15.02.2013 Views

Proceedings of the 19 GIS Research UK Annual Conference ...

Proceedings of the 19 GIS Research UK Annual Conference ...

Proceedings of the 19 GIS Research UK Annual Conference ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> <strong>19</strong> th <strong>GIS</strong> <strong>Research</strong> <strong>UK</strong><br />

<strong>Annual</strong> <strong>Conference</strong><br />

University <strong>of</strong> Portsmouth<br />

with<br />

Ordnance Survey<br />

27 th to 29 th April


<strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> <strong>19</strong> th <strong>GIS</strong> <strong>Research</strong> <strong>UK</strong><br />

<strong>Annual</strong> <strong>Conference</strong><br />

University <strong>of</strong> Portsmouth<br />

with<br />

Ordnance Survey<br />

27 th to 29 th April 2011<br />

Editors:<br />

Ca<strong>the</strong>rine Emma Jones, Lecturer, Dept <strong>of</strong> Geography, University <strong>of</strong> Portsmouth<br />

Alastair Pearson, Principal Lecturer, Dept <strong>of</strong> Geography, University <strong>of</strong> Portsmouth<br />

Glen Hart, <strong>Research</strong> Manager, Ordnance Survey<br />

Nick Groome, External <strong>Research</strong> Manager, Ordnance Survey<br />

P a g e | I


<strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> <strong>19</strong> th <strong>GIS</strong> <strong>Research</strong> <strong>UK</strong><br />

<strong>Annual</strong> <strong>Conference</strong><br />

University <strong>of</strong> Portsmouth with Ordnance Survey<br />

27th to 29th April 2011<br />

©2011 papers, except where indicated<br />

All rights reserved. The copyright on each <strong>of</strong> <strong>the</strong> papers published in <strong>the</strong>se proceedings<br />

remains with <strong>the</strong> author(s). No part <strong>of</strong> <strong>the</strong>se proceedings may be reprinted or reproduced<br />

or utilised in any form by any electronic, mechanical or o<strong>the</strong>r means without permission in<br />

writing from <strong>the</strong> relevant authors.<br />

P a g e | II


WELCOME<br />

The <strong>GIS</strong>R<strong>UK</strong> annual conference is now in its <strong>19</strong> th Year and for <strong>the</strong> first time in its<br />

history comes to Portsmouth in conjunction with <strong>the</strong> Ordnance Survey. As in previous<br />

years <strong>the</strong> conference is going from strength to strength with a wide range <strong>of</strong> papers, topics and<br />

keynote speakers. This year <strong>the</strong> topics <strong>of</strong> <strong>the</strong> papers cover a diverse range including urban modelling,<br />

historical <strong>GIS</strong> and Open Source <strong>GIS</strong> and <strong>the</strong> GeoWeb. We would like to thank Pr<strong>of</strong>essor David Martin<br />

and Dr Mordechai (Muki) Haklay for agreeing to deliver <strong>the</strong> keynote speeches.<br />

<strong>GIS</strong>R<strong>UK</strong> is a unique conference, originally created as <strong>the</strong> <strong>UK</strong>’s national <strong>GIS</strong> research conference in<br />

<strong>19</strong>93. Delegates attend from all corners <strong>of</strong> <strong>the</strong> <strong>UK</strong> toge<strong>the</strong>r with international delegates from <strong>the</strong> EU<br />

and fur<strong>the</strong>r afield. The conference is particularly welcoming to new academics and postgraduate<br />

students providing an encouraging environment for those at <strong>the</strong> beginning <strong>of</strong> <strong>the</strong>ir <strong>GIS</strong> research<br />

career.<br />

The <strong>GIS</strong>R<strong>UK</strong> conference has <strong>the</strong> following aims:<br />

� Act as a focus for <strong>GIS</strong> <strong>Research</strong> in <strong>the</strong> <strong>UK</strong><br />

� To provide a mechanism for <strong>the</strong> announcement and publication <strong>of</strong> <strong>GIS</strong> research<br />

� To act as an intermediary forum for <strong>the</strong> discussion <strong>of</strong> research ideas<br />

� To promote collaboration amongst researchers from diverse parent disciplines<br />

� To provide a framework in which postgraduate students can see <strong>the</strong>ir work in a national<br />

context<br />

All <strong>the</strong> papers submitted to <strong>the</strong> conference were reviewed by appropriate reviewers in <strong>the</strong><br />

discipline. The final papers have been revised in light <strong>of</strong> <strong>the</strong> review comments. We would like to<br />

thank all <strong>the</strong> reviewers who took part in this process.<br />

The conference would not be possible without <strong>the</strong> generosity <strong>of</strong> a number <strong>of</strong> sponsors. We are<br />

<strong>the</strong>refore pleased to acknowledge <strong>the</strong>ir contributions <strong>of</strong> awards and financial support which makes<br />

attendance affordable to postgraduate students and academics.<br />

Welcome to Portsmouth and we hope you enjoy <strong>the</strong> conference.<br />

Alastair, Kate, Glen and Nick<br />

<strong>GIS</strong>R<strong>UK</strong> Local Steering Committee<br />

P a g e | III


<strong>GIS</strong>R<strong>UK</strong> National Organising Committee<br />

Adam Winstanley National University <strong>of</strong> Ireland, Maynooth<br />

Andrew Lovett University <strong>of</strong> East Anglia<br />

Bruce Gittings University <strong>of</strong> Edinburgh<br />

Christine Dunn University <strong>of</strong> Durham<br />

David Fairburn University <strong>of</strong> Newcastle<br />

Steve Wise University <strong>of</strong> Sheffield<br />

Duncan Whyatt Lancaster University<br />

Jane Drummond University <strong>of</strong> Glasgow<br />

Jeremy Morley University <strong>of</strong> Nottingham<br />

Jo Wood City University<br />

Ka<strong>the</strong>rine Arrell University <strong>of</strong> Leeds<br />

Muki Haklay University College London<br />

Nick Mount University <strong>of</strong> Nottingham<br />

Chris Holcr<strong>of</strong>t AGI<br />

Peter Halls University <strong>of</strong> York<br />

Glen Hart Ordnance Survey<br />

<strong>GIS</strong>R<strong>UK</strong> Local Organising Committee<br />

Alastair Pearson Department <strong>of</strong> Geography, University <strong>of</strong> Portsmouth<br />

Ca<strong>the</strong>rine Jones Department <strong>of</strong> Geography, University <strong>of</strong> Portsmouth<br />

Glen Hart Ordnance Survey<br />

Martin Schaefer Department <strong>of</strong> Geography, University <strong>of</strong> Portsmouth<br />

Nick Groome Ordnance Survey<br />

P a g e | IV


Contents<br />

Modelling Fire Evacuation Behaviour Based on Fire Investigation Reports ......................................................3<br />

Reducing Exposure to Air Pollution: A Network Approach............................................................................. 10<br />

Effects <strong>of</strong> Climate Change on Solent Coastal Management, Businesses ......................................................... 16<br />

and Communities .......................................................................................................................................... 16<br />

Assessing climate change vulnerability in Small Island Developing Nations: <strong>the</strong> case <strong>of</strong> Puerto Rico ............ 27<br />

Assessing Data Completeness <strong>of</strong> OpenStreetMap in <strong>the</strong> <strong>UK</strong> through an Automated Matching Procedure for<br />

Linear Data .................................................................................................................................................... 32<br />

Open Source <strong>GIS</strong> for Small Organisations ...................................................................................................... 40<br />

Annotating Spatial Features in OpenStreetMap ............................................................................................ 52<br />

Geographical Information Integration from Disparate Sources ..................................................................... 57<br />

Projecting obesity in small area populations ................................................................................................. 63<br />

Participatory Health Surveys Using Ubiquitous Computing: Gastrointestinal illnesses application case study<br />

...................................................................................................................................................................... 69<br />

Spatio-temporal change in population and Health facility location planning: A case study <strong>of</strong> Ambulance<br />

location planning in Leicestershire. ............................................................................................................... 75<br />

Using a <strong>GIS</strong>-based network analysis to determine Saudi and non-Saudi accessibility to <strong>the</strong> Primary Health<br />

Care Centers in Buraydah City, Kingdom <strong>of</strong> Saudi Arabia ............................................................................... 81<br />

3D Urban Visibility Analysis with Vector <strong>GIS</strong> Data ......................................................................................... 89<br />

A modified two-step floating catchment area technique for measuring transit system accessibility ............. 99<br />

Modelling Local Scale Land-Use: Case Nekala .............................................................................................. 106<br />

The Changing face <strong>of</strong> land use in <strong>the</strong> British countryside since <strong>the</strong> <strong>19</strong>30s .................................................... 114<br />

Map Mash-ups: What looks good must be good? ........................................................................................ 1<strong>19</strong><br />

Using Google Maps to collect spatial responses in a survey environment ................................................... 127<br />

Using Open Source S<strong>of</strong>tware and Data to Teach Spatial Database Skills ...................................................... 135<br />

Use <strong>of</strong> Geographical Information for Non-Visual Perceptualisation ............................................................. 141<br />

Fuzzy Geographical Buffers Revisited .......................................................................................................... 147<br />

Using Morphometric Terrain Properties to Model DEM Error ...................................................................... 153<br />

Effects <strong>of</strong> Candidate Position on Ballot Papers: Exploratory Visualization <strong>of</strong> Voter Choice in <strong>the</strong><br />

London Local Council Elections 2010 .......................................................................................................... 162<br />

Delivering new digital mapping services for schools .................................................................................... 167<br />

Comparing Flickr tags to a geomorphometric classification .......................................................................... 174<br />

Effective Vector Data Transmission and Visualization Using HTML5 ............................................................ 179<br />

The Specification, Estimation, and Testing <strong>of</strong> Real-time Geodemographics using Parallel Graphics Processing<br />

Unit Architecture ......................................................................................................................................... 184<br />

Examining connectivity for recreation in rural areas – an initial methodology ............................................ <strong>19</strong>1<br />

Regional Scale Assessment <strong>of</strong> Cumbria’s Renewable Energy Resource ........................................................ <strong>19</strong>9<br />

Integrating Haptic Feedback to Pedestrian Navigation Applications ......................................................... 205<br />

Hierarchical Structures in Support <strong>of</strong> Dynamic Presentation <strong>of</strong> Multi Resolution Geographic Information for<br />

Navigation in Urban Environments .............................................................................................................. 211<br />

Evaluating <strong>the</strong> potential <strong>of</strong> Web-<strong>GIS</strong> for enhancing public participation in wind farm planning .................. 220


Relevance <strong>of</strong> Volunteered Geographic Information In A Real World Context .............................................. 230<br />

Reproducible <strong>Research</strong>: What Can Geocomputation Achieve? .................................................................... 237<br />

The Use <strong>of</strong> Consensus Clustering in Geodemographics ................................................................................ 246<br />

ESRI vs BREWER: An Evaluation <strong>of</strong> Map Use with Alternative Colour Schemes amongst <strong>the</strong> General Public 254<br />

Towards a Field Toolkit for in-field construction <strong>of</strong> 3D surface models ........................................................ 263<br />

Trajectory Similarity Analysis in Movement Parameter Space ................................................................... 270<br />

Multi-Agent Simulation <strong>of</strong> Drivers Reactions to Unexpected Incidents on Urban Road Networks ............... 280<br />

Trajectory Data Similarity with Metric Data Structures ................................................................................ 286<br />

Decision support system for optimisation <strong>of</strong> marginal rural area development based on <strong>GIS</strong> technology ... 295<br />

Understanding Global Change: Potentials for multi-national historical <strong>GIS</strong> ................................................. 301<br />

A method for detecting geographical cinema circuits using Markov Chains ................................................ 306<br />

Railroads, Visualization and <strong>the</strong> Web: A Progress Report on <strong>the</strong> ‘Digging into Data Challenge’ Project ....... 315<br />

Measuring Urban Environmental Quality across Salford using an integrated Geographic Information Systems<br />

and Remote Sensing approach .................................................................................................................... 320<br />

Study <strong>of</strong> <strong>the</strong> suburbanization <strong>of</strong> <strong>the</strong> Budapest agglomeration with <strong>GIS</strong> methods: characteristics and trends<br />

.................................................................................................................................................................... 328<br />

Rapid screening type flash flood risk assessment in Hungary ...................................................................... 341<br />

Where to draw <strong>the</strong> line? Mapping perceived neighbourhoods onto Lower Super Output Areas ................. 350<br />

Combining Geographical Potential Model with Micro Scale Network Analysis - With An Application Of<br />

Shopping Centres in Helsinki City Region ..................................................................................................... 358<br />

POSTERS<br />

SARMApp: A search and rescue mapping application for mobile devices .................................................... 363<br />

Guidelines for setting up a spatially explicit Bayesian Network for modelling land use decisions ............... 363<br />

Building Generic Quality Indicators for OpenStreetMap .............................................................................. 363<br />

An investigation <strong>of</strong> database driven production <strong>of</strong> atlases .......................................................................... 364<br />

Exploring Road Incident Data with Heat Maps ............................................................................................ 364<br />

Evaluating Household Recycling Waste Management in Malta (within <strong>the</strong> locality <strong>of</strong> Marsascala). ............ 364<br />

Exploring food networks <strong>of</strong> students using a qualitative <strong>GIS</strong> approach ....................................................... 365<br />

The World through Two Eyes: An Exploration <strong>of</strong> Stereo-3D for Geospatial Visualisations ........................... 365<br />

Application <strong>of</strong> <strong>GIS</strong> and DPT in Systematic Surveying, Inventory and Title Registration <strong>of</strong> Properties – A pilot<br />

study <strong>of</strong> Kaneshie town area in Ghana ........................................................................................................ 366<br />

Using <strong>GIS</strong> to model topographic controls on cold air drainage ..................................................................... 366


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Modelling Fire Evacuation Behaviour Based on Fire Investigation Reports<br />

Tyng-Rong Roan 1 , Muki Haklay 1 , Claire<br />

Ellul 1<br />

1 Department <strong>of</strong> Civil, Environmental and Geomatic Engineering, University College London<br />

Gower Street, London WC1E 6BT, United Kingdom t.roan@ucl.ac.uk<br />

ABSTRACT<br />

To simulate realistic human behaviour in fire situations, more accurate results can be obtained if<br />

behaviour is observed from a real situation instead <strong>of</strong> evacuation drills. We propose a new method<br />

to investigate human behaviour in real fire situations by analysing fire investigation reports instead<br />

<strong>of</strong> using traditional methods, such as video recording and questionnaires. An agent- based model is<br />

developed to achieve an improvement <strong>of</strong> realistic evacuation simulation.<br />

KEYWORDS: Fire investigation report, agent-based modelling, evacuation simulation, human<br />

behaviour<br />

1.Introduction<br />

Disasters happen almost every day all over <strong>the</strong> world and can be classified into natural disasters and<br />

man-made disasters. Man-made disasters, such as fires, attacks, or stampedes, are difficult<br />

to prevent when compared to natural disasters. Indeed, technology can now be used to alert people<br />

to prepare in advance before natural disasters occur. For example, people could be asked to<br />

evacuate before a flood or volcanic eruption and could prepare for a cold wave following a wea<strong>the</strong>r<br />

warning. In <strong>the</strong> case <strong>of</strong> man-made disasters, people generally only evacuate an environment after an<br />

emergency situation happens. In <strong>the</strong>se cases, <strong>the</strong> public expects a higher level <strong>of</strong> safety, faster<br />

rescue, and safe evacuation routes, due to <strong>the</strong> unpredictable nature <strong>of</strong> <strong>the</strong>se disasters. More<br />

specifically, according to fire statistics from different countries (Table 1), building fires result in<br />

over 70% <strong>of</strong> deaths and injuries from fires (fire in buildings, road vehicles, and outdoors). For<br />

this reason, this research focuses on building fires, as <strong>the</strong>y are <strong>the</strong> most common man-made disasters.<br />

Country Fire in Buildings/ Total<br />

Fire Incidents<br />

Table 1. Fire statistics in 2009.<br />

Deaths in Buildings/<br />

Total Deaths<br />

Injuries in Buildings/<br />

Total Injuries<br />

United States (1) 480,500/1,348,500 (35.63%) 2,695/3,010 (89.53%) 14,740/17,050 (86.45%)<br />

United Kingdom<br />

(England only) (2)<br />

76,867/252,690 (30.42%) 286/337 (84.87%) N/A<br />

Japan (3) 28,372/51,139 (55.48%) 1,352/1,877 (72.03%) 6,594/7,654 (86.15%)<br />

Sweden (4) 11,080/27,460 (40.35%) 110/1<strong>19</strong> (92.44%) N/A<br />

Finland (4) 6,241/15,057 (41.45%) 96/107 (89.72%) N/A<br />

Source: (1)National Fore Protection Association (2)Communities and Local<br />

Government(3)Fire and Disaster Management Agency(4)Nordstat.net<br />

Increasingly, <strong>the</strong> general public are more aware <strong>of</strong> safety issues, having seen reports <strong>of</strong> many disasters<br />

involving casualties or deaths. Learning from serious disasters, governments can change<br />

Page | 3


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

regulations, improve evacuation strategies, or practice evacuation drills to ensure human safety in<br />

such events. However, evacuation drills cannot recreate a real situation, in that people normally<br />

move patiently and in an orderly fashion during a drill, and indeed is also very dangerous to ask<br />

people act more realistically as people might get injured. As a result, pedestrian computer<br />

simulation has become a useful tool to check if people are safe in an environment.<br />

This paper presents a new method to investigate human behaviour in a real fire situation by analysing<br />

fire investigation reports, which contain a variety <strong>of</strong> useful information, in order to understand how<br />

humans behave in fire events. For <strong>the</strong> purpose <strong>of</strong> understanding how people act in emergency<br />

situations, an agent-based model is developed to model a more realistic evacuation simulation based<br />

on <strong>the</strong>se reports.<br />

2. Human Behaviour Study<br />

To understand how humans behave in emergency situations, video recordings and questionnaires<br />

which are collected from evacuation drills or real fire situations have been used in previous studies.<br />

Observations from videotapes yields information such as number <strong>of</strong> persons, gender, age, location,<br />

mobility, role status (staff or customer), activities, and <strong>the</strong> actions <strong>of</strong> each pedestrian at each time step<br />

(Sandberg, <strong>19</strong>97; Shields and Boyce, 2000). Fur<strong>the</strong>rmore, <strong>the</strong>y can be used to calculate<br />

individual pre-evacuation time and travel speed (Olsson and Regan, 2001). Ano<strong>the</strong>r technique<br />

is video tracking, which detects moving objects using tracking modules (Siebel and Maybank,<br />

2002) or observes abnormal crowd behaviour using a social force model (Mehran et al.,<br />

2009).Kobes et al. (2010) discovered that humans behave differently in smoke situations when<br />

<strong>the</strong>se formed part <strong>of</strong> unannounced fire drills. However, fire drills cannot represent real fire<br />

situations, because in reality people usually demonstrate different behaviour to that seen in a drill,<br />

in particular due to <strong>the</strong> high intensity <strong>of</strong> stress felt while facing a real fire (Proulx, 2002). In<br />

addition, collecting and analysing video recordings from real fire disasters can be difficult due to<br />

difficulties obtaining <strong>the</strong> video data from buildings destroyed by fire, and <strong>the</strong> complexities <strong>of</strong><br />

human behaviour identification in a scene filled with smoke. This means that analysing such<br />

real-life recordings may take a long time and may not be conclusive.<br />

Ano<strong>the</strong>r common method used to understand behaviour in emergency situation is questionnaires,<br />

which usually accompany <strong>the</strong> video recording <strong>of</strong> simulations to provide more information about<br />

human characteristics and o<strong>the</strong>r activities during evacuation (Sandberg, <strong>19</strong>97; Cheng et al., 2009).<br />

In addition, this technique is also used to identify <strong>the</strong> factors which influence human behaviour by<br />

collecting information from post-fire surveys (Zhao et al., 2009). Some pr<strong>of</strong>iles and<br />

information, including demographic information, individual knowledge about <strong>the</strong> specific building,<br />

and how <strong>the</strong>y respond and select egress routes during an evacuation are commonly included in <strong>the</strong><br />

questionnaires. According to <strong>the</strong>se studies, some differences between evacuation drills and real fire<br />

situations can be observed – for example, individual actions and pre-evacuation time changes. The<br />

questionnaires also demonstrate that such simulations do not necessarily closely represent reality, as<br />

people usually provide responses that differ from <strong>the</strong>ir actions.<br />

The current research proposes a new method <strong>of</strong> modelling emergency situations caused by fire – that<br />

<strong>of</strong> analysing fire investigation reports, which provide various details <strong>of</strong> building information, fire<br />

circumstance, and human behaviour. For example, (1) time <strong>of</strong> <strong>the</strong> day when fire occurred<br />

influences <strong>the</strong> activities that occur; (2) building type provides details <strong>of</strong> structural configuration; (3)<br />

building background including floor plans and <strong>the</strong> number <strong>of</strong> potential egress routes provides <strong>the</strong><br />

information <strong>of</strong> where people might chose to evacuate <strong>the</strong> building; (4) fire starting point provides<br />

information on how fire spreads through <strong>the</strong> space as people normally escape in a direction away<br />

from any heat or smoke; (5) total number <strong>of</strong> occupants can help in evaluating <strong>the</strong> possibility <strong>of</strong> a<br />

safe evacuation procedure; (6) location and number <strong>of</strong> deaths show human behaviour before death;<br />

(7) witness statements record occupant responses, feelings, and <strong>the</strong>ir observations within <strong>the</strong><br />

environment<br />

Page | 4


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

3. Methodology<br />

Preliminary work is based on a review <strong>of</strong> twenty fire investigation reports which were recorded by <strong>the</strong><br />

U.S. <strong>of</strong>ficial bodies, such as <strong>the</strong> National Fire Protection Association (NFPA), <strong>the</strong> National Institute<br />

<strong>of</strong> Standards and Technology (NIST) and <strong>the</strong> United States Fire Administration (USFA). These fire<br />

investigation reports record <strong>the</strong> location and number <strong>of</strong> deaths on a floor map and describe occupant<br />

behaviour which can be generalised as follows:<br />

a) People investigate <strong>the</strong> fire/smoke source at <strong>the</strong> beginning <strong>of</strong> <strong>the</strong> event.<br />

b) Panic starts when people notice <strong>the</strong> smoke. As a result, two behaviours are common:<br />

i) Calm: people move in an orderly manner.<br />

ii) Panic: people start pushing, rushing, and searching for alternative paths once <strong>the</strong>y<br />

determine that <strong>the</strong>y are in danger.<br />

c) People are aware <strong>of</strong> <strong>the</strong> fire because <strong>of</strong> <strong>the</strong>y hear <strong>the</strong> fire alarm, see <strong>the</strong> smoke, or smell<br />

burning.<br />

d) People queue/push/die around <strong>the</strong> exit as behaviour changes from calm to panic to death due<br />

to smoke inhalation.<br />

e) People escape from <strong>the</strong> fire/smoke, moving away from <strong>the</strong> hazardous area.<br />

f) Sometimes, people jump from windows or hide in a room when no alternative escape route is<br />

available.<br />

g) Staff play different roles to guests.<br />

i) Fight fire: Staff search for <strong>the</strong> original fire/smoke source and try to put out <strong>the</strong> fire if<br />

possible.<br />

ii) Give commands: Staff lead people in carrying out <strong>the</strong> correct evacuation response. Direct<br />

people out: Staff control pedestrian flow in order to use exits efficiently.<br />

To simulate <strong>the</strong> interaction between pedestrians and environment, an evacuation scenario is built to<br />

model behaviour in a virtual environment using <strong>the</strong> agent-based modelling toolkit, Repast Simphony.<br />

First, a building environment is constructed in terms <strong>of</strong> one <strong>of</strong> <strong>the</strong> disasters from <strong>the</strong> reviewed reports,<br />

and agents are defined with different characteristics, such as age, walking speed, and lung size to<br />

analyse <strong>the</strong> two selected priority results (total evacuation time and location <strong>of</strong> death). Individual<br />

total evacuation time is calculated by personal movement and walking speed which is defined<br />

according to three significant age groups (children, adult and elderly). To simulate how people<br />

faint or die in <strong>the</strong> environment, individual lung size is designed according to breath time while<br />

COHb 1 level > 50% (Goldstein, 2008). O<strong>the</strong>r factors, such as height, gender, education level,<br />

knowledge <strong>of</strong> emergency procedure, familiarity <strong>of</strong> building, pre-evacuation activities and location,<br />

which might influence <strong>the</strong>ir egress selection but have not taken into consideration at this early<br />

stage <strong>of</strong> model development. At this stage, A* and potential field navigation algorithms are<br />

designed to simulate agents moving in <strong>the</strong> environment – A* algorithm uses a distance-plus-cost<br />

heuristic function to determine a route from where each agent starts to its destination and<br />

potential field approach calculates distance values from a target to all <strong>the</strong> possible grids.<br />

Therefore, <strong>the</strong>se two navigation algorithms are tested to see if opposite directions <strong>of</strong><br />

calculations would influence pedestrian movement. Following this step, a set <strong>of</strong> agent<br />

behaviour is developed according to <strong>the</strong> identified behaviour.<br />

The current model simulates pedestrians moving toward main entrances when <strong>the</strong>y hear a fire alarm,<br />

with pedestrian agents changing <strong>the</strong>ir evacuation behaviour to panic if <strong>the</strong>y see smoke. Individual<br />

1 Carboxyhaemoglobin (COHb) is a stable complex <strong>of</strong> carbon monoxide and haemoglobin that forms in red blood cells when<br />

carbon monoxide is inhaled, and hinders delivery <strong>of</strong> oxygen to <strong>the</strong> body.<br />

Page | 5


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

attitudes change to panic when <strong>the</strong> first pedestrian agent sees <strong>the</strong> smoke, assuming <strong>the</strong>y alert each<br />

o<strong>the</strong>r within <strong>the</strong> environment. When <strong>the</strong>y start panicking, <strong>the</strong>y are <strong>the</strong>n programmed to search for<br />

alternative routes to evacuate through, as described in <strong>the</strong> fire reports. For example, some move<br />

towards emergency exits, some find a shelter room and hide <strong>the</strong>re, some wait for rescue and some<br />

may decide to jump from windows. Pedestrian agents move faster while panicking and slow<br />

down due to a decrease in visibility <strong>of</strong> <strong>the</strong> smoke – i.e. as <strong>the</strong>y move fur<strong>the</strong>r away from <strong>the</strong> danger<br />

zone. These agents are programmed to faint or die when <strong>the</strong>y are exposed to smoke, should <strong>the</strong>y<br />

inhale smoke over maximum lung size. O<strong>the</strong>r types <strong>of</strong> agents also exist. Exit agents<br />

control pedestrian flow, which leads <strong>the</strong> pedestrian agents move slower if too many people are<br />

rushing to one door; fire agents control smoke level and <strong>the</strong> interaction between pedestrians and<br />

smoke. Figure 1 shows a time line <strong>of</strong> human behaviour in this fire evacuation model.<br />

4. Results<br />

Figure 1. A time line <strong>of</strong> human behaviour in serious fire disasters.<br />

This model simulates human behaviour in a fire scene as show in Figure 2, which demonstrates five<br />

out <strong>of</strong> seven identified behaviours listed above (Section 3). First, a fire alarm is sounded, but no<br />

one knows if it is a real fire as it starts in an empty room. Then, pedestrian agents start<br />

moving in an orderly manner toward <strong>the</strong> main entrance where <strong>the</strong>y entered <strong>the</strong> building (Figure<br />

2-b). Agent evacuation behaviour changes from calm to panic while <strong>the</strong>y notice <strong>the</strong> blaze or<br />

smoke, and thus <strong>the</strong>y change <strong>the</strong>ir evacuation behaviour, starting to push each o<strong>the</strong>r around <strong>the</strong><br />

exit, selecting alternative exits, waiting for rescue, jumping from windows (Figure 2-c), or finding in<br />

a shelter room (Figure 2- d). Finally, <strong>the</strong> death commonly takes place near an exit or in a refuge<br />

room (Figure 2-e).<br />

Page | 6


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 2: Simulation <strong>of</strong> human behaviour in a fire scenario. (a) Original <strong>of</strong> fire starts (white<br />

cell) and agents‘ positions, three different colours represent three groups <strong>of</strong> ages (magenta:<br />

child, cyan: adult, pink: elderly).(b) Before panic: agents move in an orderly manner toward<br />

<strong>the</strong> main entrance. (c) After panic: agents change <strong>the</strong>ir destination toward alternative exits or<br />

windows. Green dots on windows show those people who jumped or were rescued from<br />

windows. (d) People are stuck around an exit and smoke forces agents hide in a room. (e)<br />

Location <strong>of</strong> death around exits and in a room (white dots); people who are rescued by fire<br />

fighters are shown as green dots.<br />

Page | 7


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

5. Conclusion and Fur<strong>the</strong>r Work<br />

Fire investigation reports contain different areas <strong>of</strong> details including building construction, fire<br />

spread, and human behaviour. Fur<strong>the</strong>rmore, <strong>the</strong> reports provide information which allows us to<br />

understand <strong>the</strong> interaction <strong>of</strong> fire spread and occupant movement. The fire events are<br />

described pr<strong>of</strong>essionally not only by witnesses but also by fire-fighters. Therefore, this new<br />

methodology <strong>of</strong> model development by investigating fire reports can improve two main aspects that<br />

have not received attention in <strong>the</strong> past:<br />

1) More realistic representation <strong>of</strong> egress selection<br />

Most <strong>of</strong> <strong>the</strong> existing models only consider <strong>the</strong> design <strong>of</strong> exits and configuration, and<br />

agents were only permitted to escape through doors. In real life as described in <strong>the</strong> fire<br />

reports, people usually stand at windows and wait for fire-fighters to rescue <strong>the</strong>m. As<br />

a result, a window provides ano<strong>the</strong>r possible egress route and should be included in<br />

<strong>the</strong> evacuation simulation. In addition, a hazard area could be identified if <strong>the</strong> model<br />

predicts a group <strong>of</strong> people hide or die in a room. Taking this into account, <strong>the</strong> design<br />

and configuration <strong>of</strong> a building could <strong>the</strong>n be improved or more exits and windows be<br />

installed to increase <strong>the</strong> possibility <strong>of</strong> safe evacuation.<br />

2) Improved accuracy <strong>of</strong> total evacuation time<br />

Evacuation time is calculated by taking into account an individual‘s walking speed and<br />

movement. This model additionally simulates <strong>the</strong> transition <strong>of</strong> human behaviour from<br />

calm to panic. Therefore, agent walking speed and movement changes during <strong>the</strong><br />

simulation, bringing <strong>the</strong> total evacuation time closer to reality.<br />

This model is developed mostly based on <strong>the</strong> U.S. fire investigation reports, and fur<strong>the</strong>r work is to<br />

investigate whe<strong>the</strong>r people behave differently in different countries – cultural factors such as<br />

crowd behaviour in stressful situations, methodology used for collecting statistics or different<br />

building regulations may be important here. Preliminary tests determined that walking speed in<br />

terms <strong>of</strong> ages (child, adult, elderly) does not result in a significant difference in a crowd<br />

situation, and such differentiation <strong>of</strong> <strong>the</strong> pedestrian agents may not be required. Fur<strong>the</strong>r<br />

research is also required in relation to fire/smoke behaviour, and building/interior/furnishing<br />

materials. This should take into consideration, how fire and smoke spread through an<br />

environment, how <strong>the</strong> burning furniture influences <strong>the</strong> fire event, how people interact with smoke<br />

and fire, and <strong>the</strong> impact <strong>of</strong> <strong>the</strong> rate <strong>of</strong> smoke inhalation.<br />

6. Reference<br />

Cheng, X., Zhang, Heping, Xie, Q., Zhou, Y., Zhang, Hongjiang & Zhang, C. (2009) Study <strong>of</strong><br />

Announced Evacuation Drill from a Retail Store. Building and Environment, 44 (5), p.pp.864-870.<br />

Goldstein, M. (2008) Carbon Monoxide Poisoning. Journal <strong>of</strong> Emergency Nursing, 34 (6), p.pp.538-<br />

542.<br />

Kobes, M., Helsloot, I., de Vries, B., Post, J.G., Oberijé, N. & Groenewegen, K. (2010) Way<br />

Finding during Fire Evacuation; an Analysis <strong>of</strong> Unannounced Fire Drills in a Hotel at Night.<br />

Building and Environment, 45 (3), p.pp.537-548.<br />

Mehran, R., Oyama, A. & Shah, M. (2009) Abnormal Crowd Behavior Detection using Social<br />

Force Model. In: IEEE Computer Society <strong>Conference</strong> on Computer Vision and Pattern Recognition.<br />

Los Alamitos, CA, USA, IEEE Computer Society, pp.935-942.<br />

Page | 8


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Olsson, P.Å. & Regan, M.A. (2001) A Comparison between Actual and Predicted Evacuation<br />

Times.Safety Science, 38 (2), p.pp.139-145.<br />

Proulx, G. (2002) Understanding Human Behaviour in Stressful Situations. In: Workshop to Identify<br />

Innovative <strong>Research</strong> Needs to Foster Improved Fire Safety in <strong>the</strong> United States, National Academy<br />

<strong>of</strong> Sciences, Delegate Binder Section 7, Washington, DC. p.1–5.<br />

Sandberg, A. (<strong>19</strong>97) Unannounced Evacuation <strong>of</strong> Large Retail-Stores. An Evaluation <strong>of</strong><br />

HumanBehaviour and <strong>the</strong> Computermodel Simulex. Lund University.<br />

Shields, T.J. & Boyce, K.E. (2000) A Study <strong>of</strong> Evacuation from Large Retail Stores. Fire<br />

SafetyJournal, 35 (1), p.pp.25-49.<br />

Siebel, N. & Maybank, S. (2002) Fusion <strong>of</strong> Multiple Tracking Algorithms for Robust<br />

People Tracking. In: Computer Vision — ECCV 2002. pp.373-387.<br />

Zhao, C.M., Lo, S.M., Zhang, S.P. & Liu, M. (2009) A Post-fire Survey on <strong>the</strong> Pre-<br />

Evacuation Human Behavior. Fire Technology, 45 (1), p.pp.71-95.<br />

7. Biography<br />

Tyng-Rong Roan is a PhD student in <strong>the</strong> department <strong>of</strong> Civil, Environmental and<br />

Geomatic Engineering at University College London. Her research interests include spatial<br />

modelling and simulation.<br />

Muki Haklay is a Senior Lecturer in Geographic Information Science and <strong>the</strong> director <strong>of</strong> Chorley<br />

Institute in <strong>the</strong> department <strong>of</strong> Civil, Environmental and Geomatic Engineering at University College<br />

London. His research interests include Public Access and use <strong>of</strong> Environmental Information, Human-<br />

Computer Interaction (HCI) and Usability Engineering <strong>of</strong> <strong>GIS</strong> and Societal aspects <strong>of</strong> <strong>GIS</strong> use.<br />

Claire Ellul is a Lecturer in Geographic Information in <strong>the</strong> department <strong>of</strong> Civil, Environmental<br />

and Geomatic Engineering at University College London. Her research interests include<br />

spatial databases and approaches for handling large quantities <strong>of</strong> spatial data.<br />

Page | 9


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Reducing Exposure to Air Pollution: A Network Approach<br />

Gemma Davies, Duncan Whyatt<br />

Lancaster Environment Centre, Lancaster University, Lancaster, LA1 4YQ<br />

Tel. +44(0)1524 510252<br />

gemma.davies@lancaster.ac.uk<br />

ABSTRACT<br />

In response to concerns over <strong>the</strong> health implications <strong>of</strong> journey-time exposure to air pollution, this<br />

paper uses <strong>GIS</strong> to incorporate modelled pollution concentrations into a path network, enabling<br />

network analysis to be used in order to identify low exposure routes. Results from this initial case<br />

study show that exposure levels on low exposure routes are typically 20% lower than those based on<br />

shortest distance, suggesting that <strong>the</strong> design <strong>of</strong> less exposed walking routes could have benefits for<br />

human health.<br />

KEYWORDS: journey-time exposure, network analysis, dispersion modelling<br />

1. Introduction and Review <strong>of</strong> Literature<br />

Concerns over <strong>the</strong> health implications associated with exposure to airborne pollutants have led <strong>GIS</strong><br />

techniques to be increasingly used in <strong>the</strong> analysis <strong>of</strong> human exposure to air pollution. (Gulliver and<br />

Briggs 2004; Maynard et al. 2007). Individual exposure can be significantly increased by walking in<br />

heavily trafficked environments (Greaves et al, 2008) with Kaur et al (2007) concluding that<br />

proximity to pollution source has a significant impact upon exposure and that pedestrians should<br />

make greater use <strong>of</strong> quieter back streets. Significant variations in pollution concentrations can also be<br />

found between two sides <strong>of</strong> <strong>the</strong> same road (Kaur et al, 2005). This paper presents a methodology for<br />

generating lower exposure walking routes through such environments.<br />

Various approaches to generating lower exposure routes have previously been developed. Hertal et al<br />

(2008) used annual diurnal traffic flow as a surrogate for air pollution in <strong>the</strong>ir route finder approach.<br />

However, this failed to account for <strong>the</strong> increased exposure resulting from longer journey times.<br />

Meanwhile awareness raising initiatives such as <strong>the</strong> least polluted route option developed within <strong>the</strong><br />

WALK-IT (www.walkit.com) route planning service are starting to promote <strong>the</strong> choice <strong>of</strong> less<br />

polluted routes for pedestrians within selected cities in <strong>the</strong> <strong>UK</strong>. WALK-IT uses a modelled pollution<br />

surface based on <strong>the</strong> dispersion <strong>of</strong> NO2 for a ‗typical‘ day which is <strong>the</strong>n integrated with nodes in a<br />

path network to increase <strong>the</strong> impedance <strong>of</strong> polluted segments <strong>of</strong> <strong>the</strong> network making <strong>the</strong>m less<br />

preferable to <strong>the</strong> route finder. Davies and Whyatt (2009) used an alternative raster-based least-cost<br />

method to define low exposure routes. Their approach used a mask to define areas <strong>of</strong> traversable<br />

space, however, while allowing freedom <strong>of</strong> movement through open spaces <strong>the</strong> derivation <strong>of</strong> this<br />

mask was time consuming which has implications for larger scale studies.<br />

Drawing on some <strong>of</strong> <strong>the</strong> strengths from previous approaches <strong>the</strong> methodology outlined in this paper<br />

demonstrates <strong>the</strong> potential <strong>of</strong> network analysis for defining lower exposure routes for a number <strong>of</strong><br />

scenarios taking into account day-specific meteorological conditions and background pollution<br />

concentrations. The network developed in this approach differs from previous studies in two key<br />

ways. Previous approaches such as WALK-IT and Hertal et al (2008) have adopted a road centreline<br />

approach to <strong>the</strong> network, whilst this study adopts a pavement approach enabling <strong>the</strong> choice <strong>of</strong> which<br />

side <strong>of</strong> <strong>the</strong> road to walk along, thus reflecting <strong>the</strong> significant variations found by Kaur et al (2005).<br />

To date <strong>the</strong> analysis <strong>of</strong> least-polluted routes has concentrated on ‗typical‘ pollution conditions,<br />

however, <strong>the</strong> approach presented here expands this concept to consider <strong>the</strong> potential variation in<br />

exposure under a variety <strong>of</strong> metrological conditions and background pollution concentrations. The<br />

Page | 10


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

analysis will focus on <strong>the</strong> case study <strong>of</strong> a city centre high school, surrounded by a network <strong>of</strong> busy<br />

roads and will examine <strong>the</strong> journey time exposure for a hundred year 8 pupils living within 2km <strong>of</strong><br />

<strong>the</strong> school. The work builds on a wider project exploring <strong>the</strong> school journey (Pooley et al 2010).<br />

2. Methodology<br />

The method integrates high resolution pollution surfaces with a topologically structured path network,<br />

to create new cost evaluators representing exposure. This makes it possible to compute routes<br />

designed to minimise exposure to pollution in addition to those representing shortest distance. The<br />

method used is summarised in Figure 1.<br />

2.1 Data Inputs<br />

Figure 1: Methodology<br />

The first step <strong>of</strong> <strong>the</strong> methodology was to derive pollution surfaces which could later be used to<br />

calculate exposure through <strong>the</strong> network. In this case study <strong>the</strong> pollutant NO2 was chosen, as this is<br />

one <strong>of</strong> <strong>the</strong> key pollutants known to have chronic and acute health impacts (Xia and Tong 2005). The<br />

pollution surfaces were generated using <strong>the</strong> dispersion model ADMS Urban, which calculates<br />

concentrations based on emission estimates derived from traffic counts. O<strong>the</strong>r inputs included hourly<br />

background NO2 concentrations and meteorology. Model output was used to create pollution<br />

surfaces at 1m spatial resolution, sufficiently fine to distinguish variation across <strong>the</strong> width <strong>of</strong> a road.<br />

Most examples <strong>of</strong> network analysis use a road centreline approach in defining route choice, however,<br />

in order to test subtleties in route choice, such as which side <strong>of</strong> <strong>the</strong> road to walk along, a pavement<br />

network was developed. The initial network was based on Ordnance Survey MasterMap ITN data.<br />

Page | 11


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Road centre-lines where <strong>the</strong>n buffered at a distance representing <strong>the</strong> road widths and <strong>the</strong>n converted<br />

to lines in order to define pavements ei<strong>the</strong>r side <strong>of</strong> <strong>the</strong> road. Additional footpaths not adjacent to<br />

roads were <strong>the</strong>n added to <strong>the</strong> pavement network. Road crossings were automatically included where<br />

<strong>the</strong> ends <strong>of</strong> <strong>the</strong> buffers met, which included all road junctions. Some additional road crossing options<br />

were also added to <strong>the</strong> network, including <strong>the</strong> location <strong>of</strong> pedestrian crossings. A typical walking<br />

speed <strong>of</strong> 3mph was assumed for most <strong>of</strong> <strong>the</strong> network, however, at road crossing points on busy main<br />

roads this speed was lowered in order to simulate <strong>the</strong> time required to wait before crossing <strong>the</strong> road,<br />

<strong>the</strong>refore accounting for impact on exposure <strong>of</strong> waiting times within potentially highly polluted<br />

environments.<br />

2.2 Creating a network evaluator<br />

Network analysis with ESRI‘s Arc<strong>GIS</strong> depends on <strong>the</strong> definition <strong>of</strong> evaluators, ei<strong>the</strong>r cost evaluators<br />

such as distance or time, or restrictors such as oneway streets. Within this environment a new cost<br />

evaluator, <strong>the</strong>refore, needed to be created representing exposure to NO2 for each segment <strong>of</strong> <strong>the</strong><br />

network. This was achieved by using zonal statistics to calculate <strong>the</strong> mean concentration per<br />

segment, <strong>the</strong>n subsequently calculating <strong>the</strong> total cumulative exposure per segment taking into account<br />

<strong>the</strong> length <strong>of</strong> <strong>the</strong> segment and <strong>the</strong> travel speed (3mph) in addition to <strong>the</strong> mean concentration.<br />

2.3 Network Analysis<br />

Once <strong>the</strong> network and relevant evaluators were established <strong>the</strong> Origin-Destination (OD) cost matrix<br />

tool with Arc<strong>GIS</strong>‘ network analyst was used to calculate sets <strong>of</strong> both <strong>the</strong> shortest and least exposed<br />

routes between a set <strong>of</strong> origins, in this example <strong>the</strong> homes <strong>of</strong> all year 8 pupils living within 2km <strong>of</strong> a<br />

city centre secondary school and a destination, <strong>the</strong> school. Under each scenario <strong>the</strong> accumulated<br />

exposure to pollution (NO2) was calculated.<br />

3. Results and Discussion<br />

In order to assess <strong>the</strong> potential for exposure reduction when taking a route defined by least exposure<br />

to pollution ra<strong>the</strong>r than shortest distance, <strong>the</strong> methodology was applied using day-specific<br />

meteorology and background pollution levels. The sample days chosen were selected to represent a<br />

variety <strong>of</strong> differing meteorological and background conditions observed throughout a calendar year,<br />

2006 (Table 1).<br />

Table 1. Meteorology and background concentrations for selected days in 2006<br />

Figure 2 represents <strong>the</strong> range <strong>of</strong> potential exposure experienced by <strong>the</strong> 100 pupils for each <strong>of</strong> <strong>the</strong><br />

sample days. It shows <strong>the</strong> reduction in exposure which may be achieved by choosing a least exposed<br />

route, while also highlighting <strong>the</strong> considerable variation in exposure between days. The average<br />

reduction in journey-time exposure to NO2 using a least exposed route is 20%, however, on days such<br />

as <strong>the</strong> 20 th November this can be as high as 50%. For some individual route choices this variation<br />

may be greater, with exposure reduction <strong>of</strong> up to 86% recorded for one individual journey.<br />

Page | 12


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 2: Journey-time Exposure to NO2<br />

Developing a path network with associated evaluators representing pollution exposure has a number<br />

<strong>of</strong> advantages over a least cost surface approach. While care needs to be taken to ensure that viable<br />

pathways are not omitted from <strong>the</strong> network, this editing process is easier to achieve and more<br />

adaptable than <strong>the</strong> process <strong>of</strong> ensuring an accurate analysis mask is derived to represent traversable<br />

space (Davies and Whyatt 2009). Network analysis also provides <strong>the</strong> flexibility to run from a large<br />

number <strong>of</strong> origins and destinations simultaneously. The network design is however, not without its<br />

limitations. Open spaces can, for example, only be crossed where specified pathways are defined as<br />

features in <strong>the</strong> network. Likewise while <strong>the</strong> network used in this paper is effective in enabling<br />

pavement choice, <strong>the</strong> locations at which <strong>the</strong> road can be crossed are restricted to <strong>the</strong> locations <strong>of</strong><br />

junctions and a limited number <strong>of</strong> o<strong>the</strong>r specified crossing points such as pedestrian crossings. One<br />

additional advantage <strong>of</strong> <strong>the</strong> network is that it can allow for variable speed <strong>of</strong> travel through space, for<br />

example allowing for pause points along <strong>the</strong> journey while waiting to cross a road.<br />

Results from <strong>the</strong> network analysis presented here assume a willingness to adopt suggested leastexposed<br />

routes, however, fails to take into account o<strong>the</strong>r factors which may affect decisions regarding<br />

route choice. For example an earlier study focusing on exploring <strong>the</strong> school journey with a sample <strong>of</strong><br />

<strong>the</strong> pupils from <strong>the</strong> school used in this case study suggests route choice is influenced by factors such<br />

as parental control (Walker et al 2008). For instance, many parents will not allow <strong>the</strong>ir children to<br />

walk through <strong>the</strong> park opposite <strong>the</strong> school, yet for <strong>the</strong> majority <strong>of</strong> pupils a least polluted route would<br />

take <strong>the</strong>m through <strong>the</strong> park. The model could <strong>the</strong>refore be expanded to account for restrictors<br />

accounting for areas to be avoided due to concerns such as safety, some <strong>of</strong> which only take effect<br />

during certain time periods such as after dark.<br />

The challenge <strong>of</strong> realistically representing <strong>the</strong> real world extends to <strong>the</strong> complexities <strong>of</strong> attempting to<br />

accurately calculate actual journey-time exposure, which is affected by a number <strong>of</strong> factors besides<br />

route choice. These include background pollution concentrations, meteorological conditions,<br />

emissions from point sources, traffic concentrations, breathing height and level <strong>of</strong> physical activity<br />

(Cook et al 2008; Crabbe et al 2000; Gulliver and Briggs 2005). While any attempt to model <strong>the</strong> real<br />

world is met with limitations, <strong>the</strong> analysis presented here hopefully proves that <strong>the</strong> application <strong>of</strong> <strong>GIS</strong><br />

techniques is potentially very useful in addressing <strong>the</strong> challenge <strong>of</strong> quantifying and potentially<br />

reducing journey-time exposure. In exploring <strong>the</strong> use <strong>of</strong> day specific conditions and pavements in<br />

preference to road centrelines, this approach moves us one step closer to simulating some <strong>of</strong> <strong>the</strong><br />

complexities faced when modelling reality.<br />

Page | 13


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

4. Conclusion<br />

The application <strong>of</strong> network analysis provides an efficient way <strong>of</strong> deriving least polluted routes,<br />

although <strong>the</strong> success <strong>of</strong> this approach depends on <strong>the</strong> completeness <strong>of</strong> <strong>the</strong> path network and <strong>the</strong><br />

availability <strong>of</strong> a relevant pollution surface from which to derive <strong>the</strong> network evaluator. The greatest<br />

reductions in journey-time exposure are generated when day-specific meteorological and background<br />

conditions are applied, however, applying <strong>the</strong> current methodology in real time is not yet practical<br />

due to data access and processing requirements. We are, <strong>the</strong>refore, left to assess whe<strong>the</strong>r <strong>the</strong><br />

implementation <strong>of</strong> a set <strong>of</strong> ‗typical‘ conditions may provide a useful alternative for defining least<br />

polluted routes and whe<strong>the</strong>r this in turn may in <strong>the</strong> future provide added benefit to <strong>the</strong> kind <strong>of</strong> route<br />

planning systems already in place such as WALK-IT. Fur<strong>the</strong>r analysis is, <strong>the</strong>refore, needed to assess<br />

<strong>the</strong> extent to which a set <strong>of</strong> ‗representative days‘ may be able to replace more precise day specific<br />

conditions, while still generating routes able to reduce an individual‘s exposure to airborne pollution.<br />

In this case study we have only explored a series <strong>of</strong> short journeys, seeing relatively small reductions<br />

in exposure, however, <strong>the</strong> cumulative reduction in exposure over repeated journeys through <strong>the</strong><br />

course <strong>of</strong> a year may have significant health benefits. This is especially <strong>the</strong> case with regards to<br />

exposure to particulate matter (PM) which is deposited in deep areas <strong>of</strong> <strong>the</strong> lungs and is not easily<br />

resolved by <strong>the</strong> human body (Xia and Tong 2006). In order to fur<strong>the</strong>r this work to fully appreciate<br />

<strong>the</strong> impact such reductions in journey-time exposure may have for human health, fur<strong>the</strong>r research in<br />

this field will require <strong>the</strong> input <strong>of</strong> epidemiological expertise.<br />

5. References<br />

Cook R, Isakov V, Touma J S, Benjey W, Thurman J, Kinnee E, and Ensley D (2008). Resolving<br />

local-scale emissions for modelling air quality near roadways. Journal <strong>of</strong> <strong>the</strong> Air & Waste<br />

Management Association 58 (3), 451-461.<br />

Crabbe H, Hamilton R and Machin N (2000). Using <strong>GIS</strong> and dispersion modelling tools to assess <strong>the</strong><br />

effect <strong>of</strong> <strong>the</strong> environment on health. Transactions in <strong>GIS</strong> 4 (3),235-244.<br />

Davies G and Whyatt J D (2009) A least-cost approach to personal exposure reduction. Transactions<br />

in <strong>GIS</strong>. 13(2), 229-246<br />

Greaves S Issarayandgyun T and Lui Q (2008). Exploring variability in pedestrian exposure to fine<br />

particulates (PM2.5) along a busy road. Atmospheric Environment 42, 1665-1676.<br />

Gulliver J and Briggs D J (2005). Time-space modelling <strong>of</strong> journey-time exposure to traffic-related<br />

air pollution using <strong>GIS</strong>. Environmental <strong>Research</strong> 97, 10-25<br />

Gulliver J and Briggs D J (2004). Personal exposure to particulate air pollution in transport<br />

microenvironments. Atmospheric Environment 38, 1-8.<br />

Hertel O, Hvidberg M, Matthias Ketzel M, Storm L and Stausgaard L (2008). A proper choice <strong>of</strong><br />

route significantly reduces air pollution exposure — A study on bicycle and bus trips in urban streets.<br />

Science <strong>of</strong> <strong>the</strong> Total Environment 389(1), 58-70<br />

Kaur S, Nieuwenhuijsen M J and R N Colvile (2007). Fine particulate matter and carbon monoxide<br />

exposure concentrations in urban street transport microenvironments. Atmospheric Environment 41,<br />

4781-4810.<br />

Kaur S, Nieuwenhuijsen M J and Colvile R N (2005). Pedestrian exposure to air pollution along a<br />

major road in Central London. Atmospheric Environment 39, 7307-7320.<br />

Maynard D, Coull B A, Gryparis A, and Schwartz J (2007). Mortality risk associated with short-Term<br />

exposure to traffic particles and sulfates. Environmental Health Perspectives 115(5), 751-755.<br />

Pooley C, Whyatt J D, Walker M, Davies G, Coulton P and Bamford W (2010) Understanding <strong>the</strong><br />

school journey, integrating data on travel and environment. Environment and Planning A 42, 948-<br />

965<br />

Walker M, Whyatt J D, Pooley C, Davies G, Coulton P, and Bamford W (2008). Talk, technologies<br />

and teenagers, understanding <strong>the</strong> school journey using a mixed-methods approach. Children‘s<br />

Page | 14


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Geographies, 7(2) 107-122.<br />

Xia Y and Tong H (2006) Cumulative effects <strong>of</strong> air pollution on public health. Statistics in Medicine.<br />

25, 3548-3559<br />

6. Biography<br />

Gemma Davies is <strong>the</strong> <strong>GIS</strong> Officer for Lancaster Environment Centre.<br />

Duncan Whyatt is a senior lecturer in <strong>GIS</strong> with interests in air pollution modelling at regional and<br />

local scales.<br />

Page | 15


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Effects <strong>of</strong> Climate Change on Solent Coastal Management, Businesses<br />

and Communities<br />

Sarah Percival 1 , Richard Teeuw 2<br />

1 Centre for Applied Geosciences, School <strong>of</strong> Earth and Environment Sciences,<br />

University <strong>of</strong> Portsmouth, Burnaby Building, Burnaby Road, Portsmouth, PO1 3QL<br />

Tel. +44 2392842418<br />

1 Sarah.Percival@port.ac.uk<br />

2 Richard.Teeuw@port.ac.uk<br />

Tel. +44 2392842267<br />

ABSTRACT<br />

Climate change is expected to increase <strong>the</strong> frequency and magnitude <strong>of</strong> flood events. Society‘s<br />

vulnerability to flooding has increased, due to urbanisation, extension <strong>of</strong> infrastructure and o<strong>the</strong>r landuse<br />

changes in flood-prone areas. This project will concentrate on coastal flood risks, and aims to<br />

produce an integrated <strong>GIS</strong> database that will facilitate <strong>the</strong> modeling <strong>of</strong> vulnerability maps, which will<br />

identify at-risk assets and vulnerable sectors <strong>of</strong> society. The Solent region will be used to test this<br />

methodology, particularly <strong>the</strong> island city <strong>of</strong> Portsmouth and <strong>the</strong> Havant coastal district.<br />

1. Introduction<br />

KEYWORDS: Vulnerability, Coastal, Flooding, Mapping, Solent<br />

Coastal zones have social, economic and environmental importance: <strong>the</strong>y attract settlements and<br />

economic activity, and include natural wildlife habitats that provide valuable services and functions.<br />

Coastal activities are vulnerable to climate changes. Sea-level rise and more intense storms could<br />

raise flood risk, increase coastal erosion and adversely affect ecosystem structure and functioning,<br />

especially on low-lying coasts (Nicholls et al, 2008). As sea level rises and climate conditions become<br />

increasingly variable, global climate projections indicate that pressures on <strong>the</strong> coasts will increase.<br />

The economic impacts <strong>of</strong> flood events have already increased due to growing vulnerability arising<br />

from societal changes, such as interference by land-use changes (urbanisation and infrastructural<br />

extensions) in flood-prone areas (Douben, 2006). Munich Re (<strong>19</strong>97) state that <strong>the</strong> damage caused by<br />

floods in recent decades has been extremely severe, and it is evident that both flood frequency and<br />

intensity are increasing. Of all natural hazards, floods are <strong>the</strong> globally <strong>the</strong> most frequent, cause <strong>the</strong><br />

largest amount <strong>of</strong> fatalities, and generate <strong>the</strong> largest economic losses. Climate change and progression<br />

within society will increase vulnerability for <strong>the</strong> estimated £130 billion <strong>of</strong> assets that would be at risk<br />

from coastal flooding (DEFRA, 2009).<br />

This study assesses <strong>the</strong> effects <strong>of</strong> climate change on coastal management, businesses and<br />

communities in <strong>the</strong> Solent (Figure 1), paying particular interest to <strong>the</strong> island city <strong>of</strong> Portsmouth and<br />

<strong>the</strong> coastal zone <strong>of</strong> Havant Borough (highlighted in Figure 1). The main research questions are:<br />

� How can current coastal flood vulnerability maps be improved when identifying at-risk<br />

infrastructure and sectors <strong>of</strong> society?<br />

� How can at-risk social, environmental and economic assets be classified and analysed?<br />

� Can frameworks for vulnerability analysis for flood risk, be improved?<br />

Page | 16


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 1. Map <strong>of</strong> <strong>the</strong> Solent based on <strong>the</strong> geographical area used by <strong>the</strong> Solent Forum. The<br />

map includes coastal landforms, urban distribution, important points <strong>of</strong> infrastructure (e.g.<br />

roads, railways, hospitals etc). Incorporating data from SCOPAC, <strong>the</strong> Solent Forum and<br />

Google Maps. The rectangular box indicates <strong>the</strong> Havant Borough and city <strong>of</strong> Portsmouth case<br />

study areas<br />

1.1 Climate Change and <strong>the</strong> Coast<br />

There is growing evidence that so-called greenhouse gas emissions are changing <strong>the</strong> atmosphere<br />

significantly (Hosking & McInnes, 2002). In 2007 <strong>the</strong> International Panel on Climate Change (IPCC)<br />

concluded that most <strong>of</strong> <strong>the</strong> observed increase in Global average temperatures since <strong>the</strong> mid-20 th<br />

century is due to <strong>the</strong> observed increase in anthropogenic greenhouse gas concentrations. In 2001,<br />

Standing <strong>Conference</strong> on Problems Affecting <strong>the</strong> Coastline (SCOPAC) stated that '<strong>the</strong> question <strong>of</strong><br />

climate change impacts is probably <strong>the</strong> most important issue to be faced by coastal local authorities<br />

and <strong>the</strong> communities <strong>the</strong>y represent, alongside o<strong>the</strong>r organisations in <strong>the</strong> coastal zone'.<br />

From current scientific understanding <strong>of</strong> climate change, coastal areas <strong>of</strong> <strong>the</strong> <strong>UK</strong> will experience a<br />

number <strong>of</strong> key changes in <strong>the</strong> next 100 years: this includes higher tides, increased coastal erosion,<br />

higher river flows in winter, and an increase in storm surges (Few et al. 2007).<br />

1.2 Coastal flood risk and vulnerability<br />

Coastal risks arise when society interacts with hazards associated with <strong>the</strong> physical environment, such<br />

as flooding and erosion or instability (McInnes, 2006). Risk <strong>of</strong> disaster occurs at <strong>the</strong> interaction <strong>of</strong> <strong>the</strong><br />

human environment with <strong>the</strong> physical environment (Figure 2). Climate change will increase <strong>the</strong><br />

magnitude and frequency <strong>of</strong> coastal hazards, and (according to McInnes, 2006) as urban development<br />

intensifies or spreads into hazardous areas, <strong>the</strong> potential impact <strong>of</strong> hazards increases, <strong>the</strong>reby<br />

increasing risk.<br />

Page | 17


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 2. Coastal Hazards, human activity and risk (McInnes, 2006)<br />

