Thesis Reference

Thesis

Contributions of GIS to Efficient Mine Action

LACROIX, Pierre Marcel Anselme

Abstract

La Campagne Internationale pour Interdire les Mines (International Campaign to Ban Landmines : ICBL) rapporte qu'entre 1999 et 2010 les mines terrestres ont fait plus de 80'000 victimes dans 117 pays et régions du monde. En 2010, on dénombrait 4'191 victimes, parmi lesquelles environ 75% de civils. Pour cette seule année 2010, le chiffre de 1'155 morts a été avancé, mais il est probablement en deçà de la réalité. Les systèmes d'information géographique (SIG) restent encore très peu utilisés par les acteurs du déminage humanitaire, alors qu'ils le sont de manière extensive par de nombreuses autres communautés d'utilisateurs. Cette thèse de doctorat s'intitule ‘How GIS contributes to Efficient Mine Action'. Elle examine dans quelle mesure la cartographie et les SIG peuvent contribuer à augmenter l'efficacité des activités de déminage humanitaire.

Reference

LACROIX, Pierre Marcel Anselme. Contributions of GIS to Efficient Mine Action. Thèse de doctorat : Univ. Genève, 2013, no. Sc. 4571

URN : urn:nbn:ch:unige-289967 DOI : 10.13097/archive-ouverte/unige:28996

Available at: http://archive-ouverte.unige.ch/unige:28996

Disclaimer: layout of this document may differ from the published version.

1 / 1 Université de Genève, Faculté des Sciences Prof. Anthony Lehmann Institut F.-A. Forel/enviroSPACE Centre International de Déminage Humanitaire de Genève Dr. Daniel Eriksson

Contributions of GIS to Efficient Mine Action

Thèse

Présentée à la Faculté des Sciences de l‘Université de Genève

En vue d‘obtenir le grade de Docteur ès Sciences, mention Sciences de

l‘Environnement

par

Pierre Lacroix – Université de Genève

Collonges-sous-Salève (France)

Thèse numéro

4571

Genève

Atelier d‘impression ReproMail - 2013

To my daughter

“Ego cogito, ergo sum” (Descartes)

Table of Contents

Abstract ...... 12

Acknowledgments ...... 16

Supervisory Team & External Advisors ...... 18

Chapter 1. Introduction ...... 19

1.1. Structure of the thesis ...... 20

1.2. Contributing research papers ...... 22

1.3. Mine action background ...... 23

1.3.1. International treaties ...... 24

1.3.2. Mine action ...... 25

1.3.3. Humanitarian demining ...... 25

1.3.4. Information Management ...... 25

1.3.5. Information Management System for Mine Action (IMSMA) ...... 26

1.3.6. The Geneva International Centre for Humanitarian Demining (GICHD) ...... 27

1.3.7. Glossary ...... 27

1.3.8. Photo gallery ...... 32

1.4. User needs ...... 37

1.4.1. Who are the end-users of this research? ...... 37

1.4.2. GIS needs...... 38

1.5. State of the art: GIS in mine action ...... 42

1.6. Research problem, Research questions and Hypotheses...... 44

1.7. Methodology ...... 46

1.7.1. Development cycles ...... 46

1.7.2. Data ...... 47

1.8. Overview of the different research projects ...... 49

1.8.1. SERWIS: Sharing data, maps, technologies and processes ...... 49

1.8.2. Visualising contamination ...... 50

1.8.3. Determining the impacts on human population...... 50

Table of Contents

1.8.4. Choosing the right technique...... 51

1.8.5. Calculating the Shortest Path ...... 51

1.8.6. Setting priorities ...... 51

1.8.7. Optimising GIS workflows ...... 52

1.8.8. E-learning...... 52

1.8.9. Standardising symbology ...... 52

1.8.10. Improving the Quality of data Web Services ...... 53

1.8.11. Summary ...... 53

Chapter 2. To what extent can GIS improve visualisation of contamination and its impact on population? ...... 55

Contributing research papers ...... 55

2.1. Introduction ...... 56

2.2. Visualising Contamination ...... 58

2.2.1. Abstract ...... 58

2.2.2. Introduction ...... 59

2.2.3. Background ...... 60

2.2.4. Experimental Setup...... 64

2.2.5. Visualising Hazards and Mine Hazards: State of the Art ...... 65

2.2.6. Evaluated Visualisation Methods ...... 67

2.2.7. Customising KDE-based methods (D and E): adjusting KDE bandwidth...... 72

2.2.8. Quantitative Evaluation of the Visualisation Methods ...... 75

2.2.9. Discussion ...... 78

2.2.10. Conclusions and future Outlook ...... 84

2.3. Using Clustering Techniques to improve Visualisation of Contamination ...... 86

2.3.1. Introduction ...... 86

2.3.2. Mine action data ...... 87

2.3.3. The clustering algorithm ...... 93

2.3.4. Results and discussion ...... 100

Table of Contents

2.3.5. Conclusion ...... 107

2.4. Determining the impact ...... 108

2.4.1. Abstract ...... 108

2.4.2. Introduction ...... 108

2.4.3. Objectives of the paper ...... 108

2.4.4. Estimating the population density ...... 109

2.4.5. Estimating the ERW hazard density ...... 112

2.4.6. Combining hazard density maps with population data ...... 114

2.4.7. Results and discussion ...... 115

2.4.8. Conclusion ...... 117

2.5. Highlights of Chapter 2 ...... 118

Chapter 3. What are the contributions and limits of GIS for improving decision-making in mine action? ...... 124

Contributing research papers ...... 124

3.1. Introduction ...... 125

3.2. Choosing the Right Technique...... 128

3.2.1. Abstract ...... 128

3.2.2. Introduction ...... 128

3.2.3. Objectives ...... 129

3.2.4. Inputs ...... 130

3.2.5. The Model ...... 136

3.2.6. Benefits of the Model ...... 140

3.2.7. Conclusion ...... 141

3.3. Calculating the Shortest Path ...... 142

3.3.1. Abstract ...... 142

3.3.2. Introduction ...... 142

3.3.3. Objectives ...... 143

3.3.4. The ArcGIS Network Analyst extension...... 143

Table of Contents

3.3.5. Inputs of the Model ...... 144

3.3.6. Workflow ...... 145

3.3.7. The Model ...... 150

3.3.8. Case Study Results...... 151

3.3.9. Perspectives for NAMA ...... 152

3.4. Setting Priorities ...... 154

3.4.1. Abstract ...... 154

3.4.2. Introduction ...... 154

3.4.3. Why developing MASCOT, and for whom? ...... 155

3.4.4. Design of MASCOT ...... 157

3.4.5. MASCOT ...... 162

3.4.6. Perspectives ...... 172

3.4.7. Conclusion ...... 173

3.5. Highlights of Chapter 3 ...... 175

Chapter 4. How to best build GIS capacity in mine action? ...... 181

Contributing research papers ...... 181

4.1. Introduction ...... 182

4.2. Optimising GIS workflows ...... 185

4.2.1. Abstract ...... 185

4.2.2. Introduction ...... 185

4.2.3. The Toolbar ...... 187

4.2.4. Tool Improvement and User Help ...... 188

4.2.5. Mine Action Scenario ...... 189

4.2.6. Alternative use of START outside the Mine Action Community ...... 190

4.2.7. Conclusion ...... 191

4.3. E-learning ...... 193

4.4. Standardising Symbology ...... 194

Table of Contents

4.4.1. Objectives of this research ...... 194

4.4.2. Loose standards for cartography in mine action ...... 195

4.4.3. The 2011 recommendations ...... 196

4.4.4. Validation process ...... 199

4.4.5. Perspectives of our research ...... 200

4.4.6. Conclusion ...... 200

4.5. Sharing Data, Maps, Technologies and Processes ...... 202

4.5.1. Definitions ...... 202

4.5.2. SERWIS services ...... 205

4.5.3. From IMSMANG to the web ...... 206

4.6. Improving the Quality of Web Services ...... 209

4.6.1. Abstract ...... 209

4.6.2. Introduction ...... 209

4.6.3. Geospatial data interoperability ...... 212

4.6.4. Quality of Service (QoS) ...... 213

4.6.5. Methodology of testing ...... 214

4.6.6. Results ...... 219

4.6.7. Discussion ...... 227

4.6.8. Conclusion ...... 230

4.7. Highlights of Chapter 4 ...... 231

Chapter 5. Conclusion ...... 235

5.1. Conclusion: To what extent does GIS contribute to efficient mine action? ...... 236

5.1.1. To what extent can GIS improve visualisation of contamination and its impact on population? ...... 236

5.1.2. What are the contributions and limits of GIS for improving decision-making in mine action? ...... 238

5.1.3. How to best build GIS capacity in mine action? ...... 241

5.2. Contributions of our work ...... 246

Table of Contents

5.2.1. To provide mine action users with a new framework to make better decision ...... 246

5.2.2. To develop new tools and new material ...... 246

5.2.3. To communicate on our research ...... 247

Acronyms ...... 248

References ...... 252

List of Figures ...... 274

List of Tables ...... 278

Abstract

This document presents a study that aims at answering the question how Geographic Information Systems (GIS) can contribute to efficient mine action. In a first stage, we explore to what extent GIS can meet the needs for making visible the problem of contamination by Explosive Remnants of War (ERW) and its impact on population. To that end, we conduct an analysis of the requirements for visualising ERW of four categories of humanitarian demining stakeholders (donors and the general public, directors of national mine action authorities, operations officers and database administrators) and at four geographical scales, ranging from the municipal to the global level. We show that not one but several cartographic visualisation methods should be investigated to address these requirements. We thus explore a set of seven cartographic visualisation methods and systematically evaluate their usefulness to the four categories of stakeholders at the scale where they have to make decision. We integrate a number of challenges raised by the mapping of contamination data, such as dealing with highly heterogeneous patterns and preserving data confidentiality. Three of the seven cartographic visualisation methods are extensions of traditional kernel density estimation-based mapping. We show that with these three methods there is a serious risk of under- or over-estimation of the picture of contamination in some countries, and we propose relevant solutions to keep control over this possible drift. The main outcome of our research is to show that GIS can meet users’ needs and requirements for making the problem of ERW contamination visible. However, this cannot be achieved through the use of a unique cartographic visualisation method but requires several methods. Another important outcome is to provide mine action users with a comprehensive framework for visualising ERW and making informed decision. A third outcome of our research is the prototyping of a new cartographic module designed for implementation in the Information Management System for Mine Action – Next Generation (IMSMANG), which is the standard in use in a sixty mine-affected countries. A fourth outcome is the development of a model for assessing and mapping populations at risk of ERW. Though very simple, this model clearly shows the potential for assessing population vulnerability to landmines, provided that further research is conducted. In a second stage, we investigate to what extent GIS can help mine action users reduce time-decision and set clearance priority. Assuming that users need geospatial tools and that environmental, geographic and socio-economic conditions can have substantial influence on demining activity, we present three possible GIS-based approaches to bring auxiliary data (e.g. population density, human settlements, vegetation, cropland surface, soil characteristics, slope, roads, infrastructure, health facility, schools etc.) in the analysis. The three models are: 5D, MASCOT and NAMA. The first model, 5D, is a GIS-based analytical method for Determining and Displaying a Degree of operational Difficulty of Demining. 5D classifies degrees of difficulty as low, medium, high or extreme. Different realistic terrain factors are combined on an output map. On this basis, macro statistics can be computed for each degree of difficulty and provided to decision-makers and operators. The model is applicable to any country or any province, at the national and sub-national scales and for any demining method. We show that with further work, 5D can open the door to the possibility to estimate the financial implications of users‘ operational choices. The second

