004 

Backscattering Ratio (DB) Backscattering Ratio (DB) 

Backscattering Ratio (DB) Backscattering Ratio (DB) 

SWE (mm) SWE (mm) 

2004 

2006 2006 

Figure 9. Scatter plots of Backscattering vs. SWE at 5km resolution for two images (Feb 06, 04 & Feb 02, 

06)

161 



The left scatter plots representing the complete dataset and the right ones are modified by 

eliminating the backscattering ratio higher than zero since only few percentage points are points 

have backscattering values higher than zero. This could be the effect urban areas that have high 

backscattering. The new scatter plots are shown in Figure 9. These scatter plots clearly show 

where SWE is in the range of 10mm to 60mm the backscattering range is very large. Also, the 

backscattering range between two years (2004 and 2006) is very different. This indicates that 

developing a model to estimate SWE using backscattering ratio is very difficult. A solution for 

this problem is modifying the model for each year. 

Low Resolution Active QuikSCAT and Passive SSM/I 

A. Evaluation of using NDVI and QuikSCAT in SWE estimation 

A feed forward backpropagation neural network model with 2 hidden layers, 20 neurons each 

layer, was developed. The output of the model consists of ground truth SWE data from NOHRSC. 

In order to evaluate the effect SWE on various microwave channels different combination of 

inputs were used. First, the input consisted four of SSM/I channels (19V, 19H, 37V, and 37H). Then, 

NDVI and SSM/I were introduced as the input to the model. Finally, QuikSCAT-ku along with 

SSM/I and NDVI were used as the input. For all above approaches the model was validated by a 

dependent data. In other words, the training and validation data were the same. Figure 10 shows 

the results for various approaches. SSM/I brightness temperatures have shown correlations with 

SWE to some extent. Adding NDVI to the input brings information about the land cover type for 

the ANN model and increases the accuracy of the estimate. By adding active QuikSCAT to the 

input we have an input of three independent data sets. The increase of correlation coefficients 

indicates that combining active and passive using a neural network can improve the SWE 

estimation. Figure 10 also indicates that combining SSM/I with QuikSCAT and NDVI produces 

the best results. The sudden decrease of correlation coefficients for days in February should be due 

to the existence of wet snow in parts of the study area which was already explained in chapter for 

snow cover estimation. 

1.2 

1 

0.8 

0.6 

0.4 

0.2 

0 

Dec 

6, 03 

SSM/I 

QSCAT 

SSM/I &NDVI 

SSM/I &QSCAT 

QSCAT & NDVI 

SSM/I & QSCAT & NDVI 

Dec 

13, 03 

Dec 

20, 03 

Jan 

4, 04 


11, 04 


18, 04 


25, 4 

Feb 

1, 04 

Feb 

8, 04 

Feb 

23, 04 

Figure 10. Correlation coefficients (R 2 ) between estimated SWE and the corresponding ground truth data 

(winter 2003–2004, model validated by dependent data)

162 



B. Using Active and Passive to Estimate SWE 

The results above showed that adding NDVI and QuikSCAT-ku increases the correlation 

between input of microwave channels and SWE. To investigate the capability of the model to 

estimate SWE, it was examined to independent data. The approach consisted of training the model 

with the data from the days before the selected day for estimation. Table 2 describes the approach. 

The results for correlation coefficients and RMSE in the table show an increasing trend for RMSE. 

This increase in the error originates from the increase in the average depth of snow during the 

winter season. For correlation coefficients there is an improvement in the beginning but it 

decreases after January 25, 2004. For the selected days in February especially Feb08, Feb16, and 

Feb23, The error increases dramatically. This is due to the wet snow conditions for those days. 

Wet snow can not be detected by passive SSM/I scattering channels. This section is already 

discussed in snow cover section. Figures 11 and 12 show the estimated snow for February 01, 

2004. It is observed that the model is incapable of detect and estimating deep snow. The scatter 

plot of the ground truth versus the estimate (Fig 11) illustrates the results in a quantitative way. 

The best fitted line (red line) is below the 1:1 line indicating underestimation of the estimate. 

Training Data 

(Days) 

Table 2: Estimating SWE by ANN model 

Validation 

Data (Day) 

Correlation 

Coe. 

RMSE Bias 

Dec06,Dec13 Dec 20 0.37 21 –14 

Dec06, 

Dec13,Dec20 

Jan04 0.43 17 1 

Dec06,Dec13, 

Dec20, Jan04 

Jan11 0.47 19 5 

Dec13,Dec20, 

Jan04,Jan11 

Jan18 0.44 30 –11 

Dec20,Jan04, 


Jan25 0.53 45 –34 

Jan04,Jan11, 


Feb01 0.47 44 –30 

Jan18,Jan25, 

Feb01 

Feb08 0.37 42 –31 


Feb01,Feb08 

Feb16 0.12 75 –58 

Jan25, Feb01, 

Feb08, Feb16 

Feb23 0 90 –83

180 

160 

140 

120 

100 

80 

60 

40 

20 

0 

163 



Figure 11. Estimated SWE by ANN (left) and SNODAS ground truth SWE (right), February 01, 2004 

CONCLUSION 

ANN output 

200 

180 

160 

140 

120 

100 

80 

60 

40 

20 

R = 0.533 

0 

0 50 100 

Snocover(SNODAS) 

150 200 

Figure 12. Estimated SWE by ANN vs. SNODAS ground truth SWE, February 01, 2004 

Three RADARSAT images were obtained to investigate the potential of RADARSAT SAR in 

estimating SWE. The images were processed and georefenced using PCIGeomatica. The Ground 

truth data were obtained from NOHRSC SNODAS dataset through NSIDC. The backscattering 

ratio of RADARSAT images was derived by subtracting them from a reference image. The 

analysis indicates that backscattering ratio has limited correlation with SWE (20 percent). An 

ANN model was used to explore non-linear relationships between backscattering ratio and SWE. 

The results showed low correlation between estimated and ground truth SWE. In order to 

introduce land cover characteristics, an NDVI image was added to the input of the ANN model. 

The results showed a more than 15 percent improvement in correlation coefficient. To improve the 

estimation the input classified based on SWE values. This improved the range of the estimated 

SWE although it did not change the correlation coefficients. Finally the resolution was changed to 

5km. It was concluded that where SWE is in the range of 10mm to 60mm the backscattering range 

is very large. Also, the backscattering range between two years (2004 and 2006) is very different. 

This indicates that developing a model to estimate SWE using backscattering ratio is very 

difficult. A solution for this problem is modifying the model for each year. 

180 

160 

140 

120 

100 

80 

60 

40 

20 

0

164 



In case of low resolution passive SSM/I and active QuikSCAT, the ANN model shows 

satisfactory result in dependent estimation of SWE. Also, adding QuikSCAT-Ku increased the 

accuracy of the estimated SWE by neural networks. It was concluded that estimating SWE by 

neural networks is highly dependent on training data. This can become a source of error on model 

development. In order to resolve this problem a large dataset is necessary. 


The Authors express their gratitude to Dr Jeff Hurley, a Senior Project Manager in 

RADARSAT International, for his support in processing and analyzing RADARSAT images. 

REFERENCES 

Bernier M, Fortin J (1998).The Potential of Times Series pf C-Band SAR Data to Monitor Dry 

and Shallow Snow Cover. IEEE Transaction on Geoscinece and remote sensing, 36(1): 226– 

242. 

Bernier M, Fortin J Y.Gauthier, R.Gauthier, R.Roy, and P. Vincent (1999i).Determination of 

Snow Water Equivalent using RADARSAT SAR data in eastern Canada..Hydrological 

Processes,13:3041–3051. 

Chang, A. T. C, L. Foster, D. K. Hall. (1987). Nimbus-7 SMMR derived global snow cover 

parameters. Annals Glaciology, 9:39–44. 

Derksen, E. LeDrew, A.Walker, and B. Goodison (2001). Evaluation of a Multi-Algorithm 

Approach to Passive Microwave Monitoring of Central North American Snow Water 

Equivalent. IEEE, 952–954. 

Hallikainen, M. T. (1984). Retrieval of snow water equivalent from Nimbus-7 SSMR data: effect 

of land cover categories and weather conditions. IEEE Oceanic Engineering, 9(5): 372–376. 

Hallikainene, M. T,P.Halme,M Takala, J.Pulliainen (2003). Combined Active and Passive 

Mirowave Remote Sensing of Snow in Finland. IEEE:830–832. 

Hallikainene, M. T,P.Halme,P.Lahtinen, M Takala, J.Pulliainen (2004). Retrieval of Snow 

Characteristics from Scapeborne Scatteromter Data. IEEE:1849–1852. 

Sokol. J, T.J. Pultz, and A.E. Walker (2003). Passive and Active microwave remote sensing of 

snow cover. INT. J. Remote Sensing, 24(24):5327–5344. 

Tedesco, J. Pullianinen, M. Takala, M Hallikainen, P. Pampaloni. (2004). Artificial neural 

network-based techniques for the retrieval of SWE and snow depth from SSM/I data. Remote 

Sensing of Environment 90:76–85. 

Walker, G. B. E. a. A. E. (1995). Canadian development and use of snow cover information from 

passive microwave satellite data. Passive microwave remote sensing of land–atmosphere 

interactions. B. J. Choudhury, Y. H. Kerr, E. G. Njoku and P. Pampaloni. The Netherlands, 

VSP BV 245–262.

165 



Enhancements and Forthcoming Developments to the Interactive 

Multisensor Snow and Ice Mapping System (IMS) 

ABSTRACT 

SEAN R. HELFRICH 1 , DONNA MCNAMARA 1 , BRUCE H. RAMSAY 2 , 

THOMAS BALDWIN 1 , AND TIM KASHETA 3 

The national oceanic and atmospheric administration’s national environmental satellite data, and 

information service (noaa/nesdis) interactive multisensor snow and ice mapping system (ims) has 

undergone substantial changes since the inception in 1997. These changes include the data sources 

used to generate the product, methodology of product creation, and even changes in the output. 

Among the most notable of the past upgrades to the ims are a 4 kilometer resolution grid output, 

ingest of an automated snow detection algorithm, expansion to a global extent, and a static digital 

elevation model for mapping based on elevation. Further developments to this dynamic system 

will continue as noaa strives to improve snow parameterization for weather forecast modeling. 

Several future short term enhancements will be evaluated for possible transition to operations 

before the northern hemisphere winter of 2006–07. Current and historical data will be adobted to a 

geographic information systems (GIS) format before 2007, as well. Longer-term enhancements are 

also planned to account for new snow datasources, mapping methodologies and user requirements. 

These modifications are being made with care to preserve the integrity of the long standing 

satellite derived snow record that is vital to global change detection. 

Keywords: satellite remote sensing; environmental data; snow cover; ice cover; geographic 

information systems; climate 

INTRODUCTION 

The interactive multisensor snow and ice mapping system (IMS) has been the main operational 

snow and ice charting tool of the National Oceanic and Atmospheric Administration’s National 

Environmental Satellite Data, and Information Service (NOAA/NESDIS) for almost a decade. 

This product was primarily created to improve the quality and timeliness of Northern Hemisphere 

snow maps for National Centers for Environmental Prediction (NCEP) numerical forecasting 

(Ramsay, 1998). Prior to the IMS’s operational inception in 1997, snow charts were constructed 

manually once per week. The IMS is produced daily using geographic information systems (GIS) 

technology. This system had substantial impacts on production speed, product spatial accuracy, 

and time between observations. A comparison and validation review of the product transition from 

1 

NOAA/NESDIS/OSDPD – NOAA Science Center, 5200 Auth Rd., Camp Springs Maryland, 

20746 sean.helfrich@noaa.gov 

2 

NOAA/NESDIS/STAR/CORP – NOAA Science Center, 5200 Auth Rd., Camp Springs 

Maryland, 20746 

3 

Riverside Technology, Inc. – 2290 East Prospect Road, Suite 1, Fort Collins, Colorado 80525

manual weekly to IMS daily charts was conducted between 1997 and 1999. Preliminary 

examination of the data between these periods suggests the IMS output to be superior to the 

weekly manual snow charts (Ramsay, 2000). In June 1999, the manual charting of snow extent 

was supplanted operationally with the daily IMS. Since the charts are now constructed digitally, 

their distribution has increased, with hundreds of known users viewing data each month from the 

NESDIS site and an unknown number of users obtaining the IMS data from other sources. 

While there are potentially many uses, the primary function of the product is to provide 

cryospheric input for environmental modeling. There are two operational government customers 

for this product, the NCEP / Environmental Modeling Center (EMC) and the NCEP / Climate 

Prediction Center (CPC). These customers help support and influence the direction of the product. 

The feedback from the NCEP modeling agencies and the preliminary NOAA Program 

Observational Requirements have led to advancements in the product and point toward continued 

improvement. The EMC applies the models for each three hour modeling run for North America 

and temporally coarser models for the entire planet. Snow plays an important role in model input 

and can lead to substantial error in forecast results based on incorrect representations of snow 

distribution, age, depth, snow water equivalent (SWE), and snow density (Mitchell et al., 1993; 

Sheffeld et al., 2003). 

Along with serving as an initial state of the surface cryosphere for the Northern Hemisphere for 

weather prediction, NOAA’s snow maps serve as a 40 year environmental monitoring record for 

hemispheric snow cover. This is considered the longest continuous satellite-derived record of any 

environmental variable (Robinson et al., 1993). It is vital for climate change detection and a key 

element in NOAA’s Mission Goals to “understand climate variability and change to enhance 

society’s ability to plan and respond” (USDOC/NOAA, 2005). Given the importance of this 

record, changes in the record should be considered with great care to preserve the integrity of the 

product for climate monitoring. Consultation within the snow and ice climate monitoring 

community has been sought before the integration of changes to safeguard the IMS’s 

environmental monitoring role. 

The IMS was designed to allow meteorologists to chart snow cover interactively on a daily basis 

using a variety of data sources within a common geographic system. Since first outlined by 

Ramsay (1998), there has been additional information discerned about the production 

methodology, and there have been noticeable changes in the input data sources, production 

techniques, and output format. This paper will cover changes in the input, production techniques, 

and output files since 1997, including statistics regarding the production methodology. The paper 

will also discuss the future enhancements and pending developments to the product, both short and 

long term plans. The conclusion will summarize the present and future of the product and what 

this means to the user community. 

IMS PRODUCT EVOLUTION 

System architecture enhancements 

A limiting factor of the original IMS system architecture was the inability of analysts to draw 

while looping imagery. This adversely affected the areas covered by geostationary satellites where 

imagery animation distinguishes snow and ice from fog and clouds. This limitation was changed 

in February 2004 to allow IMS analysts the freedom to loop imagery while drawing, erasing, or 

using any of the other IMS features. With this enhancement and other features such as image 

enhancement, product overlays and terrain mapping, the analysts can deduct snow and ice without 

relying on looking at a nearby system with looping imagery. 

Another feature modified within the system architecture in February 2004 was enhancing how 

the geographic area assignments of imagery were made within the system. The system before the 

change applied fixed areas for each satellite. These fixed geographic areas often only covered the 

best viewed regions. These regions also had set screen boundaries which did not allow analysts to 

recenter. The current IMS allows analysts to pan globally and select different datasets/satellite data 

166

independent of regions. This greatly enhanced the flexibility of the IMS to optimize the snow 

mapping display. 

Improved Resolution 

One of the largest changes in the product since the creation of the IMS has been the 

improvement in output resolution introduced in February 2004. This increased resolution was 

made to improve model input for the EMC, providing greater detail of snow and ice cover. These 

improvements were possible due to advancements in computer speed and imagery resolution that 

produced a higher resolution product at approximately 4km resolution (6144 × 6144 grid). 

Changes were made to all ancillary data layers such as coastlines, elevation, and vegetation cover 

to support the improved resolution. Impacts of this change are still under evaluation but positive 

feedback has been noted from National Weather Service (NWS) field stations in regard to snow 

season temperature predictions. This is in part due to the improvements in snow distribution 

mapping, as well as the resolving of the many water bodies, not possible on larger scale products. 

When not correctly mapped there can be significant forecast errors where sea or lake ice cover 

affects heat and moisture fluxes to the atmosphere (Grumbine, 2005). Figures 1A and 1B 

demonstrate the difference in resolutions over northern North America. This easily demonstrates 

the noticeable differences in the inclusion of interior lakes and more detailed coastlines. Snow on 

mountainous terrain is also better represented using the 4km versus lower resolution products. The 

4km product is also upscaled to the original previous resolution of approximately 24km resolution 

(1024 × 1024 grid). This is to maintain the satellite snow cover historical dataset. As previously 

mentioned, the IMS record is an important climate monitoring element and careful consideration 

must be taken to preserve the integrity of this snow cover record. Validation and monitoring of the 

IMS product at the 24km resolution is carried out under a joint effort by NESDIS and Rutgers 

University (Robinson, 2003). 

Added input data sources 

The IMS was designed to allow meteorologists to chart snow cover interactively on a daily basis 

using a variety of data sources within a common geographic system. The original input satellite 

data sources were outlined as NOAA polar orbiters (POES), NOAA geostationary (GOES) data, 

Japanese geostationary meteorological satellites (GMS), European geostationary meteorological 

satellites (METEOSAT), and US Department of Defense (DOD) polar orbiters (DMSP). Indirect 

satellite sources also include a weekly National Ice Center (NIC) chart and the US Air Force 

(USAF) daily snow depth & ice cover product (Ramsay, 1998). Several additional input products 

have been added to the IMS over the past decade. A few of these enhancements were outlined as 

prospects before, but have since become operational input options (Ramsay, 2000). These 

products include the Advanced Very High Resolution Radiometer (AVHRR) channel 3A, added in 

February 2001, the MOderate Resolution Imaging Spectroradiometer (MODIS) channel 1 added 

in February 2004, and an experimental automated snow mapping system over North America 

added in February 2004. Other product enhancements and their impacts are outlined in the 

following paragraphs. 

Meteosat 5 for INDOEX 

The original primary geostationary coverage left large portions of Siberia, central Asia, 

Himalayas, and the Tibetan Plateau unobserved by looping imagery. This is an important and 

difficult area for snow charting. Snow cover in this vast area has been identified as a significant 

influence on the Asian Monsoon (Hahn and Shukla, 1976; Huaqiang et al, 2004), global 

circulation (Bamzai and Marx, 2000; Clark and Serreze, 2000; Gong et al., 2003), and regional 

river discharge (Yang et al., 2003; Shaman et al., 2005). While non-geostationary satellite data 

sources such as polar orbiting imagery and microwave measurements provide mapping snow 

input, they are not the preferred data source by analysts. Microwave retrievals over the area are 

often erroneous in the winter due to atmospheric signal distortion, high elevation bare ground low 

temperatures, and/or soil grain scattering (Basist et al., 1996; Armstrong and Brodzik, 2001). 

Polar orbiters have limited over pass times, thus providing only a limited number of observations 

167

per day. This amplifies the common problem of misidentified snow and clouds in an already area 

known for its difficulty in visual identification from satellites (Clark and Serreze, 2000). 

The placement of Meteosat-5 at the equator and 63°E in 1998, helped fill the void of 

geostationary data. This satellite was moved in support of the Indian Ocean Experiment 

(INDOEX), and was first incorporated into IMS in March 2001. This platform was preferred over 

other geostationary meteorological satellites such as Feng Yun 2 (FY-2) from China and Indian 

National Satellite System 2 (INSAT 2) from India. The observation footprint is comparable for all 

these satellites, but only a single geostationary satellite is required. Meteosat-5 was chosen 

because FY2 is experimental and near the end of its serviceable period, while data from INSAT 

are available for Indian national use. The inclusion of Meteosat-5 has provided a great boost to 

Asian snow mapping during the winter season. 

Multi-functional Transport Satellite (MTSAT) 

The Japanese Meteorological Agency (JMA) MTSAT series succeed the Geostationary 

Meteorological Satellite (GMS) series as the next generation satellite series covering East Asia 

and the Western Pacific. While MTSAT’s offers only marginal improvements in visible resolution 

over GMS, the improved calibration and correction algorithms provide improved detection of 

snow. The IMS began using MTSAT in November 2005. 

National Operational Hydrologic Remote Sensing Center (NOHRSC) 

The inclusion of NOHRSC maps into IMS analysis began in February 2004. The national 

analysis provided by NOHRSC, called the SNOw Data Assimilation System (SNODAS), provides 

a 1 km resolution estimation of snow cover for the conterminous United States. SNODAS is a 

system that amalgamates NCEP modeling, multisensor, station report, and airborne information 

into a single daily or sub-daily product (Barrett, 2003). The output timing of this product 

corresponds to the IMS observation day and serves as an important winter input source when 

clouds blanket the conterminous United States. While the fine resolution and multi-source data of 

SNODAS provides reliable data, its spatial extent is limited compared to the IMS, so it provides 

only a small but nationally important area for snow mapping. 

MODIS Looping 

Soon after the inclusion of MODIS visible imagery as an IMS datasource, analysts found the 

utility of looping recent MODIS overpasses for a given location. Looping of this polar orbiter is 

available due in part to the Aqua and Terra satellites sharing the identical visible channel at 1km 

and the poles having multiple daily overpasses. While the time span between images used in the 

loop is somewhat coarse compared with geostationary observations, these loops allow the analyst 

greater ability to distinguish between the surface cyrosphere and clouds. MODIS is an 

experimental satellite and will not be followed by a direct legacy product. The merger of the DoD 

and NOAA satellites in the future, known as the National Polar-Orbiting Operational 

Environmental Satellite System (NPOESS), will provide a similar product as that of MODIS and 

hopefully can be exploited for image looping once NPOESS is launched and declared operational. 

Marine Modeling and Analysis Branch (MMAB) Sea Ice Analysis 

The tracking of sea ice cover presents many difficulties. The IMS relies primary on visible 

imagery but this method is contingent on clear sky or thin cloud during the observing periods. 

When weather or low illumination obscure visual interpretation of sea ice, microwaves play a 

greater role. IMS analysts often apply a 1/16 mesh Northern Hemispheric grid of sea ice 

concentrations from the MMAB to demarcate those locations with >50% ice cover. The MMAB 

sea ice analysis is solely based on Special Sensor Microwave Imager (SSM/I) and applies a 

modified version of the National Aeronautics and Space Administration (NASA) team algorithm 

to derive sea ice concentrations (Grumbine, 1996). All SSM/I based products suffer from melt 

water attenuation, coastal contamination, poor thin ice detection, and difficulties in identifying 

concentration along the marginal ice zone. Analysts apply the MMAB product cautiously and will 

168

slightly vary the IMS with this product when other external sources suggest the MMAB may be 

contaminated with a false emissivity. 

A B 

Figure 1. The 24km IMS product (A on the left) as remapped from the 4km IMS (B on the right) for May 19, 

2006. The 24km represents the same land/sea mask used in previous 24km products prior to February 2004. 

Note the differences in representations for mountainous snow, inland lakes, and sea ice leads. 

Daily NIC Ice Edge 

In its inception, the IMS was designed to exploit the NIC weekly ice charts for the Northern 

Hemisphere. The NIC produced detailed sea ice maps coded in an ice charting nomenclature 

known as egg code once per week, usually updating the product on Friday afternoons, until 2001. 

Since that time the NIC switched from weekly to biweekly hemispheric coverage. This decreased 

the utility of the charts for daily operational ice mapping. Furthermore, ice charts released on 

Fridays would use input data typically three to seven days old for the analysis. While the NIC 

charts could still provide a general outlook of ice thickness and ice distribution, the dynamic 

nature of ice made the charts too old for the IMS analysis. Since February 2004, a daily vector sea 

ice edge has been incorporated into the IMS. The NIC daily sea ice vector product applies visible, 

infrared, passive microwave, and radar data to outline those areas with >10% ice cover. This 

product differs from the IMS product in three key ways. First, the NIC product is vector-based and 

attempts to enclose amorphous polygons, while the IMS defines ice cover within predefined 

pixels. The NIC ice cover has no set size requirements on the polygon size or shape, thus the scale 

of what areas enclose >10% is at the NIC analyst’s digression. This often leads to smoothing along 

the ice edge in some areas, while other areas may be more detailed than the IMS. Experience has 

revealed more of the former than the latter. A second difference in the products is the ice 

concentrations captured by each product. The NIC outlines areas with >10% ice cover, while the 

IMS demarcates areas with >50% ice cover. IMS analysts must adjust for this difference when 

using the NIC daily ice edge for IMS input. A third difference in the ice analysis output is the 

IMS’s inclusion of lake ice. The NIC vector ice edge only outlines sea ice and lake ice in the Great 

169

Lakes. Other significant lake bodies that freeze (Great Bear, Great Slave, Caspian Sea, Lake 

Baikal, Aral Sea, among many others) are not included within the NIC product. 

Despite the differences, IMS analysts will bridge the analysis outputs and methodologies to use 

NIC data as an input source. NIC data is often applied with or taken in the context of the MMAB 

sea ice product to provide a best guess approach for the IMS. Figure 2 shows how the NIC ice 

cover product is able to suggest ice cover in the Hudson Bay even when imagery has cloud 

contamination. 

Figure 2. AVHRR Channel 1 with the NIC ice edge overlayed for May 29, 2006. The NIC ice edge is a 

vector file with shorelines. White lines represent shoreline and 10% ice cover isopleth. 

One can notice that each product suggests a different ice cover pattern, perhaps due to times of 

observations. But the NIC ice edge is able to provide IMS analysts information regarding the 

shape of the lead in James Bay, even through the southern part of James Bay is cloud-covered. 

Plans are being developed to bring the radar data available to NIC analysis into the IMS and to 

have an NIC ice edge product that outlines using similar criteria as that of the IMS. This will be 

discussed later in this paper in greater detail. 

Automated Snow and Ice Cover Products 

In August 1999, NOAA/NESDIS began the production of automated snow maps over North 

America. The product generates daily maps of 4km resolution based on visible, near-infrared, 

middle-infrared, infrared and microwave imagery. This imagery comes from both polar-orbiting 

and geostationary satellites. The algorithm applies a series of decision-trees to bin pixels 

containing either a majority of the area snow covered or having less then a majority of area snow 

covered (Romanov et al., 2000). While the midlatitudes maps are generally mapped using GOES 

170

imagery, the higher latitudes rely on polar orbiter spectral differences and microwave signals. 

Microwave retrievals are the default observation when shorter wavelength data is attenuated by 

clouds for several days. Since being declared operational the product has expanded beyond North 

America to the Southern Hemisphere and the remainder of the Northern Hemisphere, though the 

spatially expanded versions of the automated snow maps remain experimental at the present time 

(Romanov and Tarpley, 2003). Examples of the Northern Hemisphere multisensor product are 

demonstrated on Figure 3. The pattern closely resembles that of the IMS for the same date (IMS 

not shown). Figure 4 displays an example of the experimental automated Southern Hemisphere 

product. Validation efforts remain ongoing for both hemispheric products. Other comparison 

studies using automated snow cover mapping versus IMS suggest that IMS may have 

underestimation problems in the transition season but outperform automated products in cloudy 

areas with new snow cover and during winter (Brubaker et al., 2005). The validation efforts will 

help determine how these products will be incorporated with the IMS. Without a current Southern 

Hemisphere IMS product, an automated product could serve as the NOAA NESDIS Southern 

Hemisphere output in the future. However should the product be unable to provide a serviceable 

input for EMC or CPC modeling efforts, the output will merely serve as one input to a southern 

hemispheric IMS analysis that will need to be created. Likewise, the role the Northern 

Hemispheric automated analysis product plays in the IMS will be determined based on the 

validation results and the customer requirements. Possible scenarios include serving as another 

layer in the IMS (much like the North American product does now), serving as the initial state of 

the IMS, or even replacing the current Northern Hemispheric IMS product. 

Current Production Methodology for IMS 

Since the inception of the IMS, production methodologies and image preferences have become 

more transparent. The production of the IMS has evolved over time, with inclination in which 

imagery types are applied at certain times of year and how long the production process takes. This 

section will provide a greater insight into the production methodology used to make multisensor 

output. Production of the IMS products are not tremendously time consuming for analysts, who 

spend anywhere from one to five hours generating a daily product depending on the season, 

analyst familiarity, and satellite data available. The month of August requires the lowest average 

time to conduct an analysis, averaging between 70 and 75 minutes. Production time during the 

accumulation season (Oct–Apr) averages about 120 minutes. The ablation season (May–Sep) 

averages approximately 90 minutes for production time. Products are due to the primary customer 

by 2300z. The IMS analysis currently starts with the previous day’s analysis as the initial state. 

The analyst then reconfigures the current IMS based on the input data available at the time of the 

analysis. 

Seasons determine not only just how much time is required to generate a chart, but also what 

data will be used as input into the IMS. Geostationary data looping is the primary tool for 

determination of snow cover (Ramsay, 1998). 

171

Figure 3. Examples of the experimental NOAA automated snow and ice multisensor retrieval output for May 

30, 2006 over Eurasia (above) and North America (below). 

The geostationary looping represents an estimated 60% of the snow analysis examination areas 

during the winter (Dec–Feb). This decreases to an estimated 30% of the snow examination area 

during the summer (Jun–Aug). During the summer months, polar orbiting satellites’ visible 

channels (bands) characterize an estimated 65% of the snow analysis. Analysts generally prefer 

using visible imagery for snow extent mapping, but will use microwave data in the event that light 

is unavailable, due to cloud occlusion or low solar illumination angles. The combination of high 

albedo, static motion, and meteorological conditions provides the analyst with 80–90% of input 

data used in the analysis of observed snow cover. Even during the winter, microwave derived 

snow data generally only represents 5% of an analysis. Microwave snow extent data have welldocumented 

snow misidentification errors due to signal obstructions, snow grain size, land cover 

influences, and algorithm limitations (Chang et al., 1996; Foster et al., 1999; Foster et al., 2004). 

Analysts rely more on snow climatology to estimate snow cover in the high latitudes during the 

winter than pure microwave data. Where and when sources are available the IMS uses METAR 

and cooperative observations, NOAA NWS’ NOHRSC, and SNODAS data, and automated snow 

cover maps from both NOAA’s Center for Satellite Applications and Research (STAR) and the 

172

MODIS Land Science Team, NASA Goddard Space Flight Center (GSFC). These external data 

sources are often used to validate an area having snow or snow obscured with clouds. Often, the 

IMS analyst must use a consensus of data sources to provide an optimal “best-guess” approach to 

determining the presence of snow. 

Figure 4. Examples of the experimental NOAA automated snow and ice NOAA-17 retrieval output for May 

30, 2006 over Southern Hemisphere. This product is at 4km resolution and could expand IMS from the 

Northern Hemisphere to global coverage. Green areas represent areas scanned for snow cover, while grey 

areas represent land areas not scanned for snow cover. 

173

Ice cover analysis relies on a different approach than snow cover for charting. Ice cover 

determination must rely less on high albedo, stagnate cover, and meteorological conditions. Sea 

ice in the Northern Hemisphere winter is primarily located in areas with low solar illumination. 

New ice formation often has a low albedo until the ice thickens, becomes more opaque, and albedo 

increases (Wadhams, 2000). Furthermore, sea ice is a dynamic surface making it less discernible 

from clouds using image loops. The presence of lake and sea ice can be unrelated to current 

meteorological conditions due to ice transport, ice thickness, water temperatures, among other 

factors. The prominence of low-level stratus clouds over polar and sub-polar region also preclude 

the use of visible imagery as the primary source of ice observation. While this reduces the 

efficiency of albedo-based observation of ice, it is still a valuable input. On average, about 60% of 

changes in winter ice cover are noted via geostationary, AVHRR or MODIS observations. Much 

of the higher latitude areas are verified as being ice-covered using microwave-based retrievals. 

Microwave-based observations often represent 30–35% of the winter and autumn (Sep–Nov) ice 

cover input. Ice climatology is another tool for estimating ice cover in places where observations 

are unavailable. Since ice cover often exists in remote and dangerous areas, no station data is 

currently incorporated into the analysis. The NIC produces a sea ice edge vector file that provides 

the IMS with an external source for ice cover information. Currently, the NIC ice edge 

encompasses total polygons with greater than 10% ice cover. The IMS attempts to identify 

whether each 4km × 4km pixel contains more than 50% ice cover. These products do not 

correspond directly due to each product requiring different output specifications. Despite the 

differences between products, the NIC ice edge is used when other sources of data fail to provide 

any clear input on ice-covered ocean or Great Lakes waters. This represents about 2–10% of the 

time, depending on the season. 

Mountainous snow mapping 

Elevation plays an important role in snow generation due mostly to orographic lifting and 

temperature decreasing with increasing height. Snow often outlines higher elevation areas during 

transition season and during the winter in semi-arid, mid-latitude regions. To mimic this effect in 

mapping snow, the IMS allows analysts to chart areas dynamically as having snow based on a 

digital elevation model (DEM). The DEM resolution is 4km and matches the IMS, thus providing 

a direct relationship between elevation and IMS pixels. This provides the analyst with the ability 

to toggle the pixels within a given polygon to match the outline of snow revealed through imagery. 

The analyst can optimize the snow cover pattern based on elevation, local geography, and 

reflectance revealed through imagery. This has become a frequently applied tool in IMS snow 

mapping. 

The strengths and shortcomings of this DEM-based mapping are considered by the analyst while 

applying this tool. The strength is a more detailed and realistic mapping technique than previously 

available based on a physiographic relationship of snow with elevation. A weakness is the DEM 

based mapping does not account for other known state or physiographic factors that play a role in 

snow cover distribution, such as slope and aspect. Nor does it take solar, vegetation, or 

climatologic wind and storm patterns into account. Studies reveal that physiographic features such 

as radiation, elevation, slope, and aspect account for between 50–80% of snow depth variability in 

the Rocky Mountains, Sierra Nevadas, and Alps (Marchand and Killingtveit, 2001; Balk and 

Elder, 2000). Elevation tends to be the second largest influence on snow cover distribution next to 

radiation, but the weight of this influence is dependent on scale (Balk and Elder, 2000, Marks et 

al., 2002). Elevation and radiation appear to be greater factors at increasing scales, likely playing a 

large role in distribution variability at the 4km scale in semi-arid and mountainous environments. 

Despite the shortcoming of this tool, it is just a methodology for mapping, with analysts basing 

snow distribution on numerous input data not merely elevation. Analysts can compensate for 

inhomogeneous spatial patterns noted with regional elevation due to the other state factors that 

influence snow cover distribution. 

174

FUTURE OF IMS 

Enhancements to the IMS will continue to push the bounds of cryospheric observation and 

charting. Requirements for snow and sea ice extent differ for climate verse numerical weather 

prediction. Since surface cyrospheric climate datasets are constrained by past data with the 

previous scale limitations, data needs to be maintained at original resolutions to preserve the 

climate record integrity (Robinson, 2003). Downscaling of older datasets would be required to 

blend the old and new data into a common 4km grid with interpolated daily values from weekly 

values. Fox and Ghan (2004) point out another limitation to the current computational limits that 

fine resolution climate grids would need to overcome as well. While the IMS’s resolution for 

global climate outlooks and monitoring exceeds the current requirements by using a 4km 

resolution at one observation per day, improvements in weather prediction are predicated on the 

prediction grid resolutions and prefer up to date observations. The need for global snow 

information at improved spatial and temporal resolutions for numerical weather prediction models 

is driving the advancements in the IMS. A 4km resolution snow and ice cover has the spatial 

resolution to initialize models such as the NCEP North American Mesoscale (NAM) model with 

12km resolution and could even provide improved ice and snow initialization for finer resolution 

models such as the Fifth-Generation Penn State/National Center for Atmospheric Research 

Mesoscale Model (MM5) (Dudhia et al., 1999; Rogers et al., 2001). The temporal resolution of 

once per day could improve model results since afternoon (Eastern Standard Time, EST) model 

runs may be up to 21 hours removed from the last snow and ice cover initialization. With daily 

snow depth depletions of over 12 inches reported at SNOTEL stations, spatial distribution of snow 

cover may change drastically over one day, given ideal weather conditions for ablation. IMS will 

attempt to respond to this need for more timely information by introducing a second IMS 

observation over North America at the 4km resolution. This will be changing given time 

constraints of Satellite Analysis Branch (SAB) analysts and the window of visible imagery 

available for analysis by the late afternoon EST model run, particularly in the western United 

States in the winter. 

In addition to a second IMS daily product, the IMS will be expanding to provide global 

coverage. While the IMS provides adequate coverage in the Northern Hemisphere, the product 

failed to capture the snow and ice extent in the Southern Hemisphere. Like the improved temporal 

resolution North American IMS product, this presents a challenge to the resources required to 

provide such data. As previously mentioned, the automated snow and ice product is likely to play 

a large role in the production of southern hemispheric analysis. The completion of the Southern 

Hemisphere IMS will complete the global snow and ice coverage for model initialization. 

The IMS currently employs over 15 separate sources of data for input. This number can seem 

daunting to navigate, but each source is expertly selected to provide an optimal snow analysis. 

Still, NOAA is looking to exploit new technologies for understanding the current state of the 

surface cryosphere. To improve the output and to meet future product requirements, several new 

products are being tested for implementation into the IMS. In the short term, these products 

include snow and sea ice cover from the Advanced Microwave Scanning Radiometer-EOS 

(AMSR-E), the Northern and Southern Hemisphere automated snow mapping systems, NASA’s 

Quick Scatterometer (QuikSCAT), and ESA’s Environmental Satellite (Envisat) Advanced 

Synthetic Aperture Radar (ASAR) Global Monitoring Mode (GMM), and MetOp’s Advanced 

Scatterometer (ASCAT). MetOp’s impending launch will also offer an expansion in the platforms 

carrying the AVHRR and Advanced Microwave Sounding Unit (AMSU) sensors. Recent 

improvements to the Air Force Weather Agency (AFWA) snow depth and MMAB sea ice 

products will also be incorporated in the near future. 

The IMS output product has been available to users for almost ten years now. Archival and 

archived product dissemination has been done through cooperation with the National Snow and 

Ice Data Center (NSIDC). The NSIDC currently provides users with American Standard Code for 

Information Interchange (ASCII) output data at the original 24km product as well as the recently 

added 4km output. While these products have been a popular data source, the formatting of the 

output can be complex to novice users. To promote a broader user community, NOAA NESDIS 

175

has begun to generate GIS GeoTiff compatible output at the 4km resolution. This product will be 

archived and disseminated at the NSIDC. The GeoTiff archive will span from February 2004 until 

the latest analysis day when complete. 

Snow and ice extent and coverage has been the primary output for the IMS. However, this is far 

from the lone variable needed for modeling snow and ice behavior at regional and global scales. 

NOAA/NESDIS is at the cusp of introducing new snow products that work in conjunction with the 

IMS to improve initialization in atmospheric models. A common problem reported with IMS has 

been continuing to keep snow cover during cloud obscured periods. The IMS analysts apply many 

tools and images to produce a “best guess” approach to snow observing. However, IMS analysts 

leave conditions as they were since the last observation when clouds obscure visible observations, 

snow is too thin for microwave detection, and there are no station reports. This can be problematic 

when snow has actually melted away. Atmospheric models contain algorithms to estimate snow 

depth throughout day and predict ablation of snow cover. However, snow ablation in atmospheric 

models is reinitialized with the current snow extent from IMS. If the snow in the IMS is merely 

the result of continuance and not observed, this reinitialization can lead to false snow observations 

and propagate errors throughout the NCEP model. A file of last observation time for the IMS has 

been developed and is undergoing testing. This will allow modelers to choose to use IMS for snow 

cover observation or to base snow cover on modeled estimates. Snow water equivalent (SWE) has 

been produced by SSM/I measurements for many decades, and is a valuable snow variable for 

atmospheric and hydrologic modelers. NOAA NESDIS is currently testing a combined AMSU 

and IMS SWE product that will merge the reliable IMS snow cover observations with AMSU’s 

capacity for estimating SWE (Kongoli et al., 2006). An example of the pre-merged and merged 

products is demonstrated in Figure 5. Additional snow variables like snow depth and fractional 

snow covered area are being experimented with to improve model initialization in combination 

with IMS output (Romanov et al., 2003). Future NOAA efforts for sea and lake ice variables 

utilizing IMS ice cover such as ice concentration and ice thickness are in the planning stage. 

The enhancements of input and output are not the only short-term changes planned for the 

product’s future. Current plans are to relocate the operational production of the IMS from its 

current location of SAB to the NIC. This transition from one agency within NESDIS to another is 

being done with the hope of producing an improved product at a reduced cost for NESDIS as a 

whole. This should lead to less duplication in sea ice monitoring within NESDIS by parallel 

offices, consolidate network systems and imagery storage, free SAB personnel time for other 

products, and allow time for NIC personnel trained for IMS analysis to meet NCEP requirements 

of two IMS observations per day. All NIC personnel will undergo the same training as SAB 

personnel and meet IMS qualifications before being assigned to IMS product generation. Parallel 

product generation will be performed at the SAB and NIC until comparable output products are 

obtained between offices. After evaluation and duplication of the IMS has been achieved, 

production of the IMS will transition fully to the NIC. No production methods other then 

personnel will change during this process. 