Flood risk is defined in terms <strong>of</strong> hazard, vulnerability and exposure. Hazard describes <strong>the</strong> frequency<br />

and severity <strong>of</strong> <strong>the</strong> flood event; vulnerability is <strong>the</strong> extent to which <strong>the</strong> development (property and<br />

people) could be affected by <strong>the</strong> hazard; and exposure is defined in terms <strong>of</strong> economic and social<br />

impact (Haynes et al, 2007).<br />

Generally speaking, vulnerability to environmental hazards means <strong>the</strong> potential for loss. Vulnerability<br />

analysis involves <strong>the</strong> identification <strong>of</strong> conditions that make people or places vulnerable to extreme<br />

natural events (Cutter et al, 2003). The concept <strong>of</strong> vulnerability is at <strong>the</strong> heart <strong>of</strong> our understanding <strong>of</strong><br />

how communities and natural systems, institutional structures and social relationships are affected by<br />

climate variability and disaster (Vincent, 2004). All societies are vulnerable to floods, but under<br />

different situations: each case is unique. A vulnerability analysis, including socio-economic data and<br />

asset data, would provide levels <strong>of</strong> detail that will help future planning for coastal management and<br />

civil emergency management.<br />

1.3 The Solent<br />

The Solent (Figure 1) is <strong>the</strong> body <strong>of</strong> water that lies between <strong>the</strong> central south coast <strong>of</strong> England and <strong>the</strong><br />

Isle <strong>of</strong> Wight. It is a low energy, sediment dominated estuarine complex, consisting <strong>of</strong> 12 separately<br />

defined estuaries and harbours, draining a catchment <strong>of</strong> approximately 3000km 2 (Fletcher et al,<br />

2007). It is a densely populated area that has a long history <strong>of</strong> coastal development, commercial<br />

activities and marine and coastal recreation. The coastal zone has a high nature conservation value,<br />

and has many dedicated commercial and military ports. Fletcher et al (2007) stated <strong>the</strong> Solent is an<br />

area that is shaped by <strong>the</strong> sea, both in terms <strong>of</strong> <strong>the</strong> physical environment and <strong>the</strong> prevailing economic<br />

and social conditions. This has given <strong>the</strong> area a long history <strong>of</strong> reconciling conflicting coastal<br />

activities.<br />

1.4 Coastal Management in <strong>the</strong> Solent<br />

General coastal management encompasses all <strong>the</strong> activities and multiple users that occur in and<br />

around <strong>the</strong> coast, and <strong>the</strong>refore requires an integrated multi-disciplinary approach to provide a holistic<br />

approach to coastal protection and flood defence. There are three main tiers <strong>of</strong> management for <strong>the</strong><br />

coast: (i) Shoreline Management Plans (SMP); (ii) Coastal Defence Strategies (CDS); (iii) coast<br />

protection schemes.<br />

The multiple uses <strong>of</strong> <strong>the</strong> Solent area means that it experiences many impacts and political pressures.<br />

Decisions are <strong>the</strong>refore needed in <strong>the</strong> context <strong>of</strong>, and in response to, <strong>the</strong> latest government policy and<br />

advice; new and revised legislation; environmental change; and changing socio-economic demands<br />

(Fletcher et al, 2007).<br />

Page | 18


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

The Government sets outcome measures for <strong>the</strong> Environment Agency and o<strong>the</strong>r operating authorities<br />

that work with <strong>the</strong>m to manage flood risk. Planning Policy Statement 25 (PPS25): Development and<br />

Flood Risk (2006) sets out <strong>the</strong> Government‘s approach to <strong>the</strong> use <strong>of</strong> <strong>the</strong> planning system to reduce<br />

flood risk. Under this guidance, Regional Strategies and Local Development Frameworks must<br />

include and account for flood risk (Environment Agency, 2009). Local authorities, with <strong>the</strong><br />

Environment Agency, must carry out a Strategic Flood Risk Assessment, which forms part <strong>of</strong> <strong>the</strong><br />

evidence to Local Development Frameworks.<br />

Shoreline Management Plans (SMPs) - one <strong>of</strong> <strong>the</strong> main tiers <strong>of</strong> coastal management - are produced by<br />

coastal groups (typically maritime operating authorities, council members, Defra, Environment<br />

Agency, Natural England, port authorities). It is a non-statutory document that provides a broad<br />

assessment <strong>of</strong> <strong>the</strong> long-term risks associated with coastal processes (Southampton City Council,<br />

2010). They are also high-level planning tools, and present a long term policy framework to reduce<br />

<strong>the</strong>se risks to people and <strong>the</strong> developed, historic and natural environment in a sustainable manner<br />

(DEFRA, 2009).<br />

An SMP is a high level document that forms an important element <strong>of</strong> <strong>the</strong> strategy for flood and<br />

coastal erosion risk management. Many operating authorities go on to adopt <strong>the</strong> recommendations <strong>of</strong><br />

<strong>the</strong>ir SMP as a basis for production <strong>of</strong> individual strategic plans, monitoring programmes and studies<br />

for all or part <strong>of</strong> <strong>the</strong>ir coastline and, where proven by strategic plans, for investment in appropriate<br />

capital improvement projects (DEFRA, 2009).<br />

1.5 Case Study areas – Portsmouth and Havant Borough<br />

The city <strong>of</strong> Portsmouth (Figure 1) is a low-lying maritime island city and one <strong>of</strong> <strong>the</strong> most densely<br />

populated areas in Britain (Portsmouth City Council, 2009). It also has major industrialisation,<br />

tourism, historical importance, commercial recreation and is now <strong>the</strong> primary naval port in Britain.<br />

Portsmouth City Council (2009) have also stated it is one <strong>of</strong> <strong>the</strong> most vulnerable cities to coastal<br />

erosion in Britain, with some areas being at risk from tidal flooding and wave overtopping.<br />

The Borough <strong>of</strong> Havant (Figure 1) has a coastline <strong>of</strong> 48 kilometres, as well as an extensive network <strong>of</strong><br />

rivers, streams, and ditches. Havant Borough Council maintain <strong>the</strong> coastline with <strong>the</strong> assistance <strong>of</strong><br />

land owners whose land adjoins <strong>the</strong> shoreline (Havant Borough Council, 2008). Hayling Island has an<br />

area around 30km 2 , and is a low lying predominantly open coastal plain island, separated from <strong>the</strong><br />

mainland by shallow tidal harbours. The maximum height <strong>of</strong> <strong>the</strong> land is on average about 5m AOD<br />

(Above Ordnance Datum) (Hampshire County Council, 2010). It has a prominent sandy beach on its<br />

south coast, which in recent years has been topped with dredged shingle in an effort to reduce <strong>the</strong><br />

potential <strong>of</strong> flooding to <strong>the</strong> low lying hinterland. It is largely a residential area with water sport leisure<br />

activities (i.e. sailing and windsurfing).<br />

Both Portsmouth and Havant Borough fall under <strong>the</strong> North Solent SMP. They also are part <strong>of</strong> <strong>the</strong><br />

Partnership for Urban South Hampshire (PUSH) (See Figure 3). This was a project that was<br />

commissioned to be a joint Strategic Flood Risk Assessment (SFRA) for <strong>the</strong> sub-region in accordance<br />

to PPS25. The SFRA is a critical part <strong>of</strong> <strong>the</strong> evidence base for <strong>the</strong> sub-region and enables <strong>the</strong> Local<br />

Planning Authorities within <strong>the</strong> Push sub region to make informed decisions. Within <strong>the</strong> PUSH<br />

project an online multi-layered awareness-raising tool was created: Flood Map.<br />

Page | <strong>19</strong>


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 3. PUSH Sub-region (PUSH, 2008)<br />

The PUSH Flood Map was used within <strong>the</strong> new North Solent SMP as evidence to inform <strong>the</strong> baseline<br />

scenario for coastal defence/management schemes (example see Figure 4). It is also important that <strong>the</strong><br />

number and location <strong>of</strong> commercial and residential properties at risk from erosion and/or tidal<br />

flooding in wards was presented integrated within <strong>the</strong> maps (see Table 1).<br />

Figure 4. Havant Borough Council: Flood Zone 3 (1:200yr) 2115 (North Solent SMP, 2010)<br />

Page | 20


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Table 1. Total number and type <strong>of</strong> properties per Havant Borough Council Ward, potentially<br />

within tidal floodplain, assuming No Defences, for 2007 and 2115.<br />

2. Preliminary findings<br />

The first stage <strong>of</strong> this comparative research study is complete. Using different maps from <strong>the</strong><br />

Ordnance Survey, historical flood data from <strong>the</strong> British Geological Survey, and expected flooding<br />

data up to 2115 from <strong>the</strong> PUSH Strategic Risk Flood Assessment. The ‗assets‘ that were affected by<br />

past flooding are shown in Figure 5, as are <strong>the</strong> areas <strong>of</strong> potential future flooding (Figure 6).<br />

Page | 21


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 5. Map <strong>of</strong> Portsmouth and <strong>the</strong> Borough <strong>of</strong> Havant in greater detail, including data from Ordnance Survey and historical flooding data<br />

(source: British Geological Survey)<br />

Page | 22


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 6. Maps <strong>of</strong> Portsmouth and <strong>the</strong> Borough <strong>of</strong> Havant in more detail, including expected flooding for <strong>the</strong> year 2115 (based upon <strong>the</strong> PUSH<br />

Strategic Flood Risk Assessment - The data shown in <strong>the</strong>se figures is predicted flooding by climate change in 2115. This assessment was based<br />

on <strong>the</strong> Environment Agency's sea level rise data (from <strong>19</strong>90), Lidar data, and Flood maps. This was combined with <strong>UK</strong> Climate Impacts<br />

Programme estimate <strong>of</strong> sea-level rise from <strong>19</strong>90-2115<br />

Page | 23


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

After <strong>the</strong> initial research, it was noted that coastal flood risk maps for <strong>the</strong> Solent lacked detail, with<br />

regard to socio-economic and asset data. This study is using Ordnance Survey data sets (shown in<br />

Figure 7) to develop a more detailed methodology for assessing coastal flood risk. An integrated<br />

Arc<strong>GIS</strong> database will be used to produce impact vulnerability maps, which will identify at-risk<br />

infrastructure and vulnerable sectors <strong>of</strong> society within <strong>the</strong> test areas. Preliminary results can be seen<br />

in Figures 7 and 8.<br />

Figure 7. All <strong>of</strong> Ordnance Survey Mastermap and Address Layer data presented in Arc<strong>GIS</strong><br />

Figure 8. O.S. file showing part <strong>of</strong> Hayling Island overlaid with Mastermap Road Link and<br />

Address Layer shapefiles. An example <strong>of</strong> new visualisation for flood vulnerability mapping.<br />

Page | 24


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

3. Conclusion<br />

This project will result in producing an integrated <strong>GIS</strong> database that will that will facilitate <strong>the</strong><br />

modelling <strong>of</strong> predictive vulnerability maps, identifying at-risk infrastructure and vulnerable sectors<br />

<strong>of</strong> society within <strong>the</strong> case study areas. These digital maps will improve visualisation when assessing<br />

risk with regard to coastal management and will be <strong>of</strong> use to local authorities, contingency planners,<br />

emergency services and public education.<br />

4. Acknowledgements<br />

Thanks to <strong>the</strong> Ordnance Survey for <strong>the</strong>ir support and data, as well as to <strong>the</strong> University <strong>of</strong> Portsmouth,<br />

for logistical support.<br />

5. References<br />

Cutter S, Boruff B and Shirley W (2003). Social vulnerability to Environmental Hazards. Social<br />

Science Quarterly, 84 (2), 242-261<br />

DEFRA (2009). ‗Shoreline Management Plans‘. [Online] Available at:<br />

http://www.defra.gov.uk/environment/flooding/policy/guidance/smp.htm<br />

Douben K-J (2006). Characteristics <strong>of</strong> river floods and flooding: A global overview, <strong>19</strong>85-2003.<br />

Irrigation and Drainage, 55, S9-S21<br />

Environment Agency (2009). ‗Flooding in England: A national assessment <strong>of</strong> flood risk‘. [Online]<br />

Available at: http://publications.environment-agency.gov.uk/pdf/GEHO0609BQDS-E-E.pdf<br />

Few R, Brown K and Tompkins EL (2007). Climate Change and Coastal Management Decisions:<br />

Insights from Christchurch Bay, <strong>UK</strong>. Coastal Management, 35 (2), 255-270<br />

Fletcher S, Johnson D and Hewett T (2007). Coastal Management in <strong>the</strong> Solent: An Introduction.<br />

Marine Policy, 31 (5), 585-590<br />

Hampshire County Council (2010). ‗Hayling Island Coastal Plain: Status Draft March 2010‘. [Online]<br />

Available at: http://www3.hants.gov.uk/9h_hayling_island_coastal_plain.pdf<br />

Havant Borough Council (2008). ‗Coastal defence, land drainage and flooding‘. [Online] Available<br />

at: http://havant.gov.uk/havant-<strong>19</strong>8<br />

Haynes H, Haynes R and Pender G (2007). Integrating socio-economic analysis into decision-support<br />

methodology for flood risk management at <strong>the</strong> development scale (Scotland). Water and Environment<br />

Journal, 22 (2), 117-124<br />

Hosking A and McInnes R (2002). Preparing for <strong>the</strong> Impacts <strong>of</strong> Climate Change on <strong>the</strong> Central South<br />

Coast <strong>of</strong> England: A Framework for Future Risk Management. Journal <strong>of</strong> Coastal <strong>Research</strong>, SI36,<br />

381-389<br />

IPCC (2007). Climate Change 2007: Impacts, adaptation and vulnerability: contribution <strong>of</strong> Working<br />

Group II to <strong>the</strong> fourth assessment report <strong>of</strong> <strong>the</strong> Intergovernmental Panel on Climate Change,<br />

Cambridge University Press<br />

McInnes R (2006). Responding to <strong>the</strong> Risks from Climate Change in Coastal Zones - A good<br />

practice. Centre for <strong>the</strong> Coastal Environment. Isle <strong>of</strong> Wight Council, 1-12<br />

Munich Re (<strong>19</strong>97). ‗Flooding: Munich Re discusses loss potential and insurance solutions‘. [Online]<br />

Available at:<br />

http://munichre.com/en/media_relations/press_release/<strong>19</strong>97/<strong>19</strong>97_07_28_press_release.aspx<br />

Nicholls R, Hanson S, Mokrech M, Stansby P, Chini N, Walkden M, Dawson R, Roche N, Hall J,<br />

Nicholson-Cole S, Watkinson A, Jude S, Lowe J, Wolf J, Leake J, Rounsevell M, Fontaine C and<br />

Acosta-Michlik L (2008). The Tyndall Coastal Simulator and Interface. <strong>Proceedings</strong> <strong>of</strong> Coastal<br />

Dynamics, 1-14<br />

North Solent Shoreline Management Plan (2010). ‗North Solent Shoreline Management Plan‘.<br />

[Online] Available at:<br />

Page | 25


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

http://www.northsolentsmp.co.uk/index.cfm?articleid=9907&articleaction=nthslnt&CFID=38693563<br />

&CFTOKEN=54654086<br />

Portsmouth City Council (2009). ‗Coastal Protection‘. [Online] Available at:<br />

http://www.portsmouth.gov.uk/AtoZ/3726.html<br />

PUSH (Partnership for Urban South Hampshire) 'Strategic Flood Risk Assessment'. [Online]<br />

Available at: http:// push.atkinsgeospatial.com<br />

SCOPAC (Standing <strong>Conference</strong> on Problems associated with <strong>the</strong> Coastline) (2001). Preparing for <strong>the</strong><br />

Impacts <strong>of</strong> Climate Change: Summary Report. Halcrow<br />

Southampton City Council (2010) ‗Coastal Issues‘. [Online] Available at:<br />

http://www.southampton.gov.uk/s-environment/climatechange/coastal-issues.aspx<br />

Vincent K (2004). Creating an Index <strong>of</strong> Social Vulnerability to climate change for Africa. Tyndall<br />

Centre for Climate Change <strong>Research</strong>, Working paper 56<br />

6. Biography<br />

Sarah Percival is a PhD student in <strong>the</strong> School <strong>of</strong> Earth and Environmental Sciences at Portsmouth<br />

University, using <strong>GIS</strong> to assess climate change impacts within coastal areas <strong>of</strong> <strong>the</strong> <strong>UK</strong>. Current<br />

research is focused on vulnerability mapping.<br />

Richard Teeuw is a Principal Lecturer in <strong>the</strong> School <strong>of</strong> Earth & Environmental Sciences at<br />

Portsmouth University, using remote sensing and <strong>GIS</strong> to assess geohazards and Earth resources.<br />

Current research is focused on uses <strong>of</strong> low-cost geoinformatics for disaster risk reduction and<br />

vulnerability mapping.<br />

Page | 26


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Assessing climate change vulnerability in Small Island Developing Nations:<br />

<strong>the</strong> case <strong>of</strong> Puerto Rico<br />

Félix I. Aponte‐González<br />

PhD Candidate, School <strong>of</strong> Environment and Development, University <strong>of</strong> Manchester, Oxford<br />

Road, M13 9PL<br />

Tel. ((+44)75547 47801)<br />

Felix.Aponte-gonzalez@postgrad.manchester.ac.uk<br />

ABSTRACT<br />

This research will be aiming to promote <strong>the</strong> understanding <strong>of</strong> climate change vulnerability for small<br />

islands in <strong>the</strong> Caribbean region. Using Puerto Rico as case study, <strong>the</strong> main objectives <strong>of</strong> this research<br />

are: to develop a climate change vulnerability assessment for <strong>the</strong> Island <strong>of</strong> Puerto Rico and; to use <strong>the</strong><br />

results <strong>of</strong> <strong>the</strong> vulnerability assessments to identify areas <strong>of</strong> special attention for dealing with climate<br />

change impacts. It is expected that <strong>the</strong> research will serve as a technical basis for more in-depth<br />

analysis <strong>of</strong> <strong>the</strong> effects <strong>of</strong> climate change to <strong>the</strong> future <strong>of</strong> <strong>the</strong> country and <strong>the</strong> region.<br />

KEYWORDS: Climate Change, Vulnerability, Caribbean, Small islands, Puerto Rico<br />

1. Introduction<br />

Dealing with climate change has become a matter <strong>of</strong> great concern for many countries nowadays,<br />

particularly for Small Island Developing Nations (SIDS). While not every part <strong>of</strong> <strong>the</strong> world would be<br />

similarly affected, <strong>the</strong> expected consequences <strong>of</strong> <strong>the</strong> increased global changes can be severe for<br />

developing countries and SIDS (Allison, et. al, 2009). The adverse effects that most <strong>of</strong> <strong>the</strong>se small<br />

countries are ei<strong>the</strong>r experimenting or beginning to face have been recognized as a matter <strong>of</strong> concern<br />

for <strong>the</strong> international community (UNFCC, 2010). In <strong>the</strong> most recent United Nations climate change<br />

conferences, several agreements have been reaffirming <strong>the</strong> need <strong>of</strong> all countries to engage in<br />

assessing vulnerability, impacts and adaptation to climate change on a national and regional scale<br />

(UNFCC2, 2010).<br />

The Caribbean has been identified as a region that is prone to be adversely impacted by climate<br />

change (UNEP, 2005). Some <strong>of</strong> <strong>the</strong> most pressing challenges that this region has to face are:<br />

increasing pressures on coastal and marine environments; limited land and freshwater resources; a<br />

strong dependency <strong>of</strong> <strong>the</strong> productive sector on <strong>the</strong>ir limited natural resources; high reliance on<br />

imports, particularly for food, fuel and o<strong>the</strong>r strategic imports and; shifting rainfall patterns and<br />

hurricanes (UNEP, 2008). These issues exacerbate <strong>the</strong> vulnerability <strong>of</strong> <strong>the</strong> Caribbean islands to<br />

extreme climate events and o<strong>the</strong>r natural disasters (UNEP, 2008).<br />

The island-nation <strong>of</strong> Puerto Rico, one <strong>of</strong> <strong>the</strong> 26 territories <strong>of</strong> <strong>the</strong> insular Caribbean (see Figure 1), is<br />

one <strong>of</strong> <strong>the</strong> dozens <strong>of</strong> SIDS that will be affected by climate change. The political, geographical, and<br />

socio-economical characteristic <strong>of</strong> this small island elevates its propensity towards hazards,<br />

particularly in coastal areas (Bush, <strong>19</strong>95). Yet, <strong>the</strong>re is a lack <strong>of</strong> studies that looks into <strong>the</strong><br />

vulnerability <strong>of</strong> this territory to some <strong>of</strong> <strong>the</strong> related impacts <strong>of</strong> climate change.<br />

Page | 27


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 1: Location <strong>of</strong> Puerto Rico (dark grey) in <strong>the</strong> Caribbean region<br />

This research intends to promote <strong>the</strong> understanding <strong>of</strong> climate change vulnerability for small islands<br />

in <strong>the</strong> Caribbean region. Using Puerto Rico as a case study, <strong>the</strong> main objectives <strong>of</strong> this research are:<br />

1) to develop a climate change vulnerability assessment for <strong>the</strong> Island <strong>of</strong> Puerto Rico and; 2) to use<br />

<strong>the</strong> results <strong>of</strong> <strong>the</strong> vulnerability assessment to spatially identify areas <strong>of</strong> special attention for dealing<br />

with climate change impacts. It is expected that <strong>the</strong> results <strong>of</strong> this paper will serve as a geographical<br />

basis for more in-depth analysis <strong>of</strong> climate change issues in <strong>the</strong> country and on <strong>the</strong> region.<br />

2. Vulnerability in <strong>the</strong> climate change context<br />

Vulnerability as a concept has been widely defined and categorized, depending upon its functionality<br />

and <strong>the</strong> extent <strong>of</strong> its analysis. In <strong>the</strong>ir latest assessment, <strong>the</strong> IPCC defines vulnerability as ―…<strong>the</strong><br />

degree to which a system is susceptible to, and unable to cope with, adverse effects <strong>of</strong> climate change,<br />

including climate variability and extremes. Vulnerability is a function <strong>of</strong> <strong>the</strong> character, magnitude,<br />

and rate <strong>of</strong> climate change and variation to which a system is exposed, its sensitivity and its adaptive<br />

capacity.‖ (Parry, et. al, 2007: p.883).<br />

From this perspective, vulnerability to climate change is seen as a function <strong>of</strong> exposure, sensitivity<br />

and adaptive capacity. This interpretation exemplifies <strong>the</strong> evolution <strong>of</strong> vulnerability assessments for<br />

dealing with climate change impacts. The exposure factor is related to <strong>the</strong> nature and degree in which<br />

a system is exposed to a series <strong>of</strong> changes, particularly on <strong>the</strong> valued constituents (Metzger, et. al,<br />

2006). The focus is centred mainly upon <strong>the</strong> interactions between hazards and <strong>the</strong> features <strong>of</strong> interest<br />

on a particular area <strong>of</strong> study.<br />

Sensitivity can be understood as ―<strong>the</strong> degree to which a system is affected, ei<strong>the</strong>r adversely or<br />

beneficially, by climate variability or change.‖ (Parry, et. al, 2007). While this definition is centred<br />

upon climate change vulnerabilities, it implies that sensitivity is seen as a measurement <strong>of</strong> how<br />

certain hazards can produce a series <strong>of</strong> effects on a particular system being studied. The adaptive<br />

capacity is related to <strong>the</strong> potential <strong>of</strong> a system to develop and integrate certain measures in order to<br />

effectively deal with <strong>the</strong> expected impacts (Parry, et. al, 2007). The output <strong>of</strong> this factor will vary<br />

depending on <strong>the</strong> characteristics <strong>of</strong> <strong>the</strong> hazards and <strong>the</strong> effects on <strong>the</strong> characteristics <strong>of</strong> <strong>the</strong> study area,<br />

as well as <strong>the</strong> temporal reference in which <strong>the</strong> assessment is established.<br />

Page | 28


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

2.1 Analyzing vulnerabilities through integrated assessments<br />

Integrated assessments are continuously being recognized as a useful approach for analyzing climate<br />

change vulnerabilities (Schröter, et. al, 2005; Füssel and Klein, 2006), and it has been used as a<br />

framework by <strong>the</strong> IPCC on <strong>the</strong>ir third and fourth Assessment Reports. Engaging in vulnerability<br />

assessment must be focused on a specific purpose; whe<strong>the</strong>r it is directed towards improving<br />

adaptation techniques, establish mitigation problems, address social injustice issues, to conduct<br />

scientific research or a combination <strong>of</strong> <strong>the</strong> aforementioned (Patt et. al, 2010). This type <strong>of</strong><br />

assessments needs to be able to produce results that can be used for generating sounding policies<br />

related to climate change issues. Analysis directed towards adaptation measures are becoming more<br />

common on integrated assessments, gaining importance due to its pertinence and necessity for<br />

countries worldwide (Moss, Pahl-Wostl and Downing, 2001; Brooks, Adger and Kelly, 2005;<br />

Mehrotra, et. al, 2009; Dawson, et. al, 2009).<br />

3. Proposed methodology:<br />

The proposed methodology will provide <strong>the</strong> means to gain a better understanding <strong>of</strong> how <strong>the</strong><br />

complexities <strong>of</strong> climate change impacts are expected to manifest in Caribbean cities. It is important to<br />

point out that this study will be generating a first look into <strong>the</strong> vulnerabilities <strong>of</strong> Puerto Rico towards<br />

climate impacts. It is not <strong>the</strong> intention to develop a comprehensive study <strong>of</strong> climate change<br />

vulnerabilities and adaptation. Still, <strong>the</strong> results from this research should be a contributing factor<br />

towards a better understanding <strong>of</strong> <strong>the</strong> vulnerabilities in <strong>the</strong> island-nation and for developing sound<br />

adaptation measures.<br />

The proposed methodology has been developed taking into consideration <strong>the</strong> guidelines and<br />

techniques used for case-study research, vulnerability assessments and index creation. The<br />

combination <strong>of</strong> <strong>the</strong>se guidelines (ranging from <strong>the</strong> social and natural sciences) is used to create <strong>the</strong><br />

necessary methodological approach to deal with this complex issue. The proposed methodology will<br />

consist <strong>of</strong> three main phases. These are: i) pr<strong>of</strong>ile and background analysis ii) vulnerability<br />

assessment and iii) Feedback and recommendations.<br />

i) Pr<strong>of</strong>ile and Background Analysis<br />

This initial phase will consist <strong>of</strong> developing a background analysis <strong>of</strong> <strong>the</strong> climate<br />

change characteristics for <strong>the</strong> Caribbean region, which include <strong>the</strong> island <strong>of</strong> Puerto<br />

Rico. Four main characteristics will be studied: temperature changes, sea-level rise,<br />

rainfall patterns and extreme wea<strong>the</strong>r events (hurricanes and droughts). These<br />

characteristics were selected as <strong>the</strong>y provide a general overview <strong>of</strong> some <strong>of</strong> <strong>the</strong><br />

associated hazards related to climate change on <strong>the</strong> region. Information will be<br />

ga<strong>the</strong>red from secondary sources.<br />

ii) Vulnerability Assessment<br />

The next phase will consist <strong>of</strong> <strong>the</strong> development <strong>of</strong> a vulnerability assessment, given a<br />

series <strong>of</strong> spatial indicators. This assessment should serve as a tool to compare how<br />

<strong>the</strong> Island-nation is prone to be affected by climate change effects, to get a better<br />

understanding <strong>of</strong> what and where are main expected impacts, <strong>the</strong>ir severity and <strong>the</strong><br />

capacity to cope with <strong>the</strong> outcomes. This tool will look at <strong>the</strong> vulnerability <strong>of</strong> Puerto<br />

Rico in terms <strong>of</strong> <strong>the</strong> exposure, sensitivity and adaptive capacity towards climate<br />

change related hazards that are prone to impact <strong>the</strong> territory. The vulnerability<br />

assessment will take into account characteristics <strong>of</strong> <strong>the</strong> biophysical environment, built<br />

environment, socio-economic characteristics <strong>of</strong> <strong>the</strong> country.<br />

The assessment will be realized using Geographical Information Systems (<strong>GIS</strong>) to engage in <strong>the</strong><br />

analysis <strong>of</strong> <strong>the</strong> data. The data used for <strong>the</strong> indicators <strong>of</strong> exposure, sensitivity and adaptive capacity <strong>of</strong><br />

vulnerability will be normalized and reclassified in order to engage in <strong>the</strong> spatial analysis <strong>of</strong> <strong>the</strong><br />

Page | 29


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

assessment. The results <strong>of</strong> <strong>the</strong> indicators will be <strong>the</strong>n processed through statistical analysis to generate<br />

a raster file depicting <strong>the</strong> climate-change vulnerable areas <strong>of</strong> Puerto Rico. Table 1 presents <strong>the</strong> list <strong>of</strong><br />

indicators.<br />

Table 1: List <strong>of</strong> Data Components for Vulnerability Assessment<br />

Once <strong>the</strong> vulnerability assessment is generated, an analysis will be made to identify <strong>the</strong> most<br />

vulnerable areas. This will be realized using <strong>GIS</strong> cartographic tools to aid in <strong>the</strong> spatial representation<br />

<strong>of</strong> <strong>the</strong> vulnerabilities to climate impacts for <strong>the</strong> country. The results will be displayed through<br />

interactive maps that will portrait <strong>the</strong> vulnerable areas as well as <strong>the</strong> data from <strong>the</strong> indicators,<br />

following <strong>the</strong> framework used for <strong>the</strong> GRaBS Adaptation Action Planning Toolkit for European cities<br />

(see http://www.ppgis.manchester.ac.uk/grabs/). Then <strong>the</strong> results will be analyzed to understand <strong>the</strong><br />

human, environment and infrastructural implications <strong>of</strong> <strong>the</strong> potential hazards for <strong>the</strong> selected cities.<br />

iii) Feedback and Recommendations<br />

This final phase consist on generating an overall review <strong>of</strong> <strong>the</strong> vulnerability<br />

assessment. This will be done in order to evaluate possible constrains and limitations<br />

as a evaluation tool for climate change impacts, as well as providing feedback to<br />

improve its applicability for analyzing vulnerabilities in small island nations on <strong>the</strong><br />

Caribbean region. Also, this phase will include a series <strong>of</strong> normative and operational<br />

recommendations for coping with climate changes. The main findings and <strong>the</strong><br />

limitations <strong>of</strong> <strong>the</strong> research will be also presented and discussed in this final phase.<br />

4. Expected Results<br />

The results from this research will bring out not only <strong>the</strong> climate change vulnerable areas <strong>of</strong> Puerto<br />

Rico, but it could serve as a baseline for a framework to develop climate vulnerability assessments on<br />

Caribbean islands. The vulnerability assessment, in conjunction with its evaluation and feedback<br />

process, can present a first glance <strong>of</strong> <strong>the</strong> susceptible areas <strong>of</strong> <strong>the</strong> country. This, in turn, will serve as<br />

an initial step for a more in-depth exploration <strong>of</strong> <strong>the</strong> vulnerabilities <strong>of</strong> climate change in <strong>the</strong> islandnation.<br />

Also it should contribute to develop a more intensive analysis <strong>of</strong> <strong>the</strong> potential adaptation<br />

measures to cope with <strong>the</strong> expected climate impacts.<br />

The combination <strong>of</strong> data variables in conjunction with <strong>the</strong> use <strong>of</strong> <strong>GIS</strong> technologies should provide<br />

results <strong>of</strong> great importance for planners and policymakers in order to expand <strong>the</strong>ir understanding <strong>of</strong><br />

climate impacts for <strong>the</strong> country. <strong>GIS</strong> technologies have been proven to analyze complex amounts <strong>of</strong><br />

data to provide sound spatial information. Through <strong>the</strong> use <strong>of</strong> <strong>the</strong>se technologies, it is possible to<br />

facilitate <strong>the</strong> understanding <strong>of</strong> climate change impacts on a particular territory. This in turn can aid <strong>the</strong><br />

process to develop plans and policies by local governments and community groups not only in Puerto<br />

Page | 30


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Rico, but elsewhere.<br />

5. References:<br />

Allison, I., Bind<strong>of</strong>f, N.L., Bindschadler, R.A., Cox, P.M., de Noblet, N., England, M.H., Francis,<br />

J.E., Gruber, N., Haywood, A.M., Karoly, D.J., Kaser, G., Le Quéré, C., Lenton, T.M., Mann, M.E.,<br />

McNeil, B.I., Pitman, A.J., Rahmtorf, S., Rignot, E., Schellnhuber, H.J., Schneider, S.H., Sherwood,<br />

S.C., Somerville, R.C.J., Steffen, K., Steig, E.J., Visbeck, M., Weaver, A.J. (Eds.). (2009). The<br />

Copenhagen Diagnosis: Updating <strong>the</strong> World on <strong>the</strong> Latest Climate Science. The University <strong>of</strong> South<br />

Wales Climate Change <strong>Research</strong> Centre. Australia.<br />

Birkmann, J. (2006.) Measuring vulnerability to promote disaster-resilient societies: Conceptual<br />

frameworks and definitions. In Birkmann, J. (ed.) Measuring Vulnerability to Natural Hazards:<br />

Toward disaster resilient societies. United Nations University Press. Hong Kong.<br />

Brooks, N., Adger, W. N. and Kelly, P. M. (2005). The Determinants <strong>of</strong> Vulnerability and Adaptive<br />

Capacity at <strong>the</strong> National Level and <strong>the</strong> implications for Adaptation. Global Environmental Change.<br />

15(2), 151-163.<br />

Bush, D.M. (<strong>19</strong>95). Living with Puerto Rico Shore. Duke University Press. United States <strong>of</strong> America<br />

Dawson, R. J., Hall, J. W., Barr, S. L., Batty, M., Bristow, A. L., Carney, S., Dagoumas, A., Evans,<br />

S., Ford, A., Harwatt, H., Köhler, J., Tight, M. R., Walsh, C. L. and Zanni, A. M. (2009). A Blueprint<br />

for <strong>the</strong> Integrated Assessment <strong>of</strong> Climate Change in Cities. Tyndall Centre for Climate Change<br />

<strong>Research</strong> Working Paper 129. United Kingdom.<br />

Füssel, H-M., Klein, R. (2006). Climate Change Vulnerability Assessment: An Evolution <strong>of</strong><br />

Conceptual Thinking. Climatic Change. 75(3), 301-329.<br />

Mehrotra, S., Natenzon, C.E., Omojola, A., Folorunsho, R., Gilbride, J., Rosenzweig, C. (2009).<br />

Framework for City Climate Risk Assessment: Buenos Aires, Delhi, Lagos and New York. Fifth urban<br />

research symposium 'Cities and climate change: responding to an urgent agenda'. World Bank<br />

Commissioned <strong>Research</strong>. France.<br />

Metzger, M. J., Rounsevell, M.D.A., Acosta-Michlik, L., Leemans, R., Schröter, D. (2006). The<br />

Vulnerability <strong>of</strong> Ecosystem Services to Land Use Change. Agriculture, Ecosystems and Environment.<br />

114(1), 69-85.<br />

Moss, S., Pahl-Wostl, C. and Downing, T. (2001). Agent-based integrated Assessment Modelling:<br />

The Example <strong>of</strong> Climate Change. Integrated Assessment. 2(2) 17-30.<br />

Parry, M.L., Canziani, O.F., Palutik<strong>of</strong>, J.P., van der Linden, P.J., and Hanson, C.E. (eds.). (2007)<br />

Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution <strong>of</strong> Working Group II to<br />

<strong>the</strong> Fourth Assessment Report <strong>of</strong> <strong>the</strong> Intergovernmental Panel on Climate Change. Cambridge<br />

University Press. United Kingdom and United States <strong>of</strong> America.<br />

Patt, A. G., van Vuuren, D.P., Berkhout, F., Aaheim, A., H<strong>of</strong>, A.F., Isaac, M., Mechler, R. (2010).<br />

Adaptation in Integrated Assessment Modeling: Where do We Stand? Climatic Change. 99 (3-4),<br />

383-402.<br />

Schröter, D., Polsky, C., Patt, A. (2005). Assessing Vulnerabilities to <strong>the</strong> Effects <strong>of</strong> Global Change:<br />

An Eight-Step Approach. Mitigation and Adaptation Strategies for Global Change. 10 (4), 573-595.<br />

United Nations Environmental Programme. (2005). Caribbean Environment Outlook. Accessed on<br />

November 2009. http://www.unep.org/geo/pdfs/Caribbean_EO.pdf •<br />

United Nations Environmental Programme. (2008). Climate Change in <strong>the</strong> Caribbean and <strong>the</strong><br />

Challenge <strong>of</strong> Adaptation. United Nations Environmental Programme Regional Office for Latin<br />

America and <strong>the</strong> Caribbean. Panama.<br />

United Nations Framework Convention on Climate Change. (2010). Report <strong>of</strong> <strong>the</strong> <strong>Conference</strong> <strong>of</strong> <strong>the</strong><br />

Parties on its Fifteenth Session, Held in Copenhagen From 7 to <strong>19</strong> December 2009. Accessed on May<br />

2010.http://unfccc.int/resource/docs/2009/cop15/eng/11a01.pdf#page=4<br />

United Nations Framework Convention on Climate Change2. (2010). Outcome <strong>of</strong> <strong>the</strong> work <strong>of</strong> <strong>the</strong> Ad<br />

Hoc Working Group on long-term Cooperative Action under <strong>the</strong> Convention. Accessed on December<br />

2010. http://unfccc.int/files/meetings/cop_16/application/pdf/cop16_lca.pdf<br />

Page | 31


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Assessing Data Completeness <strong>of</strong> OpenStreetMap in <strong>the</strong> <strong>UK</strong> through an<br />

Automated Matching Procedure for Linear Data<br />

Thomas Koukoletsos 1 , Mordechai (Muki) Haklay 2 , Claire Ellul 3<br />

1,2,3 University College London, Gower Street, London, WC1E 6BT, <strong>UK</strong><br />

+44 20 7679 2745<br />

1 thomas.koukoletsos.09@ucl.ac.uk , 2 m.haklay@ucl.ac.uk , 3 c.ellul@ucl.ac.uk<br />

ABSTRACT<br />

OpenStreetMap‘s (OSM) increasing popularity and density urges researchers on studying its data quality.<br />

When this includes comparison with a reference dataset, a data matching is necessary for <strong>the</strong> comparison to<br />

be meaningful, usually performed manually at data preparation stage. We propose an automated matching<br />

method specifically designed for OSM in <strong>the</strong> <strong>UK</strong>, based on a multi-stage approach that combines geometric<br />

and attribute constraints. We apply it on rural and urban areas and we show how <strong>the</strong> results can be used to<br />

evaluate data completeness <strong>of</strong> OSM and <strong>the</strong> reference dataset.<br />

1. Introduction<br />

KEYWORDS: VGI, OSM, Spatial Data Quality, Data completeness, Data matching<br />

OpenStreetMap (OSM) is an open source web mapping application that is based on volunteered effort to<br />

create a free and worldwide spatial database. The increasing importance and acceptance <strong>of</strong> OSM makes it a<br />

valuable free geographical source but also creates <strong>the</strong> necessity <strong>of</strong> a tool to evaluate its quality, so that we<br />

know if it is fit for our purpose. Moreover, such a tool should enable <strong>the</strong> future re-evaluation <strong>of</strong> data quality,<br />

as OSM data is rapidly updated, with <strong>the</strong> minimum effort. Previous studies on OSM quality (Haklay, 2010;<br />

Girres and Touya, 2010) evaluate OSM by comparing it with an <strong>of</strong>ficial dataset <strong>of</strong> known quality. However,<br />

some <strong>of</strong> <strong>the</strong>ir more detailed methods are difficult to replicate and scale due to <strong>the</strong> need <strong>of</strong> manually matching<br />

features between <strong>the</strong> two datasets.<br />

An <strong>of</strong>ficial spatial dataset can be very different from <strong>the</strong> OSM dataset for <strong>the</strong> area. Any information with low<br />

or no commercial value compared to <strong>the</strong> difficulty in collecting, such as ‗cycleways‘ or ‗steps‘, may not be<br />

present in an <strong>of</strong>ficial dataset, unlike OSM. On <strong>the</strong> o<strong>the</strong>r hand, a uniform density can be guaranteed for all<br />

o<strong>the</strong>r data collected, even in rural areas, unlike OSM. Due to <strong>the</strong>se differences, all objects not present in both<br />

datasets have to be removed before comparison. This makes data matching an unavoidable task before any<br />

fur<strong>the</strong>r quality analysis, usually included in <strong>the</strong> ‗data preparation‘ stage.<br />

An automated matching procedure is <strong>of</strong>fered by Ludwig, Voss and Krause-Traudes (2010) for OSM in<br />

Germany, however <strong>the</strong>y exclude roads <strong>of</strong> no interest to geomarketing or with no name attribute and<br />

additionally <strong>the</strong>ir approach is not suitable for <strong>the</strong> general case due to <strong>the</strong> datasets involved.<br />

We propose an automated way <strong>of</strong> matching linear data for <strong>the</strong> OSM dataset in <strong>the</strong> <strong>UK</strong>, which can be<br />

generalised to o<strong>the</strong>r cases, based on a combination <strong>of</strong> geometric and attribute constraints that are applied<br />

differently during a multi-stage approach. After presenting <strong>the</strong> method, we test it in urban and rural areas and<br />

<strong>the</strong> results are used to evaluate data completeness <strong>of</strong> OSM and <strong>the</strong> reference dataset.<br />

2. How to customise a matching process for OSM<br />

The reference dataset that we selected is Ordnance Survey‘s (OS) ITN layer <strong>of</strong> Master Map. ITN is <strong>the</strong><br />

highest accuracy <strong>of</strong>ficial dataset in <strong>the</strong> <strong>UK</strong> for roads centre line. The evaluation method was developed to<br />

have <strong>the</strong> following characteristics:<br />

� It is a multi-stage approach, applying different constraints in each stage in order to find <strong>the</strong> best<br />

matching candidate where previous stages fail.<br />

Page | 32


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

� It accepts <strong>the</strong> inherent heterogeneity <strong>of</strong> OSM data by splitting <strong>the</strong> datasets into smaller areas, using<br />

<strong>the</strong> OS‘s 1 km2 national grid, so comparison is localised.<br />

� It uses OSM information partially. For example, we could perform a task on OSM primary roads, but<br />

we should also consider that some features may not have a road type attribute, or it may be<br />

differently expressed.<br />

� It combines distance, direction and length geometric constraints with road name and road type<br />

attributes.<br />

3.The matching process<br />

The process starts by splitting each OSM and ITN feature into minor linear parts <strong>of</strong> two points, called<br />

segments. Segmentation occurs whenever part <strong>of</strong> a feature changes direction, resulting in splitting features<br />

when necessary in many straight lines. An eight-stage approach follows, with <strong>the</strong> first 4 stages dealing with<br />

segments. Then <strong>the</strong> initial features are composed and <strong>the</strong> procedure continues with <strong>the</strong>m (table 1).<br />

Table 1. The proposed multi-stage approach<br />

3.1. Stage 1<br />

For each ITN and OSM segment we calculate length and orientation based on its coordinates. Since direction<br />

<strong>of</strong> segments is not important, orientation is converted to a range from 0 to 180° to simplify computation. For<br />

each ITN segment:<br />

• Distance: identify <strong>the</strong> corresponding OSM segments within a distance <strong>of</strong> D (Equation 1).<br />

(1)<br />

where α is an assumed average GPS= 2accuracy,×+ଶ<br />

doubled to cover older devices or satellite image ortho-rectification errors, and w is an assumption <strong>of</strong><br />

road width based on road type, to cover cases where <strong>the</strong> mapper walks on <strong>the</strong> side <strong>of</strong> <strong>the</strong> road instead <strong>of</strong><br />

along <strong>the</strong> axis.<br />

• Orientation: Within <strong>the</strong> above distance, OSM segments with similar orientation are identified, based on an<br />

angular tolerance φ (as Equation 2), calculated for each ITN segment based on a GPS accuracy assumption<br />

and a worst case scenario regarding <strong>the</strong> orientation. (Figure 1).<br />

Page | 33


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 1. Calculation <strong>of</strong> angular tolerance<br />

• Length: only OSM segments with length less than three times <strong>the</strong> ITN segment length are<br />

considered.<br />

The ITN and OSM pairs that are matched are marked. This stage deals with cases where we find only one<br />

OSM matching candidate.<br />

3.2. Stage 2<br />

This stage relies mostly on road name attribute. For every non-matched ITN segment, we follow <strong>the</strong> same<br />

Distance and Orientation constraints as in stage 1. For each OSM candidate an exact match in road name<br />

attribute is carried out. We exclude considering some OSM road types that we are definitely sure that <strong>the</strong>y<br />

are not present in ITN dataset (like ‗steps‘, ‗bridleways‘), however this is only to speed up <strong>the</strong> process.<br />

3.3. Stage 3<br />

This is <strong>the</strong> same as stage 2, applied to ITN segments not matched so far, using text similarity between road<br />

names, considering those with a score above 70% and accepting <strong>the</strong> candidate with <strong>the</strong> maximum value<br />

(equation 3). This level aims to deal with misspelling and use <strong>of</strong> abbreviations in OSM dataset.<br />

3.4. Stage 4<br />

This applies to segments without name, as well as with those unmatched so far. For every non-matched ITN<br />

segment, <strong>the</strong> same Distance and Orientation constraints as in previous stages are calculated. For every<br />

OSM segment found, <strong>the</strong> distances between ITN and OSM segment start-points and end-points are<br />

calculated correspondingly, accepting as a match <strong>the</strong> OSM segment with <strong>the</strong> lowest sum <strong>of</strong> values.<br />

3.5. Stage 5<br />

The segments are composed to create <strong>the</strong> initial features, based on <strong>the</strong> matching information ga<strong>the</strong>red. This<br />

stage deals with possible errors <strong>of</strong> <strong>the</strong> automatic procedure so far, leading to different classification <strong>of</strong><br />

segments comprising <strong>the</strong> same feature. For each OSM and ITN feature we calculate <strong>the</strong> length found as a<br />

match, and if it is above 50% <strong>of</strong> <strong>the</strong> feature length, <strong>the</strong> feature is considered as matched.<br />

3.6. Stage 6<br />

(2)<br />

(3)<br />

Page | 34


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

For each non-matched OSM feature that has a name attribute, and within a distance <strong>of</strong> twice <strong>the</strong> assumed<br />

GPS accuracy, ITN features with text similarity in road name are searched, using <strong>the</strong> same function as in<br />

stage 3. If we find any that score above 75%, we consider <strong>the</strong> OSM feature to have a match. We may exclude<br />

some OSM road types as in stages 2, 3.<br />

3.7. Stage 7<br />

For each non-matched OSM feature (excluding some OSM road types as in stages 2, 3, 6), a buffer <strong>of</strong> <strong>the</strong><br />

GPS accuracy width is applied. The length <strong>of</strong> ITN features found inside this buffer is calculated, as well as<br />

<strong>the</strong> length for matched OSM features. The examined feature‘s length is compared with <strong>the</strong> two lengths, and a<br />

match is considered when inside this buffer <strong>the</strong>re is an ITN line with similar length and at <strong>the</strong> same time<br />

<strong>the</strong>re is no OSM line with similar length accepted as a match so far (equations 4 and 5). For example, <strong>the</strong><br />

non-matched OSM feature shown in figure 2 will be considered to have a match.<br />

ITN length inside buffer – 2 x GPS accuracy > 0.8 x OSM feature length (4)<br />

OSM matched length inside buffer – 2 x GPS < 0.9 x OSM feature length (5)<br />

3.8. Stage 8<br />

Figure 2. Matching procedure - Stage 7<br />

A final stage deals with errors when features run close and along with <strong>the</strong> cell boundary. We merge all nonmatched<br />

features, as well as all matched ones. If a feature with <strong>the</strong> same id is found in both tables it means<br />

that it has a considerable length matched somewhere, so it is accepted as a match. However, this gives<br />

additional matching information for <strong>the</strong> whole dataset and should not be applied on <strong>the</strong> cells involved,<br />

because <strong>the</strong>re will usually be a reason why a part <strong>of</strong> feature is not found with a match for a specific cell (e.g.<br />

<strong>the</strong> corresponding line is in <strong>the</strong> adjacent cell).<br />

4. Evaluating <strong>the</strong> Method<br />

The method is tested by manually calculating <strong>the</strong> length <strong>of</strong> <strong>the</strong> misjudged features for both datasets and<br />

comparing it with <strong>the</strong> dataset‘s length. So far, a small number <strong>of</strong> tiles has been examined (around 4% and<br />

10% <strong>of</strong> urban and rural area <strong>of</strong> England respectively), revealing errors <strong>of</strong> less than 2% (urban) and 4% (rural)<br />

accordingly. In rural areas errors are doubled because <strong>the</strong> constraints are sometimes too strict regarding <strong>the</strong><br />

accuracy <strong>of</strong> <strong>the</strong> satellite image used by OSM as well as <strong>the</strong> ITN dataset‘s reduced accuracy; corresponding<br />

roads were found at bigger than <strong>the</strong> searching distances, (e.g. 70 m), and were classified as unmatched. Even<br />

so <strong>the</strong> error values remain low, demonstrating <strong>the</strong> efficiency and suitability <strong>of</strong> <strong>the</strong> method as a first stage <strong>of</strong><br />

an OSM automated quality assessment. More tiles are currently examined so that <strong>the</strong> evaluation is more<br />

representative.<br />

5. Applying <strong>the</strong> Method<br />

The method is applied in urban and rural areas <strong>of</strong> Greater London and west <strong>of</strong> Newcastle, comparing data for<br />

Page | 35


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

1686 km2 and 1270 km2 respectively (Figure 3). Table 2 provides information <strong>of</strong> <strong>the</strong> significance <strong>of</strong> <strong>the</strong><br />

stages 1-4 that deal with <strong>the</strong> segments. As shown, stage 4 is important for rural areas as it deals efficiently<br />

with <strong>the</strong> missing name attribute <strong>of</strong> OSM dataset, while stage 2 is significant in urban areas where OSM name<br />

attribute is usually present. Table 3 shows percentages <strong>of</strong> <strong>the</strong> length matched by stages 1-4 (segments) and 5-<br />

8 (features) compared to <strong>the</strong> whole dataset.<br />

Figure 3. Selected areas<br />

Table 2. Stages 1-4 contribution in matched segments<br />

Table 3. Stages contribution in total dataset<br />

The information ga<strong>the</strong>red can provide valuable information on <strong>the</strong> extent to which <strong>the</strong> two datasets agree<br />

with each o<strong>the</strong>r.<br />

6. Results<br />

A high matching percentage in OSM dataset means that most <strong>of</strong> its features are also present in ITN dataset,<br />

although it gives no information on completeness. If we also consider <strong>the</strong> ITN matching percentage for <strong>the</strong><br />

same cell, however, a better picture <strong>of</strong> <strong>the</strong> datasets is captured. For example, a low matching percentage in<br />

ITN dataset means that <strong>the</strong>re are ITN features with no match in OSM dataset. Combined with a high OSM<br />

matching percentage, it means that <strong>the</strong> ITN dataset is denser in <strong>the</strong> area examined. For each cell we can<br />

create a mixed percentage as <strong>the</strong> sum <strong>of</strong> <strong>the</strong> two matching percentages divided by 2, so that we can see where<br />

both datasets have high matching percentages. We can <strong>the</strong>n roughly distinguish 4 cases, as shown in table 4.<br />

Page | 36


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Table 4. General cases <strong>of</strong> matching score for each cell<br />

Case 2 dominates in <strong>the</strong> tested rural area, while case 3 usually appears in <strong>the</strong> urban area. This is also <strong>the</strong><br />

reason why <strong>the</strong> total matching percentage (table 3) is much lower for <strong>the</strong> ITN dataset in rural area and higher<br />

for <strong>the</strong> urban one.<br />

We classify each cell according to <strong>the</strong> matching percentage. The manual classification selected focuses high<br />

on <strong>the</strong> percentage scale, as most tiles achieve scores above 70%. Using <strong>the</strong> same classification in all graphs<br />

enables visual comparison and a better understanding <strong>of</strong> <strong>the</strong> results (figures 4 and 5).<br />

Figure 4. OSM, ITN and mixed matching results for <strong>the</strong> Urban Area tested<br />

Figure 5. OSM, ITN and mixed matching results for <strong>the</strong> Rural Area tested<br />

OSM is denser in <strong>the</strong> urban area and sparser in <strong>the</strong> rural area. This is in agreement with <strong>the</strong> results <strong>of</strong><br />

previous studies on OSM. The high values <strong>of</strong> <strong>the</strong> mixed results map show where both datasets are almost<br />

complete compared to one ano<strong>the</strong>r, however, since datasets may contain different information, <strong>the</strong> best<br />

expression would be that ‗<strong>the</strong>y agree with each o<strong>the</strong>r‘.<br />

The evaluation showed bigger errors in rural areas. As a result, <strong>the</strong> rural area was re-processed, using looser<br />

constraints regarding <strong>the</strong> searching distance and angular tolerance (almost double searching distance and<br />

slightly bigger angular tolerance). This led to slightly better matching results (table 5), however, rural<br />

inaccuracies are sporadic and unpredictable; <strong>the</strong>y may appear only to some features in an area, depending on<br />

Page | 37


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

<strong>the</strong> OSM data source (GPS or satellite image) as well as <strong>the</strong> reduced ITN rural accuracy. We rarely know <strong>the</strong><br />

OSM data source, as well as where we move from <strong>the</strong> ITN‘s urban to <strong>the</strong> rural accuracy. Additionally, <strong>the</strong>se<br />

looser constraints will not be suitable for urban areas.<br />

7. Future Work<br />

Table 5. Comparison <strong>of</strong> matching results when using looser constraints in rural area<br />

Additional manual evaluation <strong>of</strong> <strong>the</strong> method will give more representative results <strong>of</strong> its efficiency. We also<br />

intend to automate <strong>the</strong> distinction between urban and rural areas, so that different constraints are applied and<br />

better results are produced for rural areas. If selected OSM road types not present in ITN dataset are<br />

excluded before applying <strong>the</strong> method, it remains to be tested how results will be more representative <strong>of</strong> data<br />

completeness. Finally, we plan to move on to automate <strong>the</strong> evaluation <strong>of</strong> o<strong>the</strong>r OSM data quality elements as<br />

well, such as positional accuracy and attribute quality.<br />

8. Acknowledgments<br />

We thank <strong>the</strong> Ordnance Survey and OSM for <strong>the</strong> data used in this work. All figures and tables using OS data<br />

are ©Crown Copyright/database right 2011, an Ordnance Survey/EDINA supplied service.<br />

9. References<br />

Girres J-F and Touya G (2010). Quality Assessment <strong>of</strong> <strong>the</strong> French OpenStreetMap Dataset. Transactions in<br />

<strong>GIS</strong>, 14(4):435-459. DOI:10.1111/j.1467-9671.2010.01203.x<br />

Haklay M (2010). How good is OpenStreetMap information? A comparative study <strong>of</strong> OpenStreetMap and<br />

Ordnance Survey datasets for London and <strong>the</strong> rest <strong>of</strong> England, In Environment and Planning, 37(4):682-703<br />

DOI:10.1068/b35097.<br />

Ludwig, I., Voss, A. and Krause-Traudes, M. (2010). How Good is OSM? - Method and Results for<br />

Germany. In Sixth International <strong>Conference</strong> on Geographic Information Science 2010, Zurich, Switzerland<br />

14-17 Sep 2010<br />

10. Biography<br />

Thomas Koukoletsos is a Captain in <strong>the</strong> Greek Army, serving at Hellenic Military Geographical Service.<br />

Currently is an Mphil Student at UCL in <strong>the</strong> department <strong>of</strong> Civil, Environmental and Geomatic Engineering.<br />

His research interests are in VGI data quality.<br />

Mordechai (Muki) Haklay is a senior lecturer in Geographical Information Science in <strong>the</strong> department <strong>of</strong><br />

Civil, Environmental and Geomatic Engineering at UCL. His research interests are in public access to<br />

environmental information, Human-Computer Interaction (HCI) and Usability Engineering for <strong>GIS</strong>, and<br />

Societal aspects <strong>of</strong> <strong>GIS</strong> use.<br />

Claire Ellul is a Lecturer in Geographic Information in <strong>the</strong> department <strong>of</strong> Civil, Environmental and<br />

Geomatic Engineering at University College London. Her research interests include spatial databases and<br />

approaches for handling large quantities <strong>of</strong> spatial data.<br />

Page | 38


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Page | 39


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Open Source <strong>GIS</strong> for Small Organisations<br />

A.B.G Scott 1 , Dr S.H. Hallett 1<br />

1 Cranfield University, Cranfield, Bedfordshire MK43 0AL<br />

Tel. +44 (0) 1234 750111 Fax +44 (0) 1234 750875<br />

antony.scott@cranfield.ac.uk, s.hallett@cranfield.ac.uk, http://www.cranfield.ac.uk<br />

ABSTRACT<br />

With recent growth in <strong>the</strong> scope and quality <strong>of</strong> Free and Open Source <strong>GIS</strong> S<strong>of</strong>tware (FOSS4G), existing<br />

commercial <strong>GIS</strong> users now benefit from new options and business models for <strong>GIS</strong> deployment.<br />

FOSS4G has also provides opportunities for organisations who are not traditional <strong>GIS</strong> users, particularly <strong>the</strong><br />

case in small organisations with limited budgets. This paper reports on a study considering <strong>the</strong> potential<br />

uptake <strong>of</strong> FOSS4G tools by six SMEs in <strong>the</strong> context <strong>of</strong> <strong>the</strong>ir organizational objectives. Also presented is a<br />

suggested technological architecture supporting common goals <strong>of</strong> spatial data management and webmapping<br />

dissemination.<br />

1. Introduction<br />

KEYWORDS: FOSS4G, Open Source <strong>GIS</strong> adoption, spatial data infrastructure<br />

There has been a steady growth in <strong>the</strong> scope and quality <strong>of</strong> Free and Open Source <strong>GIS</strong> S<strong>of</strong>tware (FOSS4G)<br />

in recent years. This has not only provided existing users <strong>of</strong> commercial <strong>GIS</strong> s<strong>of</strong>tware with new options and<br />

business models for <strong>GIS</strong> deployment, but also provided opportunities for organisations which have not<br />

used <strong>GIS</strong> before. This is particularly relevant for small organisations having limited budgets.<br />

This paper describes a study carried out in 2010 which examined <strong>the</strong> implications for smaller<br />

organisations <strong>of</strong> FOSS4G, covering a number <strong>of</strong> aspects <strong>of</strong> <strong>GIS</strong> adoption and implementation relevant to<br />

small organisations. These included requirements definition, application definition and<br />

development, system architecture and design, s<strong>of</strong>tware component selection, and lastly support. Some six<br />

SME organisations participated in <strong>the</strong> research. Reference is made to a number <strong>of</strong> sample case studies<br />

selected from <strong>the</strong>se organisations, representing a range <strong>of</strong> sectors and size bands, and each at differing stages<br />

<strong>of</strong> FOSS4G adoption.<br />

Conclusions are drawn relating to <strong>the</strong> viability and sustainability <strong>of</strong> FOSS4G for smaller organisations, and<br />

recommendations for steps which could be taken to support organisations in taking this approach. A<br />

suggested technological architecture is presented, capable <strong>of</strong> supporting common goals identified in <strong>the</strong> case<br />

study organisations for spatial data management and web-mapping dissemination.<br />