Abstract

model builds on multi-criteria analysis: we develop MASCOT, a participatory Multi-criteria Analytical SCOring Tool designed for setting clearance priority at the national and the sub-national scale. The novelty of this Spatial Decision-Support System (SDSS) is that input features (typically, ERW) receive a score in function of their Euclidian distance to real-world scoring objects. Another novel contribution of MASCOT lies in its capacity of processing vector and raster data in the same workflow. MASCOT integrates the Analytic Hierarchy Process (AHP) for achieving the weighting process. The third model that we develop builds on a GIS-based Network Analysis for Mine Action (NAMA), which could find useful applications in victim assistance and road clearance priority setting. We show that these three tools combined together provide an intuitive toolbox for mine action users to make better decision. We also show that this user community of users is not yet ready to integrate such tools and models in their everyday work. In a third stage we analyse the requirements for an optimal use of the models and tools introduced previously, and more broadly for making GIS more accessible to the mine action community. Requirements include increasing the access to core mine action and auxiliary geospatial data, optimising GIS workflows, increasing GIS resources (training and software), facilitating the dissemination of understandable maps for users inside and outside the mine action community, and encouraging standardisation of GIS processes. To address these requirements, we develop a GIS toolbox that allows enhancing preparation of geospatial data for further GIS analysis and map design. We participate in the drafting of an online training that teaches the fundamentals users need to know in order to create maps to support land-release efforts. As a contribution to standardisation of visual communication within the mine action community, we revise the IMSMA current collection of cartographic symbols. We design a prototype Spatial Data Infrastructure (SDI) for mine action including configuration of a geoportal and implementation of the whole process from obfuscating the IMSMA data by raster density calculation to their publication as web services. We provide data providers with best practices for supplying responsive map and data services. We demonstrate that these key structuring elements altogether can contribute to best build GIS capacity in mine action. Nevertheless, we show that collaboration among key mine action actors is a limiting factor towards this. In a fourth stage, we discuss limitations and perspectives of our work and we provide guidance and recommendations.

Remerciements

En préface à cette dissertation, je souhaiterais remercier un certain nombre de personnes qui ont grandement contribué à son succès. Tout d‘abord mes proches qui m'ont apporté leur support et encouragé tout au long de cette grande aventure. Celle-ci a débuté en février 2010 lorsque le Prof. Anthony Lehmann et le Dr. Daniel Eriksson ont mis en place un partenariat académique entre l‘Université de Genève et le Centre International de Déminage Humanitaire de Genève (GICHD). Je suis très reconnaissant au Prof. Lehmann, qui a supervisé cette thèse, pour son soutien et ses conseils. Je souhaite également remercier le Dr. Eriksson, co- superviseur de ce travail de thèse, pour m‘avoir offert l‘opportunité de servir cette noble cause qu‘est le déminage humanitaire. Nos nombreux échanges ont considérablement élargi mon horizon, tant d‘un point de vue scientifique et technique qu‘humain. Un acteur important au cours de ces trois années de recherche a été le Prof. Robert Weibel, membre de mon Jury et avec qui j‘ai eu l‘opportunité de collaborer à un travail portant sur la représentation des mines et restes explosifs de guerre. Sa disponibilité et ses précieux conseils m‘ont été très précieux. Merci au Prof. Hy Dao pour avoir accepté de participer à mon jury de thèse et pour son retour sur mon travail. Je voudrais exprimer ma gratitude au Prof. Martin Beniston, qui dirige le groupe Climate Change and Climate Impacts à l‘Institut des Sciences de l‘Environnement de l‘Université de Genève, le Prof. Walter Wildi, Directeur de l‘Institut Forel de cette même Université ainsi que l‘Ambassadeur Stéphane Husy, Directeur du GICHD, qui m‘ont accueilli dans leur institution respective. Je remercie sincèrement le Dr. Nicolas Ray pour ses précieux conseils et sa collaboration autour de deux papiers présentés dans cette monographie. Mon travail de thèse n‘aurait pu être accompli sans l‘assistance du Dr. Grégory Giuliani. Son enthousiasme et ses encouragements m‘ont accompagné tout au long de ces trois années. J‘ai beaucoup appris de lui, notamment lors de notre collaboration autour de la qualité des services géographiques. Je voudrais aussi exprimer ma gratitude à Valentina Bigoni, Olivier Cottray, Inna Cruz, Alain Dubois, Rocío Escobar, Eva Fernandez, Gissela Girón, Zoë Goodman, Jonas Herzog, Diana Joaqui Lopez, Aurora Martinez, Anne-Li Nauclér, Pablo de Roulet, Jean-Paul Rychener, Helder Santiago, et Julia Schwank qui ont toutes et tous contribué à leur manière à l‘accomplissement de ce travail. Merci aux institutions suivantes pour m‘avoir donné accès à leurs bases de données : le Directorate of Mine Action (DMA), le Lebanon Mine Action Centre (LMAC), le Mine Action Centre in Cyprus (MACC), le Mine Action Coordination Centre of Afghanistan (MACCA), le Programa Presidencial Para la Acción Integral contra Minas Antipersonal (PAICMA), le Tajikistan Mine Action Centre (TMAC), et United Nations Mine Action Service (UNMAS). Ma gratitude aux personnes suivantes pour leur support: Karin Allenbach, Dr. Andrea de Bono, Prof. Hy Dao, Emanuele Gennai, Dr. Stéphane Goyette, Yaniss Guigoz, Dr Enrique Moran, Fabio Oliosi, Dr Kazi Rahman, Ana Silva, Olivier Travaglini et Ron Witt.

Remerciements

Je remercie également le staff technique et administratif de l‘Institut des Sciences de l‘Environnement, de l‘Institut Forel ainsi que celui du GICHD, qui m‘ont permis d‘accomplir ce travail dans des conditions très favorables. Enfin, je remercie trois enseignants qui ont compté au cours de mes études: Michèle Béguin (Université de Paris - Sorbonne), le Dr. Claude Chambon (Ecole Nationale Supérieure des Mines de Nancy) et Gérard Chappart (Ecole Nationale des Sciences Géographiques).

Supervisory Team & External Advisors

Prof. Anthony Lehmann Professeur Associé Directeur du Laboratoire enviroSPACE Institut des Sciences de l‘Environnement Université de Genève

Dr. Daniel Eriksson Head, Management Consulting Geneva International Centre for Humanitarian Demining

Prof. Robert Weibel Professor of Geographic Information Science Head, Geographic Information Systems Division Zürich University

Prof. Hy Dao Professeur titulaire Département de Géographie et Environnement Université de Genève

Chapter 1. Introduction

Chapter 1: Introduction

1.1. Structure of the thesis

This thesis on the contributions of geographic information systems (GIS) for efficient mine action is divided into five chapters. Each of them contributes to answer the different research questions that will be asked later in this introduction. Chapter 1 provides an overview of basic concepts underlying mine action, to give to the reader all the necessary knowledge to go through the rest of the thesis. In this chapter, we put in evidence strong needs for GIS in the mine action community. We identify different categories of stakeholders that could significantly benefit from GIS at the geographical scale at which they have to make decisions. Chapter 1 also defines the three research questions to be addressed and gives an overview of the projects that were conducted to support the associated research and of the methodology that was chosen to conduct these projects. In Chapter 2, we investigate and discuss how maps can make the problem of contamination by Explosive Remnants of War (ERW) visible, and to what extent GIS can meet the needs of humanitarian demining stakeholders. We integrate a number of scientific and technical challenges raised by the mapping of contamination data, such as finding a good compromise between providing close-to-reality representations and preserving non-disclosure, and avoiding under- or over-estimating the picture of contamination in some countries. We provide mine action stakeholders with a novel and comprehensive framework for visualising ERW at the geographical scale at which they have to make decisions, as well as customised visualisation methods and recommendations to help them make informed decisions. We investigate in Chapter 3 to what extent GIS can improve decision-making in mine action. Assuming that environmental, geographic and socio-economic conditions, as well as human activity, have substantial influence on demining activities, we present possible approaches to combine contamination data with other geospatial factors through multi-criteria and transportation network analysis models. To help users integrate these models in their everyday work, we implement them as user-friendly and flexible GIS tools. Chapter 4 analyses the requirements for making GIS more accessible to mine action users. We show that an optimal use of the GIS models developed in Chapter 3 requires improving the access of mine action stakeholders to geospatial data. Other requirements include increasing GIS users‘ expertise, optimising preparation of GIS data, facilitating the production and sharing of understandable maps, and encouraging standardisation of GIS processes. As a response to these requirements, we develop tools for optimising GIS workflows, we publish an online course that teaches the basics of map creation to support land release efforts, we revise the existing collection of mine action cartographic symbols and propose the updated version as a standard and common visual language for GIS users, and we design the technical specifications of a Spatial Data Infrastructure (SDI) for mine action. Finally, we provide guidance to data providers to supply responsive data and map services.

Chapter 1: Introduction

Each one of Chapters 2, 3 and 4 is followed by a few pages titled ‗Highlights of Chapter N‘ which summarise the key ideas and results of the Chapter. Chapter 5 concludes this thesis by answering the three research questions, discussing perspectives of the projects and making recommendations for the future.