Longer-term plans for enhancements to the IMS input data revolve around the future 

deployment of NPOESS and GOES-R. NPOESS will be a joint military and civilian satellite 

replacing many of the existing U.S. polar orbiting sensors with improved sensors such as the 

Visible Infrared Imager Radiometer Suite (VIIRS), Conical Scanning Microwave Imager/Sounder 

(CMIS), and Advanced Technology Microwave Sounder (ATMS). 

GOES-R will be the next generation of NOAA geostationary satellites that will aide snow and 

ice observations. Geostationary satellites are regarded by analysts as the most valuable input 

source. IMS analysts assessing surface cryospheric coverage will benefit greatly by the 

advancements in remote sensing provided on GOES-R, particularly the Advanced Baseline Imager 

(ABI). The ABI will have 16 spectral bands, compared with five on the current GOES imagers. 

The ABI will improve the spatial coverage from 1km to 0.5km at nadir for broadband visible and 

from 4km to 2km for the infrared bands. For snow and ice detection this will improve the ability 

of the analyst to confirm the presence of ice on the surface through recognition of spatial patterns 

more discernable at higher resolutions, such as dendritic spatial patterns of snow on mountains, ice 

floe shapes that indicate certain ice thickness, or ice fractures. The ABI also includes spectral 

176

information never present before on GOES imagers. One band of particular relevance to snow and 

ice detection will be centered at 1.61μm, which would expand the ice/cloud discrimination 

sampling beyond the temporally coarse polar orbiters (Schmit et al., 2005). This should improve 

the IMS accuracy and reduce the amount of time required for detection. This channel differencing 

would also improve automated snow and ice detection. GOES-R will increase the coverage 

acquisition rate by nearly fivefold, allowing closer to real-time observations and increased 

discrimination of relatively static surface features from highly dynamic atmospheric features. 

Figure 5. Examples of the experimental NOAA merged AMSU SWE-IMS output (bottom) for February 1, 

2006 over the Northern Hemisphere. The pre-merged AMSU SWE (top) has erroneous or questionable 

signals masked by using the IMS. 

177

SUMMARY 

The IMS underwent enhancements to the resolutions, input data, methodology, and output 

formats that augmented the product’s utility. These changes have been largely beneficial to the 

NCEP EMC modeler (Mitchell, K., 2006, professional conversation). The changes involve an 

improved architecture, superior output resolution, expanded input sources, and topographic 

mapping capabilities. The current system architecture allows IMS analysts to quickly record snow 

and ice, thus speeding the production time. The advanced resolution product begun in February 

2004 allows for greater detail in the snow and ice information to be conveyed. While product 

evaluations are still ongoing at NCEP, improvements in the resolution are likely to have a positive 

impact on numerical weather prediction. The expanded imagery from which the user may derive 

data sources allows for an increase in the likelihood of correct snow and ice identification. The 

more accurately the surface cryosphere is depicted, the better chance a weather prediction model 

should have at accurate forecasts. 

The enhancements made to the IMS were mostly driven by the need for better NCEP EMC 

initializations. Even the planned changes in the product are geared to improving NCEP EMC 

forecasts. Unfortunately, the impacts of the enhancements to NCEP CPC climate evaluation 

remain unknown. The product has likely become more accurate due to improved spatial 

resolution, enhanced methodologies, and improved input sources. However, the impact of an 

improved product may cause heterogeneity in the snow and ice data record is still under 

evaluation. While consultation of change impacts is sought by the CPC and non-federal 

researchers, formal examinations are to date forthcoming. This concern is not without note and is 

in need of further investigation. 

Hopefully, providing the data in new formats and with accompanying products will expand the 

user community for the products. Further advances in the products are predicated on customer 

requirements, new sensors, and advanced technologies. The advancements in the IMS, as well as 

links to past snow mapping, should continue to make this a viable snow mapping system for years 

to come. 


The authors thank Dr. Dan Tarpley and Dr. Peter Romanov, Office of Research and 

Applications, NOAA/NESDIS; and Dr. Cezar Kongoli, QSS Group Inc. for their direct 

contributions to this work. The authors also wish to acknowledge the important contributions and 

support of, Dr. Ken Mitchell, NOAA/NWS; Mr. Ricky Irving, PIB NOAA/NESDIS; and Dr. 

David Robinson, Rutgers University. 

REFERENCES 

Armstrong, R. L. and M. J. Brodzik, 2001. ‘Recent Northern Hemisphere snow extent: a 

comparison of data derived from visible and microwave satellite sensors.’ Geophysical Res. 

Letters, 28(19), 3673–3676. 

Balk, B.C., and K. Elder. 2000. ‘Combining binary decision tree and geostatistical methods to 

estimate snow distribution in a mountain watershed.’ Water Resources Research., 36, 13–26. 

Bamzai, A. S. and L. Marx, 2000: ‘COLA AGCM simulation of the effect of anomalous spring 

snow over Eurasia on the Indian summer monsoon.’ Quart. J. Roy. Meteor. Soc., 126, 2575. 

Barrett, A., 2003. National Operational Hydrologic Remote Sensing Center SNOw Data 

Assimilation System (SNODAS) Products at NSIDC. NSIDC Special Report 11. Boulder, CO, 

USA: National Snow and Ice Data Center. 19 pages 

Basist, A., D. Garrett, R. Ferraro, N. Grody, and K. Mitchell, 1996. ‘A comparison between snow 

cover products derived from visible and microwave satellite observations.’ J. Appl. Meteorol., 

35, 163–177. 

178

Brubaker, K.L., R. Pinker and E. Deviatova, 2005. ‘Evaluation and Comparison of MODIS and 

IMS Snow Cover Estimates for the Continental United States Using Station Data’. J. 

Hydrometeorology, 6(6), 1002–1017. 

Chang, A.T.C., J.L. Foster and D.K. Hall, 1996. ‘Effects of forest on the snow parameters derived 

from microwave measurements during the BOREAS winter field campaign.’ Hydrol. 

Processes, 10, 1565–1574. 

Clark, M.P., and M.C. Serreze, 2000. ‘Effects of variations in East Asian snow cover on 

modulating atmospheric circulation over the North Pacific Ocean.’ J. Climate, 13, 3700–3710. 

Dudhia, J., D. Gill, Y-R. Guo, D. Hansen, K. Manning, and W. Wang, 1999. PSU/NCAR 

Mesoscale Modeling System Tutorial Class Notes and Users' Guide: MM5 Modeling System 

Version 2. Mesoscale and Microscale Meteorology Division, NCAR, 182 pages. 

Foster, J.L., Hall, D.K., Chang, A.T.C., Rango, A., Wergin, W. and Erbe, E. 1999. ‘Effects of 

snow crystal shape on the scattering of passive microwave radiation.’ IEEE Trans. Geosci. 

Remote Sens., 37, 1165–1168. 

Foster, J., C. Sun, J.P. Walker, R. Kelly, J. Dong, and A. Chang, 2004. ‘Mapping random and 

systematic errors of satellite-derived snow water equivalent observations in Eurasia’, Proc. 

SPIE Int. Soc. Opt. Eng.,, M. Owe, G. D'Urso, B.T. Gouweleeuw, and A.M. Jochum, Eds., 

5568, Maspalomas, Gran Canaria, Spain, September 14–16, 2004, 150–158. 

Gong, G, D. Entekhabi, and J. Cohen. 2003. ‘Relative impacts of Siberian and North American 

snow anomalies on the winter Arctic Oscillation’, Geophysical Res. Letters, 30(16), 1848. 

doi:10.1029/2003GL017749. 

Grumbine, R. W., 1996. Automated Passive Microwave Sea Ice Concentration Analysis at NCEP 

OMB Tech. Note 120, 13 pages. 

Grumbine, R. W., 2005. http://meted.ucar.edu/nwp/pcu2/avice1.htm. 

Hahn, D. J. and J. Shukla, 1976. ‘An apparent relationship between Eurasian snow cover and 

Indian monsoon rainfall.’ J. Atmos. Sci., 33, 2461–2462. 

Hall, D.K., J.L. Foster, V.V. Salomonson, A.G. Klein and J.Y.L. Chien, 2001. "Development of a 

Technique to Assess Snow-Cover Mapping Errors from Space," IEEE Transactions on 

Geoscience and Remote Sensing, 39(2), 432–438. 

Hall, D.K., A.B. Tait, J.L. Foster, A.T.C. Chang and M. Allen, 2000. ‘Intercomparison of satellitederived 

snow-cover maps.’ Annals of Glaciology, 31, 369–376. 

Huaqiang, L., S. Zhaobo, W. Ju, and M. Jinzhong. 2004 ‘A Modeling Study of the Effects of 

Anomalous Snow Cover over the Tibetan Plateau upon the South Asian Summer Monsoon.’ 

Advances in Atmospheric Sciences, 21(6), 964–975. 

Kongoli, C, C.A.Dean, S.R. Helfrich, and R.R.Ferraro, 2006. ‘The retrievals of snow cover extent 

and snow water equivalent from a blended passive microwave – interactive multi-sensor snow 

product’ Proceedings of the 63rd Eastern Snow Conference. 7–9 June, 2006. Newark, 

Delaware. 

Marchand, W-D and Å. Killingtveit, 2001 ‘Analyses of the Relation between Spatial Snow 

Distribution and Terrain Characteristics.’ Proceedings of the 58th Eastern Snow Conference. 

14–17 May, 2001. Ottawa, Canada. 

Marks, D.G., Winstral, A.H., and Seyfried, M.S. 2002. ‘Simulation of Terrain and Forest Shelter 

Effects on Patterns of Snow Deposition, Snowmelt and Runoff over a Semi-Arid Mountain 

Catchment.’ Hydrol. Processes, 16, 3605–3626. 

Mitchell, K. E., G. J. DiMego, and T. L. Black, 1993. ‘Case study on the impact of improved snow 

cover and sea ice on wintertime forecasts of NMC’s regional NGM and Eta models.’ Preprints, 

13th Conf. of Weather Analysis and Forecasting, Vienna, VA, Amer. Meteor. Soc., 516–517. 

Ramsay, B., 1998. ‘The interactive multisensor snow and ice mapping system.’ Hydrol. Processes, 

12, 1537–1546. 

Ramsay, B., 2000. ‘Prospects for the Interactive Multisensor Snow and Ice Mapping 

System.(IMS)’ Proceedings of the 57th Eastern Snow Conference. 18–19 May 2000, Syracuse, 

New York. 161–170. 

Robinson, D.A., K.F. Dewey and R.R. Heim Jr., 1993. ‘Global snow cover monitoring: an 

update.’ Bull. Am. Meteorol. Soc., 74, 1689–1696. 

179

Robinson, D.A., 2003. ‘Recent variability of northern hemisphere snow cover.’ Preprints: Seventh 

Conference on Polar Meteorology and Oceanography, Hyannis, MA, American 

Meteorological Society, paper 13.12, 6. 

Rogers, E., T. Black, B. Ferrier, Y. Lin, D. Parrish, and G. DiMego, 2001. ‘Changes to the NCEP 

Meso Eta Analysis and Forecast System: Increase in resolution, new cloud microphysics, 

modified precipitation assimilation, modified 3DVAR analysis.’ NWS Technical Procedures 

Bulletin 488, NOAA/NWS, Washington DC. 

Romanov, P., Gutman G. and Csiszar I., 2000. ‘Automated monitoring of snow cover over North 

America with multispectral satellite data’ J. Appl. Meteorol., 39, 1866–1880. 

Romanov, P., D. Tarpley, G. Gutman, and T. Carroll, 2003. ‘Mapping and monitoring of the snow 

cover fraction over North America.’ J. Geophys. Res., 108(D16), 8619. 

Romanov, P. and D.Tarpley, 2003. ‘Automated monitoring of snow cover over South America 

using GOES Imager data. Int. J. Remote Sens., 24(5), 1119–1125. 

Schmit T.J., M. M. Gunshor, W.P. Menzel, J. Li, S. Bachmeier, J.J. Gurka, 2005. ‘Introducing the 

Next-generation Advanced Baseline Imager (ABI) on GOES-R’. Bull. Amer. Meteor. Soc., 8, 

1079–109. 

Shaman, J., M.A.Cane, A.Kaplan, 2005. ‘The relationship between Tibetan snow depth, ENSO, 

river discharge and the monsoons of Bangladesh.’ Int. J. Remote Sens., 26(17), 3735–3748. 

Sheffeild, J.; Pan, M.; Wood, E.F.; Mitchell, K.E.; Houser, P.R.; Schaake, J.C.; Robock, A.; 

Lohmann, D.; Cosgrove, B.; Duan, Q.; Luo, L.; Higgins, R.W.; Pinker, R.T.; Tarpley, J.D.; 

Ramsay, B.H., Journal of Geophysical Research-Atmospheres, November 21, 2003 (Vol. 108, 

No. D22, 8849, Paper No. 10.1029/2002JD003274) 

United States Department of Commerce, National Oceanic and Atmospheric Administration 

(USDOC / NOAA), 2005. New Priorities for the 21 st Century – NOAA’s Strategic Plan, 

Updated for FY 2006 – FY 2011. NOAA's Strategic Planning Office, Silver Spring, MD. 

Wadhams, P., 2000. Ice in the ocean. Gordon and Breach Science Publishers. 351. 

180

181 



Retreat of Tropical Glaciers in Colombia and Venezuela 

from 1984 to 2004 as Measured from ASTER and Landsat Images 

ABSTRACT 

JENNIFER N. MORRIS, 1 ALAN J. POOLE, 2 AND ANDREW G. KLEIN 3 

Like glaciers throughout the world, tropical glaciers in the northern Andes have retreated during 

the last century. In this study, the retreat of Andean glaciers in Colombian and Venezuela has been 

mapped from Landsat and ASTER images acquired between 1984 and 2004. Glacier retreat has 

been mapped for three glaciated regions in Columbia, the Sierra Nevada de Santa Marta, the Sierra 

Nevada del Cocuy, and the Ruiz-Tolima Massif as has the retreat of the single remaining glacier 

on the Pico Bonpland Massif in Venezuela. From the Landsat archives, several satellite images 

spanning the Landsat record were selected based on cloud cover over the glaciers and which 

provided adequate temporal coverage. All selected satellite images of the study sites were coregistered 

to each other with a root mean square (RMS) error for the ground control points of less 

than 20 meters. Snow and ice extent in each image was classified using the Normalized Difference 

Snow Index (NDSI) method and a density slice was used to create a binary snow/ice map. The 

total glaciated area on each individual peak was then determined from the classified snow and ice 

areas. This approach was found to be effective in determining glacier area. Incorporating historical 

data, a time series of glacier retreat from the early 1950s to 2003 was established. All four studied 

regions studied showed glacier retreat over this period. In the 1950s, the overall area of glaciers 

studied in Colombia and Venezuela glaciers was 91.36 km 2 in the 1950s. By 2003, the glaciers 

had retreated to approximately 62.07 km 2 which represents a total ice loss of 32%. 

Keywords: Remote Sensing, Colombia, Venezuela, Tropical Glaciers 

INTRODUCTION 

As one indicator of global change, glaciers around the world have been studied. Topical glaciers 

are of particular interest to study because the distinct characteristics of tropical climates make 

glacier-climate interactions different from the mid- and high-latitudes (Kaser 1999). However, not 

all tropical glaciers have been comprehensively studied because their remoteness can limit 

accessibility and extensive cloud cover can preclude extended temporal studies from remote 

sensing. The recent increase in availability of satellite images over glacier regions has increased 

the likelihood of studies throughout these regions. 

The tropical glaciers of Colombia and Venezuela, the focus of this study, were at a recent 

maximum extent during the Little Ice Age and have receded since this time (Kaser 1999). 

1 

Department of Geography, MS 3147, Texas A&M University, College Station TX 77843-3147 

USA email: jenmorris@tamu.edu. 

2 


USA email: alanpoole@tamu.edu 

3 


USA email: klein@geog.tamu.edu

Utilizing historical data and satellite observations, glacier retreat in Columbia and Venezuela is 

measured over the past four decades. 

The objective of this study is to quantify changes in the area of glaciers in several areas of 

Colombia and Venezuela using both historical data and measurements from satellite images to 

construct a time series starting in the early 1950’s and ending in 2004. Glaciers in the following 

four regions have been studied: the Pico Bonplad Massif in Venezuela, the Sierra Nevada de Santa 

Marta, the Sierra Nevada de Cocuy and the Ruiz-Tolima Massif in Columbia. The study locations 

are illustrated in Figure 1. Glacier retreat on the Nevado del Huila, the only other glaciated region 

in Colombia, was not studied. 

Figure 1: Map of Study Area 

STUDY SITES AND THEIR GLACIAL HISTORY 

The glaciers of the northern tropical Andes Mountains studied are located in Venezuela and 

Colombia along the western edge of South America and can be broken into four distinct glacial 

regions (Figure 1). 

The Pico Bonpland Massif, of Venezuela, was formally the site of numerous glaciers extending 

around the peak with an elevation of 4,983 meters; now it is site of the last measurable Venezuelan 

glacier, Sinigüis located at 8º32′ N and 72º00′ W. 

The Sierra Nevada de Santa Marta in northwestern Colombia is part of the Cordillera Central 

branch of the Andes Mountains extending into Colombia. Located near the Pacific coast, the 

Sierra Nevada de Santa Marta is the site of numerous glaciers located on a single peak at 10º34Ń 

and 73º43´W. 

182

The Sierra Nevada del Cocuy region of the Cordillera Oriental in the Columbian Andes is 

located in north-central Colombia, south of the Venezuelan border. The glaciated range trends 

north–south, and is located at 6º27′ N and 72º18′ W. 

The Ruiz-Tolima Massif is located in the Colombian Parque Nacional de los Nevados on the 

Cordillera Central chain of the Andes Mountains at 4º50Ń and 75º20´W. In 1959, the Parque 

Nacional de los Nevados contained 53 glaciers when the glaciers were measured by aerial 

photography (Hoyas-Patiño 1998). 

Work by Thouret et al. (1996) suggests that the glacier extent of the Ruiz-Tolima Massif, Sierra 

Nevada de Santa Marta, and Sierra Nevada del Cocuy of Colombia were approximately 1500 km 2 , 

1500 km 2 , and 2000 km 2 , respectively, during the last glacial maximum which occurred 

approximately 27,000–24,000 years BP. 

A later major glaciation occurred before 13,000–12,400 years BP at which time the glaciers of 

the Ruiz-Tolima Massif had an approximate area of 800 km 2 . During the late neoglacial period, 

during the Little Ice Age which occurred from the 1600’s to the 1900’s, the glacierized area was 

reduced to 100 km 2 . The Sierra Nevada de Santa Marta had an approximate glacier area of 850 

km 2 around 13,000–12,400 years BP, and 107 km 2 during the late neoglacial period. Glaciers in 

the Sierra Nevada del Cocuy showed similar retreat with an approximate glacier size of 1000 km 2 

around 13,000–12,400 years BP and 150 km 2 during the Little Ice Age. 

Similar to the Colombian glacier regions and other glaciers in the tropical and temperate zones, 

Venezuelan glaciers were affected by Quaternary glaciations. Between 20,000–13,000 years BP, 

glaciers in the area advanced. This period is denoted as the Mérida Glaciation. The Cordillera de 

Mérida branch of the northern Andes Mountains is the site to high peaks tall enough to support 

glacier conditions. At the end of the Merida Glaciation period, the Cordillera de Mérida chain had 

an approximate glacier area of 600 km 2 . Since 13,000 years BP, significant decreases in glacial 

size are apparent (Schubert and Clapperton 1990). 


Satellite Images 

ASTER and Landsat TM and ETM+ images from 1984 to 2004 were used in this study. Images, 

listed in Table 1, were selected based on minimum amounts of snow and cloud cover and on 

which produced the best time series for each region. All images available for each region were 

reviewed and critiqued based on the following criteria: seasonal precipitation, cloud cover, extent 

of snow cover, and image acquisition date. Selection of images was limited by seasonal patterns in 

precipitation and temperature, causing increased cloud cover in the area. Following visual analysis 

of the initially screened images, only those images with minimal snow and cloud cover were 

selected for further analysis. 

ASTER (Advanced Space-borne Thermal Emission and Reflection Radiometer) is an imaging 

instrument aboard NASA’s Terra satellite which was launched in 1999 as a part of the Earth 

Observing System (EOS). Each ASTER scene covers an approximate area of 60 km by 60 km 

with 15 spectral bands at 3 spatial resolutions. The visual near infrared (VNIR) bands, short-wave 

infrared (SWIR) bands, and thermal infrared (TIR) bands of ASTER have 15, 30, and 90 meter 

resolution, respectively (Abrams 2000). 

Each Landsat scene covers a larger area, approximately 185 km by 185 km. The Landsat 

Thematic Mapper (TM) on Landsat 4 and 5 has seven bands; six bands in the visible to short-wave 

infrared have a spatial resolution of 30 meters while the remaining thermal bands have 120 meter 

spatial resolution. The Enhanced Thematic Mapper Plus (ETM+), carried on the Landsat 7 

satellite, has eight multispectral bands with six bands at 30 meter spatial resolution, two bands 

with 60 meter resolution, and one panchromatic band at 15 meter resolution (Landsat Project 

Science Office 2006). 

183

Colombia 

Venezuela 

Table 1. Optimum images of the glaciers of Venezuela and Colombia 

Glacier Series Date Sensor Type Path Row 

Ruiz-Tolima 

Massif 

Sierra Nevada 

de Santa Marta 

Sierra Nevada 

del Cocuy 

Pico Bonpland 


(Siniguis) 

01/01/1959 * Aerial — — 

02/01/1976 * Landsat 1 9 57 

10/24/1997 Landsat 5 9 57 


01/01/1957 * Aerial — — 




12/20/2000 ASTER 8 53 

01/14/2004 ASTER 8 53 

01/01/1959 * Aerial — — 





01/14/2001 ASTER 7 56 

01/30/2001 ASTER 7 56 

03/06/2002 ASTER 7 56 

03/02/2003 ASTER 7 56 

01/01/1952 * Aerial — — 




02/11/2002 ASTER 6 54 

03/02/2003 ASTER 6 54 

02/01/2004 ASTER 6 54 

* Measurements taken from Hoyos-Patiño (1998) and Schubert (1998)I 

Image Preprocessing 

For ASTER scenes, the six SWIR bands were spatially oversampled to 15 meter resolution to 

match the VNIR bands using nearest neighbor resampling. For the Landsat images, all bands were 

oversampled from 28.5 meter to 15 meter resolution. Resizing all images to 15 meters allowed for 

better comparisons of snow and ice classification between the ASTER and Landsat platforms and 

facilitated visual interpretation of glacier area and comparisons of the classified glacier extent 

between platforms. 

The images for each glacier region were coregistered using ground control points (GCPs) 

spaced throughout each scene. A minimum of 9 GCPs were used with all images. Images for each 

glacier area were coregistered to a master image (Table 2) selected for each glacier series. All 

images were projected into Universal Transverse Mercator, Zone 19N with a WGS-84 datum. 

Once the images in a glacial series were coregistered, they were overlaid to visualize glacial 

retreat over time for each area. 

184

Colombia 

Venezuela 

Glacier 

Series 

Ruiz-Tolima 


Sierra 

Nevada de 

Santa Marta 

Sierra 

Nevada del 

Cocuy 

Pico 

Bonpland 


(Sinigüis) 

Table 2. Master Images for Image Coregistration 

Date 

185 

Sensor 

Type 

Landsat 

WRS Path 

Landsat 

WRS Row 


01/14/2004 ASTER 8 53 

03/06/2002 ASTER 7 56 

02/11/2002 ASTER 6 54 

Snow and Ice Classification 

Many image processing techniques could have been used in this study including manual 

digitization methods, band ratios and normalized difference ratios. Because this study focuses on 

digital area assessments, manual techniques were not considered. Further analysis of the band ratio 

method and normalized difference method were completed before selecting the best technique to 

use for all satellite images. 

The Normalized Difference Snow Index (NDSI) and Band Math Ratio were both considered and 

computed for all ASTER images for comparison. The NDSI (equation 1) is a commonly employed 

snow detection method that helps to differentiate between cloud cover and snow/ice (Hall et al. 

1995). The Band Math Ratio (equation 2) is a simple ratio between two bands, a and b 

( NIR − Red) 

( NIR + Red) 

NDSI = (1) 

Band a 

Band Math Ratio = (2) 

Band b 

Both methods were analyzed for accuracy using several ASTER band combinations: one and 

five, two and five, and three and five. After visually assessing both methods and all band 

combinations, the NDSI method using ASTER band two (0.63–0.69 µm) and five (2.145–2.185 

µm) was selected as best for determining glacial area for these study areas. The NDSI method was 

then applied to the Landsat scenes using the Landsat band combination, three (0.63–0.69 µm) and 

five (1.55–1.75 µm), that best approximated the ASTER bands. 

To determine the glacial area in each image using the NDSI ratio, a density slice was performed 

to classify pixels containing snow and ice. For both ASTER and Landsat, pixels with NDSI values 

ranging between 0.4 and 1.0 were classified as snow and ice. To compute the total glaciated area 

for each area, a region of interest (ROI) was drawn around glaciated area to determine the number 

of pixels with NDSI values in the selected range. This ROI approach helped eliminate clouds and 

other features misclassified as snow/ice. In some instances, transient snow cover still may be 

misinterpreted as glacier.

RESULTS 

Glacier changes between the 1950s and 2003 for the study areas in Columbia and Venezuela 

were determined using historical information and satellite images. In the 1950s, the total glacier 

area for three study regions in Colombia was 89.33 km 2 . By 2003 the total glacier area had been 

reduced to 45.77 km 2 . From the 1950s to 2003 the calculated total ice loss in Colombia was 43.56 

km 2 . The Sierra Nevada del Cocuy contributed 52% of the total ice lost, the Ruiz-Tolima Massif 

contributed 42% of this loss and the Sierra Nevada de Santa Marta contributed 6%. These ice loss 

percentages are directly related to the total glacier area of each region. 

Out of the 10 Venezuela glaciers mapped in 1952 with a total area of 2.91 km 2 (Schubert 1998), 

only one glacier is still visible from ASTER and Landsat images as of 1985. In 2004 the last 

remaining glacier on the Pico Bonpland Massif-Sinigüis, decreased from 2.03 km 2 in 1952 to 0.29 

km 2 or 86% of its 1952 area. Unfortunately, several images analyzed of this glaciated region had 

to be eliminated due to seasonal snow cover in the area. 

Colombia 

Table 3. Glacier Areas 

Glacier Series Date Area (km 2 ) 

Ruiz-Tolima 


1959 Historic * 

10/24/1997 

01/28/2001 

33.95 

17.47 

15.81 

Sierra Nevada 

de Santa Marta 



12/20/2000 

16.26 

14.10 

13.66 

1959 Historic * 39.12 

Sierra Nevada 

del Cocuy 


05/25/1999 

03/06/2002 

28.00 

20.39 

16.30 

03/02/2003 16.30 

1952 Historic * 2.03 

Pico Bonpland 


Sinigüis 

03/24/1985 

01/20/1988 

01/29/2000 

02/11/2002 

0.66 

0.58 

0.38 

0.33 

03/02/2003 0.29 

* Measurements taken from Hoyos-Patiño (1998) and Schubert (1998) 

Venezuela 

Ruiz-Tolima Massif 

In 1959, five peaks with glaciers existed in the Parque Nacional de los Nevados, but as of 1976 

only three glaciated peaks are still visible, Ruiz, Tolima, and Santa Isabel (Hoyas-Patiño, 1998). In 

1959, glacier area on the Ruiz-Tolima Massif was 33.95 km 2 and was reduced to 15.81 km 2 by 

2001 (Figure 2). This represents a 53% loss in ice over a 42-year time span. Individually, glaciers 

on the Nevado del Ruiz decreased from 21.4 km 2 to 10.92 km 2 , a total loss of 49%. The Nevado 

del Santa Isabel glaciers decreased from 9.78 km 2 to 3.61 km 2 , losing 63% of their area. The 

Nevado del Tolima glaciers declined from 2.22 km 2 in 1959 to 1.26 km 2 in 2001, a total ice loss of 

43% (Tables 3 and 4). 

186

Table 4. Glacier Areas of the Ruiz-Tolima Massif 

Glacier Date Area (km 2 ) 

Nevado de 1959 Historic * 9.78 

Santa Isabel 10/24/1997 4.19 

01/28/2001 3.61 

Nevado del Ruiz 1959 Historic * 21.40 

10/24/1997 12.13 

01/28/2001 10.92 

Nevado del 1959 Historic * 2.22 

Tolima 10/24/1997 1.15 

01/28/2001 1.26 

*Measurements taken from Hoyos-Patiño(1998) 

Figure 2. Time Series for Ruiz-Tolima Massif Glaciers 

Sierra Nevada de Santa Marta 

The Sierra Nevada de Santa Marta was the second Colombian Glacier region studied (10º34′ N 

and 73º43′ W). As determined from aerial photography, the region had 88 glaciers in 1957 with a 

total glacial area of 16.26 km 2 (Hoyas-Patiño 1998). Based on the 2000 ASTER image the total 

glaciated area was 13.66 km 2 . Using the 2000 ASTER image, the most recent available snow free 

image, the total ice loss for the Sierra Nevada de Santa Marta is 2.86 km 2 or 16% (Table 3 and 

Figure 3). However, the 2000 estimate of glacier area still may be impacted by some transient 

snow cover. 

187

Figure 3. Time Series for Sierra Nevada de Santa Marta 

Sierra Nevada del Cocuy 

The Sierra Nevada del Cocuy is also a glaciated area in Colombia (6º27′ N and 72º18′ W). The 

total glaciated area in 1959 was 39.12 km 2 (Hoyas-Patiño 1998) and had retreated to 16.3 km 2 in 

2003. In 44 years the area had a total ice loss of 58% (Table 3 and Figure 4). 

Figure 4. Time Series for Sierra Nevada del Cocuy Glaciers 

188

Pico Bonpland Massif–Sinigüis Glacier 

The Pico Bonpland Massif glacier region is located in Venezuela at 8º32′ N and 71º00′ W. In 

1952 two glaciers, Siniguis and Nuestra Senora, existed on Pico Bonpland/Humbolt. In a Landsat 

5 image acquired for the Pico Bonpland Massif, on March, 24, 1985, the only visible glacier was 

the Sinigüis glacier. The total glacier area assessment in 1952 was 2.03 km 2 and by 2003 the 

glacier area had decreased to 0.29 km 2 . This is a total area loss of 1.77 km 2 or 87% of the 1952 

area (Table 3 and Figure 5). 

DISCUSSION 

Figure 5. Time Series for Pico Bonpland Massif – Sinigüis Glacier 

Aerial photography and Landsat Multispectral Scanner (MSS) images provide historical 

measurements of the Colombian and Venezuelan glaciers. Although these measurements help to 

expand the time span used in this study, it is unknown exactly how the areas were calculated. In 

this study, the Normalized Difference Snow Index (NDSI) provided a strong foundation for 

estimating the areas of the glacier regions by discriminating the snow/ice in the images through 

the use of a 0.4 threshold. By providing a uniform criterion to delineate snow and ice, it is possible 

to construct a glacier retreat time series. Examining the trend for each glacier area, it is evident 

that each region has experienced a decrease in glacier area over the past fifty years. 

The glaciated region of the Sierra Nevada de Santa Marta illustrates the commission errors 

caused by increases in seasonal snow cover as several images could not be used due to transient 

snowfall obscuring glacier boundaries. Nevertheless, the overall trend on this glacier series 

confirms a general decrease in glacier area from 1957 through 2004. 

In the case of the glacier area Ruiz-Tolima Massif, volcanic activity in the area is a partial cause 

for the decline in glacial area. As part of a series of stratovolcanoes, extending along the Andes 

from the southern tip of Chile to the northern portion of Venezuela, Nevado del Ruiz is currently 

in an active eruptive state. Recent eruptions in the mid-1980s affected temperatures in the glacier 

area enough to trigger snow and ice melt, which is supported by the decline in areal extent of the 

Ruiz-Tolima Massif glaciers between the mid-1980’s and 2001 (Linder and Jordan 1991, Linder et 

al. 1994). 

On November 13, 1985, an eruption on Nevado del Ruiz caused South America’s deadliest 

eruption, a result of devastating lahars. The last known eruption of Nevado Del Ruiz occurred in 

189

1991. Unlike Ruiz, Nevado del Tolima and Nevado de Santa Isabel, the other stratovolcanoes in 

the Ruiz-Tolima Massif, are not currently in an eruptive state at this time (Hoyas-Patiño 1998). 

Figure 4 illustrates the time series of area measurements, and shows a general decline in glacier 

size from 1959 to 2001. 

The series of images from the glaciers of Sierra Nevada del Cocuy make up the largest time 

series in this study, 9 images over 44 years. Originally consisting of numerous distinct glaciers, 

Sierra Nevada del Cocuy has continually shown a significant loss in glacier size over the study 

period, resulting in steep general decrease in glacial area. The observed retreat of the Sierra 

Nevada del Cocuy glaciers confirm notions of tropical glacier loss in this region. 

Finally, the glaciers of Venezuela are also disappearing. The sole remaining glacier, Sinigüis, 

showed a steady reduction in size between 1988 and 2003 after a dramatic decrease in size from 

1952 to 1985. The computed loss rate from 1952 to 1985 and from 1988 to 2003 is 41515 m 2 /year 

and 19333 m 2 /year, respectively. Overall, a general decrease in glacier size is shown in Figure 5 

during the 51-year period. 

CONCLUSION 

Overall a decline in glacial area can be seen throughout all four studied glacier regions of the 

tropical Colombian and Venezuelan Andes. Steady to extreme decreases in area are captured by 

the Normalized Difference Snow Index (NDSI) calculations completed in this study. From the 

1950’s to present, significant snow and ice loss over the area portray proof of a regional glacier 

recession. Broadening the scale, glaciers throughout the tropics are showing similar retreating 

characteristics (Kaser 1999, Kincaid and Klein 2004), and globally, glaciers around the world, 

from different regions, climates, and human interferences, confirm the results of this survey (IPCC 

2001). 


This work was funded by NASA Young Investigator Program Grant 02-000-0115. This project 

was completed as an undergraduate directed studies course in the Department of Geography at 

Texas A&M University. 

REFERENCES 

Abrams, M. 2000. The Advanced Spaceborne Thermal Emission and Reflection Radiometer 

(ASTER): data products for the high spatial resolution imager on NASA’s Terra platform. 

International Journal of Remote Sensing 21: 847–849. 

Hall DK, Riggs GA, Salomonson VV. 1995. Development of methods for mapping global snow 

cover using moderate resolution imaging spectroradiometer data. Remote Sensing of 


Hoyos-Patiño F. 1998. Glaciers of Columbia. In Williams RS Jr. and Ferrigno JG (eds) Satellite 

Image Atlas of Glaciers of the World – South America. USGS Professional Paper 1386-I. 

United States Government Printing Office: Washington, DC: 11–30. 

IPCC, 2001: Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the 

Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge 

University Press: Cambridge; 881pp. 

Kaser G. 1999. A review of the modern fluctuations of tropical glaciers. Global and Planetary 

Change 22: 93–103. 

Kincaid JL, Klein AG. 2004. Retreat of the Irian Jaya glaciers from 2000 to 2002 as measured 

from IKONIS Satellite Images. In Proceedings of the 61 st Eastern Snow Conference held June 

9–11, Portland Maine, USA, 147–157. 

Landsat Project Science Office 2006. The Landsat-7 Science Data User's Handbook. 

http://landsathandbook.gsfc.nasa.gov/handbook.html 

190

Linder W, Jordan E, 1991. Ice-mass losses at the Nevado del Ruiz, Columbia, under the effect of 

the volcanic eruption of 1985; a study based on digital elevation models. Santa Fé de Bogotá 

Revista Cartográfica 59: 479–484. 

Linder W, Jordan E, Christke K. 1994. Post-eruptive ice losses on the Nevado del Ruiz, Columbia. 

Zentralblatt für Geologie und Paläontologie 1: 479–484. 

Schubert C. 1998. Glaciers of Venezuela. In Williams RS Jr. and Ferrigno JG (eds) Satellite Image 

Atlas of Glaciers of the World – South America. USGS Professional Paper 1386-I. United 

States Government Printing Office: Washington, DC: 1–10. 

Schubert C, Clapperton C. 1990. Quaternary Glaciations of the Northern Andes (Venezuela, 

Colombia and Ecuador). Quaternary Science Reviews 9: 123–135. 

Thouret JC, Van der Hammen T, Salomons B, Juvigné E. 1996, Palaeoenvironmental changes and 

glacial stades of the last 50,000 years in the Cordillera Central, Colombia. Quaternary 

Research 46: 1–18. 

191


192

Snowpack Processes 

193


194



Snow Cover Patterns and Evolution at Basin Scale: GEOtop Model 

Simulations and Remote Sensing Observations 

STEFANO ENDRIZZI, 1 GIACOMO BERTOLDI, 2 MARKUS NETELER 3 AND RICCARDO 

RIGON 1 

ABSTRACT: Remote sensing data can provide images of snow covered areas and, therefore, 

it is possible to follow the time evolution of snow melting spatial patterns with increasing 

spatial and temporal resolution. Snow cover patterns are dominated by the complex 

interplay of topography, radiation forcings and atmospheric turbulent transfer processes. 

The snow cover evolution in an alpine basin in Trentino (Italy) is here studied, comparing 

the simulations of the distributed hydrological model GEOtop with remotely sensed data. 

GEOtop describes the soil-snow-atmosphere energy and mass exchanges, taking into account 

the snow physics and the topographic effects of elevation, slope and aspect on solar radiation 

and air temperature. the Snow cover extent has been provided by MODIS 8-day composite 

maps with a resolution of 500 metres. The model reproduces the physical features of snow 

melting reasonably and shows a fair agreement with the data. The relative importance of 

precipitation, solar radiation, and temperature to control the snow accumulation and 

melting processes is also investigated. 

Keywords: snow water equivalent, distributed modelling, remote sensing, topography 

INTRODUCTION 

To have information about the evolution of the snow cover extent and the snow water equivalent 

is very useful for the water resource management, and in relation to climate change. So far 

temperature index models, like SRM (Martinec and Rango, 1986), based on a statistical relation 

between snow melting and temperature, have been widely used. Although many of these models 

have evolved so that topographic effects could be considered (Cazorzi and Dalla Fontana, 1995), 

distributed physically based models, like ISNOBAL (Marks et al., 1999) can provide more 

detailed information about snow physics and improved predictions of snow cover. Their 

application also allows to study what phenomena may be important in snow accumulation and 

melting, and what may be the sensitivity to changes of atmospheric forcing. 

At the same time, remote sensing data are available with improved spatial and temporal 

resolution (for example MODIS (Hall et al., 2002) and NOHRSC (Hall et al., 2000)). Several 

models, using remote sensing data as input (Turpin et al., 1999) or to perform data assimilation 

(Carroll et al., 2006), have been recently developed. However, in mountain basins, remote sensing 

data can be affected by some errors, and consistent physical modelling maintains great 

1 Department of Civil and Environmental Engineering, University of Trento, Via Mesiano 77, 

38050 Povo (TN), Italy, email: stefano.endrizzi@ing.unitn.it 

2 Department of Civil and Environmental Engineering, Duke University, Durham, NC, USA 

3 ITC-IRST, Center for Scientific and Technological Research, 38050 Povo (TN), Italy 

195

importance, because snow cover patterns are dominated by the complex interplay of topography, 

radiation forcing and atmospheric turbulent transfer processes (Pomeroy et al., 2003). 

The GEOtop model (Rigon et al., 2006) is a distributed hydrological model that solves the 

energy and water balance on a landscape whose topographical surface is described by a digital 

elevation model. The model has been conceived to be applied to mountain basins characterized by 

complex topography, where snow accumulation and melting have to be accounted. GEOtop 

includes a snow module, the first version (version 0.875) of which was the object of a previous 

work (Zanotti et al., 2004), where its capability to predict the snow water equivalent (SWE) 

evolution in a point was tested, and the results were compared with local measurements. In Zanotti 

et al. (2004) an application on a small mountain basin with a surface area of a few square 

kilometres was then presented, but the results could not be checked in more than one measurement 

point. Since then, the snow cover module in GEOtop has been improved (version 0.9375) such 

that a multilayer representation has been implemented, similar to the one of the CLM land surface 

model (Oleson et al., 2004). The GEOtop snow module is quite similar to ISNOBAL (Marks et al., 

1999), as it solves the snow energy balance, but it is part of a complete hydrological model that 

considers the whole soil-snow system. The next step is to predict the snow cover evolution and its 

variability at the distributed scale. 