2. Methodology<br />

The study commenced with a broad review <strong>of</strong> FOSS4G, including a review <strong>of</strong> recent literature and current<br />

applications. A candidate <strong>GIS</strong> s<strong>of</strong>tware architecture was developed, covering four key major stages in <strong>GIS</strong><br />

processing and publishing, namely: Data Discovery; Content Creation and Management; Publishing<br />

Management; and Content Publication.<br />

In parallel, six SME organisations were selected as sources <strong>of</strong> application requirements, drawn from a<br />

number <strong>of</strong> sectors, and at different stages <strong>of</strong> FOSS4G adoption. These included public bodies and voluntary<br />

and commercial organisations, all with fewer than 100 employees.<br />

Formal interviews were held with <strong>the</strong> organisations, covering <strong>the</strong>ir experience <strong>of</strong> using FOSS4G where<br />

relevant, and drawing out requirements for <strong>GIS</strong> applications. These requirements were documented<br />

Page | 40


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

and distilled into three sample Use Cases.<br />

From <strong>the</strong>se Use Cases, <strong>the</strong> s<strong>of</strong>tware review, and <strong>the</strong> candidate architecture, sample applications were<br />

designed, implemented in prototype form and evaluated.<br />

3. The state <strong>of</strong> FOSS4G<br />

It is contended that ―Open Source Web <strong>GIS</strong> s<strong>of</strong>tware systems have reached a stage <strong>of</strong> maturity,<br />

sophistication, robustness and stability, and usability and user friendliness rivalling that <strong>of</strong> commercial,<br />

proprietary <strong>GIS</strong> and Web <strong>GIS</strong> server products‖ (Buolos & Honda 2006). The number <strong>of</strong> applications<br />

available has been growing steadily, with <strong>the</strong> Free<strong>GIS</strong> database (http://www.freegisorg) currently listing over<br />

300 packages, including an ―Application‖ category with values <strong>of</strong> ―Miscellaneous, Web <strong>GIS</strong>, Desktop <strong>GIS</strong>,<br />

Library, Database, Mobile <strong>GIS</strong>‖. Development is most noticeable in <strong>the</strong> Desktop <strong>GIS</strong> and Web <strong>GIS</strong><br />

categories. However categorisation is not straightforward. Sherman (2008) observes that ―many applications<br />

are <strong>of</strong> <strong>the</strong> ―Swiss Army knife‖ variety, providing a wide range <strong>of</strong> applications‖ and <strong>the</strong>re are many libraries<br />

and packages which are used as components in more than one application.<br />

The development <strong>of</strong> Web <strong>GIS</strong> has to both been fuelled by, and is a significant contributor to, <strong>the</strong> Web 2.0<br />

movement. Google Maps, which although proprietary is free at <strong>the</strong> point <strong>of</strong> use in many cases, has been one<br />

<strong>of</strong> <strong>the</strong> major drivers <strong>of</strong> Web 2.0, and is a major influence on contemporary open source web mapping. O<strong>the</strong>r<br />

mapping services such as Bing maps and OpenStreetMap underpin this evolving development.<br />

A fur<strong>the</strong>r enabling factor for <strong>the</strong> growth <strong>of</strong> FOSS4G, and a critical one for smaller organisations, has been<br />

<strong>the</strong> continuing trend towards open data, with <strong>the</strong> <strong>UK</strong> Government‘s data.gov.uk website and <strong>the</strong> Ordnance<br />

Survey‘s OpenData initiative representing two relatively new sources <strong>of</strong> extensive open geographical data<br />

relating to <strong>the</strong> <strong>UK</strong>. The availability <strong>of</strong> this data is for some organisations, lacking <strong>the</strong> scale to justify<br />

licensing data at what is <strong>of</strong>ten a high minimum cost threshold, a fur<strong>the</strong>r enabling factor in <strong>GIS</strong> adoption and<br />

use. The result is that for many organisations, opportunities are arising for <strong>the</strong> first time to not only publish<br />

<strong>the</strong>ir own data without incurring licence costs, but also to integrate it with both general and specialise<br />

external data published by third parties. This is particularly important from a resourcing perspective for<br />

voluntary organisations, but also lends weight to <strong>the</strong> role <strong>of</strong> consultancies, including those in <strong>the</strong> case study<br />

organisations described below, who noted that <strong>the</strong>y<br />

have a role to play in advising <strong>the</strong>ir clients on appropriate data sources, and how to integrate <strong>the</strong>m with <strong>the</strong>ir<br />

own systems – both issues which small organisations may find difficult to address unaided.<br />

4. Requirements and Use Cases<br />

As part <strong>of</strong> a discussion on both actual and potential use <strong>of</strong> FOSS4G, requirements for <strong>GIS</strong> applications were<br />

articulated by <strong>the</strong> case study organisations, and are summarised in Table 1.<br />

Page | 41


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Table 1: Case Study Organisations<br />

Three Use Cases were developed which represented a sample <strong>of</strong> <strong>the</strong>se requirements. These are described<br />

below.<br />

4.1 Use Case 1: Publish Area-based Statistics<br />

This Use Case, shown in Figure 1, covers <strong>the</strong> process <strong>of</strong> publishing online <strong>UK</strong>-based statistics relating to<br />

poverty, in a way that is accessible to both <strong>the</strong> specialist and general end user. Data sources used are public,<br />

and <strong>the</strong> value added by <strong>the</strong> site is to publish map-based representations <strong>of</strong> <strong>the</strong> statistics which are easy to use,<br />

navigate and extract data from. The inclusion <strong>of</strong> an „Interface Designer‟ role in <strong>the</strong> process is <strong>the</strong>refore<br />

important to realise this goal.<br />

Page | 42


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

4.2 Use Case 2: Publish Property Portfolio Data<br />

Figure 1: Use Case 1 - Publish Area-based Statistics<br />

This Use Case, shown in Figure 2, addresses <strong>the</strong> processes <strong>of</strong> making available data related to property<br />

portfolios through a mapping interface. It assumes that analysis is required to obtain data <strong>of</strong> value for<br />

individual properties, and to represent that data in an appropriate way at higher levels <strong>of</strong> geography, for<br />

example Postcodes or administrative areas.<br />

Page | 43


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 2: Use Case 2 - Publish Property Portfolio Data<br />

4.3 Use Case 3: Renewable Resource Assessment<br />

This Use Case, shown in Figure 3, describes <strong>the</strong> process <strong>of</strong> recording energy demand information for a<br />

geographical area at as low a level <strong>of</strong> granularity as possible, with <strong>the</strong> purpose <strong>of</strong> estimating <strong>the</strong> feasibility <strong>of</strong><br />

siting generation facilities in <strong>the</strong> area, typically a Combined Heat and Power (CHP) plant. Data is ga<strong>the</strong>red<br />

where possible at <strong>the</strong> individual building level, including actual consumption figures, and estimated<br />

consumption based on building size and type. For larger sites, where individual building-level data ga<strong>the</strong>ring<br />

was not possible, aggregated area statistics could be used.<br />

The main output <strong>of</strong> <strong>the</strong> Use Case would be a geographically-referenced dataset <strong>of</strong> heat demand which could<br />

be used to model <strong>the</strong> feasibility <strong>of</strong> one or more generation sites. The dataset could also be used to provide a<br />

visual mapped representation <strong>of</strong> <strong>the</strong> demand data.<br />

Page | 44


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

5. Implementation<br />

5.1 System Architecture<br />

Figure 3: Use Case 3 - Renewable Resource Assessment<br />

It was important to devise an architectural schema to guide <strong>the</strong> selection <strong>of</strong> s<strong>of</strong>tware tools and capabilities,<br />

and Figure 4 shows <strong>the</strong> architecture developed to meet <strong>the</strong> requirements <strong>of</strong> <strong>the</strong> Use Cases. The focus on web<br />

publishing reflects <strong>the</strong> requirements <strong>of</strong> many <strong>of</strong> <strong>the</strong> scenarios described. The architecture is never<strong>the</strong>less<br />

flexible, and subsets <strong>of</strong> it are capable <strong>of</strong> independent deployment.<br />

Four top-level process groups are identified at <strong>the</strong> top <strong>of</strong> <strong>the</strong> diagram. Below <strong>the</strong>se, grey boxes represent subtasks<br />

or processes which make up <strong>the</strong> group, and yellow boxes s<strong>of</strong>tware components required to fulfill <strong>the</strong><br />

task.<br />

Page | 45


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

5.2 System Components<br />

Figure 4: System Architecture<br />

Application components were selected to populate <strong>the</strong> architecture, and to allow an integrated application<br />

infrastructure to be put into place. The suggested FOSS4G s<strong>of</strong>tware tools are shown in Figure 5.<br />

Page | 46


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

5.3 Prototype Applications<br />

Figure 5: System Architecture with Components Used<br />

Prototype applications were developed to demonstrate how <strong>the</strong> needs <strong>of</strong> <strong>the</strong> Use Cases could be addressed,<br />

and screenshots from <strong>the</strong>se applications are shown below. Note that all source data is stored in a<br />

PostgreSQL/Post<strong>GIS</strong> database<br />

Figure 6 shows <strong>the</strong> Standard Assessment Procedure (SAP) energy ratings <strong>of</strong> a property stock portfolio,<br />

averaged at Postcode and management area levels, generated by Quantum <strong>GIS</strong>.<br />

Figure 6: Quantum <strong>GIS</strong> Property Stock Map<br />

Page | 47


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 7 uses <strong>the</strong> same dataset as in Figure 6, presenting this data in a web interface with a popup showing<br />

detail, using GeoServer and MapFish.<br />

Figure 7: Web Property Stock Map with Popup<br />

Figure 8 shows <strong>the</strong> GeoServer management interface, with available layers (including those published in<br />

Figure 7) from all sources displayed.<br />

Figure 8: GeoServer Layer Configuration<br />

Page | 48


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Figure 9 shows a Quantum <strong>GIS</strong> view <strong>of</strong> a CHP feasibility study, with polygons <strong>of</strong> candidate buildings and<br />

Ordnance Survey OpenData background. The s<strong>of</strong>tware dialogues allow attribute data for individual buildings<br />

to be edited.<br />

Figure 9: Quantum <strong>GIS</strong> Interface with Edit Dialogue for CHP Feasibility<br />

Figure 9 shows a web-based view <strong>of</strong> <strong>the</strong> data in Figure 9, with a popup showing attribute data for individual<br />

buildings.<br />

Figure 10: Web Presentation <strong>of</strong> CHP Estimator<br />

Page | 49


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

6. Discussion and Conclusions<br />

With <strong>the</strong> rapid development <strong>of</strong> FOSS4G, barriers to entry for small organisations represented by licensing<br />

costs have largely disappeared. A similar impact is emerging from <strong>the</strong> decision <strong>of</strong> many public-sector bodies<br />

to release geospatial datasets at little or no cost. However a decision on adoption <strong>of</strong> FOSS4G should be<br />

guided by a wider range <strong>of</strong> criteria than cost. Perhaps surprisingly, it appears that FOSS can bring benefits in<br />

areas such as customisation and modularity, and <strong>the</strong> limited scope <strong>of</strong> some open source <strong>GIS</strong> applications can<br />

itself be an attraction to an organisation for which commercial <strong>GIS</strong> products appear daunting.<br />

The following conclusions are drawn from this study:<br />

� The FOSS4G market has reached a level <strong>of</strong> maturity where production-quality implementations are<br />

realistic and possible. Many applications can be installed and brought into use with minimum<br />

configuration. This has been demonstrated by <strong>the</strong> sample applications developed for <strong>the</strong> study, and<br />

by real-world case studies.<br />

� The extent and complexity <strong>of</strong> <strong>the</strong> market means that selecting appropriate applications from <strong>the</strong> wide<br />

range <strong>of</strong> tools available remains difficult. There is no panacea. Instead, a range <strong>of</strong> <strong>of</strong>ferings, each<br />

with specific functionalities and advantages, may be considered as potential candidates for adoption.<br />

This is somewhat mitigated by support provided by organisations such as OSGeo, but <strong>the</strong><br />

component-based nature <strong>of</strong> FOSS, contrasted with <strong>the</strong> comprehensive packages provided by<br />

commercial vendors, requires decisions on each component <strong>of</strong> an architecture, coupled with <strong>the</strong> need<br />

to ensure <strong>the</strong>se decisions are compatible.<br />

� The limited size <strong>of</strong> an organisation should not preclude a thorough analysis and articulation <strong>of</strong><br />

requirements before decisions are made on implementation. There <strong>the</strong> easy availability <strong>of</strong> FOSS4G<br />

could encourage an over-experimental approach, with resource time squandered on non-essential<br />

applications. The discipline <strong>of</strong> formal systems analysis approaches and clear<br />

requirements definition, and documentation such as Use Cases, helps to focus scarce resources on<br />

core functionality.<br />

� FOSS <strong>of</strong>fers an opportunity to build applications incrementally, and subject to <strong>the</strong> caveat on<br />

requirements definition above, some experimentation within <strong>the</strong> scope <strong>of</strong> stated requirements will<br />

assist <strong>the</strong> learning process and inform <strong>the</strong> feasibility stage.<br />

� External assistance with <strong>the</strong> selection and implementation <strong>of</strong> appropriate applications could represent<br />

a worthwhile investment. Without <strong>the</strong> marketing framework or widely available peer experience<br />

which characterises commercial s<strong>of</strong>tware, <strong>the</strong> task <strong>of</strong> evaluating and deploying ―best <strong>of</strong> breed‟<br />

applications for a specific scenario will be beyond <strong>the</strong> capability <strong>of</strong> most small organisations.<br />

Whe<strong>the</strong>r on a voluntary, mutual or commercial basis, organisations should seek to draw on <strong>the</strong><br />

experience and expertise <strong>of</strong> o<strong>the</strong>rs before committing <strong>the</strong>mselves to FOSS4G. Networks will emerge<br />

to aid this process, and <strong>the</strong>re are great opportunities for emerging service providers to enter this<br />

market.<br />

Recommendations can also be made for <strong>the</strong> wider FOSS4G community to ease <strong>the</strong> entry <strong>of</strong> small<br />

organisations into <strong>the</strong> field.<br />

� Greater availability <strong>of</strong> resources such as case studies, training events, online materials and specialist<br />

consultancy support would all help to give confidence and impart knowledge to small organisations<br />

approaching FOSS4<strong>GIS</strong> for <strong>the</strong> first time. OSGeo and its <strong>UK</strong> chapter, OSGeo:<strong>UK</strong> have laid <strong>the</strong><br />

groundwork for <strong>the</strong>se resources to be made available, but contributions are needed from o<strong>the</strong>rs to a<br />

resource library for potential users.<br />

� Support in application selection and deployment could be enhanced by <strong>the</strong> greater availability <strong>of</strong><br />

standard <strong>GIS</strong> packages, drawn from OSGeo projects, and backed by commercial support where<br />

required. Given <strong>the</strong> limited extent <strong>of</strong> OSGeo resources, this would probably need to be a<br />

commercially-backed initiative ra<strong>the</strong>r than an OSGeo one, but <strong>the</strong>re could be value to consultancies<br />

in assembling such a package and <strong>of</strong>fering support to users.<br />

� Pr<strong>of</strong>essional bodies such as <strong>the</strong> Association for Geographic Information (AGI), Royal Geographical<br />

Page | 50


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Society (RGS) and <strong>the</strong> Royal Institution for Chartered Surveyors (RICS) could extend <strong>the</strong>ir services<br />

to <strong>of</strong>fer greater support to small member organisations and individuals<br />

for whom FOSS is appropriate. Through <strong>the</strong> provision <strong>of</strong> education, conferences and written<br />

resources, <strong>the</strong>se organisations could extend entry to <strong>GIS</strong>, and <strong>the</strong>ir own membership, embracing <strong>the</strong><br />

FOSS movement and acknowledging <strong>the</strong> opportunities it represents.<br />

In some sectors, joint approaches will provide <strong>the</strong> appropriate means to adopt FOSS4G, and domain (as<br />

opposed to <strong>GIS</strong>-focussed) organisations may have a role to play (similarly to <strong>the</strong> emergence <strong>of</strong> personal<br />

website domains, such as Geocities, in <strong>the</strong> last decade). For example, from discussion with <strong>the</strong> bat<br />

conservation group (organisation 3 above), it would be <strong>of</strong> benefit to o<strong>the</strong>r bat conservation groups<br />

throughout <strong>the</strong> country if <strong>the</strong> overarching Bat Conservation Trust were able to package a standard <strong>GIS</strong><br />

application suite for <strong>the</strong> purposes <strong>of</strong> bat movement monitoring. This model would be applicable to o<strong>the</strong>r<br />

cases in <strong>the</strong> voluntary sector.<br />

7. Acknowledgements<br />

The authors <strong>of</strong>fer grateful thanks to <strong>the</strong> organisations which made <strong>the</strong>mselves available to be<br />

interviewed for this study.<br />

8. References<br />

Buolos, M. & Honda, K., (2006). Web <strong>GIS</strong> in practice IV: publishing your health maps and connecting to<br />

remote WMS sources using <strong>the</strong> Open Source UMN MapServer and DM Solutions MapLab. International<br />

Journal <strong>of</strong> Health Geographics, 5(6).<br />

Sherman, G., (2008). Desktop <strong>GIS</strong>: Mapping <strong>the</strong> Planet with Open Source Tools, Pragmatic<br />

Bookshelf.<br />

9. Biographies<br />

Antony Scott has recently completed a Masters Degree in Geographical Information Management at<br />

Cranfield University, having worked for a number <strong>of</strong> years in <strong>the</strong> information management, s<strong>of</strong>tware<br />

development and consultancy fields. He is currently working with a small environmental consultancy on an<br />

open source <strong>GIS</strong> implementation programme.<br />

Dr Steve Hallett acts as Group Manager <strong>of</strong> <strong>the</strong> Cranfield University National Soil Resources Institute‟s Soil<br />

Resources Group, where he holds responsibility for Cranfield‟s national Land Information System (LandIS).<br />

He has expertise in <strong>the</strong> application <strong>of</strong> spatial computer-based methodologies and <strong>GIS</strong> to <strong>the</strong> environmental<br />

sciences with particular reference to soil and natural environmental systems.<br />

Page | 51


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

Annotating Spatial Features in OpenStreetMap<br />

Peter Mooney and Padraig Corcoran<br />

Department <strong>of</strong> Computer Science, National University <strong>of</strong> Ireland Maynooth,<br />

Maynooth,Co. Kildare. Ireland.<br />

email {peter.mooney,padraig.corcoran}@nuim.ie Tel: 353 (1) 2680100, Fax: 353 (1) 2680<strong>19</strong>9<br />

http://www.cs.nuim.ie/˜pmooney<br />

ABSTRACT<br />

OpenStreetMap (OSM) is, potentially, <strong>the</strong> most famous example <strong>of</strong> Volunteered Geographic Information<br />

(VGI) on <strong>the</strong> Internet today. OSM volunteers contribute spatial con-tent to <strong>the</strong> global OSM database. These<br />

contributors are encouraged to `tag' content under <strong>the</strong> guidance <strong>of</strong> a flexible community endorsed ontology<br />

<strong>of</strong> spatial object tags provided on <strong>the</strong> OSM Wiki. This paper explores how OSM contributors edit spatial<br />

objects with an emphasis on 'tagging' generated from an examination <strong>of</strong> <strong>the</strong> historical evolution <strong>of</strong><br />

geographic features such as lakes, rivers, roads, forests, in OSM.<br />

1. Introduction<br />

KEYWORDS: OpenStreetMap, Quality, Web <strong>GIS</strong>, VGI<br />

The OpenStreetMap object (id = 24015216 - see Forest (2010)) is a forest near Stuttgart in Germany. It was<br />

first contributed in April 2008. Currently (December 2010) it has received 43 subsequent revisions <strong>the</strong> last<br />

<strong>of</strong> which was performed in August 2010. Upon closer inspection <strong>of</strong> <strong>the</strong> history <strong>of</strong> <strong>the</strong> revisions to this feature<br />

one sees that <strong>the</strong> object was repeatedly tagged as ei<strong>the</strong>r ―landuse=forest‖ or as ―natural=wood‖ and on one<br />

occasion as a ―highway=primary‖. Depending on when o<strong>the</strong>r users downloaded <strong>the</strong> OSM data for this region,<br />

for mapping and/or <strong>GIS</strong> analysis, <strong>the</strong>y would have been presented with a different tag for this object. Which<br />

version is correct? The ambiguity between versions is one example <strong>of</strong> data quality issues surrounding <strong>the</strong><br />

OpenStreetMap (OSM) project. OSM is a collaborative project to create a fully free and openly accessible<br />

map <strong>of</strong> <strong>the</strong> world. Volunteers in <strong>the</strong> OSM community collect geographic information and submit this to<br />

<strong>the</strong> global OSM database (Ciepluch et al.; 2009). Currently OSM is used primarily for rendering various<br />

map visualizations (Auer et al.; 2009). Over et al. (2010) comment that <strong>the</strong> greatest obstacle to more wider<br />

use <strong>of</strong> OSM is spatial data inhomogenity which is preventing OSM being used in ―geomatics applications‖.<br />

Contributors are <strong>the</strong> cornerstone <strong>of</strong> OSM. This paper provides result <strong>of</strong> an analysis <strong>of</strong> contributions to two<br />

large OSM databases.<br />

2. Working with OpenStreetMap data<br />

OSM data is represented by adhering to a relatively simple data model comprised <strong>of</strong> three basic elements<br />

- nodes, ways and relations (Auer et al.; 2009). The creator <strong>of</strong> OpenStreetMap Steve Coast comments that<br />

―metadata in OSM is open ended and simple‖ (Coast; 2010). There are no restrictions whatsoever regarding<br />

<strong>the</strong> use <strong>of</strong> attributes or attribute values to annotate elements in OSM. Consequently users can create arbi-<br />

trary attributes or attribute values (Auer et al.; 2009). The Map Features page on <strong>the</strong> OpenStreetMap Wiki<br />

Page | 52


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1a: Disaster Risk Management<br />

(OSM; 2010) describes <strong>the</strong> OSM community agreed ontology <strong>of</strong> terms, or ―Tags‖, to describe <strong>the</strong> geograph-<br />

ical features in <strong>the</strong> OSM database. However <strong>the</strong> use <strong>of</strong> <strong>the</strong> ontology is not strict and is provided as a guide.<br />

Coast (2010) comments that by not constraining contributors with an ontology two things are possible. First,<br />

creative and unexpected types <strong>of</strong> geo data can be added and secondly one exposes a playful aspect <strong>of</strong> <strong>the</strong><br />

project which is to allow experimentation. This is broadly in agreement with Wang et al. (2010) who remark<br />

that ―people seem to like <strong>the</strong>se collaborative projects because <strong>the</strong>y enjoy <strong>the</strong> openness <strong>of</strong> social media‖. In<br />

relation to <strong>the</strong> example presented at <strong>the</strong> beginning <strong>of</strong> this paper Comber et al. (2005) comment that as ―<strong>the</strong><br />

number <strong>of</strong> nonspecialist users <strong>of</strong> GI increases and spatial data are used to answer more questions about <strong>the</strong><br />

environment, <strong>the</strong> need for users to understand <strong>the</strong> wider meaning <strong>of</strong> <strong>the</strong> data concepts becomes more urgent‖.<br />

Some authors have found that ―indexing content through tagging is prone to unsystematic and inconsistent<br />

metadata that can potentially harm retrieval performance and make fur<strong>the</strong>r analysis difficult‖ (Wang et al.;<br />

2010). In this paper we provide some results from an analysis <strong>of</strong> tagging and editing <strong>of</strong> geographic features in<br />

OpenStreetMap databases by analysing <strong>the</strong> history <strong>of</strong> edits to over 4000 geographic objects. There is little<br />

literature published, to our current knowledge, which deals specifically with <strong>the</strong> metadata or tagging<br />

behaviour <strong>of</strong> contributors to OpenStreetMap. A considerable body <strong>of</strong> literature exists on tagging and annota-<br />

tion behaviour <strong>of</strong> contributors in social networks and enterprise systems such as Flickr, YouTube, Delicious,<br />

etc. This literature provides very helpful and informative support studies. The OpenStreetMap ontology is<br />

best described as a folksonomy where ―folksonomies and social tagging‖ provides a cheaper and more natural<br />

way <strong>of</strong> organising web objects (Gupta et al.; 2010). In Gupta et al. (2010) <strong>the</strong> authors describe a folksonomy as<br />

(folk (people) + taxis (classification) + nomos (management)) - a user-generated classification emerging from<br />

a bottom up consensus. Crucially unlike formal taxonomies <strong>the</strong> concept <strong>of</strong> a folksonomy is one where no<br />

explicity relation is defined between terms. It has been shown that since users <strong>the</strong>mselves tag <strong>the</strong> objects,<br />

sometimes from a suggested list <strong>of</strong> possible terms, <strong>the</strong> folksonomy directly reflects <strong>the</strong> vocabulary <strong>of</strong> <strong>the</strong><br />

user/contributor (Milicevic et al.; 2010). Ambiguity arises in folksonomies because different users apply<br />

different tags to different objects. Acronyms can lead to ambiguity as can spelling errors or <strong>the</strong> combination <strong>of</strong><br />

several words into a single word tag. In OpenStreetMap objects are permitted multiple tags. However each<br />

tag can only be assigned a single value. In <strong>the</strong> next section we analyse how annotation and edits <strong>of</strong> spatial<br />

objects in OSM are performed.<br />

3. Experimental Analysis<br />

OSM data is freely available, in OpenStreetMap XML format, from <strong>the</strong> GeoFabrik website http://<br />

download.ge<strong>of</strong>abrik.de/. This data is updated almost hourly so <strong>the</strong> most up-to-date version <strong>of</strong> <strong>the</strong><br />

OpenStreetMap database is always available. We downloaded <strong>the</strong> OSM-XML for England and Germany‘s<br />

Baden-Wü rttemberg region. These two locations were chosen because <strong>the</strong>y have two <strong>of</strong> <strong>the</strong> most active<br />

OSM communities in Europe subsequently providing us with a large set <strong>of</strong> geographic objects for analysis.<br />

This could be easily extended to o<strong>the</strong>r regions with less active OSM communities. We extracted all ways<br />

with at least 20 versions <strong>of</strong> edits. For England <strong>the</strong>re are 3250 ways with at least 20 versions while Baden-<br />

Wü rttemberg has 909 such ways. For each way w we compute a number <strong>of</strong> characteristics for each version<br />

wv <strong>of</strong> w including: wvn - <strong>the</strong> number <strong>of</strong> nodes in wv, wvT - <strong>the</strong> set <strong>of</strong> tags (key,value) pairs annotating wv,<br />

wvu - <strong>the</strong> user id <strong>of</strong> <strong>the</strong> user who created wv, and wvt <strong>the</strong> timestamp <strong>of</strong> <strong>the</strong> edit <strong>of</strong> wv. Table 1 shows <strong>the</strong><br />

distribution <strong>of</strong> time between consecutive edits wvt1 and wvt2 for all ways in both OSM datasets. There are<br />

a number <strong>of</strong> interesting observations. Almost 42% <strong>of</strong> consecutive edits are separated by an editing time <strong>of</strong> 1<br />

week to 1 month. Almost 38% <strong>of</strong> consecutive edits have 1 hour and 24 hours between <strong>the</strong>m.<br />

There are a number <strong>of</strong> additional observations, extracted from this analysis, worthy <strong>of</strong> fur<strong>the</strong>r discussion.<br />

These are summarised as follows:<br />

• In Baden-Württemberg 93 (10%) <strong>of</strong> polygons (from 909) have no tags at some stage <strong>of</strong> <strong>the</strong>ir evolution<br />

Page | 53


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

Table 1: Distribution <strong>of</strong> time between consecutive edits to all ways<br />

� In England 531 (16%) <strong>of</strong> polygons (from 3250) have no tags at some stage <strong>of</strong> <strong>the</strong>ir evolution<br />

� The name <strong>of</strong> a geographic feature is one <strong>of</strong> <strong>the</strong> most basic metadata attribute values for<br />

spatial data. 2827 objects use <strong>the</strong> name tag. The number <strong>of</strong> objects tagged with <strong>the</strong> name tag<br />

at <strong>the</strong>ir first version is 1<strong>19</strong>0 or (42%)<br />

� Of <strong>the</strong> objects which use <strong>the</strong> name tag 823 <strong>of</strong> <strong>the</strong>se objects are first tagged with a name tag as<br />

late as <strong>the</strong> 10TH version <strong>of</strong> <strong>the</strong>ir evolution<br />

� In total 114 objects are created with no name tag but <strong>the</strong>ir current version contains a valid<br />

value for <strong>the</strong> name tag<br />

� There are 3332 unique editors. The mean number <strong>of</strong> editors for all objects is 5.892. The<br />

median number <strong>of</strong> editors <strong>of</strong> 5.00 with a standard deviation <strong>of</strong> 3.794. This goes some way<br />

towards supporting <strong>the</strong> anecdotal claim that only a small number <strong>of</strong> editors do most <strong>of</strong> <strong>the</strong><br />

editing in OSM.<br />

� Only 30 objects have twenty or more editors - 7 <strong>of</strong> <strong>the</strong>se objects represent road polylines with<br />

<strong>the</strong> remainder representing forest features<br />

� There are 120929 unique object versions. 231 users contributed 71% <strong>of</strong> <strong>the</strong>se versions. 39<br />

―super contributors‖ (those who contributed more than 500 edits) are responsible for 39.5%<br />

<strong>of</strong> all edits. Haklay et al. (2010) argue that in OSM <strong>the</strong>re is a ―decreased gain in terms <strong>of</strong><br />

positional accuracy when <strong>the</strong> number <strong>of</strong> contributors passes about 10 or so‖.<br />

� In table 2 an example is provided <strong>of</strong> where <strong>the</strong> ―name‖ tag <strong>of</strong> a street (located in Sou<strong>the</strong>ndon-Sea,<br />

England) changes multiple times. The current version is V26. The name changes 4<br />

times from it's original ―Thames Drive‖ to current version <strong>of</strong> ―Grande Parade‖. Three<br />

distinct users are involved.<br />

� Changes to Name Tags: There are 285 polygons in our test set which exhibit a ―name‖ tag<br />

which is changed 3 or more times. For each way we clustered <strong>the</strong> assigned name tags into<br />

chronological groups and <strong>the</strong>n compared <strong>the</strong> transformation <strong>of</strong> tags into one ano<strong>the</strong>r using<br />

two well known string matching metrics to quantify how similar <strong>the</strong> name tags were. The<br />

Levenshtein distance is defined as <strong>the</strong> minimal number <strong>of</strong> characters you have to replace,<br />

insert or delete to transform from one string to ano<strong>the</strong>r (Yujian and Bo; 2007). The<br />

JaroWinkler distance (Bilenko et al.; 2003) is a similar metric used mostly for duplicate<br />

detection in databases. The metric is normalized such that 0 equates to no similarity and 1 is<br />

an exact match between <strong>the</strong> two strings. In Figure 1 we show a plot <strong>of</strong> <strong>the</strong> mean<br />

Table 2: An example <strong>of</strong> <strong>the</strong> ―name‖ tag changing on a street polyline (4803031) in OSM<br />

Page | 54


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

Levenshtein distance against <strong>the</strong> mean JaroWinkler distance for each qualifying spatial object. Most<br />

objects are clustered around a mean Levenshtein distance <strong>of</strong> 10 and mean JaroWinkler distance <strong>of</strong> 0.5<br />

which indicates that <strong>the</strong> changes from one name tag to <strong>the</strong> next name tag are substantially different.<br />

This is potentially caused by contributors : spelling place names incorrectly, providing local<br />

variations on <strong>of</strong>ficial place names, incorrect naming <strong>of</strong> streets, and correction or spelling.<br />

� The polylines representing ―highways‖ were analysed. In our test set <strong>the</strong>re are 2889 polylines<br />

tagged as highways (trunk, motorway, residential, etc). Of <strong>the</strong>se highway polygons 1143<br />

changed highway designation at least once - for example <strong>the</strong>ir tag changed from primary to<br />

secondary. Close inspection <strong>of</strong> <strong>the</strong>se 1143 polylines show interesting tagging behaviour: 594<br />

changed designation once, 293 changed twice, 127 changed three times, 60 changed four<br />

times. The remaining 69 polylines have between 5 and 10 designation changes. Incredibly<br />

three polylines exist with 23, 41, and 73 designation changes.<br />

� There are 548 polygons tagged as ―land use‖ polygons. There is less tag changing amongst<br />

<strong>the</strong>se polygons. Only 40 <strong>of</strong> <strong>the</strong>se polygons experience changes to <strong>the</strong>ir original land use tag.<br />

Figure 1: Using <strong>the</strong> Levenshtein distance and JaroWinkler distance metrics to visualise<br />

changes to name tags <strong>of</strong> spatial objects in OpenStreetMap<br />

4. Conclusions and Fur<strong>the</strong>r Work<br />

This paper has described <strong>the</strong> results <strong>of</strong> an analysis <strong>of</strong> how spatial features are edited and annotated in<br />

OpenStreetMap by extracting <strong>the</strong> entire history <strong>of</strong> all contributions to spatial features in two large<br />

OSM databases. The majority <strong>of</strong> <strong>the</strong> quality analysis <strong>of</strong> OSM reported (such as Haklay (2010);<br />

Mooney et al. (2010)) in <strong>the</strong> literature base <strong>the</strong>ir analysis on <strong>the</strong> current available version <strong>of</strong> <strong>the</strong> OSM<br />

database or a recently downloaded version <strong>of</strong> <strong>the</strong> database. For users <strong>the</strong> problems associated with<br />

annotation <strong>of</strong> spatial features in OpenStreetMap are <strong>of</strong> great importance. Potential users <strong>of</strong> OSM data<br />

will require some measures <strong>of</strong> certainty that <strong>the</strong> current version <strong>of</strong> <strong>the</strong> OSM data has evolved to a<br />

Page | 55


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

stable and agreed-upon representation <strong>of</strong> <strong>the</strong> real-world features modelled in <strong>the</strong> data. The changing<br />

<strong>of</strong> names <strong>of</strong> features (by different contributors), for example, could cause <strong>the</strong>se users to lose<br />

confidence with OSM. An example <strong>of</strong> this is shown in table 2. Section 3. showed that <strong>the</strong>se OSM<br />

databases are continuously evolving and changing. The spatial characteristics and <strong>the</strong> attribute<br />

metadata can change quickly <strong>of</strong>ten within a very short period <strong>of</strong> time. While <strong>the</strong> results <strong>of</strong> only two<br />

regions are presented in this paper <strong>the</strong> analysis is not restricted to <strong>the</strong>se regions. Any OSM database<br />

can be provided as input. The computational-based analysis <strong>of</strong> how tags describing names or highway<br />

designations change is very informative and shows in <strong>the</strong> case <strong>of</strong> some features that <strong>the</strong>re is<br />

disagreement and ambiguity surrounding <strong>the</strong> naming <strong>of</strong> features from local contributors. Local<br />

knowledge or OSM contributor input to why names changed is needed more completely understand<br />

our initial observations made. Future work will pursue a number <strong>of</strong> research questions including: an<br />

analysis <strong>of</strong> how different users change and edit spatial attributes for certain objects; <strong>the</strong> correlation<br />

between <strong>the</strong> number <strong>of</strong> contributors and <strong>the</strong> number <strong>of</strong> changes (locally and globally); and finally to<br />

find and subsequently quantify evidence <strong>of</strong> ―tag wars‖ where contributors constantly disagree about<br />

<strong>the</strong> correct values for tags for certain objects.<br />

5. Bibliography<br />

Auer, S., Lehmann, J. and Hellmann, S. (2009). Linkedgeodata: Adding a spatial dimension to <strong>the</strong><br />

web <strong>of</strong> data, in A. Bernstein, D. R. Karger, T. Heath, L. Feigenbaum, D. Maynard, E. Motta and K.<br />

Thirunarayan (eds), International Semantic Web <strong>Conference</strong>, Vol. 5823 <strong>of</strong> Lecture Notes in<br />

Computer Science, Springer, pp. 731–746.<br />

Bilenko, M., Mooney, R., Cohen, W., Ravikumar, P. and Fienberg, S. (2003). Adaptive name<br />

matching in information integration, Intelligent Systems, IEEE 18(5): 16 – 23.<br />

Ciepluch, B., Mooney, P., Jacob, R. and Winstanley, A. C. (2009). Using openstreetmap to deliver<br />

location-based environmental information in ireland, SIGSPATIAL Special 1: 17–22.<br />

Coast, S. (2010). Openstreetmap: The best map?, OpenGeoData.org - http://opengeodata.org/<br />

openstreetmap-<strong>the</strong>-best-map Last Checked - December 2010.<br />

Comber, A., Fisher, P. and Wadsworth, R. (2005). What is land cover?, Environment and Planning<br />

B: Planning and Design 32(1): <strong>19</strong>9–209.<br />

Forest (2010). A forest feature in <strong>the</strong> openstreetmap database - located in nor<strong>the</strong>rn germany, Spatial<br />

Content in <strong>the</strong> OpenStreetMap database - http://www.openstreetmap.org/browse/way/24015216.<br />

Gupta, M., Li, R., Yin, Z. and Han, J. (2010). Survey on social tagging techniques, SIGKDD Explor.<br />

Newsl. 12: 58–72.<br />

Haklay, M. (2010). How good is volunteered geographical information? a comparative study <strong>of</strong> openstreetmap<br />

and ordnance survey datasets, Environment and Planning B: Planning and Design 37(4):<br />

682– 703.<br />

Haklay, M., A<strong>the</strong>r, A. and Basiouka, S. (2010). How many volunteers does it take to map an area<br />

well?, in M. Haklay, J. Morely and H. Rahemtulla (eds), <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> <strong>GIS</strong> <strong>Research</strong> <strong>UK</strong> 18th<br />

<strong>Annual</strong> <strong>Conference</strong>, University College London, London, England, pp. <strong>19</strong>3–<strong>19</strong>6.<br />

6. Biography<br />

Dr. Peter Mooney is a research fellow at <strong>the</strong> Department <strong>of</strong> Computer Science NUI Maynooth and he<br />

is funded by <strong>the</strong> Irish Environmental Protection Agency STRIVE programme (grant 2008-FS-DM-14-<br />

S4).<br />

Dr. Padraig Corcoran is a lecturer and post-doctoral researcher also at <strong>the</strong> Department <strong>of</strong> Computer<br />

Science NUI Maynooth. Dr. Corcoran is part <strong>of</strong> STRAT-AG which is a Strategic <strong>Research</strong> Cluster<br />

grant (07/SRC/I1168) funded by Science Foundation Ireland under <strong>the</strong> National Development Plan.<br />

The authors gratefully acknowledge this support.<br />

Page | 56


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

Geographical Information Integration from Disparate Sources<br />

Heshan Du 1 , Suchith Anand 1 , Mike Jackson 1 , Didier Leibovici 1 , Jeremy<br />

Morley 1 , Glen Hart 2<br />

1 Centre for Geospatial Science, University <strong>of</strong> Nottingham, <strong>UK</strong><br />

2 Ordnance Survey, <strong>UK</strong><br />

Tel: +44(0)115 846 8411<br />

Email: psydhd@nottingham.ac.uk ; URL: http://www.nottingham.ac.uk/cgs/<br />

ABSTRACT<br />

The paper describes a research project which explored <strong>the</strong> technical issues for integrating<br />

unstructured volunteered geographic information (VGI) with Ordnance Survey data, <strong>the</strong> ultimate<br />

objective being to develop methodologies for change intelligence operations <strong>of</strong> National Mapping<br />

Agencies using VGI. Techniques for geographical information verification from disparate sources are<br />

described. An ontology based methodology which was developed and implemented for geospatial<br />

information linking and merging is also presented. The purpose <strong>of</strong> <strong>the</strong>se techniques is to enrich <strong>the</strong><br />

geometric and metadata information available for mapped objects by linking geospatial information<br />

from different datasets. The developed methodology can deal with geometry inconsistency and<br />

provide more flexibility for geospatial information merging.<br />

KEYWORDS: VGI, ontology, geospatial, inconsistency resolution, data conflation, information<br />

fusion<br />

1. Introduction<br />

The context <strong>of</strong> this paper is <strong>the</strong> progress <strong>of</strong> national and international spatial data infrastructures, such<br />

as <strong>UK</strong> location Programme and INSPIRE, contrasted against crowd-sourced geospatial databases,<br />

such as OpenStreetMap (Anand et al, 2010). Crowd-sourced data sources rely on volunteers to collect<br />

data. Although typically not as structured as most authoritative data, crowd-sourced data may provide<br />

a rich source <strong>of</strong> information which may be more current and which can incorporate interesting userbased<br />

information. Though currently being relatively independent, authoritative and crowd-sourced<br />

communities need to communicate and collaborate to improve <strong>the</strong> overall quality (richness,<br />

consistency, accuracy, and timeliness) <strong>of</strong> geospatial information. It is desirable but challenging to<br />

generate an overview <strong>of</strong> all available information <strong>of</strong> any object, from disparate sources, with differing<br />

conceptual, contextual and topographical representations. Fur<strong>the</strong>rmore, in <strong>the</strong> ever-changing world,<br />

<strong>the</strong>re is an increasing need for <strong>the</strong> representation <strong>of</strong> knowledge <strong>of</strong> objects to be fluent, changing<br />

during its use (Bundy, 2006).<br />

Ontology, in information science, refers to a formal representation <strong>of</strong> knowledge by a set <strong>of</strong> concepts<br />

and <strong>the</strong>ir relationships within a domain. It is hailed as a mechanism to make better use <strong>of</strong> <strong>the</strong> Web, by<br />

<strong>of</strong>fering a shared definition <strong>of</strong> a domain that computers can understand enough to meaningfully<br />

process data automatically. Ontology is expected to play a major role in <strong>the</strong> Semantic Web, which is<br />

to add a level <strong>of</strong> meaning to <strong>the</strong> Web (Wilson, 2004). Geo-ontology, as a sub-concept, refers to <strong>the</strong><br />

formalization <strong>of</strong> geospatial concepts and knowledge (Yang et al, 2006), resulting from analysis and<br />

modelling <strong>of</strong> ontology in geo-spatial applications (Wang et al, 2007). Describing <strong>the</strong> characteristics <strong>of</strong><br />

data and resources and data acquiring modes, a geo-ontology can provide a uniform framework for<br />

data integration, sharing, and updating.<br />

Linked data is about using <strong>the</strong> Web to create typed links between data from disparate sources (Heath<br />

et al., 2009). It uses <strong>the</strong> Resource Description Framework (RDF) to link conceptually related things in<br />

<strong>the</strong> world, resulting in <strong>the</strong> Web <strong>of</strong> Data (Semantic Web). Based on Linked Data concepts,<br />

Data.gov.uk, a part <strong>of</strong> <strong>the</strong> <strong>UK</strong> Government‘s Transparency programme, aims to open up data<br />

collected for <strong>of</strong>ficial purpose for free reuse (data.gov.uk, 2010). These developments make it easier to<br />

Page | 57


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

obtain geospatial data from different sources for fur<strong>the</strong>r analysis, extraction, and reasoning. There is a<br />

need to be able to integrate <strong>the</strong>se apparently disparate sources <strong>of</strong> information and conflate <strong>the</strong>m<br />

geometrically and semantically. We propose to deliver to <strong>the</strong> user different levels <strong>of</strong> information. The<br />

first level quantifies <strong>the</strong> agreement between <strong>the</strong> sources, whilst <strong>the</strong> o<strong>the</strong>r levels address specificities <strong>of</strong><br />

datasets. The methodology to be adopted may be quite data specific but whilst using OpenStreetMap<br />

and Ordnance Survey data as a use case we strived towards a generic approach. We envisage that <strong>the</strong><br />

potential users for this will be NMAs, government agencies etc.<br />

2. Methodology<br />

To provide high quality geospatial information to users by taking <strong>the</strong> best <strong>of</strong> available information,<br />

this research project focuses on two main problems:<br />

1. how to link information from disparate sources<br />

2. how to merge <strong>the</strong> linked information.<br />

For <strong>the</strong> first problem, disparate geospatial information sources may employ different conceptual<br />

representations. In o<strong>the</strong>r words, different expressions may have <strong>the</strong> same meaning or <strong>the</strong> same<br />

expression may refer to different concepts. The data structures for Ordnance Survey and Open Street<br />

Map are very different. In addition, <strong>the</strong> coordinates delimiting an object may not be exactly <strong>the</strong> same<br />

in <strong>the</strong> crowd sourced and authoritative datasets. Hence it is important to develop methodologies and<br />

techniques for bridging <strong>the</strong>se disparate worlds. In doing so, some translation and tolerance is required.<br />

When merging linked information, it is also necessary to ensure that <strong>the</strong> merged information is<br />

consistent. Consistency means <strong>the</strong>re are no logical conflicts. Ideally, <strong>the</strong> geometries <strong>of</strong> <strong>the</strong> same<br />

object in disparate datasets should be <strong>the</strong> same. If <strong>the</strong>re is any inconsistency, some selection or<br />

amendments are <strong>of</strong>ten required based on <strong>the</strong> accuracy and timeliness <strong>of</strong> information.<br />

The methodology used is based on an ontology, which refers to a logical conceptual framework for<br />

<strong>the</strong> representation <strong>of</strong> information in a particular domain. OWL, a W3C standard web ontology<br />

language is used to represent ontology (www.w3.org, 2010). Pellet, a <strong>the</strong>orem prover and OWL<br />

reasoner, is used to discovery inconsistency (clarkparsia.com, 2010). To solve <strong>the</strong> above problems for<br />

road networks, a graph model is employed. Within this model, road networks are simplified as a<br />

graph, made <strong>of</strong> edges and vertexes. Each road is seen as an individual <strong>of</strong> base class ―edge‖, while<br />

each end point <strong>of</strong> a road is represented as an individual <strong>of</strong> base class ―vertex‖. Information, including<br />

attributes and geometry, about each individual is stored as data properties, and relationships between<br />

different individuals are represented as object properties.<br />

3. The algorithm and s<strong>of</strong>tware implementation<br />

When linking information about an object, linked data principles (Heath et al., 2009) are applied by<br />

ensuring <strong>the</strong> IRI <strong>of</strong> <strong>the</strong> same object in disparate sources are <strong>the</strong> same. To merge geometries from<br />

disparate sources <strong>of</strong> <strong>the</strong> same object, an inconsistency resolution algorithm (Figure 1) is designed.<br />

Ordnance Survey Integrated Transport Network (ITN) data and OpenStreetMap (OSM) road data for<br />

Portsmouth, <strong>UK</strong> were used as test data for this case study.<br />

The prototype s<strong>of</strong>tware (Figure 2) was implemented employing <strong>the</strong> methodology and algorithm above<br />

using Java. It imports a Java API developed in JUMP. JUMP Unified Mapping Platform is a Java<br />

open source application for viewing, editing, and processing geospatial data<br />

(www.vividsolutions.com, 2006). It allows users to import OWL or Shapefile data (it will translate<br />

Shapefile data into OWL data) and visualizes OWL data as a graph. When a road name is searched<br />

information both about attributes and geometry will be shown. In <strong>the</strong> current version, <strong>the</strong> features are<br />

selected by road name but we are working on adding new functionality for applying this for <strong>the</strong> whole<br />

dataset at once.<br />

For instance, Figure 2 shows <strong>the</strong> state <strong>of</strong> <strong>the</strong> s<strong>of</strong>tware when Adstone Lane is searched, importing<br />

Ordnance Survey ITN data and OpenStreetMap data <strong>of</strong> Portsmouth. Since <strong>the</strong> Ordnance Survey data<br />

is used as <strong>the</strong> main reference, <strong>the</strong> OpenStreetMaps layers are made invisible to give users a clean<br />

view. In addition, <strong>the</strong> s<strong>of</strong>tware allows users to merge information from different datasets, and store it<br />

Page | 58


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

in an OWL ontology. If <strong>the</strong> newly generated ontology is inconsistent (Inconsistency is defined quite<br />

strict currently, and will arise when <strong>the</strong> geometries <strong>of</strong> a same road in different data sets are different.),<br />

<strong>the</strong> geometry amendments algorithm (Figure 1) will be employed to deal with it. Users can specify<br />

different degrees <strong>of</strong> belief, an integer between 0 to100 inclusive, to <strong>the</strong> first dataset. If this degree <strong>of</strong><br />

belief equals 100, <strong>the</strong> geometry from <strong>the</strong> first dataset will be <strong>the</strong> output geometry. If it equals 0, <strong>the</strong><br />

output geometry will be that from <strong>the</strong> second dataset. O<strong>the</strong>rwise, this algorithm will output a<br />

geometry going through some intermediate coordinates calculated according to weight specified.<br />

Input: target, // searched edge or vertex<br />

Algorithm:<br />

fstwt, sndwt // indicating <strong>the</strong> preference or degree <strong>of</strong> belief to <strong>the</strong>se two datasets<br />

fuzzy //fuzzy tolerance applied<br />

// select <strong>the</strong> basic geometry from <strong>the</strong>se two geometries available, depending on input wt.<br />

GeometryBase gb�select(fstwt, sndwt)<br />

Coordinate[] coords� gb.getCoordinates()<br />

For each coordinate in coords<br />

If coordinate.hasTwoGeometry(fuzzy)<br />

// exist in both dataset given a tolerance<br />

coordinate � (firstcoord*fstwt+sndcoord*sndwt)/(fstwt+sndwt)<br />

gb.update(coordinate)<br />

// angle checking to ensure <strong>the</strong> amendments do not generate strange shapes<br />

gb.smooth()<br />

return gb<br />

Figure 1: Pseudocode for Geometry Amendments Algorithm<br />

Figure 2. Example result shown for doing a road name search<br />

Page | 59


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

For example, within <strong>the</strong> Line Viewer in Figure 3, <strong>the</strong> green line shows <strong>the</strong> geometry from <strong>the</strong> first<br />

dataset, while <strong>the</strong> purple line shows <strong>the</strong> geometry from <strong>the</strong> second dataset. The three blue lines, with<br />

red points (matched points) on <strong>the</strong>m, show <strong>the</strong> newly generated geometries given different degrees <strong>of</strong><br />

belief, 80, 50, and 20. This s<strong>of</strong>tware allows users to try various degrees <strong>of</strong> belief and select <strong>the</strong> one<br />

<strong>the</strong>y have most confidence in to fix geometry inconsistency. For example, if degree <strong>of</strong> belief is<br />

specified to 80, <strong>the</strong> red line shown in <strong>the</strong> Line Viewer in Figure 4 will be <strong>the</strong> final output geometry.<br />

The result message now shows <strong>the</strong> ontology is consistent (Figure 4).<br />

Figure 3. Inconsistency Resolution: Trial<br />

Figure 4. Inconsistency Resolution: Fix<br />

Page | 60


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

4. Conclusions and Future work<br />

This paper explores <strong>the</strong> development <strong>of</strong> techniques for geographical information fusion from<br />

disparate sources. Only <strong>the</strong> first step <strong>of</strong> looking at data agreement or disparity and its weighting<br />

according to user confidence for conflation has been described. The results are promising but more<br />

work needs to be done in refining <strong>the</strong> process <strong>of</strong> linking information and in inconsistency resolution.<br />

National Mapping Agencies may benefit immensely from such developments but research is needed<br />

to understand how to tap into this huge potential opportunity and to obtain a consistent, quality and<br />

verifiable product from <strong>the</strong> data so acquired within <strong>the</strong> terms <strong>of</strong> use <strong>of</strong> <strong>the</strong> crowd-sourced data. This<br />

can <strong>the</strong>n be used to develop different models, for example, for change intelligence operations. There<br />

will be scope for deriving products based on <strong>the</strong> volunteered and vernacular geographic data collected<br />

from <strong>the</strong> crowd sourced communities. Future work will concentrate on developing more robust and<br />

sound strategies for inconsistency resolution to solve different real life problems<br />

5. Acknowledgements<br />

The authors express thanks for useful advice from Natasha Alechina.<br />

The authors express thanks for <strong>the</strong> Ordnance Survey, <strong>UK</strong> and OSM for <strong>the</strong> data used in this work. All<br />

figures in this text using OS data are ©Crown Copyright/database right 2010. An Ordnance<br />

Survey/EDINA supplied service.<br />

6. References<br />

Anand, S., Morley, J., Jiang, W., Du, H., Hart, G. and Jackson, M. (2010). When worlds collide:<br />

Combining Ordnance Survey and OSM data. AGI Geocommunity <strong>Conference</strong><br />

Association for Geographic Information (2010). AGI Foresight Study: The <strong>UK</strong> Geospatial Industry in<br />

2015. [Online] Available at:<br />

http://www.agi.org.uk/storage/AGI%20Foresight%20Study%20Summary%20Report%201.1.pdf .<br />

Bizer, C., Heath, T. and Berners-Lee, T., (2009). Linked Data-The Story So Far. [online] Available<br />

at: http://tomheath.com/papers/bizer-heath-berners-lee-ijswis-linked-data.pdf .<br />

Bundy, A., & McNeill, F. (2006). Representation as a fluent: An AI challenge for <strong>the</strong> next half<br />

century. IEEE Intelligent Systems.<br />

Clarkparsia (2010). Pellet: OWL 2 Reasoner for Java [online] Available at:<br />

http://clarkparsia.com/pellet [6 Oct 2010].<br />

Du H., Jiang W., Anand S., Morley J., Hart G., Jackson M. (2010). ‗Ontology Based Approach for<br />

Geospatial Data Integration‘. In International Cartography <strong>Conference</strong> 2011, Paris. (submitted).<br />

Vividsolutions (2006). JUMP Unified Mapping Platform [online] Available at:<br />

[8 Oct 2010]<br />

Wang, Y., et al (2007) Geo-ontology Design and its Logic Reasoning, Geoinformatics 2007:<br />

Geospatial Information Science.<br />

Wilson, R (2004) The Role <strong>of</strong> Ontologies in Teaching and Learning. TechWatch Reports, Citeseer,<br />

2004<br />

W3C (2004). OWL Web Ontology Language Overview [online] Available at:<br />

http://www.w3.org/TR/owl-features [6 Oct 2010]<br />

Yang, K., et al (2006). The <strong>Research</strong> and Practice <strong>of</strong> Geo-Ontology Construction. <strong>Proceedings</strong> <strong>of</strong><br />

International Symposium on Spatio-temporal Modeling, Spatial Reasoning, Analysis, Data Mining<br />

and Data Fusion.<br />

7. Biography<br />

Heshan Du is a third year undergraduate at University <strong>of</strong> Nottingham. She is an Internship student<br />

at <strong>the</strong> Open Source Geospatial Lab at <strong>the</strong> Centre for Geospatial Science.<br />

Suchith Anand is Ordnance Survey <strong>Research</strong> Fellow at <strong>the</strong> Centre for Geospatial Science, University<br />

Page | 61


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

<strong>of</strong> Nottingham. His research interests are in automated generalization, optimization and open source<br />

geospatial technologies.<br />

Glen Hart is Principal <strong>Research</strong> Scientist <strong>of</strong> Ordnance Survey, <strong>UK</strong>.<br />

Jeremy Morley is Deputy Director <strong>of</strong> <strong>the</strong> Centre for Geospatial Science, University <strong>of</strong> Nottingham.<br />

Didier Leibovici is research Fellow in geospatial modelling and analysis at <strong>the</strong> Centre for Geospatial<br />

Science, University <strong>of</strong> Nottingham<br />

Mike Jackson is <strong>the</strong> Director <strong>of</strong> <strong>the</strong> Centre for Geospatial Science, University <strong>of</strong> Nottingham.<br />

Page | 62


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

Projecting obesity in small area populations<br />

Belinda Wu 1 , Mark Birkin 1<br />

1 School <strong>of</strong> Geography, University <strong>of</strong> Leeds<br />

Leeds LS2 9JT, <strong>UK</strong><br />

Tel. (0044) 113 34 33300 Fax (0044) 113 34 33308<br />

B. Wu@Leeds.ac.uk<br />

ABSTRACT<br />

Public health reflects complex interactions between population characteristics and <strong>the</strong> environment.<br />

The understanding <strong>of</strong> place effects on health helps explain <strong>the</strong> cause, facilitate prevention and/or<br />

treatment, enable <strong>the</strong> design and implementation <strong>of</strong> targeted interventions and policies. Obesity is a<br />

fast growing problem in <strong>the</strong> <strong>UK</strong> and worldwide. Moses is a dynamic spatial MicroSimulation Model<br />

(MSM) that simulates <strong>UK</strong> individuals through discrete demographic processes within a fine spatial<br />

scale for 30 years from 2001 to 2031. Within <strong>the</strong> health process, three ―what if‖ scenarios <strong>of</strong> obesity<br />

have been projected to facilitate public health planning and various explorations.<br />

1. Introduction<br />

KEYWORDS: MSM, dynamic, spatial, obesity, decision support<br />

Public health reflects complex interactions between population characteristics and <strong>the</strong> environment.<br />

Although genetics can predispose certain health conditions, economic status, access to health care and<br />

life styles also have important impact on people‘s health. In fact, Wilkinson and Marmot (2003)<br />

pointed out that environmental factors are <strong>the</strong> most common causes <strong>of</strong> <strong>the</strong> ill health and <strong>the</strong>ir changes<br />

have a more immediate impact than <strong>the</strong> genetic changes, as <strong>the</strong>y reflect <strong>the</strong> changes in <strong>the</strong> way people<br />

live. Public policies <strong>the</strong>refore play an important role in shaping <strong>the</strong> social environment to promote<br />

public health (Marmot, 2005). Compared to <strong>the</strong> traditional medical approach, promoting people‘s<br />

well-being through studies about who <strong>the</strong>y are, where <strong>the</strong>y live, and how <strong>the</strong>y live within a spatial<br />

framework provides a more holistic approach to public health. Place is clearly important when<br />

considering public health and social inequalities, eg. access to pubic health services. The<br />

understanding <strong>of</strong> place effects on individual health can also help to explain <strong>the</strong> cause <strong>of</strong> diseases, as<br />

well as providing additional information to facilitate preventions and/or treatments. Such a targeted<br />

method is more effective (Lang and Rayner, 2007).<br />

Obesity has become a fast growing problem in <strong>the</strong> <strong>UK</strong> and worldwide. The health, economic and<br />

social issues associated with obesity are costly; it is also closely linked to many diseases, including<br />

cancers, cardiovascular and metabolic diseases (Stamatakis, 2006). Reducing obesity and health<br />

inequalities are at <strong>the</strong> centre <strong>of</strong> <strong>the</strong> <strong>UK</strong> government‘s health policy. The government‘s health White<br />

Paper (DoH, 2004) recognises that policy making needs to move away from considering disease<br />

groupings in isolation, towards a population approach that considers <strong>the</strong> determinants <strong>of</strong> health<br />

problems. WHO (2010) has now made it a regional priority goal to address obesity through healthy<br />

environments, physical activity and healthy diet. Although creating supportive environments have<br />

been at <strong>the</strong> centre <strong>of</strong> health promotion, strategies to address obesity have been primarily focused on<br />

behavioural, educational and medical interventions to encourage individuals to eat more healthily and<br />

exercise more. This individual focus can overlook <strong>the</strong> complex social contexts and become an<br />

impediment to successful public health policy development (Pearce and Witten, 2010).<br />

Some previous obesity studies use Multi Level Models (MLMs), in <strong>the</strong> attempt to capture<br />

environmental impact in modelling <strong>the</strong> health outcomes/behaviour in areas (Moon et al. 2007, Pearce<br />

and Witten, 2010, Twigg and Moon, 2002). Such MLMs normally consider <strong>the</strong> proportion <strong>of</strong> cases<br />