Chapter 1: Introduction

1.2. Contributing research papers

 Lacroix P., Herzog J., Eriksson D., Weibel R. (2013). Methods for Visualising the Explosive Remnants of War. Applied Geography, 41:179-194. Available from: http://www.sciencedirect.com/science/article/pii/S0143622813001021  Lacroix P., Herzog J., Eriksson D. (2011). Mapping Populations at Risk of ERW. The Journal of ERW and Mine Action, 15(1). Available from: http://maic.jmu.edu/journal/15.2/specialrpt/lacroix/lacroix.htm  Lacroix P., Escobar R. (2012). 5D: a GIS-based approach for Determining and Displaying a Degree of operational Difficulty of Demining. The Journal of ERW and Mine Action, 16(3). Available from: http://maic.jmu.edu/journal/16.3/rd/lacroix.htm  Lacroix P., De Roulet P., Escobar R., Cottray O. (2013). NAMA: A GIS-based Network Analysis approach for Mine Action. Accepted for publication by the Journal of ERW and Mine Action  Lacroix P., Santiago H., Ray N. MASCOT: Multi-criteria Analytical SCOring Tool for ArcGIS Desktop. Submitted to the International Journal of Information Technology and Decision Making  Lacroix P., de Roulet P., Ray N. (2013). Simplified Toolbar to Accelerate Repeated Tasks (START) for ArcGIS: Optimising Workflows in Humanitarian Demining. Accepted for publication by the International Journal of Applied Geospatial Research.  Giuliani G., Dubois A., Lacroix P. (2013). Testing OGC Web Feature and Coverage Services performance: towards efficient delivery of geospatial data. Accepted for publication by the Journal of Spatial Information Science. Available from http://www.josis.org/index.php/josis/article/view/112

Chapter 1: Introduction

1.3. Mine action background

This section will present an overview and provide definitions of basic concepts underlying mine action. We will focus on the five pillars of mine action, information management, the main types of landmines, their characteristics and how they are stored in Relational Database Management Systems (RDBMS). With the following glossary, the reader will have all the necessary knowledge to go through the monograph. This background chapter on mine action will briefly evoke the international treaties, the five pillars of mine action, the work of the GICHD and the issue of information management in mine action. About eighty countries are affected by landmine and ERW contamination situated on five continents (Figure 1). Thousands of people are injured or killed in accidents worldwide every year with an estimated 75 percent of civilian casualties. For the year 2010, 4‘191 victims where reported, among which 1‘155 were killed. ICBL (International Campaign to Ban Landmines) reports that the actual number is however likely higher, as not all accidents and victims are systematically reported (ICBL 2011a). The issue of contamination by ERW will remain an issue as long as there are armed conflicts. This situation means that expertise on the different aspects of mine action will still be needed in the long term.

Figure 1: Mine contamination as of 2011 (source: ICBL 2011b)

Chapter 1: Introduction

The cost of clearance is variable depending on the area and the demining technique, but can be estimated to a minimum of $1‘000‘000 per square kilometre. In 2010, more than 200 square kilometres were cleared worldwide by states, Non-governmental organisations (NGOs) and commercial companies in which 400‘000 anti-personnel and 30‘000 anti-vehicle mines were destroyed.

1.3.1. International treaties

Several treaties and protocols rule on the regulation or ban of landmines and conventional weapons. The three most important of these treaties are the mine Ban Treaty (also commonly called the Ottawa Convention), the Protocols II and V of the Convention on Certain Conventional Weapons (CCW) and the Convention on Cluster Munition (CCM). The CCW was signed in 1980 to regulate the use of conventional weapons. Only two of the five Protocols of the Convention relate to mine action, the amended Protocol II of 1996 to landmines and booby traps and the Protocol V of 2003 to cluster munitions. In the framework of the CCW which aims to reduce the ―excessively injurious‖ harm and is a general pledge to disarmament, both protocols regulate the use of these types of weapons, including by forbidding the use of some deemed ―inhuman‖ models. It does not however forbid altogether its use, trade and stockpiling. This limitation of the CCW led several states and representatives of civil society to develop treaties that would in fact ban these types of weapons: the 1997 International Mine Ban Treaty and the 2008 Convention on Cluster Munitions (Maslen 2001). The Mine Ban Treaty currently has 160 state parties. State parties of the treaty agree never to use, develop, produce, stockpile or trade anti- personnel landmines. It must also clear all laid anti-personnel landmines and destroy its stockpile, in respectively ten and four years for each task. Within their means, states should also provide assistance for mine awareness, stockpile destruction, and victim assistance activities worldwide. The convention on cluster munitions was adopted in 2008 as a stronger alternative to Protocol V of the CCW. Inspired by the Mine Ban Treaty this new treaty did not have the same wide success, with 76 state parties to date. The CCM prohibits the use, stockpiling production and transfer of cluster munition. Separate articles in the Convention concern assistance to victims, clearance of contaminated areas and destruction of stockpiles (Björk 2012). The implementation of the Mine Ban Treaty signifies in most states a significant reduction of stockpiles, and its complete destruction in dozens of states. Although, a majority of countries worldwide are members of the Mine Ban Treaty, several non-signatory states still lay anti-personnel mines. Worryingly, three non-signatories countries that had nevertheless stopped laying landmines years ago restarted in recent years. Despite a global and steady reduction of both stockpiles and contaminated land, the fact that thirteen states1 still refuse to prohibit production of anti-personnel mines still poses a risk of a renewal of contamination in the future.

1 China, Cuba, India, Iran, Myanmar, North Korea, Nepal, Pakistan, Russia, Singapore, South Korea, United States of America, and Vietnam

Chapter 1: Introduction

1.3.2. Mine action

Mine action is generally defined as ―activities which aim to reduce the social, economic and environmental impact of mines and UXO‖ (UNMAS 2003, 04.10). With this definition, mine action is not limited to demining but aims at reducing the risk of landmines to societies. For this aim, it is conventionally divided into five pillars with the goal of reducing the impact of mines and ERW (Borrie 2009). The five pillars are the following:  Humanitarian demining: the removal of mines and ERW as well as the marking and fencing of contaminated areas. It includes aspects related to information management and mapping.  Mine risk education: helps people living in contaminated areas understand the risks they are exposed to, how to identify mines and ERW and how to stay safe.  Victim assistance: it includes all medical assistance to victims, such as rehabilitation and reintegration, including job skills.  Stockpile destruction: all tasks related to the destruction of mines stockpile to comply with the requirements of the mine ban treaty.  Advocacy against the use of anti-personnel mines: it includes encouraging countries to participate to international meetings and treaties and conventions to end the production, trade and use of landmines.

1.3.3. Humanitarian demining

Humanitarian demining, also called land release or land clearance, is concerned with all aspect of the removal of mines or ERW that have been laid in the field. Unlike military demining, which simply aims at clearing a path to pursue operations, the goal of humanitarian demining is to render civilian areas safe. As a general pledge it aims at helping civilians to return to normal life after a conflict. It includes the removal of landmines and ERW, but also survey, marking, mapping and more generally information management (Björk 2012). Mine clearance is done by operation officers and managers, along with survey and marking.

1.3.4. Information Management

The crucial aspect of information management includes data collection, data preservation, the use of data and data dissemination. Overall, information management in mine action aims to ―support operational staff by giving them access to more relevant information on which to base their decisions‖ (Eriksson 2008). Information management follows a cycle of four main steps, starting with an assessment of the information needs and followed by data collection, data analysis and information dissemination. The end of the cycle means that a new assessment covering new needs has to be done along with all information updates. For this reason mine action programmes use RDBMS to cover their information management

Chapter 1: Introduction

needs.

1.3.5. Information Management System for Mine Action (IMSMA)

Countries that have signed the anti-personnel Mine Ban Treaty are obliged to collect, analyse and report spatial data on mine action. The Information Management System for Mine Action - Next Generation (IMSMANG) application uses ArcGIS Engine for its spatial functionality. The software has been available in its current form since 2006. The first generation of the software was based in different technology. The first deployment in Kosovo 1999 only focused on supporting business processes and did not have GIS functionality. In the current generation, the application offers useful cartography functions such as visualising the hazards in the database, and designing layouts. IMSMANG uses an integrated MySQL server for data storage. Each country can store data in a customised way. The data types, scales and coordinate systems are different from one country to another. The data may also be heterogeneous inside a country where contaminated areas in one part of the country might mainly be cluster bomb strikes and minefields in another. The data collection forms have to be adjustable to allow for such differences. IMSMANG comes with sets of functionality for information management of contaminated areas, hazard reduction activities (such as clearance and marking), accidents, victims (as well as rehabilitation), risk education (activities to inform the population how to live more safely near contaminated areas), and quality management (to control the activities and the information). The core principle in the development of IMSMA was to provide a comprehensive software package that could be used by national governments in the developing world. A decade later, the software package has emerged as one of the most successful information management project in the humanitarian domain. In 2011, the GICHD and its IMSMA team was awarded the ―Making a difference award‖ by Esri at the user conference in San Diego2. The tool has been translated to numerous languages and is installed in 60 countries on 1‘200 computers and approximately 3‘000 users have been trained. The challenges that had to be overcome were the very limited computer literacy in some organisations and the vast differences in information needs between the various user organisations. The IMSMA package solved this by being highly customizable, modular and enabling gradual growth in complexity of the used functionality. All information contained in IMSMA can be referenced with geospatial information in the form of pair of XY coordinate points, either to mark a single object, a line or an area. It allows however only little possibilities in terms of GIS and its main strength lay in the updates capabilities of the software in the information management cycle. The data update frequency in IMSMA, however, is extremely variable according to the country and its level of contamination. A few updates per year only are necessary for countries like Nicaragua, Zambia and Kosovo. In the other hand strongly affected countries like Afghanistan require several thousands of updates per year.

2 http://www.flickr.com/photos/esri/5930839475/

Chapter 1: Introduction

1.3.6. The Geneva International Centre for Humanitarian Demining (GICHD)

Dozens of states, NGOs and commercial companies are involved in mine action and humanitarian demining. Among those actors, the GICHD is an international NGO of experts established in April 1998 and legally based in Switzerland (GICHD 2012). The GICHD works for the elimination of anti-personnel mines, cluster munitions and other ERW. It wishes to contribute to the social and economic well-being of people and communities in affected countries. The Centre respects the lead of the national mine action programmes working closely with them, cooperating with other mine action organisations, and following humanitarian principles of humanity, impartiality, neutrality and independence. A key task of the GICHD is to provide advice and support capacity building. It undertakes applied research on different aspects of mine action, disseminates knowledge and best practices, and develops standards. The goal of its activities is to enhance performance and professionalism in mine action, and supports the implementation of the Mine Ban treaty, the CCM and other instruments of international law related to landmines and ERW. Among other activities the GICHD is responsible for the development of the IMSMA software and providing training. Another important role of the GICHD in mine action worldwide is its implication in the International Mine Action Standards. Mine action, like all specialised fields, has specific terms. In order for the large array of states, NGOs and commercial companies to best coordinate, the International Mine Action Standards (IMAS) were established to find common practices and definitions. The first IMAS were issued in 2001 and were updated several times since, to adapt to the changing practices. The standards are prepared by the GICHD on behalf of UNMAS. They are adapted to the requirement of the Mine Ban Treaty (UNMAS 2003, 01.10).

1.3.7. Glossary

1.3.7.1. Hazard and Hazardous area

Hazard: potential source of harm (UNMAS 2003, 04.10, 3.126). In the context of mine action, it can typically refer to a minefield or an Unexploded Ordnance (UXO).

Hazardous area: a generic term for an area perceived to have mines and/or ERW (UNMAS 2003, 04.10, 3.127).

Confirmed Hazardous Area (CHA) is an area identified by a non-technical survey in which the necessity for further intervention through either technical survey or clearance has been confirmed (UNMAS 2003, 04.10, 3.47). Defined Hazardous Area (DHA) is an area, generally within a CHA, that requires full clearance: a DHA is normally identified through thorough survey (UNMAS 2003, 04.10, 3.56). The DHA term is not anymore in use by IMAS. Suspected Hazardous Area (SHA) is an area suspected of having a mine/ERW hazard (UNMAS 2003, 04.10, 3.276). A SHA can be identified by an impact survey, other

Chapter 1: Introduction

form of national survey, or a claim of presence of explosive hazard.