Some works have been published about the temporal distribution of snow cover, for example 

Alfnes et al. (2004) using statistical tools. In order to do this, a distributed field of the 

meteorological forcing as input data is needed, or, if only a few measurement stations are available 

throughout the basin, we need to find some criteria to spatially extrapolate the measurements, 

although they might lead to some errors. 

An application in an Alpine basin with surface area of about 250 square kilometres is shown 

here. The snow cover extension area is compared with corresponding maps provided by remote 

sensing techniques. In particular, MODIS maps (Hall et al., 2002; Riggs et al., 2003) have been 

used because the Aqua and Terra satellites overpass locations daily. Snow products are delivered 

at a spatial resolution of 500 metres, which is sufficient to keep the signature of the topographic 

features in an Alpine environment (Cline et al., 1998) where the snow cover spatial distribution is 

strongly dependent on aspect and elevation. In addition, the MODIS data are easily available for 

applicative uses. 

Snow cover maps provided by remote sensing have already been used, mainly as an auxiliary to 

the models (Lee et al., 2005; Turpin et al., 1999), but here they are used only to check the model 

results. The problem of the initial SWE distribution does not exist as the simulation starts from a 

late summer initial condition, when in the considered basin no snow is present. 

Our aim in this paper is to test the capability of the model to reproduce a realistic snow cover 

evolution during a whole winter season and to investigate the relative importance of precipitation, 

solar radiation, and temperature to control the snow accumulation and melting processes in an 

Alpine environment, with reference to its time evolution and its dependence on elevation and 

aspect. We want to develop a modelling framework as physically based as possible, though 

parsimonious in its input data requirements and, therefore, easily applicable for operational use 

(Carroll et al, 2006). 

THE MODEL 

The GEOtop model has been fully described in Rigon et al. (2006) and in Bertoldi et al. (2006). 

This model has been conceived to be an integration of a rainfall-runoff model and a land surface 

model as it solves the three-dimensional soil water budget equation together with the onedimensional 

energy budget equation, so that it can calculate the spatial distribution of many hydrometeorological 

variables (such as soil moisture, surface temperature, convective and radiative 

fluxes) without losing the main purpose of hydrological models, that is predicting the water 

discharge at a specific closure section. However, this paper is focused on the simulation of the 

spatial patterns and of the time evolution of the snow covered area and of the SWE at basin scale. 

196

In the following paragraph the main improvements of the snow module with reference to the 

version presented in Rigon et al. (2006) are briefly discussed. 

In the version 0.9375 of the snow module in GEOtop, the snow cover is described with a 

multilayer scheme. For each snow layer the liquid and solid water budget equations and the energy 

budget equation are solved together with the continuity equation and a formula linking the phase 

change with the temperature. So five equations for the following five unknowns are available: 

liquid and ice contents, porosity, snow temperature, and amount of water changing phase. 

The water budget equations, for liquid water and ice, are expressed as follows: 

! 

! 

" w = 1 

# w 

" i = 1 

# i 

$W 

$t + $Q % ( w 

' * 

& $z ) 

$W 

$t + $Q % ( i 

' * 

& $z ) 

where θw and θi are the nondimensional liquid water and ice content, t is time, z the vertical 

coordinate (positive upwards), ρw and ρI the density of liquid water and ice, respectively, Qw and 

Qi the vertical flux of liquid water and ice (the latter is equal to 0 except at the snow-atmosphere 

interface where it becomes the snow precipitation), and W is the amount of water changing phase 

(positive in the case of melting, negative in the case of freezing). As the liquid water flow inside 

the snow cover is mainly due to gravity, the equations are solved only in the vertical direction, 

neglecting the lateral flow inside the snowpack, but not in the underlying soil. Qi depends on the 

liquid water fraction in the snow exceeding the capillary retention as in Colbeck and Anderson 

(1982). 

The energy budget equation is the following: 

! 

C "T 

"t + L f 

"W 

"T 

( ) 

" # "T & 

= % k ( + 

"z $ "z ' 

" QwU w 

"z 

where T is the temperature, Lf the latent heat of fusion, Uw the specific energy content of the liquid 

water, and k and C the thermal conductivity and the thermal capacity, respectively, calculated as 

averages of the values of the different phases present. The equation (3) is solved with the 

following boundary condition at the snow-atmosphere interface, where the exchange fluxes occur: 

! 

# 

% 

$ 

k "T 

"z 

(1) 

(2) 

(3) 

& 

( = )Rn + H + L (4) 

' 

where Rn is the net radiation, H the sensible heat flux and L the latent heat flux (both positive 

upwards). The treatment of radiation is essentially the same as in Zanotti et al. (2004). The effects 

of shadowing, slope and aspect, which are extremely important in a complex topography, are 

taken into account for direct shortwave radiation, and the sky view factor is considered for the 

diffuse shortwave radiation and for the longwave radiation. Shortwave radiation is parameterized 

as in Iqbal (1983). Longwave radiation is calculated using Stefan-Boltzmann formula. Whereas 

the outcoming longwave radiation is clearly dependent on the surface temperature, there are some 

uncertainties in calculating the incoming longwave radiation, as the latter depends on the entire 

profile of the air temperature above the snowpack. Different experimental formulae, as those of 

Brutsaert (1975), Satterlund (1979), and Idso (1981), that parameterize this component of 

radiation in function of the skin temperature, have been used. The convective heat fluxes are 

calculated after Monin-Obukhov similarity theory, using the stability function of Businger et al. 

(1981). As they depend also on surface temperature, which is unknown as long as the energy 

budget is not solved, some iterations are needed. 

The continuity equation correlates the different phases in the snowpack: 

! 

" w + " i + " v =1 (5) 

197

! 

where θv is the nondimensional gas and vapour content, that is the porosity. 

Finally, we need to find a relation between the snow temperature and the phase change. We 

have chosen that the phase change should occur when the temperature in each snow layer is equal 

to 273.15 K: neither the temperature is allowed to increase as long as ice is present nor can it 

decrease until liquid water is entirely frozen. Therefore, 

W " 0 if T = 273.15K ; W = 0 if T " 273.15K (6) 

Thus the system of equations (1)-(2)-(3)-(5)-(6) is solved with a finite difference method with 

respect to the unknowns T, W, θw, θi, θv. 

The snow layers have a varying thickness in accordance with the snow precipitation and snow 

metamorphism (Anderson, 1976). The algorithms to combine and subdivide snow layers have 

been taken from Oleson et al. (2004), which makes it possible to describe the density changes of 

the snow pack. 

APPLICATION 

The model has been applied to the upper part of Brenta Basin (Trentino, Italy), in the Eastern 

Italian Alps, a basin with a surface area of about 250 square kilometres. Its elevation ranges from 

340 m a.s.l. to 2303 m a.s.l., and the land cover is very variable. The bottom of the valley is 

urbanized, and many small villages are present, although there is still a wide countryside area. 

Two lakes, with total area of about 6 square kilometres, are also present. They play an important 

role in the regulation of the discharge, but, as here our main interest is to follow the snow cover 

evolution, they are neglected. As elevation increases, woodland prevails, even if there are some 

small resort villages and wide grassland and pasture areas. The vegetation limit ranges from 1900 

m a.s.l. to 2000 m a.s.l.. The snow cover has been simulated for the season 2003-2004, namely 

from 16 th October 2003 to 16 th June 2004, using a model grid resolution of 250 m. 

METEOROLOGICAL DATA 

For the time considered, hourly precipitation measurement data were available for 15 stations 

located in the basin or in the nearby areas. Precipitation was then distributed according to kriging 

techniques, and it was split into liquid and solid form in conformity with the air temperature, 

defining a threshold air temperature TL above which the precipitation is always liquid and another 

threshold temperature TS below which the precipitation is always solid, as in U.S. Army Corps of 

Engineers (1956). These threshold temperatures are considered tuneable and adjustable 

parameters. In fact, as discussed later, in order to catch the snow events with precision, it results 

very important to distribute the air temperature in the basin accurately. 

Hourly air temperature observations performed in 17 meteorological stations in and around the 

basin were used to find a linear dependence of the temperature on the elevation, which is 

considered variable in time, but valid all over the basin, and an hourly value of the lapse rate was 

found. 

Figure 1 shows the relation between the air temperature measurements averaged on the whole 

season and the station elevations. The temperature variation presents a clear linear trend with a 

lapse rate of 3.9 °C/km (regression constant equal to 0.94), which is lower than the lapse rate 

normally used (6.5 °C/km). This means that at the annual scale the temperature measurements do 

not exhibit evident and regular variability with the horizontal coordinates. However, some stations 

located approximately at the same elevation can record temperature differences of some degrees 

(at the most 2 °C) due to local factors (for example shadowing) and to particular meteorological 

events. This could be a measure of the error in the temperature estimation. 

198

Figure 1: Variation of the season average of the air temperature measurements with the elevations of the 

meteorological stations (dots), and regression line. 

Figure 2: Time evolution of the hourly and daily average lapse rate inferred by the measurements. 

But if the time evolution of the lapse rate, daily and hourly averaged (Figure 2) is plotted, it can 

be noticed that during the winter there are frequent periods of thermal inversion, with a nearly 

isotherm atmosphere, but during the spring time the lapse rate is more clearly defined and reaches 

values around the adiabatic lapse rate. The lapse rate also shows a typical daily cycle, with a 

marked nocturnal inversion in winter. The definition of a correct value of the lapse rate is critical 

to capture the correct snow-rain limit. Therefore, we have chosen to use an hourly varying lapse 

rate as input data for the model. 

199

The relative humidity and the wind velocity are kept constant all over the basin. 

PARAMETERS OF THE MODEL 

The parameters of the model were guessed in order to obtain the best agreement with the remote 

sensing data. By all means, only a few parameters are important for the snow cover description. 

They are the temperature roughness length, the wind velocity roughness length, the capillary water 

amount retained by the snow, and the threshold temperatures aforesaid. The following table 

reports the value of these parameters with the reference. 

Table 1. Main parameters of the snow module in GEOtop 

Value Parameter Description Reference 

0.05 z0T Temperature roughness length [m] Calibration 

0.5 z0V Wind velocity roughness length [m] Calibration 

0.04 Sr Liquid water in snow retained by capillarity 

forces, as fraction of porosity [-] 

Jordan (1991) 

2.0 TL Temperature above which all the precipitation is 

rain [°C] 

Calibration 

0.5 TS Temperature below which all the precipitation is 

snow [°C] 

Calibration 

MODIS 

The Moderate Resolution Imaging Spectroradiometer (MODIS) is a 36-channel sensor 

measuring the spectral range from visible to thermal-infrared. It was launched as part of the 

payload of the Terra (12/1999) and Aqua (5/2002) satellites. Among other products, daily and 8day 

composite snow related maps are produced at 500 m and 1000 m resolution. The maps created 

over land include daily snow albedo and 8-day maximum snow extent. For the latter multiple days 

of observations for a cell are examined. If snow cover is found for at least one day the cell is 

indicated as snow. Other values are lake ice, cloud, ocean, inland water and land. If no snow is 

found in the cell, the most frequent value of the other classes is assigned. Due to the 8-day 

compositing technique, the impact of clouds is minimized (Riggs et al., 2003). 

In this study we have used the MOD10A2 8-day composite maximum snow extent data at level 

V004. The data were processed in GRASS GIS (Neteler, 2005). 

However, MODIS snow cover maps can be affected by some errors, the most common of which 

are the errors of commission. This means mapping cells as snow covered when the occurrence of 

snow is extremely unlikely. These errors are usually associated with the least favorable conditions 

of illumination and cloud type for optimal snow mapping (Riggs and Hall, 2006). 

COMPARISON BETWEEN THE RESULTS OF THE MODEL AND MODIS DATA 

Snow covered area spatial distribution 

In this paragraph the MODIS 8-day maximum snow extent maps and the corresponding 8-day 

GEOtop model output maps are compared. The snow free area is reported in green, and the snow 

covered area is in white. Whereas MODIS can only provide information about the presence of 

snow, the model gives results about the amount of SWE, which is reported in the pictures in 

white-blue colour scale. Five pairs of maps have been chosen as representative of the main 

qualities and problems of the models. 

200

Figure 3: 8-day composite maps (24 th October 2003) for snow cover obtained by MODIS (left) and by the 

model (right). The snow free area is reported in green, the snow covered area is in white in MODIS map, in 

white-blue colour scale (legend on the right) in the model map. 

Figure 4: 8-day composite maps (17 th November 2003) for snow cover obtained by MODIS (left) and by the 



The first pair of pictures (Figure 3) regards the autumn season: snow cover is present only at the 

highest elevations, and the model predicts reasonably the snow limit. However, in the model 

output map the snow limit strictly follows the elevation contour lines, because snow precipitation 

mainly depends on the air temperature, which, in turn, depends only on elevation. In the MODIS 

map the snow limit is close to the elevation contour, but there are some deviations probably due to 

local variations of air temperature, or, simply, to the coarser resolution of the MODIS maps (500 

m grid cell) than in the model maps (250 m grid cell). 

201

Figure 5: 8-day composite maps (17 th January 2004) for snow cover obtained by MODIS (left) and by the 


white-blue colour scale (legend on the right) in the model map. The pixels for which it was not possible to 

detect snow due to cloud cover are in yellow. 

Figure 6: 8-day composite maps (29 h March 2004) for snow cover obtained by MODIS (left) and by the 



The second pair (Figure 4) is an example of the late autumn season, after several days without 

solid precipitation. The snow patterns are controlled here also by slope and aspect, and there are 

some white spots inside the snow free area in the pixels receiving a less amount of radiation. The 

agreement is fairly good, although not all the white spots agree in detail, but it has to be taken into 

account that MODIS maps may be affected by some commission errors (for example the white 

spot located near the lake in the bottom of the valley is very unlikely). 

202

The third pair (Figure 5) represents the snow cover after a snowfall in winter. It is difficult for 

the model to reproduce the snow in the bottom of the valley, as the air temperature field can be 

uneven and snow precipitation occurs usually at low elevations when the air temperature is close 

or even slightly above 0 °C. For this reason the threshold temperature above which the 

precipitation is always liquid and the threshold temperature below which the precipitation is 

always solid have been fixed to 2 °C and 0.5 °C, respectively, after some calibrations. These values 

are a little different from the values proposed in US Army Corps of Engineers (1956), respectively 

3 °C and -1 °C. The areas located at the bottom valley are quite large, but they are usually covered 

by a small amount of snow, which melts quickly. 

The fourth pair (Figure 6) is relative to early springtime. The snow patterns are dominated by 

elevation, as a consequence of late snowfall events, as well as by slope and aspect, as it is possible 

to appreciate the differential melting in north facing and south spacing slopes, even though in the 

two maps the snow limit can be slightly different. In this time SWE reaches its maximum value at 

the highest elevations. 

Figure 7: 8-day composite maps (16 h May 2004) for snow cover obtained by MODIS (left) and by the model 

(right). The snow free area is reported in green, the snow covered area is in white in MODIS map, in whiteblue 

colour scale (legend on the right) in the model map. 

Finally, the fifth pair (Figure 7) refers to late springtime: snow cover is present only at the 

highest elevations, and the snow patterns in the two maps are similar. However, the model result 

map shows a little faster melting on the south facing slopes than the MODIS map, which, on the 

other hand, may also be affected by commission errors, because it shows a snow limit lower in the 

south facing slopes than in the north facing slopes. 

In conclusion, the model seems to catch quite well the main patterns of the snow cover extent, 

which are dominated firstly by elevation, and secondly by slope and aspect. In order to reproduce 

the snow cover in the best way, it seems crucial to catch the snowfall elevation limits, and to 

calculate accurately the radiation fluxes and the radiation topographic controls, which play a 

fundamental role in melting. 

Snow covered area as a fraction of total area 

The model simulates SWE as a continuous variable and, therefore, a SWE threshold value 

should be chosen to consider the pixel as entirely covered by snow. Figure 8 reports the evolution 

in time of the snow covered surface as fraction of the total basin area, using threshold values of 1 

203

mm or 10 mm. This has been done to draw a better comparison between the model and the remote 

sensing data, as MODIS cannot detect snow if SWE is very small. Moreover, the MODIS data are 

sometimes affected by uncertainty due to the persistence of cloud cover, and the related range of 

variability in the snow cover area is shown in the figure. 

The agreement is good, in particular in the melting season. However, the limitations of the 

MODIS 8 day product have to be considered. Large uncertainties during the precipitation periods 

due to the cloud cover can occur, and the product has a resolution which does not always allow to 

appreciate the detailed elevation patterns of snow cover, as the mountain basin considered is 

characterized by strong elevation gradients. 

The model underestimates the snow cover extent in the early snow events, probably because it 

does not predict well the snowfall in the bottom of the valley, which covers quite a wide area. As 

it has been said in the previous paragraph, it is difficult to distinguish the rain and snow events, 

because they are not only dominated by the air temperature at the ground, but by the entire 

temperature profile. The MODIS curve agrees more with the 1 mm threshold model result curve: 

this probably means that a thin layer of fresh snow in the bottom of the valley, where vegetation is 

not dense, is easily recognized by remote sensing techniques. 

The model snow cover extent is much less sensitive to the threshold value during the spring, 

when the averaged SWE distribution is higher and more strongly dependent on elevation, as 

discussed later. 

Figure 8: Fraction of the total basin area covered by at least 1 mm (grey continuous line) and 10 mm (black 

continuous line) of SWE, according to the model results, and snow covered area according to MODIS 

(dashed black lines) with uncertainty range due to cloudiness against time. 

DEPENDENCE OF SNOW COVER ON ELEVATION AND ASPECT 

In this paragraph the model results are used to find some relations between SWE, elevation and 

aspect for the basin considered. The basin is here divided into 10 elevation ranks, each spanning 

200 metres. SWE is then averaged in each rank. Figure 9 shows the time evolution of SWE for 4 

different elevation ranks, and it can be noticed that its maximum value shifts later into spring as 

elevation gets higher. At the bottom of the valley a strong snowfall occurred at the beginning of 

March, but at higher elevations other strong snowfalls occurred later. 

204

Figure 9: Time evolution of SWE averaged in 4 elevation ranks 

The slope of the SWE profile with respect to elevation changes after every snowfall and with 

the season. Here some examples after 3 different snowfall events - in autumn, in winter, and in 

early springtime - are shown in Figures 10, 11, and 12. 

Figure 10: Daily averaged SWE-elevation chart for 4 days after a strong autumn snowfall (occurred on 8 th 

November 2003) 

205

Figure 11: Daily averaged SWE-elevation chart for 4 days after a strong winter snowfall (occurred on 30 th 

December 2003) 

Figure 12: Daily averaged SWE-elevation chart for 5 days after a strong spring snowfall (occurred on 11 th 

March 2004) 

For the autumn snowfall (Figure 10), at the lower elevations SWE exhibits little dependence on 

elevation. In fact, snowfall has more or less the same intensity at the elevations higher than the 

snow limit (ranging from 700 m to 1000 m a.s.l.), but at the lower elevations snow was not 

formerly present at the ground. On the contrary, at the higher elevations other snowfalls had 

already occurred earlier with different snow limits, and this results in a linear relation. Some days 

after the snowfall, an increase in temperature causes rainfall events at the lower elevations, where 

206

melting occurs, whereas at the higher elevations other snowfall events still occur, and the curve 

becomes steeper. For the winter snowfall (Figure 11), the relation is almost linear for all the 

elevations, as many other snowfall events had formerly occurred, and there is no melting. 

Concerning the early spring snowfall (Figure 12), just after the snowfall the relation is linear but 

after some days differential melting occurs at the lower elevations, this causes a steepening of the 

curve and an increase in its concavity, whereas at the higher elevations there is still snow 

accumulation. 

In conclusion, the relation between SWE and elevation is more or less linear, but in autumn 

SWE is weekly dependent on elevation, whereas the dependence is very strong in spring, due to 

differential melting and to the succession of the snowfall events. 

Figure 13: Time evolution of the ratio between SWE averaged only in north facing pixels and SWE 

averaged, for 2 elevation ranks 

In Figure 13 the dependence of SWE on aspect is explored. The ratio between the SWE 

averaged in the north facing pixels and the SWE averaged in all pixels for each of the elevation 

ranks can be an index to evaluate the importance of the effects exerted by radiation and aspect. We 

consider as north facing pixels those characterized by aspect ranging from -20° to +20°, being 0° 

the north aspect. Figure 13 shows the time evolution of this ratio for 2 different elevation ranks, 

one characteristic of the bottom of the valley and the other of the higher elevations. When the ratio 

is close to 1 no significant relation of SWE with aspect is present. When differential melting 

between N and S facing slopes occurs, the ratio increases to very high values. It can be noticed 

that in the bottom of the valley differential melting is always present all through the winter, except 

immediately after snowfalls. At the higher elevations, on the contrary, melting is aspect dominated 

only in autumn and in spring. 

207

CONCLUSION 

The comparison of the GEOtop SWE results with MODIS snow extent maps shows a fairly 

good agreement. The problems of the model are mainly due to uncertainties about the distribution 

of the meteorological forcing in space and time, as only measurements in some points are 

available. Only total precipitation is measured, and a criterion based on two thresholds on the air 

temperature has been used to distinguish solid and liquid precipitation. However, threshold values 

a little higher than the values found in literature were needed to reproduce the snow cover at the 

bottom of the valley. An hourly varying lapse rate was inferred by the temperature measurements, 

and was used as input data for the model. 

Though characterized by some commission errors, MODIS maps can be valuable tools to 

validate distributed models. However, images of finer resolution and more frequent overpassing 

time would be better tools. 

With the results of the model we investigated the relationship between SWE and elevation and 

aspect in different seasons. It has been shown that the dependence of SWE on elevation is more or 

less linear, but tends to be weaker in autumn and stronger in spring. Moreover, differential melting 

in north and south facing slopes occurs all through the winter at lower elevation (approximately 

below 800 m a.s.l.), whereas it is present only in autumn and in spring at the higher elevations. 

However, these relations should be verified in other applications to different and wider basins. 

ACKNOWLEDGEMENT 

We thank prof. John Albertson who supported the visit of the first author at the Department of 

Civil and Environmental Engineering of Duke University, and Joseph Tomasi who corrected a 

first version of the manuscript. 

REFERENCES 

Alfnes E, Andreassen LM, Engeset RV, Skaugen T, Udnaes HC. 2004. Temporal variability in 

snow distribution. Annals of Glaciology 38: 101-105. 

Anderson EA. 1976. A point energy and mass balance model of a snow cover. Office of 

Hydrology National Weather Service. 

Bertoldi G, Rigon R, Over TM. 2006. Impact of Watershed Geomorphic Characteristics on the 

Energy and Water Budgets. Accepted to Journal of Hydrometeorology. 

Brutsaert W. 2004. On a derivable formula for long-wave radiation from clear sky. Water 

Resources Research 11: 742-744. 

Businger JA, Wyngaard JC, Izumi Y, Bradley EF. 2004. Flux-profile relationship in the 

atmospheric surface layer. Journal of Atmospheric Sciences 28: 181-189. 

Carroll T, Cline D, Olheiser C, Rost A, Nilsson A, Fall G, Bovitz C, Li L. 2006. NOAA’s national 

snow analyses. Paper presented at the 74 th Annual Meeting of the Western Snow Conference 

2006. 

Cazorzi F, Dalla Fontana G. 1996. Snowmelt modeling by combining air temperature and a 

distributed radiation index. Journal of Hydrology 181: 169-187. 

Cline DW, Bales CB, Dozier J. Estimating the spatial distribution of snow in mountain basins 

using remote sensing and energy balance modeling. Water Resources Research 34(5): 1275- 

1285. 

Cline D, Elder K, Bales R. 1998. Scale effects in a distributed snow water equivalence and snow 

melt model for mountain basins. Hydrological Processes 12: 1527-1536. 

208

Colbeck SC, Anderson EA. 1982. The permeability of a melting snow cover. Water Resources 

Research 18(4): 904-908. 

Hall DK, Andrew BT, James LF, Alfred TCC, Milan A. 2000. Intercomparison of satellite-derived 

snow-cover maps. Annals of Glaciology 31: 369-376. 

Hall KH, Riggs GA, Salomonson VV, DiGirolamo NE, Bayr KJ. 2002. MODIS snow-cover 

products. Remote Sensing of Environment 83: 181-194. 

Idso SB. 1981. A set of equation for full spectrum and 8 to 14 µm and 10.5 to 12.5 µm thermal 

radiation from cloudless skies. Water Resources Research 17(2): 295-304. 

Iqbal M. 1983. An introduction to solar radiation. Academic Press. 

Jordan R. 1991. A one dimensional temperature model for a snowcover: Technical documentation 

for SNTHERM 89, Spec. Rep. 657, User’s guide. US Army Cold Regions Research and 

Engineering Laboratory, Hanover, NH, USA. 

Lee S., Klein AG, Over TM. 2005. A comparison of MODIS and NOHRSC snow-cover products 

for simulating streamflow using the Snowmelt Runoff Model. Hydrological Processes 19: 

2951-2972. 

Marks D, Domingo J, Susong D, Link T, Garen D. 1999. A spatially distributed energy balance 

snowmelt model for application in mountain basins. Hydrological Processes 13: 1935-1959. 

Martinec J, Rango A. 1986. Parameter values for snowmelt runoff modeling. Journal of 

Hydrology 84: 197-219. 

Neteler M. 2005. Time series processing of MODIS satellite data for landscape epidemiological 

applications. International Journal of Geoinformatics 1(1): 133-138. 

Oleson KW, Dai Y, Bonan G, Bosilovich M, Dickinson R, Dirmeyer P, Hoffman F, Houser P, 

Levis S, Niu G, Thornton P, Vertenstein M, Yang Z, Zeng X. 2004. Technical description of 

the Community Land Model (CLM). NCAR, Boulder, CO, USA. 

Pomeroy JW, Toth B, Granger RJ, Hedstrom NR, Essery RLH. 2003. Variation in Surface 

Energetics during Snowmelt in a Subartic Mountain Catchment. Journal of Hydrometeorology 

4: 702-719. 

Riggs GA, Hall DH, Salomonson VV. 2003. MODIS Snow Products User Giude. Electronic 

document: http://modis-snow-ice.gsfc.nasa.gov/sugkc.html. 

Riggs GA, Hall DK. 2006. Evaluation of snow errors in the MODIS snow data products. Poster 

presented at the 63 rd Annual Meeting of the Eastern Snow Conference 2006. 

Rigon R, Bertoldi G, Over TM. 2006. GEOtop: A Distributed Hydrological Model with Coupled 

Water and Energy Budgets. Accepted to Journal of Hydrometeorology. 

Satterlund DR. 1979. An improved equation for estimating longwave radiation from the 

atmosphere. Water Resources Research 15(6): 1649-1650. 

Turpin O, Ferguson R, Johansson B. 1999. Use of remote sensing to test and update simulated 

snow cover in hydrological models. Hydrological Processes 13: 2067-2077. 

US Army Corps of Engineers. 1956. Snow hydrology, Summary report of the snow investigations. 

US Army Corps of Engineers, North Pacific Division, Portland, OR, USA. 

Zanotti F, Endrizzi S, Bertoldi G, Rigon R. 2004. The GEOTOP snow module. Hydrological 

Processes 18: 3667-3679. 

209


210

ABSTRACT 

211 



Microstructural Characterization of Firn 

I. BAKER 1 , R. OBBARD 1 , D. ILIESCU 1 , AND D. MEESE 1,2 

In this paper, we use a scanning electron microscope (SEM) coupled with x-ray spectroscopy 

and electron back-scattered diffraction patterns to examine firn in cores retrieved by the United 

States International Trans-Antarctic Scientific Expedition. From grain boundary grooves we were 

able to see where the previously-existing snow crystals were joined, and can determine grain sizes. 

From the SEM images, the porosity and the surface area per unit volume of the pores were 

measured. Finally, we have shown that we can determine the microchemistry of impurities in firn 

and demonstrated that we can determine the orientations of the firn crystals. 

Keywords: Firn, snow, porosity, surface area, scanning electron microscopy, x-ray microanalysis 

INTRODUCTION 

In a number of recent papers, we have used a scanning electron microscope coupled with 

energy-dispersive x-ray microanalysis (EDS) to determine the microstructural location of 

impurities in ice cores (Cullen and Baker 2000; 2001; 2002a; 2002b; Cullen et al. 2002; Iliescu et 

al. 2002; Baker et al. 2003; Baker and Cullen 2003a; 2003b; Obbard et al. 2003a; 2003b; Iliescu 

and Baker 2004; Baker et al. 2006). A key feature of this work is that the ice specimens were 

uncoated, and examined at -80 ± 10 °C so that slight sublimation from the ice prevented charge 

build-up. This is a technique that we have also used to examine snow (Iliescu and Baker 2002) and 

that Barnes and co-workers used to examine both ice cores and snow from Antarctica (Barnes et 

al. 2002a; 2002b; 2003). This approach has an advantage over the use of aluminum or gold 

coating on ice to prevent charging in the SEM (Mulvaney, Wolff and Oates 1988; Wolff, 

Mulvaney and Oates 1988; Barnes et al. 2002b) since the metal coatings can obscure weaker Xray 

fluorescence signals from the sample. 

In other work, we have demonstrated that Raman spectroscopy coupled with a confocal 

scanning optical microscope can be used to analyze impurities in grain boundaries and triple 

junctions that cannot be detected by EDS (Iliescu et al. 2004). MicroRaman spectroscopy was first 

used to look at impurities in a triple junction in ice by Fukazawa and co-workers (1998). 

More recently, we have demonstrated that we can obtain the complete orientations of crystals in 

polycrystalline ice (not simply the c-axis direction) with high angular (0.1 o ) and spatial resolution 

(50 nm) by using electron back-scattered patterns (EBSPs) obtained in a SEM (Iliescu et al. 2004; 

2005; Obbard et al. 2006). 

In this paper, we show, for the first time, that we can obtain SEM images, EDS and EBSPs from 

uncoated firn using the same approach. Ultimately, such information can provide an efficient and 

accurate method of measuring porosity, internal surface area and grain size and will allow us an 

1 

Thayer School Of Engineering, Dartmouth College, Hanover, NH 03755, Email: Ian Baker 

(Ian.Baker@Dartmouth.Edu) 

2 

Climate Change Institute, University Of Maine, Orono, ME 04469

understanding of how impurities are redistributed and how fabric forms in the shallow layers of 

ice sheets. 

EXPERIMENTAL 

We examined six firn specimens from depths of 9.71 m, 19.545 m and 34.465 m from core 02- 

SP, and depths of 13.019 m, 23.028 m and 38.533 m, from core 02-5 obtained on the United States 

International Trans-Antarctic Scientific Expedition (ITASE). Core 02-SP is at the South Pole 

while core 02-5 is the next core along the route, see the route map at 

http://www2.umaine.edu/USITASE/html/map.html for details. 

For examination in the SEM, ~25-mm- × 25-mm- × 10-mm-thick specimens were cut 

perpendicular to the core axis, shaved flat with a razor blade under a HEPA-filtered, laminar-flow 

hood at -10°C and frozen onto a brass plate. Specimens were then either sealed in a small 

container for later examination or mounted onto a cold stage for immediate observation. For 

imaging and microanalysis, the uncoated specimens were examined at -80°C ± 10°C using a JEOL 

5310 low-vacuum SEM operated at 10 kV equipped with a Princeton Gamma-Tech IMIX EDS 

system utilizing a lithium-drifted silicon thin-window detector. For orientation determination, 

specimens were frozen to a purpose-built pre-tilted brass sled that was then mounted on a cold 

stage in a field emission gun FEI XL-30 environmental SEM. The SEM was operated at 15 kV 

with a beam current of 0.15 nA, and the ice was again examined at temperatures between -80°C ± 

10°C at a pressure of approximately 5 × 10 –4 Pa. Secondary electron imaging was used on both 

SEMs. EBSPs were captured using the techniques described by Iliescu and others (Iliescu et al. 

2004; 2005) and indexed using HKL Technologies’ CHANNEL 5 software. 


Figure 1 shows low-magnification secondary electron images of firn from three different depths. 

The images were obtained as soon as possible after the specimens were inserted into the SEM to 

reduce the degree of sublimation, which can affect the appearance of the specimens. Several 

features are worth noting. First, shaving the ice produced some debris, which fell into the pores in 

the ice. This was much more noticeable in the higher porosity shallow ice, presumably because the 

grains were less strongly bonded together and the pore volume is larger. Examples of such debris 

are labeled “D” in Figure 1(a). 

Second, grain boundaries grooves were observed in some images. These indicate where the 

snow crystals are joined together, see Figure 1 (a). By allowing the specimens to sublimate in the 

SEM it is possible to see grain boundary grooves between all the individual ice crystals and, 

hence, it should be possible to determine how the grain size changes as a function of depth in the 

firn. However, one needs to be aware that longer sublimation times, due to loss of mass, 

drastically affect the appearance of the microstructure and hence a delicate compromise needs to 

be achieved between the need to depict the microstructure in an accurate unadulterated manner 

and the desired to better visualize the details within the grain boundary region. 

Third, it is straightforward to determine the % areal porosity from these images by dividing the 

area occupied by the shaved flat regions by the area of the image. The pore area and length of 

boundary area were measured on the SEM images using Image SXM (Barrett 2005), a derivative 

of NIH Image (Rasband 1997). The boundaries of pore areas were traced with the drawing tool, 

and the perimeter and enclosed area measured with the pixel-counting measurement utility, see 

Figure 2. As is quite common in the SEM, the edges of the flat surface tended to charge. This 

analysis was performed from multiple images at each depth (except for the 9.71 m specimen, 

where only one image was available). The porosity values obtained at each depth are shown in 

Table 1. As might be expected, (except for the 9.71 m specimen), the porosity at a particular site 

decreases with increasing depth. 

212

A B 

Figure 1. Low-magnification secondary electron images of firn samples from depths of (a) 9.71 m, (b) 23.028 

m, and (c) 34.465 m. The features labeled “D” in (a) are debris produced by shaving. Some grain boundaries 

where the original snow crystals joined together, as indicated by grooves, are labeled “GB”. Some air bubbles 

are indicated by “A”. 

Table 1. Porosity, measured length of internal surface around the pores in 16 mm2 areas, length of 

(internal surface) line per unit area, LA, and internal pore surface area per unit volume, SV, for 

various depths at two sites from the US ITASE measured from SEM images. 

Site Depth 

(m) 

% 

Porosity 

C 

Measured length 

of surface (mm) 

213 

LA (mm -1 ) SV (mm -1 ) 

02-SP 9.71 80 33.2 2.08 2.65 

02-5 13.019 85 28.7 1.79 2.28 

02-SP 19.545 87 44.5 2.78 3.54 

02-5 23.028 58 30.4 1.98 2.52 

02-SP 34.465 36 40.7 2.54 3.23 

02-5 38.533 38 40.6 2.54 3.23

Fourth, by measuring the length of the surface around the pores per unit area it is possible to 

obtain the internal surface area per unit volume, see Figure 2. Microstructural examinations are 

normally performed on firn by infiltrating the firn with a liquid, e.g. dimethyl Phthalate (Albert 

and Shultz 2002; Rick and Albert 2004), allowing it to set, and then sublimating the firn. The 

problems with this method are that the viscous liquid cannot easily penetrate small pores, which 

increase in frequency with depth; and, of course, the liquid cannot infiltrate closed off pores, 

which also increase in frequency with depth. The lengths of projected surface around the pores 

measured from images of area 16 mm 2 are shown along with the length of (internal surface) line 

per unit area, LA, in Table 1. The internal surface area per unit volume, SV, was calculated from 

LA using: SV = (4/π) LA (Underwood 1970). In contrast to the volume porosity, the internal surface 

area per unit volume of the pores was generally found to increase with increasing depth. This is an 

unexpected result. It may indicate that the pores are more convoluted at greater depth even though 

their percentage of the volume is less. A more detailed SEM analysis would involve taking both 

horizontal and vertical sections. This becomes more important with increasing depth where the 

pore and grain structures become more inhomogeneous due to the flattening of the microstructure 

from the increasing overburden. Thus, it is possible that this sectioning issue may be at the root of 

the unexpected increase in internal surface area with increasing depth. 

Fifth, the edges of the flats shaved on the specimens preferentially etch before the rest of the 

surface, see Figures 1(b) and 1(c). 

And, sixth, some air bubbles can be observed, which have been trapped in the ice, see Figures 

1(b) and 1(c). 

Figure 2. Secondary electron image of firn sample from 13.019 m. The boundaries of the pore areas were 

traced with the drawing tool, and the perimeter and enclosed area measured with the pixel-counting 

measurement utility of Image SXM (Barrett 2005). 

Figure 3 shows a higher magnification image of two grains in firn. In this case, ridges 

(indicated) appear to have formed on either side of the grain boundary groove. Because the ridges 

were well below the surface of the ice, EDS analysis was not possible. Adams et al. (2001) 

observed a grain boundary ridge on snow crystals that were sintering together, a feature that they 

suggested was indicative of direct evidence for grain boundary diffusion as a sintering mechanism 

in ice. However, Barnes (2003) later pointed out that the ridge may have been a grain boundary 

impurity filament of the kind observed previously in ice (Baker and Cullen 2003a; 2003b; Baker et 

al. 2003; Baker et al. 2006; Barnes et al. 2002a; 2002b; Cullen and Baker 2000; 2001; 2002a; 

2002b; Cullen et al. 2002; Iliescu et al. 2002; Iliescu and Baker 2004; Obbard et al. 2003a; 2003b). 

214

GB ridge 

GB groove 

Figure 3. Higher-magnification secondary electron images of firn samples showing ridging on either side of a 

grain boundary groove (arrowed). The vertical lines are scan faults. 

Figure 4 shows x-ray spectra taken from white spots formed during the sublimation of the 

surface of firn specimens. The oxygen peak is from the ice, while the carbon peak arises from the 

breakdown of diffusion pump oil under the electron beam in the SEM. In the shallow samples 

(9.71 m and 13.019 m), for example Figure 4(a), only small, barely-detectable quantities of 

impurities were found in the white spots. In this case, S and Cl were found. Spectra from white 

spots on deeper firn specimens showed more pronounced peaks. The example spectra shown in 

Figure 4(b) contain Cl, K, Na, S, Si and possibly Ca. Such impurities have often been noted in 

white spots in ice specimens from Greenland and Antarctica (Cullen and Baker 2001; 2002b Baker 

and Cullen 2003a; Barnes et al. 2002a; 2002b; Obbard et al. 2003a; 2003b; Baker et al. 2006). It is 

evident that impurities are spread throughout the firn. We did not find any impurities segregated to 

the grain boundaries as has been found in ice from Antarctica and Greenland (Cullen and Baker 

2000; 2001; 2002b; Cullen et al. 2002; Barnes et al. 2002a; 2002b; Baker et al. 2003; Baker and 

Cullen 2003a; 2003b; Obbard et al. 2003a; 2003b; Baker et al. 2006). 

Finally, Figure 5 shows an EBSP from firn from 3 m. The reds lines indicate the centers of the 

indexed Kikuchi bands. The pattern, while not as good quality as patterns previously presented for 

ice (Cullen and Baker 2002a; Iliescu et al. 2004; 2005; Obbard et al. 2006) is of sufficient quality 

that it is indexable. Various poles are indicated on the pattern. Each EBSP contain the complete 

three-dimensional orientation information of the crystal from whence it came and together the 

patterns enable the misorientations between grains to be determined. 

215 

GB ridge

A 

B 

Figure 4. X-ray spectra taken from white spots, such as that indicated in Figure 1(c), on the surface of firn 

samples from depths of: (a) 9.71 m; and (b) 38.533 m. The full scale counts are 1130 for (a) and 230 from the 

three spectra in (b). 

216

CONCLUSION 

Figure 5. Indexed electron backscatter diffraction pattern from firn from a depth of 3 m. 

In conclusion, we have shown that it is possible and useful, using various advanced electronoptical 

techniques and x-ray spectroscopy measurements in a SEM, to examine firn at high 

resolution. By observing grain boundary grooves, one can determine where snow crystals were 

joined; their orientations can be determined from EBSPs, allowing the misorientations between 

grains to be calculated; the porosity, approximate grain size and internal pore surface area can be 

measured; and the microchemistry of impurities can be determined by EDS. 

ACKNOWLEDMENTS 

The cores were collected during the 2005 US International Trans Antarctic Scientific Expedition 

and samples provided from NSF Grant 9980434. Support by the U.S. National Science Foundation 

Office of Polar Programs Grant 0440523 and U.S. Army Research Office under contract DAAD 

19-03-1-0110 are gratefully acknowledged. The views and conclusions contained herein are those 

of the authors and should not be interpreted as necessarily representing official policies, either 

expressed or implied, of the National Institute of Standards and Technologies, or the U.S. 

Government. 