Page | 63


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

explained by various individual factors (eg. age and sex), <strong>the</strong>n introduce area characteristics (eg<br />

deprivation) to explain <strong>the</strong> remaining cases. Based on this, <strong>the</strong> model‘s parameter estimates can be<br />

applied to <strong>the</strong> individuals and areas to create new prevalence estimates (Twigg and Moon, 2002).<br />

However, such estimates remain as <strong>the</strong> zonal average that cannot truly reflect <strong>the</strong> individual<br />

characteristics and different zoning system may produce different spatial patterns. As we shall see, an<br />

individual level simulation approach also provides a powerful means for projecting <strong>the</strong> dynamics <strong>of</strong><br />

healthcare needs within a population.<br />

2. Moses model<br />

A spatial Micro Simulation Model (MSM) projects <strong>the</strong> population with individual characteristics in a<br />

local context. Although certain demographic characteristics are <strong>of</strong>ten found to persist in areas, it is<br />

difficult to determine <strong>the</strong> exact cause and model <strong>the</strong> process, as it may be <strong>the</strong> outcome <strong>of</strong> complex<br />

interactions between many factors. Location in a spatial MSM can provide a useful proxy variable for<br />

<strong>the</strong> simultaneous operation <strong>of</strong> many variables at both individual and area level, such as socioeconomic,<br />

ethnic, life style and (physical/social) environment variables, without getting into too much<br />

<strong>the</strong>oretical and practical difficulty. A spatial MSM allows us to link different data based on a common<br />

variable, providing small area populations with a rich set <strong>of</strong> attributes in previously unavailable<br />

combinations. A dynamic spatial MSM updates each attribute <strong>of</strong> each individual at each simulation<br />

step, using location specific transitional probabilities. Therefore <strong>the</strong> simulation is based on distinctive<br />

individual characteristics, not <strong>the</strong> zonal averages. As <strong>the</strong> simulation result is <strong>the</strong> basis for <strong>the</strong> next<br />

step, <strong>the</strong> impact <strong>of</strong> previous changes is captured in <strong>the</strong> model, including <strong>the</strong> impact <strong>of</strong> government<br />

interventions on <strong>the</strong> simulated populations at <strong>the</strong> individual/household level. The individual based<br />

projections also help to overcome <strong>the</strong> problem <strong>of</strong> different geographical patterns produced by<br />

different zoning systems, as <strong>the</strong> re-aggregation is based on individuals/households not zonal<br />

boundaries. Indeed <strong>the</strong> MSM output can be aggregated to any geographical scale and analysed at<br />

multi levels.<br />

As obesity is <strong>the</strong> result <strong>of</strong> complex interactions between individual and environmental, a dynamic<br />

spatial MSM, Moses, has been used to provide <strong>the</strong> capacity to model <strong>the</strong> two dimensions<br />

simultaneously. Moses simulates individuals in households through discrete demographic transitions<br />

at a fine spatial scale <strong>of</strong> ward from year 2001 to 2031. The scale <strong>of</strong> ward, similar to MSOA, is chosen<br />

instead <strong>of</strong> OA/LSOA to reduce <strong>the</strong> impact from <strong>the</strong> Small Cell Adjustment Method (SCAM) used in<br />

<strong>the</strong> Census 2001, where <strong>the</strong> information <strong>of</strong> a small cell population is adjusted to provide<br />

confidentiality protection (ONS, 2001). Individuals in <strong>the</strong> MSM are simulated through demographic<br />

processes <strong>of</strong> Ageing, Mortality, Fertility, Health, Household Formation and Migration. It models <strong>the</strong><br />

demographic lifecycle at an annual interval and all probabilities are spatially disaggregated at <strong>the</strong><br />

ward level so that <strong>the</strong> impact <strong>of</strong> individual and local characteristics can be observed in <strong>the</strong> model<br />

results, even at <strong>the</strong> aggregated level.<br />

Within <strong>the</strong> Health module, <strong>the</strong> projected individuals are simulated with obesity probabilities<br />

calculated on <strong>the</strong> basis <strong>of</strong> obesity data from <strong>the</strong> Health Survey for England <strong>19</strong>93-2008 (NHS, 2009).<br />

In <strong>the</strong> following section we will discuss <strong>the</strong> findings from <strong>the</strong> model results, using <strong>the</strong> changes in<br />

Leeds ward population as an example in three ―what if‖ scenarios where <strong>the</strong> obesity patterns continue<br />

as current; with significant improvement (eg. due to suitable interventions) or deterioration (by<br />

applying 1% weight annually).<br />

3. Result analysis<br />

Obesity changes have been projected for populations in Leeds wards and we can see <strong>the</strong> changes in<br />

<strong>the</strong> spatial pattern clearly in <strong>the</strong> maps below. In Figure 1, <strong>the</strong> map suggests that at <strong>the</strong> beginning <strong>of</strong> <strong>the</strong><br />

simulation, obesity is a more serious problem in <strong>the</strong> suburban areas, while <strong>the</strong> city centre is much<br />

better. This may be due to <strong>the</strong> demographic composition: due to <strong>the</strong> migration impact, <strong>the</strong> population<br />

at <strong>the</strong> city centre is much younger than <strong>the</strong> suburban areas. We can also see that <strong>the</strong>re are small<br />

differences between <strong>the</strong> four types <strong>of</strong> areas (about 1000 counts).<br />

Page | 64


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

However, this has changed after 30 years‘ simulation. In Figure 2, all three sets <strong>of</strong> scenario results in<br />

2030 suggest that <strong>the</strong> obesity has become a more serious problem in city centre areas, compared to<br />

that in 2001. In fact, <strong>the</strong> high degree <strong>of</strong> obesity impact seems to persist in <strong>the</strong> city centre areas and <strong>the</strong><br />

nor<strong>the</strong>ast <strong>of</strong> <strong>the</strong> city in all scenarios. The areas with <strong>the</strong> least obesity remain to be <strong>the</strong> areas that are<br />

just outside / surrounding <strong>the</strong> city centre areas (in light grey). This is consistent both in 2001 and<br />

2030, as well as in all scenarios. The maps also reveal a clear distinction between <strong>the</strong> east and <strong>the</strong><br />

west <strong>of</strong> <strong>the</strong> city: <strong>the</strong> west <strong>of</strong> <strong>the</strong> city seems to have a much worse obesity issue than <strong>the</strong> east. In <strong>the</strong><br />

2030 results, <strong>the</strong> gaps between <strong>the</strong> four types <strong>of</strong> areas have increased to 2000-4000 counts. All 3<br />

scenarios suggest <strong>the</strong> same trend, although variances have been found in small areas.<br />

Figure 1. Obesity in Leeds wards 2001<br />

Although with a substantial reduction in obesity counts, <strong>the</strong> improvement scenario map (right) in<br />

Figure 2 indicates a similar pattern as in <strong>the</strong> scenario where <strong>the</strong> current obesity trends continue (left).<br />

The two maps look almost identical, except in four wards in <strong>the</strong> city centre. The right map suggests a<br />

slightly more concentrated obesity environment in <strong>the</strong> city centre than <strong>the</strong> left. This may indicate that<br />

it is harder to improve <strong>the</strong> obesity in <strong>the</strong> city centre areas. The deterioration scenario (middle),<br />

however, reveals a different pattern than <strong>the</strong> two. Although with a substantial increase in obesity<br />

counts in small areas (up to 15460 counts in one ward), it indicates that in <strong>the</strong> suburban areas, obesity<br />

deterioration has much less impact than in o<strong>the</strong>r areas. Particularly it seems to have a significantly<br />

less impact in <strong>the</strong> east and <strong>the</strong> north <strong>of</strong> <strong>the</strong> city, compared to <strong>the</strong> o<strong>the</strong>r two scenarios (Figure 2).<br />

However, compared to <strong>the</strong> simulation results in 2001 (Figure 1), all <strong>the</strong> east and north <strong>of</strong> <strong>the</strong> city are<br />

not in <strong>the</strong> top band <strong>of</strong> <strong>the</strong> obesity, except <strong>the</strong> two wards in <strong>the</strong> nor<strong>the</strong>ast.<br />

Page | 65


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 1b: Open Source and Web 2.0<br />

Figure 2. Obesity in Leeds wards 2030: current, deterioration and improvement scenarios<br />

Page | 66


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

4. Conclusion<br />

In this paper, we introduced a dynamic spatial microsimulation model, Moses. It dynamically<br />

simulates <strong>the</strong> modelled population in small areas through important demographic changes. The<br />

individual and local characteristics are captured and impacts from previous simulation steps are built<br />

into <strong>the</strong> next. Such holistic modelling approach produces a better reflection <strong>of</strong> <strong>the</strong> studied populations<br />

that allows us to better understand <strong>the</strong> population changes and provides <strong>the</strong> basis for various strategic<br />

planning and policy making. The projection <strong>of</strong> <strong>the</strong> obesity changes in small areas using Leeds<br />

scenarios demonstrates that Moses can provide a new approach to allow <strong>the</strong> promotion <strong>of</strong> <strong>the</strong> well<br />

being <strong>of</strong> people, through studies about who <strong>the</strong>y are, where <strong>the</strong>y live, and how <strong>the</strong>y live within a<br />

spatial framework. Spatial variances have been found in small areas in <strong>the</strong> projections. Such<br />

information is important to understand population trends at a more aggregate level. In <strong>the</strong> obesity<br />

scenarios, we have seen such changes clearly.<br />

At <strong>the</strong> beginning <strong>of</strong> <strong>the</strong> study, <strong>the</strong> suburban areas have a more serious obesity issue. However, this<br />

has changed after 30 years‘ simulation and city centre have become <strong>the</strong> hotspots. In fact, <strong>the</strong> city<br />

centre areas and <strong>the</strong> nor<strong>the</strong>ast areas <strong>of</strong> <strong>the</strong> city have consistently has more hotspots in all scenarios.<br />

On <strong>the</strong> o<strong>the</strong>r hand, <strong>the</strong> areas surrounding <strong>the</strong> city centre areas have been consistently <strong>the</strong> better <strong>of</strong>f<br />

areas in all scenarios since <strong>the</strong> beginning <strong>of</strong> <strong>the</strong> simulation. However, <strong>the</strong> rest <strong>of</strong> north and all east<br />

suburban areas seem to improve more than <strong>the</strong> rest <strong>of</strong> <strong>the</strong> city. We are particularly encouraged by<br />

finding that obesity patterns persist in <strong>the</strong> city centre and northwest areas. Such projections reveal<br />

similar patterns as in previous obesity studies in Leeds. Identifying obesogenic environment in Leeds,<br />

Edwards and Clarke (2009) pointed out that poverty has an important impact on obesity in Leeds:<br />

―<strong>the</strong>re are higher rates in <strong>the</strong> central, deprived, areas <strong>of</strong> Leeds, with some isolated cases in <strong>the</strong> north <strong>of</strong><br />

Leeds‖. The projected obesity hotspots in <strong>the</strong> northwest are also consistent with <strong>the</strong>ir identification <strong>of</strong><br />

areas with ―high levels <strong>of</strong> poor diet and low activity amongst children … in more affluent and rural<br />

parts <strong>of</strong> Leeds‖. Although <strong>the</strong>y use a determinist model with a focus on <strong>the</strong> child obesity and use<br />

specialised obesity datasets in <strong>the</strong>ir model, such consistent patterns have been found in <strong>the</strong> projections<br />

produced by Moses.<br />

Like any MSM, Moses relies on good microdata to produce robust projections. To address this issue<br />

and enable more behaviour modelling, a hybrid modelling approach has been adopted to bring <strong>the</strong><br />

Agent Based Model (ABM) insight to streng<strong>the</strong>n such areas <strong>of</strong> <strong>the</strong> model (Wu et al. 2008). The<br />

performance <strong>of</strong> <strong>the</strong> models is also constrained by uncertainties about future changes in lifestyle and<br />

behaviour which may determine health status, as well as variant assumptions about migration and<br />

demographic change. Fur<strong>the</strong>r refinements to <strong>the</strong> simulation outputs could potentially be achieved by<br />

an ensemble modelling approach using methods already introduced in <strong>the</strong> context <strong>of</strong> climate<br />

simulations (Murphy, 2004).<br />

Moses as a population based model can be used for various strategic planning applications where<br />

population changes are important. Public health is one <strong>of</strong> <strong>the</strong> potential areas. There are three main<br />

<strong>the</strong>mes in <strong>the</strong> medical geography: disease ecology, health care delivery and environment and health.<br />

The obesity application model <strong>of</strong> <strong>the</strong> Moses demonstrates <strong>the</strong> potential <strong>of</strong> this model in such areas: it<br />

can be used to study how a disease distributed spatially over time, identify <strong>the</strong> hotspots and design <strong>the</strong><br />

health care provision accordingly. It can also be used explore <strong>the</strong> relationship between <strong>the</strong> people‘s<br />

health and where <strong>the</strong>y live. Finally as Moses projects <strong>the</strong> studied populations into <strong>the</strong> future from<br />

2001 to 2031, <strong>the</strong> projection results are particularly useful in providing <strong>the</strong> groundwork for various<br />

explorations or facilitating medium and longer term strategic planning.<br />

5. References<br />

Department <strong>of</strong> Health (DoH) (2004) Choosing Health. Making Healthy Choices Easier. London:<br />

Department <strong>of</strong> Health<br />

Edwards, K. and Clarke, G. (2009) The design and validation <strong>of</strong> a spatial microsimulation model <strong>of</strong><br />

obesogenic environments for children in Leeds, <strong>UK</strong>: SimObesity, Social Science & Medicine, 69:<br />

1127–1134<br />

Page | 67


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Lang, T. and Rayner, G. (2007) Overcoming policy cacophony on obesity: an ecological public health<br />

framework for policymakers. Obesity reviews, 8 (1): 165-181<br />

Moon, G., Quarendon, G., Barnard, S., Twigg, L. and Blyth, B. (2007) Fact nation: deciphering <strong>the</strong><br />

distinctive geographies <strong>of</strong> obesity in England. Social science and medicine, 65:25-31<br />

Murphy, J., Sexton, D., Barnett, D, Jones, G, Webb, M, Collins, M. and Stainforth, D. (2004)<br />

Quantification <strong>of</strong> modelling uncertainties in a large ensemble <strong>of</strong> climate change simulations, Nature,<br />

430, 768-772.<br />

National Health Service (NHS) (2009) Health Survey for England – 2008 Trend Tables at<br />

www.ic.nhs.uk/pubs/hse08trends (accessed 12/12/2010).<br />

ONS (2001) Census 2001 Disclosure Protection Measures,<br />

http://www.statistics.gov.uk/census2001/discloseprotect.asp, last access 22/02/2011<br />

Pearce, J. and Witten, K. (2010) (ed.) Geographies <strong>of</strong> obesity: environmental understandings <strong>of</strong> <strong>the</strong><br />

obesity epidemic, Aldershot: Ashgate, ISBN: 9780754676<strong>19</strong>5<br />

Stamatakis, E. (2006) Chapter 5: Obesity, eating and physical activity, Focus on Health 2006,<br />

http://www.statistics.gov.uk/downloads/<strong>the</strong>me_compendia/foh2005/05_Obesity_Eating_PhysicalActi<br />

vity.pdf (accessed 12/12/2010)<br />

Swinburn, B. and Egger, G. (2002) Preventive strategies against weight gain and obesity. Obesity<br />

Reviews, 3: 289-301<br />

Twigg, L. and Moon, G. 2002 Predicting small-area health-related behaviour: a comparison <strong>of</strong><br />

multilevel syn<strong>the</strong>tic estimation and local survey data. Social science and medicine, 50:1109-20.<br />

Wilkinson, R. and Marmot, M. (2003) The Solid Facts, Copenhagen: World Health Organisation.<br />

WHO (2010) Parma Declaration on Environment and Health, Parma, Italy, 10–12 March 2010<br />

http://www.euro.who.int/__data/assets/pdf_file/0011/78608/E93618.pdf (accessed 22/02/2011)<br />

Wu, B. M., Birkin, M. H. and Rees, P. H. (2008) A spatial microsimulation model with student<br />

agents, Journal <strong>of</strong> Computers, Environment and Urban Systems (32) pp. 440-453 DOI information:<br />

10.1016/j.compenvurbsys. 2008.09.013<br />

8. Biography<br />

Belinda Wu is a research <strong>of</strong>ficer in School <strong>of</strong> Geography, University <strong>of</strong> Leeds. Main research<br />

interests include: microsimulation, agent based model, complex social systems, decision support<br />

systems and <strong>GIS</strong>. Recent research projects: Genesis (http://www.genesis.ncess.ac.uk/), Moses<br />

(http://www.ncess.ac.uk/nodes/moses/). Email: B.Wu@Leeds.ac.uk<br />

Mark Birkin is a Pr<strong>of</strong>essor <strong>of</strong> Spatial Analysis and Policy in <strong>the</strong> School <strong>of</strong> Geography, University <strong>of</strong><br />

Leeds, with research interests in spatial analysis, social simulation, ma<strong>the</strong>matical modelling and<br />

planning support systems. Email: M.H.Birkin@leeds.ac.uk.<br />

Page | 68


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Participatory Health Surveys Using Ubiquitous Computing:<br />

Gastrointestinal illnesses application case study<br />

Natalie Adams 1 , Chaoyu Ye 2 , Wenchao Jiang 2 , Sergiusz<br />

Pawlowicz, Suchith Anand 2 , Didier G. Leibovici 2, and Mike<br />

Jackson 2<br />

1 Department <strong>of</strong> Health Science, University <strong>of</strong> Nottingham, U.K.<br />

Email, Natalie.Adams@nottingham.ac.uk<br />

2 Centre for Geospatial Science, University <strong>of</strong> Nottingham, U.K.<br />

ABSTRACT<br />

Participatory online health surveys are widespread in <strong>the</strong> United States but very few examples <strong>of</strong> such<br />

surveys exist in <strong>the</strong> <strong>UK</strong>. The development <strong>of</strong> <strong>the</strong> health survey for mobile phone application could<br />

enable <strong>the</strong> users to have instant and up-to-date information regarding <strong>the</strong>ir risk status based on <strong>the</strong>ir<br />

current locations. This can provide better healthcare services by alleviating <strong>the</strong> pressure on primary<br />

health workers but also better diagnostic assessment as symptoms can be recorded as <strong>the</strong>y happen to<br />

<strong>the</strong> user. An interoperable framework allowing wide availability but also seamless re-use <strong>of</strong> <strong>the</strong><br />

application service and <strong>the</strong> data generated a presented concerning gastrointestinal disease in<br />

Nottinghamshire.<br />

KEYWORDS: VGI, participatory <strong>GIS</strong>, public health, gastrointestinal illness, location-based services<br />

1. Introduction<br />

The aim <strong>of</strong> <strong>the</strong> study is to assess <strong>the</strong> relevance <strong>of</strong> participatory health surveys in developing public<br />

understanding <strong>of</strong> a disease and to alleviate some pressures on <strong>the</strong> public healthcare system. Can<br />

location-based services support public health through dynamically mapping <strong>the</strong> prevalence and<br />

incidence <strong>of</strong> a contagious disease with provision <strong>of</strong> relevant information to <strong>the</strong> user and susceptible<br />

members <strong>of</strong> <strong>the</strong> community?<br />

Contagious diseases with a syndrome which may be caused by few focal epidemic points such as<br />

Gastrointestinal (GI) illness can beneficiate <strong>the</strong> most from this setting. GI illnesses are primarily<br />

spread via <strong>the</strong> faecal-oral route and are <strong>of</strong>ten caused by contaminated food or water. Norovirus is <strong>the</strong><br />

most common <strong>of</strong> <strong>the</strong> GI viruses (NHS online, 2010a). Rotavirus is <strong>the</strong> leading cause <strong>of</strong> GI infection in<br />

children (NHS online, 2010b). The Health Protection Agency (HPA) already has a comprehensive<br />

disease surveillance system for <strong>the</strong> U.K. involving observation, local health protection <strong>of</strong>ficers and<br />

<strong>the</strong> use <strong>of</strong> a hierarchical system for disease reporting. A trawling health survey, usually via telephone,<br />

is conducted when a significant outbreak <strong>of</strong> unknown source is identified. This is <strong>the</strong>n used to narrow<br />

down <strong>the</strong> potential outbreak sources and attempt to find a direct link between cases. However, <strong>the</strong><br />

HPA does not make particular use <strong>of</strong> <strong>GIS</strong> or online mapping to provide information for <strong>the</strong>mselves,<br />

<strong>the</strong>ir partners or indeed, <strong>the</strong> general public. Outbreak detection for GI can be <strong>of</strong> particular relevance<br />

in <strong>the</strong> case <strong>of</strong> dining-out, as, in a study by Green et al (2004), 22% <strong>of</strong> those infected with a GI illness<br />

attributed it to <strong>the</strong>ir latest eating-out experience.<br />

In order to answer <strong>the</strong> above question, it is proposed to enable <strong>the</strong> general public to use a simplistic,<br />

personalised online ‗symptom checker‘ application to evaluate and record <strong>the</strong>ir symptoms as <strong>the</strong>y<br />

occur, in detail and help guide <strong>the</strong>m through <strong>the</strong> self-care guidelines established for GI illness by <strong>the</strong><br />

NHS or to enable <strong>the</strong>m to take appropriate action, such as seeking medical attention, if <strong>the</strong> symptoms<br />

warrant this. Using this pre-diagnostic check combined with mapping information about <strong>the</strong> current<br />

status <strong>of</strong> any current epidemic history in <strong>the</strong> area, a refined feedback can be delivered to <strong>the</strong> user,<br />

Page | 69


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

which can also be used directly by <strong>the</strong> GP <strong>the</strong> user might consult later on.<br />

Through <strong>the</strong> example <strong>of</strong> GI illness in Nottinghamshire, this paper investigates <strong>the</strong> potential uses <strong>of</strong><br />

location-based services, mashups and participatory health surveys to better manage an outbreak <strong>of</strong><br />

infectious disease for both service users and public health <strong>of</strong>ficials. A review <strong>of</strong> issues to overcome<br />

when developing such system is presented, and <strong>the</strong>n a mobile phone application is presented. Results<br />

from simulations and real data surveyed, which are ongoing, will be presented at <strong>the</strong> conference.<br />

2. Existing issues towards participatory health surveys<br />

Participatory surveys can be considered as ‗systematic inquiry, with <strong>the</strong> collaboration <strong>of</strong> those<br />

affected by <strong>the</strong> issue being studied, for purposes <strong>of</strong> education and taking action or effecting change‘ (I<br />

Green online, 2010). Rains and Ray (<strong>19</strong>95) conducted a study on <strong>the</strong> use <strong>of</strong> participatory action<br />

research in <strong>the</strong> community for <strong>the</strong> purpose <strong>of</strong> health promotion. The study indicated multiple benefits<br />

<strong>of</strong> including <strong>the</strong> community in health research from <strong>the</strong> initial framing <strong>of</strong> <strong>the</strong> research questions to <strong>the</strong><br />

analysis <strong>of</strong> findings. This may indicate a potential necessity to include <strong>the</strong> public in <strong>the</strong> creation <strong>of</strong> a<br />

participatory health survey on GI illnesses.<br />

2.1 Online Symptom Checkers<br />

Online symptom checkers are extremely popular and <strong>the</strong>re are a huge number <strong>of</strong> different websites<br />

hosting a variety <strong>of</strong> health surveys. It is important to establish <strong>the</strong> difference between a symptom<br />

checker <strong>of</strong>fering medical advice and a health survey collecting data. It is anticipated that <strong>the</strong> general<br />

public will be more willing to contribute to a health survey if <strong>the</strong>y will benefit from <strong>the</strong> results.<br />

Therefore, it is necessary to provide appropriate feedback such as epidemiological data in <strong>the</strong> form <strong>of</strong><br />

cartographic visualisation. To be relevant real-time feedback is important. The HPA reports on GI<br />

illness (HPA online, 2010), as well as <strong>the</strong> aforementioned NHS pages provide medical information to<br />

aid <strong>the</strong> creation <strong>of</strong> <strong>the</strong> participatory health survey. Fur<strong>the</strong>rmore, <strong>the</strong> HPA health questionnaires<br />

specific to GI illness can be used to establish <strong>the</strong> important epidemiological information required<br />

from participants (Adams, 2010). The HPA questionnaires can be used as a basis to assess <strong>the</strong><br />

important epidemiological data required but can be reduced in length and complexity to make it<br />

compatible with an online survey or mobile application (Adams, 2010).<br />

2.2 Disease Surveillance<br />

The World Health Organisation (WHO) is responsible for global disease surveillance and reporting.<br />

They are primarily reliant on <strong>the</strong>ir regional departments to report <strong>the</strong> disease status <strong>of</strong> <strong>the</strong>ir region and<br />

to follow protocol established by <strong>the</strong> WHO. At a smaller geographical scale, countries have public<br />

health responsibilities for surveillance <strong>of</strong> disease such as <strong>the</strong> United States‘ Centre for Disease<br />

Control (CDC) and <strong>the</strong> U.K.‘s HPA. Using <strong>the</strong> direct example <strong>of</strong> <strong>the</strong> HPA, surveillance is conducted<br />

at <strong>the</strong>ir head-<strong>of</strong>fice where data and information is input by <strong>the</strong>ir regional HPA <strong>of</strong>fices located<br />

countrywide. Weekly reports are produced as to <strong>the</strong> current disease status and this is used to build up<br />

a general epidemiological pr<strong>of</strong>ile.<br />

Disease surveillance allows for <strong>the</strong> use <strong>of</strong> data from health services in developing an understanding <strong>of</strong><br />

<strong>the</strong> spatial distribution <strong>of</strong> a particular disease. In this way it enables researches to fur<strong>the</strong>r <strong>the</strong>ir<br />

understanding <strong>of</strong> <strong>the</strong> pathways <strong>of</strong> <strong>the</strong> disease in question and understand its epidemiology.<br />

Cartographic representations <strong>of</strong> disease are key in this information-pathway between healthcare<br />

pr<strong>of</strong>essionals and <strong>the</strong> general public, as it allows for <strong>the</strong> visualisation <strong>of</strong> at-risk areas and helps inform<br />

behaviour to reduce risk to susceptible individuals.<br />

2.3 Privacy in Health Data<br />

Much <strong>of</strong> <strong>the</strong> literature on health data addresses <strong>the</strong> concerns surrounding protection <strong>of</strong> privacy and<br />

confidentiality. Al-Iwihan and Leibovici (2010) discuss <strong>the</strong> role <strong>of</strong> web services standards about<br />

insuring data protection and copyright issues when handling health information. They also raise <strong>the</strong><br />

Page | 70


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

concern <strong>of</strong> accessing relevant and accurate health information. A solution is presented in <strong>the</strong> form <strong>of</strong><br />

<strong>the</strong> on-going standardisation <strong>of</strong> GeoXACML to control user access to services and data. Gaining<br />

public trust is important as allowing better quality data, (Leibovici et al. 2010).<br />

To preserve confidentiality Gao et al. (2008) advocate ei<strong>the</strong>r aggregation, removal <strong>of</strong><br />

geographical identifiers, small scale relocation <strong>of</strong> individual, and limitation <strong>of</strong> access. In order to<br />

accurately map <strong>the</strong> data provided by <strong>the</strong> participatory health survey, different levels <strong>of</strong> spatial data<br />

can be used: house address, post-code, GPS tracking systems once concerns over privacy have been<br />

addressed.<br />

2.4 Interactive Mapping<br />

Again, <strong>the</strong>re is much use <strong>of</strong> <strong>GIS</strong> for disease surveillance in <strong>the</strong> United States (e.g. http://gis.cdc.gov/),<br />

but this is relatively limited in <strong>the</strong> U.K., despite efforts from <strong>the</strong> NHS and HPA to increase<br />

implementation <strong>of</strong> disease mapping. It is becoming more widely used, but <strong>the</strong>re is still much progress<br />

to be made in order to ensure <strong>the</strong> use <strong>of</strong> <strong>GIS</strong> and interactive mapping to its full potential.<br />

There are several key interactive online mapping programmes available for general use. Healthmap<br />

and InstantAtlas are two examples <strong>of</strong> <strong>the</strong>se. They provide current health data in cartographic form,<br />

allowing <strong>the</strong> user to identify a key disease or location for greater focus.<br />

Some well known <strong>GIS</strong> desktop applications, such as gvSIG (see OSGeo, 2010) have been ported<br />

under phones with Android operational system, but any smartphone using a browser can benefit from<br />

existing web <strong>GIS</strong> tools such <strong>the</strong> ones developed under <strong>the</strong> open source community as long as a <strong>the</strong><br />

interface can be adapted to mobile devices.<br />

2.5 Mobile Phone Applications for Health Data<br />

Using <strong>the</strong> example <strong>of</strong> an iPhone, <strong>the</strong>re are already several applications devoted to health. The ‗Health<br />

Vitals‘ application includes facilities for monitoring blood pressure, blood sugar and cholesterol<br />

levels (Apple online, 2010a). Similarly, <strong>the</strong> ‗Quick Health Calculator‘ application is able to calculate<br />

personal risk levels for cardio-vascular disease, type-II diabetes and hypertension (Apple online,<br />

2010b). The availability <strong>of</strong> such applications clearly indicates a demand for health-related mobile<br />

phone applications.<br />

3. Proposed Surveying Methodology<br />

For <strong>the</strong> preliminary study, data on <strong>the</strong> prevalence <strong>of</strong> GI illness in Nottinghamshire is required. The<br />

collaboration established with Health Protection Agency is important in this respect for GI<br />

surveillance, at different geographical levels in <strong>the</strong> U.K. Input <strong>of</strong> data from <strong>the</strong> general public on <strong>the</strong>ir<br />

symptoms about GI illness is collected <strong>the</strong>refore allowing refining <strong>of</strong> <strong>the</strong> map <strong>of</strong> <strong>the</strong> local incidence<br />

<strong>of</strong> such illnesses.<br />

Initial data comes from <strong>the</strong> Health Protection Agency Centre for Infections, GI Illness Unit. This data<br />

can be used to map <strong>the</strong> incidence <strong>of</strong> <strong>the</strong>se illnesses for Nottinghamshire. Information from <strong>the</strong> mobile<br />

participatory health survey (Figure 1) will <strong>the</strong>n be amalgamated with relevant information to build up<br />

an epidemiological pr<strong>of</strong>ile <strong>of</strong> GI illnesses through <strong>the</strong> use <strong>of</strong> online, interactive mapping.<br />

A prototype for <strong>the</strong> mobile surveyor is presented on Figure 1. Location-aware self-driven surveys are<br />

a major step forward in receiving user feedback in connection with health-related monitoring. Current<br />

smart-phone market is rapidly growing (Raento at al. 2008), and a competition between producers<br />

makes applications incompatible between devices. To eliminate this problem, this research is built on<br />

Web 2.0 technology utilizing Internet connection and a web browser, which is generally and openly<br />

standardized. The survey system is independent <strong>of</strong> ubiquitous hardware producer, utilizing any<br />

HTML5 enabled browser, e.g., tablets, smart-phones, netbooks, laptops, giving a participation<br />

opportunity to a wider public, and was tested on Android and Iphone devices. Literature review<br />

Page | 71


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

suggests <strong>GIS</strong> model should be based on open standards (Lee and Percivall 2008), and be simple as<br />

possible to avoid unnecessary overhead in data exchange.<br />

Figure 1. Html5 based survey on mobile<br />

phone for public health<br />

The information technology platform built for this research and o<strong>the</strong>r, VGI or with pr<strong>of</strong>essionals,<br />

surveying projects (Pawlowicz at al. 2010) is open and consists <strong>of</strong> several subsystems connected by<br />

Open Geospatial Consortium (OGC) standardized protocols and interfaces, e.g., WPS, WFS, WFS-T,<br />

and WMS.<br />

4. Discussion and conclusion<br />

The impact <strong>of</strong> such system on public health in <strong>the</strong> U.K. can be substantial by alleviating some <strong>of</strong> <strong>the</strong><br />

pressures caused by GI illnesses dealt with by primary health workers which could safely be dealt<br />

with in <strong>the</strong> home. GI illnesses are usually self-limiting, and <strong>the</strong>refore, this type <strong>of</strong> illness would<br />

greatly benefit from this programme, as it can be managed at home with <strong>the</strong> right information and<br />

guidance. However, its clinical presentation can be very similar to that <strong>of</strong> o<strong>the</strong>r, potentially more<br />

serious, illnesses and as such it requires precise and specific details in order to generate direct advice<br />

and management instructions.<br />

Page | 72


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

The use <strong>of</strong> GPS tracking with mobile phones to assess <strong>the</strong> location-based risk <strong>of</strong> contracting a GI<br />

illness is also a potential area <strong>of</strong> interest, and one that could be used alongside <strong>the</strong> online participatory<br />

health survey.<br />

It is hoped that this project in collaboration with <strong>the</strong> HPA can provide a more comprehensive<br />

epidemiological surveillance <strong>of</strong> GI illnesses in <strong>the</strong> U.K. This project has <strong>the</strong> potential to be developed<br />

for use in a pandemic situation, for example Influenza outbreaks. If this system can bring direct<br />

benefit to <strong>the</strong> users ei<strong>the</strong>r before or after <strong>the</strong>y see <strong>the</strong>ir GP, <strong>the</strong>re are none<strong>the</strong>less few important points<br />

remaining to be controlled in order for such participatory health survey to be used in disease<br />

surveillance: data accuracy, validated cases and data conflation with o<strong>the</strong>r sources.<br />

5. Acknowledgements<br />

The first author is grateful to <strong>the</strong> HPA Nottinghamshire for welcoming us and allowing a close<br />

collaboration for this study.<br />

6. References<br />

Adams N (2010). Participatory Health Survey using Ubiquitous Computing. University <strong>of</strong><br />

Nottingham, Centre for Geospatial Science, Summer Internship Report, 39pp<br />

Apple (2010a). Health Vitals Tracker [online] Accessed 6th July 2010. Available at:<br />

http://www.apple.com/webapps/utilities/healthvitalstracker.html<br />

Apple (2010b). Quick Health Calculator [online] Accessed 6th July 2010. Available at:<br />

http://www.apple.com/webapps/calculate/quickhealthcalculator.html<br />

Al-Iwihan R and Leibovici DG (2010). Privacy Protected Health Maps. Horizon Doctoral Training<br />

Centre, report, pp.<br />

Gao S Mioc D Anton F Yi X and Coleman C (2008). Online <strong>GIS</strong> Services for Mapping and Sharing<br />

Disease Information. International Journal <strong>of</strong> Health Geographics 7:8<br />

Green L Selman C Jones T Scallan E Marcus R and EHS-Net Population Survey Working Group<br />

(2004). Beliefs About Sources <strong>of</strong> Gastrointestinal Illness: What Factors are Associated with People‘s<br />

Beliefs that a Meal Eaten Outside <strong>of</strong> <strong>the</strong> Home Made <strong>the</strong>m Sick? Presented at <strong>the</strong> 91st <strong>Annual</strong> Int‘l<br />

Association for Food Protection Mtg.<br />

Healthmap (2010) Healthmap website Available at: http://healthmap.org/en/<br />

Health Protection Agency (2010). Gastrointestinal Disease [online] Accessed 6th July<br />

2010. Available at:<br />

http://www.hpa.org.uk/web/HPAweb&Page&HPAwebAutoListName/Page/1<strong>19</strong><strong>19</strong>42150117<br />

Green I (2010). Guidelines and Categories for Classifying Participatory <strong>Research</strong> Projects in<br />

Health.<br />

Available at: http://www.lgreen.net/guidelines.html<br />

Instantatlas (2010). Public Health Reporting [online] Accessed 5th July 2010. Available at:<br />

http://www.instantatlas.com/public-health_background.xhtml<br />

NHS (2010a). Gastroenteritis [online] Accessed 5th July 2010. Available at:<br />

http://www.nhs.uk/conditions/Gastroenteritis/Pages/Introduction.aspx<br />

NHS (2010b). Rotavirus [online] Accessed 5th July 2010. Available at:<br />

http://www.nhs.uk/Conditions/Rotavirus-gastroenteritis/Pages/Introduction.aspx?url=Pages/What-isit.aspx<br />

Lee C and Percivall G (2008). Standards-based computing capabilities for distributed geospatial<br />

applications. Computer 41, no. 11: 50-57.<br />

Leibovici DG Anand S Swan J Goulding J Hobona G Bastin L Pawlowicz S Jackson M and James,<br />

Richard (2010) Workflow issues for Health mapping "mashups" <strong>of</strong> OGC. University <strong>of</strong> Nottingham,<br />

CGS Technical report, 2010DL1, 6 pp<br />

OSGeo (2010) The Open Source Geospatial Foundation http://www.osgeo.org<br />

Page | 73


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Pawlowicz S Leibovici DG, Saull R Haines-Young R and Jackson M (2011) Dynamic model<br />

adjustment from crowd-sourced observations (to be submitted)<br />

Rains J and Ray D (<strong>19</strong>95). Participatory Action <strong>Research</strong> for Community Health Promotion. Public<br />

Health Nursing, 12:4<br />

Raento M Oulasvirta A and Eagle N (2009). Smartphones: An Emerging Tool for Social<br />

Scientists. Sociological Methods <strong>Research</strong> 37, no. 3 (February 1): 426-454.<br />

7. Biography.<br />

Natalie Adams, after finishing her degree in Geography, is studying a Master in Public Health at <strong>the</strong><br />

University <strong>of</strong> Nottingham, she is much interested in health and geography and <strong>the</strong> use <strong>of</strong> new<br />

technologies in <strong>the</strong>se fields. She did internship at <strong>the</strong> Centre for Geospatial Science on this project.<br />

Chaoyu Ye is a third year undergraduate at University <strong>of</strong> Nottingham. He is an Internship student at<br />

<strong>the</strong> Open Source Geospatial Lab at <strong>the</strong> Centre for Geospatial Science.<br />

Wenchao Jiang is a third year undergraduate at University <strong>of</strong> Nottingham. He is an Internship<br />

student at <strong>the</strong> Open Source Geospatial Lab at <strong>the</strong> Centre for Geospatial Science.<br />

Sergiusz Pawlowicz is in <strong>the</strong> third year <strong>of</strong> his PhD at <strong>the</strong> Centre for Geospatial Science. He has<br />

acquired considerable experience in managing large scale computing systems whilst working for <strong>the</strong><br />

BBC for few years. His motivations are now to use <strong>the</strong>se skills within <strong>the</strong> research context <strong>of</strong><br />

volunteered Geographical Information systems.<br />

Dr Suchith Anand is Ordnance Survey <strong>Research</strong> Fellow at <strong>the</strong> Centre for Geospatial Science,<br />

University <strong>of</strong> Nottingham. His research interests are in open source <strong>GIS</strong>, automated<br />

mapgeneralization, geohydroinformatics, mobile<strong>GIS</strong>, location based services, optimization<br />

techniques and asset management.<br />

Dr Didier G. Leibovici is a <strong>Research</strong> Fellow in geospatial modelling and analysis, with previous<br />

posts as a statistician in epidemiological/medical imaging research and as a geomatician for<br />

landscape changes in agro- ecology. His interests in Multivariate spatial interaction analysis refer<br />

also to interoperability and conflation models for cross-scales <strong>of</strong> integrated modelling applications<br />

within an interoperable framework chaining web services.<br />

Pr. Mike Jackson is Director <strong>of</strong> <strong>the</strong> Centre for Geospatial Science. Prior to this he worked in<br />

industry at QinetiQ, Hutchison 3G, Laser Scan in various geospatial specialist and executive roles<br />

and in research for NERC. Mike is non-executive director <strong>of</strong> <strong>the</strong> Open Geospatial Consortium, and<br />

has research interests in combining new technologies such as positioning, pervasive computing and<br />

location based services for geo-informatics applications<br />

Page | 74


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Spatio-temporal change in population and Health facility location<br />

planning: A case study <strong>of</strong> Ambulance location planning in Leicestershire.<br />

Emeka Chukwusa. 1 , Alexis Comber. 2 , Chris Brunsdon. 3<br />

Department <strong>of</strong> Geography, University Of Leicester, University Road, Leicester.<br />

LE1. 7RH.Telephone: +44(0)1162523823, Fax: +44(0)1162523854<br />

ec102@le.ac.uk 1 , ajc36@le.ac.uk 2 , cb179@le.ac.uk 3<br />

ABSTRACT<br />

Ambulance location must be sensitive to spatio-temporal variation in demand in order for users to<br />

derive maximum benefit from ambulatory care. Previous studies have ignored spatio-temporal<br />

variation in demand, when modelling location for health facilities (ambulances). This study addressed<br />

this gap by showing how to locate and allocate ambulances at different time states, when demand is<br />

spatio-temporally varying. Using travel to work data; this study applied a modified P-median model<br />

to optimise ambulance location for day and night time population. A subset <strong>of</strong> 12 ambulance<br />

locations were selected from a given set <strong>of</strong> 20 ambulance station locations and each was allocated to<br />

population in 583 Lower super output areas <strong>of</strong> Leicestershire.<br />

KEYWORDS: Spatial-temporal planning; Ambulance location; Location-allocation model; Pmedian.<br />

1. Introduction:<br />

Strategic planning <strong>of</strong> health facility location is a decisive factor to improve accessibility, demand<br />

coverage or response times (in <strong>the</strong> case <strong>of</strong> ambulances). The location problem is a spatial resource<br />

problem (Brandeau and Chiu, <strong>19</strong>89) that involves making strategic decisions <strong>of</strong> where to locate, how<br />

to allocate and what time to locate facilities to ensure adequate coverage <strong>of</strong> demand. For example a<br />

typical ambulance location problem involves making decisions on where to locate ambulances or<br />

paramedic facilities to minimise response time. In order to solve this problem, several location<br />

modelling techniques have been proliferated in <strong>the</strong> literature. These techniques coupled with <strong>GIS</strong><br />

spatial analysis, have proven to be very efficient in addressing location-allocation problems.<br />

However, one basic limitation in <strong>the</strong>ir application is <strong>the</strong> assumptions that demand (population) that<br />

uses public health facilities are fixed. This is contrary to real life scenarios where demands are<br />

generally spatially mobile at various times <strong>of</strong> <strong>the</strong> day.<br />

Given this limitation, this study addresses <strong>the</strong> problem <strong>of</strong> finding optimum ambulance locations for<br />

two time states (day & night) using a modified P-median that accounts for spatio-temporal variation<br />

in demand. This study demonstrates an example <strong>of</strong> where to locate ambulances or paramedic facilities<br />

by choosing a subset <strong>of</strong> 12 ambulance locations from a given set <strong>of</strong> 20 ambulance stations in <strong>the</strong><br />

study area. The objective is to minimise <strong>the</strong> distance between ambulance locations and population<br />

weighted centroids <strong>of</strong> 583 Lower super output areas (LSOAs) in Leicestershire. Locating and<br />

allocating emergency medical services strategically to account for spatio-temporal variation in<br />

demand will <strong>of</strong>fer improved access to ambulatory care and minimise ambulance response time at<br />

various time states.<br />

Two categories <strong>of</strong> location models have been identified in <strong>the</strong> literature: Deterministic and Stochastic<br />

location models.<br />

Page | 75


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

1.2 Deterministic models<br />

Deterministic location models assume static input such as fixed travel time or distances along<br />

network edges. These models have been used to solve many health facility location problems in <strong>the</strong><br />

literatures. One aspect <strong>of</strong> research interest is <strong>the</strong> area <strong>of</strong> accessibility improvement and evaluation <strong>of</strong><br />

future health location choices <strong>of</strong> fixed health facilities. For example, Bennett (<strong>19</strong>81) using a distance<br />

minimising model (quantitative) toge<strong>the</strong>r with questionnaire survey (qualitative), demonstrated <strong>the</strong><br />

efficiency <strong>of</strong> location models in improving present and future access to health services in Lansing<br />

area <strong>of</strong> Michigan. In a similar study, <strong>the</strong> authors used <strong>GIS</strong> techniques with <strong>the</strong> objective <strong>of</strong><br />

minimising weighted distance and improving coverage to improve accessibility to health services in<br />

rural Ghana (Moller-Jensen and K<strong>of</strong>ie, 2001). Oppong and Hodgson (<strong>19</strong>94) demonstrated in <strong>the</strong>ir<br />

study that facility addition without efficient location choices is not enough to guarantee improved<br />

access to health services. Similar outcomes have been reported in a study on access to rural health<br />

care in Nigeria conducted by Ayeni, et al. (<strong>19</strong>87).<br />

Location models have also been used to plan <strong>the</strong> location <strong>of</strong> mobile health facilities (e.g; blood banks<br />

and Ambulances). For example; Sasaki, et al. (2010) used a modified genetic to optimise <strong>the</strong> location<br />

<strong>of</strong> present and future health planning. Also, McAleer and Naqvi (<strong>19</strong>94) used a P- median model to<br />

identify <strong>the</strong> relocation <strong>of</strong> ambulance site in Belfast. In addition, Jacobs, et al. (<strong>19</strong>96) used a<br />

capacitated P- median model to optimise <strong>the</strong> operations <strong>of</strong> blood bank for American Red Cross in<br />

Norfolk, Virginia.<br />

1.2 Stochastic models<br />

Stochastic models are used when <strong>the</strong>re is uncertainty in data inputs or parameters (Murray, 2010). For<br />

example Niko<strong>of</strong>al and Sadjadi (2010) in <strong>the</strong>ir study proposed a robust optimisation model for solving<br />

a P- median model when edge or links on a network are fraught with uncertainties. Murawski and<br />

Church (2009) formulated <strong>the</strong> Maximal covering network improvement model to account for<br />

situation where links or edges are under development. The authors tested this model on dataset <strong>of</strong><br />

Suhum district <strong>of</strong> Ghana and outcomes showed an improvement in access to health services.<br />

O<strong>the</strong>r study, have also considered scenarios with stochastic demand. For example, Ruslim and Ghani<br />

(2006) used a P- median problem model to evaluate uncertainty in <strong>the</strong> demand for emergency medical<br />

services in Austin, Texas. In addition, Kutangila and Verdegay (2005) proposed a model for solving a<br />

P- median problem when demand nodes and edge lengths are fuzzy or uncertain.<br />

2. Modelling day and night population variation in Leicestershire<br />

Travel to work data is a matrix showing temporary migration <strong>of</strong> people from <strong>the</strong>ir residence (origins)<br />

to <strong>the</strong>ir workplaces (destinations) derived from http://cider.census.ac.uk/cider/wicid. Travel areas<br />

considered include LSOAs in Leicester, Blaby, Charnwood, Harborough, Hinckley & Bosworth,<br />

Melton, North West Leicestershire and Oadby & Wigston. This data was based on 2001 census<br />

estimates.<br />

Travel to work data was used to illustrate spatio-temporal variation in demand for two time states<br />

(day and night). Day time represents time intervals between 9:00 am – 5:00pm, when people are at<br />

<strong>the</strong>ir place <strong>of</strong> work, Night time falls within 5:01pm – 8.59am, when people are at <strong>the</strong>ir residence. Day<br />

and Night time includes Monday – Friday, <strong>the</strong>se are days that people mostly travel to work.<br />

Population estimates for both time states in each LSOA were derived from Equation (1). LSOAs<br />

consist <strong>of</strong> homogenous census units with a mean population <strong>of</strong> 1500 people. Day and night time<br />

populations were based on working-age 16 to 74, weighted for each LSOA to generate demand<br />

centroid points.<br />

Night time population represents <strong>the</strong> total population from equation 1. In- commuters are people<br />

Page | 76


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

living in o<strong>the</strong>r LSOAs but have <strong>the</strong>ir work place in a receiving lower super output area. For example,<br />

people commuting from Nuneaton to <strong>the</strong>ir workplace in Leicester. Out-commuters consist <strong>of</strong><br />

outward-bound people leaving <strong>the</strong>ir residents to work places in o<strong>the</strong>r neighbouring LSOAs.<br />

3. P- median model for spatio-temporal varying demand<br />

The P- median selects subset <strong>of</strong> facilities known as P facilities from a given set <strong>of</strong> candidate facilities<br />

that minimises <strong>the</strong> aggregate travel time or distance between demand points and nearest facility<br />

locations (Fo<strong>the</strong>ringham, et al., <strong>19</strong>95). In this study, <strong>the</strong> classical P- median model first espoused by<br />

ReVelle and Swain (<strong>19</strong>70), was modified to account for spatio-temporal variation by incorporating a<br />

spatial variation (xi, yi) and time component ( tn ) for day and night time populations .<br />

P = subset <strong>of</strong> 12 ambulance locations from 20 ambulance stations<br />

I....m. = Set <strong>of</strong> demand locations (583 population weighted centroid points). J....n. = Set <strong>of</strong><br />

ambulance station location (20 ambulance stations).<br />

(xi , yi) = Location coordinates (showing spatial variation in demand).<br />

tn = Time (showing temporal change in demand during <strong>the</strong> Day or Night).<br />

dij = Shortest distance between demand and ambulance station locations ( from <strong>GIS</strong> network<br />

analysis).<br />

ai = weight <strong>of</strong> demand node i at time tn., Z is <strong>the</strong> objective function to be minimised.<br />

Criteria:<br />

� There are no restrictions on <strong>the</strong> amount <strong>of</strong> demands each ambulance location can serve.<br />

� Only one location can be allocated to each LSOA at a time.<br />

� No new locations are allowed, only given locations are considered.<br />

� There are no capacity constraints on <strong>the</strong> number on <strong>the</strong> maximum or minimum numbers <strong>of</strong><br />

ambulance per station.<br />

A typical location-allocation problem involves selecting optimum location choices from a pool <strong>of</strong><br />

candidate location and allocating demand to <strong>the</strong>se points. In this study, <strong>the</strong> pool <strong>of</strong> candidate location<br />

consists <strong>of</strong> 20 ambulance stations in Leicestershire, with a choice <strong>of</strong> selecting 12 ambulance stations<br />

to allocate to 583 LSOAs centroids representing demands. Finding solution to this type <strong>of</strong> problem is<br />

computationally difficult because <strong>the</strong> solution search space is large. For example, choosing 12<br />

ambulance locations from a given set <strong>of</strong> 20 locations require a solution search space <strong>of</strong> 20! /12! (20-<br />

12)! possible solutions. Deriving solution for this problem involve <strong>the</strong> application <strong>of</strong> heuristic. Teitz<br />

and Bart heuristic was applied to solve <strong>the</strong> P- median problem (Equation 2). It is an interchange<br />

heuristic that selects a set <strong>of</strong> initial random solutions and improves <strong>the</strong>ir outcome by swapping until<br />

<strong>the</strong>re are no fur<strong>the</strong>r improvements in <strong>the</strong> solution <strong>of</strong> <strong>the</strong> objective function (Teitz and Bart, <strong>19</strong>68).<br />

4. Results<br />

Outcome from location analysis using P-median, shows difference in location preference for<br />

ambulances during day and night. Figure 1and 2 show <strong>the</strong> optimum locations for ambulance for night<br />

and day time population respectively. The location model suggests a better way to allocate ambulance<br />

resources during <strong>the</strong> day and night to ensure optimal coverage. Optimum locations for ambulances are<br />

represented with proportional circles. Each circle is located on <strong>the</strong> areas <strong>the</strong>y serve, with <strong>the</strong>ir sizes<br />

relative to <strong>the</strong> proportion <strong>of</strong> demand <strong>the</strong>y cover. For example, a larger circle signifies more demand<br />

allocations. LSOAs with similar hue have same optimum coverage<br />

Page | 77


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Figure 1. LSOAs with same optimum coverage are represented with same colour (Symbols<br />

represent optimum locations & sizes are proportional to <strong>the</strong> amount <strong>of</strong> demand allocation)<br />

Figure 2, LSOAs with same optimum coverage are represented with same colour (Symbols<br />

represent optimum locations & sizes are proportional to <strong>the</strong> amount <strong>of</strong> demand allocation)<br />

Figure 1 and 2, shows difference in location preference for day and night time population. Day time<br />

population estimates show more allocation in central LSOAs <strong>of</strong> Leicestershire compared to Night<br />

time population with fewer allocations. This is obviously as a result <strong>of</strong> <strong>the</strong> large number <strong>of</strong> people<br />

that commute to work from o<strong>the</strong>r LSOAs into central Leicestershire during <strong>the</strong> day. In both time<br />

states <strong>the</strong>re are more ambulances allocated to central parts <strong>of</strong> Leicestershire due to <strong>the</strong> high<br />

concentration <strong>of</strong> demands.<br />

Page | 78


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Table 1. Optimum Location for Ambulances and number <strong>of</strong> demand served.<br />

NIGHT TIME POPULATION DAY TIME POPULATION<br />

Ambulance Demand Relative Ambulance Demand Relative<br />

Location Demand Location Demand<br />

1 89083 0.596 1 81185 0.675<br />

14 59231 0.396 14 55698 0.463<br />

3 149423 1.000 3 113069 0.939<br />

4 125126 0.837 4 96956 0.806<br />

18 115159 0.771 5 86035 0.715<br />

6 76521 0.512 6 72909 0.606<br />

7 68447 0.458 7 65653 0.546<br />

8 96900 0.648 8 96343 0.801<br />

9 <strong>19</strong>224 0.129 9 21694 0.180<br />

2 23380 0.156 10 33988 0.282<br />

11 47866 0.320 11 45624 0.379<br />

15 <strong>19</strong>147 0.128 20 120353 1.000<br />

Table 1 shows <strong>the</strong> index <strong>of</strong> optimum locations; allocated demands and relative demand for day and<br />

night time population. From Table 1, Ambulance 3 has <strong>the</strong> largest demand allocation for night time<br />

population, with a relative demand <strong>of</strong> 1. This implies that more demand is allocated to this ambulance<br />

station at night. Also for day time population, Ambulance 20 has <strong>the</strong> highest demand allocation.<br />

Differences in day and night time ambulance allocation reflects shift in demand at various time states.<br />

5. Summary and Conclusion<br />

In this study, P- median model was used to model optimum locations for ambulances in<br />

Leicestershire. This analysis is based on <strong>the</strong> notion that demand changes position at different times.<br />

Results from P-median suggest differences in location preferences for day and night time population.<br />

This is an indication that ambulance location planning is sensitive to changes in demand.<br />

From this study, it is obvious that spatio-temporal variation in demand influences <strong>the</strong> location and<br />

allocation <strong>of</strong> ambulances. Given this outcome, it is important for emergency service planners to<br />

understand <strong>the</strong> changing locations <strong>of</strong> demand when planning or allocating ambulance services. This is<br />

necessary to ensure that more rational decisions are made on where to locate, how to allocate and<br />

what time to allocate ambulances prior to <strong>the</strong> event <strong>of</strong> an emergency. In addition, outcomes from this<br />

study can be used to formulate or evaluate policies on how to optimise scarce or limited health<br />

resources.<br />

Finally, <strong>the</strong> result from this study demonstrates <strong>the</strong> importance <strong>of</strong> location-allocation models in<br />

addressing modelling optimum location and <strong>the</strong> role <strong>of</strong> <strong>GIS</strong> in solving location problems.<br />

6. Acknowledgements<br />

Special thanks go to my supervisors Dr Lex Comber and Pr<strong>of</strong>. Chris Brunsdon for <strong>the</strong>ir support and<br />

encouragement during <strong>the</strong> first year <strong>of</strong> my PhD.<br />

Page | 79


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

7. References<br />

Ayeni, M. A., Rushton, G., & McNulty, M. L. (<strong>19</strong>87) Improving <strong>the</strong> geographical accessibility <strong>of</strong><br />

health care in rural areas: A Nigerian case study. Social Science & Medicine, (25) 1083–1094.<br />

Bennett, W. D. (<strong>19</strong>81) A location-allocation approach to health care facility location: A study <strong>of</strong><br />

<strong>the</strong> undoctored population in Lansing, Michigan. Social Science & Medicine, (15D), 305–312.<br />

Brandeau, M.L. and Chiu, S.S., (<strong>19</strong>89) An overview <strong>of</strong> representative problems in location<br />

research. Management, 35(6), pp.645-674.<br />

Fo<strong>the</strong>ringham, A. S, Curtis A, & Densham , P. J (<strong>19</strong>95) The zone definition problem and<br />

location– allocation modelling. Geographical Analysis 27:60–77.<br />

Jacobs, D.A., Silan, M.N. & Clemson, B.A. (<strong>19</strong>96), An Analysis <strong>of</strong> alternative locations and service<br />

areas <strong>of</strong> American Red Cross blood facilities. Interfaces, 26, 40-50.<br />

Kutangila, D. and Verdegay, J.L.(2005). P-Median Problems in a Fuzzy Environment. Mathware and<br />

S<strong>of</strong>t computing 12, 97-106.<br />

McAleer, W.E. and Naqvi I.A. (<strong>19</strong>94), The relocation <strong>of</strong> ambulance stations: A successful case study,<br />

European Journal <strong>of</strong> Operational <strong>Research</strong>, 75, 582-588.<br />

Møller-jensen, L. and K<strong>of</strong>ie, R. Y., (2001) Exploiting available data sources: location/allocation<br />

modeling for health service planning in rural Ghana. Geografisk Tidsskrift,101, 145-153.<br />

Murawski, L. and Church, R. L.(2009). Improving accessibility to rural health services: The<br />

maximal covering network improvement problem. Socio-Economic Planning Sciences, 43(2),<br />

pp.102-110.<br />

Murray, A.T. (2010). Advances in location modelling: <strong>GIS</strong> linkages and contributions. Journal<br />

<strong>of</strong> Geographical Systems, 12(3), 335-354.<br />

Niko<strong>of</strong>al, M.E. and Sadjadi, S.J., (2010). A robust optimization model for p-median problem with<br />

uncertain edge lengths. The International Journal <strong>of</strong> Advanced Manufacturing Technology, 50(1-<br />

4), .391-397.<br />

Oppong, J. R., and M.J. Hodgson (<strong>19</strong>94). Spatial accessibility to health facilities in Suhum<br />

district, Ghana, Pr<strong>of</strong>essional Geographer 46 (2) <strong>19</strong>9±209.<br />

ReVelle, C.S. and Swain, R.W. (<strong>19</strong>70) Central facilities location. Geographical Analysis, 2: 30-42.<br />

Ruslim, N.M. and Ghani, N.A. (2006). An Application <strong>of</strong> <strong>the</strong> P- median problem with uncertainty<br />

in demand in emergency medical services. <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> 2nd IMT-GT Regional <strong>Conference</strong><br />

on Ma<strong>the</strong>matics, Statistics and Applications. Malaysia.<br />

Sasaki. S., Comber A. J., Suzuki. H, & Brunsdon, .C (2010) Using Genetic Algorithms to<br />

optimise current and future health planning – The example <strong>of</strong> ambulance locations. International<br />

Journal <strong>of</strong> health geographics , 9:4 doi:10.1186/1476-072X-9-4.<br />

Teitz, M. B., and Bart. P (<strong>19</strong>68). Heuristic Methods for Estimating <strong>the</strong> Generalized Vertex Median<br />

<strong>of</strong> a Weighted Graph. Operations <strong>Research</strong> 16:955–6.<br />

8 Biographies<br />

Emeka Chukwusa is a First year, PhD research Student at <strong>the</strong> Department <strong>of</strong> Geography,University<br />

<strong>of</strong> Leicester. His research interest is on Health facility location planning, Accessibility to health<br />

services, location-allocation modelling using open source s<strong>of</strong>tware (R s<strong>of</strong>tware) and<br />

Geodemographic pr<strong>of</strong>iling.<br />

Alexis Comber is a senior lecturer in Geographic information at Department <strong>of</strong> Geography,<br />