These three types of areas defined as such since 2003 are the product of different steps in the land release process. SHAs are identified by the claim of mine or ERW presence through an impact survey or by the local population. It does generally not have a precisely known perimeter. A non-technical survey is done on a SHA to define the borders of one or several CHAs, generally represented as polygons, within the SHA. Parts of the SHA that are not included in the CHAs are called cancelled areas. The characteristics of the CHA, such as the type of contamination, are known more precisely. Thorough survey, generally a technical survey on a CHA allows to identify one or several DHA. The CHA can be released, while the DHA requires full clearance. In future, new terms will likely replace current IMAS definitions (APOPO et al. 2012). SHA will be replaced by ATS (Area Targeted for Survey). DHA will be suppressed. The term ―CHA‖ will include both what is currently understood as CHA and DHA.

1.3.7.2. Contamination types

Anti-Personnel mines (AP): a mine designed to be exploded by the presence, proximity or contact of a person and that will incapacitate, injure or kill one or more persons (UNMAS 2003, 04.10, 3.15). Mines designed to be detonated by the presence, proximity or contact of a vehicle as opposed to a person that are equipped with anti-handling devices. Other types of landmines are the anti-vehicle mines and the anti- tank mines.

Explosive Remnant of War (ERW): the term refers to explosive munitions left behind after a conflict has ended. They include unexploded artillery shells, grenades, mortars, rockets, air-dropped bombs, and cluster munitions. ERW consist of unexploded ordnance (UXO) and abandoned explosive ordnance, but not mines.

Sub-munition: any munition that, to perform its task, separates from a parent munition. Any mine or munition that form part of a cluster bomb unit, an artillery shell or a missile payload (UNMAS 2003, 04.10, 3.272).

Unexploded Ordnance (UXO): Explosive Ordnance that has been primed, fuzzed, armed or otherwise prepared for use or used. It may have been fired, dropped, launched or projected yet remains unexploded either through malfunction or design or for any other reason (UNMAS 2003, 04.10, 3.293).

1.3.7.3. Processes

Battle Area Clearance (BAC): the systematic and controlled clearance of hazardous areas where the

Chapter 1: Introduction

hazards are known not to include mines (UNMAS 2003, 04.10, 3.20).

Community Liaison (CL): also called ―community mine action liaison‖. Liaison with men and women in mine/ERW affected communities to exchange information on the presence and impact of mines and ERW, create a reporting link with the mine action programme and develop risk reduction strategies. Community liaison aims to ensure that the different community needs and priorities are central to the planning, implementation and monitoring of mine action operations (UNMAS 2003, 04.10, 3.44). Community liaison is based on an exchange of information and involves men, women, boys and girls in the communities in the decision making process (before, during and after demining), in order to establish priorities for mine action. In this way mine action programmes aim to be inclusive, community focused and ensure the maximum involvement of all sections of the community. This involvement includes joint planning, implementation, monitoring and evaluation of projects. Community liaison also works with communities to develop specific interim safety strategies promoting individual and community behavioural change. This is designed to reduce the impact of mines/ERW on individuals and communities until such time as the hazard is removed.

Hazard reduction: the term does not have an IMAS definition. It concerns all type of activities directed towards rendering safe a hazardous area, including all types of surveys and clearance.

Hazard status: different status is defined for a hazard: active, closed, transitional/ongoing and cancelled. An active hazard refers to an area or object that has been deems hazardous and is recorded as such. A closed hazard has been cleared. A transitional/ongoing hazard corresponds to an area where hazard reduction is in the process. A cancelled hazard has been declared safe following a Non-Technical Survey (NTS).

Impact Survey: an assessment of the socio-economic impact caused by the actual or perceived presence of mines and ERW, in order to assist the planning and prioritisation of mine action programmes and projects (UNMAS 2003, 04.10, 3.137).

Landmine Impact Survey (LIS): refers to Impact Survey as defined by IMAS (Björk 2012).

Mine Risk Education (MRE): activities that seek to reduce the risk of injury from mines/ERW by raising awareness of men, women, and children in accordance with their different vulnerabilities, roles and needs, and promoting behavioural change including public information dissemination, education and training, and community mine action liaison (UNMAS 2003, 04.10, 3.1839).

Non-Technical Survey (NTS): survey activity that involves collecting and analysing new and/or existing

Chapter 1: Introduction

information about a suspected hazardous area. Its purpose is to confirm whether there is evidence of a hazard or not, to identify the type and extent of hazards within any hazardous area and to define, as far as is possible, the perimeter of the actual hazardous areas without physical intervention. A non-technical survey does not normally involve the use of clearance or verification assets. The results from a non- technical survey can replace any previous data relating to the survey of an area (UNMAS 2003, 04.10, 3.197).

Quality Assurance (QA): part of Quality Management focused on providing confidence that quality requirements will be fulfilled (UNMAS 2003, 04.10, 3.228).

Quality Control (QC): part of Quality Management focused on fulfilling quality requirements (UNMAS 2003, 04.10, 3.229). QC relates to the inspection of a finished product. In the case of humanitarian demining, the ―product‖ is safe cleared land.

Task: the term ―Task‖ does not have an IMAS definition. It refers to an action being taken in the field of mine action that may cover one or several hazardous areas. It may include diverse type of activities, such as MRE, Surveys, and Clearance Quality Assurance. Similarly as for hazard reduction, possible status for a task is:  Ongoing: the task is being carried out by a mine action personnel.  Suspended: the task is temporarily stopped.  Cancelled: the task issued is cancelled. Cancelled is used here in its literal meaning and should not be confused with a hazard being cancelled through a NTS.  Issued: an issued task refers to assignment to this task to a mine action personnel (e.g. NGO, commercial company).  Planned: a planned task refers to the strategic decision in the long term.  Completed: the task that was assigned is finished according to the terms of reference when it was issued.

Technical survey (TS): detailed intervention with clearance or verification assets into a CHA, or part of a CHA. It should confirm the presence of mines/ERW leading to the definition of one or more DHA and may indicate the absence of mines/ERW which could allow land to be released when combined with other evidence (UNMAS 2003, 04.10, 3.281).

1.3.7.4. Demining Methods

Three major methods are used for land clearance:  Animal detection refers to the use of animals trained to detect mines through the scent of

Chapter 1: Introduction

explosives. Mine Detection Dogs are traditionally used for this task, but rats are also increasingly used for these tasks. A proposal to include Animal Detection System as an IMAS definition is currently under review.  Manual demining refers to the manual removal of mines or ERW from a terrain, using metal detectors and digging tools. It is the most common method for land clearance. It has the particularity of being the least expensive of the different methods for clearance, but it is also slow. Important limitations occur when mines contain little or no metal, and/or when the terrain contains high level of metal, which is often the case in dwelling areas.  Mechanical demining operations refer to the use of machines in demining operations and may involve a single machine employing one mechanical tool, a single machine employing a variety of tools or a number of machines employing a variety of tools (UNMAS 2003, 04.10, 3.166).

Chapter 1: Introduction

1.3.8. Photo gallery

The following pictures illustrate some of the concepts and definitions described above.

Chapter 1: Introduction

1.4. User needs

1.4.1. Who are the end-users of this research?

Before analyzing end users‘ needs, it is important to specify who they are. Within the overall mine action process, four main groups of stakeholders may be distinguished:  Users outside the core mine action domain are the donors (public and private organisations and individuals) and the general public, considered as a potential donor. They need a reliable indicator of the progress of mine action activities in order to decide which country to fund as well as which activity (e.g. landmine clearance, mine risk education). In 2009, 83% of the funding for mine action came from international sources (Devlin and Naidoo 2010). In 2010, the international donors funded mine action activities up to US$ 480 million, including US$ 100 million for Afghanistan only. The top six recipient states were Afghanistan, Angola, Iraq, Sudan, Sri Lanka, and Cambodia, representing 55% of all international contributions. Donors no longer consider mine action as an immediate humanitarian response, but as part of a broader process including conflict prevention, protection, socio-economic impact, reintegration (Devlin 2010), humanitarian assistance and care for survivors (Devlin and Naidoo 2010). Donors have a priori low GIS expertise.  The directors of national mine action authorities ensure that mine action data is collected for their country in compliance with international standards and policies (GICHD 2007, GICHD and UNMAS 2011). They also enable strategic decisions at country level by undertaking landmine impact surveys (LIS) to assess socio-economic impacts of ERW on communities, and they coordinate the regional activities of demining organisations. They work in collaboration with other international and national bodies, governments, communities and field operators, and regularly produce overviews of their goals and achievements for distribution to donors and the broader mine action community (UNDP 2011). In its guidelines for policy and programme development intended for mine-affected states (GICHD 2009a), the GICHD recommends national mine action authorities to strengthen information sharing and cross-sector collaboration with different actors.  The operations officers in a mine action authority may be military, NGOs and commercial demining companies employing local individuals specifically trained for clearance activities. Operations officers intervene in operational planning as well as in the demining operations themselves. They are part of a small to large scale prioritisation process: they first refer to information at the regional or sub-regional level (e.g. the results of a LIS) to prioritise the areas to survey or to clear. In a second step, other elements such as local infrastructures, land cover,

Chapter 1: Introduction

vegetation, and topography may help operations officers decide how to access these areas. Operations officers are experts in mine action and explosive ordnance disposal and do not necessarily have GIS expertise.  The database administrators are in charge of probing the national IMSMANG repository for incompleteness or inaccuracy. Checking spatial data attributes such as coordinates, area type and area are common tasks undergone at large scales (between 1:50‘000 and 1:5‘000) in collaboration with field operators. Database administrators are experts in IMSMANG and not necessarily in GIS.

1.4.2. GIS needs

In this paragraph we summarise users‘ needs relative to cartography and GIS. Needs analysis was achieved from different sources:  Literature review,  Two focus group meetings held in 2011 and 2012 and regrouping a twenty mine action or GIS experts or both3.  A two-day end-user workshop involving a dozen users from information management, national programme management, operations and database management4.  The results of the Space Assets for Demining Assistance (SADA) feasibility study. Conducted by the European Space Agency (ESA) and advised by the GICHD, the SADA activities aim at proving the viability and sustainability of an integrated set of services supporting land release in mine action, with a focus on space assets and GIS technologies (ESA 2012). Following an open competition, three consortia5 have been asked to identify users‘ needs and requirements and to propose technological and operational solutions. Within this framework, a survey covering 37 end-user organisations has been organised.  Discussions and brainstorming with dozens of users from organisations involved in humanitarian demining and/or in GIS, including national mine action programme management, strategic management, information management, operations, academic institutions, NGOs and others. In particular, numerous discussions were held with GICHD staff who are in frequent contact with decision makers in the community as part of their work and who are also responsible for the development of the GIS functionalities in IMSMANG.  A one-week field immersion in direct contact with the Albanian Mine Action Executive (AMAE), meeting demining and medical staff, governmental authorities and GIS users.