REFERENCES 

Adams EE, Miller DA, Brown RL. 2001. Grain boundary ridge on sintered bonds between ice 

crystals. Journal of Applied Physics 90: 5782–5785. 

Albert M, Shultz EF. 2002. Snow and firn properties and air–snow transport processes at Summit, 

Greenland. Atmospheric Environment 36: 2789–2797. 

Baker I, Cullen D. 2003a. The Structure and Chemistry of 94m GISP2 ice. Annals of Glaciology 

35: 224–230. 

Baker I, Cullen D. 2003b. SEM/EDS observations of impurities in polar ice: artifacts or not? 

Journal of Glaciology 49(16): 184–190. 

217

Baker I, Cullen D, Iliescu D. 2003. The microstructural location of impurities in ice. Canadian 

Journal of Physics 81: 1–9. 

Baker I, Iliescu D, Obbard R, Chang H, Bostick B, Daghlian CP. 2006. Microstructural 

Characterization of Ice Cores. Annals of Glaciology, in press. 

Barnes PRF, Mulvaney R, Robinson K, Wolff EW. 2002a. Observations of polar ice from the 

Holocene and the glacial period using the scanning electron microscope. Annals of Glaciology 

35: 559–566. 

Barnes PRF, Mulvaney R, Wolff EW, Robinson K. 2002b. A technique for the examination of 

polar ice using the scanning electron microscope. Journal of Microscopy 205: 118–124. 

Barnes PRJ. 2003. Comment on "Grain boundary ridge on sintered bonds between ice crystals" [J. 

Appl. Phys. 90, 5782 (2001)]. Journal of Applied Physics 93: 783–785. 

Barnes PRJ, Wolff EW, Mallard DC, Meider HM. 2003. SEM Studies of the Morphology and 

Chemistry of Polar Ice. Microscopy Research and Technique 62: 62–69. 

Barrett S. 2005. Image SXM, Surface Science Research Centre, University of Liverpool, 

Liverpool, UK. 

Cullen D, Baker I. 2000. The Chemistry of Grain Boundaries in Greenland Ice. Journal of 

Glaciology 46: 703–706. 

Cullen D, Baker I. 2001. Observation of Impurities in Ice. Microscopy Research and Technique 

55: 198–207. 

Cullen D, Baker I. 2002a. Sulfate Crystallites in Vostok Accretion Ice. Materials Characterization 

48: 263–270. 

Cullen D, Baker I. 2002b. Scanning Electron Microscopy of Vostok Acrretion Ice. In Proceedings 

of Microscopy and Microanalysis; 1546-7CD. 

Cullen D, Iliescu D, Baker I. 2002. SEM/EDS Studies of Impurities in Natural Ice. In Proceedings 

of Microscopy and Microanalysis; 1398-9CD. 

Fukazawa H, Sugiyama K, Mae S, Narita H, Hondoh T. 1998. Acid ions at triple junction of 

Antarctic ice observed by Raman scattering. Geophysical Research Letters, 25: 2845. 

Iliescu D, Baker I. 2002. Imaging of Uncoated Snow Crystals Using a Low-Vacuum Scanning 

Electron Microscope. Journal of Glaciology 48(162): 479–480. 

Iliescu D, Baker I, Cullen D. 2002. Preliminary Microstructural and Microchemical Observations 

of Pond and River Accretion Ice. Cold Regions Science and Engineering 35: 81–99. 

Iliescu D, Baker I. 2004. Characterization of the microstructure and mechanical properties in 

seasonal lake and river ice. In Proc. 24 th Army Science Conference, Orlando, FL; OS-26. 

Iliescu D, Baker I, Chang H. 2004. Determining the Orientations of Ice Crystals using Electron 

Backscatter Patterns. Microscopy Research and Technique 63: 183–187. 

Iliescu D, Baker I, Daghlian CP. 2005. Orientation Mapping in Polycrystalline Ice Using Electron 

Backscatter Patterns. In Proceedings of Microscopy & Microanalysis, Honolulu, HI; 1500– 

1501. 

Mulvaney R, Wolff EW, Oates K. 1988. Sulphuric acid at grain boundaries in Antarctic ice. 

Nature 331: 247. 

Obbard R, Iliescu D, Cullen D, Chang J, Baker I. 2003a. SEM/EDS Comparison of Polar and 

Seasonal Temperate Ice. Microscopy Research and Technique 62: 49–61. 

Obbard R, Iliescu D, Baker I, Cullen D. 2003b. A Technique for the Scanning Electron 

Microscopy and Microanalysis of Uncoated Ice Crystals. In Electron Microscopy: Its Role in 

Materials Science, The Mike Meshii Symposium, TMS, Warrendale, PA; 133–140. 

Obbard R, Baker I, Sieg K. 2006. Using Electron Backscatter Diffraction Patterns To Examine 

Recystallization in Polar Ice Sheets. Submitted to Journal of Glaciology. 

Rasband W. S. 1997–2006. NIH ImageJ, http://rsb.info.nih.gov/ij/. 

Rick U, Albert, M. 2004. Microstructure of West Antarctic Firn and its Effect on Air Permeability. 

ERDC/CREEL Report TR-04-16. 

Underwood EE. 1970. Quantitative Stereology, Addision-Wesley Publishing Co., Reading, Mass, 

U.S.A., p24. 

Wolff EW, Mulvaney R, Oates K. 1998. The location of impurities in Antarctic ice. Annals of 

Glaciology, 11: 194. 

218

219 

63rd EASTERN SNOW CONFERENCE 

Delaware, DE, USA 2005 

Computational Time Steps of Winter Water Balance 

for Snow Losses at United States Meteorological Stations 

ABSTRACT 

S.R. FASSNACHT 1 

When estimating the water balance for a cold region watershed, that is one that receive a 

substantial portion of its annual precipitation as snow, accumulation and other winter hydrological 

processes must be considered. For many of theses watersheds, all but the most fundamental 

meteorological data (temperature and precipitation), are either not measured or not measured at a 

reasonable time step. Of particular importance are wind data, as wind influences underestimates of 

precipitation due to wind undercatch and losses of snow from the snowpack, specifically, 

snowpack sublimation, and the occurrence and magnitude of blowing snow. Estimating snow 

accumulation to yield snowmelt amounts requires summing of gauged precipitation and gauge 

undercatch, and subtracting minus snowpack sublimation and blowing snow transport. The first 

two components are computed on a daily time step, while the latter two are computed on an hourly 

time step. From five National Weather Service meteorological stations, the variations in computed 

snowpack mass losses minus undercatch using data at different time intervals show that at most 

sites it is difficult to use monthly time steps for computations derived using hourly or daily data. 

At the relative dry and cold Leadville, Colorado site the computations were transferable between 

time steps. 

Keywords: solid precipitation, meteorological data, undercatch, sublimation, blowing snow 

INTRODUCTION 

Snow is disturbed by wind, from snowfall to movement through redistribution to sublimation. 

Snowfall quantities are underestimated from precipitation gauges due to undercatch from wind, 

wetting of gauges, and to a lesser degree evaporation (Goodison et al., 1998). Undercatch due to 

wind is caused by the deformation of the wind field around the gauge orifice. As well, falling 

snow crystals are more easily blown away from the gauge orifice than rain drops. Snow on the 

ground can be redistributed based on wind characteristics, upwind and downwind fetch length, and 

the history of the snowpack surface (Pomeroy et al., 1991). Wind across a snowpack (or across 

snow held by vegetation) can sublimate snow away from or towards the surface depending upon 

temperature and humidity profiles (Sverdrup, 1936). 

A watershed analysis uses the water balance to partition water storage and movement into 

different components of the hydrological cycle. At the end of the accumulation period, the 

remaining snow melts to contribute to runoff. Accumulation, given as snow water equivalent 

(SWE), is typically estimated as the cumulative precipitation occurring at air temperatures colder 

than freezing (0 degrees Celsius), without correcting for precipitation underestimation due to 

gauge undercatch, nor snowpack losses due to sublimation or blowing snow. 

1 

Watershed Science Program, College of Natural Resources, Colorado State University, Fort 

Collins, Colorado 80523-1472 USA.

Fassnacht (2004) used equations derived from field measurements to estimate gauge undercatch 

and compared it to sublimation and blowing snow from six weather stations across the United 

States (U.S.) in order to adjust monthly and seasonal accumulation. With these considerations, the 

amount of SWE that accumulates can be computed as 

SWE = P + P m F m q 

(1) 

g 

U 

E 

BS 

where Pg is the measured amount of precipitation, PU is the estimated amount of gauge 

underestimation (hereinafter assumed to be mainly due to undercatch), FE is the amount of 

sublimation (away from or towards the snowpack), and qBS is the amount of blowing snow 

redistributed (scoured away from or deposited at the snowpack). The precipitation (measured plus 

undercatch) is an accumulation of snow, as estimated from a gauge, whereas sublimation and 

blowing snow are losses from the snowpack which is has accumulated beside a precipitation 

gauge. These components can be computed for a point location using meteorological data. 

Fassnacht (2004) compared these components to see if measured precipitation, without 

consideration of undercatch or other biases, could be used as an estimate of snowpack 

accumulation after sublimation and blowing snow had reduced accumulation. 

For U.S. National Weather Service (NWS) automated surface observation stations (ASOS), 

meteorological data are reported over an hourly interval (to be used to compute FE and qBS in 

equation 1). The NWS cooperative (COOP) stations data are reported over a daily interval (to be 

used to compute Pg and PU in equation 1). These data and monthly summaries are available online 

via the NWS National Climate Data Center (NCDC, 2006). Data are presented as quantities, with 

the exception of precipitation events that are less than 0.254 mm (0.01 inches), which are reported 

as trace events. Fassnacht (2004) assumed that these trace events yielded precipitation at one half 

of the minimum detection (0.127 mm). Yang et al. (1998a) stated that trace events can be 

significant in drier environments, such as Alaska. 

Fassnacht (2004) scrutinized the validity of undercatch, sublimation and redistribution estimates 

in trying to determine if and where PU is approximately equal to FE plus qBS, so that SWE can be 

set to Pg for equation 1. Considering that water balance computations are typically made for 

monthly intervals, this paper compares the components of equation 1 for individual winter months 

and the entire winter season as computed using different time steps. Specifically the objectives are 

1) to compare the transferability of computed snow loss rates (precipitation undercatch, snowpack 

sublimation and blowing snow transport) over different time scales (hourly, daily, and monthly); 

2) for monthly undercatch to determine if there is a difference using monthly average (for 

temperature and wind speed, with totals for precipitation) of the daily data (hereinafter called 

average monthly data) versus using monthly data adjusted for the monthly probability of each 

precipitation type (snow, mixed precipitation, or rain) together with the average wind speed during 

each precipitation type (hereinafter called monthly phase partitioned data); and 3) to determine if 

monthly or seasonal gauged precipitation can be used to estimate discrepancies in computations of 

Pg, PU, FE and qBS from different time steps. Since the precipitation undercatch equations were 

derived from data at daily interval, undercatch was not computed using hourly data. 

STUDY SITES 

Four of the six meteorological stations across the conterminous U.S. used by Fassnacht (2004) 

were analysed in this study (Table 1). Pullman WA was been substituted for the Stanley ID 

station, since there were no observed trace events at Stanley during the study period. Pullman WA 

has a similar climate to Stanley (Table 2 and Fassnacht, 2004), receiving 6 mm more precipitation 

per winter month, being warmer (–0.6 degrees C average air temperature versus –5.9 degrees C), 

more humid (a vapour pressure of 4.9 mb versus 3.3 mb), and more windy (4.1 m s –1 average wind 

speed versus 1.3 m s –1 ), but having the same vapour pressure deficit. The South Lake Tahoe 

station was not used, as the no suitable undercatch equation has been derived for the heating 

tipping bucket gauge used to estimate daily precipitation. 

220

Meteorological data for the winter (October–June) of three water years (2000–2002) were 

retrieved from the NCDC online database (NCDC, 2006). Data were available only for water years 

2001 and 2002 for the Rawlins WY station (Table 1). 

Snow depths were not recorded, thus, hourly temperature and precipitation data were examined 

for each year for each station to determine when snow started to accumulate and when it ablation 

was likely complete. These dates were rounded to the nearest month (Table 1). While the phase of 

precipitation was not known for the study sites, it was observed that only in the later winter 

months did precipitation occasionally occur at air temperatures warmer than 0 degrees C. This 

factor also helped determine the start and end of ablation. The monthly average precipitation, 

temperature, wind speed and vapour pressure for the winter months are summarized in Table 2, for 

the winter months given in Table 1. 

Table 1. Summary of stations used in analyses, and the periods considered winter for each of the three 

water years of interest (2000–2002). The precipitation gauge type is denoted as SRN for the NWS 

standard 8" rain gauge or BUG for the Belfort Universal Recording Rain Gauge. 

elevation latitude longitude 

winter period precipitation 

station state (m) (N) (W) 2000 2001 2002 gauge type 

Pullman WA 778 46Ε45' 117Ε7' Dec–Jan Nov–Feb Dec–Feb SRN 

Rawlins WY 2053 41Ε48' 107Ε12' no data Nov–Mar Nov–Mar SRN 

Leadville CO 3029 39Ε14' 106Ε19' Dec–Apr Nov–Apr Nov–Apr SRN 

Rhinelander WI 487 45Ε38' 89Ε28' Dec–Feb Nov–Mar Dec–Apr SRN 

Syracuse NY 125 43Ε7' 76Ε6' Jan–Feb Dec–Mar Dec–Feb BUG 

Table 2. The average (mean) and coefficient of variation (COV) of the station meteorology for the 

winter periods listed in Table 1. Note: † precipitation is corrected using daily data. 

precipitation temperature humidity 

vapour 

pressure 

(mm)† (ΕC) (mb) deficit (mb) wind (m/s) 

station mean COV mean COV mean COV mean COV mean COV 

Pullman, WA 37.3 0.37 –0.63 –1.57 4.94 0.06 1.04 0.301 4.1 0.176 

Rawlins, WY 19.2 0.552 –4.71 –0.62 3.12 0.218 1.36 0.294 5.6 0.170 

Leadville, CO 27 0.552 –5.24 –0.74 2.44 0.250 1.76 0.449 3.6 0.111 

Rhinelander, WI 26.7 0.749 –6.64 –0.51 3.19 0.251 0.97 0.278 3.4 0.097 

Syracuse, NY 89.9 0.473 –1.66 –1.69 4.32 0.183 1.48 0.250 4.3 0.086 

METHODOLOGY 

To address the objectives of this paper, precipitation gauge undercatch was estimated using 

daily and monthly data. The monthly mean of daily data produced the average monthly data. 

Precipitation was summed for each month. To generate undercatch estimates from the monthly 

phase partitioned data, daily data were used to identify the form of the precipitation, and yielded a 

fraction of the monthly precipitation. The average wind speed during each precipitation type was 

used together with the fraction of the monthly precipitation type. Snowpack sublimation and 

blowing snow transport were estimated from data at hourly, daily, and average monthly, as 

detailed in Fassnacht (2004). 

The amount of gauge undercatch due to wind was computed as a function of measured 

precipitation and wind speed (Uz) (Yang et al., 1998b): 

U 

( P U ) 

P = f , . (2) 

g 

z 

221

The height of the anemometer at each station was assumed to be at 6m, and the height of the 

gauge orifice was assumed to be at 2m. The wind speed was converted from a 6m height to a 2 m 

height, as per Goodison et al. (1998) using the snowpack aerodynamic roughness of 0.005 m 

(Fassnacht et al., 1999). 

For solid precipitation, wetting losses were assumed to be small as compared to wind induced 

losses, and for monthly computations were assumed to be minimal. Similarly, evaporation losses 

were assumed to be negligible (Goodison et al., 1998). Undercatch equations were derived from 

daily data for snowfall when the daily air temperature (Ta) was colder than freezing and for mixed 

precipitation when Ta was between 0 and 3 degrees C (Goodison et al., 1998). The specific 

equations derived by Yang et al. (1998b) for the unshielded 8" NWS Standard precipitation gauge 

with respect to the Double Fence Intercomparison Reference gauge (DFIR) were used for all sites 

except Syracuse (Table 1). This station used a Belfort Universal gauge. As per Groisman et al. 

(1999), the Yang et al. (1998b) equation was used for snowfall undercatch of the Belfort Universal 

gauge. Mixed precipitation undercatch was increased by 7% for the Belfort Universal gauge used 

at Syracuse (Groisman et al., 1999). As per Fassnacht (2004) and Bogart et al. (2006), a maximum 

wind speed of 6.5 m/s was used in the undercatch equations due to the increased uncertainty at the 

higher wind speeds. 

Sublimation and the occurrence of blowing snow are episodic and were thus computed using 

hourly data. Sublimation was estimated using the bulk transfer approach for the latent heat flux as 

a function of humidity (RH), air temperature, wind speed, and station pressure (PR): 

( RH, 

T , U , PR) 

FE = f a z , (3) 

as initially formulated by Sverdrup (1936). The occurrence of blowing snow (BSY/N) was 

initially estimated as a function of wind speed and different temperature considerations, as per Li 

and Pomeroy (1997): 

Y / N = . (4) 

( U T ) 

BS f , 

z 

a 

Once blowing snow was determined to have initiated, the quantity of blowing snow was 

computed as a function of wind speed: 

BS 

( U ) 

q = f , (5) 

z 

using the equation derived by Pomeroy et al. (1991). Sublimation and blowing snow quantities 

were summed to yield net snowpack loss estimates. 

The NWS denotes trace events (PT) as precipitation amounts less than 0.01 inches or 0.254 mm 

per hour or day (NWS, 2005), while Legates et al. (2005) called all measurements less than half 

the measurable precipitation depth (0.005 inches or 0.127 mm) as a trace event. In this paper, daily 

trace events will be assigned a value of 0.127 mm. 

RESULTS 

Hourly, daily and monthly meteorological data were used to compute PU (Figure 1) and 

snowpack losses (FE plus qBS) (Figure 2). The estimated difference between losses and undercatch 

using the time step specific to the derived equations versus using a monthly time step is presented 

in Figure 3 for monthly totals, and Figure 4 for seasonal totals. For the difference comparison, the 

same net result would appear along the 1:1 line, while data at the origin would indicate that 

measured precipitation could be used as an estimate of snow on the ground. Below the x-axis or to 

the left of the y-axis, more snow is actually accumulating than estimated from the measured 

precipitation alone. The difference between the two time step estimates and gauge precipitation for 

individual months is illustrated in Figure 5, and for each season is illustrated in Figure 6. 

222

monthly probability 

adjusted data 40 

[% of δt 

cumulative] 

90 

80 

70 

60 

50 

30 

20 

10 

0 

100 

0 

0 10 20 30 40 50 

daily data 

60 70 80 90 100 

[% of δt 

cumulative] 

100 

223 

90 

80 

70 

60 

average monthly data 

[% of δt 

cumulative] 

50 

Pullman WA 

Rawlins WY 

Leadville CO 

Rhinelander WI 

Syracuse NY 

Figure 1. Comparison of monthly precipitation undercatch estimates using daily and monthly data. Each 

estimate is presented as a percentage of the total of the three different datasets. The values for undercatch 

have been derived for daily data. The average monthly data were derived from the mean of the daily data, 

whereas the monthly probability adjusted data were derived using the monthly precipitation distribution 

(snow, mixed precipitation, or rain) and the average wind speed during each type of precipitation. 

average 

monthly data 40 

[% of δt 

cumulative] 

90 

80 

70 

60 

50 

30 

20 

10 

0 

100 

0 

0 10 20 30 40 50 60 70 80 90 100 

hourly data 

[% of δt 

cumulative] 

100 

90 

80 

70 

60 

50 

40 

40 

30 

30 

20 

20 

10 


Rawlins WY 

Leadville CO 


Syracuse NY 

daily data 

[% of δt 

cumulative] 

Figure 2. Comparison of relative monthly snowpack loss (sublimation plus blowing snow) estimates using 

hourly, daily, and monthly data. Each estimate is presented as a percentage of the total of the three different 

time step estimates. The values have been derived for hourly data. 

10


Rawlins WY 

Leadville CO 


Syracuse NY 

Syracuse 

(-94.4,-178.4) 

Figure 3. Monthly total comparison of the difference between snowpack losses (sublimation plus blowing 

snow) and precipitation undercatch estimated using average monthly meteorological data versus the time step 

for which the values were derived (hourly for losses and daily for undercatch). The dashed line represents the 

1:1 relationship. 


Rawlins WY 

Leadville CO 


Syracuse NY 

Syracuse 

(-191,-337) 

Figure 4. Seasonal total comparison of the difference between snowpack losses (sublimation plus blowing 

snow) and precipitation undercatch estimated using average monthly meteorological data versus the time step 

for which the values were derived (hourly for losses and daily for undercatch). The dashed line represents the 

1:1 relationship. 

224

225 


Rawlins WY 

Leadville CO 


Syracuse NY 

Figure 5. Monthly total difference between monthly derived and equation appropriate time step losses minus 

undercatch versus monthly gauge precipitation. 


Rawlins WY 

Leadville CO 


Syracuse NY 

Figure 6. Seasonal total difference between monthly derived and equation appropriate time step losses minus 

undercatch versus monthly gauge precipitation. 

With the exception of January 2002 at Pullman WA, March 2002 at Rawlins WY, and 

December 2001 at Syracuse NY, undercatch estimates from daily data were at least comparable to 

those from average monthly and from monthly partitioned data (Figure 1). Undercatch estimated 

from monthly average data was representative of those estimated from daily data, while the 

monthly partitioned data was more representative. The results from both monthly dataset were 

most similar for Leadville CO, Rhinelander WI and Syracuse NY, yet for Pullman WA and 

Rawlins WY the monthly partitioned data yielded undercatch estimates more similar to using the 

daily data than using the average monthly data. The mean monthly undercatch estimate for all 

stations and months using daily data was 29.3 mm (standard deviation, s of 27.4, and a maximum 

of 127), using average monthly data was 35.8 mm (s of 40.6 and a maximum of 229), and using 

the monthly partitioned data was 31.5 mm (s of 31.5 and a maximum of 168). 

Snowpack loss estimates were more similar for the different time steps (Figure 2), except for the 

25% of the months when monthly data yielded no loss estimates. For 13 months, monthly derived

estimates were equal to the sum of hourly and daily derived estimates, primarily due to very small 

hourly estimates. The average monthly snowpack losses were 22.7 mm (s of 12.3 and a maximum 

of 54.9), 24.3 mm (s of 13.4 and a maximum of 57.1), and 27.5 mm (s of 22.8 and a maximum of 

113) for hourly, daily and monthly time steps. 

The net difference between losses and undercatch was greater or equal for Rhinelander WI 

when the monthly time step was used compared to the appropriate time step, whereas they tended 

to be equal or less for Syracuse NY (Figure 3 for monthly totals and Figure 4 for seasonal totals). 

As shown by Fassnacht (2004), losses tended to be larger than undercatch, i.e., only 4 months of 

data are in the third quadrant of Figure 3, and only 1 season in the second quadrant of Figure 4. 

Data for Leadville CO and to a lesser extent Rawlins WY plotted along or close to the 1:1 line in 

Figures 3 and 4 (estimates were similar for both time steps). At Rawlins WY, some months had 

greater differences derived from the appropriate time step in the first quadrant and below the 1:1 

line. 

The difference between the y-axis (monthly time step) and x-axis (equation appropriate time 

step) (Figure 3 and 4 for monthly and seasonal totals) is presented as a function of gauged 

precipitation for monthly and seasonal time steps (Figures 5 and 6, respectively). Points along the 

x-axis in Figures 5 and 6 correspond with points along the 1:1 line in Figures 3 and 4. For monthly 

totals (Figure 5), there is no systematic trend, except that the monthly difference may decrease as 

gauged precipitation increases. The Leadville CO data are clustered around the x-axis more 

closely than other stations. 

DISCUSSION 

For some months at some stations estimates of month undercatch and snowpack losses 

(snowpack sublimation and blowing snow transport) are similar using data at an hourly, a daily, or 

a monthly time step. The appropriate time step is daily for undercatch and hourly for sublimation 

and blowing snow. To estimate accumulated SWE from gauge precipitation (equation 1), data at a 

monthly time step could be used for the Leadville CO station, which is a dry environment (low 

humidity) with moderate precipitation and wind (Table 2). Computationally this occurs in part 

since snowpack losses are precipitation limited for some months, i.e., FE plus qBS are equal to Pg 

plus PU for some months. However, using the bulk transfer method has been shown to 

overestimate sublimation (Hood et al., 1999). In particular, the latent heat flux equation uses the 

vapour pressure deficit to compute sublimation, which is drier environments can be greater than 

the available energy flux. As well, blowing snow estimates are likely larger than actual. 

Information on the snowpack history may assist in improving blowing snow estimates. The 

occurrence of blowing snow equations from Li and Pomeroy (1997) are based on the initiation of 

blowing snow on an hourly basis. Finer temporal resolution data (and observations) could improve 

blowing snow estimates. However, archived meteorological data are usually not available at 

shorter time intervals than hourly. 

Both mean monthly undercatch and snowpack loss estimates increased as the temporal 

resolution of the data decreased, since there is more variation (s is larger) due to more large 

values. In particular, there were substantially larger monthly estimates from average monthly data 

for Syracuse NY, which had a persistent wind and various large precipitation events. Removal of 

the March 2001 Syracuse NY (six large events) estimate reduced the mean monthly average 

undercatch of the remaining 69 station-months by 1.6, 3.4, and 2.3 mm using the daily, average 

monthly and monthly partitioned data. This was also the month with the most gauged precipitation 

(138 mm), as illustrated in Figure 5. The monthly mean undercatch estimates for the other four 

stations were 22.9, 24.9, and 24.5 mm. 

Undercatch estimates for December 2000 at Syracuse NY were similar for the different time 

intervals, but it should be noted that snow occurred on 27 days of the month with an average 

monthly wind speed of 5.3 m s –1 . For five days with snow, the daily average wind speed was 

greater than 7 m s –1 . This was the only month where the monthly difference was greater than the 

equation difference (Figure 5). 

226

The daily wind speed can exceed 6.5 m s –1 (achieved at all stations), which would result in an 

undercatch ratio in the order of 20% (collecting one-fifth of the actual snowfall). The undercatch 

equation is based on data from a number of stations. To improve this and the estimates of 

sublimation and blowing snow requires field observations. The precipitation undercatch computed 

in this paper is for unshielded NWS gauges, except at Syracuse. The NWS unshielded gauge is 

typical, but results were similar for the Belfort Universal gauge (Groisman et al., 1999). Net 

precipitation increased by 0.2 to 1% when the Groisman et al. (1999) increase to mixed 

precipitation was considered for the Syracuse site. 

Using data at a monthly time step produced an average wind speed that was more than the 

threshold for blowing snow for only one station for one month (January 2002 at Rawlins WY with 

an average Uz of 7 m s –1 ). Using daily data, the blowing snow threshold wind speed was only 

achieved 41% of the time. With hourly data, it was achieved at least twice each month at each 

station. Snow blowing into the gauge is a consideration for certain gauge and/or shield 

configuration, such as the Tretyakov (Goodison et al., 1998). However, this has not been observed 

to be a problem for the NWS Standard Rain gauge nor the Belfort Universal gauge (Yang et al., 

1998b). 

The appropriate time step versus monthly time step (Figures 3 and 4) yielded similar estimates 

for most months at Leadville CO and some months Pullman WA, Rawlins WY, and Rhinelander 

WI. Additional winters of data should be examined at these and other stations with a variety of 

climates. 

At present, monthly or seasonal gauged precipitation cannot be used to estimate discrepancies in 

computations from different time steps. With more months and stations, it may be possible to 

identify a systematic trend, at least for seasonal data. 

Averaging of data to coarser temporal resolutions does not always produce smaller estimates of 

undercatch, while sublimation estimates tend to be smaller. This was due in part to difference in 

wind speed during precipitation events compared to when no precipitation occurs, as well as the 

nature of averaging, as indicated by comparing undercatch estimated with average monthly data 

versus monthly partitioned data. 

Gauge precipitation is not a useful indicator of the difference between monthly and equation 

appropriate time steps losses minus undercatch, i.e., it cannot be used to systematically adjust 

monthly time step estimates of losses minus undercatch for monthly or seasonal totals (Figure 5 

and 6). Thus, the appropriate time step should be used for most locations for the computations. At 

some stations for some months, there is even a substantial difference between monthly totals 

estimated using hourly versus daily data (Figures 1 and 2). 

Trace events have been illustrated to be important for determining monthly and seasonal 

precipitation amounts (e.g., Yang et al., 1998a). The amount of precipitation assigned to trace 

events is especially important for drier environments. Undercatch during trace events is a problem, 

as it may be uncertain how much snow is actually falling, and thus what the actual undercatch 

ratio is. The frequency of occurrence of measurable versus trace precipitation events could be used 

as a guide to highlight the importance of trace events. 

To quantify the monthly total precipitation, cumulative monthly precipitation should be 

measured at gauging sites, together with daily (and hourly) measurements. While gauge 

evaporation was not considered as it tends to be small for snow (Goodison et al., 1998), it can 

account for some non-wind undercatch. Finer resolution precipitation measurements may need to 

quantify gauge evaporation and wetting losses. The resolution of the non-recording gauge that is 

used at thousands of NWS Cooperative sites across the U.S. is 0.254 mm (0.01 inches) thus 

making the assignment of a value to trace events important. For automated sites, especially those 

at airport (presented in the paper), finer resolution gauges, such as the Geonor (2006) vibrating 

wire gauge, should be explored, especially for hourly measurements. 

227

CONCLUSIONS 

Precipitation gauge undercatch of snowfall should be estimated using daily data, unless new 

relationships between undercatch, wind speed, etc., are developed at different temporal 

resolutions. The inaccuracies of using monthly averaged data are evident from adjusting the 

average data for the monthly probability of each precipitation type and considering wind speeds 

during each type. However, these data must be derived from daily data. For snowpack losses 

(sublimation and blowing snow), it is reasonable to use daily data in lieu of hourly data when 

computing monthly or seasonal values. 

Average monthly data yielded no blowing snow, expect for one month at Rawlins WY. No 

sublimation was estimated 25% of the time using monthly data. For Leadville CO and Rhinelander 

WI, monthly sublimation plus blowing snow was computed to be very similar using each of the 

time steps, for most months. 

The net accumulation difference, i.e., sublimation plus blowing snow minus undercatch, 

computed from the time step of data for which the equations were formulated (hourly for 

sublimation and blowing snow, daily for undercatch) can be estimated from monthly time step 

data at Leadville CO and for several months at Rhinelander WI. Seasonally, monthly time step 

data overestimate for 2 of 3 seasons at Rhinelander WI. Underestimates occur at Rawlins WY, for 

1 season at Pullman WA, and 2 of 3 seasons at Syracuse NY. 

Neither monthly nor seasonal gauged precipitation can be used to estimate discrepancies in 

computations from different time steps, as no systematic trend is obvious. More stations and years 

are required, especially focusing on specific climates. 


The insight and thoughtful comments of two anonymous reviewers and Special Editor Dr. 

Andrew Klein improved this paper and are appreciated. 

REFERENCES 

Bogart TA, Legates DR, Yang D. 2006. Sensitivity and spatial variability of Arctic precipitation 

bias adjustments. Oral presentation at the 63 rd Annual Eastern Snow Conference, Newark, 

Delaware, June 7–9, 2006. 

Fassnacht SR, Soulis ED, Kouwen N. 1999. Shape characteristics of freshly fallen snowflakes and 

their short-term changes. Interactions between the Cryosphere, Climate and Greenhouse Gases 

(Proceedings IUGG 99 Symposium HS2, Birmingham, July 1999), IAHS, 256: 111–122. 

Fassnacht SR. 2004. Estimating Alter-shielded gauge snowfall undercatch, snowpack sublimation, 

and blowing snow transport at six sites in the coterminous United States. Hydrological 

Processes, 18(18): 3481–3492 [doi:10.1002/hyp.5806]. 

Geonor. 2006. Geonor Meteorological Equipment. Geonor Industries, Millbank PA, URL: 

, last accessed: June 20, 2006. 

Goodison BE, Louie PYT, Yang D. 1998. WMO Solid Precipitation Measurement 

Intercomparison Final Report. WMO Instruments and Observing Methods Report No. 67, 

WMO/TD No. 872. 

Groisman PY, Peck EL, Quayle RG. 1999. Intercomparison of recording and standard 

nonrecording U.S. gauges. Journal of Atmospheric and Oceanic Technology, 16: 602–609. 

Hood E, Williams M, Cline D. 1999. Sublimation from a seasonal snowpack at a continental, midlatitude 

alpine site. Hydrological Processes 13: 1781–1797. 

Legates DR, Yang D, Quiring S, Freeman K, Bogart T. 2005. Bias adjustment to arctic 

precipitation: a comparison of daily versus monthly bias adjustments. 8th Conference on Polar 

Meteorology and Oceanography, San Diego, CA, American Meteorological Society, 6 pages. 

Li L, Pomeroy JW. 1997. Estimates of threshold wind speeds for snow transport using 

meteorological data. Journal of Applied Meteorology 36: 205–213. 

228

NCDC. 2006. National Climate Data Center. National Climate Data Center, Ashville NC, 

, page last updated June 19, 2006. 

Pomeroy JW, Gray DM, Landine PG. 1991. Modelling the transport and sublimation of blowing 

snow on the prairies. In Proceedings of the 48 th Eastern Snow Conference, Guelph, Ontario, p 

175–188. 

Sverdrup HU. 1936. The eddy conductivity of the air over a smooth snow field. Geofysiske 

Publikasjoner 11(7): 69pp. 

Yang D, Goodison BE, Ishida S. 1998a. Adjustment of daily precipitation data at 10 climate 

stations in Alaska: Application of World Meteorological Organization intercomparison results. 

Water Resources Research 34(2): 241–256. 

Yang D, Goodison BE, Metcalfe JR. 1998b. Accuracy of NWS 8" Standard Nonrecording 

Precipitation Gauge: Results and Application of WMO Intercomparison. Journal of 

Atmospheric & Oceanic Technology 15: 54–68. 

229


230

ABSTRACT 

231 



Shaped Solution Domains for Snow Properties 

RAE A. MELLOH, 1 SALLY A. SHOOP, 2 AND BARRY A. COUTERMARSH 2 

The objective of this work was to develop a method for distributing snow properties on a 

landscape that is less dependent on extensive mass and energy balance modeling, yet provides a 

realistic distribution of snow depth and density across the environmental gradients of elevation, 

slope, azimuth, and forest type. In this paper, we present important progress toward development 

of such an approach. A shaped solution domain was identified for snow depth, water equivalent, 

and density. The pinched-cone shape describes the differentiation of the snow properties with 

slope and azimuth and is approximated by an analytical equation with only two coefficients. 

Knowledge of the solution domain shape permits a few model runs or measurements to be 

exploited to define a continuous solution for all slope–azimuth combinations. The shaped 

solutions morph over time and environmental gradients. 

Keywords: Shaped solution domain, Snow cone, Snow distribution, Snow properties 

INTRODUCTION 

This research effort arose from a requirement for distributed snow depth and density for use in 

high-resolution ground vehicle mobility assessment models used by the U.S. Army (Shoop et al. 

2004, Richmond et al. 2005). The requirement was for a realistic distribution of snow depth and 

density across the environmental gradients of elevation, slope, azimuth, and forest type. An 

approach that is widely applicable and that decreases the dependence on extensive snow property 

modeling is needed because the computational and data storage aspects of the mobility model 

cannot be overburdened by snow process computations and snow solution storage. Yet, the 

distribution of snow properties across a landscape is a complex problem that requires distribution 

of mass and energy balance in a high-resolution (30-m) landscape. 

One way to visualize this problem is to determine the likely snow depth at a remote point, given 

a snow depth at a point on a landscape (Fig. 1). The known point might be where we are standing, 

at a low elevation and on a south-facing slope. The point in question may be across the next ridge 

on a north-facing slope and at higher elevation. We know there is likely to be more snow at the 

remote point, but we do not know how much more. A method for making this extrapolation to an 

unknown point (or every point in the landscape) without resorting to extensive calculations would 

be quite useful. 

1 Correspondence to Rae A. Melloh, U.S. Army Engineer Research and Development Center, 

Cold Regions Research and Engineering Laboratory, 72 Lyme Road, Hanover, NH USA 03755- 

1290, e-mail rmelloh@crrel.usace.army.mil. 

2 U.S. Army Engineer Research and Development Center, Cold Regions Research and 

Engineering Laboratory, 72 Lyme Road, Hanover, NH USA 03755-1290.

In this paper, we present important progress toward development of such an approach. We have 

found that the solution domains for snow depth, snow water equivalent, and snow density have a 

predictable shape that can be approximated by an analytical equation with only two coefficients. 

Figure 1. Snow depth across a landscape showing shallow snow (dark) at south-facing point A 

and deeper snow (light tone) on north-facing slope and higher elevation point B. 

BACKGROUND AND APPROACH 

The approach taken was to investigate snow properties as a function of slope, azimuth, 

elevation, and forest canopy transmissivity using a snow model that has a proven record of 

accurately representing snowpack processes. The model solution domain was analyzed for 

structure in snow property differentiation without mapping to any particular landscape. 

SNTHERM (Jordan 1991), SHAW (Flerchinger et al. 1994), UEB (Tarboton and Luce 1996), and 

ISNOBAL (Marks 1999) are examples of physically based models that accurately simulate mass 

and energy transfer in snowpacks. The complexity and computational intensity of the models and 

meteorological driver requirements usually restrict their use to point simulations or small research 

drainage basins (Marks 1999). With the exception of ISNOBAL, these models are not intended for 

explicit mapping (pixel by pixel). We chose SNTHERM (Jordan 1991) because of past experience 

with the model (Melloh et al. 2004) and used a recent update (SLTHERM) that includes mass and 

energy transfer between the snow and soil. 

We designed a method that would be broadly applicable to a hilly, forested landscape in New 

Hampshire or Vermont similar to the landscapes of the Sleepers River Research Watershed near 

Danville, Vermont; Hubbard Brook Experimental Forest near Thornton, New Hampshire; and the 

Ethan Allen Firing Range near Jericho, Vermont. The first planned application of the method 

developed here is for a high-resolution mobility model of the Ethan Allen Firing Range. Hubbard 

Brook, Sleepers River, and Ethan Allen all have similar climate, topography, and elevation ranges 

from approximately 200 to 1000 m. The tree species are predominantly northern deciduous 

hardwoods, including sugar maple (Acer sacharum), beech (Fagus grandifoia) and yellow birch 

(Betula allegheniensis). White ash (Fraximus americana) is found at middle and lower elevations. 

Red spruce (Picea rubens), balsam fir (Abies balsamea), and white birch (Betula papyrifera var. 

cordifolia) occur at the higher elevations and on rock outcrops. Hemlock (Tsunga canadensis) is 

found along the streams. 

Solar radiation, when calculated as a function of slope and azimuth and visualized in an array, 

displays a gradually changing symmetrical pattern across the array and with calendar progression 

(Fig. 2). It seemed plausible that snowmelt differentiation due to slope and azimuth would also 

follow a tractable pattern. Slope–azimuth combinations were selected by examining plots of clearsky 

solar radiation over the snow season (Fig. 2). South- and north-facing exposures were 

emphasized over east- and west-facing exposures in the selection of slope–azimuth combinations 

because the magnitude of daily solar radiation varies less with terrain slope for east and west 

exposures. Modifications were made to the base meteorological data to drive the snowpack energy 

232

and mass balances for 540 environments: the product of 27 slope–azimuth combinations, five 

forest canopy transmissivities, and four elevations. 

55o 55o 0o 0o Oct Oct 

Nov Nov 

Dec Dec 

N 

233 

Jan Jan 

E 

Feb Feb 

Figure 2. (Top) Relative clear sky radiation for the months of October through February plotted for terrain 

slopes of 0° to 55° for each month, and for all azimuths. (Bottom) SLTHERM solutions were obtained for the 

center points of these 27 slope–azimuth regions. 

METHODS 

Distributing the snowpack energy balance 

Terrain elevation, slope, azimuth, and forest canopy greatly influence the mass and energy 

balance of the snowpack and the spatial distribution of snowpack properties. Our approach to 

distributing a mass and energy balance in the hilly, forested terrain of New England was to modify 

meteorological data measured at an open site so that it represents the meteorological gradients 

imposed by elevation, slope, azimuth, and forest cover. The energy balance of a snowpack may be 

expressed as 

Q + Q = Q + Q + Q + Q + Q + Q 

(1) 

M Δ K L E H P G 

where 

QM = snowmelt, 

Q� = change in stored heat, 

QK = solar (or shortwave) radiation, 

QL = terrestrial (or longwave) radiation, 

QE = latent heat transfer (evaporation and condensation), 

S 

W 

N

QH = sensible heat transfer, 

QP = heat advected by rainwater, and 

QG = conduction of ground heat. 