University <strong>of</strong> Leicester, Leicester. His research interest is includes Accessibility, equity <strong>of</strong> access and<br />

optimisation.<br />

Chris Brunsdon is a pr<strong>of</strong>essor <strong>of</strong> Geographical information at <strong>the</strong> Department <strong>of</strong> Geography,<br />

University <strong>of</strong> Leicester, Leicester. His research interest includes; Spatial statistics; exploratory data<br />

analysis; data visualisation; house price modelling and crime pattern analysis<br />

Page | 80


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Using a <strong>GIS</strong>-based network analysis to determine Saudi and non-Saudi<br />

accessibility to <strong>the</strong> Primary Health Care Centers in Buraydah City,<br />

Kingdom <strong>of</strong> Saudi Arabia<br />

Ibrahim Alshwesh 1 , Alexis Comber 2 , Chris Brunsdon 3<br />

1 Department <strong>of</strong> Geography, University <strong>of</strong> Leicester, Leicester LE1 7RH, <strong>UK</strong><br />

Tel. (00441162523823) Fax (00441162523854)<br />

Email, ia98@le.ac.uk 1 , ajc36@le.ac.uk 2 , cb179@le.ac.uk 3<br />

ABSTRACT<br />

Primary health care centers (PHCCs) are <strong>the</strong> first choice for patients, according to health systems in<br />

<strong>the</strong> Kingdom <strong>of</strong> Saudi Arabia (KSA). Saudi and non-Saudi accessibility to PHCCs in Buraydah city<br />

was analysed using a network analysis in a geographical information system (<strong>GIS</strong>). This paper<br />

revealed that <strong>the</strong> Saudis, who represent <strong>the</strong> largest proportion <strong>of</strong> <strong>the</strong> population in <strong>the</strong> city <strong>of</strong><br />

Buraydah, have 46.9% less access to PHCCs than non-Saudis. In addition, <strong>the</strong> use <strong>of</strong> network<br />

analysis in a <strong>GIS</strong> can be linked with statistical methods to support <strong>the</strong> decision-makers in detecting<br />

<strong>the</strong> accessibility and geographical distribution <strong>of</strong> health services.<br />

KEYWORDS: PHCCs; accessibility; Network analysis; <strong>GIS</strong>; geographical distribution.<br />

1. Introduction<br />

The overall aim <strong>of</strong> this study is to analyse Saudi and non-Saudi accessibility to primary health care<br />

centers (PHCCs) in <strong>the</strong> city <strong>of</strong> Buraydah through a <strong>GIS</strong>-based network analysis. The PHCCs in <strong>the</strong><br />

KSA are <strong>the</strong> main service in <strong>the</strong> defence against diseases and <strong>the</strong>ir most important responsibilities are<br />

preventive and curative. According to <strong>the</strong> World Health Organization (2008), in order for health<br />

services to be fair and efficient with universal coverage, <strong>the</strong>y must be made available to all people<br />

according to <strong>the</strong>ir health needs regardless <strong>of</strong> <strong>the</strong>ir ability to pay. In 2009, <strong>the</strong> Ministry <strong>of</strong> Health in<br />

KSA began an important phase focusing on <strong>the</strong> renewal <strong>of</strong> PHCCs, and provided services at different<br />

standards; in addition it provided <strong>the</strong> service to all city areas. Thus, PHCCs play an important role in<br />

<strong>the</strong> delivery <strong>of</strong> health services and are intended to be available to all residents but uneven<br />

geographical distribution <strong>of</strong> health services may result in access deprivation <strong>the</strong>m between Saudis and<br />

non-Saudis. Therefore, this study seeks to assess <strong>the</strong> accessibility ratio between Saudis and non-<br />

Saudis, according to <strong>the</strong> current status based on <strong>the</strong> locations <strong>of</strong> <strong>the</strong>se services.<br />

2. Background and Literature Review<br />

2.1 Access to PHCCs<br />

There are no specific criteria that can guide in <strong>the</strong> assessment <strong>of</strong> <strong>the</strong> access to PHCCs in <strong>the</strong> KSA, but<br />

<strong>the</strong>re is a limitation on <strong>the</strong> size <strong>of</strong> a population that can be served by PHCC, as classified by <strong>the</strong><br />

Ministry <strong>of</strong> Health; e.g. a PHCC classification (M1) can serve up to 32000 people, whilst in contrast,<br />

a PHCC classification (B3) can serve between 2000 and 12000 people (General Directorate <strong>of</strong> Health<br />

Affairs in Al Qassim, 2010). Although <strong>the</strong>re have been a number <strong>of</strong> studies conducted on PHCCs in<br />

KSA, <strong>the</strong>re are few studies that have evaluated accessibility and geographical distribution within<br />

cities. Al Ghamdi (<strong>19</strong>81) suggested developing a program for governmental medical centres that are<br />

accessible within a journey <strong>of</strong> not more than ten minutes for each population group in <strong>the</strong><br />

neighbourhoods <strong>of</strong> <strong>the</strong> city <strong>of</strong> Jeddah, KSA. In addition, Al Rabdi (<strong>19</strong>90) presented <strong>the</strong> first study on<br />

medical services and <strong>the</strong> benefits from <strong>the</strong>m in <strong>the</strong> Al Qassim province, by focusing on <strong>the</strong><br />

characteristics <strong>of</strong> PHCCs. Moreover, Al Dlghan (<strong>19</strong>92) presented an analysis <strong>of</strong> <strong>the</strong> current pattern <strong>of</strong><br />

spatial distribution and access to PHCCs in Riyadh city.<br />

Page | 81


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

2.2 Access to PHCCs by using <strong>GIS</strong><br />

The Ministry <strong>of</strong> Health in KSA has focused in recent times on <strong>the</strong> importance <strong>of</strong> <strong>GIS</strong> maps in <strong>the</strong><br />

field <strong>of</strong> health, where it has begun to establish a database for health services in some <strong>of</strong> <strong>the</strong> bigger<br />

cities such as Riyadh and Jeddah. In <strong>the</strong> face <strong>of</strong> <strong>the</strong> absence or unavailability <strong>of</strong> census data for<br />

smaller areas and buildings in KSA, some studies have relied on larger spatial units such as<br />

neighbourhoods or blocks, to assess <strong>the</strong> access or <strong>the</strong> geographical distribution <strong>of</strong> PHCCs. And <strong>the</strong><br />

neighbourhoods that are residential areas are identified to divide <strong>the</strong> city into a number <strong>of</strong> areas,<br />

which in turn are <strong>the</strong>n divided into number <strong>of</strong> blocks, which represent <strong>the</strong> smaller unit area <strong>of</strong><br />

neighbourhoods. Typically, <strong>the</strong> average number <strong>of</strong> people in <strong>the</strong>se neighbourhoods is between 5000<br />

and 15000 people. For instance, Al Shahrani (2004) suggested several locations for PHCCs in<br />

Riyadh, according to <strong>the</strong> demanded ‗catchment area' by using questionnaires and population density<br />

data. In addition, Al Dossari (2009) highlighted <strong>the</strong> use <strong>of</strong> <strong>GIS</strong> applications to evaluate accessibility<br />

to PHCCs in Riyadh city, by calculating <strong>the</strong> distance to, and a blocks‘ accessibility in a catchment<br />

area to, <strong>the</strong> nearest PHCCs.<br />

<strong>GIS</strong> analyses <strong>of</strong> PHCCs have varied greatly. Some authors have focused on access to <strong>the</strong>se centres,<br />

such as Møller-Jensen and K<strong>of</strong>ie (2001) presenting scenarios aimed at improving accessibility to<br />

health services in Ghana, using location-allocation modelling tools such as closest facility through a<br />

network. In addition, Luo and Wang (2003) <strong>of</strong>fered an assessment <strong>of</strong> <strong>the</strong> spatial differences between<br />

access to PHCCs based on <strong>the</strong> attractiveness <strong>of</strong> doctors and several o<strong>the</strong>r variables in <strong>the</strong> Chicago<br />

region. Rosero-Bixby (2004) presented a <strong>GIS</strong>-based study evaluating equity in access to health care<br />

among Costa Ricans using traditional measurements, such as distance to <strong>the</strong> nearest health facility;<br />

and suggested that an index <strong>of</strong> access to healthcare should be based on <strong>the</strong> characteristics <strong>of</strong> <strong>the</strong><br />

population and <strong>the</strong> nearest health facility. Tanser et al. (2006) presented an analysis <strong>of</strong> costs within<br />

<strong>the</strong> geographic information system in order to determine <strong>the</strong> time it takes (in any particular place) to<br />

reach a primary healthcare centre. Bagheri et al. (2006) demonstrated a new approach for access to<br />

primary healthcare by identifying <strong>the</strong> distance to <strong>the</strong> nearest primary healthcare facilities across <strong>the</strong><br />

road network by using <strong>the</strong> centre <strong>of</strong> population distribution within each polygon. In <strong>the</strong> <strong>UK</strong>, Langford<br />

et al. (2007) showed <strong>the</strong> impact <strong>of</strong> alternative models, such as population distribution, to determine<br />

spatial access to primary healthcare services, for example, raster map data with database and mailing<br />

information in order to increase accuracy in determining <strong>the</strong> residential areas and primary healthcare<br />

centres in <strong>the</strong> study area <strong>of</strong> Cardiff, South Wales.<br />

2.3 Access to health services by using <strong>GIS</strong><br />

The Locations <strong>of</strong> health services have been studied since <strong>19</strong>60 by many sciences, for example,<br />

geography, spatial planning, industry, engineering, and public administration (Teixeira and Antunes,<br />

2008). Accordingly, several authors have addressed locations <strong>of</strong> health services when trying to<br />

maximizing access to <strong>the</strong>se facilities. For example; <strong>the</strong> use <strong>of</strong> models <strong>of</strong> network analysis methods<br />

such as travel time, ra<strong>the</strong>r than o<strong>the</strong>r traditional models such as Euclidean distance is an effective<br />

analysis to examine and to evaluate <strong>the</strong> supply and demand for health services (Walsh et al. <strong>19</strong>97).<br />

Martin et al. (<strong>19</strong>98) used catchment areas analysis in <strong>GIS</strong> to calculate <strong>the</strong> travel distances to renal<br />

replacement <strong>the</strong>rapy in England. They found that <strong>the</strong> use <strong>of</strong> travel distance more effective than <strong>the</strong><br />

crow-fly distances. In addition, Schuurman et al. (2006) presented <strong>the</strong> methodology <strong>of</strong> <strong>the</strong> model <strong>of</strong><br />

use <strong>of</strong> catchments areas with <strong>the</strong> network analysis based on travel time to reduce <strong>the</strong> costs <strong>of</strong> <strong>the</strong><br />

locations <strong>of</strong> rural hospitals in British Columbia's. Moreover, Haynes et al. (2006) used <strong>the</strong> analysis <strong>of</strong><br />

travel time in <strong>GIS</strong> to compare this estimates with <strong>the</strong> real access by car for some patients to <strong>the</strong> clinics<br />

<strong>of</strong> 8 hospitals in <strong>the</strong> North <strong>of</strong> England. They found that <strong>the</strong>re was a strong association between <strong>the</strong><br />

estimated time and real access.<br />

Little research has been conducted that has examined geographic access to PHCCs for different ethnic<br />

groups. This paper shows that <strong>the</strong> use <strong>of</strong> network analysis in a <strong>GIS</strong> can be linked with statistical<br />

methods to support decision-makers to resolve disparities in health accessibility through analysis <strong>of</strong><br />

<strong>the</strong> geographical distribution <strong>of</strong> <strong>the</strong>se services in relation to demand. In addition, this study <strong>of</strong>fers a<br />

fur<strong>the</strong>r application for <strong>the</strong> method presented by Comber et al. (2008), when combining <strong>the</strong> network<br />

Page | 82


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

analyses in <strong>GIS</strong> with statistical analyses.<br />

3. Method<br />

3.1 Study area<br />

Al Qassim province is one <strong>of</strong> thirteen areas in <strong>the</strong> KSA; Buraydah city is <strong>the</strong> largest city in Al Qassim<br />

Figure 1. According to <strong>the</strong> latest population census in 2004, <strong>the</strong> population <strong>of</strong> Buraydah city<br />

numbered 394.009 people, following <strong>the</strong> addition <strong>of</strong> new parts to <strong>the</strong> city. Additionally, according to<br />

this data it was estimated that Saudis comprise 85.5% <strong>of</strong> <strong>the</strong> total population <strong>of</strong> Buraydah. Non-<br />

Saudis make up <strong>the</strong> remaining percentage; <strong>the</strong>y can be described as residents (non-citizens) males and<br />

females <strong>of</strong> any o<strong>the</strong>r nationality.<br />

3.2 The data sources and network analysis<br />

Data sources used in this study are listed in Table 1:<br />

Table 1. The data sources<br />

Maps and Data format Produced by<br />

Road network maps Shape file (line) Al Qassim Municipality<br />

Neighbourhoods maps Shape file (polygon) Al Qassim Municipality<br />

PHCCs maps Shape file (point) Al Qassim Municipality<br />

Demographic data Excel file Ministry <strong>of</strong> Economic & Planning<br />

The total number <strong>of</strong> PHCCs in <strong>the</strong> city <strong>of</strong> Buraydah is 32, distributed throughout 82 neighbourhoods.<br />

This information is based on a network analysis in <strong>GIS</strong> data, collected using <strong>the</strong> following steps:<br />

� Building <strong>the</strong> road network dataset.<br />

� Creating <strong>the</strong> PHCCs‘ access points.<br />

� Creating Output Areas neighbourhoods‘ centroids.<br />

� Analysing <strong>the</strong> origin–destination (OD)<br />

� Using <strong>the</strong> results <strong>of</strong> OD analysis (Output Areas) linked with census polygons to analyse<br />

accessibility.<br />

Outputs <strong>of</strong> <strong>the</strong> previous steps in <strong>the</strong> form <strong>of</strong> tables were used in combination with a maximum<br />

distance applied <strong>of</strong> no more than 300 meters from <strong>the</strong> nearest PHCCs to <strong>the</strong> centre <strong>of</strong> <strong>the</strong><br />

neighbourhoods, or a centroid point. The choice <strong>of</strong> 300 meters as a measure for analysis <strong>of</strong><br />

accessibility relates to <strong>the</strong> small size <strong>of</strong> <strong>the</strong> most <strong>of</strong> neighbourhoods in <strong>the</strong> centre <strong>of</strong> <strong>the</strong> city <strong>of</strong><br />

Buraydah and to demonstrate <strong>the</strong> method.<br />

Page | 83


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

4. Results<br />

Figure 1. Buraydah city<br />

The results <strong>of</strong> applying a network analysis can be combined with Mosaic plots to detect <strong>the</strong><br />

relationship between accessibility or deprivation, for Saudis and non-Saudis with regards to PHCCs.<br />

This analysis uses <strong>the</strong> set <strong>of</strong> ‗5% most deprived‘ and ‗5% least deprived‘, where Hartigan and Kleiner<br />

(<strong>19</strong>81) have proposed <strong>the</strong> first Mosaics plots for contingency tables. The scale which is applied in this<br />

analysis specifies no more than 300 meters from <strong>the</strong> nearest PHCCs to <strong>the</strong> centres <strong>of</strong> <strong>the</strong><br />

neighbourhoods (Figure 2). Mosaic plots are a graph that can be used to study <strong>the</strong> relationship<br />

between two variables or more to clarify <strong>the</strong>ir relationship with o<strong>the</strong>r variables. In this study <strong>the</strong><br />

Mosaic plots showed <strong>the</strong> Saudis and non-Saudis in terms <strong>of</strong> access: False or True and <strong>the</strong> percentage.<br />

They can use this analysis as applied by Comber et al. (2008), seen below:<br />

Page | 84


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

Figure 2. Access to PHCCs by Ethnicity in Buraydah city<br />

Higher than average combinations <strong>of</strong> access and deprivation are represented by <strong>the</strong> blue tiles and<br />

those tiles shaded deep blue represent residuals greater than +4, and <strong>the</strong> red less than -4, upon<br />

comparison with a model <strong>of</strong> proportional access for deprivation.<br />

The Poisson regression model was applied through <strong>the</strong> following Equations; 1:<br />

E(c ij) � log( r � Ai<br />

� Fj<br />

)<br />

(1)<br />

Where: cij is Poisson distribution, r is an intercept term, Ai is <strong>the</strong> effect <strong>of</strong> <strong>the</strong> column and Fj is <strong>the</strong><br />

effect <strong>of</strong> <strong>the</strong> row. The comparison was made against <strong>the</strong> following model Equation 2:<br />

E( cij ) � log( r � Ai<br />

� Fj<br />

� Iij<br />

)<br />

(2)<br />

Where: Iij is <strong>the</strong> impact <strong>of</strong> <strong>the</strong> interaction between <strong>the</strong> columns and rows, when <strong>the</strong>re is a significant<br />

difference from zero; this indicates a degree <strong>of</strong> correlation between <strong>the</strong> effect <strong>of</strong> <strong>the</strong> column and <strong>the</strong><br />

row. In addition, testing <strong>the</strong> relationship between deprivation and access to PHCCs, among Saudis<br />

and non-Saudis can occur with <strong>the</strong> previous model and information detailing access to PHCCs from<br />

Table 2:<br />

Table 2. Access to PHCCs in Buraydah city<br />

Access non-Saudi Saudi<br />

False 11450 108060<br />

True 45645 228854<br />

This work used an R statistical s<strong>of</strong>tware package to estimate <strong>the</strong> value <strong>of</strong> Iij and <strong>the</strong> data for<br />

deprivation by applying Equation 2. The index <strong>of</strong> access for Saudis and Non-Saudis used <strong>the</strong><br />

following formula; Equation 3:<br />

(3)<br />

Each column category j is compared with a ‗reference‘ category; with 0 suggesting parity <strong>of</strong> access<br />

for category j and <strong>the</strong> reference category, +50 access is one-and-a-half times as likely and a value <strong>of</strong> -<br />

Page | 85


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

50 that it is half as likely, etc. The reference categories used were in <strong>the</strong> median 50% <strong>of</strong> <strong>the</strong> scores for<br />

deprivation.<br />

The results in Table 3 showed that Saudis have 46.9% less access to PHCCs than non-Saudis. This<br />

indicates that <strong>the</strong> largest proportion <strong>of</strong> <strong>the</strong> population in <strong>the</strong> city <strong>of</strong> Buraydah is about half as likely to<br />

have access to PHCCs, when compared to non-Saudis within a distance <strong>of</strong> no more than 300m.<br />

Table 3 Percentage access to PHCCs within no more than 300 meters in Buraydah city<br />

Ethnicity Estimate Access<br />

true<br />

Std. Error z value Pr(>|z|)<br />

Saudi -0.6325067 -46.8741 0.035026269 -3.055360264 0.002247903<br />

*Compared to <strong>the</strong> ‗non-Saudi‘.<br />

5. Discussion<br />

The findings <strong>of</strong> this study can be explained by <strong>the</strong> concentration <strong>of</strong> non-Saudis and PHCCs in <strong>the</strong><br />

older neighbourhoods that are located in<br />

<strong>the</strong> middle <strong>of</strong> <strong>the</strong> city; Figure 3. At <strong>the</strong><br />

present <strong>the</strong> Saudi population lives in <strong>the</strong><br />

new neighbourhoods in <strong>the</strong> east, north and<br />

west <strong>of</strong> <strong>the</strong> city. In <strong>the</strong> future data and<br />

analysis will be used to help to determine<br />

accessibility to PHCCs more accurately<br />

than has previously been possible in <strong>the</strong><br />

majority <strong>of</strong> <strong>the</strong> studies in <strong>the</strong> KSA. For<br />

example, future work will estimate <strong>the</strong><br />

population data by using areal<br />

interpolation techniques to estimate <strong>the</strong><br />

number <strong>of</strong> people (demand for PHCCs)<br />

living in <strong>the</strong> neighbourhoods to <strong>the</strong><br />

smallest unit area, i.e. in terms <strong>of</strong> blocks or<br />

parcels. In addition; future work will apply<br />

<strong>the</strong> new classifications <strong>of</strong> PHCCs, which<br />

will be implemented by <strong>the</strong> Ministry <strong>of</strong><br />

Health in 2011, and will <strong>the</strong>n determine<br />

accessibility, according to those<br />

classifications. These analyses can support<br />

<strong>the</strong> decision-makers in detecting <strong>the</strong><br />

accessibility and geographical distribution<br />

<strong>of</strong> health services. In addition, identify <strong>the</strong><br />

optimal spatial arrangement <strong>of</strong> health<br />

facilities that maximizes access to help<br />

planners to identifying <strong>the</strong> optimal location<br />

<strong>of</strong> health facilities according to <strong>the</strong><br />

geographical distribution <strong>of</strong> population in<br />

urban centres.<br />

Figure 3. PHCCs in Buraydah city<br />

6. Conclusions<br />

Planning for health services is important in many countries, and <strong>the</strong> majority <strong>of</strong> States make it a<br />

priority <strong>of</strong> any development plans. This work is to fill <strong>the</strong> gap in <strong>the</strong> literature pertaining to health<br />

service planning as has been identified in this study. In addition, this study is an important link to <strong>the</strong><br />

studies <strong>of</strong> health services in general and <strong>GIS</strong> applications in particular. Moreover, this work will<br />

support <strong>the</strong> decision-makers by providing different methods for planning health services in <strong>the</strong> KSA.<br />

Page | 86


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

7. Acknowledgements<br />

The authors would like to thank <strong>the</strong> Al Qassim Municipality, General Directorate <strong>of</strong> Health Affairs in<br />

Al Qassim and <strong>the</strong> Ministry <strong>of</strong> Economy and Planning for providing both <strong>the</strong> data and <strong>the</strong> maps.<br />

8. References<br />

Al Dlghan Abdulaziz (<strong>19</strong>92). The spatial dimensions <strong>of</strong> health care centers services in <strong>the</strong> city <strong>of</strong> Riyadh.<br />

Master Thesis, Department <strong>of</strong> Geography, King Saud University, Riyadh.<br />

Al Dossari Hamad (2009). Primary health care accessibility in Riyadh. Master Thesis, Department <strong>of</strong><br />

Geography, University <strong>of</strong> Leicester.<br />

Al Ghamdi Abdulaziz (<strong>19</strong>81). An Approach to Planning a Primary Health Care Delivery System in Jeddah,<br />

Saudi Arabia. Ph.D. Thesis, Michigan State University, U.S.A.<br />

Al Qassim Municipality (2010). Digital maps <strong>of</strong> AL Qassim province. Department <strong>of</strong> Urban Planning, Al<br />

Qassim.<br />

Al Ribdi Mohammed (<strong>19</strong>90). The Geography <strong>of</strong> Health Care in Saudi Arabia: Provision and Use <strong>of</strong> Primary<br />

Health Facilities in Al Qassim province. Ph.D. Thesis, University <strong>of</strong> Southampton, U.K.<br />

Al Shahrani H (2004). The Accessibility and Utilization <strong>of</strong> Primary Health Care Services in Riyadh, Kingdom <strong>of</strong><br />

Saudi Arabia. Ph.D. Thesis, University <strong>of</strong> East Anglia, Norwich.<br />

Bagheri N Benwell G L and Holt A (2006). Primary health care accessibility for Rural Otago: A Spatial<br />

Analysis. Health Care and Informatics Review OnlineTM, Retrieved September 3, 2006, [Online]. Available<br />

from: http://hcro.enigma.co.nz/website/index.cfm?fuseaction=articledisplay&FeatureID=010906.pdf.<br />

[Accessed 18/10/09].<br />

Comber A Brunsdon C and Green E (2008). Using a <strong>GIS</strong>-based network analysis to determine urban greenspace<br />

accessibility for different ethnic and religious groups. Landscape and Urban Planning, 86,103–114.<br />

General Directorate <strong>of</strong> Health Affairs in Al Qassim (2010). Unpublished data for <strong>the</strong> classification <strong>of</strong> primary<br />

health care centers. Department <strong>of</strong> health centers, Al Qassim.<br />

Hartigan J and Kleiner B (<strong>19</strong>81). Mosaics for contingency tables. In W. F. Eddy (Ed.). Computer Science and<br />

Statistics: <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> 13th Symposium on <strong>the</strong> Interface. New York: Springer-Verlag.<br />

Haynes R Jones A P Sauerzapf V and Zhao H (2006). Validation <strong>of</strong> travel times to hospital estimated by <strong>GIS</strong>,<br />

International Journal <strong>of</strong> Health Geographics, 5, 40.<br />

Langford M Higgs G Radcliffe J and White S D (2007). Urban Population Distribution Models And<br />

Service Accessibility Estimation, Computers, Environment and Urban Systems, 32,(1), 66-80.<br />

Luo W and Wang F (2003). Measures <strong>of</strong> spatial accessibility to health care in a <strong>GIS</strong> environment: syn<strong>the</strong>sis and<br />

a case study in <strong>the</strong> Chicago region. Environment and Planning B: Planning and Design, 30,( 6), 865-884.<br />

Martin D Roderick P Diamond I Clements S and Stone N (<strong>19</strong>98) Geographical aspects <strong>of</strong> <strong>the</strong> uptake <strong>of</strong> renal<br />

replacement <strong>the</strong>rapy in England. Int. J. Popul. Geogr, 4, 227–242.<br />

Ministry <strong>of</strong> Economy & Planning (2004). Population <strong>of</strong> Kingdom <strong>of</strong> Saudi Arabia in 2004. Central Department<br />

<strong>of</strong> Statistics & Information. Riyadh.<br />

Møller-Jensen Lasse and K<strong>of</strong>ie Richard (2001). Exploiting available data sources: location/allocation modeling<br />

for health service planning in rural Ghana. Danish Journal <strong>of</strong> Geography, (101), 145-153.<br />

Rosero-Bixby L (2004). Spatial access to health care in Costa Rica and its equity: A <strong>GIS</strong>-based study. Social<br />

Science and Medicine, (58), 1271-1284.<br />

Schuurman N Fiedler Robert Grzybowski Stefan and Grund Darrin (2006) Defining rational hospital<br />

catchments for non-urban areas based on travel-time, International Journal <strong>of</strong> Health Geographics, 5:43.<br />

Tanser F C Gijsbertsen B and Herbst A (2006). Modeling and understanding primary health care<br />

accessibility and utilization in rural KwaZulu‐Natal South Africa: an exploration using a geographical<br />

information system. Soc Sci Med, (63), 691–705.<br />

Teixeira Joao C and Antunes Antonio P (2008). A hierarchical location model for public facility planning.<br />

European Journal <strong>of</strong> Operational <strong>Research</strong>. (185), 92–104.<br />

Walsh S J Page P H and Gesler W M (<strong>19</strong>97). Normative Models and Healthcare Planning: Network-Based<br />

Simulations Within a Geographic Information System Environment. Health Services <strong>Research</strong>, 32, 243-260.<br />

World Health Organization (2008). A Summary <strong>of</strong> <strong>the</strong> 2008 World Health Report ―Primary Health Care: Now<br />

More Than Ever‖. [Online]. Available from: http://www.who.int/whr/2008/summary.pdf. [Accessed 28/11/09].<br />

9. Biography<br />

Ibrahim Alshwesh 1 : PhD student in Geographical Information Science, Department <strong>of</strong> Geography, University<br />

<strong>of</strong> Leicester. <strong>Research</strong> interests in <strong>the</strong> analysis <strong>of</strong> <strong>the</strong> spatial distribution <strong>of</strong> health facilities and spatial analysis<br />

Page | 87


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2a: Health Geography and <strong>GIS</strong><br />

<strong>of</strong> policy and planning.<br />

Alexis Comber 2 : Senior Lecturer in Geographic Information, Department <strong>of</strong> Geography, University <strong>of</strong><br />

Leicester. <strong>Research</strong> interests in two primary areas: issues associated with uncertainty and representation in<br />

spatial data and <strong>the</strong> use <strong>of</strong> spatial analyses to evaluate policy.<br />

Chris Brunsdon 3 : Pr<strong>of</strong>essor <strong>of</strong> Geographic Information, Department <strong>of</strong> Geography, University <strong>of</strong> Leicester.<br />

<strong>Research</strong> interests include <strong>the</strong> methodologies underlying spatial statistical analysis and geographical<br />

information systems, and <strong>the</strong>ir application in a number <strong>of</strong> subject areas.<br />

Page | 88


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

3D Urban Visibility Analysis with Vector <strong>GIS</strong> Data<br />

Suleiman Wassim 1 , Joliveau Thierry 1 , Favier Eric 2<br />

1 CRENAM-ISIG CNRS/UMR 5600, Université Jean Monnet - Saint-Etienne.<br />

E-mail: {wassim.suleiman, thierry.joliveau}@univ-st-etienne.fr<br />

2 DIPI EA 37<strong>19</strong> ENISE (École Nationale d'Ingénieurs de Saint-Etienne). E-mail:<br />

eric.favier@enise.fr<br />

ABSTRACT<br />

The visibility estimation has an important impact in many economical and aes<strong>the</strong>tic fields. A mixed<br />

environment which contains artificial objects like buildings laying on a natural ground surface is a<br />

challenge for visibility calculation. This paper presents a new method to solve this problem, based on<br />

vector <strong>GIS</strong> data. The use <strong>of</strong> vector data gives <strong>the</strong> possibility <strong>of</strong> calculating <strong>the</strong> intervisibility indices as<br />

well as delineating a viewshed in a mixed environment. The new method could identify obstacles<br />

(relief, buildings) that may block <strong>the</strong> visibility for an observer situated in any point <strong>of</strong> a 3D<br />

environment. The intervisibility impact <strong>of</strong> a specific building can also be calculated.<br />

KEYWORDS: 3D environment, field <strong>of</strong> vision, visibility, viewshed simulation, urban visibility<br />

analysis, <strong>GIS</strong>.<br />

1. Introduction<br />

The economic and aes<strong>the</strong>tic value <strong>of</strong> buildings is highly related to <strong>the</strong> visibility field that it could <strong>of</strong>fer<br />

to its citizens (Miller D. 2001). People prefer <strong>the</strong> locations that <strong>of</strong>fer large views on green space (Jim<br />

& Chen 2006), water (Luttik 2000) and forests (Gueymard 2006). Hence it is very important to choose<br />

<strong>the</strong> position which gives a new building a good and pleasant sight. On <strong>the</strong> o<strong>the</strong>r hand, adding a new<br />

building at a specific position could disturb <strong>the</strong> intervisibility <strong>of</strong> <strong>the</strong> whole area, while locating <strong>the</strong><br />

same building in a different position may not.<br />

Isovist and viewshed are utilized in view assessments, but <strong>the</strong>y differ in <strong>the</strong>ir methodologies and <strong>the</strong>ir<br />

applications (Sander & Manson 2007). The isovist term is used generally in built environment with<br />

vector data, to calculate <strong>the</strong> subset points <strong>of</strong> space that could be seen from a vantage point (Benedict<br />

<strong>19</strong>79). A visibility graph can be calculated with different characteristics, like view quality and<br />

perimeter compactness (Turner et al. 2001). The viewshed term is more commonly used to calculate<br />

<strong>the</strong> intervisibility in a rural environment, with raster data, (Floriani & Magillo 2003).<br />

Isovist and viewshed both face some difficulties when dealing with 3D environment composed <strong>of</strong><br />

man-made objects like buildings laying on a non-flat topographical surface. They fail in addressing<br />

some questions like finding <strong>the</strong> visible facade <strong>of</strong> buildings from a vantage point or finding <strong>the</strong><br />

obstacles (buildings, relief) that block <strong>the</strong> intervisibility between two points. This challenging<br />

question is very important for many applications like positioning radio antennas, wifi bornes,<br />

surveillance cameras, advertisement posters, etc..<br />

This article is organized in this way: <strong>the</strong> second part presents <strong>the</strong> related work which explores <strong>the</strong><br />

methods <strong>of</strong> viewshed calculation; <strong>the</strong> third part presents a new method for calculating intervisibility<br />

and viewshed in 3D vector data environment; <strong>the</strong> forth section is dedicated to an evaluation <strong>of</strong> our<br />

work and results; <strong>the</strong> fifth and last section indicates <strong>the</strong> future work necessary to extend <strong>the</strong> proposed<br />

method.<br />

Page | 89


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

2. Related works<br />

2.1 2D Isovist<br />

The use <strong>of</strong> vector data for outdoor environment visibility is mentioned by (Rana 2006) who developed<br />

a program using Arc<strong>GIS</strong> s<strong>of</strong>tware from ESRI. The program is based on a ray tracing algorithm<br />

(Figure 1) that computes a visibility polygon which encloses <strong>the</strong> visible area.<br />

Figure 1 : Tracing ray algorithm<br />

The proposed s<strong>of</strong>tware works on vector data building information. A later application was developed<br />

by (Joliveau & Rana 2008); <strong>the</strong> viewshed was tested in Lyon (France) by coupling <strong>the</strong> isovist analysis<br />

to a database <strong>of</strong> georeferenced photos. The s<strong>of</strong>tware is limited to 2D environment and cannot deal<br />

with non flat surfaces nor with 3D building.<br />

2.2 3D Isovist<br />

Ano<strong>the</strong>r idea is to integrate <strong>the</strong> buildings in <strong>the</strong> terrain, by adding <strong>the</strong> building height to <strong>the</strong> DEM<br />

(Digital Elevation Model) information. This operation creates what is <strong>of</strong>ten called a 2.5 D<br />

environment. The visibility could <strong>the</strong>n be calculated in raster mode using lines <strong>of</strong> sight and ray tracing<br />

(Lake et al. 2000) (Brossard et al. 2008) (Figure 2).<br />

Figure 2: Visibility in raster mode with regular pixelisation (Source Brossard & Wieber)<br />

A regular pixelisation grid is used to create a raster image, and <strong>the</strong>n a tracing ray algorithm is applied<br />

to test <strong>the</strong> intervisibility between <strong>the</strong> vantage point (observer point) and <strong>the</strong> o<strong>the</strong>r pixels. As it is<br />

shown in <strong>the</strong> Figure 2, a big resolution image needs a considerable calculation time and a low<br />

resolution image will lead to imprecise results. A later idea was to use high resolution in a limited area<br />

around <strong>the</strong> vantage point and a low resolution in <strong>the</strong> o<strong>the</strong>r part <strong>of</strong> <strong>the</strong> image (Figure 3).<br />

Page | 90


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

Figure 3: Visibility in raster mode with variable pixelisation. (Source Brossard &<br />

Wieber)<br />

With this method, <strong>the</strong> building identification is lost. Moreover it is not easy to answer a series <strong>of</strong> very<br />

interesting real life questions. What is <strong>the</strong> general intervisibility impact <strong>of</strong> a specific building? What is<br />

<strong>the</strong> hidden area from a vantage point caused by a specific building? What are <strong>the</strong> obstacles (building,<br />

hills, banks…) that block <strong>the</strong> visibility <strong>of</strong> a specific point in <strong>the</strong> space from a vantage point? Or, more<br />

simply, what can we see from <strong>the</strong> third floor <strong>of</strong> a specific building facade?<br />

O<strong>the</strong>r researches were developed in a 3D-isovist-like approach (Pyysalo et al. 2009) (Figure 4)<br />

(Morello & Ratti 2009) (Figure 5). The basic idea is to use voxel models 2 with adaptive transparency<br />

to add buildings and vegetation <strong>the</strong>n to calculate <strong>the</strong> 3D isovist. This kind <strong>of</strong> calculation needs 3D<br />

laser scanner data.<br />

Figure 4: Voxel model (Pyysalo et al. 2009)<br />

Figure 5: Voxel model (Morello & Ratti 2009)<br />

The 3D isovist calculation is based on a ray tracing idea to scan <strong>the</strong> complete space. To obtain a good<br />

precision <strong>of</strong> <strong>the</strong> 3D isovist, it is necessary to execute a very complicated calculation process<br />

(360*180= 64800 tracing rays).<br />

In <strong>the</strong> next section we will propose a new a method to calculate <strong>the</strong> intervisibility based on<br />

considering <strong>the</strong> 3D environment as a constellation <strong>of</strong> polygons.<br />

2 http://en.wikipedia.org/wiki/Voxel<br />

Page | 91


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

3. Intervisibility and Viewshed in 3D vector environment<br />

3.1 The 3D environment<br />

The main idea in this article is to treat <strong>the</strong> <strong>GIS</strong> data (terrain, buildings) as a constellation <strong>of</strong> 3D<br />

polygons. To obtain <strong>the</strong> terrain model a TIN (Triangular Irregular Network) derived from Delaunay<br />

triangulation (Yan & Lianhe 2009) (Figure 6) was built.<br />

Figure 6: Delaunay triangulation<br />

The terrain is modelled as an aggregation <strong>of</strong> 3D plane triangular surfaces; every triangle has a 3D<br />

position precisely defined.<br />

To obtain <strong>the</strong> buildings model,<br />

classical 2D footprints with a height<br />

extrusion value associated were used.<br />

In this kind <strong>of</strong> data, <strong>the</strong> buildings‘<br />

walls and ro<strong>of</strong>s can be regarded as a<br />

set <strong>of</strong> addressed plan surfaces. These<br />

surfaces have a height <strong>of</strong>fset from <strong>the</strong><br />

correspondent level <strong>of</strong> footprints<br />

building.<br />

Figure 7: The 3D environment<br />

In this kind <strong>of</strong> model, <strong>the</strong> 3D world becomes a list <strong>of</strong> polygon plane facets; each <strong>of</strong> <strong>the</strong>se facets has a<br />

type (terrain, building) and an identification number that can refer to a triangle Id <strong>of</strong> <strong>the</strong> TIN or a<br />

building Id combined with a wall (facade) or a ro<strong>of</strong> number.<br />

The Error! Reference source not found. presents an extraction <strong>of</strong> a 3D environment concerning a<br />

art <strong>of</strong> Saint-Etienne city.To achieve <strong>the</strong> conceptual test, we wrote a program using Matlab version 7.<br />

The mapping extension was used, because it can deal with <strong>GIS</strong> data (import, export) and it contains<br />

many useful ma<strong>the</strong>matical functions and modelling facilities that are not available in <strong>the</strong> usual <strong>GIS</strong><br />

tools.<br />

3.2 The intervisibility calculation<br />

A grid <strong>of</strong> points is needed to calculate <strong>the</strong> intervisibility between <strong>the</strong>m and to present <strong>the</strong> results on a<br />

map. The grid gives us a new way to delineate a viewshed by calculating for each point <strong>of</strong> <strong>the</strong> grid if it<br />

is visible or not visible from <strong>the</strong> vantage point. An interpolation <strong>of</strong> <strong>the</strong> results is realised between <strong>the</strong><br />

points <strong>of</strong> <strong>the</strong> grid in order to produce a continuous representation. The complexity <strong>of</strong> this algorithm is<br />

(Number <strong>of</strong> grid points * Number <strong>of</strong> polygons) operations <strong>of</strong> (segment/polygon intersections) in <strong>the</strong><br />

3D environment. On <strong>the</strong> o<strong>the</strong>r hand, <strong>the</strong> complexity <strong>of</strong> <strong>the</strong> ray tracing technique is (Number <strong>of</strong> rays *<br />

Number <strong>of</strong> polygons) operations <strong>of</strong> (line/polygon intersections) in 3D environment. The accuracy <strong>of</strong><br />

<strong>the</strong> calculation <strong>of</strong> our method is based on <strong>the</strong> number <strong>of</strong> <strong>the</strong> points while it is based on <strong>the</strong> number <strong>of</strong><br />

<strong>the</strong> tracing rays in <strong>the</strong> ray tracing method.<br />

Page | 92


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

In this approach, Intervisibility between two points is considered reciprocal. In our algorithm two<br />

points can see each o<strong>the</strong>r if <strong>the</strong> segment between <strong>the</strong>m is not intersected with any surface plane<br />

(polygon) <strong>of</strong> <strong>the</strong> 3D environment. That means that if <strong>the</strong> point A sees <strong>the</strong> point B, <strong>the</strong>n <strong>the</strong> point B can<br />

see <strong>the</strong> point A.<br />

Polygon/segment intersection algorithm in 3D environment is based on <strong>the</strong> calculation <strong>of</strong> <strong>the</strong><br />

intersection point between <strong>the</strong> 3D line which lies on <strong>the</strong> segment, and <strong>the</strong> 3D plan which lies on <strong>the</strong><br />

3D polygon. After finding this point, we check if <strong>the</strong> intersection point is in <strong>the</strong> polygon as 2D point<br />

and 2D polygon, because <strong>the</strong> intersection point is on <strong>the</strong> polygon plan. Finally we check if this<br />

intersection point is located between <strong>the</strong> two ends <strong>of</strong> <strong>the</strong> segment. If all <strong>the</strong>se conditions are verified,<br />

we assume that <strong>the</strong> 3D polygon and <strong>the</strong> 3D segment intersect, and <strong>the</strong>re is no intervisibility between<br />

<strong>the</strong> extremities <strong>of</strong> <strong>the</strong> segment.<br />

In <strong>the</strong> test presented in this paper, <strong>the</strong> height <strong>of</strong> every point <strong>of</strong> <strong>the</strong> grid is set to 1.60 metre above <strong>the</strong><br />

ground level, which corresponds to an average human height. This consideration could be changed to<br />

meet <strong>the</strong> application need. For example, <strong>the</strong> observer points could be at 1.60 metre where o<strong>the</strong>r grid<br />

points are at 0 metre if <strong>the</strong> observer is a human that looks at an object located on <strong>the</strong> ground, or 2.5<br />

metres if <strong>the</strong> person looks at an advertisement poster for example. In this calculation, a percentage <strong>of</strong><br />

<strong>the</strong> visibility or a visibility coefficient can be assigned to every point <strong>of</strong> <strong>the</strong> grid (Turner et al. 2001)<br />

with <strong>the</strong> identification <strong>of</strong> points that could be seen from <strong>the</strong> considered point (Figure 8). The result is<br />

exported as a shape file <strong>of</strong> points.<br />

Page | 93


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

Figure 8: The intervisibility interpolated map using Inverse Distance Weighting (IDW),<br />

VisPersnetN is <strong>the</strong> visibility percentage, VisPointsIDC is <strong>the</strong> visible points IDs from <strong>the</strong><br />

according point<br />

Through an option <strong>of</strong> <strong>the</strong> program, <strong>the</strong> user can adjust <strong>the</strong> role <strong>of</strong> <strong>the</strong> points <strong>of</strong> <strong>the</strong> grid that are located<br />

within <strong>the</strong> buildings polygons. He can consider that those points do not interfere with <strong>the</strong><br />

intervisibility or he can regard <strong>the</strong>m as points located on <strong>the</strong> ro<strong>of</strong> <strong>of</strong> a building. In this case <strong>the</strong>y will<br />

affect <strong>the</strong> intervisibility calculation.<br />

As it was shown before, <strong>the</strong> Intervisibility in this 3D environment is reciprocal. It can be used to find<br />

<strong>the</strong> best place for an advertisement poster located at a height <strong>of</strong> 4 metres to be viewed by a person who<br />

is 1.60 tall. But <strong>the</strong> computation works also <strong>the</strong> o<strong>the</strong>r way with <strong>the</strong> observer located at a height <strong>of</strong> 4<br />

metres when <strong>the</strong> grid points are at a height <strong>of</strong> 1.60 metres.<br />

The values <strong>of</strong> <strong>the</strong> points on <strong>the</strong> border <strong>of</strong> <strong>the</strong> grid are biased by <strong>the</strong>ir location. They could be highly<br />

visible but from points situated outside <strong>the</strong> region.<br />

3.2 The viewshed calculation<br />

The viewshed can be defined as <strong>the</strong> set <strong>of</strong> environment points which are visible from <strong>the</strong> observer<br />

point. The method supports <strong>the</strong> possibility to associate to each point <strong>of</strong> <strong>the</strong> grid <strong>the</strong> obstacles which<br />

block <strong>the</strong> visibility from <strong>the</strong> observer point. To achieve <strong>the</strong> calculation, we need to identify <strong>the</strong><br />

observer point and its height above <strong>the</strong> ground level, and to identify <strong>the</strong> grid points (position, and<br />

height). The result is a shape file <strong>of</strong> points with a visibility data field that indicates if <strong>the</strong> grid point is<br />

visible or not. An obstacle data field lists <strong>the</strong> parts <strong>of</strong> <strong>the</strong> terrain or <strong>the</strong> buildings that block <strong>the</strong> view<br />

from <strong>the</strong> observer position (Figure 9).<br />

Page | 94


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

Figure 9: Viewshed interpolated map using Inverse Distance Weighting (IDW), with<br />

obstacles extractions, in <strong>the</strong> obstacle field <strong>the</strong> number 0 is <strong>the</strong> earth formation, o<strong>the</strong>r<br />

numbers present <strong>the</strong> building identification<br />

3.3 Building intervisibility effect<br />

By using this method, we have also <strong>the</strong> possibility to calculate <strong>the</strong> intervisibility effect <strong>of</strong> a specific<br />

building on <strong>the</strong> 3D environment. The process is simple: <strong>the</strong> target building is selected and <strong>the</strong><br />

intervisibility effect <strong>of</strong> this building is calculated. The result is a shape file <strong>of</strong> <strong>the</strong> grid intervisibility<br />

points. A new attribute ―Affected‖ is created in <strong>the</strong> table who indicates if <strong>the</strong> target building has or not<br />

an impact. If some points are affected, ano<strong>the</strong>r attribute lists <strong>the</strong> hidden points due to <strong>the</strong> target<br />

building (Figure 10).<br />

Page | 95


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

Figure 10: The building effect on intervisibility interpolated map using Inverse<br />

Distance Weighting (IDW), <strong>the</strong> affected field give if <strong>the</strong> point affected from <strong>the</strong><br />

presence <strong>of</strong> <strong>the</strong> red building, <strong>the</strong> HiddenPointsC give <strong>the</strong> hidden point for <strong>the</strong><br />

according point because <strong>of</strong> <strong>the</strong> red building<br />

4. Results<br />

In our work we use Matlab 7 simulation tools with mapping toolbox to load <strong>the</strong> data from <strong>the</strong> DEM<br />

files and <strong>the</strong> building shape file, to create <strong>the</strong> 3D environment, and to calculate <strong>the</strong> intervisibility, <strong>the</strong><br />

viewshed and <strong>the</strong> building intervisibility effect. Then we export <strong>the</strong> result as a shape file <strong>of</strong> points<br />

with attributes that stock <strong>the</strong> visibility percentage, <strong>the</strong> visible points, <strong>the</strong> hidden points and <strong>the</strong><br />

obstacles. This file can be opened in classical <strong>GIS</strong> tools.<br />

This method gives us a new opportunity to visualise <strong>the</strong> effect <strong>of</strong> <strong>the</strong> building layer on <strong>the</strong><br />

intervisibility and viewshed. And <strong>the</strong> result could be exported as a shape file to be accessible for all<br />

<strong>GIS</strong> user.<br />

The calculation process depends on two elements:<br />

� The number <strong>of</strong> polygons in <strong>the</strong> 3D environments P (building faced or triangle tessellation).<br />

� The number <strong>of</strong> <strong>the</strong> intervisibility grid points N.<br />

The complication <strong>of</strong> <strong>the</strong> viewshed process method is <strong>of</strong> <strong>the</strong> order N*P intersection segment-3D<br />

polygon. On <strong>the</strong> o<strong>the</strong>r side <strong>the</strong> complexity <strong>of</strong> <strong>the</strong> intervisibility map is <strong>of</strong> <strong>the</strong> order <strong>of</strong> N*(N*P), so it is<br />

very important for <strong>the</strong> intervisibility map to reduce N, <strong>the</strong> number <strong>of</strong> points in order to compute <strong>the</strong><br />

intervisibility grid.<br />

As shown above, <strong>the</strong> number <strong>of</strong> polygons in <strong>the</strong> 3D environment plays an essential role in <strong>the</strong><br />

complexity <strong>of</strong> <strong>the</strong> intervisibility calculation. So we need to reduce <strong>the</strong> number <strong>of</strong> polygons as much as<br />

possible with reserving <strong>the</strong> main forms <strong>of</strong> <strong>the</strong> 3D environment. This operation is not obligatory to<br />

achieve <strong>the</strong> calculation but it significantly accelerates <strong>the</strong> process.<br />

4.1 Simplifying <strong>the</strong> DEM<br />

There are many methods in <strong>the</strong> literature for simplifying Mesh triangle compression (Deering <strong>19</strong>95)<br />

(Touma & Gotsman <strong>19</strong>98). The DEM in our test was simplified by grouping <strong>the</strong> points which had <strong>the</strong><br />

Page | 96


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

same level (IsoLevel) within 1 metre height interval. Then <strong>the</strong> extra points were eliminated. The idea<br />

here is that <strong>the</strong>re is no need <strong>of</strong> 100 DEM points to describe a flat surface. The interval height could be<br />

adapted regarding <strong>the</strong> user needs (Figure 11).<br />

a) DEM b) DEM with IsoLevel simplification<br />

Figure 11: Simplifying <strong>the</strong> DEM<br />

4.2 Simplifying <strong>the</strong> buildings polygon<br />

In our test for modelling <strong>the</strong> polygon buildings using Matlab 7, <strong>the</strong> test shows that for one real<br />

segment <strong>of</strong> <strong>the</strong> building (facade) <strong>the</strong>re are <strong>of</strong>ten a lot <strong>of</strong> vertices between two nodes. Actually this is<br />

due to <strong>the</strong> digitalisation process. The simplifying <strong>of</strong> <strong>the</strong>ses polygons is addressed in <strong>the</strong> literature as<br />

<strong>the</strong> model generalisation question (Favier <strong>19</strong>94) (Anne Ruas <strong>19</strong>99) (Qingsheng et al. 2002). In order<br />

to simplify <strong>the</strong> polygons <strong>of</strong> <strong>the</strong> buildings, extra collinear points are eliminated. Then <strong>the</strong> points that<br />

form a facade arc smaller than 2 metre long were removed from <strong>the</strong> polygon. The user can modify this<br />

value (<br />

Figure 12).<br />

a) Original building polygon b) Simplified building polygon<br />

Figure 12: Simplifying <strong>the</strong> buildings polygon<br />

In <strong>the</strong> test, an extraction <strong>of</strong> map was taken with 88 intervisibility grid points and 564 3D polygons<br />

(424 building facades for 63 polygon buildings, 140 triangle tessellations for TIN). The method was<br />

applied using Matlab 7 on a AMD Athlon 30 Dual core 4800+ PC. Calculation time was 7 min. With<br />

<strong>the</strong> same files without simplification <strong>the</strong> calculation time was 16 min.<br />

5. Future work<br />

A future extension <strong>of</strong> this work is to find a way to add a vegetation layer which is not considered here.<br />

Ano<strong>the</strong>r dimension not considered in <strong>the</strong> current calculation is <strong>the</strong> nature <strong>of</strong> <strong>the</strong> building facades. A<br />

grid <strong>of</strong> points could be added on <strong>the</strong> facade to figure <strong>the</strong> windows for example. Those points would be<br />

added to <strong>the</strong> observery points. The application was made using Matlab environment that is useful to<br />

apply <strong>the</strong> idea <strong>of</strong> combining a TIN Model for <strong>the</strong> terrain and a 3D models for <strong>the</strong> buildings in a unique<br />

constellation <strong>of</strong> 3D polygons. The 3D models used in classical <strong>GIS</strong> s<strong>of</strong>twares don‘t permit yet easily<br />

Page | 97


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

this kind <strong>of</strong> implementation. To be accessible to all geographic users, it is necessary to dispose <strong>of</strong> 3D<br />

<strong>GIS</strong> data organized according to this conceptual idea. CityGML is a step forward in that direction for<br />

<strong>the</strong> building part <strong>of</strong> <strong>the</strong> question.<br />

6. References<br />

Anne Ruas, <strong>19</strong>99. Modèle de généralisation de données urbaines à base de contraintes et d´autonomie.<br />

Benedict, M.L., <strong>19</strong>79. To Take Hold <strong>of</strong> Space: Isovists and Isovists Fields. Environment and<br />

Planning. , pp.6, 47–65.<br />

Brossard, T., Joly, D. & Tourneux, F., 2008. Modélisation opérationnelle du paysage. In Paysage et<br />

information géographique. Lavoisier, pp. 117-137.<br />

Deering, M., <strong>19</strong>95. Geometry compression. In <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> 22nd annual conference on<br />

Computer graphics and interactive techniques. ACM, pp. 13-20. Available at:<br />

http://portal.acm.org/citation.cfm?id=218391.<br />

Favier, E., <strong>19</strong>94. Contribution de l'analyse multi-résolution à la description des contours et des<br />

textures, l'Université de Saint-Etienne.<br />

Floriani, L.D. & Magillo, P., 2003. Algorithms for visibility computation on terrains: a survey.<br />

Environment and Planning B: Planning and Design, 30(5), pp.709 – 728.<br />

Gueymard, S., 2006. Facteurs environnementaux de proximité et choix résidentiels. Développement<br />

durable et territoires. Available at: http://developpementdurable.revues.org/index2716.html.<br />

Jim, C. & Chen, W.Y., 2006. Impacts <strong>of</strong> urban environmental elements on residential housing prices<br />

in Guangzhou (China). Landscape and Urban Planning, 78(4), pp.422-434.<br />

Joliveau & Rana, 2008. Coupler photographies et information géographique pour décrire l‘espace<br />

visible. In SAGEO. Montpellier, p. 15P. Available at: http://sdhsageo.teledetection.fr/index.php?option=com_docman&task=doc_download&gid=12&Itemid=35.<br />

Lake, I.R. et al., 2000. Using <strong>GIS</strong> - and large-scale digital data to implement hedonic pricing studies.<br />

International Journal <strong>of</strong> Geographical Information Science, 14(6), p.521.<br />

Luttik, J., 2000. The value <strong>of</strong> trees, water and open space as reflected by house prices in <strong>the</strong><br />

Ne<strong>the</strong>rlands. Landscape and Urban Planning, 48(3-4), pp.161-167.<br />

Miller D., 2001. A method for estimating changes in <strong>the</strong> visibility <strong>of</strong> land cover. Landscape and<br />

Urban Planning, 54, pp.93-106.<br />

Morello, E. & Ratti, C., 2009. A digital image <strong>of</strong> <strong>the</strong> city: 3D isovists in Lynch‘s urban analysis.<br />

Environment and Planning B: Planning and Design, 36(5), pp.837 – 853.<br />

Pyysalo, U., Oksanen, J. & Sarjakoski, T., 2009. Viewshed analysis and visualization <strong>of</strong> landscape<br />

voxel models. 24th International Cartographic <strong>Conference</strong>, Santiago, Chile.<br />

Qingsheng, G., Brandenberger, C. & Hurni, L., 2002. A progressive line simplification algorithm.<br />

Geo-spatial Information Science, 5(3), pp.41-45.<br />

Rana, S., 2006. Isovist analyst: an arcview extension for planning visual surveillance. Available at:<br />

http://eprints.ucl.ac.uk/2104/.<br />

Sander, H.A. & Manson, S.M., 2007. Heights and locations <strong>of</strong> artificial structures in viewshed<br />

calculation: How close is close enough? Landscape and Urban Planning, 82(4), pp.257-270.<br />

Touma, C. & Gotsman, C., <strong>19</strong>98. Triangle Mesh Compression. Graphics Interface 98 <strong>Conference</strong><br />

<strong>Proceedings</strong>, pp.26–34.<br />

Turner, A. et al., 2001. From isovists to visibility graphs: a methodology for <strong>the</strong> analysis <strong>of</strong><br />

architectural space. Environment and Planning B: Planning and Design, 28(1), pp.103-121.<br />

Yan, L. & Lianhe, Y., 2009. Based on Delaunay Triangulation DEM <strong>of</strong> Terrain Model. Available at:<br />

http://ccsenet.org/journal/index.php/cis/article/view/1810.<br />

Page | 98


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

A modified two-step floating catchment area technique for measuring<br />

transit system accessibility<br />

Mitch Langford, Richard Fry, Gary Higgs<br />

<strong>GIS</strong> <strong>Research</strong> Centre and Wales Institute <strong>of</strong> Social and Economic <strong>Research</strong>, Data and<br />

Methods, Faculty <strong>of</strong> Advanced Technology, University <strong>of</strong> Glamorgan, Pontypridd, Wales,<br />

CF37 1DL. Tel. (01443 654146)<br />

mlangfor@glam.ac.uk<br />

ABSTRACT<br />

Measuring accessibility to public transit is important for planning and decision-making purposes.<br />

Existing measures lack sophistication and rely on simple buffering and overlay operations. We<br />

introduce a new metric based on <strong>the</strong> ‗floating catchment area‘ technique. After suitable modification to<br />

reflect <strong>the</strong> needs <strong>of</strong> a transit service accessibility measure, this metric captures aspects <strong>of</strong> proximity, <strong>the</strong><br />

balance between service supply and demand, cumulative opportunity, and temporal availability. We<br />

illustrate its application through a case study in South Wales, integrating publicly available digital<br />

transport timetables with bus stop locations, to create a realistic appraisal <strong>of</strong> bus transit accessibility.<br />

1. Introduction<br />

KEYWORDS: Accessibility, public transport, floating catchment area<br />

A well designed transportation system is essential for economic development and social interaction.<br />

Adequate transit provision can reduce social exclusion and geographic isolation (Preston and Raje,<br />

2007), and for households with limited financial resources public transit provides access to essential<br />

services to help meet social, health, education and recreational needs (Church et al., 2000; Witten et<br />

al. 2003; Currie et al., 2009).<br />

This paper is concerned with <strong>the</strong> measurement <strong>of</strong> bus transit system accessibility, and specifically<br />

access to a transport system, ra<strong>the</strong>r than access by a transport system. The capability to monitor,<br />

evaluate and model transit system performance is essential both for policy development and to ensure<br />

best operational practise. Accessibility has been defined as ‗<strong>the</strong> ease with which people can reach<br />

opportunities and services‘ (Lei and Church, 2010; p. 284), but no clear consensus exists on how to<br />

measure it. Thill and Kim (2005) identify four main approaches: distance, gravity, cumulativeopportunity,<br />

and space-time. This paper develops a new technique for measuring transit system<br />

accessibility based on <strong>the</strong> floating catchment area methodology which is popular within <strong>the</strong> field <strong>of</strong><br />

healthcare geography. The limitations <strong>of</strong> simple population-to-provider ratios recorded within fixed<br />

geographical zones, or alternative proximity measures to nearest service provision points have been<br />

widely recognised (e.g. Higgs, 2004). The two-step floating catchment area (2SFCA) methodology<br />

accounts for potential interaction between patients and physicians across administrative boundaries by<br />

evaluating <strong>the</strong> ratio between supply and demand, both <strong>of</strong> which are determined within travel-time<br />

catchments (Luo and Wang, 2003). We demonstrate that, with appropriate modifications, this<br />

approach can provide a powerful and flexible tool for measuring transit service accessibility.<br />

2. Previous approaches to measuring transit system accessibility<br />

Most previous studies <strong>of</strong> transit system accessibility adopt distance based measures and focus on <strong>the</strong><br />

concept <strong>of</strong> transit coverage (i.e. <strong>the</strong> percentage <strong>of</strong> population deemed to be served). This is logical to<br />

an extent since bus stops are mostly used by pedestrians and consequently have limited service areas.<br />

<strong>Research</strong> has shown that proximity is a primary determinant in <strong>the</strong> decision to use public transit<br />

(O‘Sullivan and Morrall, <strong>19</strong>96; Biba et al., 2010), and empirical studies suggest <strong>the</strong> maximum<br />

Page | 99


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

distance pedestrians will walk to reach a bus is around 400m (e.g. Atash <strong>19</strong>94, Nyerges <strong>19</strong>95,<br />