3 These focus groups will be described in detail in Section 2.2 4 This end-user workshop will be described in detail in Section 4.2 5 Led respectively by Infoterra UK, RadioLabs and Ingenerio y Servicios Aeroespaciales (INSA)

Chapter 1: Introduction

1.4.2.1. Needs for maps

Mine clearance is one of the five pillars of mine action6. Mine clearance includes surveying, clearance of unexploded ordnance and mines, marking of unsafe areas, and mapping. Maps that are produced by impact and technical surveys supply baseline data for clearance organisations to plan the operations in the field (UNMAS 2010). But the needs for maps in the humanitarian demining field are not limited to operational activities. In particular, countries that have signed the anti-personnel mine ban treaty are obliged to collect, analyse and report spatial data on mine action. As such, directors of national mine action authorities regularly provide donors and the broader mine action community with overviews of their goals and achievements. Similarly, donors and the general public need a global overview of the contamination problem to decide which country to fund.

1.4.2.1.1. Data disclosure policies in the context of mine action

Several major issues rise when wanting to map mine action-related information. One issue relates to the conflictual and controversial nature of mapping in a post-conflict context, both for question of national sensibilities, concerning the location of borders and on the practical aspect of the defence policies of countries. The question of the access to data is important and is strongly influenced by the very nature of the contamination. Landmines and ERW are deployed in the case of conflicts. Although demining operations generally occur in post-conflict periods, past armed confrontations often still are present and their political origin stays partly unresolved. Ancient belligerents, might still be in a frozen conflict, might not have diplomatic relations, and even in the cases where there are, the relationship between countries may be still very cold. Ongoing existence of the conflict affects particularly the sensitivity to geographic representations in cases where a territory is disputed. The map of the outline border of a country may be very controversial, if, for example it does not include the territory claimed by one or the other warring party. As pointed out by O'Shea (1994), cartography constitutes an extremely powerful language in the constitution of a national discourse and is very likely to be strongly opposed if it is not deemed satisfying by one of the parties. In this context, the notion and goal of neutral mapping become extremely sensitive, as a border drawing is potentially subject to controversy. In fact where the line a border might be seen as a harmless element in a small scale map by someone not directly concerned, national representatives of different countries in conflict can easily only focus on this issue over all other questions at stake (Raleigh et al. 2010). This may harmfully divert attention from the proper humanitarian concern at stake. Another issue is the question of data privacy and disclosure, sensitive because potentially touching private issues for victims and affected communities. Countries may not want to show the extent of remaining contamination. As for any type of military-related information, the presence of ERW and

6 The other four are mine risk education, victim assistance, stockpile destruction and advocacy

Chapter 1: Introduction

landmines is very sensitive, and countries may not want to disclose it, while in the same time, they may want to show that their country is affected. The question of disclosure is especially problematic in countries that are not signatories of the mine ban treaty, as they consider landmines as part of the defence policy (ICBL 2011a). Similarly, one other key important reason motivating non-disclosure of data related to landmines is its intricated relationship with health issues. As pointed out by Andersson and Mitchell (2006), the question of data privacy for victims of ERW is a key ethical requirement for analysis, and a limited level of detail might even benefit planning, giving possibly wider access to the data. It is very legitimate that victims of ERW want to protect their records from being made public. As suggested by Taylor (2002), methods for collecting and storing data in the context of mine action victims should be attentive to protect the privacy of health-care recipients. And logically, it is also legitimate that the GIS treatment of victim-related mine action data is required to follow guidelines protecting victims from their records being exposed nominally, or in a way in which they may get recognised by location. Finally, it is very important for the mine action community that maps that are shown online do not give the exact location due the high data sensitivity. Exact locations could be used by civilians or criminals to either attempt to navigate through contaminated areas or to steal landmines to sell the explosives on the black market.

1.4.2.2. Needs for auxiliary data

Mapping mine action information requires accessing not only data from the core mine action domain, but auxiliary data as well. The results of the SADA feasibility study (ESA 2012) show that about 50% of the mine action centres see as very high the relevance of general mapping (e.g. infrastructures, roads, villages, rivers) for their organisation and region/country to support land release monitoring and reporting to donors. This study also suggests that environmental factors (e.g. humidity, slope, surface, roughness of a hazardous area, soil classification and land cover including vegetation) can have significant influence on mine action resource selection, and that maps combining mine information with information indicating areas with high density of human activity can help decision-makers to prioritise areas and to plan demining activities. Likewise, the study stresses the needs for quick access to socio-economic data for producing risk maps and determining socio-economic impacts of ERW.

1.4.2.3. GIS capacity needs

GIS capacity is heterogeneous across the sixty mine-affected countries. The division roughly contrasts poor and rich programmes. Well-funded programmes have good resources in terms of GIS licences and trained staff. It is the case of UNMAS-managed programmes – there are about ten of them – and programmes in developed countries like Argentina, Chile, Finland, Israel and Turkey. Providing a

Chapter 1: Introduction

detailed picture of software capacity in the mine action community is quite complex. Many programmes are equipped with GIS tools, and most of them with the Esri suite. Among them, numerous unofficial ArcGIS 9.3.x and 10.x licences are circulating. Although 150 ArcGIS Desktop licences have already been provided by GICHD to Esri7 to be used in national programmes, they have yet not been distributed. In some countries like Bosnia and Herzegovina the main cartographic software remains MapInfo (Grujic 2011). Many other GIS utilities and plug-ins are in use within the mine action community (e.g. MGRS Conversion Utility (Mentor Software INc. 2012), XTools (Data East 2012) etc.), especially for specific GIS functions requiring transformation of geographic coordinates from one format to another (e.g. Military Grid Reference System (MGRS) to decimal degrees (DD)) or feature conversion from one geometry type to another (e.g. transformation of ERW points into polygons). GIS expertise too is disparate across different countries. Some programmes are using GIS tools on a daily basis, while in others GIS expertise is very low. Within this context, Eriksson (2011) points out an increasing amount of requests from national mine action programmes for the use of advanced information technology, including ―functionality that is too complex or costly to include in the regular IMSMANG‖.

1.4.2.4. Looking for more efficient mine action

The needs for improving efficiency in mine action are palpable. A basic search of the term ―efficiency‖ 8 on the GICHD website search engine returns 1‘060 results. In comparison, the acronym ―GICHD‖ is displayed 22‘200 times. The word ―effective‖ carves itself a place of predilection with 1‘200 results, ―speed‖ is displayed 1‘000 times, ―efficient‖ 663 times, and ―reduce/reducing costs‖ 769 times.9 The conclusions of the SADA feasibility study (ESA 2012) are in line with these results. Recognizing that between 90% and 97.5% of the suspected land proves in hindsight to be non-contaminated, the authors suggest (1) focusing efforts on clearing minefields that are most threatening and costly to society, (2) avoiding the unnecessary deployment of clearance activities in non-contaminated areas, and (3) reducing the cost of detection and clearance per unit of land area. From a scientific and technical perspective, they see in new methodologies and technologies like remote sensing and GIS relevant support for contributing to the achievement of these three goals. More concretely, they point out strong needs for improving efficiency in collection and integration of field level data, for increasing the accuracy of data georeferencing and for optimising data transfer from in-field workers to decision-makers. In line with the latter need, workshops and discussions with end-users (see Section 1.4.2) have highlighted the complexity of frequent GIS workflows. In particular, the design of maps from mine action data often requires repetition of numerous tasks such as data extraction from the IMSMANG repository,

7 Based on personal communication with Esri 8 http://www.gichd.org 9 Figures are of September 2012

Chapter 1: Introduction

conversion of XY coordinates (or bearings and distances) to points/lines/polygons, assignment of appropriate spatial reference10, clipping to study area boundaries, merge of IMSMANG data with in-field data and auxiliary data, and integration of meaningful cartographic elements. Users are seeking to reduce the time spent on these workflows.

1.5. State of the art: GIS in mine action

As of 2013, a few number of attempts have been made to apply GIS and cartographic methodologies in the humanitarian demining field. Most studies concentrate on the use of sensors to detect individual landmines using ground-penetrating radar (Havens et al. 2009) or hyper-spectral imaging (Zare et al. 2008, Wong 2009). Benini (2000) and later Benini et al. (2003) combine contamination data with social and economic data at country level to help setting clearance priority. Williams and Dunn (2003) explore the potential for using GIS in participatory approaches to assess the risk of landmines to local communities in Cambodia. Riese et al. (2006) propose a GIS-based approach to make probabilistic forecasts about the presence of ERW, with the goal of supporting decision-making in the allocation of demining resources, relocation of refugees and peacekeeping operations. As a further probabilistic approach, Vistisen (2006) employs Bayesian inference to develop a risk model quantifying to what extent a minefield poses a risk to a society. Andersson and Mitchell (2006) use inverse-distance weighted interpolation to generate population-weighted raster maps for use in the evaluation of mine risk education. More recently, Alegría et al. (2011) use a variety of geostatistical techniques as well as kernel density estimation (KDE) to analyse and map ERW risk, with a focus on exploring the utility of different analytical tools offered by a particular software package (CrimeStat; Levine 2010) in ERW risk mapping. This study clearly puts in evidence the potential of density-based methods in landmine mapping. In terms of visualisation, few publications have been made to make visible the problem of contamination by landmines and its socio-economic impacts. If in-field operators and database analysts extensively use printed one-to-one dot maps on the level of field operations, few maps are available on the Internet (e.g. ICBL 2011b, ITF 2001, Lokey 2001, and Rekacewicz 2003). Most of them are hardly legible, not interactive and presumably not up-to-date, based on their dates of publication. The worldwide choropleth map of contamination by ICBL (2011b) shows four degrees of contamination at the country level: very heavy, heavy, low and none. Two websites (Sasi and Newman 2006, Hennig 2011) show cartograms displaying landmine casualties at the global level. UNMAS provides monthly updated information about UXO removal and mine risk education on an interactive web application (UNMAS 2012). This application opens promising perspectives for the dissemination of geospatial mine action information on a global geoportal, as many organisations involved e.g. in humanitarian assistance and disaster and risk management have already provided information specific to their field of activity (see for instance ESA 2010, GEO 2011, Giuliani and Peduzzi 2011, and ICRC 2012).

10 E.g. WGS 84, UTM and national mapping systems

Chapter 1: Introduction

Likewise, few references are available on the use of GIS-based multi-criteria solutions in mine action. One good reference is Knezic and Mladineo (2006), who describe and investigate a new approach to priority setting within mine action, with the introduction of a hierarchy GIS-based Decision-Support System (DSS). Mladineo (2012) goes further by implementing in ArcGIS Server a multi-criteria application for setting priorities in humanitarian demining. This GIS-based web application allows users to combine layers at the local level through predefined scenarios. Even if it is demanding in terms of GIS skills and licence, the application clearly highlights the potential of GIS-based multi-criteria analysis in priority setting in the humanitarian demining field. A more specific state of the art on the use of cartographic and GIS methods and tools in mine action is provided in different parts of this thesis, as it compiles several papers that were submitted or published in scientific journals. For more details we invite the reader to refer to Sections 2.2, 3.4, 4.2 and 4.6.