South-facing slopes receive more solar radiation (Fig. 2). Forest canopies transmit only a 

percentage of the above canopy radiation to the snowpack; decrease longwave losses under cold, 

clear sky conditions; emit longwave radiation; and reduce wind speeds that reduce latent and 

sensible heat transfers. Precipitation and temperature lapse rates combine to accumulate more 

snow at higher elevations. 

The meteorological data used were collected by the Forest Service at the Hubbard Brook 

Experimental Forest (HBEF) at 252-m elevation in an open field during the fall of 2002 and 

winter of 2003 (USDA 2004). Meteorological variations were created from the original base 

meteorology for 540 combinations of 27 slope–azimuths, four elevations (300, 500, 700, and 900 

m) and five canopy solar transmissivities (0.14, 0.3, 0.5, 0.75, and 1.0). Canopy transmissivity 

(Tr) is a continuous variable rather than discrete; however, most forest cover maps are categorical 

and these discrete values were chosen to represent conifer, mixed conifer-deciduous, deciduous, 

sparse canopy, and no canopy, respectively. 

Solar and terrestrial radiation 

Solar radiation on slopes was calculated within SLTHERM using a cosine correction for 

illumination angle. Subcanopy shortwave radiation (KSC) was calculated from solar radiation 

measured in the open (KIN): 

KSC IN 

= K Tr . (2) 

Reflected solar radiation was calculated using a constant albedo of 0.78: 

K = 0. 

78 K . (3) 

SCout 

SC 

Terrestrial radiation under the forest canopy (LINforest) was calculated as 

INforest 

4 

( − Tr) 

TempK ( Tr L ) 

L = 1 ε + 

(4) 

IN 

where emissivity (ε) of the forest is taken as 0.96, the effective temperature of the forest is 

estimated by the air temperature (TempK), and σ is the Stephan–Boltzman constant (5.67 × 

10 –8 Wm –2 K –4 ). LIN is terrestrial radiation in the open (no canopy) and is calculated for three 

conditions: 100% clear (LINclear), 100% cloudy (LINcloud), and partly cloudy (LINpart). 

L TempK 

4 

INcloud = εσ . 

LINclear INcloud 

= L − A + OFF . (6) 

0.5 0.5 

( VP VP ) 

A=− 0.4842 [228 + 11.6 S − A ] . (7) 

LINpart INcloud 

( A − OFF) 

= L − CI 

. (8) 

Equations 5, 6, and 7 were adapted from Anderson and Baker (1967) with a locality offset 

(OFF) determined for a site near Danville, Vermont (Melloh et al. 2004). SVP and AVP are 

saturation and actual vapor pressures. The clear-sky index (CI) is the ratio of observed solar 

radiation KIN to clear-sky radiation KCS calculated by methods presented by Dingman (1993). 

234 

(5)

Temperature, precipitation, windspeed 

An average temperature lapse rate of –6.5 °C over 1000-m elevation rise was adopted. 

Precipitation gage undercatch was corrected in the base meteorology by adding between 0 to 41 

percent of the measured precipitation over the windspeed range of 0 to greater than 7 ms –1 for a 

Universal gage with an Alter shield (Goodison 1978). A precipitation multiplier (Pm) applied with 

increased elevation (E) 

P 

M 

⎛E−252 ⎞ 

= 1+ ⎜ ⎟0.75 

⎝ 1000 ⎠ 

was adapted from Dingman’s (1988, 1993) study of mean annual precipitation–elevation trends in 

New Hampshire and Vermont. Precipitation interception loss (PI) was expressed in terms of 

canopy transmissivity (Tr) as 

I 

( Tr) 

PM 

P = 1 − 

(10) 

and PI was limited to a maximum of 

PIm ax 

( ) 3 

1− 

= 0. 0005 Tr . (11) 

Precipitation type was assumed to be rain if the temperature was above 275.5 K, snow below 

273.2 K and a mix of snow and rain between these two temperatures (Jordan 1991). The initial 

snow-grain diameter was set at 0.0000254 m. The results of using Equations 9, 10, and 11 (Fig. 3) 

are increased precipitation (rain and snow) and increased snowfall with increased elevation. There 

is also decreased precipitation and snowfall under fuller canopies. The larger difference between 

the 300- and 500-m elevation snowfalls indicates a snowline occurs, at times, below the 500-m 

elevation (Fig. 3). 

A wind lapse rate was set at 15% increase for each 200-m elevation gain. Wind modification by 

canopy was an adaptation of an equation suggested by Dunne and Leopold (1978) 

( Tr) 

W = w [ 1− 

0. 

8 1− 

(12) 

where w is wind measured in the open and W is wind in the subcanopy. 

Precipitation (m) 

0.6 

0.4 

0.2 

0 

300m 

500m 

700m 

900m 

conifer mixed decid other open 

235 

Snowfall (m) (m) 

0.4 

0.3 

0.2 

0.1 

0 

(9) 

conifer mixed decid other open 

Figure 3. Precipitation and snowfall values are water equivalent summed over Julian days 300 to 110. (Left) 

Increased precipitation with elevation and increased interception with increased canopy fullness. (Right) 

Increased snowfall with elevation, increased interception with increased canopy fullness, and snowline 

development below 500 m. Only snowfalls for temperatures below 273.2 K are included.

SLTHERM Model initiation 

Snowpack properties for the 540 meteorological sub-environments were modeled with 

SLTHERM. SLTHERM is a new version of SNTHERM (Jordan 1991) that has enhanced 

capability to model moisture and energy transfer between the snowpack and soil. SNTHERM is a 

well-tested and internationally known physically based snow process model. The simulation 

period initiated on 27 October (Julian day 300) and continued through snow accumulation and 

finally snowmelt. An initial snow temperature profile was available from Sleepers River Research 

Watershed (SRRW) at 900-m elevation and lapsed at a rate of 6.5 °C over 1000 m to the lower 

elevations. 

Snow depth (m) 

1 

0.8 

0.6 

0.4 

0.2 

0 

300 320 340 360 15 35 55 75 95 115 

Julian day 

Figure 4. The modeled snow depth time-series for each of the 27 slope–azimuth cases for sparse forest. 

The top line is the most north-facing, and the bottom line the most south-facing modeled time series. 

RESULTS 

Shaped solution domains for snow depth 

The snow model solutions (Fig. 4) for sparse canopy at 900-m elevation for each of the 27 

slope–azimuth cases illustrate that the snow depths differentiate with time following snowfall 

events, and that the differentiation becomes more pronounced in late winter and spring as the 

snowpack ablates. The slope–azimuth and canopy dependence of solar radiation is the driving 

force behind snow property differentiation. On day 60 the snowpack has differentiated mildly with 

azimuth and terrain slope (Figs. 4, 5), by day 82 additional snow has fallen and further 

differentiation has occurred, and on day 110 the snowpack is shallow and highly differentiated. 

The 540 model solutions generated output for 20 time-series plots such as Figure 4, one for each 

of the five forest and four elevation combinations. 

A three-dimensional (3-D) solution domain in the shape of a pinched cone is suggested by the 

terrain slope versus snow-depth plot (Fig. 5) if one visualizes the bold lines (east-facing azimuths) 

to be in the foreground and the dashed lines (west-facing azimuths) to be in the background. The 

cone vertex position along the x-axis (Fig. 5) represents the snow depth for the no-slope case and 

the vertex translates to the right along the x-axis in response to new snow (day 60 to day 82) or to 

236

the left in response to snow compaction or ablation (day 82 to day 110). The increased radius of 

the cone indicates increased snow depth differentiation by slope–azimuth with increased terrain 

slope. Elongation along the north-south axis of the cone (Fig. 6) occurs because snow depth 

differentiates more for north- to south-facing (x-axis) than for east- to west-facing slopes (y-axis). 

A pinched-cone shape that approximates the solution domain (Fig. 6) is given by the equation 

2 2 2 

x + y x 

z = − 

(13) 

2 

2 

a b 

where x and y are Cartesian coordinates that determine the radius 

r + 

2 2 

= x y 

(14) 

and where the radius represents the difference in the snow depth from the 0° terrain slope case. 

The slope azimuth (θ) direction is given by a radial coordinate system about the z-axis (Fig. 6). 

Coefficients a and b satisfy the equations 

a = 

y 

z 

when x = 0 and y = minimum radius (15) 

−1 

⎛1z⎞ b = ⎜ − ⎟ when y = 0 and x= maximum radius. 

⎝a x ⎠ 

(16) 

Terrain slope (degrees) 

25 

20 

15 

10 

5 

0 

18 o 

342 o 

Day 110 

54 o 

126 o 

234 o 

306 o 

90 o 

270 o 

162 o 198 o 

237 

Day 60 

Day 82 

0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 

Snow depth (m) 

Figure 5. Modeled snow depths (x-axis) for days 60, 82, and 110 plotted against terrain slope. The circles 

on day 110 indicate the snow model solutions. Bold lines are east-facing azimuths and dashed lines are 

west-facing azimuths. The 18° azimuth is the most south-facing and 198° the most north-facing.


-15 

-10 

S 

-5 

E 

0 

θo θo θo θo θo r 

ΔΔ ΔΔ Δ snow snow snow snow snow depth depth depth depth depth (x) (x) (x) (x) (x) 

5 

z 

238 

10 

W 

15 

5 

N 

0 

5 

10 

ΔΔ ΔΔ Δ snow snow snow snow snow depth depth depth depth depth (y) (y) (y) (y) (y) 

Figure6. The squashed-cone shaped solution in Cartesian and cylindrical coordinate systems, 

representing the snow-depth differentiation with slope and azimuth. 

The SLTHERM model solutions plot along the z = 8.5° and z = 20° contours (Fig. 7) of the 

pinched cone and verify that equation 13 approximates the shape of the solution domain. Southfacing 

Δ snow depths are negative and north-facing Δ snow depths are positive (Fig. 6 and 7) and, 

as a result, the shape given by equation 13 must be corrected to pass through r = 0 on the 

minimum axis. The available SLTHERM solutions suggest that an 18° rotation corrects for eastwest 

asymmetry. The long axis is along 18° and 198° rather than directly north-south and the 

minimum axis is along 108 o and 288 o azimuths (rather than directly east-west). The asymmetry 

likely arises because west-facing slopes receive the afternoon sun through a more transmissive 

atmosphere than east-facing slopes that receive the morning sun. A gradual correction to r=0 at the 

108° and 288° azimuths modifies the cone shape within ±18 o of the minimum axis (between 90 o 

and 126 o , and between 270 o and 306 o ) where SLTHERM solutions were not computed (Fig. 7).

Pinched-cone dimensions for each slope–azimuth–forest environment were determined by finding 

the coefficients (a and b) needed to match the SLTHERM solutions for each respective 

environment. Some of the SLTHERM model solutions do not lie as close to the contours as others 

(Fig. 7). Additional model studies at additional azimuths are needed to refine the rotation and 

fundamental shape if necessary, and shorter time steps (< 1hr) may be necessary to obtain more 

consistent results. Equation 13 (with the r=0 at the minimum axis correction) is a reasonable first 

approximation of the solution domain shape. An equation that predicts r (Δ snow depth) as a 

function of z and θ o will be developed for future applications (Melloh et al. in review). 

E 

306o 306o 234o 234o 342o 342o 198o 198o S 

N 

Figure 7. Contour plot of the cone at terrain slopes of z = 20° (outer contour) and 8.5° (inner contour) 

with day 110 sparse forest SLTHERM solutions (circles) shown. Solid circles are for z = 20° and open circles 

for z = 8.5° terrain slopes. The facing azimuths are indicated on the compass. The Δ snow depths (cm) are 

positive beneath and negative above the east-west axis. An r=0 correction is sketched in at the minimum axis. 

Shaped solution transitions over time and canopy gradients 

The shaped solutions for deciduous forest for days 60, 82, and 110 (Fig. 8 top) present an 

expanding cone over time as the snowpack ablates. The radii of the cone-shaped solutions contract 

progressively across the decreasing canopy transmittance gradient of open to conifer (Fig. 8 

bottom). The cone-shaped solutions over the canopy transition nest neatly inside one another when 

the vertices are co-located. 

239 

18o 18o 162o 162o 54o 54o 126o 126o 0 5 10 15 

15 

Δ snow depth (cm) 

W

deciduous 



20 

2.8 

66-cm 82-cm 41-cm 

1 

6.3 

day 60 day 82 

day 110 

240 

1.2 

9 

1.6 

open open sparse 

sparse deciduous deciduous mixed 

mixed conifer 

conifer 

9 

14 

4 

9 

1.6 

17 

17 32 

32 

41 

41 41 

41 40 

40 

Snowdepth Snowdepth (cm) 

(cm) 

Figure 8. (Top) Snow depth shaped solutions over time in the deciduous forest at 900-m elevation. The snow 

depths for 0° terrain slopes are shown above each cone for days 60, 82, and 110. In both top and bottom, the 

arrows at the vertices represent the maximum (north-south) and minimum (east-west) Δ snow depths (cm) 

relative to the 0° slope case. (Bottom) Snow depth shaped solutions over the forest environmental range on 

day 110 at 900-m elevation. Snow depths for 0° slopes are listed on the x-axis. Corrections on minimum axes 

not shown. 

Shaped solutions for snow water equivalent and density 

Snow water equivalent (SWE) variation with terrain slope and azimuth (Fig. 9) displays 

similarly to snow depth variation (Fig. 5). The SWE solution domain can also be approximated by 

the pinched-cone equation (Fig. 10). A snow density shaped solution was calculated by dividing 

the axes dimensions of the snow water equivalent cone by those of the snow depth cone (Fig. 10). 

DISCUSSION 

Shaped solution domains for snow properties are a way to extend what we know about snow 

property differentiation with solar radiation and forest cover. The snow cone is an interpolation 

surface; a few model runs or measurements are exploited to define a continuous solution. The 

pinched-cone shape can be defined by three model runs: 1) the no-slope case, 2) a slope–azimuth 

combination that defines the maximum diameter of the cone, and 3) a slope–azimuth combination 

5 

1.5 

2 

1



a. 

14-cm SWE 

at 0 

b. 

o a. 

14-cm SWE 

at 0 slope 

b. 

o slope 

-10 

-5 

25 

20 

15 

10 

5.5 

5 

0 

S 

ΔΔ SWE SWE (x) (x) 

1.4 

0 

5 

10 

162 

18 342 

198 

N 

5 

0 

5 

10 

ΔΔ SWE SWE (y) (y) 

54 306 126 

241 

90 

Terrain Terrain slope (degrees) 

100 

270 

40 

450-kg m-3 density 

at 0o 450-kg m 

slope 

-3 density 

at 0o slope 

S 

10 

0 

ΔΔ Density Density (x) (x) 

234 

0.08 0.1 0.12 0.14 0.16 0.18 0.2 

Snow water equivalent (m) 

Figure 9. Modeled snow water equivalent (x-axis) for day 110 plotted against terrain slope. 

The circles indicate the SLTHERM solution points. 

-100 

N 

0 

ΔΔ Density Density (y) (y) 

Figure 10. (Left) Shaped solution for snow water equivalent for day 110 in the sparse forest at 900-m 

elevation. (Right) Shaped solution for snow density derived from the depth and water equivalent cones. 

The arrows at the vertices represent the maximum (north-south) and minimum (east-west) Δ SWE (cm) 

and Δ density (kg m –3 ), respectively, relative to the 0° slope case. The arrows are not to scale. Corrections to 

r=0 on minimum axes are not shown.

that defines a second point on the cone. The three points defining the shaped solution do not have 

to be these specific ordinates because any three points on the cone will define the shape; however, 

points near these extremes are recommended. The fundamental shape of the solution domain can 

be used to help plan snow measurement networks. In the absence of model solutions, 

measurements at a few key locations in a basin might be used to define a cone shape and provide 

an estimate of the snow properties for any slope–azimuth combination of similar canopy and 

elevation. Locations in contrasting canopy or elevation can define the change in cone dimension 

with these environmental gradients. 

Mathematics on the solution domains provides another interpolation route. Our first application 

of the shaped solution domains will be to distribute snow depth and density across the landscape 

of the Ethan Allen Firing Range near Jericho, Vermont, for a vehicle mobility model. Interpolating 

between the shaped solution domains of the 20 slope–azimuth–canopy–elevation environments 

permits creation of a continuous snow property map across the landscape if continuous maps of 

each of the environmental variables are available (slope, azimuth, canopy transmissivity, and 

elevation). The forest map at Ethan Allen Firing Range is categorical, thus our forest component 

interpolation will also be categorical. Other examples of solution domain mathematics include 

snow density computations from snow water equivalent and snow depth cones, and melt rate 

calculations by differencing snow water equivalent cones over time. Equation 13 (with a shape 

correction to r=0 at the minimum axis) is a good approximation of the fundamental shape for our 

intended application. Additional model studies at shorter time steps to improve model consistency 

and potentially refine the azimuths of the major axes are warranted. An equation that predicts r (Δ 

snow depth) as a function of z and θ o will be developed for future applications (Melloh et al. in 

review). 

The shaped solution domains were developed to map snow properties in a hilly forested terrain 

typical of New Hampshire and Vermont and over a relatively small basin. The method is likely to 

be useful in other situations, though adaptations are needed to deal with large basins, topographic 

shading, and blowing snow. The presented method will not be especially useful on plains and 

prairies where the topography is flat, and where there is significant blowing snow. Using a cone 

method over larger basins will require adaptations to account for latitudinal and longitudinal 

meteorological gradients. The cone does not account for topographic shading, an important 

consideration in mountainous basins. Also of interest is how the cone shape changes with 

changing maximum sun angle at lower and higher latitudes. It is also important to recognize that 

the cone solutions represent mean snow properties, and on any given slope in the field there will 

be variability about the mean due to microclimate and local deposition. 

CONCLUSION 

A shaped solution domain was identified for snow depth, water equivalent, and density. The 

pinched-cone shape describes the differentiation of the snow properties with slope and azimuth 

relative to the no-slope case represented by the vertex of the cone. The shaped solution domain is 

approximated by an analytical equation with only two coefficients. A time series of snow 

properties over a snow season can be described efficiently by a time series of three numbers: the 

snow property for the no-slope case, and the two coefficients. Interpolating between the shaped 

solution domains of slope–azimuth–canopy–elevation environments permits creation of a 

continuous snow property map across the landscape if continuous maps of each of the 

environmental variables are available (slope, azimuth, canopy transmissivity, and elevation). 

Knowledge of the shape of the solution domain can be exploited to portray snow property 

differentiation across environments and may lead to less model-intensive approaches of estimating 

snow properties across a landscape. 

242


This work was funded by the U.S. Army AT42 High-Fidelity Ground Platform and Terrain 

Mechanics Modeling Program. The meteorological data used in this publication were obtained by 

scientists of the Hubbard Brook Ecosystem Study; this publication has not been reviewed by those 

scientists. The Hubbard Brook Experimental Forest is operated and maintained by the 

Northeastern Forest Experiment Station, U.S. Department of Agriculture, Radnor, Pennsylvania. 

We thank Susan Frankenstein and Paul Richmond for their reviews of this manuscript. 

REFERENCES 

Anderson EA, Baker DR. 1967. Estimating incident terrestrial radiation under all atmospheric 

conditions. Water Resources Research 3(4): 975–988. 

Dingman SL. 1988. Application of kriging to estimating mean annual precipitation in a region of 

orographic influence. Water Resources Bulletin 24(2): 329–339. 

Dingman SL. 1993. Physical Hydrology. Englewood Cliffs, NJ: Prentice Hall, 575 pp. 

Dunne T, Leopold LB. 1978. Water in Environmental Planning. San Francisco, CA: W.H. 

Freeman and Co. 

Flerchinger GN, Cooley KR, Deng Y. 1994. Impacts of spatially and temporally varying snowmelt 

on subsurface flow in a mountainous watershed: 1. Snowmelt simulation. Hydrological 

Sciences 39(5): 507–520. 

Goodison BE. 1978. Accuracy of Canadian snow gage measurements. Journal of Applied 

Meteorology 27: 1542–1548. 

Jordan R. 1991. A one-dimensional temperature model for a snow cover: Technical documentation 

for SNTHERM.89. U.S. Army Cold Regions Research and Engineering Laboratory, Special 

Report 91-16. 

Marks D, Domingo J, Susong D, Link T, Garen D. 1999. A spatially distributed energy balance 

snowmelt model for application in mountain basins. Hydrological Processes 13: 1925–1959. 

Melloh RA, Richmond P, Shoop S, Coutermarsh B, Affleck R. in review. Continuous snow 

property mapping for mobility models. Cold Regions Science and Technology. 

Melloh RA, Hall TJ, Bailey RN. 2004. Radiation data corrections for snow-covered sensors: Are 

they needed for snowmelt modeling? Hydrological Processes 18: 1113–1126. 

Richmond, P., Reid A, Shoop S, Mason G. 2005. Terrain Surface Codes for an All-Season, Off- 

Road Ride Motion Simulator, MSIAC M&S Journal On-Line. 

Shoop S, Coutermarsh B, Reid A. 2004. All-Season Virtual Test Site for a Real-Time Vehicle 

Simulator, Paper No. 2004-01-2644, SAE Transactions 113(2): 333–343. 

Tarboton DG, Luce CH. 1996. Utah Energy Balance Snow Accumulation and Melt Model (UEB) 

Computer Model Technical Description and Users Guide. Utah Water Research Laboratory 

and USDA Forest Service Intermountain Research Station: 64 p. 

U.S. Department of Agriculture, Forest Service, Northeastern Research Station, Hubbard Brook. 

2003. Web page (http://www.hubbardbrook.org), GIS coverages, and overview (guidebook). 

243


244

Glacial and Periglacial 

Processes 

245


246

247 



Quantifying the Effect of Anisotropic Properties in Snow 

for Modelling Meltwater Retention 

INTRODUCTION 

C. E. BØGGILD 1 

In cold snowpacks of the Arctic meltwater retention is a significant factor for timing and 

magnitude of runoff. Despite being an important component in Arctic hydrology and glacier mass 

balance the meltwater infiltration and subsequent re-freezing in cold snow is rather little quantified 

in literature. The quantity of super-imposed ice (SI) has previously been modeled in onedimensional 

vertical profiles. But, since these approaches rely on few points the spatial 

distribution of SI still needs to be determined more precisely. The problem of moving interface 

associated with SI growth calls for high precision in numerical models, which again greatly 

enhance the computational demand. This computational demand can likely be the reason for the 

lack of SI treatment in most cryospheric models. 

Bøggild (submitted) proposes an analytical solution to the ice warming, from which the 

temporal SI-quantity can be derived directly without comprehensive numerical modeling. 

However, this solution is only valid for isothermal and isotropic condition, which likely never 

occurs in nature. Here is presented the result from statistical fitting of thermal gradients and 

thermal diffusivity to the resulting effect on meltwater refreezing by SI formation. The purpose is 

to extend the existing analytical solution to anisotropic condition also. And, since the theory 

behind an analytical solution is only valid for isotropic condition the approach here has to rely on 

statistical methods. 

METHODS 

Bøggild et al (2005) found that for isothermal and isotropic condition the growth rate of SI is: 

H ( t) 

= 

2γ 

αt 

where H is SI, t is time α and γ are constants. With a temperature gradient and anisotropic 

conditions SI is modelled by temporal heat diffusion modelling followed by: 

dH 

dt 

[ ρsi 

− ρws( 

1− 

ω ) ] 

L 

= xm 

∂T 

ciρi 

∫ dx 

∂t 

0 

1 

The University Centre in Svalbard (UNIS), PO Box 156, N-9171 Longyearbyen, Norway 

e-mail: carl.egede.boggild@unis.no

where ρ is density, ω is water content, L is Latent heat of fusion, k i is thermal conduction of ice 

and c i is specific heat of ice. 

A numerical model has been developed based on the above equation and description. The model 

is used for analyzing the effect of variable temperature gradients and thermal properties on SI 

formation. Sets of results have been produced, from which gradients can be derived using linear 

regression. 

Effect of variable temperature gradient 

In the analysis of the effect of temperature gradient in snow on the resulting superimposed ice 

formation the aim has been twofold, namely 1) an expression valid for different ‘average’ 

temperatures in the snow and 2) derivation based on realistically temperature gradients as 

expected to occur in nature. Based on 1) and 2) the experiments performed have first been carried 

out with an average temperature of –5 °C and then with an average temperature of –10 °C. As for 

2) the experiments have been based on the following temperature gradients from the ice surface 

and downward of –1, –0.5, –0.25, 0, +0.25, +0.5 °C per meter. Beyond this range of gradients 

values are considered to be highly rare or non-existing in nature for longer time spans. 

Effect of variable thermal diffusivity 

Since the highest thermal diffusivity in a combined ice/snow system occurs with ice only, the 

thermal diffusivity has been set as a fraction of the diffusivity of ice. These values have been 

inserted into the numerical model and resulting output are presented in the next section. 

RESULTS 

A set of results was produced using the a –5 °C ice/snow interface temperature but changing the 

temperature gradient inside the ice. This Approach was repeated for an interface temperature of – 

10 °C (Fig. 1). And, it was found that constants did show invariable with temperature as long, as the 

temperature gradient was accounted for in a separate term. 

Fitting of gradients 

0,00035 

0,0003 

0,00025 

0,0002 

0,00015 

0,0001 

0,00005 

0 

-12 -10 -8 -6 -4 -2 0 

Temperature range 

Figure 1. The change in SI formation as a function of snow/ice-interface temperature and ranges of 

temperature gradients inside the ice. 

248

From linear regression the constants could be determined. The resulting expression with a 

variable temperature gradient is: 

Accounting for thermal diffusivity did prove more complex because variable diffusivity does 

not relate linearly to H. Instead the logarithm of the ratio between the adjusted thermal diffusivity 

to the diffusivity of ice(Diff/Diff ice ) did prove invariable with H. Fig 2 shows this relation for 

T ice = –10 °C. 

For T(ice)=-10 °C 

0,0007 

0,00065 

0,0006 

0,00055 

0,0005 

0,00045 

0,0004 

0,00035 

0,0003 

-1,5 -1 -0,5 0 

Figure 2. The change in SI formation as a function of the property Diff/Diffice The resulting expression with 

variable diffusivity becomes. 

When combining the contribution from temperature gradient and thermal diffusivity, 

respectively, the final equation becomes: 

CONCLUSIONS 

⎛ dT ⎞ 

H ( t) 

= ⎜k 

2 − Tsk1⎟ 

⎝ dx ⎠ 

⎛ ⎛ ⎜ ⎛ ⎛ 

f 

⎞ 

H t = ⎜Ts⎜ 

⎜ ⎜κ 

⎟ 

( ) k1− 

k 

⎜ ⎜ 

⎜ 3ln⎜ 

⎟ 

⎜ ⎜ ⎜ κi 

⎟ 

⎝ ⎝ ⎝ ⎝ ⎠ 

t 

⎞ 

⎞ 

⎟ 

⎟ 

⎟ 

⎟ 

⎟ 

⎟ 

⎠ ⎠ 

⎞ 

⎟ 

⎠ 

⎛ ⎛ dT ⎜ ⎛ ⎛ 

f 

H t = ⎜k 

Ts 

⎜ 

− ⎜ 

⎜ ⎜κ 

( ) 2 

k1− 

k ⎜ 

⎜ 3ln⎜ 

i 

⎝ dx ⎜ ⎜ ⎜ κ 

⎝ ⎝ ⎝ 

A parameterization as addition to the Neuman solution has been derived valid for temperature 

gradients ranging from –1 to 0.5 K/m. It is believed that gradients exceeding this range do not 

occur over longer time in nature. The constants were found to be robust over a temperature range 

of at least 10 K. 

As a second addition to the Neumann solution the effect of variable thermal diffusivity is 

examined and parameterized. Here a logarithmic term did prove to be the best solution. The 

motivation for deriving the diffusivity term is because the parameterization then also becomes 

t 

249 

⎞ 

⎞ ⎞⎟ 

⎟ ⎟ 

⎟ 

⎟ 

⎟ 

⎟ ⎟ 

⎟ 

⎠ 

⎟ 

⎠⎠ 

⎞ 

⎟ 

⎠ 

t

valid for formation of ice lenses where SI is formed on top of e.g. firn with lover diffusivity. This 

lower diffusivity result in slower SI formation rates, as long as the temperature gradient is 

remaining. 

REFERENCES 

Bøggild, C.E. 1991: En smeltende snepakkes masse- og energifluxe – belyst ved 

beregningsmetoder. Unpublished Thesis. Univ. Copenhagen. 94 pp. 

Bøggild, C.E., 2000 Preferential flow and meltwater retention in cold snow packs in West- 

Greenland, Nordic Hydrology. 31 (4/5), 287–300. 

Bøggild, C.E., Forsberg, R., Reeh, N. 2005. Meltwater retention in a transect across the Greenland 

ice sheet. Ann. Glac., 40: 102–105. 

250

251 



The Equilibrium Flow and Mass Balance of the Taku Glacier, 

Alaska, 1950–2005 

ABSTRACT 

M.S. PELTO 1 , , G.W. ADEMA 2 , M.J. BEEDLE 2 , S.R. MCGEE 2 , 

M.M. MILLER 2 , K.F. SPRENKE 2 , AND M. LANG 3 

The Taku Glacier, Alaska, has advanced 7.5 km since the late nineteenth century, while all other 

primary outlet glaciers of the Juneau Icefield are in retreat. The Juneau Icefield Research Program 

(JIRP) has completed field work on the Taku Glacier annually since 1946. The thickest known 

alpine temperate glacier, it has a maximum measured depth of 1480 m. The Taku is a tidewater 

glacier that formerly calved, but is now advancing slowly over its outwash delta. Velocity 

measured over a twelve-month span and annual summer velocity measurements completed at 

Profile 4 slightly above the ELA from 1950–2004 indicate insignificant variations in velocity 

seasonally or from year to year . The consistency of velocity over the 50-year period indicates that 

in the vicinity of the equilibrium line, the flow of the Taku Glacier has been in an equilibrium 

state. 

Surface mass balance was positive from 1946–1988 averaging +0.42 m/a. This led to glacier 

thickening. From 1988–2005 an important change has occurred the annual balance has been 

–0.17m/a, and the glacier thickness has stopped increasing. 

Field measurements of ice depth and surface velocity allow calculation of the volume flux. 

Volume flux is then compared with the surface balance flux from the accumulation zone 

determined annually in the field. On each profile the mean surface balance flux 1946–2005 is 

positive versus the annually determined volume flux, leading to glacier thickening. From 1950– 

2000, Taku Glacier thickened by 20–30 m at Profile 4, 33 km above the terminus. At this profile, 

the expected surface flux for equilibrium is 5.50 × 10 8 m 3 /a (±10%), while the calculated volume 

flux range is 5.00–5.47 × 10 8 m 3 /a. At Profile 7, 43 km above the terminus, the observed surface 

flux above Profile 7 is 1.90 × 10 8 m 3 /a, and the volume flux range is 1.72–1.74 × 10 8 m 3 /a 

INTRODUCTION 

Taku Glacier is a temperate, maritime valley glacier in the Coast Mountains of Alaska. With an 

area of 671 km 2 , it is the principal outlet glacier of the Juneau Icefield. It attracts special attention 

because of its continuing, century-long advance (Motyka and Post, 1995), while every other outlet 

glaciers of the Juneau Icefield is retreating. Taku Glacier is also noteworthy for its positive mass 

balance from 1946–1988 (Pelto and Miller, 1990), during a period when alpine glacier mass 

balances have been dominantly negative (Dyugerov and Meier, 1997). Finally, it is unique as the 

1 

Nichols College Dudley, MA 01571 

2 

Glaciological and Arctic Sciences Institute, University of Idaho, Moscow ID 83843 

3 

Institute of Geodesy, Universitat der Bundeswehr, Werner Heisenberg Weg, D-85577 

Neubiberg, Germany

thickest alpine glacier yet measured, with a fjord extending 38–48 km upglacier from its terminus 

(Nolan and others, 1995). 

The Juneau Icefield Research Program (JIRP) has completed field work on the Taku Glacier 

annually since 1946 (Miller, 1963; Pelto and Miller, 1990). In this paper we present a data set for 

the Taku Glacier that is unique in its temporal and spatial extent containing: 

1) Multi-year surface transverse velocity data from three profiles, spanning 50 years on one 

profile; 2) Seismic profiling depth data on the same profiles; 3) Centerline longitudinal velocity 

transects from the glacier divide to the ablation zone; and 4) Multi-year surface mass balance data. 

From this data we can determine surface balance and volume balance transfers. This provides a 

field-based quantitative determination of the volume flux at multiple locations and hence 

constraints on the future behavior of the Taku Glacier. 

The glacier is divided into three zones that describe both mass balance and flow dynamics: (1) 

The ablation zone, below the mean annual ELA of 925 m (113 km 2 ), descends the trunk valley 

with no tributaries joining the glacier, and only the distributary tongue, Hole in the Wall, leaving 

the glacier 11 km above the terminus. (2) The lower neve zone, extending from the ELA at 925 m 

to 1350 m, is a zone where summer ablation is significant (178 km 2 ). All the main tributaries 

(Southwest, West, Matthes, Demorest, and Hades Highway) join in this zone. (3) The upper neve 

zone extends from 1350 m to the head of the glacier (380 km 2 ), comprising the principal 

accumulation region for each tributary except the Southwest Branch. Ablation is limited in this 

zone, with much of the summer meltwater refreezing within the firnpack. This results in a unique 

signature in SAR imagery (Ramage et al., 2000). 

The Taku Glacier has been advancing since 1890: It advanced 5.3 km between 1890 and 1948 

(Moytka and Post, 199; Pelto and Miller, 1990). The glacier advanced 1.8 km from 1948–1988, 

and 0.4 km from 1988 to 2003. The advance rate measured by distance is slowing. The rate of 

advance is best assessed in terms of area as the terminus lobe is spreading out on a terminus shoal. 

Motyka and Post (1995) noted that the rate from 1948–1963 was 0.428 km 2 /year, 0.345 km 2 /year 

from 1963–1979 and 0.11 km 2 /year from 1979–1988. The slowing of the advance has been 

attributed to the impedance of the terminus outwash plain shoal (Motyka and Post, 1995), but it 

has also been conjectured as due to the inability of the mass balance to sustain this advance. With 

an AAR of 82, Taku Glacier had a continuously positive mass balance from 1946–1994, that has 

driven the continued advance (Pelto and Miller, 1990). From 1988–2005 mass balance has been 

slightly negative. 

252

DATA COLLECTION 

Figure 1. Location map for Taku Glacier. 

Surface Mass Balance 

JIRP has measured the annual balance of the Taku Glacier from 1946 to 2005 (Pelto and Miller, 

1990; Miller and Pelto, 1999). Glacier annual mass balance is the difference between the net snow 

accumulation and net ablation over one hydrologic year. On non-calving glaciers, such as the 

present Taku Glacier, surface mass balance observations are used to identify changes in glacier 

volume. JIRP has relied on applying consistent methods at standard measurement sites (Pelto and 

Miller, 1990; Miller and Pelto, 1999). Taku Glacier mass balance measurements are similar to 

those used on the Lemon Creek Glacier. The primary difference between the two is the much 

larger extent of the Taku Glacier (671 km 2 ). On the Taku Glacier, JIRP digs 17 test pits at fixed 

sites, monitors the migration of the transient snow line, and locates the final ELA position at the 

end of the balance year (Pelto and Miller, 1990). These measurements of retained accumulation 

are taken during late July and early August and must be adjusted to end of the balance year values. 

253

This is done via daily ablation rates derived through stake measurements and migration of the 

transient snow line (Miller and Pelto, 1999). 

Possible errors for the Taku Glacier mass balance record include sparse density of measurement 

points (1 per 13 km 2 ), extrapolation to the end of the balance year, infrequent measurements of 

melting in the ablation zone, and measurements carried out by many different investigators. 

However, Pelto and Miller (1990), suggest that these sources of error are mitigated by annual 

(since 1946) measurements at 17 fixed locations, using nine years of ablation data to extrapolate 

mass balance in the ablation zone, using a balance gradient derived from the 17 fixed sites and 

known values for the ablation zone that shifts in altitude from year to year based on the ELA, and 

through supervision of field work by at least one experienced researcher (i.e., Matt Beedle from 

2003–2005). 

The measurement network consists of 17 locations where mass balance has been assessed in test 

pits annually since 1946. The majority of the snowpits are in the region from 950–1400 m. In 

1984, 1998 and 2004, JIRP measured the mass balance at an additional 100–500 points with 

probing transects in the accumulation area to better determine the distribution of accumulation 

around the snowpit locations. Measurements were taken along profiles at 100–250 m intervals. 

The standard deviation for measurements sites within 3 km, with less than a 100 m elevation 

change, was ±0.09 m/a; this indicates the consistency of mass balance around the snowpit sites. 

Another possible source of error is the assumption that the density measured at test pits is 

representative of a larger area. However, a study at 40 points within 1 km 2 at different elevations 

in different years resulted in a standard deviation of ±0.07 meters of water equivalent (m w.e.) in a 

snow pack of 1 to 2 m, displaying the highly uniform density of snow on the Taku Glacier (Pelto 

and Miller, 1990). 

An independent check of mass balance is now available in the form of direct measurement of 

the surface elevation of the glacier at specific points. The elevation has been determined annually 

since 1993 at fixed locations along Profile 4 using differential GPS as part of the velocity 

surveying program. GPS annual elevation change measurements along Profile 4 at 1100 m show a 

strong correlation with annual mass balance measurements. This would be expected as elevation at 

the mean ELA is likely to rise with increased accumulation during years of positive mass balance, 

and fall with increased ablation during years of negative mass balance. For the period 1993 to 

2004, correlation between the average surface elevation change of a 31-point profile across the 

Taku Glacier and net mass balance is 0.77 (95% significance). This provides an independent 

validation for Taku Glacier record (Table 1). 

GPS Survey Methods 

Standard rapid-static and real-time differential GPS methods have been employed for all survey 

work from 1996–2004. A key objective of the surveying program is to collect data that allows 

quantitative comparison of surface movements and surface elevation change from year to year. In 

order to ensure the consistency of year-to-year movement and elevation data, all survey flags are 

located within one meter of the standard point coordinates (Lang, 1997; McGee, 2000). After 

establishment of each survey profile is complete, each profile is surveyed two times, with the time 

differential between the surveys ranging from 6 to 9 days. For all surveys, a reference receiver is 

centered and leveled over an appropriate bedrock benchmark. A roving receiver is mounted on an 

aluminum monopole inserted into the same hole that the survey flag is placed. The height above 

the snow surface of the antenna is noted. For rapid-static work, the roving receiver collected 

readings at 15-second intervals for 10 to 20 minutes at each flag. Real-time methods require only 

enough time at each flag sufficient to obtain a position fix from the reference receiver. 

254

Table 1. Comparison of GPS measured height change on Profile 4 and the mean annual balance of the 

Taku Glacier. The GPS data is strictly in height, and the field measurement in meters of water 

equivalent. 

Year Taku bn Pro 4 Ht. Ch. 

1994 0.09 0.09 

1995 –0.76 –1.33 

1996 –0.96 –0.67 

1997 –1.34 –0.46 

1998 –0.98 –1.06 

1999 0.4 0.59 

2000 1.03 1.42 

2001 0.88 0.9 

2002 0.45 –0.7 

2003 –0.9 –1.64 

2004 –0.23 0.63 

2005 0.02 –0.29 

Correlation = 0.77 

The major focus of the survey program is to continue the annual survey of standard movement 

profiles on the Taku Glacier and its main tributaries. In this study we focus on transects 4 and 7. 

Profile 4 has an upper line and a lower line separated by 0.5 km. In 1999, 2000 and 2004, a 

longitudinal profile down the centerline of the Matthes Branch and the Taku Glacier, from the 

glacier divide (located 65 km from the terminus), down to 24 km above the terminus. The surface 

velocity was observed at survey locations spaced 0.5 km apart. The surface slope was determined 

between each survey location. 

Seismic Methods 

Seismic methods are required to determine ice depths on transverse profiles because of the 

thickness of the Taku Glacier (Nolan and others, 1995). The seismic program completed 

measurements of ice thickness along eight transects across the glacier, each following the same 

transects and using the same points used in the GPS movement and elevation surveys. 

The seismic methods used for determining ice thickness are typical. A Bison 9024 series 

seismograph was used with 24 high-frequency (100 Hz) geophones to record the seismic signals 

produced by explosive charges. The geophones were spaced at 30 meter intervals along profile 

lines that are perpendicular to the direction of glacier flow, covering 690 meters with each 

geophone spread. Explosive detonations (shots) were generally made at 500 meter intervals from 

each end of the geophone spread, to a maximum distance of 2000 meters. Shot and geophone 

locations were surveyed using standard differential GPS surveying techniques, accurate to ±5 cm. 