O‘Sullivan and Morrall <strong>19</strong>96, Peng <strong>19</strong>97, Murray 2001, Phillips and Edwards, 2002). Thus, <strong>the</strong><br />

‗standard‘ procedure for reporting system accessibility with a <strong>GIS</strong> is to evaluate <strong>the</strong> percentage<br />

population lying within this tolerable walking distance using Euclidean spatial buffers around bus<br />

stops, overlaid with census tracts, and using simple areal interpolation.<br />

An example <strong>of</strong> this approach is provided by Murray et al. (<strong>19</strong>98) who examined transit service<br />

accessibility in SE Queensland, Australia, which showed that government policy targets were not<br />

being met. A number <strong>of</strong> important issues are raised by this simplistic approach. Horner and Murray<br />

(2004) have shown it to be possible to get transit coverage values ranging from 0% to 100%<br />

depending on choices made during <strong>the</strong> analysis in terms <strong>of</strong> spatial depiction (point versus line<br />

objects), scale (census zoning level), distance measure (Euclidean or network), and interpolation<br />

method. Even when restricting <strong>the</strong>mselves to ‗sound‘ choices outcomes <strong>of</strong> between 67% and 87%<br />

were reported. Ano<strong>the</strong>r criticism <strong>of</strong> <strong>the</strong> buffer method is its reliance on <strong>the</strong> assumption <strong>of</strong> uniform<br />

population distribution within census tracts. This prompted O‘Neill et al. (<strong>19</strong>92) to propose a networkratio<br />

method whereby census tract population is evenly distributed along road segments. Zhao et al.<br />

(2003) fur<strong>the</strong>r refined <strong>the</strong> process using a property tax database to allocate population to individual<br />

dwelling units. Applying this idea to a study area in Dallas <strong>the</strong>y report that traditional buffer technique<br />

gave inflated coverage estimates around 48% higher than <strong>the</strong> parcel-based method.<br />

2.1 The Two-Step Floating Catchment Area technique<br />

The two-step floating catchment area (2SFCA) technique for measuring accessibility was introduced<br />

by Luo and Wang (2003). Service area catchments are first placed around service provision points<br />

(e.g. GP practices) and from <strong>the</strong> estimated contained population and measure <strong>of</strong> service volume (e.g.<br />

number <strong>of</strong> GPs available) a population-to-provider ratio computed. Service area catchments are <strong>the</strong>n<br />

constructed around demand points (e.g. census tract centroids) and accessibility is reported as <strong>the</strong> sum<br />

<strong>of</strong> all population-to-provider ratios contained within it. The 2SFCA method is essentially a gravity<br />

type measure in which availability and cumulative opportunity are combined.<br />

We argue that a 2SFCA based approach is <strong>of</strong> value in measuring transit service accessibility. Firstly,<br />

its use <strong>of</strong> service areas defined by a simple distance will be familiar to <strong>the</strong> transportation modelling<br />

community. Secondly, it is able to incorporate <strong>the</strong> notion <strong>of</strong> service supply by utilising a measure such<br />

as <strong>the</strong> number <strong>of</strong> bus visits per week at any given bus stop. We argue that proximity to a bus stop is in<br />

itself <strong>of</strong> little value without some such measure <strong>of</strong> service quality. Thirdly, it captures <strong>the</strong> concept <strong>of</strong><br />

demand during step one. Excessive demand could result in queues and boarding delays that would<br />

impact upon perceived accessibility. Finally, <strong>the</strong> metric is easily understood since it is simply a<br />

measure <strong>of</strong> <strong>the</strong> number <strong>of</strong> available buses per head <strong>of</strong> population able to reach <strong>the</strong>m. Modifications <strong>of</strong><br />

<strong>the</strong> original two-step floating catchment algorithm are necessary in order to meet <strong>the</strong> needs <strong>of</strong> a transit<br />

service accessibility measure, and are described in <strong>the</strong> following section.<br />

3. A modified 2SFCA algorithm for transit service assessment<br />

The original 2SFCA algorithm is defined in Equations 1 and 2:<br />

Step 1<br />

At each service supply location j, search all population locations k that lie within threshold distance<br />

<strong>of</strong> location j and compute population-to-provider ratio<br />

(1)<br />

Page | 100


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

is a measure <strong>of</strong> <strong>the</strong> service supply at location j, <strong>the</strong> population count at location k that lies within<br />

service area catchment j (i.e. ), <strong>the</strong> distance between locations j and k, and <strong>the</strong><br />

threshold distance.<br />

Step 2<br />

At each population location k, find all service supply locations j within threshold distance <strong>of</strong><br />

location k. Compute accessibility as <strong>the</strong> sum <strong>of</strong> all contained values<br />

For <strong>the</strong> measurement <strong>of</strong> transit service accessibility we must consider <strong>the</strong> relationship between bus<br />

stops and bus services – Figure 1 illustrates a typical situation, with three bus routes, a number <strong>of</strong><br />

labelled bus stops and a single population centroid. Assume all bus stops shown lie within <strong>the</strong><br />

threshold distance <strong>of</strong> <strong>the</strong> population centroid. Route A serves stops {a, b, c, d, e, f} with 30<br />

buses a day; route B serves stops {X, e, Y} with 8t buses a day; route C serves stops {f, e, α} with 15<br />

buses per day.<br />

Figure 1: Example relationships between bus routes, stops and service demand centres<br />

In <strong>the</strong> original 2SFCA formula, in Step 1 each bus stop computes using <strong>the</strong> number <strong>of</strong> buses per<br />

day as . For each stop <strong>the</strong> denominator includes <strong>the</strong> population <strong>of</strong> <strong>the</strong> centroid shown, and possibly<br />

o<strong>the</strong>r population centres lying beyond <strong>the</strong> limits <strong>of</strong> <strong>the</strong> area should <strong>the</strong>y fall within <strong>the</strong> threshold<br />

distance <strong>of</strong> that particular stop. Stops used by several routes (e, for example) compute using <strong>the</strong><br />

sum total <strong>of</strong> all buses (i.e. routes A, B, and C) that stop <strong>the</strong>re each day. In Step 2 <strong>the</strong> accessibility score<br />

<strong>of</strong> <strong>the</strong> centroid would be <strong>the</strong> sum total <strong>of</strong> all <strong>the</strong>se scores since, as stated earlier, all lie within <strong>the</strong><br />

threshold distance <strong>of</strong> this location.<br />

This metric over-estimates accessibility because many bus stops serve <strong>the</strong> same route, and a rider only<br />

utilises a service only via one stop (typically <strong>the</strong> nearest). For route A, for example, stop c is <strong>the</strong><br />

(2)<br />

Page | 101


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

nearest so it would be inappropriate for <strong>the</strong> scores <strong>of</strong> stops {a, b, d, e, and f} to be included in <strong>the</strong><br />

accessibility score since <strong>the</strong>y are not used. However, if we record <strong>the</strong> <strong>of</strong> only <strong>the</strong> nearest stop,<br />

similar to <strong>the</strong> classic nearest-distance-to-provider accessibility metric, we under-estimate transit<br />

service accessibility since we ignore <strong>the</strong> presences <strong>of</strong> o<strong>the</strong>r routes lying within <strong>the</strong> tolerable walking<br />

distance. O<strong>the</strong>r stops, which are not <strong>the</strong> nearest but which lie within tolerable walking distance, may<br />

<strong>of</strong>fer access to additional bus services travelling to alternative destinations. E.g. bus stop e is not <strong>the</strong><br />

nearest to <strong>the</strong> population centroid, but is <strong>the</strong> nearest stop to <strong>of</strong>fer access to routes B, and C. Any<br />

measure <strong>of</strong> transit service accessibility for <strong>the</strong> population centroid shown ought to reflect <strong>the</strong>ir<br />

opportunities to access route A via stop c (with 30 buses a day) and route B via stop e (8 buses a day)<br />

and route C via stop e (15 buses a day). A modified algorithm to achieve this is presented below in<br />

Equations 3 and 4..<br />

Step 1<br />

At each bus stop j, we record <strong>the</strong> set <strong>of</strong> bus services that utilise it. is computed for each service<br />

that utilises <strong>the</strong> stop. The measure <strong>of</strong> service supply is <strong>the</strong> number <strong>of</strong> buses relating to<br />

service s that visit stop j.<br />

is <strong>the</strong> population count at location k that lies within <strong>the</strong> service area catchment j (i.e. );<br />

<strong>the</strong> distance between locations j and k; <strong>the</strong> threshold distance.<br />

Step 2<br />

At each population location k, summate scores from Step 1 for each service s, using <strong>the</strong> minimum<br />

network distance between population location k to a bus stop j at which that service s is available<br />

s is a service from <strong>the</strong> full set available at bus stop j;<br />

location k to bus stop j at which service s is available.<br />

(3)<br />

(4)<br />

<strong>the</strong> minimum network distance from<br />

Step 1 measures a supply-to-demand ratio at each bus stop, computed independently for each service<br />

route. Step 2 measures cumulative opportunity reporting <strong>the</strong> total availability <strong>of</strong> service at each<br />

population demand centre, with each individual bus route counted only once using its nearest<br />

available access point.<br />

4. Case Study<br />

A case study was conducted in Merthyr Tydfil County Borough, South Wales (Figure 2). An area <strong>of</strong><br />

recent economic decline, it has <strong>the</strong> highest % long-term unemployed, highest % limiting long-term<br />

illness, and highest % <strong>of</strong> households without access to a car, amongst all Welsh Unitary authorities.<br />

Thus it is reasonable to expect that public transit is an important means <strong>of</strong> access to essential services<br />

and amenities for a substantial proportion <strong>of</strong> its resident population.<br />

Page | 102


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

Using bus stop data (based on National Public Transport Access Node (NaPTAN) sources) and digital<br />

timetable information (based on Association <strong>of</strong><br />

Transport Co-ordinating Officers (ATCO)<br />

Common Interchange Format (CIF) files),<br />

toge<strong>the</strong>r with Ordnance Survey‘s Integrated<br />

Transport Network (ITN) layer, it was possible<br />

using ArcObjects/VB.NET to reconstruct <strong>the</strong><br />

spatial and temporal coverage <strong>of</strong> all bus routes<br />

operating Monday-Friday and originating within<br />

<strong>the</strong> Merthyr Tydfil County boundary. This<br />

yielded 25 unique bus routes and all <strong>the</strong>ir stops,<br />

and, crucially for this study, a count <strong>of</strong> how many<br />

times each bus route visits each stop. Population<br />

distribution was modelled using 2001Output Area<br />

population-weighted centroids downloaded from<br />

Casweb.<br />

The modified 2SFCA procedure was executed<br />

using a bespoke Arc<strong>GIS</strong> TM VBA script, using<br />

Network Analyst TM extension to compute an<br />

Origin-Destination (O-D) matrix between all<br />

population centroids and any bus stops located<br />

within a user-specified distance (400m was<br />

adopted in this study). A map <strong>of</strong> 2SFCA scores is<br />

presented in Figure 4. The map reveals complex<br />

patterns <strong>of</strong> bus transit accessibility levels across<br />

<strong>the</strong> town‘s neighbourhoods. Each Output Area<br />

receives an individual score which are observed<br />

to vary widely both across <strong>the</strong> town in general,<br />

and <strong>of</strong>ten within localised neighbourhoods too.<br />

Figure 4: Transit system accessibility across Merthyr Tydfil<br />

Figure 2: The study area - Merthyr Tydfil<br />

County Borough<br />

Page | 103


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

5. Conclusions<br />

This new metric provides a rich assessment <strong>of</strong> accessibility which incorporates measures <strong>of</strong> service<br />

quality (number <strong>of</strong> buses in any chosen time frame), potential service demand (population within bus<br />

stop catchment), cumulative opportunity (sum <strong>of</strong> population-to-provider ratios), and proximity (since<br />

<strong>the</strong> entire analysis is conducted within <strong>the</strong> framework <strong>of</strong> a ‗tolerable walking distance‘). Fur<strong>the</strong>rmore,<br />

<strong>the</strong> modified algorithm is fine-tuned to <strong>the</strong> specific scenario <strong>of</strong> bus transit accessibility, taking into<br />

account <strong>the</strong> diversity <strong>of</strong> service (specifically responding to <strong>the</strong> number <strong>of</strong> bus routes that are available)<br />

whilst avoiding over-estimation <strong>of</strong> accessibility by repeated counting <strong>of</strong> service provision at stops<br />

which duplicate access to <strong>the</strong> same route. The metric is intuitive, being based on service area<br />

catchments, and returns a score that represents <strong>the</strong> number <strong>of</strong> available buses-per-head <strong>of</strong> population<br />

able to access <strong>the</strong>m.<br />

6. Acknowledgements<br />

<strong>Research</strong> supported by <strong>the</strong> Wales Institute <strong>of</strong> Social and Economic <strong>Research</strong>, Data and Methods<br />

(WISERD), funded by ESRC (Grant Reference: RES-576-25-0021) and HEFCW. We also<br />

acknowledge <strong>the</strong> following data sources: ESRC/JISC Census Programme, Census Geography Data<br />

Unit (<strong>UK</strong>BORDERS), DIGIMAP, EDINA (University <strong>of</strong> Edinburgh); 2001 Census: Standard Area<br />

Statistics (England and Wales) Source: Office for National Statistics, ESRC/JISC Census Programme,<br />

Census Dissemination Unit, MIMAS (University <strong>of</strong> Manchester), Department for Transport<br />

(NaPTAN data). We thank Martyn Dunn <strong>of</strong> Traveline Cymru for <strong>the</strong> supply <strong>of</strong> CIF files for bus<br />

timetables and routes for Merthyr Tydfil, and Andrew Olden <strong>of</strong> <strong>the</strong> Wales Transport <strong>Research</strong> Centre<br />

for <strong>the</strong> supply <strong>of</strong> bespoke s<strong>of</strong>tware that facilitates transfer <strong>of</strong> CIF information into Arc<strong>GIS</strong>.<br />

7. References<br />

Atash, F., <strong>19</strong>94. Redesigning suburbia for walking and transit: emerging concepts. Journal <strong>of</strong> Urban<br />

Planning and Development. 120, pp.48-57.<br />

Biba, S., Curtin, K., and Manca, G., 2010. A new method for determining <strong>the</strong> population with walking<br />

access to transit. International Journal <strong>of</strong> Geographical Information Science, 24, pp.347-364.<br />

Church, A., Frost, M., and Sullivan, K., 2000. Transport and social exclusion in London, Transport<br />

Policy, 7, pp.<strong>19</strong>5-205.<br />

Currie, G., Richardson, T., Smyth, P., Vella-Brodrick, D., Hine, J., Lucas, K., Stanley, J., Morris, J.,<br />

Kinnear, R. and Stanley, J., 2009. Investigating links between transport disadvantage, social exclusion<br />

and well-being in Melbourne – preliminary results, Transport Policy, 16, pp.97-105.<br />

Higgs, G., 2004. A literature review <strong>of</strong> <strong>the</strong> use <strong>of</strong> <strong>GIS</strong>-based measures <strong>of</strong> Access to Health Care<br />

Services, Health Services and Outcome <strong>Research</strong> Methodology, 5, pp.1<strong>19</strong>-139.<br />

Lei, T. and Church, R., 2010. Mapping transit-based access: integrating <strong>GIS</strong>, routes and schedules.<br />

International Journal <strong>of</strong> Geographical Information Science, 24, pp.283-304.<br />

Luo, W. and Wang, F., 2003. Measures <strong>of</strong> spatial accessibility to healthcare in a <strong>GIS</strong> environment:<br />

Syn<strong>the</strong>sis and a case study in Chicago region. Environment and Planning B, 30, pp.865-884.<br />

Murray, A., Davis, R., Stimson, R. and Ferreira, L., <strong>19</strong>98. Public Transportation Access.<br />

Transportation <strong>Research</strong> D, 3, pp.3<strong>19</strong>-328.<br />

Murray, A., 2001. Strategic analysis <strong>of</strong> public transport coverage. Socio-Economic Planning Sciences,<br />

35, pp.175-188.<br />

Nyerges, T., <strong>19</strong>95. Geographic information system support for urban/regional transportation analysis.<br />

In: S. Hanson (Ed.), The geography <strong>of</strong> urban transportation (New York: The Guildford Press),<br />

pp.240-265.<br />

O‘Neill, W., Ramsey, R. and Chou, J., <strong>19</strong>92. Analysis <strong>of</strong> transit service areas using geographic<br />

information systems. Transportation <strong>Research</strong> Record, 1364, pp.131-138.<br />

O‘Sullivan, S. and Morrall, J., <strong>19</strong>96. Walking distances to and from light-rail transit stations.<br />

Transportation <strong>Research</strong> Record, 1538, pp.<strong>19</strong>-26.<br />

Peng, Z., <strong>19</strong>97. A methodology for design <strong>of</strong> a <strong>GIS</strong>-based automatic transit traveller information<br />

system. Computers, Environment, and Urban Systems, 21, pp.359-372.<br />

Page | 104


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

Phillips, C.G. and Edwards, H.R., 2002. Socioeconomic, community-based approach for developing<br />

integrated mass transit systems application to city <strong>of</strong> Baltimore, Maryland. Transportation <strong>Research</strong><br />

Record, 1797, pp.71-79.<br />

Preston, J. and Raje, F., 2007. Accessibility, mobility and transport-related social exclusion, Journal<br />

<strong>of</strong> Transport Geography, 15, pp.151-160.<br />

Thill, J. and Kim, M., 2005. Trip making, induced travel demand, and accessibility. Journal <strong>of</strong><br />

Geographical Systems, 7, pp.229-248.<br />

Wang, F. and Luo, W., 2005. Assessing spatial and nonspatial factors for healthcare access: towards<br />

an integrated approach to defining health pr<strong>of</strong>essional shortage areas. Health and Place, 11, pp.131–<br />

146.<br />

Witten, K., Exeter D. and Field, A., 2003. The Quality <strong>of</strong> Urban Environments: Mapping Variation in<br />

Access to Community Resources. Urban Studies, 40, pp.161-177.<br />

Zhao, F., Chow, L., Li, M. and Ubaka, I., 2003. Forecasting transit walk accessibility: regression<br />

model alternative to buffer method. Transportation <strong>Research</strong> Record, 1835, pp.34-41.<br />

8. Biography<br />

Dr Mitch Langford is a Principal Lecturer in <strong>the</strong> Faculty <strong>of</strong> Advanced Technology, University <strong>of</strong><br />

Glamorgan. His current research interests include dasymetric mapping, population modelling, and<br />

geospatial analysis within <strong>the</strong> fields <strong>of</strong> healthcare, social equality and environmental justice.<br />

Dr. Richard Fry is a researcher for <strong>the</strong> WISERD <strong>GIS</strong>/Data Integration team at <strong>the</strong> Faculty <strong>of</strong><br />

Advanced Technology, University <strong>of</strong> Glamorgan. His current research interests include geospatial<br />

analysis, OpenSource <strong>GIS</strong> s<strong>of</strong>tware, accessibility modelling and linked data within <strong>the</strong> fields <strong>of</strong><br />

transportation and <strong>the</strong> social sciences.<br />

Pr<strong>of</strong>essor Gary Higgs is currently Director <strong>of</strong> <strong>the</strong> <strong>GIS</strong> <strong>Research</strong> Centre in <strong>the</strong> Faculty <strong>of</strong> Advanced<br />

Technology, University <strong>of</strong> Glamorgan and a co-Director <strong>of</strong> <strong>the</strong> Wales Institute <strong>of</strong> Social and<br />

Economic <strong>Research</strong>, Data and Methods (WISERD) at Glamorgan. Over-arching research interests<br />

are in <strong>the</strong> application <strong>of</strong> <strong>GIS</strong> in social and environmental studies, most recently in <strong>the</strong> areas <strong>of</strong> health<br />

geography and emergency planning.<br />

Page | 105


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

Modelling Local Scale Land-Use: Case Nekala<br />

Partanen, Jenni 1<br />

1 Tampere University <strong>of</strong> Technology, School <strong>of</strong> Architecture, Urban Planning and Design,<br />

POB 600, 33101 Tampere , FINLAND<br />

Tel. +358405184405<br />

jenni.partanen@tut.fi<br />

ABSTRACT<br />

This paper presents a case study <strong>of</strong> a CA-based local scale land use model that operates in <strong>GIS</strong><br />

environment. Using <strong>the</strong> dynamic model we demonstrate <strong>the</strong> capacity <strong>of</strong> self-organizing behaviour in<br />

areas with special potential for autonomous generation. The simulations indicate that <strong>the</strong> model under<br />

scrutiny is able to capture locally invisible logics and emergent patterns, which are o<strong>the</strong>rwise<br />

unaccessible. Understanding <strong>the</strong> dynamics <strong>of</strong> <strong>the</strong>se self-organizing areas is <strong>of</strong> great importance in<br />

achieving <strong>the</strong> functional (dynamic) stability <strong>of</strong> <strong>the</strong> city and <strong>the</strong>reby facilitating <strong>the</strong> dynamics <strong>of</strong> <strong>the</strong><br />

wider urban region.<br />

KEYWORDS: Self-organization, Local scale, Land-use Model, Simulation, Urban planning,<br />

Complexity<br />

1. Introduction<br />

This paper presents a CA-based, urban land use simulation model that works in <strong>GIS</strong> environment. Its<br />

dynamics is based on interactions between local scale actors. Resulting emergent global patterns are<br />

studied in a special type <strong>of</strong> area with high potential for self-organization. These generative areas are<br />

considered to have a major impact on urban dynamics.<br />

The use <strong>of</strong> CA-models has become established practice, and <strong>the</strong> number <strong>of</strong> land use models has also<br />

increased in <strong>the</strong> past decade. Yet many <strong>of</strong> <strong>the</strong>m operate on a regional scale, <strong>of</strong>ten contemplating urban<br />

growth or urban form. (See Batty (<strong>19</strong>99&2004), Clarke (<strong>19</strong>97), Yeh&Xia (2002). In <strong>the</strong> urban<br />

planning discourse <strong>the</strong>re are also well known lower scale models, such as <strong>the</strong> classics <strong>of</strong> Portugali<br />

(2000) and Schelling (<strong>19</strong>71, <strong>19</strong>78) that study <strong>the</strong> social phenomena and dynamics, but leave <strong>the</strong> broad<br />

spectrum <strong>of</strong> economic actors‘ residual dependences intact. With <strong>the</strong> model presented here, <strong>the</strong>se<br />

interactions between urban actors are studied on local scale using <strong>the</strong> MapInfo program.<br />

The research hypo<strong>the</strong>sis is that, in addition to <strong>the</strong> self-generation <strong>of</strong> <strong>the</strong> urban structure as a whole,<br />

<strong>the</strong>re is a special case <strong>of</strong> self-organization in certain urban enclaves. Moreover, <strong>the</strong> internal dynamics<br />

<strong>of</strong> <strong>the</strong>se enclaves is (seemingly) chaotic, non-causal and unpredictable, following its own laws.<br />

Certain mechanisms can be discerned behind this kind <strong>of</strong> emergent behaviour, in which <strong>the</strong> local<br />

instabilities <strong>of</strong> <strong>the</strong> system constantly regenerate new demarcated areas <strong>of</strong> growth and progress,<br />

simultaneously generating higher level order and structure in a system. This characteristic dynamics<br />

has again been perceived earlier in urban studies by several authors (Portugali(2000), Shane(2005),<br />

Oswald&Baccini(2003)). In <strong>the</strong> innovation-led society, <strong>the</strong>se self-organizing areas are also potentially<br />

<strong>of</strong> great economic significance, given <strong>the</strong>ir potential to form a ―production space‖ <strong>of</strong> creative<br />

encounters for small firms and activities, especially in <strong>the</strong> culture industry.<br />

The model introduced <strong>of</strong>fers an applicable method to examine <strong>the</strong> effect <strong>of</strong> boundary conditions on<br />

<strong>the</strong> qualitative and quantitative alternations <strong>of</strong> land uses in <strong>the</strong> target area, and most importantly, to<br />

observe <strong>the</strong> emerging higher level patterns and regularities.<br />

Page | 106


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

2. The Model<br />

2.1 Cellular automata<br />

Because <strong>the</strong> research hypo<strong>the</strong>sis is based on local interactions between neighbours, <strong>the</strong> cellular<br />

automaton was considered <strong>the</strong> most appropriate model type. It also enables a simple representation <strong>of</strong><br />

<strong>the</strong> global behaviour emerging from local interactions.<br />

Cellular automata are simplified ma<strong>the</strong>matical idealizations <strong>of</strong> natural systems. In <strong>the</strong>ir most simple<br />

form <strong>the</strong>y consist <strong>of</strong> a lattice <strong>of</strong> identical cells with discrete stages. The values <strong>of</strong> <strong>the</strong> cells are <strong>the</strong>n<br />

changed every time-step according to <strong>the</strong> deterministic rule that defines <strong>the</strong> new state <strong>of</strong> <strong>the</strong> cell<br />

depending on <strong>the</strong> states <strong>of</strong> cells in <strong>the</strong> neighbourhood. Despite <strong>the</strong>ir simple structure, cellular automata<br />

are applicable for describing bottom-up organized complex structures.<br />

In <strong>the</strong> present study <strong>the</strong> principles <strong>of</strong> formal CA have been relaxed for a better correspondence<br />

between simulation and urban reality. The following modifications were chosen to avoid losing<br />

elegance and simplicity in <strong>the</strong> model frame. First <strong>of</strong> all, contrary to baseline CA, <strong>the</strong> states <strong>of</strong> <strong>the</strong> cells<br />

(building sites) are defined qualitatively and quantitatively instead <strong>of</strong> binary (on-<strong>of</strong>f) mode. Secondly,<br />

in this model <strong>the</strong> form <strong>of</strong> <strong>the</strong> cell is irregular, non-uniform cell-space. The legal site division based on<br />

land register maps is used as a basic grid (Figure 1). The un-extended neighbourhood is formed by<br />

buffering <strong>the</strong> sites (r=24m). The size <strong>of</strong> <strong>the</strong> buffer is taken to be half <strong>of</strong> <strong>the</strong> ―block‖, which is based on<br />

average block dimensions in local traditional plans. This is also apparent in <strong>the</strong> older, livelier part <strong>of</strong><br />

Nekala area (Figure 2). This dimension was considered walkable, and thus attractive to customers<br />

from neighbouring firms, who could benefit from agglomeration. Accordingly, <strong>the</strong> number <strong>of</strong><br />

neighbours varies from site to site due to <strong>the</strong>ir diverse shapes and sizes. Finally, it is assumed that<br />

more complex transition rules (Figure 3) better reflect <strong>the</strong> micro-scale economic geography <strong>of</strong> <strong>the</strong><br />

target area. Yet, unlike many urban models, <strong>the</strong> transition rules <strong>of</strong> this model are stationary, indicating<br />

that <strong>the</strong> complex behaviour <strong>of</strong> <strong>the</strong> model can be considered self-organization instead <strong>of</strong> evolution, and<br />

<strong>the</strong> <strong>the</strong>ory <strong>of</strong> self-organization referred to above may be applied (Allen 2004). Never<strong>the</strong>less, <strong>the</strong><br />

calibration <strong>of</strong> <strong>the</strong> model between iterations may naturally be carried out.<br />

Figure 1. Legal site division and existing buildings and aerial photograph.<br />

Page | 107


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

2.2 Rules chosen<br />

The rules <strong>of</strong> <strong>the</strong> model are based on empirical and <strong>the</strong>oretical premises: synergetic or similar activities<br />

tend to gravitate towards each o<strong>the</strong>r. The clustering that exceeds <strong>the</strong> threshold value causes<br />

―overpopulation‖, which leads to behaviour where part <strong>of</strong> <strong>the</strong> activities re-locate <strong>the</strong>mselves (Figure<br />

2). For example, if <strong>the</strong> site‘s floor area ratio (FAR) is between 0 and 0.1, it is classified as ―empty‖.<br />

Thus <strong>the</strong> probability <strong>of</strong> <strong>the</strong> site ―remaining‖ as before is 0.1, and <strong>the</strong> probability <strong>of</strong> it being ―filled-up‖<br />

is 0.3 etc. The volume <strong>of</strong> a particular activity (in square metres) on <strong>the</strong> site is related to <strong>the</strong> volume <strong>of</strong><br />

<strong>the</strong> same activity on <strong>the</strong> neighbouring sites in a previous modelling stage. Activities are classified into<br />

six categories, according to <strong>the</strong> degree <strong>of</strong> <strong>the</strong>ir interaction with <strong>the</strong> environment (Table 1).<br />

Consequently, <strong>the</strong>re may be zero to six different actor(s) on each site depending on <strong>the</strong> states <strong>of</strong> <strong>the</strong><br />

neighbours and former state <strong>of</strong> <strong>the</strong> site itself.<br />

Figure 2. Cells mode <strong>of</strong> transformation according to <strong>the</strong>ir utilization rates.<br />

P-1: ―empty‖, FAR=0-0.1; P-2: ―half-empty‖, FAR=0.1-0.3; P-<br />

3:‖half-full‖, FAR= 0.3-0.7; P-4: ―full‖, FAR= 0.7-1.<br />

Table 1. Hypo<strong>the</strong>tical classification <strong>of</strong> activities according to <strong>the</strong> probable interactions<br />

between activities and environment. This guided <strong>the</strong> classification <strong>of</strong> data at <strong>the</strong> beginning, and<br />

for estimating <strong>the</strong> proximity preference values (table 2.). Accessibility=threshold <strong>of</strong> entering,<br />

Interference= intentional or unintentional interaction with neighbourhood, Flow= flows <strong>of</strong> traffic<br />

and transport (goods&people)<br />

Accessibility Interference Flow<br />

Housing(H) Low(1) Low(1) Medium(3)<br />

Retail(R) High(5) Low-medium(2) High(5)<br />

Services(S) Medium-high(4) Low-medium(2) Medium(3)<br />

Business(B) Low-medium(2) Low(1) Medium(3)<br />

Industry(I) Low-medium(2) Medium-high(4) Medium-high(4)<br />

Warehouses(W) Low(1) Low(1) High(5)<br />

2.3 Structure <strong>of</strong> <strong>the</strong> model (Figure 3)<br />

In <strong>the</strong> beginning, <strong>the</strong> utilization rate <strong>of</strong> <strong>the</strong> site is defined, by which <strong>the</strong> sites are classified into four<br />

categories with potential values (P-1 to P-4) indicated in Figure 2. The new state <strong>of</strong> <strong>the</strong> cell is defined<br />

Page | 108


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

depending on <strong>the</strong>se modes <strong>of</strong> transformation: <strong>the</strong> site may remain as it is (RM), it may fill up (F)<br />

according to <strong>the</strong> percentage growth rate (GR) defined by <strong>the</strong> user, activities on <strong>the</strong> site may change<br />

(C) while <strong>the</strong> volume remains <strong>the</strong> same, or <strong>the</strong> volume and activities may be totally reconstructed<br />

(RC).<br />

In addition, in <strong>the</strong> last two cases (C and RC) <strong>the</strong> model may be ―calibrated‖ – i.e. fitted to fit roughly<br />

with <strong>the</strong> observed reality – and enforce certain activities to dominate or decrease, simulating <strong>the</strong><br />

existing pressures in planning or decision-making. The weight factors are defined separately for both<br />

cases.<br />

The matrix defining preferences <strong>of</strong> proximity between diverse activity types forms a user interface.<br />

The tolerance <strong>of</strong> each activity with regard to <strong>the</strong> same and differing activity is represented by a<br />

preference factor (Table 2), that can be altered to test emergent influences <strong>of</strong> various scenarios in<br />

planning processes.<br />

Table 2. Proximity preferences: An example <strong>of</strong> matrix values emphasizing <strong>the</strong> benefit <strong>of</strong><br />

housing and retail on <strong>the</strong> one hand, and business & small industry on <strong>the</strong> o<strong>the</strong>r. Values can<br />

be outlined to simulate and test variety <strong>of</strong> planning preferences. H=housing, R=retail,<br />

S=services, B=business, I=industry, W=warehouses.<br />

H R S B I W<br />

H 8 12 4 1 1 1<br />

R 20 16 1 4 1 1<br />

S 1 5 4 1 1 1<br />

B 2 1 2 3 12 1<br />

I 1 1 1 10 24 1<br />

W 1 1 1 1 1 1<br />

Figure 3. Structure <strong>of</strong> <strong>the</strong> model.<br />

Page | 109


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

2.4 The study area<br />

This model was tested in a case study <strong>of</strong> <strong>the</strong> Nekala industrial area, Tampere, Finland. This area <strong>of</strong><br />

approximately 80 sites was planned for heavy industry and <strong>the</strong> processing <strong>of</strong> agricultural products in<br />

<strong>the</strong> late <strong>19</strong>30‘s. In recent decades this quite centrally located area has been surrounded with housing,<br />

and today forms a unique enclave within <strong>the</strong> urban fabric. The area has demonstrated a surprising<br />

capability for self-organization, and has been able to flexibly adjust to <strong>the</strong> current mode <strong>of</strong> production<br />

– from mainly industrial to a gradually more complex mixture <strong>of</strong> service, information technology and<br />

cultural industry (Castells 2000).<br />

Several essential features indicating <strong>the</strong> potential for self-organization can be listed. First <strong>of</strong> all, strict<br />

boundaries with <strong>the</strong> adjacent housing area. Second, high accessibility <strong>of</strong>fering sufficient flow <strong>of</strong><br />

energy and matter, and, third, increasing complexity and diversity (Partanen 2010, Shalizi,<br />

Shalizi&Haslinger 2004) which has been remarkable (Figure 4,Table 3), and finally, empirically<br />

discovered internal dynamics on which <strong>the</strong> rules have been built (Partanen 2010).<br />

Figure 4. Shares <strong>of</strong> activity types in <strong>the</strong> Nekala area from <strong>19</strong>71 to 2008.<br />

Table 3. Changes in counts <strong>of</strong> activities in <strong>the</strong> Nekala area from <strong>19</strong>71 to 2008.<br />

<strong>19</strong>71 <strong>19</strong>82 <strong>19</strong>93 2008<br />

Agriculture 6 4 3 3<br />

Traditional Industry 79 29 42 32<br />

Services 48 57 92 142<br />

Information Technology 0 0 13 30<br />

Cultural Industry 3 2 10 37<br />

Missing data 0 0 11 24<br />

All 136 92 171 268<br />

Gradual completion and changes in activities and to a lesser degree <strong>the</strong> total reconstruction can be<br />

considered <strong>the</strong> main structural logics in <strong>the</strong> area (Partanen 2005).<br />

The typical spatial logics <strong>of</strong> <strong>the</strong> area were studied using samples <strong>of</strong> historical statistical data on<br />

Page | 110


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

workplaces. First, data from <strong>19</strong>71, <strong>19</strong>82, <strong>19</strong>93 and 2008 were combined with location information<br />

using <strong>GIS</strong> to discover <strong>the</strong> essential dynamics <strong>of</strong> <strong>the</strong> area. The original data lacked information on <strong>the</strong><br />

extent <strong>of</strong> activities, and this data was supplemented by reference to <strong>the</strong> building permission archives<br />

<strong>of</strong> <strong>the</strong> City <strong>of</strong> Tampere. Consequently, <strong>the</strong> up-to-date quantitative data on which <strong>the</strong> model is based<br />

was not available for every site.<br />

3. Findings<br />

It was discovered that <strong>the</strong> dynamic states <strong>of</strong> <strong>the</strong> system were heavily dependent on <strong>the</strong> initial values,<br />

and very difficult to achieve. This finding concurs with <strong>the</strong> findings <strong>of</strong> Partanen & Joutsiniemi (2007)<br />

with <strong>the</strong> same type <strong>of</strong> model. Yet <strong>the</strong> classic dynamics <strong>of</strong> <strong>the</strong> CA captured in this earlier research –<br />

static, periodic, complex – was not yet discernible. There could be several reasons for this: First <strong>of</strong> all,<br />

chaotic systems‘ dependence on initial conditions makes it quite difficult to iterate <strong>the</strong> ―favourable‖<br />

initial values. Secondly, in <strong>the</strong> previous experiment, o<strong>the</strong>r factors such as site morphology and<br />

fragmentation as well as quality and quantity <strong>of</strong> static cells had a decisive impact on <strong>the</strong> behaviour <strong>of</strong><br />

<strong>the</strong> system. The effects <strong>of</strong> <strong>the</strong>se features are beyond <strong>the</strong> scope <strong>of</strong> this paper and a subject for fur<strong>the</strong>r<br />

study.<br />

Never<strong>the</strong>less, in this research it was perceived that <strong>the</strong> different emphases on <strong>the</strong> limits and boundary<br />

values affect <strong>the</strong> dynamics <strong>of</strong> <strong>the</strong> model in manifold ways. The emergent behaviour <strong>of</strong> <strong>the</strong> model was<br />

perceived since it was obvious that <strong>the</strong> values in <strong>the</strong> preference matrix affected not only <strong>the</strong> number<br />

<strong>of</strong> activities, but also <strong>the</strong> nature <strong>of</strong> <strong>the</strong>ir dynamics (Figures 5-7). In this example <strong>the</strong> combination <strong>of</strong><br />

high weight between ―retail‖ and ―housing‖ on <strong>the</strong> one hand, and emphasizing <strong>the</strong> work activities<br />

(I&B) on <strong>the</strong> o<strong>the</strong>r had unpredictable impacts on <strong>the</strong> dynamics <strong>of</strong> <strong>the</strong> model, causing for example,<br />

unstable, escalating progress in housing (Figure 5), and a surprising shift in industry (Figure 7). The<br />

relatively stable development <strong>of</strong> <strong>the</strong> retail trade is understandable, yet not self-evident, compared to<br />

almost identical weight factors <strong>of</strong> industrial use (Table 3).<br />

Figure 5. Housing sector: Iterations t=1 to t=20.<br />

Page | 111


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

4. Discussion<br />

Figure 6. Retail sector: Iterations t=1 to t=20.<br />

Figure 7. Light industry sector: Iterations t=1 to t=20.<br />

These preliminary results indicate that this type <strong>of</strong> relaxed CA model could be useful in operative<br />

urban planning and may be an applicable tool for examining <strong>the</strong> unpredictable global effects <strong>of</strong> <strong>the</strong><br />

lower level actors‘ residual dependences. In <strong>the</strong> next research phase, comparing <strong>the</strong> differently<br />

weighted iterations <strong>of</strong> <strong>the</strong> simulation will enable us to discover certain limiting values for planning<br />

that ei<strong>the</strong>r inhibit <strong>the</strong> development or allow <strong>the</strong> self-organization <strong>of</strong> <strong>the</strong> area, enabling it to adapt to<br />

<strong>the</strong> modes <strong>of</strong> production, and most <strong>of</strong> all, generate behaviour that supports <strong>the</strong> dynamic stability <strong>of</strong><br />

<strong>the</strong> whole urban region.<br />

The most valuable findings are related to self-organizing phenomena, in which, especially in areas<br />

like Nekala, with its high degree <strong>of</strong> self-sufficiency, it is important to avoid evolutionary dead-ends<br />

as much as possible. For <strong>the</strong> characteristics <strong>of</strong> <strong>the</strong> area, diversity, flexibility and pluralism are to be<br />

encouraged; decisions likely to reduce <strong>the</strong>se should be thoroughly examined. Any plan should be<br />

understood as part <strong>of</strong> <strong>the</strong> evolutionary process. In this case, this simulation <strong>of</strong>fers an appropriate tool<br />

for <strong>the</strong> processes <strong>of</strong> planning, negotiating and decision-making.<br />

Page | 112


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

5. References:<br />

Allen P (2004). Cities and regions as self-organizing systems. Taylor&Francis, Oxon.<br />

Batty M (2007). Cities and Complexity. MIT press, Cambridge.<br />

Batty M, Xiu Y and Sun Z (<strong>19</strong>99). Modelling urban dynamics through <strong>GIS</strong>-based cellular<br />

automata. Computers, Environment and Urban Systems 23, 205-233.<br />

Castells M (2000). The Information Age: Economy, Society and Culture. Volume 1. The Rise<br />

<strong>of</strong> Network Society. Blackwell Publishing,Oxford<br />

Clarke K C, Hoppen S and Gaydos L (<strong>19</strong>97). A self-modifying cellular automaton model for<br />

historical urbanization in <strong>the</strong> San Francisco Bay area. Environment and Planning B: Planning<br />

and Design 24 247-261.<br />

Oswald F and Baccini P (2003). Netzstadt – designing <strong>the</strong> urban. Birkhäuser, Basel-Boston-Berlin.<br />

Partanen J and Joutsiniemi A (2007). Simulaatio kaupungin kompleksisen kehityksen<br />

hallinnassa.Yhdyskuntasuunnittelu 2/2007<br />

Partanen J (2005). Nekalan teollisuusalueen itse-organisoitumisen malleja. Master‘s <strong>the</strong>sis, Tampere<br />

University <strong>of</strong> Technology. (Tampere: Department <strong>of</strong> Architecture, TUT, 2005)<br />

Partanen J . (to be published in December 2010) Reclaiming <strong>the</strong> Fallow Resources: Potential for<br />

Breeding Ground or Over-Exploitation in: Kimmo Ylä-Anttila ed. City Scratching. Datutop<br />

series. (Department <strong>of</strong> Architecture: Tampere, Finland).<br />

Portugali J (<strong>19</strong>99). Self-organization and <strong>the</strong> city. Springer-Verlag, Berlin Heidelberg New York<br />

Schelling T C <strong>19</strong>71. Dynamic Models <strong>of</strong> Segregation. Journal <strong>of</strong> Ma<strong>the</strong>matical Sociology 1:143-<br />

186.<br />

Schelling T C <strong>19</strong>78. Micromotives and macrobehavior. Norton, New York.<br />

Shane D G (2005). Recombinant Urbanism – Conceptual Modeling in Architecture, Urban<br />

Design and City Theory. John Wiley & Sons ltd, London.<br />

Shalizi C R, Shalizi K L and Haslinger R(2004). Quantifying Self-Organization with Optimal<br />

Predictors. Phys. Rev. Lett. 93.<br />

Yeh A G-O and Xia L (2002). A Cellular automata model to simulate development for urban<br />

planning. Environment and Planning B: Planning and Design 29, 431-450.<br />

Page | 113


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

The Changing face <strong>of</strong> land use in <strong>the</strong> British countryside since <strong>the</strong> <strong>19</strong>30s<br />

Paula Aucott 1 , Humphrey Southall 2<br />

1 University <strong>of</strong> Portsmouth, Dept. <strong>of</strong> Geography, Buckingham Building, Lion Terrace,<br />

Portsmouth, PO1 3HE<br />

Tel. +44 (0)23 9284 2500, Fax +44 (0)23 9284 2512<br />

Paula.aucott@port.ac.uk, http//www.gbhgis.org<br />

2 University <strong>of</strong> Portsmouth<br />

ABSTRACT<br />

Interaction between humans and <strong>the</strong>ir surrounding landscape has long been a subject <strong>of</strong> interest and<br />

debate. Since pre-history man has impacted on his immediate environment through activities<br />

designed to fulfil his basic needs. Both <strong>the</strong> changing natural environment <strong>of</strong> <strong>the</strong> British countryside<br />

and man‘s efforts to shape his immediate surroundings to his own advantage, have left a lasting<br />

legacy. During <strong>the</strong> twentieth Century this rate <strong>of</strong> change has significantly increased and this paper<br />

will explore those changes through <strong>the</strong> historical maps created by <strong>the</strong> Land Utilisation Survey run by<br />

Stamp in <strong>the</strong> <strong>19</strong>30s and modern satellite imagery available today.<br />

KEYWORDS: Land use, Historical <strong>GIS</strong>, Historic maps, Satellite imagery, Land Utilisation Survey<br />

1. Introduction<br />

Man has always had an impact on his immediate environment, ranging from building<br />

accommodation, military defences, cultural focal points, clearing and using areas for agriculture and<br />

creating transport links. All <strong>the</strong>se things leave an indelible impact on <strong>the</strong> landscape, but as Trevor<br />

Rowley comments ―during <strong>the</strong> twentieth century, <strong>the</strong> rural landscape was subjected to change and<br />

pressure on an unprecedented scale‖ 3 . This paper explores <strong>the</strong> ways in which <strong>the</strong> landscape has<br />

developed in <strong>the</strong> past seventy years using a combination <strong>of</strong> historical Land Utilisation maps, satellite<br />

imagery and <strong>GIS</strong> techniques to provide a detailed commentary on this change.<br />

1.2 Background<br />

A recent report by Pr<strong>of</strong>essor Sir Peter Hall for <strong>the</strong> Government Office <strong>of</strong> Science describes Robin<br />

Best‘s <strong>19</strong>81 book Land Use and Living space as ―<strong>the</strong> definitive study on <strong>the</strong> subject … <strong>of</strong> land use in<br />

Britain‖. Despite its thoroughness much has happened since <strong>the</strong> book was published and this must be<br />

taken into consideration because: all Best's maps are pre <strong>19</strong>71; <strong>the</strong> book concentrates on urbanisation<br />

<strong>of</strong> farmland, ignoring o<strong>the</strong>r changes; satellite imagery is now available; and <strong>GIS</strong> technology has<br />

developed to allow new comparison methods. The amount known today about land use change is<br />

surprisingly limited compared to knowledge on o<strong>the</strong>r historical impacts such as changing population.<br />

This dearth is not caused by a lack <strong>of</strong> resources per se as <strong>the</strong>re are plenty <strong>of</strong> historical maps with<br />

contemporary analysis such as Stamp‘s <strong>19</strong>48 book arising from <strong>the</strong> Land Utilisation Survey,<br />

historical statistics and significant satellite coverage since <strong>the</strong> mid <strong>19</strong>80s. What has been missing is<br />

analysis <strong>of</strong> long runs <strong>of</strong> data forming a time series, probably due to <strong>the</strong> diversity <strong>of</strong> <strong>the</strong> data available.<br />

This project team is well placed to carry out this analysis having completed a series <strong>of</strong> previous<br />

feasibility studies, pilot projects and surveys <strong>of</strong> sources. They have published all <strong>the</strong> map sheets <strong>of</strong><br />

<strong>the</strong> first Land Utilisation Survey for <strong>the</strong> first time via <strong>the</strong> website A Vision <strong>of</strong> Britain through time as<br />

well as building a demonstration website, Land <strong>of</strong> Britain, to show what is possible by combining all<br />

<strong>the</strong>se different types <strong>of</strong> land use datasets for <strong>the</strong> area around Brighton.<br />

3 Rowley, T. (2006), p247<br />

Page | 114


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

1.3 Objectives<br />

The current research aims to create a national overview <strong>of</strong> spatial land use change through <strong>the</strong><br />

following methods. Firstly <strong>the</strong> comparison <strong>of</strong> national historical Land Utilisation mapping with<br />

modern satellite mapping, with specific analysis focused on <strong>the</strong> contrast provided between localised<br />

case studies against <strong>the</strong> national overview. Also a comparison <strong>of</strong> ground survey with modern satellite<br />

mapping for <strong>the</strong> area around Brighton will analyse similarities and differences created through <strong>the</strong><br />

use <strong>of</strong> alternative collection techniques <strong>of</strong> field surveying and satellite imagery. Finally <strong>the</strong> national<br />

mapping analysis results are compared with county-level Farm Census statistics. Doing this type <strong>of</strong><br />

analysis helps us to understand land use change and <strong>the</strong> implications <strong>of</strong> this for society and culture<br />

existing within this space.<br />

Through investigating this underutilised set <strong>of</strong> resources by employing innovative analysis to image<br />

processing and <strong>GIS</strong> techniques a series <strong>of</strong> historical data is produced coming up to <strong>the</strong> present.<br />

Substantive findings that actually reflect <strong>the</strong> changing nature <strong>of</strong> land use in England and Wales will<br />

be presented, as long runs <strong>of</strong> farm census data are analysed toge<strong>the</strong>r taking into account large<br />

changes in both what topics were reported on and <strong>the</strong> geographies <strong>the</strong> statistics were reported in. The<br />

results <strong>of</strong> this research <strong>of</strong>fers <strong>the</strong> opportunity to impact on policy makers and <strong>the</strong>ir decision making<br />

by lending weight and gravitas to policy debate on land management, maintenance and future<br />

development.<br />

2.1 National Analysis<br />

A vector version <strong>of</strong> <strong>the</strong> Stamp‘s 10miles to 1inch<br />

Land Utilisation National summary map for<br />

England and Wales has been created (Figure 1).<br />

The sou<strong>the</strong>rn part used <strong>the</strong> black ink colour<br />

separations created during <strong>the</strong> printing process to<br />

create vector layers which were <strong>the</strong>n merged<br />

toge<strong>the</strong>r. For <strong>the</strong> nor<strong>the</strong>rn part no colour<br />

separations were available, so <strong>the</strong> vectorisation was<br />

done from <strong>the</strong> more cluttered published map sheet<br />

and manually tidied.<br />

This coverage is compared against <strong>the</strong> Land Cover<br />

2000 raster dataset, from <strong>the</strong> Centre for Ecology<br />

and Hydrology, to examine differences exposed<br />

over <strong>the</strong> 70 year period. It is also assessed in<br />

relation to <strong>the</strong> land use statistics available from <strong>the</strong><br />

annual Farm Census which began in 1866.<br />

2.2 Map Creation Analysis<br />

Figure 13. Vector National Summary LUS<br />

map<br />

Compared to field surveys which are done by people on <strong>the</strong> ground, satellite imagery is collected<br />

using a significantly different methodology, involving automated imaging <strong>of</strong> vast swa<strong>the</strong>s <strong>of</strong> land<br />

from above. This raises concern regarding <strong>the</strong> accuracy <strong>of</strong> <strong>the</strong> comparisons being made between <strong>the</strong><br />

historical maps and <strong>the</strong> satellite maps. A calibration method was devised using a comparison<br />

between <strong>the</strong> Land Use <strong>UK</strong> survey done using traditional field survey techniques during <strong>the</strong> <strong>19</strong>90s<br />

(Walford, <strong>19</strong>97), and <strong>the</strong> Land Cover 2000 vector dataset. As <strong>the</strong> two datasets are roughly <strong>the</strong> same<br />

date, this ensures <strong>the</strong>re is little actual physical difference in land use on <strong>the</strong> ground and facilitates <strong>the</strong><br />

recognition <strong>of</strong> differences caused by disparities in <strong>the</strong> data collection methods. This in turn is used to<br />

calibrate <strong>the</strong> o<strong>the</strong>r findings at both local level and more generally at a national scale.<br />

Page | 115


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

2.3 Local Analysis<br />

Detailed comparisons <strong>of</strong> patterns <strong>of</strong> land use change between <strong>the</strong> National summary sheets and <strong>the</strong><br />

Land Cover 2000 raster dataset are impractical as <strong>the</strong> spatial resolution is not high enough. Therefore<br />

a case study scenario was devised to investigate more precise variations in land use using <strong>the</strong> Land<br />

Utilisation Survey and Satellite data. Birmingham and Dartmoor were focused on in order to provide<br />

two contrasting sets <strong>of</strong> results.<br />

Birmingham was chosen as <strong>the</strong> focus to examine urban growth, in particular in relation to <strong>the</strong> Green<br />

Belt policy as surrounding councils have tried to limit <strong>the</strong> city‘s growth, population expansion and<br />

in-filling or brown-field site development. Dartmoor is a mainly pastoral landscape with dispersed<br />

settlement. Methods <strong>of</strong> moorland improvement for agricultural use have been practiced from prehistory<br />

to <strong>the</strong> present day and <strong>the</strong> analysis indicates changes over time to <strong>the</strong>se areas. Analysis <strong>of</strong> <strong>the</strong><br />

variation in land cover between <strong>the</strong> two dates and in contrast between <strong>the</strong> areas will enable<br />

assessment <strong>of</strong> <strong>the</strong> success <strong>of</strong> <strong>the</strong> methodology. Brighton was discounted from this analysis because it<br />

has Down-land ra<strong>the</strong>r than moorland, and <strong>the</strong> very presence <strong>of</strong> <strong>the</strong> Downs have limited urban<br />

expansion, <strong>the</strong>refore <strong>the</strong> city is too small and unrepresentative <strong>of</strong> <strong>the</strong> ‗norm‘ to depict more general<br />

patterns <strong>of</strong> change.<br />

The selection <strong>of</strong> <strong>the</strong>se two particular areas was arbitrary in some senses because <strong>the</strong> project had<br />

already compiled a full set <strong>of</strong> scans <strong>of</strong> <strong>the</strong> colour separations for each sheet. The colour separations<br />

<strong>the</strong>mselves are a by-product <strong>of</strong> <strong>the</strong> printing process. They are monochrome sheets that were compiled<br />

to show areas <strong>of</strong> a single colour relating to land use. For example brown is arable, whilst green is<br />

stippled for pasture and solid for woodland. These were used to create <strong>the</strong> printing plates from which<br />

successive layers were printed in colour over <strong>the</strong> base map.<br />

Figure 14. Vector coverage <strong>of</strong> Wood and Pasture LUS<br />

for part <strong>of</strong> Dartmoor<br />

The colour separations were used to<br />

produce a polygon coverage for each<br />

map sheet (Figure 2). They provided a<br />

simple monochrome sheet for each<br />

printed colour that meant <strong>the</strong> extra<br />

noise associated with wording and<br />

black lines from <strong>the</strong> published maps<br />

were excluded, <strong>the</strong>reby making <strong>the</strong><br />

vectorisation much simpler. The<br />

disadvantage was <strong>the</strong> overlap <strong>of</strong><br />

colours necessary for printing<br />

purposes meant <strong>the</strong> vector versions <strong>of</strong><br />

<strong>the</strong>se maps when added toge<strong>the</strong>r<br />

always total just over a hundred<br />

percent. Birmingham has <strong>the</strong> full<br />

range <strong>of</strong> colour separations, whereas<br />

Dartmoor lacks urban and suburban<br />

sheets as <strong>the</strong>y were irrelevant to that<br />

area. Once <strong>the</strong> vector coverages had<br />

been created <strong>the</strong> <strong>19</strong>30s land use categories were assigned to <strong>the</strong> 2000 data in order to produce a<br />

comparison <strong>of</strong> land areas that were <strong>the</strong> same and different between <strong>the</strong> two datasets. The historical<br />

land use categories had to be assigned to <strong>the</strong> modern land uses as <strong>the</strong>re were far fewer categories in<br />

<strong>the</strong> earlier survey.<br />

3. Results<br />

In Birmingham, as might be predicted, <strong>the</strong>re is a significant amount <strong>of</strong> constancy in <strong>the</strong> central area<br />

for urban (non-productive) land use with growth on <strong>the</strong> peripheries, likewise growth <strong>of</strong> <strong>the</strong> suburbs is<br />

Page | 116


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

also mainly on <strong>the</strong> peripheries. There are some signs <strong>of</strong> changing classifications from suburban to<br />

urban as areas within <strong>the</strong> city became more built up and both classes increased <strong>the</strong>ir overall<br />

percentage <strong>of</strong> land area (see Table 1). Toge<strong>the</strong>r urban growth and suburban sprawl, as shown in <strong>the</strong><br />

data, increased from 27.5% <strong>19</strong>34 to 42.9% in 2000. As might be expected for an urban area, <strong>the</strong>re are<br />

only small dispersed areas <strong>of</strong> heath-land and woodland. Very little change occurs in ei<strong>the</strong>r category.<br />

There is some arable land on <strong>the</strong> rural edges <strong>of</strong> <strong>the</strong> urban area and <strong>the</strong> changes that occurred were all<br />

increases in arable land. The vast majority <strong>of</strong> land outside <strong>of</strong> <strong>the</strong> main urban area is grassland<br />

reflecting <strong>the</strong> predominant farming economy <strong>of</strong> <strong>the</strong> area. However, this comparison shows <strong>the</strong><br />

grassland area to be getting significantly smaller. Some <strong>of</strong> <strong>the</strong> differences in grassland area might be<br />

due to variation in classification <strong>of</strong> land use – indeed <strong>the</strong>re was much debate around <strong>the</strong> classification<br />

<strong>of</strong> permanent grass, rough grazing and rotational fallow land in <strong>the</strong> <strong>19</strong>30s, leading later surveys to be<br />

more explicit in <strong>the</strong>ir instructions to surveyors.<br />

Table 1. Results <strong>of</strong> Comparison <strong>of</strong> Land Use Classes in Birmingham<br />

Land Use Classes<br />

% <strong>of</strong> Total<br />

Area <strong>19</strong>34<br />

% <strong>of</strong> Total<br />

Area 2000 Difference<br />

Meadow & Permanent Grass 58.0 21.9 -36.1<br />

Suburban 15.5 24 8.5<br />

Arable 13.8 21.6 7.8<br />

Urban 12.0 18.9 6.9<br />

Forest & Woodland 3.1 10.3 7.2<br />

Heath & Moor 2.5 2.5 0.0<br />

Water<br />

4. Conclusions<br />

0.8 0.8<br />

This analysis indicates <strong>the</strong> study area in Birmingham has not had large amounts <strong>of</strong> woodland or<br />

heath in <strong>the</strong> past 70 years. As expected it confirms <strong>the</strong> urbanisation within <strong>the</strong> central Birmingham<br />

area as a progressive shift from suburban to urban and highlights <strong>the</strong> spread <strong>of</strong> <strong>the</strong> suburban<br />

province, although perhaps not to <strong>the</strong> extent that might have been predicted. Interestingly <strong>the</strong><br />

grassland area has decreased considerably whilst <strong>the</strong> arable area has significantly increased. In part<br />

this is probably due to changes and improvements in <strong>the</strong> classification procedure, but o<strong>the</strong>r factors<br />

may also provide a stimulus; farm subsidies having a greater impact than urbanisation, <strong>the</strong> influence<br />

<strong>of</strong> planning restrictions on urban expansion and <strong>the</strong> reuse <strong>of</strong> central sites ra<strong>the</strong>r than green-field<br />

peripheral sites, or generally increased levels <strong>of</strong> agricultural production since <strong>the</strong> early <strong>19</strong>30s<br />

Depression. How far any <strong>of</strong> <strong>the</strong>se factors are important to <strong>the</strong> results is currently unclear. Overall <strong>the</strong><br />

results could well show <strong>the</strong> success <strong>of</strong> <strong>the</strong> green belt policy in this area, <strong>the</strong> aim <strong>of</strong> which is to<br />

prevent urban sprawl by keeping land permanently open. Observations can be made more generally<br />

on how visible urban expansion is on <strong>the</strong> national overview compared to <strong>the</strong> localised study.<br />

Fur<strong>the</strong>r work in this area will be to complete <strong>the</strong> currently funded work by making fur<strong>the</strong>r<br />

comparative analysis and publishing on both <strong>the</strong> methodologies used and <strong>the</strong> findings. It is also<br />

hoped to do more analysis once <strong>the</strong> new Land Cover 2007 data is released, using that in fur<strong>the</strong>r<br />

comparison work with material analysed under this research to extend <strong>the</strong> timeline as close as<br />

possible to <strong>the</strong> present day.<br />

Page | 117


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 2b: Urban and rural planning and modelling<br />

5. Acknowledgements<br />

Thanks go to The Frederick Soddy Trust for <strong>the</strong>ir funding <strong>of</strong> this work, Paul Carter for his assistance<br />

with <strong>the</strong> vector mapping and Brian Baily for his guidance and involvement.<br />

References<br />

Best R.H. (<strong>19</strong>81) Land Use and Living Space. Methuen, London.<br />

Rowley, T. (2006) The English Landscape in <strong>the</strong> Twentieth Century. Hambledon Continuum,<br />