Chapter 1: Introduction

1.6. Research problem, Research questions and Hypotheses

Recognizing the needs mentioned previously, we explore in this thesis to what extent GIS can contribute to efficient mine action and we build our argument on three research questions. Each of the three questions comes along with a set of working hypotheses. They will be confirmed or refuted in the conclusive section of this thesis (Section 5.1).

Q1. To what extent can GIS improve visualization of contamination and its impact on population? In other words: To which degree can GIS meet the needs of humanitarian demining users? Working hypothesis H1.1: A unique cartographic visualization method should be developed for mine action. Working hypothesis H1.2: Perhaps the IMSMA data patterns hold a certain degree of clustering. Working hypothesis H1.3: There is a conceivable risk of over- or under-estimating the contamination in some regions. Working hypothesis H1.4: As it is the case in environmental sciences (e.g. disasters management, flood vulnerability assessment) GIS make it possible to assess population vulnerability to landmines.

Q2. What are the contributions and limits of GIS for improving decision-making in mine action? Working hypothesis H2.1: Decision-making in mine action is very specific to this topic. Working hypothesis H2.2: Like many communities of users, the mine action community needs geospatial applications, and these applications should be simple. Working hypothesis H2.3: Perhaps environmental, geographic and socio-economic conditions have substantial influence on demining activities. Working hypothesis H2.4: The mine action community is ready to integrate GIS-based decision support tools in their everyday work.

Q3. How to best build GIS capacity in mine action? In other words: Which plans and strategy can be proposed to help mine action users take the full step to GIS? Working hypothesis H3.1: Mine action experts lack key structuring elements to improve their use of GIS. Working hypothesis H3.2: Maybe data access and sharing is too limited for efficient implementation of GIS. Working hypothesis H3.3: Collaboration among key actors is likely to be a limiting factor.

Since only few scientific publications have been made on these subjects in the humanitarian demining

Chapter 1: Introduction

field, we assume that our work is likely to bring out new light to the mine action community. More widely, we put effort on developing methodologies and tools that could bring about novel contributions to the broader scientific community.

Chapter 1: Introduction

1.7. Methodology

With the goal of expanding research and development, the GICHD has since 2010 significantly strengthened its collaborations with academic and non-academic institutions involved in GIS, including Esri Inc., Esri Switzerland, European Space Agency, Geneva University, GISCorps, James Madison University, Kansas University, Massachusetts Institute of Technology, Tilburg University, United Nations Institute for Training and Research (UNITAR) and Zürich University. Within this framework, the GICHD and the University of Geneva signed in 2010 a memorandum of understanding, where the Geneva University commits itself to support GICHD activities dealing with GIS. The starting point of this thesis was a terms of reference with a series of deliverables, including (1) the development of a Server for Explosive Remnants of War Information Systems (SERWIS) designed to serve maps for showing ERW contamination and determining its impact on populations, and (2) the development of an E-learning solution teaching the basics of GIS. These projects were followed by a second phase in the thesis aiming to design user-oriented models, tools and cartographic items. A third phase was more directed towards the synthesis of all aspects of the research and the communication of the projects to the mine action community and beyond. It included the writing and publication of the results, and description of models and tools. In terms of cartography, one key aspect is that projects were directed towards mapping rather than predictive models like the one introduced by Lehmann et al. (2003): this thesis is about mapping and visualising the recorded reality on the ground, and not about predicting the location of landmines. Another important aspect of this research is that all models were designed to be multi-scalable, as will be discussed in more detail in Chapter 3 of this thesis. They should all be usable at least on two different scales between the global and the local level. Finally, effort was put on developing tools for bridging the IMSMA server with ArcGIS, in order to spatialize the IMSMA data thus to open them to more complex geospatial models. This section will explain the development process of the models and discusses the data that was used in the framework of this thesis.

1.7.1. Development cycles

This research was set with a short time line of three years coupled with the need to create products directly usable by mine action professionals. Recognizing the strong needs for GIS mentioned previously, the main thread of the development was a wide exploration of different topics applied to the context of mine action, based on geospatial data: cartography, semiology, data clustering, GIS analysis, multi- criteria analysis, E-learning, and SDI. The length of the different projects was variable. Some were conducted on a long time period, more or less continuously, whereas two others were irregular designed to explore a specific domain of GIS (See

Chapter 1: Introduction

Sections 2.4 and 3.3). These two one-shot projects only show the promise of using GIS in the mine action field and deserve deeper analysis, as will be discussed further. An important aspect of the project was the interaction with users, as will be described and discussed in detail later in this thesis. No formal interviews were conducted but a constant interaction with mine action professionals was mediated. Three main categories of humanitarian demining actors helped the development of tools: (1) the GICHD Information Management section, (2) other sections of the GICHD including Strategic Management and Operations, and (3) mine action professionals in affected countries. The contact with the two first groups, especially the first one, was mediated through regular meetings at the GICHD. Representative of the third group were met during focus group meetings and workshops in Geneva, an online survey, regular contacts by E-mail and Skype, and a field trip in Albania. Beyond the use of GIS, the demining personnel in Albania also helped understanding the more general problems of landmines and ERW contamination (victims of ERW and medical personnel were also met during a field trip). However, the discussions with the Albanian demining authorities did not have effect on the development of the projects, but they influenced the redaction and presentations of the projects. Interaction with humanitarian demining actors during the projects was not always the same as the involvement of the different categories of professionals differed. Most projects involved the Information Management section of the GICHD and to a less extent other sections. The participative nature of these projects helped to highlight the cross-section interest of GIS for mine action, even if still much work needs to be achieved in this area. The focus group meetings and workshops also involved users in mine- affected countries. Finally, the project for developing bridging tools was conducted in coordination with the Information Management section along with workshops. The participation of end-users involved in actual demining activities helped understand at best the potential of these tools for real operations. Two projects, on web mapping and using data clustering for mapping ERW contamination were conducted separately from the GICHD, at the Geneva University, and only results were presented to the mine action community. The overall research was user-oriented and resulted in the prototyping of technological solutions. This responded to the concrete needs of humanitarian demining actors. As the users do not have many resources they needed simple, user-friendly and operational tools. An assumption at the beginning was that the needs were directed towards the most complex tools resulted false. In fact, in most cases it was the simple tools that were the most demanded.

1.7.2. Data

The data used for this thesis comes from different sources depending on the needs of the research and their availability. The data is of three different types: core mine action data, auxiliary data and self made data. All this data was used for the development of GIS models and tools and for its use by the mine action community as demonstration data.

Chapter 1: Introduction

Core mine action data comes from IMSMA databases and consists of tabular information with geographical coordinates. The fact that this study was done at the GICHD gave access to dozens of mine action databases. However, depending on the country and its confidentiality policy, data has constraints in the way it can be used: (1) some can only be used on GICHD computers, and nothing can be shown of them; (2) some can be shown as density raster data, but not the original vector data and their exact location; (3) some can be shown entirely, but can not be published; and (4) for some countries, there is no constraint. The core mine action data on which this research bases come from seven different countries, which shows out of sixty countries: Afghanistan, Colombia, Cyprus, Iraq, Lebanon, South Central Somalia and Tajikistan. This is about ten percent of all mine-affected countries in the world. This also includes both small and large countries as well as high and low contamination degrees are represented. For these reasons we can consider it as representative of the dozens of IMSMA repositories used worldwide. Note that during the focus group workshops, some national mine action programmes represented had the occasion to test the new tools with data from their own IMSMA repository. Auxiliary data coupled with core mine action data is also used for GIS analysis. Finding geospatial data that responds to a series of characteristics making it suitable for the analysis was an important research work. The two key criteria were that auxiliary data (1) must be easy to access (e.g. on an Internet platform for free) and (2) have a good resolution (e.g. usable at sub-national level). All used data is free and publishable, which means that an analysis can be made public, provided sources are cited. The data itself needed to be the best available quality data and it is presented in a list (to be exact, several lists) of online databases to potential users. This does not mean that the data is necessarily readily usable for applications in mine action. It may, in fact, need preparation work before interpolation, as will be discussed later. The geospatial data that was used and catalogued in the process includes raster and vector datasets, such as digital elevation models (DEM), population, land use, soil characteristics and environment data. The availability of other types of datasets was also investigated, including for example scent contamination and data on dangerous animal presence. However, those appear not to be available, and were equally considered too specific for a GIS analysis for mine action. Another important point in the selection of auxiliary data was the completeness of the database in relation to mine affected countries. It appeared that in all the selected datasets, the coverage included more than 98% of the total area of the mine-affected countries. To develop our models, auxiliary data with different extents were used: Afghanistan, Cyprus, three provinces in Colombia, and Mozambique. For the needs of the project entitled ―Improving Quality of Web Services‖, random screen sized requests were performed, on commonly used scales and extents corresponding to the bounding box of various countries. A set of fictive data was also used for testing models and algorithms. Fictive data can be divided into two categories. One is made of purely geometric shapes, without geographical position and other fictive data is made of real geographic data within false political boundaries, such as ―Sandland‖, an invented political entity located in the Central Sweden. The former is designed to test algorithms, the latter to test realistic mine action scenarios. Data on non-existent country is used for political reasons, allowing

Chapter 1: Introduction

showing publicly the whole range of possible GIS functions on a territory, without pointing to an existing country. This solution reduces the risk to offend citizens from affected countries. Fictive data encompasses different types of geographical data. Its range is wide enough to be used for many different GIS operations, including the virtual campus designed to help mine action professionals learning ArcGIS. This data comprises geographical data, but also tabular, such as Excel files and an SQL dump used to test a tool. The overall characteristics of the data include notably (1) small and big datasets, (2) high and low level of clustering, (3) very different characteristics in the geography of the studied countries, (4) countries with different degrees of technical advancement in demining, and (5) great heterogeneity on the different IMSMA databases. This data was essential to the development of the projects, and their relevance to each type of test on the models and tools was carefully examined and calibrated. The complex and lengthy process of collecting and creating these datasets was worthwhile both for allowing testing models and to present possibilities of using auxiliary data to mine action professionals. An in-depth description of the three categories of data mentioned above will be provided later in this thesis.

1.8. Overview of the different research projects

The research presented in this thesis is supported by ten GIS projects. Before the reader goes further through this monograph, we see as very important to provide him/her with a detailed description of all these projects. This is the purpose of this Section 1.8, which is extracted from an article entitled ―How can GIS benefit demining activity?‖ (Lacroix and Eriksson 2011). At the end of this section, we also relate each project to the research question(s) that it might contribute to address.