Up to twelve shots were taken on each profile, with up to four reflectors evident on each shot's 

record. The seismograph was normally set to record two seconds of data, recording at a 0.25 ms 

sampling rate. The energy for a shot was produced by 4 to 20 sticks of Kinepak (1/3 stick) 

explosive (ammonium nitrate and petroleum distillate combination), buried approximately 1 meter 

deep in the firn. 

Reflections from the glacier bed were generally clear and easy to recognize on the records by 

their frequency, character, and distinct moveout times. Migrations were completed using the 

common-depth-point technique described in Dobrin (1960) and adapted by Sprenke and others 

(1997). Calculations were made using a constant ice velocity of 3660 m/s, a value determined 

from P-wave first arrival times. The migration and geomorphic profiling process is based on the 

simplifying assumption that the glacier cross-sections are two-dimensional. The results on Profile 

4 match closely the results of Nolan and others (1995) with a maximum depth of 1450 m in this 

study versus 1400 m. 

255

DATA OBSERVATIONS 

Mass Balance 

The annual balance record shows a markedly positive trend from 1946–1988 period. The mean 

annual balance for the 42-year period is 0.42 m/a, representing a total thickening of 17.5 m w.e or 

20 m in total ice thickness. From 1988–2005 mass balance changed significantly and has been 

slightly negative averaging –0.17 m/a, a 3 m w.e. loss, or 3.3 meters of glacier thickness change 

(Figure 2). A comparison of surface height change along Profile 4 from 1993–2004 and mean 

surface balance yield –3.5 m and –2.7 m respectively. The next task in this research program is to 

determine the height changes at survey location along additional profiles over an extended period, 

thus providing additional surface balance data each year. 

m we 

20 

18 

16 

14 

12 

10 

8 

6 

4 

2 

0 

-2 

1946 

1949 

1952 

1955 

1958 

1961 

Juneau Icefield Mass Balance 

Taku Cumulative bn Taku bn 

1964 

1967 

1970 

1973 

1976 

Year 

Figure 2. Annual and cumulative mass balance of the Taku Glacier 1946–2005. 

With the distribution of annual balance more accurately mapped in the lower neve section of the 

glacier based on the 1998, 2003 and 2004 combined snowpit and probing measurements, mean 

annual surface flux was determined both for the glacier region above Profile 4 and Profile 7, 

summing the products of the area observed between each 0.2 m mass balance interval and the 

annual balance for that interval. The surface flux accumulating above at Profile 4 was calculated to 

be 5.5 × 10 8 m 3 a –1 , and 1.90 × 10 8 m 3 a –1 above Profile 7. 

256 

1979 

1982 

1985 

1988 

1991 

1994 

1997 

2000 

2003 

2 

1.5 

1 

0.5 

0 

-0.5 

-1 

-1.5

Transverse Velocity Profiles 

Surface velocity has been constant over a 50-year period on the Taku Glacier at Profile 4 (Table 

2) (Miller, 1963; Dallenbach and Welsch, 1993; Lang, 1995 and 1997 and McGee, 1998 and 

2000). In addition, velocity shows no significant variations seasonally, based on year round 

measurement of the movement of the top of a glacier borehole and the associated semi-permanent 

camp from 1950–1953 along Profile 4 (Miller, 1963). Movement of a meteorologic station 

instrument that endured from 1997 to 1998 on Profile 4 provided a second measure of mean 

annual velocity. In the former case mean annual velocity was 0.60 m/day, and in the latter case 

0.61 m/day. The mean observed velocity for these same locations during the summer is 0.61 

m/day. A magnet was buried in the glacier in July 2003 and then checked in July 2004 to identify 

annual glacier velocity at Profile 4. The annual velocity was 0.587 m/day and the summer velocity 

was 0.588 m/day. Observations of velocity at specific stake locations reoccupied each summer 

along Profile 4 indicate remarkable uniformity of flow from year to year (Table 3; Figure 3). 

Standard deviation ranges from 0.010–0.020 m/day. Along Profile 7 the variation is slightly less, 

though spanning the same range. 

Table 2 Mean summer surface velocity measured in the center of Profile 4 from 1950–2004. Survey 

locations are identical from 1996–2004. The velocity for the other years is the mean of locations 

between the current locations of Point 10 and Point 24. 

Year 

Velocity 

(m/day) 

1950 0.58 

1952 0.51 

1953 0.51 

1960 0.53 

1967 0.52 

1986 0.52 

1987 0.51 

1996 0.53 

1997 0.53 

1998 0.52 

1999 0.53 

2000 0.55 

2001 0.55 

2004 0.52 

st dev 0.02 

257

Velocity m/day 

0.7 

0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

0 

Taku Glacier Profile 4 Velocity 

2 4 6 8 10 12 14 16 18 20 22 24 26 28 

Point 

258 

1996 

1997 

1998 

1999 

2000 

2001 

2004 

Figure 3. Observed surface velocity along Taku Glacier Profile 4 Lower Line. The lack of significant change 

is evident. 

Most temperate glaciers have a substantial component of glacier sliding that depends on bed 

hydrology, hence displaying seasonal variations. Taku Glacier, however, has exceptionally thick 

ice, and a low basal gradient. The flow law for internal deformation suggests that negligible basal 

sliding is taking place in the accumulation zone (Nolan and others, 1995). The lack of seasonal 

velocity changes noted in this study and the remarkable uniformity in velocity suggest that sliding 

is a minor part of the glacier velocity at Profile 4. It is not reasonable to expect a glacier of this 

size to slow down each fall or winter, and then accelerate to exactly the same speed the following 

summer. This tendency has been noted on four other repeat profiles on the Taku Glacier (McGee, 

2000). 

Table 3. Observed velocity each summer at specific survey locations on Profile 4. Note the nearly 

identical velocity at each location. Mean standard deviation in velocity is below 0.02 m/day. 

Point 1996 1997 1998 1999 2000 2001 2004 st dev 

10 0.42 0.43 0.41 0.43 0.43 0.5 0.46 0.031 

12 0.53 0.53 0.5 0.52 0.53 0.57 0.54 0.021 

14 0.58 0.57 0.56 0.56 0.59 0.61 0.55 0.024 

16 0.59 0.59 0.58 0.58 0.6 0.62 0.57 0.016 

18 0.59 0.59 0.58 0.6 0.61 0.61 0.57 0.015 

20 0.57 0.59 0.56 0.59 0.6 0.59 0.55 0.019 

22 0.53 0.54 0.52 0.54 0.55 0.52 0.54 0.011 

24 0.43 0.43 0.43 0.45 0.45 0.4 0.4 0.021 

mean 0.53 0.53 0.52 0.53 0.55 0.55 0.52 0.013 

Longitudinal Profile 

The variation in velocity along the longitudinal profile from Goat Ridge at 800 m elevation, 12 

km from the terminus, up the main trunk of the glacier, along the Matthes Branch, to the ice divide

is shown in Figure 4. This variation indicates increasing velocity with distance down glacier from 

the divide at Point 94, to Goat Ridge just below the ELA at Point 13. The one notable deviation 

from this pattern occurs where the glacier steepens, causing longitudinal extension. At this point 

Taku Glacier leaves the high plateau and enters the narrower valley of the Matthes Branch. The 

surface velocity increase lags the change in surface slope by several kilometers. The glacier then 

slows under longitudinal compression as the surface slope declines. The velocity along this 

longitudinal profile has been repeated in 2001 and 2004 the maximum velocity change was 0.02 

m/day and the mean change for each point was 0.004 m/day. Again, indicating the annual and 

seasonal consistency of velocity along the glacier and the equilibrium nature of its flow (Figure 4). 

m/day 

1.00 

0.90 

0.80 

0.70 

0.60 

0.50 

0.40 

0.30 

0.20 

0.10 

0.00 

13 

17 

21 

25 

29 

33 

Taku Glacier Longitudinal Velocity Profile 

37 

41 

45 

49 

259 

53 

57 

61 

Point Location 

Figure 4. Comparison of gradient and velocity along a longitudinal profile, Taku Glacier, Alaska. 

Thickness 

The greatest thickness of the Taku Glacier was noted to be 1477 m at Goat Ridge, 22 km above 

the terminus (Nolan and others, 1995). The centerline depth of the glacier remains thicker than 

1400 m at Profile 4, 33 km up glacier, and 1100 m at Profile 7, 44 km upglacier. It is likely that 

the Taku Glacier centerline depth is greater than 1100 m in thickness the entire distance between 

these points, based on the consistency in the velocity increase from Profile 7 to Goat Ridge. The 

minimum basal elevation at Profile 4 is approximately –350 m, and is 300 m at Profile 7. Given 

the relatively uniform changes in slope of the glacier and velocity between the profiles it is likely 

that the fjord threshold is near the mid-point between the locations. 

The increase in slope from Point 54 to Point 50 (Figure 3) and then decrease is the most likely 

location of the sea level threshold. Point 49–52 are 39–40 km from the terminus slope, supporting 

the hypothesis of Nolan and others (1995) that the threshold was from 38–48 km above the 

terminus. The steady increase in velocity with distance below this point and the consistency of 

velocity with time both argue for an equilibrium flow of the Taku Glacier. 

65 

69 

73 

77 

81 

85 

89 

93 

6 

5 

4 

3 

2 

1 

0 

velocity 

Velocity 

gradient

The transverse bed profile at Profile 4 indicates benches on both the east and west sides of the 

glacier (Figure 5). The bench on the east side in an extension of the North Basin that is at the base 

of Taku B and just north of Camp 10. The bench on the west side lacks a clear surface topographic 

connection. Profile 7 lacks any benches and has a much more u-shaped profile. 

Figure 5. Surface elevation of stations along Profile IV and seismically determined bottom topography along 

the profile, Taku Glacier, Alaska. 

CALCULATION OF VOLUME FLUX 

Profile IV 

Figure 5. Surface elevation of stations along Profile IV and seismically determined bottom topography along 

1400 

the profile, Taku Glacier, Alaska. 

1200 

1000 

800 

600 

400 

200 

0 

-200 

-400 

-600 

Elevation (m) 

0 1000 2000 3000 4000 5000 6000 

Distance, looking upglacier (m) 

With direct measurement of surface velocity, ice thickness and width for each increment of 

glacier width on the profiles, the only unknown in determining volume flux is determination of 

depth average velocity. Several points led Nolan and others (1995) to conclude that basal sliding is 

minimal; most importantly, calculation of basal shear stresses yielded values of 125 kPa. We 

determined basal shear stress to be 120–180 kPa along Profile 4, and 75 to 100 kPa along Profile 

7. These values are beyond that at which basal sliding would be anticipated. In addition, the 

consistency in velocity from year to year at each point indicates that there is probably negligible 

seasonal fluctuation in velocity in the accumulation zone. This has been confirmed by annual 

velocity observations. It is not reasonable to expect velocity each summer to be within ±5% if 

seasonal variations in velocity were significant. Following the lead of Nolan and others (1995) and 

Nye (1965) we have assumed the depth averaged velocity is 0.8 the observed surface velocity. 

The mean velocity between each two survey flags is used to represent the average velocity for 

that width increment of the glacier. The mean depth for that width increment from the seismic 

profile is then determined. The product of the width of the increment and depth of the increment 

provide the mean cross-sectional area. The mean surface velocity for each increment is converted 

to a mean depth averaged velocity by multiplying by 0.8. 

260

Table 4. The calculated volume flux at Profile 4. Volume flux is in m 3 /year. Annual values are followed 

by a comparison of the mean volume flux, mean observed surface flux and the difference between 

them. The difference in this case is a positive surface flux. The volume flux is determined from annual 

measurements. The surface flux is the mean for the 1946–2004 period. 

1996 1997 1998 1999 2000 2001 2004 Mean Surface Difference 

4LL 5.25 5.34 5.07 5.38 5.42 5.41 5.20 5.27 5.5 0.20 

4UL 5.33 5.22 5.19 5.23 5.33 5.39 5.17 5.25 5.5 0.23 

The volume flux was determined separately for parallel survey lines along Profiles 4 (upper line 

and lower line) and for the Profile 7 lower line. For profile 4, 33 km above the terminus, the 

expected surface flux was 5.50 × 10 8 m 3 a –1 (+10%), the volume flux range was 5.00–5.47 × 10 8 

m 3 a –1 , with a mean of 5.25 × 10 8 m 3 a –1 for the upper line and 5.27 × 10 8 m 3 a –1 for the lower line. 

This indicates a slightly positive balance and glacier thickening above Profile 4. Glacier 

thickening in the range of 30 m has been suggested for this section of the glacier (Motyka, 

personal communication). For Profile 7, 43 km above the terminus the observed surface flux was 

1.90 × 10 8 m 3 a –1 and the volume flux range was 1.72–1.74 × 10 8 m 3 a –1 , again indicating a slight 

glacier thickening above this profile due to a positive mass balance. The calculated volume flux is 

based on the 1946–2004 average mass balance profile for the glacier and not for a given year. The 

surface flux during recent negative balance years would obviously give a lower surface flux value. 

CONCLUSION 

The results indicate that Taku Glacier has had an equilibrium flow, with no significant annual 

velocity changes in the last 50 years. Furthermore, although seasonal variations had been expected 

(Miller, 1963), observations of velocity throughout the year indicate no seasonal variations, 

probably due to high basal shear stress which prevents sliding. The surface mass balance 

accumulating above each profile in the last half century is in excess of the volume flux through 

each profile. The result, supported by both survey results of JIRP and radio echo sounding by the 

University of Alaska–Fairbanks (Nolan et al., 1995), is glacier thickening. The sustained 

thickening, positive balance, and consistent flux of the 1946–1988 period suggested that the 

glacier terminus would continue to advance (Pelto and Miller, 1990). From 1988–2005 the mass 

balance has been negative, though the volume flux at Profile 4 has not declined appreciably. The 

reduced mass balance, if it continues, along with the proglacial delta and expanding front of the 

glacier Post and Motyka (1995), should lead to a reduction in the advance rate. The significant 

change in glacier mass balance beginning in 1988 is expected to influence the glacier velocity, 

volume flux and eventually the terminus, if it is sustained. Given the slow response time of this 

glacier to climate change, if sustained it would not for sometime. The glacier velocity did not 

change appreciably as the glacier thickened by 10–20 m at Profile 4 and it is expected that it 

would take a thinning of more than this to substantially alter glacier velocity. 

261

REFERENCES 

Daellenbach, K. and W. Welsch. 1993. Determination of Surface Velocities, Strain Rates, and 

Mass Flow Rates on the Taku Glacier, Juneau Icefield, Alaska. Zeitschrift fur Gletscherkunde 

und Glazialgeologie, Band 26, Heft 2, p. 169–177. 

Dobrin, M.B. 1960. Introduction to Geophysical Prospecting, 3rd Edition, McGraw-Hill, New 

York, p 201–254, 630 pages. 

Dyugerov, M.B. and M.F. Meier, 1997. Year to year fluctuations of global mass balance of small 

glaciers and their contributions to sea level. Arctic and Alpine Res., 29: 392–402. 

Echelmeyer, K., M. Nolan, R. Motyka, and D. Trabant. 1995. Ice thickness Measurements of Taku 

Glacier, Alaska, USA, and their Relevance to it Recent Behavior. J. Glaciol., 41 (139) 541– 

552. 

Lang, M. 1997. Geodetic Activities of the 1997 Juneau Icefield Research Program Field Season. 

Open File Survey Report. Juneau Icefield Research Program, Foundation for Glacier and 

Environmental Research, Moscow, Idaho. 110 pp. 

McGee, S. 1996. Geodetic Activities of the 1996 Juneau Icefield Research Program Field Season. 


Environmental Research, Moscow, Idaho. 79 pp. 

McGee, S. 1998. Geodetic Activities of the 1998 Juneau Icefield Research Program Field Season. 


Environmental Research, Moscow, Idaho. 

McGee, S. 2000. Juneau Icefield GPS Movement Profile Coordinates. JIRP Open File Survey 

Report. Juneau Icefield Research Program, Foundation for Glacier and Environmental 

Research. Moscow, Idaho. September, 2000. 54 pp. 

Miller, M.M. 1963. Taku Glacier evaluation report. State of Alaska, Dept. of Highways and 

Bureau of Public Roads, U.S. Dept. of Commerce. 

Miller, M. M. 1997. The Juneau Icefield Research Program and its Surveying Mission. In: 

Geodetic Activities Juneau Icefield, Alaska 1981–1996. Schriftenreihe des Studiengangs 

Vermessungswesen, Universität der Bundeswehr München, Heft 50. p 45. 

Miller, M.M. and M.S. Pelto. 1999. Mass Balance measurements on the Lemon Creek Glacier, 

Juneau Icefield, AK 1953–1998. Geogr. Ann. 81A, 671–681. 

Nye, J.F. 1965. The flow of a glacier in a channel of rectangular, elliptic or parabolic cross 

section. J. Glaciol., 5(41) 661–690. 

Pelto, M. and M.M. Miller. 1990. Mass Balance of the Taku Glacier, Alaska from 1946 to 1986. 

Northwest Science. 64 (3)121–130. 

Post A. and R.J. Motyka, 1995. Taku and Le Conte Glaciers, Alaska: Calving speed control of 

late-Holocene asynchronous advances and retreats. Phys. Geogr., 16; 59–82. 

Poulter, Thomas (1950) The Poulter Seismic Method of Geophysical Exploration. Geophysics, 15( 

2) 181–207. 

Ramage, J.M., B.L. Isacks, and M.M. Miller. 2000. Radar Glacier zones in southeast Alaska, 

USA: Field and satellite observations. J. Glaciol. 46(153), 287–296. 

Sprenke, K.F., M.M. Miller, F. McDonald, C. Haagen, G. Adema, M. Kelly, S. Barbour, and B. 

Caceres. 1997. Geophysical investigation of the Taku-Llewellyn Divide: A NASA Earth 

Systems Field Research Project, Juneau Icefield, Alaska. Juneau Icefield Research Program 

Geophysics Open File Report 97-1. (Moscow, Idaho: Glaciological and Arctic Science 

Institute, University of Idaho). 

Welsch Walter, Martin Lang and M.M. Miller, 1996. Geodetic Activities—Juneau Icefield, Alaska 

1981–1996. Schriftenreihe des Studiengangs Vermessungskunde. Universität der Bundeswehr, 

Heft 50. 

262

263 



An Embedded Sensor Network for Measuring 

Hydrometeorological Variability Within an Alpine Valley 

ABSTRACT 

ROBERT Å. HELLSTRÖM 1 AND BRYAN G. MARK 2 

Conditions of glacier recession in the seasonally dry tropical Peruvian Andes motivate research 

to better constrain the hydrological balance in alpine valleys. Studies suggest that glacial mass 

balance in the outer tropics of the Andes is particularly sensitive to variations between the dry and 

wet season humidity flux. In this context, we introduce a novel embedded network of low-cost, 

discrete temperature microloggers and an automatic weather station installed in the Llanganuco 

valley of the Cordillera Blanca. This paper presents data for distinct dry and wet periods sampled 

from a full annual cycle (2004-2005) and reports on modeled estimations of evapotranspiration 

(ET). The transect of temperature sensors ranging from about 3500 to 4700 m revealed seasonally 

characteristic diurnal fluctuations in up-valley lapse rates that promote up-slope warm air 

convection that will affect the energy balance of the glacier tongue. Nocturnal rainfall dominated 

the wet season. Strong solar forcing dominated during both dry and wet periods, but extreme 

seasonal variations in soil water content and cooler wet season near-surface air temperature 

suggests the importance of considering the process of ET. Estimates of potential ET using the 

widely applied Penman-Monteith FAO model suggest nearly twice as much for the dry period, and 

we attribute this primarily to the five times higher dry period vapor pressure deficit. We ran a 

process-based water balance model, BROOK90, to estimate actual ET, which was nearly 100 

times greater for the wet season. These results reinforce the importance of diurnal cloud cover 

variability in regulating ET in the Peruvian Andes. 

Keywords: tropical, alpine, embedded sensors, evapotranspiration, diurnal, seasonal modeling 

INTRODUCTION 

Tropical Andean glacier recession over the past century has profound local consequences for 

water resources, and motivates further hydroclimatic research (Francou et al., 1997; Hastenrath 

and Kruss, 1992; Mark and Seltzer, 2003, 2005). 

While observations of continued glacier recession exist throughout the Andes, only a few efforts 

have been made to quantify the hydroclimatic changes on a scale most relevant to human impact. 

Previous studies primarily focus on the rates, controls, and flux of glacier melt water directly from 

and within the glaciers (Francou et al., 1995; Ribstein et al., 1995; Wagnon et al., 1999; Wagnon 

et al., 1998), but few studies look at the fate of the water once it leaves the glacier system. 

Moreover, lack of accompanying precipitation and stream discharge data also preclude analyses in 

1 

Bridgewater State College, Department of Geography, Conant Science Building, Bridgewater, 

MA 02325, rhellstrom@bridgew.edu 

2 

The Ohio State University, Department of Geography, 1036 Derby Hall, 154 N Oval Mall, 

Columbus, OH 43210, mark.9@osu.edu

all but a handful of sites. There is an outstanding need to better quantify the volume of 

contributions from all hydrologic components in this dynamic climate on the regional scale where 

humans utilize water resources, and where future management decisions must be made. 

Only initial attempts have been made to account for the regional flux of hydrologic components 

in the valleys of the tropical Andes, utilizing major simplifying assumptions for lack of empirical 

data. It has proven very challenging to maintain regular meteorological observations due to human 

disturbance and technical issues in the extreme environment (Georges and Kaser, 2002). 

Groundwater and evaporation have been assumed negligible in glacierized valleys, given high 

relief and steep slopes (Caballero et al., 2004; Kaser et al., 2003). The hydrological processes in 

these high tropical mountain valley deposits remain understudied, and there is a need for 

systematic research to better understand the impact of runoff down valley by different 

morphological zones (Caballero et al., 2002). 

The hydrologic balance of tropical glaciers at various time scales (Hastenrath and Ames, 1995; 

Kaser et al, 2003; Francou et al., 2000; Francou et al., 2003) has been the focus of recent research, 

but measurements and model simulations of the role of evapotranspiration (ET) in the pro-glacial 

valleys are lacking. The low barometric pressure in high elevations promotes ET, but incident 

solar radiation, wind, humidity, precipitation and temperature are primary regulating components 

of meteorological forcing. Studies of mid-latitude pro-glacial valleys suggest significant 

contributions of ET to the hydrologic balance. For example, Konzelmann et al. (1996) found that 

the type of vegetation and surface texture strongly control the rate of ET at different elevations in 

the Dischma valley, near Davos, Switzerland. Konzelmann et al. concluded that ET in this alpine 

valley was regulated by soil moisture and the physiological behavior of the vegetation. However, 

the influence of soil, vegetation, and local meteorological forcing on ET in tropical pro-glacial 

valleys is largely unverified and assumed negligible. Current ground-based meteorological 

observations in the Peruvian Andes, such as Hardy et al. (1998) and Vuille et al. (2003) do not 

extend to pro-glacial valleys, so it is necessary to develop and deploy a new measurement 

network. ET is largely controlled by a combination of available surface moisture, atmospheric 

vapor pressure, air temperature and wind speed. Vuille et al. (2003) report significant increases in 

near-surface air temperature throughout most of the tropical Andes. Interestingly, the temperature 

increase varies markedly between the eastern and western Andean slopes with a much larger 

temperature increase to the west. 

In this paper, we introduce a new network of monitoring instruments from the northeastsouthwest 

trending Llanganuco pro-glacial valley of the western Cordillera Blanca, and report on 

model estimations of ET. We present data for dry and wet periods from a full annual cycle (2004- 

2005) and report on the sensor network design, including a transect of automatic temperature 

sensors ranging from about 3500 to 4700 m. We applied the Penman-Monteith FAO method of 

estimating hourly potential ET (ET0) during the wet and dry periods. We ran a process-based 

water balance model (Brook90: Federer, 2003) to examine the influence of meteorological forcing 

on ET rates and compare the contributing sources of ET, and made comparisons with the ET0 

estimations. Unlike most research on processes affecting ET, which targets mid-latitude and 

subtropical sites, we present methods and results for a hydroclimatologically sensitive region at 

the tropical-sub-tropical interface, the outer tropics. 

Study area 

The Andean Cordillera Blanca is home to the greatest concentration of tropical glaciers on earth 

(Fig. 1). It is the largest and most northerly mountain range in Peru, trending NW-SE over 130 km 

between 8º–10º S latitude (Ames, 1998) along the Andean continental divide. Most of the 

glacierized area in the Cordillera Blanca discharges northwest via the Santa River, that has the 

least variable monthly runoff of all Pacific draining rivers. 

264

Figure 1. Cordillera Blanca regional map. The Llanganuco is one of several pro-glacial valleys in the region. 

The Llanganuco valley is a classic U-shaped hanging valley tributary draining SW to the Santa 

River (Figs. 2 and 3). Its mouth is flanked by steep walls of the glacially sculpted granodiorite 

bedrock that comprises the Upper Miocene batholithic core of the range (McNulty et al., 1998). 

Summits that border the catchment include Huascarán (6768 m) to the south, highest in Peru, the 

Huandoy massif (6395 m) to the north, and Chacraraju (6113 m) at the NE valley head. The valley 

contains two lakes, and is a well-visited tourist attraction within the Huascarán National Park and 

International Biodiversity Reserve. The road along the long axis of the valley forms one of three 

principal transect routes over the Cordillera Blanca and reaches a high pass at the Portachuelo de 

Llanganuco (4767 m) and it continues down the east side. 

265

3469 

3862 

3850 

3871 

Figure 2. Llanganuco valley and location of weather instruments: note the glaciers north and south of SW-NE 

trending valley. 

Figure 3. SW-looking view of Llanganuco valley from Portachuelo site near the highest elevation iButton 

logger. Note the highland grasses, steep rocky walls, terminus of glacier above the north wall, and Santa 

River valley in the distance. 

266 

3948 

4148 

4559 

4344 

4742

MATERIALS AND METHODS 

Field Measurements 

In July 2004, we installed an automatic weather station (Onset HOBO®) in the Llanganuco 

tributary valley (3850 m.a.s.l.). Instrumentation specifications are available: 

http://www.onsetcomp.com/Products/Product_Pages/weatherstation/weather_station_logger.html. 

Hourly data are being collected in collaboration with the University of Innsbruck including soil 

moisture, soil temperature, air temperature, wind speed and direction, relative humidity, solar 

radiation, and precipitation. 

In June 2005, we installed a set of 9 temperature loggers (iButton Thermochron®) at different 

elevations ranging from 3469 to 4742 m above mean sea-level (Fig. 2). Each logger is powered by 

an internal battery (1 year lifetime) and was programmed to record hourly intervals of air 

temperature. The reported iButton resolution is 0.5°C, accuracy is ±1°C with a range of -40°C to 

+85°C (http://www.maxim-ic.com). Each iButton logger was placed inside a specially designed 

¾” PVC shields that allowed us to embed the sensors in small trees and other inconspicuous 

locations (Figs. 4 and 5). 

Figure 4. HOBO weather station at base of lower lake, looking SW. Note the Polylipus (Queñual) trees and 

grass vegetation cover. The precipitation logger to the right was set up by Kaser, U. Innsbruck. 

267

Figure 5. (a) ¾” PVC ventilated radiation shield for iButton Thermochron logger. Note that the elbow 

sections are detachable for installation in vegetation. (b) Typical installation of iButton temperature loggers, 

one atop 1.5 m PVC post (top) and one embedded at 1.5 m inside a Queñual tree. 

One iButton was calibrated under actual field conditions against the air temperature from the 

HOBO weather station (Fig. 6). The calibration is within the tolerance of the iButton sensors 

(approximately 3%). Every two months, each iButton was checked and the data were downloaded 

to a laptop computer using the iButton USB port adaptor (Jesus Gomez, INRENA, Perú). 

iButton Temperature (°C) 

16 

14 

12 

10 

8 

6 

4 

2 

(a) 

iButton Temperature Callibration 

iButton = 1.0211(HOBO) - 0.0561 

R 2 = 0.988 

2 4 6 8 10 12 14 16 

HOBO Temperature (°C) 

Figure 6. Calibration for iButton sensor inside PVC radiation shield. 

268 

(b) 

(c)

Data analysis techniques 

We synthesized all data from the 8 iButton temperature loggers on the western slope, the HOBO 

weather station, and the precipitation logger into hourly values to create 24-day periods during the 

2005 dry, 17 June to 11 July, and wet, 7 to 31 December, seasons. These sample periods were 

chosen based on data availability and central proximity to the dry and wet seasons. We created 

composite averages of 24 hourly measurements for all 24 days making up the wet and dry 

periods—the diurnal cycle. Hence, composites are averages for each hour over the 24-day periods. 

The vector components of wind speed were calculated to create composite averages of speed and 

direction. Because we were interested in microscale variability, particularly elevation effects, we 

concentrated on analyzing and comparing the composite diurnal cycles of meteorological 

components for the wet and dry periods. 

Estimating Evapotranspiration 

It is difficult to measure ET accurately, particularly in remote mountainous regions with steep 

topography; most measurements require expensive equipment and frequent maintenance (refs). 

Consequently, there are a multitude of methods for estimating ET, most of which require site 

specific parameters and basic meteorological measurements. Evaporation rates, ignoring 

transpiration, are commonly estimated by energy balance, aerodynamic, and a combination of both 

methods, such as Penman (1945; 1963) and Penman-Monteith. Xu and Singh (1998; 2002) report 

on meteorological forcing of and compares methods of estimating ET based on input data from 

Changines station in Switzerland. We used a similar modeling approach and applied it to the 

Llanganuco valley. Allen et al. (1989) reports on the FAO methods for estimating potential ET 

(ET0) based on Penman combination equations for reference surfaces, such as uniform 120 mm 

tall grass used herein. Because of it’s inclusion of several commonly used models for estimating 

ET0, we elected to use the REF-ET software developed by Allen et al. (1998) and selected the 

Penman-Monteith FAO method for estimating hourly ET0 during the dry and wet periods. This 

will serve as a base-line for comparison to a more realistic physics-based model for ET. The 

equations for the FAO methods are explained in detail by Allen et al. (1998). 

We selected the BROOK90 (v.4.4e) model (Federer et al., 2003; Federer, 1995) to estimate the 

actual ET during the dry and wet season composite days. BROOK90 includes strong physicallybased 

determination of ET, a graphic user interface, and the visual basic code is available for 

modification. The model simulates deposition and sublimation of frozen water and assumes 

snowfall for near-surface air temperature below –1.5°C. The model meteorological input includes 

solar radiation, air temperature, vapor pressure, wind speed, and precipitation. Soil and vegetation 

parameters were measured in the field. Additional parameters were taken from the literature as 

reported by Federer et al. (2003). Model output includes hydrological balance components of the 

vegetation canopy and soil surface; we here report on the ET components. 

BROOK90 simulates evaporation and soil-water movement using a process-oriented approach 

for sparse canopies at a single location within a watershed. The model estimates interception and 

transpiration from a single-layer plant canopy, soil and snow evaporation/sublimation, snow 

accumulation and melt, and soil water movement through multiple soil layers. Potential 

evaporation rates are obtained using the Shuttleworth and Wallace (1985) modification of the 

Penman-Monteith combination equation. 

Actual transpiration is the lesser of potential transpiration and a soil water supply rate 

determined by the resistance to liquid water flow in the plants and on root distribution and soil 

water potential in the soil layers (Federer, 1979). For potential transpiration, canopy resistance 

depends on maximum leaf conductance, reduced for humidity, temperature, and light penetration. 

Each soil layer and the roots of vegetation have a resistance to water flow based on field 

observations and published results. Aerodynamic resistances are modified from Shuttleworth and 

Gurney (1990); they depend on leaf area index (LAI), which can vary seasonally, and on canopy 

height, which determines stem area index (SAI). Potential transpiration or potential interception 

are obtained using the actual or existing soil surface wetness in the Shuttleworth-Wallace 

equations. The equations then provide the total soil or ground evaporation. The total ET is the sum 

of precipitation evaporated from the canopy, soil evaporation, and transpiration from the canopy. 

269

In contrast to the six ET0 models, the BROOK90 model provided more realistic representation 

of vegetation and soil parameters and allowed evaluation of the hydrologic components affecting 

ET. The components included infiltration, surface evaporation, transpiration, and canopy 

intercepted evaporation. 

RESULTS 

Field Measurements 

We report on the results of intercomparisons for dry and wet composite diurnal cycles of air 

temperature, relative humidity, insolation, precipitation, wind speed and direction, and iButton 

temperatures. 

HOBO automated weather station 

We were particularly careful with our data syntheses from the humidity sensor (naturally 

aspirated), which was affected by supersaturation during portions of the wet season. We computed 

the dew point temperatures from humidity and relative humidity measurements and adjusted 

erroneous humidity values according to our corrections. Table 1 summarizes the results from the 

HOBO weather station. The ratio of dry to wet was applied to demonstrate differences. 

Table 1. Comparison of average daily statistics for variables measured by HOBO weather station 

during the dry and wet periods. 

HOBO Weather Station 

(daily statistics from hourly data) 

Dry Wet 

Dry/ 

Wet 

Air Temp. (°C) 7.6 6.6 1.15 

Max. Temp. (°C) 16.0 12.7 1.26 

Min. Temp. (°C) 1.5 3.6 0.42 

Temp. Range. (°C) 14.5 9.1 1.59 

RH (%) 59.5 92.4 0.64 

Vap. Press. (kPa) 0.62 0.90 0.69 

VP Deficit (kPa) 0.43 0.08 5.38 

Insolation (MJ/m 2 ) 16.82 13.21 1.27 

Wind (m/s) 1.42 1.18 1.20 

Wind Direction (°) 39 233 N/A 

Wind Constancy 0.54 0.72 0.75 

0.10 m Soil Moist. (m 3 /m 3 ) 0.005 0.119 0.04 

0.10 m Soil Temp. (°C) 13.8 12.8 1.08 

Precip. (mm/day) 0.08 7.36 0.01 

Comparisons of the HOBO weather station data for the composite dry and wet periods suggest 

significant differences between all variables of meteorological forcing. The average day during the 

wet period is 1°C cooler than the dry period, given the lack of rainfall and greater insolation 

during the dry period. Because of increased cloud cover associated with higher rainfall, the diurnal 

temperature range is 5.4°C less during the wet period. The average relative humidity of 59.5% for 

the dry period is 33% less than that of the wet season. The vapor pressure of 0.62 kPa for the dry 

period was 69% of that for the wet period, 0.90 kPa. 

Figs. 7 through 12 illustrate the diurnal cycles for the dry and wet periods. The persistent valley 

winds (Figs. 10b, 11b and 12b) during the daylight hours of the wet period controls the efficiency 

of the impact of turbulent sensible and latent heat flux on ET rate. Assuming a saturated 

evaporating surface, the aerodynamic term of the Penman-Monteith method maximizes ET0 with 

concurrently high wind speed and vapor pressure deficit. This assumption is good for the wet 

270

season, but unacceptable for the dry season when the surface is extremely dry (Fig. 8). The 

process of near-surface water vapor removal depends to a large extent on wind speed and air 

turbulence. When vaporizing water, the air above the evaporating surface becomes gradually 

saturated. If this air is not continuously replaced with drier air, the driving force for water vapor 

removal and the ET rate decreases. Hence, the dominance of daytime valley winds during the wet 

season plays a critical role in controlling ET within the pro-glacial valley, particularly given the 

steep confining walls. Furthermore, drier surface conditions in the lower valley will enhance the 

potential for ET throughout the valley, which is plausible given the predominantly easterly 

synoptic flow and the rain shadow effect to the west of the mountain range. 

(a) 

Air Temperature (°C) 

Dry Period Diurnal Variation of Average Air Temperature 

and Relative Humidity 

Relative Humidity Air Temperature 

20 

100 

16 

12 

8 

4 

0 

0:00 

2:00 

4:00 

6:00 

8:00 

10:00 

Time of Day (hours) 

12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

80 

60 

40 

20 

0 

Relative Humidity (%) 


271 

20 

16 

12 

8 

4 

0 

0:00 

2:00 

Wet Period Diurnal Variation of Average Air 

Temperature and Relative Humidity 

Relative Humidity Air Temperature 

4:00 

6:00 

8:00 

10:00 


12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

Figure 7. Composite diurnal cycle of air temperature and relative humidity for dry (a) and wet (b) periods. 

Soil Temperature (°C) 

30 

26 

22 

18 

14 

10 

6 

0:00 

2:00 

Dry Period Diurnal Variation of Average Soil 

Temperature and Soil Water Content 

Soil Water Content Soil Temperature 

4:00 

6:00 

8:00 

10:00 


12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

0.11 

0.09 

0.07 

0.05 

0.03 

0.01 

-0.01 

Water Content (m 3 /m 3 ) 

Soil Temperature (°C) 

30 

26 

22 

18 

14 

10 

6 

0:00 

2:00 

Wet Period Diurnal Variation of Average Soil 

Temperature and Soil Water Content 

Soil Water Content Soil Temperature 

4:00 

6:00 

8:00 

10:00 


12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

Figure 8. Composite diurnal cycle of incoming soil moisture and temperature for dry (a) and wet (b) periods. 

Precipitation, swe (mm) 

(a) 

(a) 

1 

0.8 

0.6 

0.4 

0.2 

0 

Dry Period Diurnal Variation of Average Precipitation 

and Insolation 

Precipitation Insolation 

1000 

0:00 

2:00 

4:00 

6:00 

8:00 

10:00 


12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

800 

600 

400 

200 

0 

Incident Solar (W/m 2 ) 

Precipitation, swe (mm) 

1 

0.8 

0.6 

0.4 

0.2 

0 

100 

80 

60 

40 

20 

0 

0.11 

0.09 

0.07 

0.05 

0.03 

0.01 

-0.01 

Wet Period Diurnal Variation of Average Precipitation 

and Insolation 

Precipitation Insolation 

1000 

0:00 

2:00 

4:00 

6:00 

8:00 

10:00 


12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

Figure 9. Composite diurnal cycle of incoming solar radiation and precipitation for dry (a) and wet (b) 

periods. 

800 

600 

400 

200 

0 

Water Content (m 3 /m 3 ) 

Relative Humidity (%) 

Incident Solar (W/m 2 ) 

(a) 

(b)

(a) 

Wind Direction (°) 

Wind Speed (m/s) 


360 

300 

240 

180 

120 

60 

(a) 

(a) 

0 

3.5 

2.5 

1.5 

0.5 

20 

16 

12 

8 

4 

0 

3 

2 

1 

0 

Dry Period Diurnal Variation of Wind Speed and 

Direction 

0:00 

2:00 

Wind Dir. 

Wind Speed 

4:00 

6:00 

8:00 

10:00 

12:00 

Time (hours) 

14:00 

16:00 

18:00 

20:00 

22:00 

3 

2.5 

2 

1.5 

1 

0.5 

0 



272 

360 

300 

240 

180 

120 

60 

0 

Wet Period Diurnal Variation of Wind Speed and 

Direction 

0:00 

2:00 

4:00 

6:00 

8:00 

10:00 

12:00 

Time (hours) 

Wind Dir. 

Wind Speed 

14:00 

16:00 

18:00 

20:00 

22:00 

Figure 10. Composite diurnal cycle of wind speed and direction for dry (a) and wet (b) periods. 

Dry Period Wind Speed versus Direction 

Valley Orientation 

55° => 235° 

0 45 90 135 180 225 270 315 360 



3.5 

3 

2.5 

2 

1.5 

1 

0.5 

0 

Wet Period Wind Speed versus Direction 


55° => 235° 

0 45 90 135 180 225 270 315 360 


Figure 11: Composite wind speed versus direction for dry (a) and wet (b) periods. 

Dry Period Air Temperature versus Wind Direction 


55° => 235° 

0 45 90 135 180 225 270 315 360 



20 

16 

12 

8 

4 

0 

Wet Period Air Temperature versus Wind Direction 


55° => 235° 

0 45 90 135 180 225 270 315 360 


Figure 12: Composite wind speed versus air temperature for dry (a) and wet (b) periods. 

iButton temperatures 

Table 2 summarizes the iButton temperature profile observed during the wet and dry periods. 

With exception of the iButton at 3948 m, which is cooler during the wet season due to shading 

from the south wall off the valley, the general trend is for stronger seasonal contrasts at higher 

elevation. Although the up-valley “lapse rate” is not the conventional atmospheric lapse rate, it is a 

strong indicator of sensible heat flux within the surface boundary layer, thereby affecting the rates 

of ice and snow ablation. Figs. 13 provides a unique perspective on the diurnal variability of 

temperature as function of elevation. The strong nocturnal inversion below the level of the 

meltwater lakes during the dry period is not present during the wet period. In general, the lapse 

rate is greater below the lakes than above for both seasons, with greatest diurnal variability during 

the dry period. Fig. 14 shows the composite lapse rate for both seasons. The average dry period 

zero degree isotherm is at 5377 m above sea-level, 344 m higher than the wet period freezing level 

(Fig. 14). Hence, the local valley impact on ablation rate of the glacial tongue is possibly 

3 

2.5 

2 

1.5 

1 

0.5 

0 

Wind Speed (m/s)

significant, although more information on the global and synoptic forcing is necessary to make 

conclusions at this time. 