London.<br />

Stamp L.D. (<strong>19</strong>48) The Land <strong>of</strong> Britain: Its use and misuse Longman Green, London.<br />

Walford R. (ed.) (<strong>19</strong>97) Land-Use <strong>UK</strong>: A Survey for <strong>the</strong> 21st Century Geographical Association,<br />

Sheffield.<br />

Website: A Vision <strong>of</strong> Britain through time (http://www.vision<strong>of</strong>britain.org.uk/maps/)<br />

Website: Land <strong>of</strong> Britain (http://riga.iso.port.ac.uk/django_projects/home/)<br />

Biographies<br />

Paula Aucott is a Senior <strong>Research</strong> Associate, involved with <strong>the</strong> GBH<strong>GIS</strong> Project since 2000. After<br />

graduating from <strong>the</strong> University <strong>of</strong> Birmingham she obtained an MA from Leicester and an MSc in<br />

<strong>GIS</strong> from Portsmouth. Her main research interests include historical <strong>GIS</strong>, historical land use,<br />

historical gazetteers and administrative unit ontologies.<br />

Humphrey Southall is a historical geographer with an MA and a PhD in Geography from <strong>the</strong><br />

University <strong>of</strong> Cambridge. Originally focused on British labour markets history and <strong>the</strong> origins <strong>of</strong> <strong>the</strong><br />

north-south divide, he has led <strong>the</strong> GBH<strong>GIS</strong> project since <strong>19</strong>94, moving from Queen Mary, University<br />

<strong>of</strong> London, to Portsmouth.<br />

Page | 118


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Map Mash-ups: What looks good must be good?<br />

Nurul Hawani Idris 1, 2,3 , Mike J. Jackson 2 , Robert J. Abrahart 3<br />

1 Department <strong>of</strong> Geoinformatics, Faculty <strong>of</strong> Geoinformation and Real Estate,<br />

Universiti Teknologi Malaysia, 81310 UTM Skudai Johor Malaysia<br />

Tel: +44 (0) 115 951 5445, Fax: + 44 (0) 115 951 3881<br />

Lgxhi1@nottingham.ac.uk<br />

2 Centre for Geospatial Science,<br />

The Nottingham Geospatial Building, University <strong>of</strong> Nottingham, Triumph Road,<br />

Nottingham NG72TU<br />

Tel: +44 (0) 115 951 5445, Fax: + 44 (0) 115 951 3881<br />

mike.jackson@nottingham.ac.uk<br />

1. Introduction<br />

3 School <strong>of</strong> Geography,<br />

Sir Clive Granger Building, University Park, Nottingham NG7 2RD<br />

Tel: +44 (0) 115 846 6145, Fax: + 44 (0) 115 951 5249<br />

bob.abrahart@nottingham.ac.uk<br />

KEYWORDS: map mash-ups, credibility, trust, web mapping, GeoWeb<br />

The Web 2.0 revolution, ‗Digital Earth‘ vision and recent technological advancements enabling <strong>the</strong><br />

general public to easily capture location-based data through, for example, smart phones, have all<br />

had a major impact on <strong>the</strong> culture <strong>of</strong> web-based mapping. Development <strong>of</strong> web mapping<br />

applications has been made simpler and can be achieved at little or no cost by <strong>the</strong> availability <strong>of</strong> free<br />

mash-up technology and mapping APIs. Map mash-ups can be developed and accessed by any webenabled<br />

citizen without <strong>the</strong> need for programming, mapping experience or geographic knowledge<br />

and as a consequence <strong>the</strong> number and range <strong>of</strong> users and uses <strong>of</strong> map mash-ups has grown<br />

exponentially.<br />

There is currently little research-based guidance to aid <strong>the</strong> amateur or ―citizen‖ producer or user <strong>of</strong><br />

map mash-ups and a paucity <strong>of</strong> information on how to design a mash-up so as to correctly and<br />

convincingly convey <strong>the</strong> data and ‗truth‘ <strong>of</strong> any scenario presented. Equally for <strong>the</strong> general user<br />

<strong>the</strong>re is minimal guidance on to how to evaluate <strong>the</strong> correctness, currency or value for extrapolation<br />

/ general applicability <strong>of</strong> <strong>the</strong> map mash-up data presented. Designers are exposed to <strong>the</strong> risks <strong>of</strong><br />

producing a product which fails to convey <strong>the</strong> message intended, which conveys <strong>the</strong> wrong message<br />

or which irrespective <strong>of</strong> <strong>the</strong> integrity <strong>of</strong> <strong>the</strong> data is not trusted or conveys an impression <strong>of</strong> low<br />

credibility purely due to <strong>the</strong> design and visualisation techniques adopted. At <strong>the</strong> same time, good<br />

design or an intuitive flair for effective communication through graphic material can give a level <strong>of</strong><br />

credibility and generate user trust which is not warranted by <strong>the</strong> data presented.<br />

The purpose <strong>of</strong> <strong>the</strong> research presented in this paper is to examine <strong>the</strong> main factors that impact users‘<br />

credence and trust in map mash-up information. The research investigated <strong>the</strong> relationship between<br />

users‘ judgment and critical elements <strong>of</strong> map design. It concludes that unless active design measures<br />

are taken <strong>the</strong> viewer <strong>of</strong> a map mash-up will instinctively assume that ―what looks good must be<br />

Page | 1<strong>19</strong>


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

good‖. This finding held even for geography-trained users <strong>of</strong> map mash-ups. The research also<br />

concluded that merely tagging <strong>the</strong> presented map mash-up with metadata, referencing <strong>the</strong> source <strong>of</strong><br />

included datasets, is insufficient to change most users‘ perceptions <strong>of</strong> <strong>the</strong> credibility <strong>of</strong> a mash-up.<br />

As a consequence our research is now pursuing lessons learnt in <strong>the</strong> field <strong>of</strong> media advertising and<br />

marketing on how to communicate a message to non-specialist consumers <strong>of</strong> a service.<br />

2. <strong>Research</strong> Context<br />

The massive availability <strong>of</strong> spatial information, downloaded at very little cost and with few<br />

limitations placed on access, provides a major social and technological advancement: although such<br />

material could be produced by individuals who have no knowledge or authority or credentials with<br />

regard to such things. This opportunity has led to increased concerns being expressed over <strong>the</strong> issue<br />

<strong>of</strong> believability and quality <strong>of</strong> information being conveyed. Hence, research on trust and credibility<br />

have become a major factor in <strong>the</strong> realm <strong>of</strong> web information quality (David and Jason, 2008) and<br />

human-computer interaction (HCI) (Fogg and Tseng, <strong>19</strong>99). This issue has been widely debated in<br />

areas such as e-commerce applications, health information, online media (news) and collaborative<br />

applications.<br />

The emergence <strong>of</strong> user-generated spatial content such as in <strong>the</strong> application <strong>of</strong> OpenStreetMap and<br />

map mash-ups pose similar issues <strong>of</strong> concern pertaining to trust, credibility, reliability and quality <strong>of</strong><br />

information produced in such products. A small number <strong>of</strong> researchers have investigated part <strong>of</strong> this<br />

particular issue e.g. Bishr and Mantelas (2008), Haklay (2010), and Stark (2010). Skarlatidou et al.<br />

(2010) have also investigated how web users trust <strong>the</strong> online map provided in a government<br />

domain. However, very little is understood about how web users evaluate <strong>the</strong> credibility<br />

(believability) <strong>of</strong> online map information, particularly in terms <strong>of</strong> <strong>the</strong> map mash-up medium.<br />

Therefore, <strong>the</strong> research questions for this present study are: (i) how web users evaluate <strong>the</strong><br />

credibility <strong>of</strong> online map mash-up information; (ii) what elements on <strong>the</strong> map mash-up influenced<br />

<strong>the</strong>ir believability (trust) judgements?<br />

The study proceeded by means <strong>of</strong> experimental questionnaires. The online map based questionnaire<br />

was chosen as a method to stimulate users in <strong>the</strong> context <strong>of</strong> making judgements based on <strong>the</strong>ir<br />

perceived credibility between two map mash-ups and obtain users‘ responses based on open and<br />

closed ended questions. This method was adopted from <strong>the</strong> HCI study <strong>of</strong> Fogg et al. (2003).<br />

3. Sample<br />

The sample in this study was a group <strong>of</strong> young adult web users aged 18 to 35. The sample drawn by<br />

applying a ‗convenience‘ (volunteer) sampling technique (Black, 2009) to members and nonmembers<br />

<strong>of</strong> <strong>the</strong> University <strong>of</strong> Nottingham. The self-completed questionnaires were distributed<br />

online using <strong>the</strong> student portal and mailing lists. This sample might suffer from poor coverage <strong>of</strong><br />

young (< 18) and middle aged (≥35) web users. The sample was split into two groups <strong>of</strong><br />

respondents: geoliterate and non-geoliterate users. The groups were coded based on <strong>the</strong> background<br />

information (academic and pr<strong>of</strong>essional courses attended or still ongoing) given by <strong>the</strong> respondents<br />

under user demographics. Geoliterate users were respondents who had a background in geography,<br />

cartography, remote sensing, land surveying or geographic information science: <strong>the</strong> remainder were<br />

classified as non-geoliterate users.<br />

Table 1 show <strong>the</strong> independent variables used in each experiment. These variables, adopted from<br />

Fogg et al. (2003) and a pilot survey (conducted earlier), were designed as a list <strong>of</strong> factors that might<br />

influence users judgement in a set <strong>of</strong> multiple choice and ranking questions (ranked based on <strong>the</strong><br />

level <strong>of</strong> importance). Table 2 shows <strong>the</strong> respondents‘ demographic information for each survey.<br />

Page | 120


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Experiment<br />

1<br />

Experiment<br />

2<br />

Table 1: Factors explored in each experiment<br />

Visual cues (peripheral factors) Metadata (critical factors)<br />

Colour (colour convention, colour Map producer (logo)<br />

coding)<br />

Symbol (clarity, symbol convention)<br />

Visual attractiveness<br />

Colour (scheme)<br />

The supplier <strong>of</strong> foreground spatial<br />

Design Look (pr<strong>of</strong>essional, amateur) data (textual descriptions)<br />

Usefulness (fit for purpose)<br />

The affiliated organisation with <strong>the</strong><br />

Website (logo)<br />

Table 2: The respondents‘ demographic information for each experiment<br />

Experiment 1<br />

(n = 133)<br />

Experiment 2<br />

(n = 102)<br />

Number <strong>of</strong> respondents Age<br />

Geoliterate<br />

users<br />

Non-geoliterate users 18 to 21 years 22 to 35 years<br />

31 102 62 % 38 %<br />

43 59 33 % 67 %<br />

Figure 1 shows <strong>the</strong> dataset (map mash-up) used in Experiment 1. The red circles highlight <strong>the</strong><br />

elements that differ between <strong>the</strong> two maps. In this experiment, respondents were required to suggest<br />

<strong>the</strong> safest route for ambulance access from <strong>the</strong> map mash-up that <strong>the</strong>y chose as having more<br />

believable information.<br />

3. Results<br />

3.1 Analysis <strong>of</strong> Experiment 1<br />

Figure 1: The map mash-up used for Experiment 1.<br />

Respondents‘ decisions were most influenced by symbol type (94%) and colour coding (92%);<br />

followed by visual attractiveness (75%) and label <strong>of</strong> map producer (author) (38%). There is no<br />

evidence to suggest that geoliterate and non-geoliterate user responses are statistically different on<br />

<strong>the</strong> question <strong>of</strong> map producer. From <strong>the</strong> Mann- Whitney statistical test (Field, 2009:550), <strong>the</strong><br />

importance levels <strong>of</strong> map producer element rated by <strong>the</strong> geoliterate users (median = 1.0, 95%<br />

Page | 121


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

confidence interval (CI)= 1.0 (not important) to 2.0(slightly important)) did not differ significantly<br />

from <strong>the</strong> non-geoliterate users (median = 1.0, 95% CI= 1.0 to 1.0); Mann-Whitney test (U) =<br />

1440.5, z-score (z) = -0.8, effect size (r) =-0.07.<br />

3.2 Analysis <strong>of</strong> Experiment 2<br />

Frequency analysis was applied to <strong>the</strong> results. Figure 2 indicates <strong>the</strong> results in percentage format<br />

according to <strong>the</strong> experimental context. Each context in this experiment was different in terms <strong>of</strong><br />

colour scheme, symbol design, supplier <strong>of</strong> foreground data and site affiliation (<strong>the</strong>se two factors are<br />

referred to as ―authority element‖) (See Table 3). In this experiment, respondents were required to<br />

choose a map mash-up to assist <strong>the</strong>ir navigation on a self-guided campus tour and had to select and<br />

rank those factors that influenced <strong>the</strong>ir decision.<br />

Context 2<br />

Table 3: The series <strong>of</strong> maps used in Experiment 2<br />

Site was affiliated with: Ordnance Survey University <strong>of</strong> Nottingham<br />

Context 3<br />

Foreground data supplier: City Council Students‘ Union<br />

Context 5<br />

Site was affiliated with: Starbucks C<strong>of</strong>fee Shop Google<br />

Page | 122


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

users' perceived credibility judgements in each experimental<br />

context colour scheme<br />

100<br />

90<br />

80<br />

70<br />

60<br />

50<br />

40<br />

Figure 2: The frequencies <strong>of</strong> factors that influenced users‘ perceived credibility judgement<br />

according to <strong>the</strong> experimental contexts.<br />

These findings show that visual cues (colour scheme and design look) were <strong>the</strong> most frequent<br />

elements that users measured as a basis in <strong>the</strong>ir credibility judgements in each experimental context.<br />

A higher percentage on <strong>the</strong> usefulness aspect indicates <strong>the</strong> occurrence <strong>of</strong> cognitive judgment on <strong>the</strong><br />

assessment. Table 4 indicates <strong>the</strong> findings according to <strong>the</strong> constraint on <strong>the</strong> context.<br />

Table 4: The results according to <strong>the</strong> constraint on <strong>the</strong> experimental map<br />

Map constraint References Results<br />

The two comparable maps<br />

(mash-ups) were significantly<br />

different in terms <strong>of</strong> <strong>the</strong> use <strong>of</strong><br />

colour scheme on <strong>the</strong> map<br />

Percent The Frequencies <strong>of</strong> factors (in percentage) that influenced<br />

30<br />

20<br />

10<br />

The comparable maps were in<br />

<strong>the</strong> same level <strong>of</strong> visual<br />

appearances (for example <strong>the</strong><br />

aspect <strong>of</strong> colour scheme and<br />

design look were not an issue<br />

if <strong>the</strong> two maps were using<br />

combinations <strong>of</strong> dull colours on<br />

<strong>the</strong> features)<br />

0<br />

context 1 context 2 context 3 context 4 context 5<br />

colour scheme 87 88 55 76 86<br />

usefulness 80 55 83 78 61<br />

Design look 79 68 61 60 67<br />

Authority 36 52 48 64 54<br />

Context 2 and Context 5 Average <strong>of</strong> 80 % <strong>of</strong> samples<br />

decisions were influenced by <strong>the</strong><br />

colours and 50 % influenced by<br />

<strong>the</strong> authority element<br />

usefulness<br />

Design look<br />

Authority<br />

Context 3 Average <strong>of</strong> 50 % <strong>of</strong> <strong>the</strong> samples<br />

decisions were influenced by <strong>the</strong><br />

authority element (map producer),<br />

whilst 50 % <strong>of</strong> <strong>the</strong>m influenced by<br />

<strong>the</strong> color, but approx. <strong>of</strong> 70% <strong>of</strong><br />

sample credibility judgements<br />

were based on <strong>the</strong>ir perceived <strong>of</strong><br />

map usefulness<br />

Page | 123


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

4. Discussion<br />

The two surveys indicate that most respondents used visual cues to assess <strong>the</strong> credibility <strong>of</strong> map<br />

information. The selection and combination <strong>of</strong> colours, <strong>the</strong> design <strong>of</strong> map features and overall<br />

presentation influenced <strong>the</strong> initial perception <strong>of</strong> users on a map at <strong>the</strong> first time <strong>of</strong> contact. The<br />

colour scheme applied to <strong>the</strong> map appears to be <strong>the</strong> main factor in influencing user decisions. This<br />

may be because it also plays an important role in assisting users in reading a map. It provides a<br />

greater impact on users‘ understanding on <strong>the</strong> displayed information. This is also in line with <strong>the</strong><br />

study done by Fogg et al. (2003:58) which posits that looking good is <strong>of</strong>ten interpreted as being<br />

good and <strong>the</strong>refore credible because visual cues are highly noticeable.<br />

However, according to David and Jason (2008), this effect would operate in <strong>the</strong> first few seconds <strong>of</strong><br />

observation and contribute to <strong>the</strong> users‘ first impression <strong>of</strong> <strong>the</strong> online information; never<strong>the</strong>less as<br />

users proceed into scrutinising <strong>the</strong> contents, <strong>the</strong>y would rely on cognitive levels <strong>of</strong> judgement.<br />

However, from this present study, even though users scrutinised <strong>the</strong> contents, it was less likely that<br />

<strong>the</strong> authority elements were being measured (considered). Respondents tend to measure <strong>the</strong><br />

usefulness <strong>of</strong> information during this phase. This was indicated in <strong>the</strong> finding that on average 40%<br />

(Experiment 1) and 50% (Experiment 2) <strong>of</strong> respondents measured <strong>the</strong> authority elements on <strong>the</strong><br />

map; whilst on average 70% measured <strong>the</strong> credibility judgement based on <strong>the</strong> usefulness <strong>of</strong> <strong>the</strong> map<br />

information.<br />

The findings <strong>of</strong> this study are contradicted by <strong>the</strong> study <strong>of</strong> Skarlatidou et al. (2010). From that study,<br />

user trust <strong>of</strong> a website is due to <strong>the</strong> sources (data provided by government) even if <strong>the</strong> aes<strong>the</strong>tic and<br />

usability element on <strong>the</strong> map are poor. But this is not <strong>the</strong> case if <strong>the</strong> users did not notice <strong>the</strong> sources<br />

(authority) <strong>of</strong> <strong>the</strong> map information. Indeed, according to ‗Prominence-Interpretation Theory‘ (Fogg,<br />

2003), web users would judge <strong>the</strong> credibility <strong>of</strong> online information based on <strong>the</strong> element(s) that <strong>the</strong>y<br />

noticed (looked prominent) on a website.<br />

The present study assessed <strong>the</strong> impact between <strong>the</strong> authority <strong>of</strong> foreground data supplier and mashup<br />

producer, but did not assess <strong>the</strong> influence <strong>of</strong> data accuracy at a user‘s level <strong>of</strong> judgement. This is<br />

because <strong>the</strong> dataset used on <strong>the</strong> experimental contexts were identical in terms <strong>of</strong> <strong>the</strong> accuracy <strong>of</strong><br />

positioning, relative and logical; but was slightly different in terms <strong>of</strong> data completeness.<br />

Never<strong>the</strong>less, in <strong>the</strong> previous study <strong>of</strong> Scholz-Crane (<strong>19</strong>98), <strong>the</strong> majority <strong>of</strong> users were found to<br />

assess <strong>the</strong> accuracy and scope <strong>of</strong> information as it could be accessed visually and easily, ra<strong>the</strong>r than<br />

assessing <strong>the</strong> authority and currency, which required more effort and it was time consuming to<br />

check such items on <strong>the</strong> sidebar or read <strong>the</strong> textual description. The experiments conducted in this<br />

study did not provide evidence to suggest that substantial differences existed between <strong>the</strong> responses<br />

<strong>of</strong> geoliterate and non-geoliterate users in assessing map information credibility.<br />

The findings <strong>of</strong> this study may contribute to <strong>the</strong> area <strong>of</strong> research that is investigating how people<br />

evaluate credibility <strong>of</strong> online information, specifically <strong>of</strong> online map (mash-up) information.<br />

However, <strong>the</strong> findings might be restricted to represent an assessment <strong>of</strong> young adult map users due<br />

to <strong>the</strong> range <strong>of</strong> ages in <strong>the</strong> sample. The findings also might be different if all respondents were<br />

required to have a deep engagement with <strong>the</strong> task; in a situation that needs <strong>the</strong>m to perform<br />

credibility assessments critically in order to successfully achieve a certain objective, such as in a<br />

location based treasure hunt game or in a real-world critical situation. Therefore <strong>the</strong> findings <strong>of</strong> this<br />

study are restricted to suggest <strong>the</strong> main factors that might impact on map users‘ perceived credibility<br />

in lightweight and low risk applications; for instance in assessing a map (mash-up) to locate a<br />

house for rent, a restaurant to dine in, or a place to visit.<br />

Page | 124


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

In conclusion, it is possible to increase <strong>the</strong> users‘ perceived credibility <strong>of</strong> a map (mash-up) by using<br />

techniques which get <strong>the</strong>m to focus on <strong>the</strong> visual appearance aspect. Some map users critically<br />

determine <strong>the</strong> credibility <strong>of</strong> online map (mash-up) information, but it may only occur in certain<br />

circumstances and will depend on a person‘s level <strong>of</strong> shallow or deep engagement with respect to<br />

<strong>the</strong> task in-hand and <strong>the</strong>ir ability (knowledge) to process each element. Indeed, a disincentivised<br />

group <strong>of</strong> users are at risk <strong>of</strong> obtaining inaccurate and unidentified information as <strong>the</strong>ir judgement is<br />

dominated by <strong>the</strong> visual appearance ra<strong>the</strong>r than any structured quality assessment or reference to<br />

quality / fitness for purpose metadata.<br />

5. Future Work<br />

Future research will propose a framework to support <strong>the</strong> development <strong>of</strong> automated online<br />

credibility assessment <strong>of</strong> map mash-ups. The research will focus on <strong>the</strong> basic components <strong>of</strong> <strong>the</strong><br />

system, conceptual design and issues <strong>of</strong> implementation. Understanding <strong>of</strong> this framework could<br />

suggest, for example, <strong>the</strong> development <strong>of</strong> online automated credibility assessment on map mash-up<br />

information; this may reduce <strong>the</strong> required assessment tasks that occur in <strong>the</strong> manual credibility<br />

assessment process. Hence users could evaluate <strong>the</strong> credibility <strong>of</strong> map (mash-up) by making a quick<br />

decision based-on, for example, a traditional ―traffic light‖ icon which embeds an automated and<br />

behind <strong>the</strong> scenes analysis <strong>of</strong> critical elements.<br />

6. References<br />

Bishr, M. and Mantelas, L., (2008). 'A trust and reputation model for filtering and classifying<br />

knowledge about urban growth'. GeoJournal, 72 (3),: 229-237.<br />

Black, T .R., (2009). Doing Quantitative <strong>Research</strong> in <strong>the</strong> Social Sciences. London: SAGE, Great<br />

Britain.<br />

David, R. and Jason, H., (2008). ' Aes<strong>the</strong>tics and credibility in web site design'. Information<br />

Processing & Management, 44 (1),: 386-399.<br />

Field, A., (2009). Discovering Statistics Using SPSS. Dubai: SAGE, Dubai.<br />

Fogg, B .J., (2003). ‗Prominence-interpretation <strong>the</strong>ory: explaining how people assess credibility<br />

online'. CHI '03 extended abstracts on human factors in computing systems. ACM, Ft. Lauderdale,<br />

Florida, USA.: ACM.<br />

Fogg, B .J., Cathy, S., David, R .D., Leslie, M., Julianne, S. and Ellen, R .T., (2003). ' How do users<br />

evaluate <strong>the</strong> credibility <strong>of</strong> Web sites?: a study with over 2,500 participants'. <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> 2003<br />

conference on designing for user experiences. ACM, San Francisco, California, USA: ACM.<br />

Fogg, B .J. and Tseng, H., (<strong>19</strong>99) . 'The elements <strong>of</strong> computer credibility'. <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong><br />

SIGCHI conference on human factors in computing systems: <strong>the</strong> CHI is <strong>the</strong> limit. ACM, Pittsburgh,<br />

Pennsylvania, USA.: ACM.<br />

Haklay, M. (2010) How good is volunteered geographical information? A comparative study <strong>of</strong><br />

OpenStreetMap and Ordnance Survey datasets. Environment and Planning B: Planning and Design,<br />

37, 682-703.<br />

Scholz-Crane, A., (<strong>19</strong>98) . 'Evaluating <strong>the</strong> Future: A Preliminary Study <strong>of</strong> <strong>the</strong> Process <strong>of</strong> How<br />

Undergraduate Students Evaluate Web Sources'. Journal Reference Services Review, 26 (3/4), :53-<br />

60.<br />

Skarlatidou, A., Haklay, M., Cheng, T. and Francis, N., (2010). 'Trust in Web <strong>GIS</strong>: A Preliminary<br />

Investigation <strong>of</strong> <strong>the</strong> Environment Agency's WIYBY Website with non-expert users'. In: Haklay, M.,<br />

Morley, J. and Rahemtulla, H. (eds). <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> <strong>GIS</strong> <strong>Research</strong> <strong>UK</strong> 18th <strong>Annual</strong> <strong>Conference</strong><br />

Page | 125


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

<strong>GIS</strong>R<strong>UK</strong> 2010. University College London (UCL), London: UCL, <strong>UK</strong>.<br />

Stark, H.J., (2010). ' Quality assessment <strong>of</strong> VGI based on Open Web Map Services and ISO/TC 211<br />

<strong>19</strong>100-family standards'. Free and Open Source S<strong>of</strong>tware for Geomatics <strong>Conference</strong> FOSS4G 2010.<br />

Barcelona. Sept 6th - 9 th 2010. (<strong>Conference</strong> Paper)<br />

7. Acknowledgements<br />

This work is fully funded by <strong>the</strong> Universiti Teknologi Malaysia (UTM). This paper submission is<br />

supported by <strong>the</strong> University <strong>of</strong> Nottingham research student grant.<br />

8. Biography<br />

Nurul Hawani Idris is a research student examining <strong>the</strong> credibility issues <strong>of</strong> online map (mashup)<br />

information. She holds a bachelor and a master degree in Geoinformatics (Science).<br />

Page | 126


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Using Google Maps to collect spatial responses in a survey environment<br />

Nick Bearman 1 , Katy Appleton 1<br />

1 School <strong>of</strong> Environmental Sciences, University <strong>of</strong> East Anglia, Norwich, NR4 7TJ<br />

Tel. +44 (0)1603 591346 Fax. +44(0)1603 591327<br />

Email. n.bearman@uea.ac.uk, k.appleton@uea.ac.uk, www.nickbearman.me.uk<br />

ABSTRACT<br />

This paper examines <strong>the</strong> use <strong>of</strong> Google Maps-based tools to collect spatial responses from<br />

participants during academic research surveys conducted via <strong>the</strong> Internet. Using two recent<br />

examples from <strong>the</strong> University <strong>of</strong> East Anglia it discusses <strong>the</strong> online survey context and how Google<br />

Maps was used, issues surrounding <strong>the</strong> technical implementation <strong>of</strong> <strong>the</strong>se tools, processing and use<br />

<strong>of</strong> <strong>the</strong> collected data, and concludes with considerations for future research that might employ<br />

similar methods.<br />

KEYWORDS: surveys, spatial data collection, Internet mapping, Google Maps API<br />

1. Introduction<br />

Internet surveys are now a common way <strong>of</strong> ga<strong>the</strong>ring data for academic research and although <strong>the</strong>re<br />

are caveats about <strong>the</strong>ir use in terms <strong>of</strong> representativeness (Peng, 2001), this paper specifically<br />

considers <strong>the</strong> collection <strong>of</strong> spatial data responses within a survey, as opposed to <strong>the</strong> general use <strong>of</strong><br />

online surveys <strong>the</strong>mselves. In standard HTML forms and third-party (e.g. SurveyMonkey) surveys,<br />

responses are usually limited to <strong>the</strong> selection or ranking <strong>of</strong> one or more pre-set options and <strong>the</strong> input<br />

<strong>of</strong> free text. Spatial information, if elicited, is commonly in <strong>the</strong> form <strong>of</strong> postcodes or named<br />

locations such as towns. Some research applications, however, require more precise spatial<br />

information for later <strong>GIS</strong> analysis. This is generally not catered for within third-party tools, but can<br />

be achieved within bespoke surveys by using an Application Programming Interface (API) for an<br />

online mapping service such as Google Maps (Google, 2010a). An API is a set <strong>of</strong> 'building blocks' -<br />

tools, data structures and functions - allowing relatively easy programmatic access to and<br />

customisation <strong>of</strong> a web service.<br />

This paper outlines two examples from recent research at <strong>the</strong> University <strong>of</strong> East Anglia where <strong>the</strong><br />

Google Maps API (GMAPI) has been used to provide an interface for spatial data input within a<br />

survey. It discusses issues related to <strong>the</strong>se two specific implementations, followed by more general<br />

considerations about <strong>the</strong> use <strong>of</strong> such tools and <strong>the</strong>ir implications for academic research.<br />

2. Case studies<br />

2.1 Sonification <strong>of</strong> uncertainty in spatial data<br />

This case study compared visual and sonic methods <strong>of</strong> representing uncertainty in spatial data,<br />

specifically <strong>UK</strong> Climate Projections 2009 (<strong>UK</strong>CP09) data (Jenkins et al., 2009). One way <strong>of</strong><br />

showing uncertainty information more effectively is to use sound in combination with vision, and<br />

this has been addressed from a <strong>the</strong>oretical and practical point <strong>of</strong> view (Krygier, <strong>19</strong>94; Fisher, <strong>19</strong>94).<br />

Sonification can be particularly useful to show an extra layer <strong>of</strong> data, if no more could be added<br />

visually without obscuring <strong>the</strong> underlying data (for more details, see Bearman & Lovett, 2010).<br />

Respondents were shown maps overlaid with datasets from <strong>the</strong> <strong>UK</strong>CP09 series and asked to<br />

highlight areas meeting certain criteria (e.g. exceeding a stated value) using a ‗paintbrush type‘ tool<br />

Page | 127


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

over <strong>the</strong> Google Maps interface displaying a KML (Keyhole Markup Language (OGC, 2010)) file<br />

<strong>of</strong> <strong>the</strong> <strong>UK</strong>CP09 data (Figure 1). The evaluation 4 also had a number <strong>of</strong> multiple choice questions. All<br />

responses were written to a MySQL database.<br />

Figure 1. Screenshot <strong>of</strong> Google Maps interface with <strong>UK</strong>CP09 overlay and an area<br />

highlighted. The respondents were not allowed to pan or zoom <strong>the</strong> map.<br />

An Internet-based approach was chosen due to its simpler s<strong>of</strong>tware requirements (previous work<br />

used Arc<strong>GIS</strong> 9.2 (Bearman and Lovett, 2010)). Evaluations were mostly run in small groups (6-8<br />

people) and were followed by a discussion session where qualitative data on participants views <strong>of</strong><br />

<strong>the</strong> sonification and interface were ga<strong>the</strong>red. Additionally, <strong>the</strong> survey could be completed remotely,<br />

which was particularly useful for users <strong>of</strong> <strong>the</strong> <strong>UK</strong>CP09 data (such as local authority policy-makers),<br />

who are geographically dispersed.<br />

2.2 Countryside recreation in <strong>the</strong> Norfolk Broads<br />

This case study investigated locations for countryside recreation in <strong>the</strong> River Ant catchment <strong>of</strong> <strong>the</strong><br />

Norfolk Broads. Participation was invited via email contacts and relevant online forums; 71<br />

responses were received online, plus 66 <strong>of</strong>fline (which are not considered fur<strong>the</strong>r in this paper). It<br />

4 Available at: http://sonicsurv.uea.ac.uk<br />

Page | 128


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

used a simple form-based questionnaire 5 (employing HTML, PHP and JavaScript) featuring<br />

multiple-choice and free-text responses in addition to a map-based question; user responses were<br />

written directly to a MySQL database. The map-based question (Figure 2) used GMAPI to request a<br />

single point, line or area feature to show <strong>the</strong> preferred location for a chosen activity. The spatial<br />

response was requested in order to allow locations to be identified in detail and analysed alongside<br />

existing datasets to investigate <strong>the</strong> importance <strong>of</strong> landscape and o<strong>the</strong>r factors for user preference.<br />

GMAPI was also used on <strong>the</strong> introductory page, providing a navigable map to illustrate <strong>the</strong> area <strong>of</strong><br />

interest. The survey only examined recreation preferences and did not ask respondents to comment<br />

on <strong>the</strong> map interface.<br />

Figure 2. Screenshot <strong>of</strong> Google Maps interface requesting input. The respondents were<br />

allowed to pan and zoom <strong>the</strong> map, and choose <strong>the</strong> base layer.<br />

3. Why use an online survey?<br />

An online survey was used for reasons similar to those discussed by Wherrett (<strong>19</strong>99). Primarily it<br />

was to access a wider population than would be possible from fieldwork alone, particularly in terms<br />

<strong>of</strong> including recreation away from obvious facilities such as visitor centres. Web links are easily<br />

distributed via contacts, online forums and social media, and an online survey can also be completed<br />

at a convenient time. Finally, <strong>the</strong> ability to write survey responses directly to a database both saves<br />

time and prevents data entry errors.<br />

3.1 Why Google Maps API?<br />

5 Available at: http://www.env.uea.ac.uk/crs<br />

Page | 129


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

GMAPI was chosen for <strong>the</strong>se case studies largely due to <strong>the</strong> dominance <strong>of</strong> Google Maps in online<br />

mapping (Ellul et al., 2009; Hitwise, 2010a, 2010b), with <strong>the</strong> conclusion that this would give <strong>the</strong><br />

greatest chance <strong>of</strong> existing user familiarity with <strong>the</strong> interface and base mapping style. The<br />

cartography <strong>of</strong> <strong>the</strong> base maps is clearer in many ways than alternatives Bing Maps and Yahoo!<br />

Maps (O'Beirne, 2010), both <strong>of</strong> which also <strong>of</strong>fer an API (Micros<strong>of</strong>t, 2010; Yahoo!, 2010). The<br />

existence <strong>of</strong> resources to assist with development was <strong>the</strong> o<strong>the</strong>r main motivation for choosing<br />

GMAPI; online documentation, tutorials and user forums are more developed than for <strong>the</strong><br />

alternatives, and experience within <strong>the</strong> department was also a consideration.<br />

4. Related research examples<br />

Google Maps and Google Earth are well-used for presenting spatial information to varied audiences,<br />

both in an <strong>of</strong>ficial capacity (e.g Brent Council, 2010; Westminster City Council, 2010) and via more<br />

informal "mashups" (e.g. MapTube 2010; Google Maps Mania 2010). They are also used to collect<br />

spatial information for re-presentation on <strong>the</strong> same online mapping base, i.e. Volunteered<br />

Geographic Information (see Goodchild 2007; Heipke, 2010). However, a literature search has not<br />

revealed o<strong>the</strong>r work that has, as part <strong>of</strong> a questionnaire, specifically requested (and used in later <strong>GIS</strong><br />

analysis) <strong>the</strong> sort <strong>of</strong> precise spatial information ga<strong>the</strong>red in <strong>the</strong> two case studies. Ellul et al. (2009)<br />

report on a community mapping website requesting and sharing user-entered spatial data; storage<br />

methods do allow future <strong>GIS</strong> analysis <strong>of</strong> that data, but such work is not <strong>the</strong> project's focus. Rosser<br />

(2010) used GMAPI within a Facebook application to ga<strong>the</strong>r data on vernacular areas, extending<br />

earlier work on capturing fuzzy areas (Evans & Waters, 2008) using raster data. This paper,<br />

contrastingly, focuses on <strong>the</strong> input <strong>of</strong> more definite (vector) location data.<br />

5. Discussion - use <strong>of</strong> Google Maps API within <strong>the</strong> case studies<br />

5.1 Setting up <strong>the</strong> survey - design/coding<br />

Nei<strong>the</strong>r <strong>of</strong> <strong>the</strong> authors had a great deal <strong>of</strong> coding experience and extensive use was made <strong>of</strong> online<br />

resources to make <strong>the</strong> maps look and behave as desired. The recreation survey‘s spatial question<br />

was implemented based upon two tutorials covering <strong>the</strong> creation <strong>of</strong> a digitiser function using<br />

GMAPI v2, and interfacing GM with a MySQL database (Google, 2010b, 2010c respectively), with<br />

additional assistance from <strong>the</strong> Google Groups help forum. The sonification survey used <strong>the</strong> same<br />

examples as a starting point, moving to v3 <strong>of</strong> <strong>the</strong> GMAPI and utilising Flash to handle <strong>the</strong> sound<br />

(Ribeiro Amigo, 2006).<br />

In general, learning to use <strong>the</strong> API for <strong>the</strong>se applications was not significantly complicated, but did<br />

require additional time. Some problems were encountered, such as combining code from different<br />

versions <strong>of</strong> <strong>the</strong> API and ensuring browser compatibility (see below), but <strong>the</strong>se were largely resolved<br />

through online resources. On <strong>the</strong> whole, coding a robust and user-friendly survey took longer than<br />

anticipated in both case studies; while this was not solely due to GMAPI, <strong>the</strong> additional<br />

consideration <strong>of</strong> creating a clear, map interface that would elicit <strong>the</strong> required information did add to<br />

<strong>the</strong> time taken.<br />

5.2 Compatibility and reliability<br />

Both case studies were developed in Firefox 3.5/3.6. Wider compatibility with o<strong>the</strong>r browsers was<br />

important for <strong>the</strong> recreation case study, and Internet Explorer had to be forced to work in<br />

compatibility mode. The more complex coding for sonification would have required significantly<br />

more work, and since <strong>the</strong> survey design meant that Firefox was sufficient, fur<strong>the</strong>r compatibility was<br />

not pursued. For larger surveys it would be necessary to test GMAPI implementation using<br />

Page | 130


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

alternative browsers and hardware, to ensure accessibility.<br />

Both case studies were reliant on Google Maps performing as expected when <strong>the</strong> survey was taken.<br />

For example, map tiles may be received slowly or not at all, leaving gaps in <strong>the</strong> base map or<br />

potentially causing <strong>the</strong> survey to be abandoned – feedback on this issue was not solicited.<br />

The use <strong>of</strong> any third-party map service is vulnerable to data updates. Visual content may be updated,<br />

but perhaps more important is spatial registration, particularly imagery (Goodchild, 2009) – low<br />

accuracy could be significant for surveys involving digitising. The GMAPI interface used to<br />

customise <strong>the</strong> interaction with <strong>the</strong> map may also change. The sonification study suddenly began to<br />

display additional, unwanted controls. As many o<strong>the</strong>r API users were affected, solutions were<br />

available through relevant forums, but this does highlight <strong>the</strong> issue <strong>of</strong> relying on a service that may<br />

change without notice.<br />

In general, both <strong>of</strong> <strong>the</strong> surveys worked well and <strong>the</strong>re were no known major problems with browser<br />

compatibility or Google reliability. The most significant impact was when <strong>the</strong> network connection<br />

was disrupted during some <strong>of</strong> <strong>the</strong> sonification evaluation sessions causing <strong>the</strong>m to be rescheduled.<br />

However this problem can be considered a potential drawback <strong>of</strong> any online survey methodology.<br />

5.3 The maps in use<br />

In both case studies, information was overlaid on <strong>the</strong> base map using a KML file. The KML Layer<br />

class <strong>of</strong> GMAPI is perhaps less well developed than o<strong>the</strong>r classes, and <strong>the</strong>re were some interaction<br />

problems. In <strong>the</strong> countryside survey clicking on <strong>the</strong> map to digitise also cleared <strong>the</strong> study area<br />

boundary from <strong>the</strong> map. This may have been beneficial in removing map clutter while digitising,<br />

but for <strong>the</strong> sonification study <strong>the</strong> KML needed to remain and so <strong>the</strong> click had to be handled<br />

differently. The documentation was unclear as to whe<strong>the</strong>r this behaviour was intended. The<br />

sonification data occasionally required a browser refresh to load all KML tiles, but this was less<br />

problematic in a supervised survey.<br />

The countryside survey also revealed problems digitising concave polygons: Google Maps<br />

interprets mouse clicks within <strong>the</strong> presumed area <strong>of</strong> an unfinished polygon as a click on <strong>the</strong> polygon<br />

feature itself, ra<strong>the</strong>r than a click on <strong>the</strong> underlying map, and is unable to retrieve latitude and<br />

longitude values. This is believed to be solvable with more complex coding, but time limitations<br />

meant that instead an explanatory message and request to click beyond <strong>the</strong> feature being drawn was<br />

implemented, possibly frustrating users.<br />

5.4 Data obtained<br />

The spatial data obtained from <strong>the</strong> recreation survey was written to a database table, subsequently<br />

imported to Arc<strong>GIS</strong> as a collection <strong>of</strong> points, and reconstructed as lines or polygons where<br />

necessary. This type <strong>of</strong> simple output is able to be processed using any <strong>of</strong> <strong>the</strong> increasingly available<br />

tools for GPS data.<br />

Analysis continued by buffering <strong>the</strong> point and line features and combining <strong>the</strong>m with <strong>the</strong> polygons<br />

to create a ―heat map‖ showing how many times each area had been digitised. It is clear that <strong>the</strong>re<br />

are issues with <strong>the</strong> accuracy <strong>of</strong> digitising, which will have implications for <strong>the</strong> ongoing analysis.<br />

Examination <strong>of</strong> <strong>the</strong> line data, for example, revealed examples for waterborne activities that were<br />

400m from <strong>the</strong> river. A useful refinement for any future use <strong>of</strong> this GMAPI code would be to record<br />

which zoom level and base map settings <strong>the</strong> user chose while digitising. Asking for additional text<br />

information (e.g. "riverside walk from X to Y") could provide fur<strong>the</strong>r support.<br />

Page | 131


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

The sonification case study stored users' input as individual points in <strong>the</strong> MySQL database, to be<br />

processed later into a surface layer and compared with <strong>the</strong> ‗correct‘ answer. Data collection is<br />

ongoing for this case study, and a full analysis will be included in <strong>the</strong> presentation. Initial analysis<br />

showed that using sound to reinforce data shown visually increased participants ability to select <strong>the</strong><br />

correct area.<br />

5.5 Methodological issues for academic research<br />

Academic research should be reproducible, and in addition to a standard methodology report, it is<br />

natural to include a demonstration <strong>of</strong> any online surveys used. With any work based on computer<br />

technology it is inevitable that at some point <strong>the</strong> code or files created will no longer work; for<br />

example, each major version <strong>of</strong> GMAPI will only be supported by Google for 3 years after<br />

deprecation (Google, 2010d), and browser standards and compatibilities will also develop.<br />

Therefore, as well as a copy <strong>of</strong> <strong>the</strong> source code (fully commented), a conceptual representation <strong>of</strong><br />

<strong>the</strong> processes involved, such as a flow chart, should be kept, to allow <strong>the</strong> data collection methods to<br />

be replicated with tools that are available in <strong>the</strong> future. A fur<strong>the</strong>r possible solution would be to have<br />

a video <strong>of</strong> <strong>the</strong> working application to show <strong>the</strong> user experience, including sound in <strong>the</strong> case <strong>of</strong> <strong>the</strong><br />

sonification study.<br />

6. Conclusions<br />

Google Maps API <strong>of</strong>fers a useful, reasonably accessible (to authors) and familiar (to respondents)<br />

way to elicit spatially-referenced responses within research surveys. There are some limitations to<br />

<strong>the</strong> information that can be obtained, and considerations to be made when designing a survey using<br />

GMAPI, that require careful thought about <strong>the</strong> intended use.<br />

Any survey that is to be completed unsupervised needs to be carefully designed. There must be clear<br />

instructions and robust error trapping for both <strong>the</strong> non-spatial and spatial parts <strong>of</strong> <strong>the</strong> survey but this<br />

may well be more important for map-based input because respondents are likely to be less familiar<br />

with providing this type <strong>of</strong> information. Related to this, <strong>the</strong>re is a need for research into <strong>the</strong> question<br />

<strong>of</strong> digitising accuracy, and whe<strong>the</strong>r this can be improved through on-screen instructions or map<br />

interface settings.<br />

Technically, <strong>the</strong>re are unknowns related to <strong>the</strong> persistence and consistency <strong>of</strong> <strong>the</strong> API over time that<br />

increase <strong>the</strong> importance <strong>of</strong> conceptual documentation <strong>of</strong> any use <strong>of</strong> <strong>the</strong> API within a survey, as well<br />

as a reference copy <strong>of</strong> <strong>the</strong> final code. However, if use <strong>of</strong> GMAPI for survey purposes continues to<br />

be developed and progress shared, it has <strong>the</strong> potential to become an accessible and extremely useful<br />

tool for research data collection.<br />

7. Acknowledgements<br />

Nick Bearman‘s research has been conducted as part <strong>of</strong> ESRC/NERC PhD Studentship No.<br />

ES/F012454/1 with additional financial support from Ordnance Survey. Katy Appleton's research is<br />

funded by <strong>the</strong> ESRC under an Interdisciplinary Early Career Fellowship, reference RES-229-27-<br />

0006.<br />

8. References<br />

Bearman N and Lovett A (2010). Using Sound to Represent Positional Accuracy <strong>of</strong> Address<br />

Locations. The Cartographic Journal, 47(4), 308-314.<br />

Page | 132


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Brent Council (2010). Interactive Maps. Available from http://www.brent.gov.uk/maps (accessed<br />

17 th December 2010).<br />

Ellul C, Haklay M, Francis L and Rahemtulla H (2009). A Mechanism to Create Community Maps<br />

for Non-Technical Users. <strong>Proceedings</strong> <strong>of</strong> <strong>the</strong> International <strong>Conference</strong> on Advanced Geographic<br />

Information Systems and Web Services, GEOWS 2009, Article number 4782704, p. 129-134.<br />

Evans A and Waters T (2008). Mapping vernacular geography: web-based <strong>GIS</strong> tools for capturing<br />

―fuzzy‖ or ―vague‖ entities. International Journal <strong>of</strong> Technology, Policy and Management, 7(2),<br />

134-150.<br />

Fisher P F (<strong>19</strong>94). Hearing <strong>the</strong> Reliability in Classified Remotely Sensed Images. Cartography and<br />

Geographic Information Systems, 21(1), 31-36.<br />

Goodchild M (2007). Citizens as sensors: <strong>the</strong> world <strong>of</strong> volunteered geography. GeoJournal, 69(4),<br />

211-221, DOI: 10.1007/s10708-007-9111-y.<br />

Goodchild M (2009). Virtual Geographic Environments as Collective Constructions. In: Li H and<br />

Batty M (eds) Virtual Geographic Environments. Science Press: Beijing, 15-24.<br />

Google (2010a). Google Maps API Family. Available from: http://code.google.com/apis/maps/<br />

(accessed 17 th December 2010).<br />

Google (2010b). Creating a Simple Digitizer Using <strong>the</strong> Google Maps API. Available from:<br />

http://code.google.com/apis/maps/articles/ezdigitizer.html (accessed 17 th December 2010).<br />

Google (2010c). From Info Windows to a Database: Saving User-Added Form Data. Available<br />

from: http://code.google.com/apis/maps/articles/phpsqlinfo.html (accessed 17 th December 2010).<br />

Google (2010d). Google Maps/Google Earth APIs Terms Of Service, Section 4.4: Termination <strong>of</strong><br />

<strong>the</strong> Service. Available from: http://code.google.com/apis/maps/terms.html#section_4_4 (accessed<br />

17 th December 2010).<br />

Google Maps Mania (2010). Available from: http://googlemapsmania.blogspot.com/ (accessed 17 th<br />

December 2010).<br />

Heipke C (2010). Crowdsourcing geospatial data. ISPRS Journal <strong>of</strong> Photogrammetry and Remote<br />

Sensing, 65(6), 550-557.<br />

Hitwise (2010a). Hitwise United Kingdom Data Centre. Available from:<br />

http://www.hitwise.com/uk/datacentre/main/dashboard-7323.htm (accessed 17 th December 2010).<br />

Hitwise (2010b). Hitwise US/Worldwide and Travel Data Centre. Available from:<br />

http://www.hitwise.com/us/datacenter/main/ (accessed 17 th December 2010).<br />

Jenkins G, Murphy J, Sexton D, Lowe J, Jones P and Kilsby C (2009). <strong>UK</strong> Climate Projections:<br />

Briefing report. Met Office Hadley Centre, Exeter, <strong>UK</strong>. Available from:<br />

http://ukclimateprojections.defra.gov.uk/ content/view/644/500/ (accessed 17 th December 2010).<br />

Krygier J B (<strong>19</strong>94). Sound and Geographic Visualization. In: Visualization in Modern Cartography.<br />

Elsevier Science: Oxford, <strong>UK</strong>, 149-166.<br />

MapTube (2010). Available from: http://www.maptube.org (accessed 17 th December 2010).<br />

Micros<strong>of</strong>t (2010). Bing Maps Developer Resources. Available from:<br />

http://www.micros<strong>of</strong>t.com/maps/developers/web.aspx (accessed 17 th December 2010).<br />

O'Beirne J (2010). A Brief Comparison <strong>of</strong> Google Maps, Bing Maps, & Yahoo! Maps. Available<br />

from: http://www.41latitude.com/post/557224600/map-comparison (accessed 17 th December 2010).<br />

OGC - Open Geospatial Consortium (2010). KML Standard. Available from:<br />

http://www.opengeospatial.org/ standards/kml (accessed 17 th December 2010).<br />

Page | 133


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Peng Z (2001). Internet <strong>GIS</strong> for public participation. Environment and Planning B: Planning and<br />

Design 28(6), 889-905.<br />

Ribeiro Amigo G (2006). Javascript Sound Kit. Available from: http://jssoundkit.sourceforge.net/<br />

(accessed 17 th December 2010).<br />

Rosser J and Morley J (2010). Rate-my-place: a social network application for<br />

crowd-sourcing vernacular geographic areas. <strong>Proceedings</strong> <strong>of</strong> <strong>GIS</strong>R<strong>UK</strong> 2010, London, <strong>UK</strong>, Available<br />

from: http://discovery.ucl.ac.uk/<strong>19</strong>284/1/<strong>19</strong>284.pdf (accessed 17 th December 2010), p.155-160.<br />

Westminster City Council (2010). Find it in Westminster. Available from:<br />

http://www.westminster.gov.uk/ (accessed 17 th December 2010).<br />

Wherrett J (<strong>19</strong>99). Issues in using <strong>the</strong> internet as a medium for landscape preference research.<br />

Landscape and Urban Planning, 45(4) 209-217.<br />

Yahoo! (2010) Yahoo! Maps Web Services. Available from: http://developer.yahoo.com/maps/<br />

(accessed 17 th December 2010).<br />

9. Biography<br />

Nick Bearman is studying for a PhD at UEA in Environmental Science, researching different<br />

methods <strong>of</strong> representing uncertainty using sound in a variety <strong>of</strong> spatial data environments. As well<br />

as <strong>UK</strong>CP09 data, <strong>the</strong>se include Ordnance Survey Address Layer 2 and representation <strong>of</strong> varying<br />

uncertainty <strong>of</strong> future landscape visualisation in a Virtual Reality setting.<br />

Katy Appleton is <strong>Research</strong> Officer for <strong>the</strong> SSEVREL (Social Science for <strong>the</strong> Environment - Virtual<br />

Reality Laboratory) facility in <strong>the</strong> School <strong>of</strong> Environmental Sciences at UEA. Her research interests<br />

include <strong>GIS</strong>-based landscape visualisation for environmental communication and decision-making,<br />

and multifunctional landscape management.<br />

Page | 134


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Using Open Source S<strong>of</strong>tware and Data to Teach Spatial Database Skills<br />

Claire Ellul 1<br />

1 Dept. <strong>of</strong> Civil, Environmental and Geomatic Engineering, University College London<br />

Gower Street, London, WC1E 6BT<br />

Tel. +44 (0) 20 7679 4118 Fax +44 (0) 20 7679 3042 c.ellul@ucl.ac.uk<br />

ABSTRACT<br />

This paper describes <strong>the</strong> development and evaluation <strong>of</strong> a self-paced course on Spatial Databases<br />

for students on an MSc. in Geographical Information Science. The material makes extensive use<br />

<strong>of</strong> Free and Open Source S<strong>of</strong>tware and Open Data (PostgreSQL/Post<strong>GIS</strong>, Quantum <strong>GIS</strong>, Open<br />

Street Map). Techniques including video instructions and ‗book‘ approaches to online<br />

learning were also trialled. The resulting material was evaluated by a cohort <strong>of</strong> 25 students in 2010,<br />

and <strong>the</strong>ir feedback (overall very positive) provides an interesting insight into suitable methods to<br />

employ when teaching technically focussed subjects to a cohort having differing background skill<br />

levels.<br />

1. Introduction<br />

KEYWORDS: Spatial Databases, FOSS, Open Data, Learning and Teaching<br />

The skills required to manage spatial data within a database are much sought-after by employers.<br />

This paper describes <strong>the</strong> development and assessment <strong>of</strong> a self-contained, self-paced spatial<br />

databases tutorial for students on an MSc. in <strong>GIS</strong> run by <strong>the</strong> Department <strong>of</strong> Civil, Environmental<br />

and Geomatic Engineering (CEGE) at University College London (UCL). The material developed<br />

makes use <strong>of</strong> Free and Open Source S<strong>of</strong>tware (FOSS) and data throughout and is suitable for both<br />

in-class and totally independent learning. It introduces students to <strong>the</strong> concepts <strong>of</strong> spatial (map) data<br />

and how this is stored, indexed and managed in a database, allows students to engage with this<br />

material through material that mixes <strong>the</strong>oretical and practical work and provides students with <strong>the</strong><br />

skills to take <strong>the</strong> knowledge and techniques forward and apply <strong>the</strong>m to o<strong>the</strong>r project related work<br />

including dissertations and <strong>the</strong>ses.<br />

2. Background<br />

2.1 The Importance <strong>of</strong> Self-Paced Learning<br />

The teaching <strong>of</strong> spatial databases within <strong>the</strong> context <strong>of</strong> <strong>the</strong> CEGE MSc. in <strong>GIS</strong> is not new.<br />

However, <strong>the</strong> previous approach included a mixture <strong>of</strong> classroom based lectures followed by lab-<br />

based sessions. Concepts were presented in an abstract format in class, and not immediately<br />

applied in practice – <strong>the</strong>re was a gap <strong>of</strong> a week between class and lab. The lab sessions presented<br />

<strong>the</strong> students with a well-described series <strong>of</strong> step-by-step tasks to carry out, due to <strong>the</strong> constraints<br />

<strong>of</strong> available lab time and highly differing background skills <strong>of</strong> <strong>the</strong> students, ranging from expert<br />

computer programmers to users with only basic computer skills. In some cases, this approach<br />

resulted in some students taking a mechanistic approach to <strong>the</strong> topic - <strong>the</strong> lab sessions were not<br />

assessed and <strong>the</strong> step-by-step guidelines made it easy to complete <strong>the</strong> work without a real<br />

understanding <strong>of</strong> <strong>the</strong> underlying principles or processes. The more advanced students worked<br />

through <strong>the</strong> sessions in 1-2 hours, leaving <strong>the</strong> students who were new to <strong>the</strong> subject demoralized,<br />

although <strong>the</strong>y were in fact able to finish within <strong>the</strong> allocated time.<br />

Page | 135


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

The overall result was, perhaps, a ―surface approach‖ to learning <strong>the</strong> subject (‗memorising <strong>the</strong><br />

course materials for <strong>the</strong> purposes <strong>of</strong> assessment‘, Richardson 2005). Some students <strong>the</strong>n<br />

struggled when attempting to translate <strong>the</strong> concepts encountered in class and in <strong>the</strong> lab to a response<br />

to <strong>the</strong> final assignment (which involves <strong>the</strong> design and build <strong>of</strong> a small spatial database).<br />

Given <strong>the</strong> importance <strong>of</strong> <strong>the</strong> topic to future employment in a <strong>GIS</strong> market place it was felt that it<br />

would be important to overcome <strong>the</strong> issues described and to move more students beyond a basic<br />

understanding (where learning is ‗acquiring information‘, Saljo <strong>19</strong>79 in Richardson 2005)<br />

towards <strong>the</strong> more complex internalization <strong>of</strong> learning (‗interpreting and understanding reality in a<br />

different way‘, Saljo <strong>19</strong>79 in Richardson 2005). A change in <strong>the</strong> method <strong>of</strong> teaching<br />

was required to encourage students towards a ―deep approach‖ to learning (Richardson 2005), and<br />

hence to <strong>the</strong> ability to apply what <strong>the</strong>y learn in o<strong>the</strong>r contexts and settings (<strong>the</strong> highest level <strong>of</strong><br />

learning, Biggs <strong>19</strong>96). In particular, <strong>the</strong> disparate skill sets <strong>of</strong> <strong>the</strong> students required a self-paced,<br />

stand-alone, student-centred (Shuell <strong>19</strong>86 in Biggs <strong>19</strong>96) approach. This should be delivered in<br />

such a way as to allow students to work externally to <strong>the</strong> classroom or lab, in <strong>the</strong>ir own time. FOSS<br />

and Open Data were seen as fundamental to achieving <strong>the</strong>se aims.<br />

2.2 Open Source S<strong>of</strong>tware and Open Data<br />

In <strong>the</strong> past, spatial data was produced and provided by ei<strong>the</strong>r governmental bodies, such as<br />

national mapping agencies, commercial data providers or o<strong>the</strong>r large organisations (Goodchild in<br />

Schuurman, 2009). It was provided in a top down manner as data products, such as data provided<br />

by <strong>the</strong> Ordnance Survey in <strong>the</strong> <strong>UK</strong>. Similarly, GI s<strong>of</strong>tware was provided through proprietary<br />

packages such as ESRI‘s Arc<strong>GIS</strong> and Pitney Bowes‘ MapInfo Pr<strong>of</strong>essional, incurring a license<br />

fee.<br />

However, with <strong>the</strong> emergence <strong>of</strong> <strong>the</strong> geographical information technologies forming part <strong>of</strong> Web<br />

Mapping 2.0 (Goodchild 2007, Haklay et al. 2008, Elwood 2008) modes <strong>of</strong> production and<br />

consumption <strong>of</strong> geographical information have changed. This trend has encompassed both <strong>the</strong><br />

s<strong>of</strong>tware packages with an emergence <strong>of</strong> FOSS and sources <strong>of</strong> data. Steiniger and Bocher (2009)<br />

note that Free<strong>GIS</strong>.org lists over 330 <strong>GIS</strong> FOSS products and a recent review by Sharp (2010)<br />

identified that 24 out <strong>of</strong> a possible 43 mappable quality <strong>of</strong> life indicators (from a list provided by<br />

<strong>the</strong> Audit Commission) were freely available on <strong>the</strong> <strong>UK</strong> Government Open Data service<br />

data.gov.uk.<br />

3. S<strong>of</strong>tware and Data<br />

Given <strong>the</strong> focus on developing material suitable for independent learning, it was important at <strong>the</strong><br />

outset <strong>of</strong> <strong>the</strong> project to identify s<strong>of</strong>tware and data that was freely available to students without<br />

cost. A review <strong>of</strong> <strong>the</strong> options resulted in <strong>the</strong> selection <strong>of</strong> Quantum <strong>GIS</strong> 1 and <strong>the</strong> PostgreSQL<br />

database with Post<strong>GIS</strong> 2 as suitable tools for this course. Free data from Open Street Map<br />

(OSM) 3 was also incorporated into <strong>the</strong> teaching material.<br />

Quantum <strong>GIS</strong> (Q<strong>GIS</strong>) was created in 2002 as a simple interface for <strong>the</strong> long-standing FOSS <strong>GIS</strong><br />

called GRASS. It now has a very large user community and has a growing developer base and<br />