1.8.1. SERWIS: Sharing data, maps, technologies and processes

The keystone of our research is the Server for Explosive Remnants of War Information Systems (SERWIS). It was set up to show the impact of ERW in contaminated countries through large-scale maps, without showing the ERW‘s exact location. For the mine action community it is very important that maps that are shown online do not give the exact location due the high data sensitivity. Exact locations could be used by civilians or criminals to either attempt to navigate through contaminated areas or to steal landmines to sell the explosives on the black market. During 2010, based on data from several test countries, the whole process from obfuscating vector data (by density raster calculation) to designing Web map services on ArcGIS Server was implemented. A specification on how the obfuscation may be applied on a national level and how it may be integrated into IMSMANG was completed and developed. The SERWIS project also aims at developing GIS tools and methods to identify where populations are mostly at risk as well as to better understand the operational difficulties of mine action. A comparative

Chapter 1: Introduction

analysis has been made among density rasters and global databases (e.g., population, slope, hydrology, land use, soils, climate, etc.) that could be used as input in an analysis to determine in which order to clear contaminated areas and with which tools. More generally, the development of the SERWIS platform will make it easier for users to access and share maps and GIS data, methods and tools in relation with demining activities. In addition to the SERWIS project, customised and user-friendly ArcGIS tools have been developed to facilitate the process of mine-action users to adopt GIS analysis tools outside IMSMANG. This includes (1) the development of GIS models for multi-criteria analysis and transportation network analysis, (2) the development of an ArcGIS extension for optimising data preparation and map design, (3) the publishing of a mine-action-oriented Esri Virtual Campus course, ―GIS for Humanitarian Mine Action‖, to teach the basics of ArcGIS Desktop to end users and (4) maintenance of a collection of cartographic symbols for ArcGIS Desktop.

1.8.2. Visualising contamination

The original data are available as points or polygons in IMSMANG. For reasons of confidentiality, sensitivity and visibility, density rasters are calculated from the vector data, country by country, using an ArcGIS interpolator. ArcGIS provides useful interpolators, the most pertinent being the Kernel function (see Sections 2.2 and 2.3). This estimator is used in many applications (e.g., crime analysis, medical sciences, urban networks, etc.). Each output pixel receives the sum of the contribution of the input pixels located within a given search radius, also called ―Kernel radius‖. The input pixels are weighted by an ―applicative‖ field (the area of the hazard), but also by the inverse distance to the output pixel. The Kernel method has a lot of advantages, including the ability to process heterogeneous data and provide short processing times—a few minutes of calculation for 5,000 hazards, upon an area of 1 million square kilometres with a 20-kilometer Kernel search radius and a 200-meter output pixel size. The interpolation results in a smoothed density map (grid) displaying the hazard contamination, while not showing all details. Red values (see Figure 2) correspond to a high probability of contamination and yellow values to a low probability. There are no known hazards in white areas. An interim tool was developed based on ArcGIS Engine to automate the process of creating and representing rasters from data in IMSMA or other MySQL databases.

1.8.3. Determining the impacts on human population

Showing hazards and field activity does not provide a complete vision of humanitarian mine action, so it would be interesting to combine and compare those rasters with global population datasets to determine in which areas populations are most at risk of ERW (see Section 2.4).

Chapter 1: Introduction

1.8.4. Choosing the right technique

An ongoing research process explores the possibility of determining nominal degrees of difficulty of demining (low, medium, high, and extreme) by combining datasets for population, infrastructure, hydrology, climate, elevation, slope, etc. Promising results came out and the first results are available (see Section 3.2). Other datasets are being evaluated for their potential use. This includes soil characteristic such as cation exchange capacity, texture, bulk density and available water-storage capacity.

1.8.5. Calculating the Shortest Path

The possibility of using transportation network analysis GIS tools in the humanitarian demining framework is under study. A methodology has been developed and a case study has been conducted for determining suitable location for building a new medical facility to improve medical care to mine victims (see Section 3.3). The model is based on minimisation of travel time calculation between origins (e.g. accident locations) and destinations (e.g. facilities). Other possible scenarios are being investigated, among which minimisation of travel distance and simulation of the effect of road disruption.

1.8.6. Setting priorities

In parallel, MASCOT (Multi-criteria Analytical SCOring Tool), a spatial tool is developed (see Section 3.4). This decision-help tool, targeting national or sub-national scales, aims to calculate scores for hazards with regard to their environment. The score is calculated depending on the hazard‘s location relative to e.g. population, hospitals or areas with industrial development. ―What I find appealing about the tool is the flexibility of factors that can be inputted in the form of shapefiles,‖ says Penelope Casswell, field programme and GIS manager, Action on Armed Violence. ―We hold a lot of geographical data in Western Sahara on the location of hazardous areas, accidents, water points and routes, so this tool would be ideal to incorporate all this data (in a few simple steps) to generate a high, medium or low priority. With regard to the scoring tool, it would be a great help for prioritising the cluster strikes and minefields in Western Sahara,‖ adds Casswell. ―Currently, we use our own criteria system based on proximity to water points, population centres, and primary and secondary routes. It would be useful to have a scoring tool to ensure consistency.‖ MASCOT allows the users to choose their own parameters:  Selection of features (e.g. hazards) to be scored.  Choice of criteria (e.g. roads, hospitals, markets).  Choice of calculation distance for each criterion (e.g. 300 meters for roads, 350 meters for hospitals, 250 meters for markets).  Weighting for each criteria (e.g. two points for roads, three points for hospitals, two points for markets).

Chapter 1: Introduction

Each hazard then is scored as follows: if two roads are nearer than 300 meters, and one hospital is nearer than 350 meters from the hazard, then the score is equal to: (2 x 2) + (1 x 3) = 7 points. The higher the score, the higher is the priority of clearance in that area. MASCOT can be downloaded for free at http://www.unige.ch/sig/outils/MASCOT.html

1.8.7. Optimising GIS workflows

An increasing number of mine-affected countries are using ArcGIS Desktop, but the users have limited training, so two customised, user-friendly ArcMap tools are under development. In particular, START (Simplified Toolbar to Accelerate Repeated Tasks) is developed (see Section 4.2) and includes (1) ArcMap MXT templates, (2) a user guide and contextual help, (3) standard tools for printouts and layouts, gathered in the same toolbar and (4) customised tools, including direct connections to the MySQL databases, creation of closed polygons from a list of XY points in Universal Transverse Mercator (UTM) or latitude/longitude, projection management, vector and raster clipping, etc.. START can be downloaded for free at http://www.unige.ch/sig/outils/StartToolbar.html ―START is a central resource for key spatial functions used in mine action that dramatically simplifies access to the most common geospatial needs in the community,‖ notes Noah Klemm, director, International and Homeland Security Programmes, FGM Inc. 11 ―The toolbar provides field users with limited ArcGIS experience to the most-useful GIS functions in the ArcGIS desktop products.‖

1.8.8. E-learning

For the course mine-action-oriented datasets were created, and theoretical lessons with practical exercises were written (see Section 4.3). The course is composed of the following chapters: (1) exploring data and using ArcCatalog, (2) Learning the basics of satellite imagery, (3) loading and using standard or mine- action-oriented symbol libraries, (4) designing a map and layout, (5) managing projections in World Geodetic System 1984 (WGS 84) or local Universal Transverse Mercator (UTM), (6) digitising data in a georeferenced map, (7) using vector data in ArcMap, and (8) using raster data in ArcMap. The course is free and can be completed at http://training.esri.com/gateway/index.cfm?fa=catalog.webCourseDetail&CourseID=2065

1.8.9. Standardising symbology

The ―Cartographic Recommendations for Humanitarian Demining Map Symbols in the Information Management System for Mine Action‖ were developed by GICHD and the University of Kansas in 2005. Most of the existing cartographic symbols are based on ISO standards.

11 Company in charge of developing the IMSMA software

Chapter 1: Introduction

However, some of them do not yet exist, such as in the Land Release domain, and others should integrate new methodologies and technologies developed in recent years. As a consequence, recommendations have been revised and updated. A new symbology was proposed, especially for hazards, hazard reductions and tasks (see Section 4.4). ―Development of the new IMSMA symbology set, which fully encompasses all methods of land release, is an important step forward for [information management] and its application to mine action, and it will be a great advantage for operational planning‖ notes Michael Creighton, land-release programme manager, GICHD. These changes are planned to be integrated into the standard IMSMA style-font, available in ArcGIS Desktop. Before that, they will be validated by the mine action community.

1.8.10. Improving the Quality of data Web Services

To ensure that web services published on SERWIS will be sufficiently responsive to fulfill users‘ needs and requirements, performance of a given service must be measured and monitored to track latencies, bottlenecks and errors that may negatively influence its overall quality. Consequently it is required to develop frameworks to assess the usability and performance of download services through a set of quantitative measurable criteria that allow quantification, repeatability, comparability and understandability of results. Based on these considerations the aims of this research project are (1) to present an open approach to measure the performance of different vector (WFS) and raster (WCS) services provided by two widely used software implementations, including ArcGIS, and (2) to provide some guidance to data providers aiming at improving the quality of their services.

1.8.11. Summary

The GIS projects described above are illustrated in Figure 2. Each of them contributes to answering one or several of the three research questions of this thesis: (1) To what extent can GIS improve visualisation of contamination and its impact on population? (2) What are the contributions and limits of GIS for improving decision-making in mine action? and (3) How to best build GIS capacity in mine action? Figure 2 is extracted from a poster that was designed for the GIS for the UN conference held in Geneva in April 2012 (Lacroix 2012).

Chapter 1: Introduction

Figure 2: Contributions of GIS to efficient mine action. Overview of the three research areas of this PhD thesis and the GIS research projects that were conducted

Chapter 2. To what extent can GIS improve visualisation of contamination and its impact on population?