Table 2. Average temperature recorded by iButtons during the dry and wet periods. Note the seasonal 

impact of “lapse rate” on the elevation of the freezing line (0°C). 

Elevation (m) 

(a) 

4800 

4600 

4400 

4200 

4000 

3800 

3600 

3400 

Dry Period Diurnal Change of Temperature Profiles 

iButton Dry Wet Dry/ 

Elevation (m) (°C) (°C) Wet 

4742 2.84 1.77 1.60 

4559 4.39 3.06 1.43 

4344 5.31 3.56 1.49 

4148 6.68 5.66 1.18 

3948 8.12 5.48 1.48 

3871 8.59 7.45 1.15 

3862 7.49 6.24 1.20 

3469 9.49 9.48 1.00 

Lapse Rate 

(°C/km) 

5.3 5.9 0.90 

0°C Elevation 

(m) 

5377 5033 1.07 

6:00 7:00 8:00 9:00 12:00 15:00 

16:00 17:00 18:00 21:00 0:00 3:00 

-5 0 5 10 15 20 25 

Temperature (°C) 

Elevation (m) 

273 

4800 

4600 

4400 

4200 

4000 

3800 

3600 

3400 

Wet Period Diurnal Change of Temperature Profiles 

6:00 7:00 8:00 9:00 12:00 15:00 

16:00 17:00 18:00 21:00 0:00 3:00 

-5 0 5 10 15 20 25 

Temperature (°C) 

Figure 13. Diurnal variability of near-surface air temperature profiles, “lapse rate,” for dry (a) and wet (b) periods. 

Elevation (m) 

5400 

5200 

5000 

4800 

4600 

4400 

4200 

4000 

3800 

3600 

3400 

Average Temperature Profiles: Dry and Wet Periods 

Dry Wet Linear (Wet) Linear (Dry) 

Tdry = -0.0053(Z) + 28.5 

R 2 = 0.95 

Twet = -0.0059(Z) + 29.7 

R 2 = 0.97 

0.0 2.0 4.0 6.0 8.0 10.0 


Figure 14. Composite iButton temperature profiles for dry (a) and wet (b) periods.

Model Results 

We report on the (preliminary) results from the FAO and BROOK90 model simulations of ET0 

and actual ET, respectively, during the same dry and wet composites created for the 

meteorological data analysis. Input from the HOBO weather station and measurements of site 

parameters in the field were made at one central location within the valley. The resulting 

hydrological components for the dry and wet composite days are compared in Table 3 and Figs. 15 

and 16. The magnitude of ET0 was much greater, about two times, for the dry period, which agrees 

with the lack of precipitation during the dry period (Table 1). 

Table 3 breaks down the hydrological components of ET based on the BROOK90 model runs 

for the composite hourly dry and wet periods. The BROOK90 estimate of actual ET from the wet 

period is 2.63 mm d -1 , which is 88 times that of the dry period, 0.03 mm d -1 . ET is 33% of daily 

precipitation for the dry period and 37% of the daily precipitation for the wet period, which are 

both significant portions of the hydrologic balance in the Llanganuco Valley. Transpiration is the 

greatest contribution to ET, 33% for the dry and 78% for the wet period. 

Table 3. BROOK90 Estimated Moisture Fluxes Dry & Wet Periods (assuming 50% true veg. cover) 

Period 

Pre* 

(mm) 

Inf 

(mm) 

SEv 

(mm) 

274 

Trs 

(mm) 

IEv 

(mm) 

ET 

(mm) 

Dry 0.09 0.09 0.02 0.01 0.00 0.03 

% of 

Precip. 

100 99 22 11 0 33 

Wet 7.06 6.82 0.35 2.06 0.22 2.63 

% of 

Precip. 

100 97 5 29 3 37 

Wet/Dry 78 76 18 206 n/a 88 

*Pre = precipitation; Inf = infiltration; SEv = surface evaporation; Trs = transpiration; 

IEv = intercepted evaporation; ET = evapotranspiration 

Liquid Water (mm) 

1 

0.8 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

ET: Dry and Wet Period (+ = loss from surface) 

Dry Wet DryRef WetRef 

0:00 

2:00 

4:00 

6:00 

8:00 

10:00 

12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

Time of Day (hour) 

Figure 15. Penman-Monteith FAO and BROOK90 modeled ET for composite diurnal cycle for dry (a) and 

wet (b) periods.

(a) 


Dry Period: Modeled Hydrological Components 

Precip. Soil Evap. Transpiration Intercepted Evap. Infiltration 

1 

0.8 

0.6 

0.4 

0.2 

0 

-0.2 

0:00 

2:00 

4:00 

6:00 

Figure 16. BROOK90 modeled components of ET for composite diurnal cycle for dry (a) and wet 

(b) periods. 

DISCUSSION 


8:00 

10:00 

12:00 

14:00 

16:00 

18:00 

20:00 

22:00 

Field and model results were interpreted and compared to explain the processes controlling ET 

within a pro-glacial valley. 

Meteorological Measurements 

The new sensor network initiated by this project permitted high resolution, discrete monitoring 

of air temperature at different elevations within a pro-glacial valley. Calibration of iButton 

temperature logger with the shielded temperature sensor on the HOBO weather station suggested a 

strong correlation, r 2 =0.988, and this gave us high confidence in our interpretation of the 

measurements. As expected, analysis of hourly data from 2005 revealed evidence for a strong 

diurnal and wet versus dry seasonal dependence of meteorological forcing within the valley. The 

dry/wet ratio of 1.27 for daily insolation is smaller than we expected given the absence of 

precipitation during the dry period and much high precipitation total for the wet period. The 

hourly composite of precipitation for the wet period (Fig. 9) clearly shows the nocturnal tendency 

for precipitation and hence convective cloud formation at night, and strong solar forcing during 

daylight hours. 

However, solar forcing was surprisingly similar for both the dry and wet periods, since most 

convection (rainfall) during the wet season occurs an hour or two prior to sunset and dissipates 

prior to sunrise. This strong diurnal convective cycle coupled with a persistent up-valley wind and 

a strong daytime lapse rate during the wet season suggests the plausible influence of daytime 

surface heating west and down slope of the pro-glacial valley. Furthermore, since clouds 

accompany precipitation, we concluded that nocturnal cloud cover strongly influences 

interseasonal and diurnal cycles of net radiation. Furthermore, this may establish a connection 

between cloud cover (and precipitation) and anthropogenic development in the Santa River valley, 

which receives most of its water from pro-glacial valleys along the western Cordillera Blanca, 

such as Llanganuco (Fig. 1). As demonstrated by Vuille et al. (2003), it is the western slope of the 

Cordillera Blanca that is experiencing the highest rate of warming based on weather station 

records. If this warming trend continues, we would expect to find drier and stronger up valley 

winds during both seasons. Forced by orographic uplift of the western slope, these near-surface 

winds would promote stronger nocturnal convection during the wet season. 

This is one of our arguments for continued expansion of ground-based meteorological networks 

within tropical proglacial valleys. We are also interested in more accurately estimating the 

evapotranspiration rate within the valleys. Our model results are preliminary and we will require 

additional field measurements to verify and more accurately initiate the BROOK90 model. 


275 

Wet Period: Modeled Hydrological Components 

Precip. 

1 

Soil Evap. Transpiration Intercepted Evap. Infiltration 

0.8 

0.6 

0.4 

0.2 

0 

-0.2 

0:00 

2:00 

4:00 

6:00 


8:00 

10:00 

12:00 

14:00 

16:00 

18:00 

20:00 

22:00

Importance of diurnal cycles in the Tropics 

Diurnal variation is related to the large and well-defined cycle in solar heating during a 24-hour 

period, and it represents one of the most fundamental components accounting for the variability of 

the climate system. Numerous observation studies have documented diurnal variation of deep 

convection, precipitation, cloudiness, and outgoing longwave radiation over the tropics (e.g., Gray 

and Jacobson, 1977; Duvel and Kandel, 1985; Hendon and Woodberry, 1993; Chen and Houze, 

1997; Yang and Slingo, 2001; Nesbitt and Zipser, 2003; among many others). Mapes et al. (2003) 

observed an afternoon maximum rainfall over most of South and Central America that is typically 

composed of relatively small convective cloud systems. Furthermore, Mapes et al. found a 

nocturnal maximum of rainfall over some large valleys in the Andes, and this would include the 

larger hanging valleys, such as Llanganuco. Poveda et al. (2005) use hourly records from 51 rain 

gages in the Tropical Andes of Columbia to find clear diurnal cycles in precipitation, with minima 

between 0900 and 1100 local time, and nocturnal maxima on the western flank of the Central 

Andes. In addition, Porveda concluded no relation between the timing of the strong seasonal 

variability of rainfall maxima and elevation. 

Bendix et al. (2006) used K-band rain-radar to study convection in a valley with east-west 

orientation between the southern Equadorean Andes and the Amazon basin. Results revealed that a 

great portion of rainfall is of stratiform character, and discovered the existence of embedded 

convection and/or showers produced by local heating for the overall amount of rainfall. Bendix et 

al. (2006) further suggests that cold air drainage flow from the Andes and low-level confluence 

due to the concavity of the Andean chain in this area leads to convective instability in the 

nocturnal Amazonian boundary layer, which is extended to the east-west oriented Andean valley 

the predominant easterlies in the mid-troposphere. Rain clouds with at least embedded shallow 

convection can overflow the bordering ridges of the San Francisco valley providing rains of higher 

intensity at the ECSF research station. On the contrary, Bendix et al. found that afternoon 

convective precipitation can be caused by locally induced thermal convection at the bordering 

slopes (up-slope breeze system) where the ECSF station profits from precipitation off the edge of 

these local cells due to the narrow valley. Recent remote sensing studies demonstrate pronounced 

diurnal variability of tropical rainfall intensity at synoptic and regional scales (Sorooshian et al., 

2002; Bowman et al., 2005). Hence, evidence from various scales of observation suggest a strong 

influence of diurnal cycles in tropical regions, including the Andes Mountains. 

CONCLUSION 

From our meteorological and model results, we make the following conclusions for this 

proglacial valley. Additional years and observations of nearby valleys are anticipated. It is 

important to note that we are using the term lapse rate in an unconventional sense, not the free 

atmosphere, but rather the effects within the surface boundary layer up the valley floor. First, the 

steepest lapse rates occur below the lakes for both wet and dry seasons, and the lapse rate is 

significantly smaller above the glacial lakes. We partially attribute this elevation effect to the 

heterogeneous vegetation and topographic characteristics of pro-glacial valleys. Fruthermroe, the 

dry season nocturnal inversion below the lakes is not evident during wet season. Most 

precipitation (and cloud cover) occurs between sunset and sunrise during the wet season, hence 

insolation at the ground is strong during both seasons. Up-valley winds dominate during sunlight 

hours in both wet and dry seasons, but abruptly shift to katabatic winds during the dry season at 

sunset. The combination of valley winds and steep lapse rate below the lakes suggest local warm 

air advection within the surface boundary layer will contribute to evapotranspiration and lower 

glacial melt. Furthermore, the up-slope winds may enhance convective precipitation, particularly 

during the wet season after sunset. The BROOK90 output suggests that transpiration is a 

significant source of ET when soil moisture is available during the wet season. The model results 

suggest that the predominance of cloud-free daylight conditions and relatively high solar input 

enhance ET during the wet season. ET was insignificant throughout the dry season. 

276

Future Work 

We are installing additional humidity, wind, solar, and soil moisture sensors to evaluate spatial 

variation of ET. We plan to make direct measurements of ET using lysimeters and evaporation 

pans, and to use BROOK90 to perform a formal sensitivity analysis of ET to meteorological 

forcing. In addition to site-specific parameters, the meteorological input includes solar radiation, 

air temperature, vapor pressure, wind speed, and precipitation. Model output includes various 

hydrological balance components of the vegetation canopy and soil surface. We plan to evaluate 

the reliability and realism of estimated ET and to determine the sensitivity of ET to variables 

measured at the HOBO site in the Llanganuco valley. Future modeling will focus on more 

accurate measurements of parameters for BROOK90 and the elevation effects on ET and total ET 

for the valley. 


Funding for this project was provided by The Ohio State University, Department of Geography 

and Office of International Affairs. We are grateful for the collaboration with Peruvian Institute of 

Natural Resources (INRENA), especially the timely collection and transmission of data by Jesús 

Gómez, INRENA-Huaraz, Perú. 

REFERENCES 

Allen, R.G., Jensen, M.E., Wright, J.L., and Burman, R.D. 1989. Operational Estimates of 

Reference Evapotranspiration. Agronomy Journal, 81: 650–662. 

Allen, R.G., Pereira, L.S., Raes, D., and Smith, M. 1998. Crop evapotranspiration-Guidelines for 

computing crop water requirements, FAO irrigation and drainage paper 56, FAO. ISBN 92-5- 

104219-5. 

Ames, A., 1998. A documentation of glacier tongue variations and lake developments in the 

Cordillera Blanca. Zeitschrift fur Gletscherkunde und Glazialgeologie, 34(1): 1–26. 

Barry, R. and Seimon, A., 2000. Research for mountain area development: Climatic fluctuations in 

the mountains of the Americas and their significance. Ambio, 29(7): 364–370. 

Bendix, J., Rollenbeck, R., and Reudenbach, C., 2006. Diurnal patterns of rainfall in a tropical 

Andean valley of southern Ecuador as seen by a vertically pointing K-band Doppler radar. 

International Journal of Climatology, Volume 26, Issue 6, 829–846. 

Bowman, KP, Collier, JC, North, GR, Wu, Q, Ha, E, Hardin, J, 2005. Diurnal cycle of tropical 

precipitation in Tropical Rainfall Measuring Mission (TRMM) satellite and ocean buoy rain 

gauge data. Journal of Geophysical Research, 110(d21): D21104-. 

Caballero, Y., Chevallier, P., Gallaire, R. and Pillco, R., 2004. Flow modelling in a high mountain 

valley equipped with hydropower plants: Rio zongo valley, Cordillera Real, Bolivia. 

Hydrological Processes, 18(5): 939–957. 

Caballero, Y., Jomelli, V., Chevallier, P. and Ribstein, P., 2002. Hydrological characteristics of 

slope deposits in high tropical mountains (Cordillera Real, Bolivia). Catena, 47(2): 101–116. 

Chen, S. S., and R. A. Houze Jr., 1997: Diurnal variation and life-cycle of deep convective 

systems over the tropical Pacific warm pool. Quart. J. Roy. Meteor. Soc., 123, 357–388. 

Duvel, J. P., and R. S. Kandel, 1985: Regional-scale diurnal variations of outgoing infrared 

radiation observed by METEOSAT. J. Climate Appl. Meteor., 24, 335–349. 

Federer, C.A. 1979. A soil–plant–atmosphere model for transpiration and availability of soil 

water. Water Resour Res., 15:555–562. 

Federer, C.A. 1995. BROOK90. A Simulation Model for Evaporation, Soil Water, and 

Streamflow. Version 3.1 Computer Freeware and Documentation. USDA Forest Service, P.O. 

Box 640, Durham, N.H. 03824. 

Federer, C.A., C. Vörösmarty, and B. Fekete. 1996. Intercomparison of methods for calculating 

potential evaporation in regional and global water balance models. Water Resour Res., 32: 

2315–2321. 

277

Federer, C.A., C. Vörösmarty, and B. Fekete. 2003. Sensitivity of annual evaporation to soil and 

root properties in two models of contrasting complexity. J Hydrometeorology, 4:1276–1290. 

Francou, B., Ribstein, P., Saravia, R. and Tiriau, E., 1995. Monthly balance and water discharge of 

an inter-tropical glacier: Zongo Glacier, Cordillera Real, Bolivia, 16S. Journal of Glaciology, 

41(137): 61–67. 

Georges, C. and Kaser, G., 2002. Ventilated and unventilated air temperature measurements for 

glacier-climate studies on a tropical high mountain site. Journal of Geophysical Research- 

Atmospheres, 107(D24): art. no.-4775. 

Gray, W. M., and R. W. Jacobson, 1977: The diurnal march of convective cloudiness over the 

Americas. Mon. Wea. Rev., 105, 1171–1188.Francou, B., Ribstein, P. and Pouyard, B., 1997. 

The melting of tropical glaciers. Recherche(302): 34–37. 

Hardy, D. R., M. Vuille, C. Braun, F. Keimig, and R. S. Bradley, Annual and daily meteorological 

cycles at high altitude on a tropical mountain, Bull. Am. Meteorol. Soc., 79(9), 1899–1913, 

1998. 

Hastenrath, S. and Kruss, P.D., 1992. The dramatic retreat of Mount Kenya's glaciers between 

1963 and 1987: greenhouse forcing. Annals of Glaciology, 16: 127–133.Hendon, H. H., and K. 

Woodberry, 1993: The diurnal cycle of tropical convection. J. Geophys. Res., 98, 16,623– 

16,637. 

Hinckley, T.M., J.P. Lassoie, and S.W. Running. 1978. Temporal and spatial variations in the 

water status of forest trees. Forest Science Monograph, 20, 72 p. 

Jackson, R.B. J. Canadell, J.R. Ehleringer, H.A. Mooney, O.E. Sala, and E.-D. Schulze. 1996. A 

global analysis of root distributions for terrestrial biomes. Oecologia, 108:389–411. 

Juen, I., Kaser, G. and Georges, C., 2005. Modelling observed and future runoff from a glacierized 

tropical catchment (Cordillera Blanca, Peru). Global & Planetary Change, submitted. 

Kaser, G., Juen, I., Georges, C., Gomez, J. and Tamayo, W., 2003. The impact of glaciers on the 

runoff and the reconstruction of mass balance history from hydrological data in the tropical 

Cordillera Blanca, Peru. Journal of Hydrology, 282(1–4): 130–144. 

Mapes, Brian E., Warner, Thomas T., Xu, Mei, Negri, Andrew J., 2003. Diurnal Patterns of 

Rainfall in Northwestern South America. Part I: Observations and Context. Monthly Weather 

Review, 131: 799–812. 

Mark, B.G., McKenzie, J.M. and Gomez, J., 2005. Hydrochemical evaluation of changing glacier 

meltwater contribution to stream discharge: Callejon de Huaylas, Peru. Hydrological Sciences 

Journal = Journal des Sciences Hydrologiques, 50(6). 

Mark, B.G. and Seltzer, G.O., 2003. Tropical glacier meltwater contribution to stream discharge: a 

case study in the Cordillera Blanca, Peru. Journal of Glaciology, 49(165): 271–281. 

Mark, B.G. and Seltzer, G.O., 2005. Recent glacial recession in the Cordillera Blanca, Peru (AD 

1962–1999). Quaternary Science Reviews, 24: 2265–2280. 

McNulty, B.A., Farber, D.L., Wallace, G.S., Lopez, R. and Palacios, O., 1998. Role of plate 

kinematics and plate-slip-vector partitioning in continental magmatic arcs: Evidence from the 

Cordillera Blanca, Peru. Geology, 26(9): 827–830. 

Nesbitt, S. W., and E. J. Zipser, 2003: The diurnal cycle of rainfall and convective intensity 

according to three years of TRMM measurements. J. Climate, 16, 1456–1475. 

Penman, H.L., 1948. Natural evaporation from open water, bare soil and grass. Proc. Roy. Soc. 

London. A193, 120–146. 

Penman, H.L., 1963. Vegetation and hydrology. Tech. Comm. No. 53. Commonwealth Bureau of 

Soils, Harpenden, England. 125 pp. 

Poveda, Germán, Mesa, Oscar J., Salazar, Luis F., Arias, Paola A., Moreno, Hernán A., Vieira, 

Sara C., Agudelo, Paula A., Toro, Vladimir G., Alvarez, J. Felipe, 2005. The Diurnal Cycle of 

Precipitation in the Tropical Andes of Colombia. Monthly Weather Review, 133: 228–240. 

Ribstein, P., Tiriau, E., Francou, B. and Saravia, R., 1995. Tropical Climate and Glacier 

Hydrology: A Case-Study in Bolivia. Journal of Hydrology, 165(1–4): 221–234. 

Shuttleworth, W.J., and R.J. Gurney. 1990. The theoretical relationship between foliage 

temperature and canopy resistance in sparse crops. Quart J Royal Meteorol Soc., 116: 497– 

519. 

278

Shuttleworth, W.J., and J.S. Wallace. 1985. Evaporation from sparse crops: An energy 

combination theory. Quart J Royal Meteorol Soc., 111: 839–855. 

Sorooshian, S., Gao, X., Hsu, K., Maddox, R. A., Hong, Y., Gupta, H. V., Imam, B. Diurnal 

Variability of Tropical Rainfall Retrieved from Combined GOES and TRMM Satellite 

Information. Journal of Climate, 2002 15: 983–1001. 

Vuille, M., Bradley, R.S., Werner, M. and Keimig, F., 2003. 20th century climate change in the 

tropical Andes: Observations and model results. Climatic Change, 59(1–2): 75–99. 

Wagnon, P., Ribstein, P., Kaser, G. and Berton, P., 1999. Energy balance and runoff seasonality of 

a Bolivian glacier. Global and Planetary Change, 22(1–4): 49–58. 

Wagnon, P., Ribstein, P., Schuler, T. and Francou, B., 1998. Flow Separation On Zongo Glacier, 

Cordillera Real, Bolivia. Hydrological Processes, 12(12): 1911–1926. 

Xu, C.-Y. and Singh, V.P. 2002. Cross comparison of empirical equations for calculating potential 

evapotranspiration with data from Switzerland. Wat. Resour. Manag., 16, 197–219. 

Xu, C.-Y. and Singh, V.P. 1998. Dependence of evaporation on meteorological variables at 

different time-scales and intercomparison of estimation methods. Hydrol. Process., 12, 429– 

442. 

Yang, Gui-Ying, Slingo, Julia, 2001. The Diurnal Cycle in the Tropics. Monthly Weather Review, 

129: 784–801 

279


280

Snow and Periglacial 

Processes Poster 

281


282

283 



A Treatise on the Preponderance of Designs Over Historic and 

Measured Snowfalls, or No Two Snowflakes Are Alike: 

Considerations About the Formation of Snowflakes 

and the Possible Numbers and Shapes of Snowflakes 

MARK PILIPSKI AND JACOB DIESSNER PILIPSKI 

"The reason for this macroscopic six-fold symmetry, which so exercised 

Kepler, is now readily explained in terms of the atomic architecture of ice, but the 

variety and change of the crystal shape has always been a puzzle and still awaits 

a complete explanation." 

— B.J.Mason in A Commentary on Kepler's essay ON THE SIX-CORNERED SNOWFLAKE 

ABSTRACT 

We use a generalized geometric model for the shape of snowflakes to approximate the finite 

number of possible unique snowflake configurations. Using historic geologic and meteorological 

data we approximate the number of snowflakes created throughout history. Assuming the 

development of snowflakes to be subject to guided random molecular movements, we calculate 

the probability of duplicate configurations. The factorial estimate of possible snowflake 

configurations overwhelms the countable linear quantity of snowflakes. We show, even with very 

conservative estimates, that except for extremely small snowflake fragments, no two snowflakes 

are identical. Thus, at the macro-optical scale, no two snowflakes are alike. 

PREMISE 

Simply stated, we determine from a general model for snowflakes how many possible 

snowflake configurations may exist. Using geologic and meteorological data we approximate the 

number of snowflakes created throughout history. If the total number of actual snowflakes 

exceeds the possible number of snowflake configurations, we deduce that at least two snowflakes 

throughout history have been identical. If this is so, we will be able to determine how often 

identical snowflakes fall. If the total number of possible snowflake configurations greatly exceeds 

the total number of snowflakes that have fallen throughout history, we must concede that it is 

improbable that two identical snowflakes have ever fallen. 

To evaluate this premise, we sought to answer these two questions: 

1. How many possible snowflake configurations are there? 

2. How many snowflakes have fallen? 

The answer to the first question will be the result of our modeling. We develop a model that 

approximates the shape of a snowflake. We limit our efforts to one form of snowflake; the 

dendritic plate.

The answer to the second question requires approximations. We chose such to maximize our 

result. Thus, for the total number of snowflakes that have fallen, our approximation will be higher 

than the historical value. 

To determine how many snowflakes have fallen we need to know, how much snow has fallen 

and the average size of a snowflake. Measurements of snowfall are made locally not globally. Any 

amount of snow that we project to have fallen on the earth will be, at best, a guess. 

SNOWFLAKE MODEL 

Combinations and configurations 

This is a model for a flat hexagonal snowflake and incorporates dendritic plates. Several 

simplifications make this model workable and yet remain consistent with what is known about 

real snowflakes. The model begins, as any good model, with actual snowflakes. (3) 

Our model applies to dendritic plates: 

We assume snowflakes have: 

Hexagonal radial symmetry (six-sided snowflakes), 

Two-dimensional configuration (flat snowflakes), and 

The simplest unit of an ice crystal is a hexagonal ring composed of six water molecules. 

We know that real snowflakes have several documented forms and, although, these may be 

hexagonal in nature, they are not strictly two-dimensional. They have thickness and the basic 

building blocks of any ice crystal may be far more complex than a ring of only six water 

molecules. Thus, any model we develop will underestimate the number of possible snowflake 

configurations. 

Each of the above assumptions carries several corollaries that shape our model. Hexagonal 

radial symmetry implies that: 

a. Each of the six arms are identical to each other. 

i. One sixth of the total number of units is used to compose each of the 

six arms. 

b. Each arm is bilaterally symmetric. 

(It is the same on each side of its radial axis.) 

i. There is a centerline of units. Assume the centerline (radial axis) to be 

the longest line of each arm. 

ii. The sidelines (parallel to the centerline) are paired, one on either side 

of the centerline. 

iii. The sidelines are made up of many segments of lines each shorter than 

the centerline. 

iv. The length of each sideline must be at least three units shorter than the 

next sideline closer to the centerline and at least three units longer than 

the next sideline farther from the centerline. 

284

----- -------------- 

------- ------------ ---------- sideline (3 segments) 

------- ----- - ----------- --------- sideline 

--------------------------------------------------- sideline 

Radial axis --------------------------------------------------------------- one of six arms 

--------------------------------------------------- 

------- ----- - ----------- --------- 

------- ------------ ---------- 

----- -------------- 

We know that the most common form of ice is a tetrahedral structure. Idealized hexagonal rings 

are found within the tetrahedral form. Close examination of the basal plane of an ice crystal 

reveals it is composed of a tiling of hexagonal plates each with a diameter of 5.5 angstroms, 4.5 

angstroms across the tiling diameter or minor chord and 2.75 angstroms between adjacent oxygen 

atoms. (1)(5) 

Using a simple idealized hexagonal ice crystal as a building block, we can approximate how 

many such blocks or hexagonal tiles must accumulate to form a single axis arm of a snowflake. 

Ordinary snowflakes vary in size from 0.05 cm. to 1.8 cm. in diameter. A reasonable midrange 

value for the average radius of a snowflake is 0.66 cm. Thus, a radius axis of an average 

snowflake is 

285

6.6 × 10 –1 cm. (radius axis length) = 1.5 × 10 7 hex units 

4.5 × 10 –8 cm. (hexagonal unit length) 

1.5 × 10 7 units is the number of hexagonal units along any one axis. 

The model calls for the axis and any of the sidearms to be at most filled with hexagonal 

units and in most cases present with gaps in the linear tiling. We may think of any axis or sidearm 

of actual hex units to be represented by a binary number (hex unit present = 1, hex unit absent = 

0.) 

A sidearm three units in length would have 

2 3 – 1 = 7 possible configurations of ice hex units 

111, 110, 101, 100, 011, 010, 001 

The null configuration of no hex units (000) is not possible because the resulting 

snowflake would lack any cohesion along this null line. 

A radius axis or sidearm with 1.5 × 10 7 units would have 

1.5 × 10 7 

(2 – 1) possible permutations. 

For a configuration consistent with our model we have, where N= 2 

286 

1.5 × 10 7 

2 (N–3) –1 is the number of possible permutations for the next smaller sidearm(s). Then it follows 

N/3 

2 (N) –1 . 2 (N–3) –1 . 2 (N–6) –1 . 2 (N–9) –1 …= Π 2 (N–3X) –1 

X=0 

Equals the number of possible permutations and combinations of all sidearms, i.e. the number 

of possible unique snowflakes. The multiplication of exponential quantities may be expressed as 

the sum of these exponents. We may simplify this series as 

N/3 N/3 N/3 

Σ (N–3X) Σ (3X+1) 3 Σ (X) + Σ (1) 

0 0 0 

2 –1 = 2 –1 = 2 – 1 = 

N/3 3.8 × 10 13 10 13 

Π 2 (N–3X) –1 or ≈ 2 or ≈ 10 

X=0 

That’s a one followed by ten trillion zeros, as the possible number of unique snowflake 

configurations.

No one to date has shown why snowflakes (here we are discussing only the formation of plates, 

specifically dendritic plates) demonstrate a remarkably precise hexagonal radial symmetry. The 

currently accepted explanation for this is that as snowflakes form they travel along unique random 

paths that expose the developing snowflake to micro environmental fluctuations. Crystalization 

occurs in response to these local conditions. The nature of water molecules leads to the familiar 

hexagonal crystals. Thus, the unique path taken by a developing snowflake determines its final 

configuration. (4)(9) 

Another possible contributing factor for the randomness we see in snowflake formation is the 

presence of a quasi-liquid phase as part of a developing snowflake. (7) 

We postulate another possible explanation for these guided random configurations of 

snowflakes; the development of an emerging pi-obital like electron cloud. This elaborate pi-orbital 

develops as the crystal grows creating sites favored for the attachment of the next water 

molecules. Such pi-orbitals are found in large organic molecules (i.e. planar benzene constructs) 

as well as several planar inorganic crystalline forms. We consider the formation of ice pi-orbitals, 

much like those of benzene molecules. The individual water (ice) hex plates may each have such a 

pi-orbital and as they are joined to form a larger plate this orbital expands to ‘guide’ the future 

formation of the snowflake. 

Regardless of the exact cause for the observed symmetry, we find it to be guided (following a 

symmetric pattern) and random (each crystalline growth point may or may not affix a water 

molecule or molecular complex. 

The physical implications of this model lead us to claim that there are 

1.5 × 10 7 hex units in a single radial axis; 3.8 × 10 13 hex units in a complete radial arm (a radial 

axis and all sidearms.) There are six radial arms per each snowflake and six water molecules for 

each ice hex unit. These figures give us 

(3.8 × 10 13 hex units per arm) × (6 arms per snowflake) × (6 H2O per hex unit) 

= 1.4 × 10 15 H2O molecules in a model snowflake. 

This is based upon a two-dimensional model. Actual snowflakes, as far as they might inform 

this model, have thickness. By adding several layers of parallel planes of ice above and below the 

central plane of this model, we see that we approach the measured mass of observed snowflakes. 

We also appreciate that each added plane of ice multiples our estimate of possible snowflake 

configurations to beyond the already predicted astronomical numbers. 

Throughout the development of this model we’ve assumed that there is an equivalence of the 

probability of any possible snowflake configuration. In other words, the probability of any one 

snowflake configuration is the same as any other snowflake configuration. This may not be a valid 

assumption. Indeed, there may be favored snowflake configurations. Thus, the final form of any 

actual snowflake may not be the product of strictly random development. If this is true, our 

estimate of all possible snowflake configurations is an over-estimate of any real quantity 

Photographs show that some real snowflakes are asymmetric. Whether this asymmetry is 

evident while a snowflake develops or induced owing the capture processes, symmetry prevails. 

We can argue ad nauseum about the definition of ‘alike’ or ‘identical.’ We may also accept the 

concept of ‘for all practical purposes.’ 

SNOWFALL 

There are 1.4 × 10 15 H2O molecules or 4.2 × 10 –8 gms of H2O in a model snowflake. If we 

approximate the average water density of fresh snow to be 12%, this is equivalent to saying that 

1.0 cm 3 of snow is equal to 0.12 gms of H2O. 

Using an estimate of total snowfall over the earth’s surface per year of 3.3 × 10 21 cm 3 (see 

Appendix A) and our model, we obtain 9.6 × 10 31 model snowflakes produced worldwide each 

287

year.(8) If we consider data for actual snowflakes we obtain 6.6 × 10 28 actual snowflakes 

produced worldwide each year. (2)(6) 

The earth is by currently available measures about 4.5 × 10 9 years old. Thus, for our model we 

obtain 4.3 × 10 39 snowflakes produced throughout history and for actual snowflakes we obtain 3 × 

10 38 as the number of historic snowflakes. Both numbers are close to each other and we use this 

closeness as a validation of the accuracy of the model. 

From our model we obtain a value of 10 13 for the number of possible snowflake configurations. 

(We have no evidence that there are preferred configurations.) We also estimate that only 10 40 

actual snowflakes have been produced throughout the history of the earth. (Of course there may 

have been great variations in seasonal snowfall throughout history. By maintaining a consistent 

yearly snowfall, we may over-estimate or under-estimate this value by a few orders of magnitude. 

With the current rate of snow production on earth it will be many trillions upon trillions of years 

before the production of two identical snowflakes may be considered a possibility. For now we 

may assume that there simply has not been enough snow to produce two identical snowflake 

configurations. 

288

APPENDIX A 

Northern Hemisphere Yearly Snowfall Totals (8) 

Band Area Avg Snowfall Snowfall 

(cm 2 )(x10 16 ) (cm) (cm 3 ) (x 10 19 ) 

90 o – 86 o 

86 o – 82 o 

82 o – 78 o 

78 o – 74 o 

74 o – 70 o 

70 o – 66 o 

66 o – 62 o 

62 o – 58 o 

58 o – 54 o 

54 o – 50 o 

50 o – 46 o 

46 o – 42 o 

42 o – 38 o 

38 o – 34 o 

34 o – 30 o 

30 o – 26 o 

0.627 2888 1.81 

1.866 3142 5.86 

3.068 3584 11.0 

4.212 3975 16.7 

5.277 3800 20.1 

6.233 3546 22.1 

7.081 3618 25.6 

7.766 5207 40.4 

8.328 3698 30.8 

8.708 2029 17.7 

8.932 589 5.26 

8.959 490 4.39 

8.740 140 1.22 

8.640 28 0.24 

8.011 30 0.24 

7.194 56 0.40 

289

REFERENCES 

Franks, F. The Properties of Ice, pp. 115–148 in WATER: A COMPREHENSIVE TREATISE 

Volume 1: The Physics and Physical Chemistry of Water. Plenum Press, New York, 1972 

Fujiyoshi, Y. and K.Muramoto. The Effect of Breakup of Melting Snowflakes on the Resulting 

Size Distribution of Raindrops. Journal of the Meteorological Society of Japan, 74(3), 1996 

Kepler, J.. THE SIX-CORNERED SNOWFLAKE 

(STRENA SEUDE NIVE SEXANGULA). Oxford University Press, Ely House, London W.1, 

1966 

Lynch, DK. Atmospheric Halos. SCIENTIFIC AMERICAN, 1978 

Nakaya, U.. The Formation of Ice Crystals. COMPENDIUM OF METEOROLOGY. American 

Meteorological Society, Boston pp. 207–220, 1951 

Rasmussen, R., J.Vivekanandan, J.Cole and E.Karplus. Theoretical Considerations in the 

Estimation of Snowfall Rate Using Visibility. The National Center for Atmospheric Research, 

1998. 

Sato, K. Instability of Quasi-liquid on the Edges and Vertices of Snow Crystals. arXiv:condmat/0208036 

v1 2 Aug 2002. 

Schultz, C., and L.D.Bregman. GLOBAL SNOW ACCUMULATION BY MONTHS. The Rand 

Corporation, 1988. 

Yosida, Z. ICE AND SNOW PROPERTIES, PROCESSES AND APPLICATIONS. Editor, 

W.D.Kingery. Massachusetts Institute of Technology Press, Cambridge 1962 

290

62nd ESC Papers 

291


292


Waterloo, ON, Canada 2005 

Extended Abstract: Simulations of North American Snow Cover 

by AGCMs and AOGCMs 

Keywords: snow, climate modeling 

A. FREI, 1 G. GONG, 2 AND R. BROWN 3 : 

We report on some highlights of recent evaluations of General Circulation Model (GCM) snow 

simulations in which the authors have participated. Two recent reports focus on evaluations of 

snow cover extent (SCE) (Frei et al. 2003) and snow water equivalent (SWE) (Frei et al. 2005) 

simulations for the period 1979-1995 by Atmospheric GCMs (AGCMs) participating in the 

Second Phase of the Atmospheric Model Intercomparison Project (AMIP-2). We also report on 

results of a recent preliminary evaluation of SCE simulations by coupled atmosphere-ocean GCMs 

(AOGCMs) participating in the upcoming Fourth Assessment Report of the Intergovernmental 

Panel on Climate Change (IPCC-AR4) (Frei and Gong 2005). 

Data Satellite observations of Northern Hemisphere SCE are available back to around 1967 

(Robinson 1993). Reconstructions of large scale SCE variations back to the early twentieth 

century over North America have been performed in two studies (Frei et al. 1999; Brown 2000), 

which utilized station observations of snow depth. The primary data set used for model evaluation 

of SWE is the gridded SWE data set produced specifically for the AMIP-2 project by Brown et al. 

(2003). 

Models AMIP-2 modeling groups run experiments for the years 1979-1996 with identically 

specified boundary conditions, including observed sea surface temperatures, so that discrepancies 

in model results are attributable to internal differences between atmospheric models. All IPCC- 

AR4 AOGCM simulations are forced with a set of boundary conditions determined by scenarios 

of anthropogenic emissions of carbon dioxide (CO2) and other gases that influence the global 

radiation budget. Here we consider twentieth century simulations (20C3M), which use a best 

estimate of historical emissions between roughly 1850 and 2000; and three twenty-first century 

climatic change scenarios (COMMIT, SRESA1B, SRESA2), which represent a range of 

socioeconomic developments and associated emission rates. 

Twentieth century results Mean monthly North American SWE (Figure 1a) and SCE (Figure 

1b) for observations and AMIP-2 AGCMs demonstrate that models capture the mean seasonal 

cycle, there is significant between model variability, and models have a tendency to overestimate 

the ablation rate during spring. The mean spatial pattern of SWE is reasonably well captured by 

the median value of AMIP-2 models, except for an overly smoothed representation of SWE over 

the western cordillera related to model orography, underestimation of SWE over eastern NA 

(figure 2), and the widespread underestimation of SWE during spring.(not shown). 

Between-model variability in IPCC-AR4 AOGCMs (Figure 3) is comparable to that found in 

AMIP-2 AGCMs. Figure 3, which shows ensemble mean time series, also indicates that the 

models disagree with each other, and with observations, on the timing and magnitude of decadal 

scale variations. With regards to the magnitude of decadal-scale variability, the disagreements 

between models and observations are perhaps not as great as they appear. Those models which 

1 Dept. of Geography, Hunter College, New York, NY, USA 

2 Dept. of Earth and Environmental Engineering, Columbia University, New York, NY, USA 

3 Meteorological Service of Canada, Dorval, Quebec, Canada 

293

appear to exhibit smaller decadal scale variability tend to include more ensemble members, and 

those models which exhibit greater decadal scale variability tend to include only one ensemble 

member. In fact, the decadal scale variability of ensemble means are consistently smaller than the 

that of the individual ensemble members, and that of individual members tend to be closer to 

observed values. This indicates that decadal scale variability in these models is due to internal 

dynamics, and not due to external forcings. 

a b 

Figure 1. North American SWE (a) and SCE (b). Observations from Brown et al (2003) indicated with 

asterisks; from Robinson (1993) with crosses. Box and whisker plots indicate model results from 18 

AMIP-2 AGCMs, and are interpreted as follows: middle line shows the median value; top and bottom of 

box show the upper and lower quartiles (i.e. 75 th and 25 th percentile values); and whiskers show the 

minimum and maximum model values. Figure adapted from Frei et al (2005). 

a b 

Figure 2. Seasonal (October-June) mean SWE (mm) regridded to 2.5° x 2.5° latitude - 

longitude resolution. (a) Observed, (b) model anomaly. Regridding was done using linear 

interpolation, and results are plotted on an Albers equal area projection using a logarithmic 

scale for contour lines. In (b), blue areas indicate model underestimation of SWE. Figure 

adapted from Frei et al (2005). 

Twenty first century results At the time of this writing, output from five modeling groups are 

available to evaluate North American SCE variations under the historical emission scenario 

(20C3M) and all three twenty first century scenarios (COMMIT, SRESA1B, and SRESA2). Under 

the COMMIT scenario, greenhouse gas emission rates remain constant at year 2000 values. Under 

the SRES scenarios, emission rates increase either moderately (SRESA1B) or severely (SRESA2). 