<strong>the</strong> s<strong>of</strong>tware itself and <strong>the</strong> development process are well documented – <strong>of</strong> particular interest to <strong>the</strong><br />

<strong>GIS</strong> community is <strong>the</strong> ability to develop custom extensions in <strong>the</strong> Python Language (Steiniger<br />

and Bocher 2009).<br />

1 http:// http://www.qgis.org/ [Accessed 12 December 2010]; 2 http://postgis.refractions.net/ [Accessed 12 December 2010]; 3<br />

http://www.openstreetmap.org/ [Accessed 12 December 2010]<br />

Page | 136


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

PostgreSQL has been in development for 15 years and supports standard database data types<br />

(integers, text, dates and so forth) and conforms to <strong>the</strong> ANSI:SQL 2008 standard. Post<strong>GIS</strong> is an<br />

extension to PostgreSQL, and has been designed to add support for spatial objects to <strong>the</strong> database.<br />

These objects conform to <strong>the</strong> Open Geospatial Consortium‘s (OGC) Simple Features Specification.<br />

The use <strong>of</strong> this database thus allows <strong>the</strong> students to learn both standards-based Structured Query<br />

Language (SQL), which is used in all relational databases, and to incorporate details relating to<br />

standards-based spatial data handling in such a database.<br />

The OSM project started at UCL in 2004. Volunteers (currently over 180,000) contribute to<br />

create a free editable vector map <strong>of</strong> <strong>the</strong> world (see Haklay and Weber 2008 for a detailed<br />

description). For <strong>the</strong> purposes <strong>of</strong> this material, <strong>the</strong> road network dataset for <strong>the</strong> <strong>UK</strong> was selected,<br />

along with building polygons and points <strong>of</strong> interest, all for <strong>the</strong> Euston/Bloomsbury area <strong>of</strong><br />

London.<br />

4. Developing <strong>the</strong> Teaching and Assessment Material<br />

The virtual learning environment <strong>of</strong> choice at UCL is Moodle4, which allows a variety <strong>of</strong> material to<br />

be included for a specific course, ranging from books, PDF files, videos, online quizzes and<br />

questionnaires, audio and links to o<strong>the</strong>r websites. Following s<strong>of</strong>tware and data selection, <strong>the</strong> first<br />

part <strong>of</strong> content development involved <strong>the</strong> design <strong>of</strong> a step-by-step installation guide was<br />

developed to explain <strong>the</strong> process <strong>of</strong> s<strong>of</strong>tware download and installation for both Quantum <strong>GIS</strong> and<br />

PostgreSQL/Post<strong>GIS</strong>.<br />

A second element <strong>of</strong> course content consisted <strong>of</strong> an introduction to databases and spatial database<br />

<strong>the</strong>ory and was developed as a Moodle book (Figure 1). The contents included a brief<br />

introduction to standard database <strong>the</strong>ory, followed by specialist information on spatial databases.<br />

The book approach allows students to work through <strong>the</strong> material systematically at <strong>the</strong>ir own pace,<br />

moving forwards and backwards as required. A multiple-choice quiz was developed to allow<br />

students to test <strong>the</strong>ir understanding <strong>of</strong> <strong>the</strong> <strong>the</strong>ory.<br />

Figure 1 - A Sample Page from <strong>the</strong> Moodle Book on Spatial Databases<br />

The final component <strong>of</strong> <strong>the</strong> course included practical material. Three videos were created to<br />

instruct <strong>the</strong> students on <strong>the</strong> process <strong>of</strong> opening <strong>the</strong> database and importing <strong>the</strong> downloaded data<br />

4 http://www.moodle.com [Accessed 12 December 2010]<br />

Page | 137


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

or viewing in <strong>the</strong> <strong>GIS</strong>. A second Moodle book was <strong>the</strong>n developed to take <strong>the</strong> students through a<br />

number <strong>of</strong> step-by-step exercises related to <strong>the</strong> process <strong>of</strong> creating spatial tables in <strong>the</strong> database,<br />

adding data and querying <strong>the</strong> data in <strong>the</strong> database and through <strong>the</strong> <strong>GIS</strong>. Learning on this part <strong>of</strong> <strong>the</strong><br />

course was validated through second multiple-choice (Moodle based) quiz was developed to allow<br />

students to validate <strong>the</strong>ir understanding <strong>of</strong> <strong>the</strong> topic at <strong>the</strong> conclusion <strong>of</strong> <strong>the</strong> practical exercises.<br />

Along with <strong>the</strong> two quizzes, students were assessed through an assignment which asked <strong>the</strong><br />

students to evaluate how three OGC geometry comparisons (e.g. touch, disjoint, adjacent)<br />

operations are implemented in a spatial database <strong>of</strong> <strong>the</strong>ir choice, and <strong>the</strong>n create a small spatial<br />

database detailing <strong>the</strong>ir routes around UCL as <strong>the</strong>y moved from one lecture to ano<strong>the</strong>r. SQL<br />

scripts were required for <strong>the</strong> spatial database creation and data population.<br />

5. Outcomes<br />

The course was <strong>of</strong>fered to 20 students on <strong>the</strong> MSc. in Geographical Information Science in<br />

CEGE, in April 2010. All students managed to work <strong>the</strong>ir way through <strong>the</strong> material and complete<br />

<strong>the</strong> quizzes, scoring an average mark <strong>of</strong> 8.68/10 for <strong>the</strong> first quiz and 9.08/10 for <strong>the</strong> second quiz<br />

(Figure 2).<br />

Figure 2 - Results for Quizzes<br />

The average grade for <strong>the</strong> assignment was 72%, which is higher than previous years (where<br />

averages were around 65%). This may, however, be due to issues with <strong>the</strong> quizzes – student<br />

feedback indicated that <strong>the</strong> answers had been very quickly passed around <strong>the</strong> class.<br />

The students were asked to provide feedback on <strong>the</strong> module. Overall, <strong>the</strong>y found <strong>the</strong> s<strong>of</strong>tware<br />

easy to install (rated 4.0 on a scale <strong>of</strong> 1 to 5 where 5 was <strong>the</strong> easiest), <strong>the</strong> <strong>the</strong>oretical background<br />

easy to understand (3.9 on <strong>the</strong> same scale) and <strong>the</strong> practical exercises easy (3.9 again). They were<br />

asked to specify whe<strong>the</strong>r <strong>the</strong>y preferred self-paced independent learning, in-class guided learning or<br />

both, with a total <strong>of</strong> only 5% preferring in class exercises, 24% self-paced learning and 71% a<br />

mixture <strong>of</strong> both. On a scale <strong>of</strong> 1 to 5 where 5 is a very suitable approach, Moodle Pages<br />

rated 4.0, Video rated 4.2 and Moodle Books rated 4.4.<br />

Page | 138


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

6. Evaluation/Reflection<br />

The use <strong>of</strong> Open Source s<strong>of</strong>tware and data reflects a growing trend in <strong>GIS</strong>, in part driven by <strong>the</strong><br />

downturn in <strong>the</strong> economy which means that many users cannot afford <strong>the</strong> expensive licenses for <strong>the</strong><br />

mainstream desktop <strong>GIS</strong> and associated datasets. The material delivered on this course was thus<br />

considered very relevant by <strong>the</strong> students, and <strong>the</strong>y felt that it could be opened up to a broader<br />

audience. This is enhanced by <strong>the</strong> self-paced approach taken to <strong>the</strong> teaching material, which<br />

allows students to move backwards and forwards between sections. As students may install <strong>the</strong><br />

s<strong>of</strong>tware and data onto <strong>the</strong>ir own personal computers or laptops, <strong>the</strong> course is scalable to any number<br />

<strong>of</strong> students.<br />

The success <strong>of</strong> video technique was mixed – <strong>the</strong> use <strong>of</strong> video was very well received by <strong>the</strong> students<br />

but it was difficult to create video <strong>of</strong> sufficiently high quality. Use <strong>of</strong> <strong>the</strong> material in ano<strong>the</strong>r context<br />

(a one-week training course) lead to <strong>the</strong> realization that <strong>the</strong>re is a need to convert <strong>the</strong> video<br />

instructions into text format to provide students with a handbook for <strong>the</strong> course that can be accessed<br />

entirely <strong>of</strong>fline.<br />

A number <strong>of</strong> usability issues with Moodle were also encountered, in particular <strong>the</strong> difficulty <strong>of</strong><br />

transforming pre-existing Micros<strong>of</strong>t Word documents with embedded images into Moodle<br />

documents or books (links to <strong>the</strong> images were lost). A number <strong>of</strong> technical issues were also<br />

encountered when <strong>the</strong> students attempted to install s<strong>of</strong>tware on platforms o<strong>the</strong>r than those<br />

specified in <strong>the</strong> instructions.<br />

Given <strong>the</strong> very high average marks, fur<strong>the</strong>r work is required to develop quizzes that perhaps<br />

challenge <strong>the</strong> students more and provide a more in-depth assessment <strong>of</strong> <strong>the</strong> learning process<br />

(learning is currently assessed by a separate assignment which <strong>the</strong> students are required to submit<br />

after completing <strong>the</strong> course material). Additionally, course material will require a short process <strong>of</strong><br />

re-evaluation every year to ensure that both <strong>the</strong> open source technology and <strong>the</strong> data are still<br />

available as referenced, and that no significant version changes have occurred. It is anticipated<br />

that a revision <strong>of</strong> course content will be required in 3-5 years time to incorporate updates in <strong>the</strong> field<br />

<strong>of</strong> spatial databases.<br />

However, overall delivering <strong>the</strong> course in this manner has proved to be very successful, with <strong>the</strong><br />

Moodle environment providing a well-integrated tool <strong>of</strong>fering a number <strong>of</strong> different learning<br />

approaches to <strong>the</strong> students. In particular, <strong>the</strong> ability to <strong>of</strong>fer self-paced material for a technical<br />

subject has shown great promise. Given <strong>the</strong> success <strong>of</strong> <strong>the</strong> project, we are now in <strong>the</strong> process <strong>of</strong><br />

expanding <strong>the</strong> use <strong>of</strong> Open Source s<strong>of</strong>tware and data in our teaching. Student feedback<br />

highlighted <strong>the</strong>ir understanding <strong>of</strong> <strong>the</strong> importance <strong>of</strong> <strong>the</strong> topic, and suggested that more time should<br />

be allocated to <strong>the</strong> course. The material will <strong>the</strong>refore be presented as a separate course this year,<br />

ra<strong>the</strong>r than a topic within ano<strong>the</strong>r module.<br />

7. Acknowledgements<br />

The author would like to thank UCL‘s Learning and Teaching Support Services (LTSS) for <strong>the</strong><br />

funding that made <strong>the</strong> development <strong>of</strong> <strong>the</strong> material described in this paper possible, and Patrick<br />

Weber for developing and testing a large part <strong>of</strong> <strong>the</strong> material itself. Thanks also go to <strong>the</strong> 2009-<br />

2010 CEGE MSc. in <strong>GIS</strong> cohort for <strong>the</strong>ir feedback.<br />

8. References<br />

Biggs, J. (<strong>19</strong>96), Enhancing teaching through constructive alignment, Higher Education, 32: 347-<br />

364<br />

Elwood S. (2008), Geographic information science: new geovisualization technologies: emerging<br />

questions and linkages with <strong>GIS</strong>cience research, Progress in Human Geography,<br />

Page | 139


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Haklay, M., Singleton, A., and Parker, C. (2008) Web mapping 2.0: <strong>the</strong> Neogeography <strong>of</strong> <strong>the</strong><br />

Geoweb, Geography Compass 3: 2011-2039,<br />

Goodchild, M. 2007a: Citizens as sensors: <strong>the</strong> world <strong>of</strong> volunteered geography. GeoJournal 69,<br />

211–21.<br />

Haklay, M. and Weber, P. (2008) OpenStreetMap – User Generated Street Map, IEEE Pervasive<br />

Computing, 12-18.<br />

Richardson, John T. E. (2005), Students‘ Approaches to Learning and Teachers‘ Approaches to<br />

Teaching in Higher Education, Educational Psychology, 25: 6, 673-680<br />

Schuurman, N. (2009) The new Brave NewWorld: geography, <strong>GIS</strong>, and <strong>the</strong> emergence <strong>of</strong><br />

ubiquitous mapping and data, Environment and Planning D: Society and Space 27, 571-580<br />

Sharp, R. (2010), The Effective Use <strong>of</strong> Public Sector Information for Community Benefit,<br />

Unpublished MSc. <strong>GIS</strong> Dissertation, Civil, Environmental and Geomatic Engineering, UCL<br />

Steiniger, S. and Bocher, E. (2009) An overview on current free and open source desktop <strong>GIS</strong><br />

developments, International Journal <strong>of</strong> Geographical Information Science, 23: 10, 1345 — 1370<br />

Page | 140


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Use <strong>of</strong> Geographical Information for Non-Visual Perceptualisation<br />

Paul D Kelly, Stuart Ferguson, Karen Rafferty, Jian-Xun Peng<br />

School <strong>of</strong> Electronics, Electrical Engineering & Computer Science<br />

Queen‘s University Belfast<br />

Ashby Building, Stranmillis Road, BELFAST BT9 5AH<br />

E-mail: Paul.Kelly@ee.qub.ac.uk<br />

ABSTRACT<br />

Non-visual perceptualisation <strong>of</strong> geographical information can be a valuable addition to, or even<br />

replacement for, a traditional cartographic display. This is especially true for navigation applications<br />

on mobile devices with a small screen size, where it can benefit more than just those with sight<br />

impairments. The HaptiMap project aims to encourage <strong>the</strong> incorporation <strong>of</strong> such enhanced<br />

accessibility into <strong>the</strong> user interfaces <strong>of</strong> mobile location-based applications. This paper discusses some<br />

<strong>of</strong> <strong>the</strong> issues faced by HaptiMap regarding <strong>the</strong> usability <strong>of</strong> geographical information for non-visual<br />

perceptualisation, and <strong>the</strong> solutions that have been proposed.<br />

KEYWORDS: non-visual perceptualisation, usability <strong>of</strong> geographical information, human computer<br />

interfaces, application programming interfaces, data structures<br />

1. Introduction<br />

Assistance with pedestrian navigation whilst on <strong>the</strong> move is an obvious application for mobile<br />

communications devices. Efficient communication <strong>of</strong> <strong>the</strong> relevant geographical information through<br />

a small screen has been well researched (e.g. Brown & Laurier (2005), Cheverst et al (2000)) but has<br />

also been shown to disengage <strong>the</strong> user from <strong>the</strong> environment (Leshed et al, 2008) as well as impede<br />

<strong>the</strong> geographical understanding <strong>of</strong> <strong>the</strong> environment (Aslan et al, 2006). In addition a user may have<br />

difficulty interacting via a display screen because <strong>of</strong> disability, e.g. blindness. Non-visual<br />

perceptualisation is thus worthy <strong>of</strong> research as an alternative method <strong>of</strong> communicating geographical<br />

information to <strong>the</strong> user <strong>of</strong> a mobile device.<br />

HaptiMap is a large-scale integrating project financed under <strong>the</strong> European Commission Seventh<br />

Framework Programme between September 2008 and August 2012. The full project title is ―Haptic,<br />

Audio and Visual Interfaces for Maps and Location-based Services‖ and a key aim is to encourage<br />

<strong>the</strong> incorporation <strong>of</strong> enhanced accessibility into <strong>the</strong> user interfaces <strong>of</strong> mobile mapping and locationbased<br />

applications. Developing methods <strong>of</strong> non-visual perceptualisation <strong>of</strong> geographical data and<br />

addressing <strong>the</strong> usability <strong>of</strong> geographical data for this purpose is central to this. The target application<br />

area is pedestrian navigation, with a particular (although certainly non-exclusive) emphasis on blind<br />

and visually-impaired users. This paper will discuss some <strong>of</strong> <strong>the</strong> needs that have been identified and<br />

solutions that have been proposed within HaptiMap for manipulating geographical data in order to<br />

perceptualise it non-visually.<br />

2. Non-Visual Perceptualisation<br />

Some examples <strong>of</strong> such non-visual methods developed by HaptiMap partners are:<br />

� A Haptic Belt containing multiple vibrator motors, which when worn round a user‘s waist<br />

and combined with heading information, can indicate <strong>the</strong> direction <strong>of</strong> a point <strong>of</strong> interest or a<br />

route to follow (Pielot et al, 2010).<br />

Page | 141


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

� The Viflex – a hand-held device containing a small platform that can exert a force in different<br />

directions on <strong>the</strong> user‘s fingers/thumb, or be manipulated by <strong>the</strong> user to indicate a direction<br />

or deflection, with variable levels <strong>of</strong> force feedback and/or vibration (Roselier & Hafez,<br />

2006).<br />

� Sonification, whereby pre-existing audio (e.g. music) or artificial audio may be manipulated<br />

to give <strong>the</strong> impression <strong>of</strong> coming from a particular direction when listened to through<br />

headphones (Strachan et al, 2005).<br />

What <strong>the</strong>se output methods have in common is that <strong>the</strong>ir requirements for <strong>the</strong> format in which<br />

geographical data is retrieved are substantially different from those for conventional rendering <strong>of</strong> a<br />

cartographic map. At this level, maps are typically ―visualised‖ by making logical queries regarding<br />

<strong>the</strong> position <strong>of</strong> various geographical features relative to each o<strong>the</strong>r and to <strong>the</strong> user‘s current location.<br />

The results <strong>of</strong> <strong>the</strong>se queries are <strong>the</strong>n presented using some non-visual methods such as those listed<br />

above. Types <strong>of</strong> queries could include:<br />

� What is <strong>the</strong> first geographical feature <strong>of</strong> a certain type (e.g. boundary fence/wall, building)<br />

that a straight line from my current location in a particular direction would intersect, and<br />

how far away is it?<br />

� I am lost while out hiking in difficult terrain – in which direction do I need to walk in order<br />

to get back onto a marked path via <strong>the</strong> shortest distance?<br />

� How far am I from <strong>the</strong> edge <strong>of</strong> a polygon (Ge<strong>of</strong>ence) defining an area <strong>of</strong> town regarded as<br />

dangerous?<br />

It is clear that in order to be able to make <strong>the</strong>se kinds <strong>of</strong> queries, access is needed to <strong>the</strong> raw vector<br />

data underpinning <strong>the</strong> map. For example, a Web Map Service would not be suitable as it only shows<br />

geometric features already rendered for visual interpretation, whereas with non-visual<br />

perceptualisation some <strong>of</strong> this interpretation must be done by <strong>the</strong> application – <strong>the</strong> data presented to<br />

<strong>the</strong> user has to be filtered to reduce its information content, because <strong>the</strong> ―bandwidth‖ <strong>of</strong> <strong>the</strong> nonvisual<br />

communication channels is not as high as that <strong>of</strong> <strong>the</strong> visual channel.<br />

3. Availability and Usability <strong>of</strong> Vector Data<br />

The main problem with using vector data in <strong>the</strong>se types <strong>of</strong> applications is that <strong>the</strong> amount <strong>of</strong><br />

information contained can be overwhelming for a typical human-computer interface (HCI) developer<br />

who only wishes to extract e.g. centre lines <strong>of</strong> paths in a forest, or details <strong>of</strong> open spaces to use in an<br />

urban navigation application. To get details <strong>of</strong> just one feature would involve implementing a parser<br />

for <strong>the</strong> whole file format or database and <strong>the</strong>re is a very steep learning curve to even get started.<br />

Personal observations within <strong>the</strong> HaptiMap project have shown in addition that HCI developers tend<br />

to be unfamiliar with <strong>the</strong> typical file and electronic transfer formats commonly used for transmission<br />

<strong>of</strong> geographical information and with <strong>the</strong> way such information is typically structured, i.e. in terms <strong>of</strong><br />

points, linestrings and polygons. Experimental systems such as those mentioned above by Pielot et al<br />

(2010) and Strachan et al (2005) tend to make use <strong>of</strong> geographical information only in <strong>the</strong> form <strong>of</strong><br />

discrete points, unless in <strong>the</strong> context <strong>of</strong> augmenting a visual rendering <strong>of</strong> an existing map through<br />

non-visual means (HaptiMap, 2010).<br />

This is clearly an issue in terms <strong>of</strong> encouraging reuse <strong>of</strong> existing geographical information in <strong>the</strong>se<br />

emerging multi-modal applications—and thus very relevant to <strong>the</strong> goals <strong>of</strong> <strong>the</strong> HaptiMap project.<br />

One <strong>of</strong> several methods by which HaptiMap aims to achieve its goals is through <strong>the</strong> development <strong>of</strong><br />

a toolkit for building accessible mapping and location-based applications. In effect this toolkit is a<br />

standard s<strong>of</strong>tware library with an application programming interface (API) written in <strong>the</strong> widely used<br />

C programming language. The toolkit provides, among o<strong>the</strong>r features, a standardised cross-platform<br />

API supporting Windows Mobile, iPhone, Android, Symbian and Linux/MeeGo mobile platforms.<br />

This provides simple cross-platform support for both specialised hardware for haptic/tactile output<br />

Page | 142


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

and for access to device internal sensors, such as accelerometers or a GPS receiver. The feature most<br />

relevant to this discussion however, is a system to make accessing and using geographical<br />

information relatively simple for developers who do not have <strong>GIS</strong> programming experience, but are<br />

none<strong>the</strong>less competent programmers.<br />

It is perhaps important to note that <strong>the</strong> target users <strong>of</strong> <strong>the</strong> toolkit are also s<strong>of</strong>tware developers, i.e. not<br />

―end users‖ in <strong>the</strong> conventional sense <strong>of</strong> <strong>the</strong> term. This means that we are more concerned with<br />

technical details <strong>of</strong> how <strong>the</strong> geographical information is logically structured, and <strong>the</strong> electronic<br />

formats in which it is stored and transmitted, than in how it is presented through a front-end user<br />

interface—<strong>the</strong> hope being that we can empower <strong>the</strong>se developers to make geographical information<br />

easier to interpret for end users, by in turn making it easier for <strong>the</strong>m to acquire, interpret and<br />

manipulate <strong>the</strong> information. Ideally this will enable developers without prior <strong>GIS</strong> programming<br />

knowledge to create useful mapping and location-based applications.<br />

The API functions provided by <strong>the</strong> toolkit for querying geographical data form a subset <strong>of</strong> <strong>the</strong> overall<br />

toolkit API. API usability (wherein <strong>the</strong> developers writing s<strong>of</strong>tware that uses <strong>the</strong> API are considered<br />

to be <strong>the</strong> ―users‖) is being increasingly recognised as important (Daughtry et al, 2009), and<br />

evaluation <strong>of</strong> <strong>the</strong> toolkit from a developer point <strong>of</strong> view is an important part <strong>of</strong> <strong>the</strong> HaptiMap project.<br />

This will address such issues as having a clear feature list, and consistent function naming and<br />

argument order, as well as clarity <strong>of</strong> documentation, amongst o<strong>the</strong>rs. The usability <strong>of</strong> <strong>the</strong><br />

geographical data structure exposed through <strong>the</strong> API is a more subjective issue, and formal<br />

evaluation may necessitate assessing developers‘ experiences using this and o<strong>the</strong>r sources in more<br />

depth; <strong>the</strong> exact mechanism by which this will be done has not yet been decided.<br />

4. Simplified Data Model<br />

A HaptiMap application may require geographical information to be retrieved from multiple sources<br />

with various different logical structures, file/transfer formats, co-ordinate systems, attribute models<br />

etc. A key aim <strong>of</strong> <strong>the</strong> toolkit is to hide this complexity from <strong>the</strong> end user. As most <strong>of</strong> <strong>the</strong> technical<br />

requirements related to this simplification and abstraction are specific to <strong>the</strong> individual sources <strong>of</strong><br />

data, it makes sense to perform it as close to <strong>the</strong> data source as possible. To this end an architecture<br />

has been defined that uses multiple ―geographical data plugins‖, one for each data source. The toolkit<br />

may use data from more than one source concurrently. It is <strong>the</strong> role <strong>of</strong> <strong>the</strong>se plugins to transform<br />

data into <strong>the</strong> simplified data model used by <strong>the</strong> HaptiMap toolkit, which is described below.<br />

The data model has been defined as a result <strong>of</strong> discussions between HaptiMap partners, and<br />

represents a compromise between representing <strong>the</strong> data structure from typical existing sources as<br />

closely as possible whilst providing a simpler ―user-facing‖ structure. It can be seen that <strong>the</strong>re are<br />

certain similarities with <strong>the</strong> data model used by OpenStreetMap (2011), specifically in terms <strong>of</strong> <strong>the</strong><br />

small number <strong>of</strong> highly simplified geometric primitives allowed, however in general <strong>the</strong> HaptiMap<br />

model is more specifically tuned to <strong>the</strong> needs <strong>of</strong> accessible navigation on mobile devices (e.g. integer<br />

co-ordinates for efficient calculations, structured attribute system for ease <strong>of</strong> representation <strong>of</strong><br />

standardised accessibility-related environmental features).<br />

Co-ordinates. The key concept here is that all co-ordinates accessible to <strong>the</strong> user through <strong>the</strong> toolkit<br />

API, whe<strong>the</strong>r obtained from a location sensor device (e.g. a GPS receiver) or a source <strong>of</strong> map data<br />

(e.g. a NAVTEQ MapTP server or a local authority-provided WFS server), are integer values in a<br />

regular grid with cartesian co-ordinates. Where available, height values should form a coherent part<br />

<strong>of</strong> a 3-D co-ordinate system when combined with 2-D position co-ordinates.<br />

The toolkit does not impose any requirement on how this 3-D cartesian system is implemented<br />

―behind <strong>the</strong> scenes‖; that is <strong>the</strong> job <strong>of</strong> <strong>the</strong> geographical data plugins. Typically however, a projected<br />

co-ordinate system, perhaps with values in centimetres, is used for <strong>the</strong> 2-D system and geoidal<br />

heights used for <strong>the</strong> third dimension. Obviously this is not a true 3-D system, however <strong>the</strong> xy-plane<br />

<strong>of</strong> <strong>the</strong> 2-D projected co-ordinate system coincides approximately with <strong>the</strong> geoid surface.<br />

Page | 143


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

Geometries. The toolkit supports geometric objects based on a greatly simplified subset <strong>of</strong> <strong>the</strong><br />

international standard Simple Features model (ISO, 2004), comprising only <strong>the</strong> elements Point,<br />

LineString and Polygon, as illustrated in Figure 1.<br />

Figure 1. Geometric Primitives in <strong>the</strong> HaptiMap Toolkit<br />

Note that <strong>the</strong> toolkit polygon is a simplified version <strong>of</strong> that defined in <strong>the</strong> ISO model. It consists<br />

solely <strong>of</strong> one outer closed LineString (also known as a LinearRing) defining <strong>the</strong> boundary, and no<br />

interior LinearRings.<br />

Feature Types. Ano<strong>the</strong>r output from HaptiMap in addition to <strong>the</strong> toolkit will be a set <strong>of</strong> content<br />

guidelines for accessible maps and location-based services. These will define a common model into<br />

which features relevant for accessible navigation, such as path, road and building locations, may be<br />

categorised irrespective <strong>of</strong> <strong>the</strong> original source <strong>of</strong> such data. The ―intelligence‖ for transforming <strong>the</strong><br />

features into <strong>the</strong> HaptiMap model will be contained within <strong>the</strong> geographical data plugins. This means<br />

that <strong>the</strong> user can access geographical data using a common feature model, no matter what <strong>the</strong> source.<br />

Attributes. Each feature may have an unlimited number <strong>of</strong> attributes associated with it; an attribute<br />

consists <strong>of</strong> an integer attribute code indicating what property <strong>the</strong> attribute describes, toge<strong>the</strong>r with an<br />

attribute value: ei<strong>the</strong>r a single integer, an array <strong>of</strong> multiple integers, or a string.<br />

The data model is also well suited to visual rendering; despite its simplicity it is possible to represent<br />

all <strong>the</strong> information that appears in conventional online maps (e.g. Google maps). In addition, because<br />

<strong>the</strong> data is stored as vector data ra<strong>the</strong>r than pre-rendered tiles <strong>the</strong> fast graphical processors in modern<br />

smart-phones can render <strong>the</strong> visual appearance <strong>of</strong> <strong>the</strong> map to accentuate features or only display<br />

features <strong>of</strong> interest. Over slower wireless networks <strong>the</strong> transfer <strong>of</strong> vector data is also considerably<br />

faster and local caches can hold more information making <strong>the</strong> user experience much more<br />

responsive.<br />

5. Summary<br />

The HaptiMap project aims to encourage <strong>the</strong> incorporation <strong>of</strong> enhanced accessibility into <strong>the</strong> user<br />

interfaces <strong>of</strong> mobile mapping and location-based applications. The phrase enhanced accessibility<br />

covers various features that make such applications easier (or indeed possible) to use for people with<br />

various disabilities, e.g. blindness. It is anticipated that <strong>the</strong> enhanced accessibility will also be useful<br />

to people without such disabilities, e.g. by removing <strong>the</strong> need to look at a display screen in order to<br />

perform certain tasks. Non-visual perceptualisation <strong>of</strong> geographical information is an essential part <strong>of</strong><br />

enhancing accessibility in such a manner.<br />

Observations within <strong>the</strong> HaptiMap project have revealed that accessibility <strong>of</strong> conventional sources <strong>of</strong><br />

geographical information <strong>the</strong>mselves (by HCI developers) is also an issue, due to lack <strong>of</strong> familiarity<br />

and <strong>the</strong> learning curve faced by non-specialist <strong>GIS</strong> developers when working with detailed<br />

geographical information in vector format. A solution has been proposed to this problem which<br />

involves including a simplified API for accessing geographical information from multiple sources as<br />

part <strong>of</strong> <strong>the</strong> ―HaptiMap Toolkit‖. This has now been implemented and is currently undergoing testing,<br />

Page | 144


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

with formal evaluation from <strong>the</strong> point <strong>of</strong> view <strong>of</strong> s<strong>of</strong>tware developers also to take place as part <strong>of</strong> <strong>the</strong><br />

project.<br />

6. Acknowledgements<br />

Integrating Project HaptiMap is funded by <strong>the</strong> European Commission (FP7-ICT-224675).<br />

7. References<br />

Aslan, I, Schwalm, M, Baus, J, Krüger, A and Schwartz, T (2006). Acquisition <strong>of</strong> spatial knowledge<br />

in location-aware mobile pedestrian navigation systems. Proc. 8th Int. Conf. on Human-Computer<br />

Interaction with Mobile Devices and Services Espoo, Finland, September 2006, pp105–108.<br />

Brown, B and Laurier, E (2005). Designing electronic maps: an ethnographic approach. In Meng et<br />

al (Eds): Map-based Mobile Services: Theories, Methods and Implementations Springer, Berlin,<br />

2005.<br />

Cheverst, K, Davies, N, Mitchell, K, Friday, A and Efstratiou, C (2000). Developing a context-aware<br />

electronic tourist guide: some issues and experiences. Proc. <strong>Conference</strong> on Human Factors in<br />

Computing Systems The Hague, Ne<strong>the</strong>rlands, April 2000, pp17–24.<br />

Daughtry, J M, Farooq, U, Myers, B A and Stylos, J (2009). API Usability: Report on Special<br />

Interest Group at CHI. SIGSOFT S<strong>of</strong>tware Engineering Notes Vol. 34 No. 4, July 2009.<br />

HaptiMap (2010). TouchShape – Making Virtual Tactile Control Surfaces. In HaptiMap Deliverable<br />

2.1: First prototypes and initial report on <strong>the</strong> perceptualization <strong>of</strong> map data, context sensing and<br />

reasoning, and hardware development February 2010, pp 151–156. Restricted dissemination<br />

technical report.<br />

ISO – International Organization for Standardisation (2004). ISO <strong>19</strong>125:2004. Geographic<br />

information – Simple feature access – Part 1: Common architecture.<br />

Leshed, G, Velden, T, Rieger, O, Kot, B and Sengers, P (2008). In-Car GPS Navigation:<br />

Engagement with and Disengagement from <strong>the</strong> Environment. Proc. 26th <strong>Annual</strong> CHI <strong>Conference</strong> on<br />

Human Factors in Computing Systems Florence, Italy, April 2008, pp1675–1684.<br />

OpenStreetMap (2011). Data Primitives – OpenStreetMap Wiki.<br />

http://wiki.openstreetmap.org/wiki/Data_Primitives<br />

Pielot, M, Krull, O and Boll, S (2010). Where is my Team? Supporting Situation Awareness with<br />

Tactile Displays. Proc. 28th <strong>Annual</strong> CHI <strong>Conference</strong> on Human Factors in Computing Systems<br />

Atlanta, USA, April 2010, pp1705–1714.<br />

Roselier, S and Hafez, M (2006). ViFlex: A Compact Haptic 2D Interface with Force Feedback for<br />

Mobile Devices. Eurohaptics Paris, France, July 2006.<br />

Strachan, S, Eslambolchilar, P and Murray-Smith, R (2005). gpsTunes – controlling navigation via<br />

audio feedback. Proc. 7th International <strong>Conference</strong> on Human-Computer Interaction with Mobile<br />

Devices and Services Salzburg, Austria, Sept. 2005, pp275–278.<br />

8. Biographies<br />

Dr Paul D Kelly received <strong>the</strong> degree <strong>of</strong> PhD in Electrical & Electronic Engineering from Queen's<br />

University Belfast in 2005. Since <strong>the</strong>n he has developed research interests in geographical data<br />

processing and digital audio and video processing, with a particular emphasis on C programming.<br />

He is currently employed by Queen's University Belfast as a research fellow.<br />

Dr Stuart Ferguson has over 25 years experience in computer graphics and s<strong>of</strong>tware engineering.<br />

He is <strong>the</strong> author <strong>of</strong> <strong>the</strong> book Practical Algorithms for 3D computer Graphics (2001). He is currently<br />

a lecturer at Queen's University Belfast where he is researching real-time application program<br />

Page | 145


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3b: Open Source and Web 2.0<br />

implementations for mobile devices.<br />

Dr Karen Rafferty has over 10 years experience working within <strong>the</strong> fields <strong>of</strong> image processing,<br />

computer vision, virtual reality applications and intelligent devices that can perceive and respond to<br />

<strong>the</strong>ir environment. She has authored over 30 journal and conference papers in <strong>the</strong>se areas. Her first<br />

book on Virtual Reality was published in 2007.<br />

Dr Jian-Xun Peng has over 15 years experience in avionics, nonlinear system modelling and<br />

control, artificial neural networks and computer graphics. He has authored over 40 journal and<br />

conference papers. He is currently with Queen‘s University Belfast where he is researching<br />

application context sensing and developing s<strong>of</strong>tware for <strong>the</strong> European Commission-funded<br />

HaptiMap project.<br />

Page | 146


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

Fuzzy Geographical Buffers Revisited<br />

Peter Fisher 1 , Firdos Almadani 2<br />

1 University <strong>of</strong> Leicester, Department <strong>of</strong> Geography, Leicester, LE2 3TD, <strong>UK</strong><br />

Tel. +44 116 252 3853 Fax +44 116 252 3854<br />

pff1@le.ac.uk http://www.le.ac.uk/gg/staff/academic_fisher.html<br />

2 University <strong>of</strong> Leicester, Department <strong>of</strong> Geography, Leicester, LE2 3TD, <strong>UK</strong><br />

fma7@le.ac.uk<br />

ABSTRACT<br />

Within analyses done with <strong>GIS</strong> <strong>the</strong> buffer is defined as <strong>the</strong> area within a certain distance <strong>of</strong> a<br />

geographical object, and is usually modelled as a Boolean set. Here we examine <strong>the</strong> consequences <strong>of</strong><br />

using a fuzzy set membership function to define a buffer. When <strong>the</strong> geographical object is Boolean and<br />

<strong>the</strong> buffer is fuzzy, membership is dependent only on distance, but, when <strong>the</strong> geographical object itself<br />

is fuzzy, it becomes necessary to model <strong>the</strong> fuzzy buffer for every α-cut <strong>of</strong> <strong>the</strong> object. Here <strong>the</strong> problem<br />

is illustrated by graphs, but in <strong>the</strong> presentation we will show <strong>the</strong> implementation in a standard raster <strong>GIS</strong><br />

as well.<br />

1. Introduction<br />

KEYWORDS: buffer, fuzzy sets, fuzzy numbers, fuzzy geographical objects<br />

A conceptually simple and widely used <strong>GIS</strong> operation is <strong>the</strong> buffer. In determining a buffer a target<br />

feature is chosen, and those areas which are within a threshold distance <strong>of</strong> <strong>the</strong> target are identified as<br />

being within <strong>the</strong> buffer. The target can be a point, line or polygon, and can be multiple instances <strong>of</strong> any<br />

<strong>of</strong> <strong>the</strong>se. The normal buffer operation results in a Boolean map which shows those areas which meet <strong>the</strong><br />

distance criterion from <strong>the</strong> target feature, and those that do not. There is no equivocation; no scope for<br />

uncertainty (Heywood et al., 2002, p115; Longley et al., 2011, p366). The change from within <strong>the</strong><br />

buffer to outside <strong>the</strong> buffer happens dramatically when <strong>the</strong> threshold distance is reached (Figure 1A).<br />

Until <strong>the</strong> threshold distance <strong>the</strong> membership is 1, and <strong>the</strong>reafter it is 0.<br />

A<br />

C<br />

B<br />

Figure 1 Membership functions for buffers<br />

with distance from a target object A) <strong>the</strong><br />

Boolean case, B) a simple fuzzy case and C)<br />

a fuzzy case with two gradients.<br />

Page | 147


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

Clearly <strong>the</strong> buffer is susceptible to a sorites type argument (Fisher 2000). If a location is just outside <strong>the</strong><br />

buffer, is it truly to be ignored or could it be identified as partly within <strong>the</strong> buffer; more so than locations<br />

that are fur<strong>the</strong>r away. This means that <strong>the</strong> buffer can be addressed conveniently by fuzzy sets. Guesgen<br />

and co workers (Guesgen and Hertzberg 2001; Duff and Guesgen, 2002; Guesgen et al. 2003) have<br />

examined an approach to <strong>the</strong> fuzzy buffer, and Mesgari et al (2008) and Fisher (2009) have shown <strong>the</strong><br />

use <strong>of</strong> this simplest instance <strong>of</strong> a fuzzy buffer (similar to Figure 1B). All authors ei<strong>the</strong>r implicitly or<br />

explicitly point out that any buffer has a monotonically decreasing degree <strong>of</strong> membership as a function<br />

<strong>of</strong> distance from <strong>the</strong> target.<br />

The next section sets out <strong>the</strong> situation where a fuzzy buffer is defined around a Boolean object. Next<br />

<strong>the</strong> situation <strong>of</strong> a Boolean buffer drawn around a fuzzy object is outlined and <strong>the</strong> subsequent section<br />

looks at fuzzy buffers around fuzzy objects. Examples <strong>of</strong> fuzzy target objects might include, but are not<br />

limited to, human settlements (villages or towns), farms (when <strong>the</strong>ir extent is poorly known as was <strong>the</strong><br />

case at <strong>the</strong> beginning <strong>of</strong> <strong>the</strong> last outbreak <strong>of</strong> foot and mouth in <strong>the</strong> <strong>UK</strong>), or woodland (is a fuzzy<br />

classification <strong>of</strong> land covers in a satellite image).<br />

2. A Fuzzy Buffer around a Boolean Object<br />

A fuzzy buffer can be shown as having a gradual decrease in <strong>the</strong> membership with distance away from<br />

<strong>the</strong> target ei<strong>the</strong>r as a single gradient (Figure 1B) with a value <strong>of</strong> 1 showing locations that are definitely<br />

within <strong>the</strong> buffer and gradually decreasing memberships until 0 membership is reached where <strong>the</strong><br />

location is definitely outside <strong>the</strong> buffer. It is also possible to use one or more inflection points (Figure<br />

1C), or even to use a continuous function for <strong>the</strong> grade <strong>of</strong> membership. Indeed <strong>the</strong>re is no reason that<br />

<strong>the</strong> membership function should be isotropic (<strong>the</strong> same in all directions), or that it should not fall and<br />

rise again with distance from <strong>the</strong> target (not be monotonically decreasing for whatever local reason).<br />

All cases, however, require <strong>the</strong> target object itself to be Boolean. If <strong>the</strong> target object is a fuzzy object<br />

<strong>the</strong>n <strong>the</strong> situation becomes more complex, and <strong>the</strong> purpose <strong>of</strong> this paper is to present <strong>the</strong>se more<br />

complex situations.<br />

3. A Boolean Buffer around a Fuzzy Object<br />

A fuzzy object can be viewed as a set <strong>of</strong> locations where each point has a membership in <strong>the</strong> range [0,<br />

1]; a special case <strong>of</strong> a fuzzy set only in <strong>the</strong> sense that <strong>the</strong> fuzzy membership values are a function <strong>of</strong><br />

spatial coordinates x and y. Any fuzzy object, just like a Boolean one, <strong>the</strong>refore can have a Boolean<br />

buffer drawn around it, but in this case, <strong>the</strong> Boolean buffer is not simply a zone with membership 1 for<br />

those areas included. Ra<strong>the</strong>r for each α-cut <strong>of</strong> <strong>the</strong> object, a separate Boolean buffer is present. Thus<br />

using <strong>the</strong> Boolean buffer membership function in Figure 1A, a set <strong>of</strong> buffers exist for <strong>the</strong> object is<br />

Figure 2, which illustrates <strong>the</strong> α-cuts at α = 1, 0.8, 0.6. 0.4 and 0.2.<br />

Page | 148


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

A<br />

C<br />

E F<br />

Figure 2 A)A fuzzy object is described by <strong>the</strong> dashed line, and a set <strong>of</strong> α-cut<br />

versions <strong>of</strong> <strong>the</strong> Boolean buffer around that object are presented as B) α=1, C)<br />

α=0.8, D) α=0.6. E) α=0.4 and F) α=0.2<br />

This means that at any point <strong>the</strong> degree to which that point is within <strong>the</strong> buffer from <strong>the</strong> target object<br />

varies with <strong>the</strong> α-cut value. Because <strong>the</strong> buffer is a measure <strong>of</strong> proximity to <strong>the</strong> target, a membership 1<br />

in <strong>the</strong> buffer <strong>of</strong> a particular α-cut shows that that location is within <strong>the</strong> distance buffer <strong>of</strong> a different<br />

form <strong>of</strong> <strong>the</strong> category or object-class <strong>of</strong> <strong>the</strong> target. When α = 1 this might be considered a perfect<br />

example <strong>of</strong> <strong>the</strong> mapped phenomenon, or within <strong>the</strong> buffer <strong>of</strong> a good example if α = 0.8, or a mediocre<br />

example if α = 0.4, etc. Thus at 30m from <strong>the</strong> 1.0 α-cut <strong>of</strong> <strong>the</strong> target object <strong>the</strong>re is a membership <strong>of</strong> 1<br />

in <strong>the</strong> illustrated case, but at <strong>the</strong> same location, <strong>the</strong>re is 0 membership <strong>of</strong> belonging to <strong>the</strong> buffer <strong>of</strong> any<br />

o<strong>the</strong>r α-cuts illustrated because <strong>the</strong>y are all within <strong>the</strong> object at those α-cuts. On <strong>the</strong> o<strong>the</strong>r hand, at 601<br />

metres from <strong>the</strong> target, <strong>the</strong>re is 0 membership <strong>of</strong> being within <strong>the</strong> buffer <strong>of</strong> <strong>the</strong> 1.0 α-cut (<strong>the</strong> perfect<br />

case), BUT <strong>the</strong>re is a membership <strong>of</strong> 1 <strong>of</strong> being within <strong>the</strong> buffer <strong>of</strong> all o<strong>the</strong>r α-cuts illustrated.<br />

This could be interpreted as amounting to <strong>the</strong> same as a fuzzy buffer like that illustrated in Figure 1B,<br />

but actually it contains more information. In Figure 2 <strong>the</strong> buffer has been kept <strong>the</strong> same for every α-cut,<br />

but that is not necessary. Arguments could be put forward for <strong>the</strong> width <strong>of</strong> <strong>the</strong> buffer being a function <strong>of</strong><br />

<strong>the</strong> value <strong>of</strong> <strong>the</strong> α-cut which could lead to a wider buffer for larger or smaller values <strong>of</strong> α.<br />

B<br />

D<br />

Page | 149


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

4. A Fuzzy Buffer around a Fuzzy Object<br />

Figure 3 Fuzzy buffers based on that<br />

illustrated in Figure 1C around 5 αcuts<br />

<strong>of</strong> <strong>the</strong> fuzzy object illustrated in<br />

Figure 2.<br />

The same logic as that applied to <strong>the</strong> Boolean buffer around a fuzzy object can be applied to a fuzzy<br />

buffer around a fuzzy object. So following <strong>the</strong> model presented above, for every α-cut <strong>of</strong> <strong>the</strong> target<br />

fuzzy object considered, a fuzzy buffer is constructed (Figure 3). Here <strong>the</strong> width parameters <strong>of</strong> <strong>the</strong><br />

fuzzy buffer are held constant for all α, but, as in <strong>the</strong> case <strong>of</strong> a Boolean buffer around a fuzzy object, <strong>the</strong><br />

buffer parameters could increase or decrease with α.<br />

At any point away from <strong>the</strong> location <strong>of</strong> <strong>the</strong> target location, <strong>the</strong> degree to which it is within <strong>the</strong> buffer<br />

around <strong>the</strong> target object can be reconstructed from this set <strong>of</strong> graphs just as with <strong>the</strong> Boolean buffer, but<br />

it is best illustrated as a different set <strong>of</strong> graphs (Figure 4). Here <strong>the</strong> degree to which any location is<br />

within <strong>the</strong> buffer is illustrated as <strong>the</strong> fuzzy numbers (Klir and Yuan, <strong>19</strong>95) at five locations (500, 700,<br />

900. 1100 and 1300 m away from <strong>the</strong> target). Values on <strong>the</strong> x axes show <strong>the</strong> membership <strong>of</strong> <strong>the</strong> buffer<br />

and <strong>the</strong> y axes show α-cuts <strong>of</strong> <strong>the</strong> target object.<br />

Figure 4 Fuzzy number representations based on <strong>the</strong> fuzzy buffers shown in Figure 3 around<br />

<strong>the</strong> fuzzy object shown in Figure 2 <strong>of</strong> <strong>the</strong> degree to which five locations away from <strong>the</strong> target<br />

location are within <strong>the</strong> buffer<br />

Page | 150


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

Fuzzy buffers around <strong>the</strong> α-cuts can be created as a set <strong>of</strong> raster layers within many <strong>GIS</strong> packages, and<br />

<strong>the</strong> fuzzy numbers can be recovered from <strong>the</strong>se. An example is illustrated in Figure 5, where a fuzzy<br />

rendition <strong>of</strong> <strong>the</strong> extent <strong>of</strong> <strong>the</strong> English Lake District mountain Helvellyn (Fisher et al., 2004) is shown.<br />

The version <strong>of</strong> <strong>the</strong> fuzzy Helvellyn presented here is based on <strong>the</strong> drop height to <strong>the</strong> nearest pass (Wood,<br />

2009). This is split into 9 α-cuts at 0.1 intervals <strong>of</strong> which 5 are illustrated in Figure 5 A-E. Fuzzy<br />

buffers around <strong>the</strong> five α-cuts are also illustrated (Figure 5 F-J , using <strong>the</strong> membership functions shown<br />

in Figure 5 O), and <strong>the</strong>se are combined to yield fuzzy number representations (Figure 5 N) <strong>of</strong> <strong>the</strong> fuzzy<br />

buffer <strong>of</strong> <strong>the</strong> fuzzy object, Helvellyn.<br />

α α-cuts Fuzzy buffers<br />

around <strong>the</strong> αcuts<br />

A<br />

F<br />

0.1<br />

0.3<br />

0.5<br />

0.7<br />

0.9<br />

B<br />

C<br />

D<br />

E<br />

G<br />

H<br />

I<br />

J<br />

L<br />

The fuzzy extent <strong>of</strong><br />

Helvellyn<br />

Sample points for fuzzy<br />

numbers<br />

M 3 2 5<br />

4 1<br />

N 5 4 3 2 1<br />

O<br />

Figure 5 A set <strong>of</strong> five α-cuts (A-E; out <strong>of</strong> nine α-cuts at 0.1 intervals) through an<br />

interpretation <strong>of</strong> <strong>the</strong> fuzzy extent <strong>of</strong> Helvellyn, in <strong>the</strong> English Lake District (see text for<br />

discussion <strong>of</strong> derivation). Also shown are <strong>the</strong> buffers around <strong>the</strong> α-cuts (F-J). These are<br />

presented as fuzzy numbers (N) for <strong>the</strong> illustrated points in (M) and <strong>the</strong> 9 fuzzy<br />

membership functions <strong>of</strong> <strong>the</strong> buffers for each α is shown (O).<br />

5. Fur<strong>the</strong>r work<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

1<br />

0<br />

0 0.2 0.4 0.6 0.8 1<br />

0<br />

0 1000 2000 3000 4000 5000 6000<br />

Most obviously from <strong>the</strong> fuzzy buffer it is possible to attach qualitative descriptions <strong>of</strong> <strong>the</strong> proximity <strong>of</strong><br />

locations to <strong>the</strong> fuzzy objects. Thus in terms <strong>of</strong> <strong>the</strong> set <strong>of</strong> fuzzy buffers used here (Figure 5 O) and<br />

Page | 151


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

asserting that <strong>the</strong> fuzzy buffer represents <strong>the</strong> nearness <strong>of</strong> one location to ano<strong>the</strong>r, we can see that point 1<br />

(Figure 5 N and M) has large memberships in all buffers examined and so could be considered very near<br />

to Helvellyn, while location 2 is less close. More to <strong>the</strong> point, Location 1 is near to <strong>the</strong> best examples <strong>of</strong><br />

Helvellyn as a prominent feature, but Point 2 is near to a less prominent Helvellyn, and Point 3 is only<br />

slightly near to still less prominent Helvellyn. These observations are ra<strong>the</strong>r obvious and are visible in<br />

any distance plot (any <strong>of</strong> <strong>the</strong> distance analyses required to derive <strong>the</strong> fuzzy buffers). Using <strong>the</strong> fuzzy<br />

buffers outlined here, however, it is possible to automate <strong>the</strong>se sorts <strong>of</strong> statements with <strong>the</strong> qualifiers.<br />

The challenge remaining for <strong>the</strong> research presented here is to show whe<strong>the</strong>r this fuzzy representation <strong>of</strong><br />

<strong>the</strong> buffer can yield novel analyses <strong>of</strong> geographical information. Buffer operations are typically used in<br />

site selection procedures, and Fisher (2009) shows <strong>the</strong> advantage <strong>of</strong> using fuzzy buffers around Boolean<br />

objects in site selection. Among possible avenues for fur<strong>the</strong>r work here is to establish <strong>the</strong> possibilities<br />

<strong>of</strong> <strong>the</strong> fuzzy representation presented here for site selection modelling.<br />

6. References<br />

Duff, D., and Guesgen, H.W. (2002). An evaluation <strong>of</strong> buffering algorithms in fuzzy <strong>GIS</strong>s.<br />

<strong>Proceedings</strong> <strong>of</strong> <strong>GIS</strong>cience 2002, LNCS 2478, Springer, Berlin. pp80-92.<br />

Fisher, P.F. (2000). Sorites Paradox and Vague Geographies. Fuzzy Sets and Systems 113 (1), 7-18.<br />

Fisher, P.F.(2009). The Representation <strong>of</strong> Uncertain Geographical Information. In The Manual <strong>of</strong><br />

Geographical Information Systems Edited by Marguerite Madden. American Society <strong>of</strong><br />

Photogrammetry and Remote Sensing, Be<strong>the</strong>sda, MD., pp235-264<br />

Fisher, P.F., Wood, J. and Cheng, T. (2004). Where is Helvellyn? Multiscale morphometry and <strong>the</strong><br />

mountains <strong>of</strong> <strong>the</strong> English Lake District. Transactions <strong>of</strong> <strong>the</strong> Institute <strong>of</strong> British Geographers 29, 106-<br />

128<br />

Guesgen, H.W., and Hertzberg, J. (2001). Algorithms for buffering fuzzy raster maps. <strong>Proceedings</strong> <strong>of</strong><br />

Florida AI <strong>Research</strong> Society <strong>Conference</strong> (FLAIRS) 01, AAAI Press, Menlo Park, CA. pp542-546.<br />

Guesgen, H.W., Hertzberg, J., Lobb, R., and Mantler, A. (2003). Buffering Fuzzy Maps in <strong>GIS</strong>. Spatial<br />

Cognition and Computation 3 (2-3),207-222<br />

Heywood, I., Cornelius, S. and Carver, S. (2002). An Introduction to Geographical Information<br />

Systems, 2 nd edition. Prentice Hall, Harlow.<br />

Klir, G.J., and Yuan, B. (<strong>19</strong>95). Fuzzy Sets and Fuzzy Logic: Theory and Applications, Prentice-Hall,<br />

Englewood Cliff, NJ.<br />

Longley, P.A., Goodchild, M.F., Maguire, D.J., and Rhind, D.W. (2011). Geographic Information<br />

Systems and Science, 3 rd edition. John Wiley and sons, Hoboken, NJ<br />

Mesgari, M.S., Pirmoradi, A. and Fallahi, G.R., (2008). Implementation <strong>of</strong> Overlay Function Based on<br />

Fuzzy Logic in Spatial Decision Support System. World Applied Sciences Journal, 3 (Supple 1): 60-65.<br />

Vullings, L.A.E.; Wessels, C.G.A.M.; Bulens, J.D., (2009) Fuzziness to Reduce Uncertainty. In 12th<br />

AGILE International <strong>Conference</strong> on Geographic Information Science Leibniz Universität Hannover,<br />

Germany.<br />

Wood,J. (2009) Landserf, version 2.3, www.landserf.org<br />

8. Biography<br />

Peter Fisher is Pr<strong>of</strong>essor <strong>of</strong> Geographical Information at <strong>the</strong> University <strong>of</strong> Leicester. He has<br />

researched many aspects <strong>of</strong> uncertainty in geographical information.<br />

Firdos Almadani is a post graduate student at <strong>the</strong> University <strong>of</strong> Leicester, and a Teaching Assistant in<br />

<strong>the</strong> Geography Department at King Abdul-Aziz University, Jeddah, Saudi Arabia. Work reported here<br />

relates to her PhD research.<br />

Page | 152


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

Using Morphometric Terrain Properties to Model DEM Error<br />

Stephen M Wise<br />

Department <strong>of</strong> Geography, University <strong>of</strong> Sheffield, Sheffield, S10 2TN, <strong>UK</strong><br />

Tel. +44 114 222 7940 Fax +44 114 222 7940<br />

s.wise@sheffield.ac.uk, http://www.sheffield.ac.uk/geography<br />

ABSTRACT<br />

When DEM error is modelled in order to study error propagation a single model <strong>of</strong> a Gaussian<br />

distribution with strong spatial autocorrelation is usually assumed. A method based on generating a<br />

DEM from a subsample <strong>of</strong> <strong>the</strong> original data points provides a large sample <strong>of</strong> error values which allows<br />

this assumption to be tested. It is found that <strong>the</strong>re are consistent differences in <strong>the</strong> nature <strong>of</strong> error in<br />

convex, concave and planar parts <strong>of</strong> <strong>the</strong> landscape. In addition a measure <strong>of</strong> Difference from Mean<br />

Elevation shows a strong correlation with error. These results suggest that it might be possible to<br />

produce more realistic, spatially distributed models <strong>of</strong> DEM error based on <strong>the</strong> characteristics <strong>of</strong> <strong>the</strong><br />

terrain itself.<br />

1. Introduction<br />

KEYWORDS: Geomorphometry, DEM, Error model<br />

One approach to <strong>the</strong> analysis <strong>of</strong> error propagation in Digital Elevation Models is to create multiple<br />

realisations <strong>of</strong> an error surface for <strong>the</strong> DEM and add <strong>the</strong>se to <strong>the</strong> DEM before doing <strong>the</strong> analysis. From<br />

<strong>the</strong> suite <strong>of</strong> results it is possible to say something about <strong>the</strong> likely magnitude and location <strong>of</strong> <strong>the</strong><br />

propagated error (Hunter and Goodchild <strong>19</strong>97, Fisher <strong>19</strong>98). In order to do this it is necessary to know<br />

something <strong>of</strong> <strong>the</strong> frequency distribution and spatial autocorrelation properties <strong>of</strong> DEM error. In <strong>the</strong><br />

absence <strong>of</strong> any empirical knowledge <strong>of</strong> this it is usually assumed that <strong>the</strong> error has a Gaussian<br />

distribution and a high level <strong>of</strong> spatial autocorrelation.<br />

In order to test whe<strong>the</strong>r <strong>the</strong>se assumptions are actually justified it is necessary to have a large sample <strong>of</strong><br />

error estimates for a DEM from which <strong>the</strong> actual properties <strong>of</strong> <strong>the</strong> distribution can be derived. With <strong>the</strong><br />

advent <strong>of</strong> LiDar this has started to become possible (Davis et al 2001, Kyriakidis et al <strong>19</strong>99) and <strong>the</strong><br />

results suggest that <strong>the</strong> assumption <strong>of</strong> a Gaussian distribution may be incorrect. Wise (in press)<br />

introduced ano<strong>the</strong>r method <strong>of</strong> generating a large sample <strong>of</strong> error estimates by extending <strong>the</strong> work <strong>of</strong><br />

Rees(2000) and Kidner (2003). An existing DEM is resampled by removing some proportion <strong>of</strong> <strong>the</strong> data<br />

points. The remaining points are <strong>the</strong>n used to interpolate a new DEM at <strong>the</strong> same density as <strong>the</strong> original<br />

and wherever a data point was removed <strong>the</strong>re is now <strong>the</strong> original elevation value and a new one<br />

estimated by interpolation. The difference between <strong>the</strong> two produces a value for <strong>the</strong> interpolation error<br />

at that point. Wise (in press) argued that this process is analogous to <strong>the</strong> way in which some existing<br />

DEMs have been produced, such as those which have been created by interpolation from mapped<br />

contours. The sample <strong>of</strong> error measurements by this method is very large. For instance when every o<strong>the</strong>r<br />

data point in both X and Y is removed, this means that error is estimated for just under 75% <strong>of</strong> <strong>the</strong><br />

points in <strong>the</strong> DEM.<br />

In Wise (in press) <strong>the</strong> method was applied to a section <strong>of</strong> <strong>the</strong> Ordnance Survey 50m PANORAMA DEM<br />

in <strong>the</strong> Cairngorms in Scotland (Figure 1) using a range <strong>of</strong> interpolation methods and resampling to a<br />

range <strong>of</strong> different data densities. In many cases <strong>the</strong> frequency distribution <strong>of</strong> DEM error was far more<br />

leptokurtic than a Gaussian and <strong>the</strong> spatial autocorrelation was quite weak. As <strong>the</strong> data density reduced<br />

(i.e. fewer points were retained for use in interpolation) <strong>the</strong> distribution became progressively more<br />

Gaussian, although always statistically different from true Gaussian, and <strong>the</strong> spatial autocorrelation<br />

approached 1. This suggests that <strong>the</strong> model <strong>of</strong> a strongly autocorrelated Gaussian error field may only<br />

apply in certain circumstances.<br />

Page | 153


<strong>GIS</strong>R<strong>UK</strong> 2011 Session 3a: Theory, Methods and Techniques<br />

In addition <strong>the</strong>re are good reasons for thinking that a single model <strong>of</strong> error applied across a whole DEM<br />

<