As an example, Figure 4 shows the responses of SCE to each emission scenario for one model. 

Although trends vary considerably between models, in all cases the decreasing trends for the 

SRESA1B and SRESA2 scenarios are statistically significant and comparable in magnitude to 

294

each other; and, they both decrease at greater rates than under the COMMIT scenario. For all 

models under the SRESA1B and SRESA2 scenarios, SCE decreases at a greater rate during the 

twenty first century than during the twentieth century. In contrast, under the COMMIT scenario, 

while some models do have weak but significant decreasing trends in SCE, in no model does SCE 

decrease at a greater rate during the twenty first century than during the twentieth century. The 

differences in responses are not proportional to the differences in forcing under these scenarios, 

indicating that non-linear dynamics are influencing the snow cover. 

Discussion and conclusions Significant between-model variability is found in all comparisons 

of GCM snow simulations: e.g. the range of simulated snow mass of North America by AMIP-2 

AGCMs is ±50% of the estimated value. When using a GCM to evaluate potential changes in 

regional to continental scale hydrological variations one must exercise caution. However, the 

median result from all models tends to do a reasonably good job compared to observations. Thus, 

perhaps a “superensemble” of models, when numerous simulations from different models are 

combined, may be an effective method. Preliminary evaluations of AOGCMs suggest that: 

decadal scale variability is associated with internal climatic variations and not with external 

forcings in these models (whether that is true in the real climate system is unknown); and, 

decreases in snow extent are expected under realistic scenarios of future emissions, although the 

precise response of the snow cover to possible future climate variations may be non linear. 

Figure 3. Nine-year running mean of twentieth century January NA-SCE for 11 IPCC-AR4 ensemblemean 

model simulations and for reconstructions of observed variations. NA-SCE is defined as the 

fraction of the land area from 20° N - 90°N and 190° E - 340° E covered with snow. The legend shows 

the model number, which corresponds to model numbers and ensemble sizes shown in Table 2; “B” and 

“F” correspond to Brown (2000) and Frei et al. (1999), respectively. Figure adapted from Frei and Gong 

(2005). 

Fractional Snow Covered Area 

0.9 

0.85 

0.8 

0.75 

0.7 

0.65 

0.6 

0.55 

MRI-CGCM2.3.2 

0.5 

1900 1950 2000 

Year 

2050 2100 

Figure 4. Annual time series (thin line), overlaid with nine-year running means (thick line), of ensemblemean 

January NA-SCE for 20 th and 21 st century scenarios from one AOGCM. 20C3M, COMMIT, 

SRESA1B and SRESA2 scenarios denoted by black, red, blue and green lines, respectively. Figure 

adapted from Frei and Gong (2005). 

295

References 

Brown, R. D. (2000). Northern hemisphere snow cover variability and change, 1915-1997. 

Journal of Climate 13(13): 2339-2355. 

Brown, R. D., B. Brasnett and D. A. Robinson (2003). Gridded North American monthly 

snow depth and snow water equivalent for GCM evaluation. Atmosphere-Ocean 41(1): 1-14. 

Frei, A., R. Brown, J. A. Miller and D. A. Robinson (2005). Snow mass over North America: 

observations and results from the second phase of the Atmospheric Model Intercomparison 

Project (AMIP-2). Journal of Hydrometeorology: in press. 

Frei, A. and G. Gong (2005). Decadal to century scale trends in North American snow extent 

in coupled Atmosphere-Ocean General Circulation Models. Geophysical Research Letters: in 

review. 

Frei, A., J. A. Miller and D. A. Robinson (2003). Improved simulations of snow extent in the 

second phase of the Atmospheric Model Intercomparison Project (AMIP-2). Journal of 

Geophysical Research - Atmospheres 108(D12): 4369, doi:4310.1029/2002JD003030. 

Frei, A., D. A. Robinson and M. G. Hughes (1999). North American snow extent: 1900-1994. 

International Journal of Climatology 19: 1517-1534. 

Robinson, D. A. (1993). Hemispheric snow cover from satellites. Annals of Glaciology 17: 

367-371. 

296

297 


Waterloo, ON, Canada 2005 

The Impact of Patchy Snow Cover on Snow Water Equivalent 

Estimates Derived from Passive Microwave Brightness 

Temperatures Over a Prairie Environment 

ABSTRACT 

KIM R. TURCHENEK 1 , JOSEPH M. PIWOWAR1, AND CHRIS DERKSEN 2 

Considerable seasonal and inter-annual variation in the physical properties and extent of snow 

cover pose problems for obtaining reliable estimates of quantities and characteristics of snow 

cover both from conventional and satellite measurements (Goodison and Walker, 1994; Goita et 

al., 2003). In spite of these challenges, the Climate Research Branch of the Meteorological Service 

of Canada (MSC) has developed a suite of algorithms to derive snow water equivalent (SWE) 

estimates from remotely sensed passive microwave imagery (Goodison and Walker, 1994; 

Derksen et al., 2002; Goita et al., 2003). These algorithms work particularly well over open prairie 

environments under the assumption of large areas of consistent snow cover (Derksen et al., 2002). 

While studies have documented underestimation in passive microwave estimates of snow extent in 

marginal areas when compared to optical satellite data (Derksen et al., 2003b), the accuracy in 

SWE retrievals under variable and patchy snow conditions is not well understood. 

In an effort to better understand how a variable and patchy snow cover impacts remotely sensed 

SWE retrievals, a field-based experiment was conducted over a patchy snow covered area in 

February 2005. A systematic sampling strategy was developed over a 1600 km 2 area in southern 

Saskatchewan near a calibration/validation flight line used for algorithm development in the 1980s 

(Goodison and Walker, 1994). Land cover at the sampling sites included fallow and stubble fields, 

pastures, and shelter belts. A large number of sampling sites contained snow pack layers that 

included one or more ice lenses. 

We verify that the continuous snow cover assumption embedded in the MSC passive microwave 

SWE algorithm does not produce acceptable results over a patchy snow cover. Several in-situ 

observations that appear to play an important role in affecting the satellite passive microwave data 

over a variable snow cover include the presence or absence of an ice lens, the fractional snow 

covered area, snow depth, and the ground temperature. 

Keywords: snow cover, snow water equivalent, passive microwave, remote sensing 

INTRODUCTION 

Microwave radiation is naturally emitted everywhere on the Earth. Its measurable intensity 

varies from place to place based on soil types, land covers, snow pack characteristics, and other 

variables (Goita et al., 1997; Sokol et al., 1999). At microwave frequencies above 15 GHz, the 

emitted radiation is scattered by snow particles as it passes through the snow pack (Goita et al., 

1997). Increasing the snow pack depth or grain size results in an increase in scattering and 

1 Department of Geography, University of Regina, Regina, Saskatchewan, S4S 0A2 

2 Climate Research Branch, Meteorological Service of Canada, Downsview, Ontario M3H 5T4

subsequent lower microwave brightness temperatures when measured above the surface (Goita et 

al., 1997). Wet and/or dense snow packs however, decrease the amount of scattering, and produce 

near blackbody emissions (Sokol et al., 1999). Complex snow packs, containing numerous layers 

at different densities, selectively influence microwave radiation, producing inconsistent 

measurements. 

Radiation recorded by a microwave sensor is expressed as a brightness temperature (TB) in 

Kelvin units. One parameter commonly derived from remotely sensed brightness temperatures is 

snow water equivalent (SWE), which is the amount of water stored in a snow pack that is available 

upon melt. 

Intensive research with passive microwave TB data has focused on empirically derived 

algorithms used to estimate SWE for validation of ground-truth observations (Goodison and 

Walker, 1994; Goita et al., 1997; Derksen et al., 2002; 2003a). Currently, the Meteorological 

Service of Canada (MSC) employs a suite of linear algorithms to retrieve SWE estimates from 

passive microwave sensors. The MSC algorithms vary according to land cover, with different 

coefficients used for open prairie, deciduous forest, coniferous forest, and sparse forest (Goita et 

al., 1997; Singh and Gan, 2000; Derksen et al, 2003a; 2003b). For example, the algorithm used to 

derive SWE over the open prairie is a vertically polarized TB gradient ratio (Goodison and Walker, 

1994) defined as: 

SWE (mm) = –20.7 – (37v – 19v) * 2.59 (1) 

Variables 37v and 19v are the brightness temperatures acquired from vertically polarized 

frequencies of 37 and 19 GHz, while the coefficients 2.59 and –20.7 are the slope and intercept of 

the best-fit regression line found between ground and airborne brightness temperatures (Goodison, 

1989). Numerous studies have found SWE estimates derived from this algorithm to be within 

±10–20 mm of in-situ observations (Goodison and Walker, 1994; Derksen et al., 2002; 2003b), 

however, the MSC algorithm assumes a complete snow cover and the effects of patchy snow cover 

on SWE estimates are not well understood. 

To help understand the relationship between patchy snow covers and remotely sensed SWE 

estimates, a field campaign to measure SWE in a highly variable snow pack was conducted from 

February 21 st to 23 rd , 2005 in southern Saskatchewan. The objectives of this study were to: 

i) Compare remotely sensed SWE estimates against ground-truth measurements; 

ii) Determine how well the MSC prairie SWE algorithm performs over a patchy snow cover; 

and 

iii) Identify in-situ variables that significantly influence passive microwave SWE estimates 

over a patchy snow cover. 

STUDY AREA 

The study area is located approximately 100 km south of Regina, Saskatchewan around the 

town of Radville and the villages of Pangman and Ceylon (Figure 1). This area was selected 

because previous research had been performed in this area by Environment Canada. In addition, 

the snow cover typical of this area has been found to have regions of both patchy and complete 

snow-cover (Goodison and Walker, 1994; Turchenek, 2004). 

The Missouri Coteau, a remnant glacial moraine, cuts across the southwest portion of the study 

area, forming a low, rolling topography. Although the natural vegetation in this area consists of 

short grasses, a large portion of the land is under agricultural production, where wheat and other 

grains are farmed. Pockets of trees and shrubs create shelter belts in areas with higher moisture 

supply. 

Relatively short warm summers and long cold winters are characteristic of Saskatchewan’s 

prairies (Hare and Thomas, 1979). With few perennial streams, much of the region’s water supply 

298

comes in the form of precipitation. However, the annual precipitation in the region is relatively 

low, and evapotranspiration usually exceeds the annual precipitation, creating an average water 

deficit by middle to late summer (Laycock, 1972). Most annual precipitation falls in the summer, 

while February is usually the driest month (Hare and Thomas, 1979). 

Figure 1. Study area 

The winter season provides relatively low amounts of snow. Extended periods of cold, clear 

weather are interrupted by occasional blizzards with gusting winds. Warming periods are frequent 

in the early and late winter (Laycock, 1972; Hare and Thomas, 1979; Walker et al., 1995). Wind 

re-distributes the snow cover by removing snow from one area and depositing it in another. 

Similarly, warming periods also impact the snow cover through freeze–thaw processes (Laycock, 

1972; Walker et al., 1995). As the air and ground temperatures rise, the snow pack melts. When 

the snow pack re-freezes, it becomes denser and shallower. Thus, along with topographic effects 

and changes in vegetation, weather systems can impart a considerable variability in snow pack 

depths and densities. 

REMOTE SENSING DATA 

Four sets of coincident remote sensing data were analyzed (Table 1). Three data sets were 

derived from the brightness temperatures collected from the Advanced Microwave Scanning 

Radiometer for NASA’s Earth Observing System (AMSR-E). The first AMSR-E data set includes 

TB re-sampled to the 12.5 km Equal Area Scalable Earth Grid (EASE-Grid) (Armstrong and 

Brodzik, 1995). The second includes AMSR-E TB re-sampled to the 25 km EASE-Grid and the 

third AMSR-E data set includes non-gridded TB swath data. For comparison, a fourth data set of 

SWE estimates derived from Special Sensor Microwave/Imager (SSM/I) brightness temperatures 

re-sampled to the 25 km EASE-Grid was included. Although data were acquired for each day of 

the field campaign, the only remote sensing data analyzed here were those collected on the first 

day of the campaign (Feb 21 st , 2005). Future research will incorporate the coincident remote 

sensing data from each date of the field campaign. 

299

Table 1: Remote sensing data used within this study 

Data Set Label Spatial Resolution Description 

SWE estimates derived from 

AMSR-E TB re-sampled to 12.5 km 

1) 12.5k_AMSR-E SWE 12.5 km 

EASE-Grid 

18.7v TB obtained from AMSR-E 

12.5k_AMSR-E 18.7v 12.5 km 

re-sampled to 12.5 km EASE-Grid 


12.5k_AMSR-E 36.5v 12.5 km 

re-sampled to 12.5 km EASE-Grid 


AMSR-E TB re-sampled to 25 km 

2) 25k_AMSR-E SWE 

25 km 

EASE-Grid 


25k_AMSR-E 18.7v 

25 km 

re-sampled to 25 km EASE-Grid 


25k_AMSR-E 36.5v 

25 km 

re-sampled to 25 km EASE-Grid 


AMSR-E TB that have not been re- 

3) Swath_AMSR-E SWE 24 km x 12 km sampled 


Swath_AMSR-E 18.7v 24 km 

that have not been re-sampled 


Swath_AMSR-E 36.5v 12 km 

that have not been re-sampled 


SSM/I TB re-sampled to 25 km 

4) 25k_SSM/I SWE 

25 km 

EASE-Grid 

METHODS 

The analysis procedure was divided into four steps: i) collection of the ground-truth 

observations, ii) processing of the data sets, iii) comparison with the remotely sensed SWE 

estimates, and iv) statistical tests. These are discussed in the following subsections. 

In-situ Data Collection 

The study area was systematically divided into 25 km square grid cells, with the centre of each 

cell located at 5 km intervals. Field measurements were made nearest the centre of each grid cell 

as possible. The field campaign was concentrated into a three-day period to minimize changes in 

snow pack conditions due to melt or fresh snowfall. 

Two teams, of 3–4 surveyors each, collected a total of 88 ground observations from 84 sampling 

sites that covered an area of 1600 km 2 . As a control, four of the sites were sampled on consecutive 

days to ensure data consistency. Included in the 84 sampling sites were 20 sites coincident with an 

established MSC validation/calibration snow course data archive (Flight Line 603). Comparisons 

among these data will be the subject of a future communication. The land coverage of the 

sampling sites included pastures and shelter belts, as well as fallow and stubble fields. The 

ground-truth data collected from each sampling site included: geographic locations, snow pack 

profiles, depth measurements, core samples, air and ground temperatures, site photographs, 

sampling dates and times, and land cover and weather observations. Sampling site locations were 

recorded using global positioning system (GPS) handsets. Since data for differential corrections 

were not available, the positions have an approximate accuracy of 10 metres. 

Approximately 90% of the sampling sites were found to have patchy snow coverage. As such, 

one of the first observations made at each site was a visual assessment of the percentage of snow 

cover. Each team member made this assessment independently and these values were then 

300

averaged to reduce bias in this variable. For sites with complete snow cover, a total of 4 snow core 

samples were collected. This number was reduced proportionally for partially covered sites. For 

example, only 2 cores were collected from sites determined as having 50% snow coverage, and 

just 1 core was collected from sites with 25% snow cover. This rationale was used to satisfy the 

requirement that the cores be taken randomly within each site. Thus, if a site was found to have 

50% snow cover, then the probability of randomly selecting a sampling location containing snow 

is only 50%. In this situation, the 2 core samples that were not actually collected were simply 

given zero values for their core lengths, depths, weights, and densities. 

The core samples were obtained using Eastern Snow Conference (ESC-30) snow core tubes. 

Measurements from the core samples include the actual depths of the snow packs from where the 

cores were removed, the lengths of the cores, and their weights. The lengths and weights of the 

cores were used to calculate the core densities and ground SWE measurements. Snow densities, 

represented as g/cm 3 , were based on the average of the 0 to 4 core samples. 

A snow pit was dug at each sampling site and a detailed snow pack profile was made that 

included the snow pack’s total depth, the number of layers and ice lenses within the snow pack, 

the depth and snow grain size of each layer (using Sears snow crystal screens, labeled with 1–3 

mm grids), a qualitative description of each layer, and the air and snow/ground interface 

temperatures. A total of 16 depth measurements were made around each snow pit using 15-metre 

long ropes as guides for the purpose of consistently collecting the depth measurements from 30metre 

diameter circles. Depth measurements to the nearest one-half centimetre were made using 1metre 

long depth probes. The depth measurements from each site were used to calculate the 

average depths within the sites, which were then used as representative values for the sampling 

sites. The average depths are based on the 16 random depth measurements recorded from the 

circle around the snow pit along with the 0 to 4 depth measurements recorded from the snow core 

samples. 

Other data recorded included the sampling dates and times, weather observations, and land 

cover types. 

In-situ Data Processing 

Two data sets were created from the ground sampled data in order to better understand how 

snow properties over a partial snow cover are manifested in the remotely sensed SWE estimates. 

In the “Snow-Only” data set, only those snow depth measurements that were greater than zero 

were included in the average depth, density, and ground SWE calculations. For example, if a site 

was found to have snow depth readings of 3, 4, 0, 1, and 4 cm, then the average depth for that site 

was recorded as 3.0 cm ( (3 + 4 + 1 + 4) / 4 ), excluding the zero value. Conversely, the same site 

in the “Actual-Conditions” data set would have a mean depth of 2.4 cm ( (3 + 4 + 0 + 1 + 4 / 5) ). 

From each of these data sets, four SWE estimates were calculated. The first ground SWE value, 

“Core_SWE,” represents the SWE calculated by using the mean SWE from the snow cores only. 

The second SWE value, “Derived_SWE,” is representative of the mean value for the Core_SWE 

plus the SWE derived from the 16 depth measurements using the average density for each 

sampling site. The remaining SWE values, “Fractional_Core_SWE” and 

“Fractional_Derived_SWE” are represented by the previous SWE values weighted by the 

Snow_Cover_Percent, respectively. Four SWE values were deemed necessary to investigate the 

most accurate way of representing SWE over a patchy snow cover. 

Remote Sensing Data 

The remote sensing images were re-projected to a UTM projection for analysis in ArcGIS. Each 

pixel centroid was assumed to be a point, and pixel footprints were created using Thiessen 

polygons. The Thiessen polygon algorithm segments the measurement space into polygons such 

that every polygon encloses the region closest to each pixel centroid (O’Sullivan and Unwin, 

2003). Although the algorithm does not produce perfectly squared pixels, the fact that the 

algorithm measures the mid-points between the pixel centroids ensures that the spatial resolutions 

of the remote sensing data sets are preserved. 

301

Following previous research by Goodison and Walker (1995) and Derksen et al. (2002, 2003b), 

remotely sensed SWE estimates were compared to ground SWE measurements. The remotely 

sensed estimates found to be within ±20 mm (the previously determined accuracy of the MSC 

SWE algorithm) of ground measurements were considered as equivalent. SWE estimates found 

not to be within the ±20 mm threshold were considered as anomalous. 

Statistical Comparison Tests 

Z-tests were performed between the results of the Snow-Only and Actual-Conditions data sets to 

determine if there were significant differences in algorithm performance between the two groundtruth 

representations. Linear regression models were developed between the remote sensing data 

(as the dependant variables) and the ground observations from the coincident sampling sites (as 

the independent variables). 

The linear regression outputs were interpreted following the systematic procedure proposed by 

Gupta (2000) (Figure 2). The significance of the model fit is analyzed. The model significance 

explains the deviations of the dependent variables (eg. the SWE estimates and TB). We used a 

model significance of 0.10 (90% confidence level) as the cutoff for model acceptance. Models 

with levels below the 90% confidence level were removed from further analysis. 

The next step in interpreting the regression output is to analyze the Adjusted R 2 value from the 

model summary. This value is sensitive to the addition of irrelevant variables, and is a measure of 

the proportion of the variance in the dependent variables that are explained by the variations of the 

independent variables. For example, an Adjusted R 2 value of 0.500 suggests that 50% of the 

variance in a SWE estimate is explained by the variation in the ground-truth measurements. 

The third interpretation step involves identifying the reliability of the individual coefficients for 

the independent variables. The Beta values included in the coefficients output indicate the 

predicted coefficients for the model along with their standard errors and significances. Similar to 

the significance of the model fit, if a coefficient results in a significance value above 0.10, then it 

can be concluded that the independent variable is not significant at a 90% level of confidence. 

Table 2 provides an instructive example of how an output coefficient table is analyzed. The 

SWE predicted by the model is specified by the Constant’s Beta coefficient, 30.5 mm. The 

standard error of this prediction is 1.9 mm of SWE. The next step is to identify the significance of 

the independent variables’ Beta coefficients. In this example, a coefficient of .149 for the average 

snow pack depth is found to be significant at a 95% level of confidence (sig. = .043), but the 

coefficient of –2.899 for the average density is irrelevant (sig. = .776) towards the predicted SWE 

value. 

Table 2: Example coefficient table output 

Unstandardized Coefficients 

Model 

Beta 

St. Error Significance 

Constant (predicted SWE value) 30.5 1.9 — 

Depth 0.149 0.111 0.043 

Density –2.899 10.148 0.776 

302

RESULTS 

Figure 2. Flow chart of regression analyses 

The results are presented in two parts. First, the remotely sensed SWE estimates are compared 

with the SWE measurements obtained from the coincident sampling sites. This is followed by an 

analysis of the linear regression models. 

In-situ vs. Remotely Sensed SWE Estimates 

The SSM/I and swath AMSR-E SWE data provided the closest estimates to both in-situ data 

sets. This was expected, because: i) the MSC algorithm used to derive SWE estimates was actually 

developed for SSM/I TB, and ii) the swath AMSR-E TB have not been re-sampled, thus, they are 

truer representations of the interaction between the sensor and the ground surface. Tables 3 and 4 

illustrate the number (and percentage in brackets) of sampling sites that were found to be 

equivalent (i.e. within ±20 mm SWE) to the remotely sensed SWE estimates. Table 3 shows the 

results of the Snow-Only data set, while Table 4 presents the results of the Actual-Conditions data 

set. These results also show a slight increase in algorithm agreement when only the amount of 

snow found at each sampling site is included in the ground-truth observations (i.e. Core_SWE 

from Table 3 vs. Table 4). 

303

Table 3: Number of equivalent Snow-Only coincident sites (n=88) 

12.5k 

25k 

Swath 

SWE Calculation SSM/I 

55 

AMSR-E AMSR-E AMSR-E 

Core_SWE 

(63%) 

50 

42 (48%) 20 (23%) 56 (64%) 

Fractional_Core_SWE (57%) 

58 

29 (33%) 11 (13%) 49 (56%) 

Derived_SWE 

(66%) 

45 

32 (36%) 15 (17%) 54 (61%) 

Fractional_Derived_SWE (51%) 24 (27%) 7 (8%) 45 (51%) 

Table 4: Number of equivalent Actual-Conditions coincident sites (n=88) 

SWE Calculation SSM/I 

51 

Core_SWE 

(58%) 

44 

Fractional_Core_SWE (50%) 

45 

Derived_SWE 

(51%) 

33 

Fractional_Derived_SWE (38%) 

12.5k 

AMSR-E 

34 

(39%) 

27 

(31%) 

26 

(30%) 

23 

(26%) 

304 

25k 

AMSR-E 

13 

(15%) 

10 

(11%) 

9 

(10%) 

7 

(8%) 

Swath 

AMSR-E 

52 

(59%) 

45 

(51%) 

49 

(56%) 

40 

(45%) 

By weighting the in-situ SWE values by the percentage of snow cover found at the sites (i.e. 

reading down each column) the agreement with the remote sensing estimates decreases by an 

average of approximately 10% in the Snow-Only data set, and by an average of approximately 7% 

in the Actual-Conditions data set. Further, the remote sensing SWE algorithm generally had a 

higher level of agreement with Core_SWE measurements than with Derived_SWE values. 

Therefore, for a patchy snow cover, it appears that the MSC SWE algorithm had the closest 

agreement with ground SWE measurements based only on the core samples 

Further analyses using only these Core_SWE measurements found that, on average, the remote 

sensing algorithm tended to overestimate the patchy in-situ SWE measurements in all cases (Table 

5). This was not surprising since the remote sensing algorithm was originally derived for a 

complete snow cover. 

Table 5: Mean differences in SWE values between remote sensing estimates and in-situ measurements 

(for core samples only) 

12.5k 

25k 

Swath 

SSM/I AMSR-E AMSR-E AMSR-E 

Snow-Only Core_SWE 4.8 17.7 32.1 5.4 

Actual-Conditions Core_SWE 8.7 21.6 35.9 9.3 

We also wanted to examine the effect that varying land covers had on the spaceborne SWE 

estimates. The ranges in ground SWE measurements were very high, particularly when the 

sampling sites included shelter belts and fallow fields, which were found to have drastically 

different snow conditions than stubble fields and pastures. Table 6 shows that there is little 

difference in the mean SWE values representative of stubble fields (26.1 mm) and pastures (23.0 

mm), but great disparity between these values and fallow fields (1.7 mm) and shelter belts (94.8 

mm).

Table 6: Actual-Conditions SWE, depth, and density measurements by land cover type 

Land Cover n SWE (mm) Depth (cm) Density (g/cm 3 ) 

Stubble 54 

Fallow 15 

Pasture 12 

Shelter Belt 2 

mean = 26.1 

min = 1.3 

max = 66.3 


min = 0 

max = 8.8 


min = 5.1 

max = 44.1 


min = 75.6 

max = 114.0 

305 


min = 1.0 

max = 19.0 


min = 0 

max = 4.7 


min = 1.8 

max = 16.4 


min = 22.7 

max = 30.9 


min = 0.026 

max = 0.511 


min = 0 

max = 0.265 


min = 0.110 

max = 0.364 


min = 0.267 

max = 0.316 

Linear Regression Models 

Linear regressions were performed using the Statistical Package for the Social Sciences (SPSS) 

software. The regressions were performed using the remotely sensed SWE estimates and 

brightness temperatures as the dependent variables and the in-situ observations as the independent 

variables. Analyses were again performed between the Snow-Only and Actual-Conditions data 

sets. Table 7 lists and describes the in-situ observations used in all of the regression models. 

Table 7: Measured snow properties used in linear regression models 

Independent variable Variable Type Description 

Percentage of snow cover found at 

Percent_Snow_Cover Ratio 

each sampling site. 

Depth Ordinal Mean snow pack depth of each site. 

Density Ratio Mean snow pack density of each site. 

Air_Temp Interval Air temperature recorded at each site. 

Snow/ground interface temperature 

Ground_Temp Interval recorded from the snow pit. 

Number of snow pack layers found in 

Num_Layers Ordinal the snow pit. 

Land_Cover Nominal Type of land cover. 

Indicates whether or not one or more 

ice lenses were found in the snow pack 

Ice_Lens Binary/Nominal of the snow pit. 

Number of ice lenses found in the 

Num_Lenses Ordinal snow pack of the snow pit. 

Total ice lens thickness found within 

Total_Lens_Thickness Ordinal the snow pack of the snow pit. 

The linear regression analyses found that snow pack densities from the Actual-Conditions data 

set were significantly positively correlated with SWE estimates derived from the Swath AMSR-E 

data. This was expected since dense and complex snow packs have been shown to amplify 

scattering and tend to produce remotely sensed SWE overestimates (Sokol et al., 1999). The 

Swath AMSR-E imagery was the only remote sensing data set to show such a statistically 

significant correlation, likely because — since they had not been re-sampled to the EASE-Grid — 

these data were closer representations of the original Earth radiances originally detected by the 

sensor.

Table 8 shows the regression results between the Swath AMSR-E SWE estimates and both the 

Snow-Only and Actual-Conditions data sets. The Model Fit shows that both data sets match the 

satellite estimates at 99% levels of confidence. From the Unstandardized Beta Coefficients we see 

that the only significant variables are the percentage of snow cover, and whether or not one or 

more ice lenses were found in the snow pit. Interestingly, density in the Snow-Only data set 

appears not to make a significant contribution, while it is found to be significant at a 90% level of 

confidence in the Actual-Conditions data set. The Model Summaries indicate that the proportion 

of the variance in the satellite estimates that is explained by the ground observations are just 

15.8% and 17.9% for the Snow-Only and Actual-Conditions data sets, respectively. 

Table 8: Regression results between Swath AMSR-E SWE estimates and in-situ observations 

(significant coefficients are shown in bold italics) 

Swath_AMSR-E SWE Snow-Only Actual-Conditions 

1) Model Fit 

2) Model Summary (Adjusted 

0.008 0.004 

R 2 ) 0.158 0.179 

St. 

St. 

3) Coefficients Beta Error Sig. Beta Error Sig. 

� Constant (Predicted SWE) 32.646 2.280 — 32.552 2.140 ⎯ 

� Snow_Cover_Percent 8.508 4.276 0.050 10.079 4.310 0.022 

� Depth 0.143 0.220 0.518 0.187 0.216 0.390 

� Density –6.009 6.646 0.369 19.131 11.357 0.096 

� Air_Temp –0.050 0.306 0.870 0.026 0.302 0.932 

� Ground_Temp –0.162 0.492 0.743 –0.257 0.489 0.601 

� Num_Layers 0.371 1.500 0.805 0.691 1.498 0.646 

� Land_Cover –1.555 1.376 0.262 –1.490 1.355 0.275 

� Ice_Lens –7.354 3.369 0.032 –6.738 3.267 0.043 

� Num_Lenses 0.563 3.205 0.861 –0.151 3.204 0.963 

� Total_Lens_Thickness 0.108 0.647 0.868 0.214 0.643 0.740 

Similar regressions were run between all of the remote sensing and in-situ data sets. The results 

are summarized in Table 9. Regression models derived for the 12.5k AMSR-E 18.7v and 25k 

AMSR-E SWE data were not statistically significant. Regressions from the AMSR-E TB found that 

the 18.7v TB resulted in having more significant variables than those collected from the 36.5v TB. 

While the Snow_Cover_Percent, and Ice_Lens variables were found to be significant in both TB 

regressions, Depth and Ground_Temp were found to also be significant in the 18.7v TB regression. 

With the 12.5k AMSR-E data it is interesting to note that in comparison to the non-gridded 

Swath_AMSR-E analyses a completely different set of variables, except for the binary variable 

Ice_Lens, was found to be significant. The significant variables in this data set include: Depth, 

Air_Temp, Land_Cover, and Ice_Lens. However, as these data have been re-sampled, there is less 

confidence in these regression results compared to those of the swath results. Unlike the 12.5k 

AMSR-E regression results, the 25k AMSR-E 18.7v TB were found to be significant, and the 36.5v 

TB were marginally significant. The 25k SSM/I SWE regressions produced nearly identical results 

between the Snow-Only and Actual-Conditions data sets. 

306

307 

Table 9: Regression results between remotely sensed SWE estimates and in-situ observations (· denotes a statistically significant correlation) 

Swath 

AMSR 

-E SWE 

S A 

-O -C 

Swath 

AMSR 

-E 18.7v 

S A 

-O -C 

Swath 

AMSR 

-E 36.5v 

S A 

-O -C 

12.5k 

AMSR-E 

SWE 

S A 

-O -C 

12.5k 

AMSR-E 

18.7v 

S A 

-O -C 

12.5k 

AMSR-E 

36.5v 

S A 

-O -C 

25k 

AMSR-E 

SWE 

S A 

-O -C 

25k 

AMSR-E 

18.7v 

S A 

-O -C 

25k 

AMSR-E 

36.5v 

S A 

-O -C 

25k 

SSM/I 

SWE 

S A 

-O -C 

1) Model Fit · · · · · · · · · · · · · · · · 

2) Model Summary 

(Adjusted R 2 0 0 0 0 0 0 0 0 

0 0 

0 0 0 0 0 0 

) 

.2 .2 .3 .3 .2 .3 .3 .3 

.3 .3 

.3 .3 .1 .1 .1 .1 

3) Coefficients 

� Snow_Cover_Percent 

� Depth 

� Density 

� Air_Temp 

� Ground_Temp 

� Num_Layers 

� Land_Cover 

� Ice_Lens 

� Num_Lenses 

� Total_Lens_Thickness 

· · 

· 

· · 

· · 

· 

· · 

· 

· 

· 

· 

· 

· 

· 

· 

· 

· 

· 

· 

· 

· · 

· 

· 

· 

· 

· 

· 

· 

· 

· · 

· 

· 

· 

· 

· 

· 

· 

· 

S-O: Snow-Only 

A-C: Actual Conditions

CONCLUSIONS AND DISCUSSION 

Although statistically significant models were established between many of the remote sensing 

and in-situ data sets, the proportion of the variance in the satellite estimates that could be 

explained by the ground observations (i.e. the Model Summaries) was, at most, 0.31. This 

suggests that either the ground data are insufficient for deriving SWE from spaceborne passive 

microwave observations or that the remote sensing data were inappropriate. We know from 

previous research (reviewed earlier) that it is possible to obtain reliable SWE estimates through 

remote sensing, so we must conclude that there were problems with remote sensing data we used 

in this experiment. Specifically, the continuous snow cover assumption embedded in the MSC 

passive microwave SWE algorithm does not produce acceptable results over a patchy snow cover. 

The poor performance of the MSC SWE algorithm for each remote sensing data set evaluated 

confirms that the algorithm fails under patchy and variable snow conditions. 

In spite of the poorly articulated regression models, there were several in-situ observations that 

appear to play an important role in affecting the satellite passive microwave data. The presence or 

absence of an ice lens in the snow pack was consistently identified as a significant coefficient in 

the regression analyses. Other observations that may prove to be useful include the percent snow 

cover, snow depth, and the ground temperature. These will need to be investigated further. 

Consideration of patchy snow cover is challenging from a ground sampling perspective, 

however this study shows that the actual conditions found at each sampling site must be 

incorporated in ground-truth data sets when collecting observations over a partial snow cover. 

Subsequent analysis will focus on using optical data to determine snow cover fraction within a 

passive microwave grid cell to greater quantify the impact of patchy snow cover. 


Support from Environment Canada’s CRYSYS (Cryosphere System in Canada) research 

initiative is greatly appreciated. Special thanks are extended to Natasha Neumann, Arvids Silis, 

and Peter Toose (all from the Meteorological Service of Canada) for equipment and data support. 

The EASE-Grid brightness temperatures were obtained from MSC through the EOSDIS National 

Snow and Ice Data Center Distributed Active Archive Center (NSIDC DAAC), University of 

Colorado at Boulder. Acknowledgements are also extended to Aaron Fedje, Kari Geller, Kathie 

Legault, Mark Otterson, Greg Peterson, Susan Rever, and Mauricio Jimenez Salazar (all of the 

University of Regina), and Michelle Yaskowich (Nature Saskatchewan) for collection of in-situ 

measurements and observations. 

REFERENCES 

Armstrong, R. L. and M. J. Brodzik, 1995. An Earth-gridded SSM/I data set for cryospheric 

studies and global change monitoring, Advances in Space Research 16(10): 155-163. 

Derksen, C., A. Walker, E. LeDrew, B. Goodison, 2002. Time-series analysis of passivemicrowave-derived 

central North American snow water equivalent imagery, Annals of 

Glaciology, 34: 1-7. 

Derksen, C., A. Walker, B. Goodison, 2003a. A comparison of 18 winter seasons of in situ and 

passive microwave-derived snow water equivalent estimates in Western Canada. Remote 

Sensing of Environment, 88: 271-282. 

Derksen, C., and A. Walker, E. LeDrew, B. Goodison, 2003b. Combining SMMR and SSM/I Data 

for Time Series Analysis of Central North American Snow Water Equivalent. Journal of 

Hydrometeorology, 4: 304-316. 

Goita, K., A.E. Walker, B.E. Goodison, and A.T.C. Chang, 1997. Estimation of Snow Water 

Equivalent in the Boreal Forest Using Passive Microwave Data. Proceedings, GER '97 

(International Symposium: Geomatics in the Era of Radarsat), Ottawa, Canada, May 25-30, 

1997. CD-ROM published by Natural Resources Canada and National Defence. 

308

Goita, K., A. Walker, B. Goodison, 2003. Algorithm development for the estimation of snow 

water equivalent in the boreal forest using passive microwave data. International Journal of 

Remote Sensing 24(5): 1097-1102. 

Goodison, B., 1989. Determination of areal snow water equivalent on the Canadian Prairies using 

passive microwave satellite data. Proceedings, 1989 International Geoscience and Remote 

Sensing Symposium, Vancouver. 3:1243-1246. 

Goodison, B.E. and A.E. Walker, 1994. Canadian development and use of snow cover information 

from passive microwave satellite data. In Passive Microwave Remote Sensing of Land- 

Atmosphere Interactions, Choudhury, B.J., Y.H. Kerr, E.G. Njoku, and P. Pampaloni (eds.), 

VSP, Utrecht, The Netherlands, 245-262. 

Gupta, V., 2000. Regression explained in simple terms. Obtained online on April 24, 2005, from 

http://www.telecom.csuhayward.edu/~esuess/Links/Software/RegressionExplained/regression_ 

explained.doc 

Hare, F.K. and M.K. Thomas, 1979. Climate Canada, 2 nd Ed., Canada: Wiley. 

Laycock, Arleigh H., 1972. The Diversity of the Physical Landscape. In The Prairie Provinces, 

Smith, P.J. (ed.), University of Toronto Press. 

O’Sullivan, David, and David J. Unwin, 2003. Geographic Information Analysis, Hoboken, NJ: 

Wiley. 

Singh, Purushottam Raj and Thian Yew Gan, 2000. Retrieval of Snow Water Equivalent Using 

Passive Microwave Brightness Temperature Data. Remote Sensing of Environment, 74: 275- 

286. 

Sokol, J., T.J. Pultz, and A.E.Walker, 1999. Passive and Active Spaceborne Microwave Remote 

Sensing of Snow Cover. Proceedings, 4 th International Spaceborne Remote Sensing 

Conference/21 st Canadian Symposium on Remote Sensing. 

Turchenek, Kim R., 2004. Validation of Snow Water Equivalents Derived from Passive 

Microwave Brightness Temperatures over a Prairie Environment. Unpublished Undergraduate 

Thesis, submitted to and accepted by the Department of Geography at the University of Regina, 

Saskatchewan. 

Walker, Anne, Barry Goodison, Michael Davey and David Olson, 1995. Atlas of Southern 

Canadian Prairies Winter Snow Cover from Satellite Passive Microwave Data: November 

1978 to March 1986, Downsview, Canada: Atmospheric Environment Service, Environment 

Canada. 

309


310

TO ERR IS HUMAN, TO FORGET IT WOULD BE DIVINE. 

SNO-FOO 2006–Andrew Klein 

The Eastern Snow Conference annually bestows this award on some poor hapless snow and ice researcher, who, 

while striving to push back the frontiers of science, is overcome by a bout of “bone-headedness,” ill planning, or just 

plain bad luck—in military parlance, referred to as SNAFU. 

This year’s winner has made countless questionable decisions when it comes to snow research. Although he 

professes to know something about mapping snow and ice from satellites, his geographic knowledge of snow must 

be questioned. He seemed blissfully unaware, when moving to Texas, that the state isn’t all that well known for 

having snow to study. He also seems to think that the tropics are the world’s hotbed for glaciers, which he appears to 

strenuously study while hiding in his air-conditioned office from the Texas summer heat. 

This individual claims that he actually conducts fieldwork in Antarctica. However, it is rumored (though never 

proven) that when one of his sampling sites actually is snow-covered, he moves it to a snow-free location so he can 

more easily sample the soil—and he claims to consider himself to be a snow scientist! 

None of this, however, comes close to the egregious mistakes this year’s winner has made while conducting 

ESC business. The ESC, for reasons still not fully understood, placed in the hands of this individual its most public 

face to the world, its Web site. On numerous occasions, ESC individuals have frantically pointed out mistakes in the 

“content” this individual has chosen to post on the organization’s Web site for the entire world to see. Yet knowing 

full well this individual’s organizational abilities, the ESC executive actually placed the organization of its annual 

meeting in his hands…! 

It probably is not surprising, therefore, that in the course of producing the meeting materials, this year’s winner 

has not once, but multiple times, shown complete and utter disrespect for the holder of the highest office in the 

ESC—its President. On numerous occasions, the ESC president has shaken his head at the “interesting” ways his 

name has been spelled in the program and on the Web site by this year’s award winner. Is it Claude Dugay, Claude 

Dugaey, or maybe Claude Duguay? 

In fact, this year’s winner probably can’t even be counted on to spell the name of this award correctly on the 

Web site. So, Andrew Klein, is it the SNO-FOO or SNOWFOO Award that has been bestowed upon you? 

311

Download the entire proceedings as an Adobe PDF - Eastern Snow ...

Create successful ePaper yourself

Delete template?

Save as template?