Contributing research papers  Lacroix P., Herzog J., Eriksson D., Weibel R. (2013). Methods for Visualising the Explosive Remnants of War. Applied Geography, 41:179-194. Available from: http://www.sciencedirect.com/science/article/pii/S0143622813001021  Lacroix P., Herzog J., Eriksson D. (2011). Mapping Populations at Risk of ERW. The Journal of ERW and Mine Action, 15(1). Available from: http://maic.jmu.edu/journal/15.2/specialrpt/lacroix/lacroix.htm

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.1. Introduction

In Chapter 2 of this thesis, we explore to what extent GIS can address the needs for maps in the mine action community, with a focus on visualising contamination by ERW and their impacts on populations. On this account, we demonstrate the capacity of GIS to provide users inside and outside the core mine action community with novel, realistic and comprehensive ways of making the ERW contamination problem visible. We also put in evidence a number of difficulties, requirements and constraints that have to be faced. The first difficulty is that few papers have been published on the use of visualisation methods in mine action – a state of the art on the topic has been presented in Section 1.5. A second difficulty is that mine action data are critically heterogeneous in type, degree of completeness, quality, positional accuracy and spatial distribution pattern. Moreover, the amount of data varies significantly across different countries. For example, more than 10‘000 polygon features are stored in the Laos IMSMANG repository while 8 have been recorded in Western Sahara. Another constraint relates to users‘ needs, which include a wide range of scales, from the global to the local level. Finally, users‘ GIS skills are often limited. Consequently, the message conveyed by maps should be intuitive in order to avoid misunderstandings. This constraint also leads us to focus on automation of visualisation workflows and to develop and recommend customized and user-friendly cartographic functions. The work presented in this chapter of the thesis also raises some challenges:  There are thousands of IMSMA users worldwide. More and more of them are taking the full step to GIS by using the Esri suite in connection to the IMSMANG data server. Therefore, our work should somehow consider linking these two technologies.  Requirements for visualising ERW contamination are sometimes contradictory. On the one hand, users need maps that are precise enough to show the contamination in their country. On the other hand, data confidentiality should be respected, for several reasons explained in Section 1.4.2.1.1 (e.g. respect of the sovereignty of countries, protection of civilians). Thus one challenge is to find a good compromise between providing close-to-reality representations of data and respecting non-disclosure policy of some programmes. Another issue relates to the visualisation of ERW contamination along country borders: the method chosen by one country to represent its contamination should not affect the representation in neighbouring countries. To add complexity, some country borders are disputed and cannot be shown on maps for reasons of neutrality. Since landmines are more commonly located in such contested areas, the visualisation process has to be done carefully.  Users need to be able to control and adjust the representation of the contamination. For instance, the users operating on a global level have different needs from those on national or local levels. If we decide to obfuscate the ERW data e.g. through kernel density maps, the purpose of the use of such maps would be different across national programmes. So would be the balance between

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

precise maps and obfuscation. These differences also vary between countries, which makes it impossible to propose standard solutions for each level. Consequently, the methods that need to be developed should be flexible enough to let users decide which degree of obfuscation and which level of detail might address their needs.  The risk with maps at supra-national level is to under- or over-estimate the picture of contamination in some countries. With the goal of remaining neutral, we should not emphasise or neglect any country. In particular, the choice of colours for representing contamination should be done carefully. Enlightening mine-affected areas with flashy colours (e.g. red) may attract the attention of potential donors and encourage them to provide funds. On the other side, it might also show them that the problem of contamination is not fixed yet and thus discourage them to continue funding. This chapter of the thesis is structured as follows. Section 2.2 is based on a paper entitled ―Methods for Visualising the Explosive Remnants of War‖ that was published in Applied Geography. In this paper, we investigate several visualisation methods, discuss their suitability to different categories of mine action users at different scales (ranging from global to local scale) and provide them to users with recommendations. If all of these cartographic visualization methods are established ones, two of them were extended from cartographic techniques based on Kernel Density Estimation (KDE) and were customised to users‘ needs within the framework of this PhD. One of them is also novel in the sense that it fills up polygons with points before application of KDE extension. In Section 2.3, we develop a novel mapping technique consisting of clustering the IMSMANG data before applying interpolation functions. The three KDE-based visualisation methods mentioned above generate single-layer maps. Consequently, we explore in Section 2.4 the feasibility of combining some of them with population-density data, for visualisation of at-risk populations. This study was published under the title ―Mapping populations at risk of ERW‖ in the Journal of ERW and Mine Action. Finally, the key ideas of this chapter are highlighted in Section 2.5 to give readers a short summary.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.2. Visualising Contamination

Based on: Methods for Visualising the Explosive Remnants of War

Pierre Lacroixa,b,f, Jonas Herzogc, Daniel Erikssond, Robert Weibele a University of Geneva, Institute for Environmental Sciences, enviroSPACE Lab., Battelle – Building D, 7 route de Drize, CH-1227 Carouge; b University of Geneva, Forel Institute, 10 route de Suisse, CH-1290 Versoix; c Joint Mine Action Coordination Centre, 111 Palm City, Tripoli, Libya; d Geneva International Centre for Humanitarian Demining, 7bis, avenue de la Paix, P.O. Box 1300, CH-1211 Geneva 1; e University of Zurich, Department of Geography, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland; f United Nations Environment Programme, Division of Early Warning and Assessment, Global Resource Information Database – Geneva, International Environment House, 11 chemin des Anémones, CH-1219 Châtelaine

2.2.1. Abstract

This study aimed to answer the question how GIS can help decision makers visualize the problem of contamination by explosive remnants of war (ERW). We thus explored a set of six cartographic visualization methods and systematically evaluated their usefulness with respect to four categories of stakeholders in the humanitarian demining process (i.e., database administrators, operations officers, directors of national mine action authorities, and donors) at four geographical scales, ranging from municipal to global. The main application of our work is for stakeholders involved in humanitarian demining. We provide them with a comprehensive framework for visualizing ERW hazards at the geographical scale at which they have to make decisions, as well as customized cartographic visualization tools and recommendations to help them make informed decisions. For example, we provide potential donors with a method for obtaining a global overview of ERW contamination while remaining aware of regional variation and hot spots. We also enhance cartographic visualization capabilities using traditional kernel density estimation by customizing key parameters. Specifically, we propose a method for adjusting kernel bandwidth for datasets with highly heterogeneous spatial distributions and a method for generating kernel surfaces from polygon data that consists of infilling the polygons with points before using them as inputs in the kernel density estimation.

Keywords : Humanitarian Mine Action; Explosive Remnants of War (ERW); Cartographic Visualisation; Kernel Density Estimation; Data Sharing

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.2.2. Introduction

The term Explosive Remnants of War (ERW) has its origins in international humanitarian law12. ERW includes all explosive contamination from war, such as landmines, unexploded ordnance (UXO), improvised explosive devices, and abandoned munitions storage. According to the United Nations Mine Action Service (UNMAS), ERW affect over 70 countries, and thousands of casualties are recorded each year (UNMAS 2011). In Afghanistan, in 2000 and 2001, 81% of those injured by ERW were civilians, and 46% were 16 years old or younger (Bilukha et al. 2003). The effects of ERW are not limited to the individuals who are directly impacted. They represent a significant challenge for society in affected countries. Situated at the crossroads of humanitarian and development activities, mine action aims both to reduce the impacts of the presence of ERW on local populations and to return cleared land to local communities for land rehabilitation. This paper aims to answer the question of how visualization of ERW contamination can support mine action decision-makers at the policy level, e.g., in determining national and international mine action priorities, assessing humanitarian impact, or estimating the financial costs of reducing ERW impacts. Building on the work of Lacroix et al. (2002) and Delhay et al. (2005), who suggested that maps improve the planning of demining campaigns, this paper investigates how maps can be used to help decision makers visualize the problem of contamination by ERW. This paper makes several methodological contributions intended for the abovementioned end users. We conducted an analysis of the user requirements for visualizing ERW contamination. We proposed six different cartographic visualization methods that can be used to display contamination by ERW hazards. For cartographic visualization using kernel density estimation (KDE), we proposed two extensions of the existing method: first, a simple yet effective technique of adjusting kernel bandwidth for datasets with highly heterogeneous spatial distributions; and second, a method for generating a KDE from polygon data by infilling polygons with data points before using them as inputs to the KDE. Finally, our comparative evaluation of the strengths and weaknesses of the six proposed methods provides overall recommendations for the use of each cartographic visualization technique by the four different categories of mine action stakeholders at four different scales. In this study, we examined cartographic visualization methods using existing contamination data managed by a set of national authorities with the Information Management System for Mine Action (IMSMA), which was developed by the Geneva International Centre for Humanitarian Demining (GICHD). The GICHD is a non-profit foundation established by Switzerland and several other countries in April 1998. It strives to eliminate ERW and to reduce its humanitarian impact. The GICHD develops

12 Protocol V to the Convention on Certain Conventional Weapons, , adopted in November 2003 by the Meeting of the State Parties to the Convention, defines explosive remnants of war (ERW) as unexploded ordnance (UXO) and abandoned explosive ordnance. UXO (also known as ―duds‖) ―refers to munitions (bombs, shells, mortars, grenades and the like) that have been used but which have failed to detonate as intended, usually on impact with the ground or other hard surface‖ (GICHD, 2010, p. 13)

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

standards, provides capacity-development support, and conducts research activities. The actual implementation of mine action activities is not within its scope of work. Instead, the mission of the organization is to promote tools and methods that improve the performance of mine action actors such as international organizations and the governments of affected countries. In this role, the GICHD often represents the mine action community in interactions with researchers and the academic community. In our project, the involvement of the GICHD was central to establishing contact with the community of users that would apply the methods developed in the study and in building an understanding of the needs of that community. The emphasis of this paper is on identifying and developing methods for displaying and conveying ERW contamination data using the existing data from IMSMA installations in a set of selected countries. Consequently, methods for detecting contamination or collecting data on potential unknown contamination, such as remote sensing, are excluded from the paper. Such methods are well addressed in the work of other authors, e.g., Zare et al. (2008) and Witmer and O‘Loughlin (2009). The paper is organized as follows. We begin in Section 2.2.3 by orienting the reader to the terminology and standard processes used in mine action work and by determining the requirements of contamination visualization for the four identified stakeholder groups, followed by a brief description of the experimental setup in Section 2.2.4, including the selected case study countries and their respective datasets. In Section 2.2.5, we review existing research related to the visualization of ERW contamination data. This is followed in Section 2.2.6 by the introduction of the six selected cartographic visualization methods that were tested according to the requirements set out in 2.2.3. Sections 2.2.7 and 2.2.8 elaborate on the technical elements of the cartographic visualization methods that use KDE. Section 2.2.9 provides a comparative discussion of the proposed cartographic visualization methods with respect to the criteria identified in 2.2.3. Finally, the paper ends with our conclusions and a discussion of the use of the proposed cartographic visualization methods by mine action stakeholders.

2.2.3. Background

2.2.3.1. User Focus Group

A user focus group was set up with the support of the GICHD. The focus group consisted of two dozen individuals, including mine action experts, GIS experts or both. The results of a self-assessment of the participants‘ individual areas of expertise demonstrated that the focus group broadly represented the entire user community13. The participants, who represented eight different organizations, had intervened directly or through partnerships in more than 60 mine-affected countries. The input from these experts was supplemented by input from GICHD staff, who are in frequent contact with decision makers as part

13 Information management: 11 experts; strategic management: 1 expert; national program management: 4 experts ; operational planning: 3 experts; database management: 6 experts; and GIS: 8 experts

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

of their work and are responsible for the development of the GIS component of the IMSMA, which is the most widely used GIS tool in the field (Eriksson 2011). With this in mind, we considered the panel of experts to be representative of the mine action community and an appropriate sample for our research. Two main meetings were held with the user focus group: one was held in April 2011 and the other in February 2012. The first meeting led to the formulation of requirements for ERW visualization. Following the second meeting, the end user experts were also asked to evaluate the results of our research, as discussed in Section 2.2.9.

2.2.3.2. Mine Action Stakeholders

Based on our review of the mine action literature and the discussions in the first user focus group meeting, four groups of actors were identified. Users outside the core mine action field include donors (public and private organizations and individuals) and the general public, which is considered a potential donor. These actors need a global overview of mine action to decide which country to fund. In 2009, 83% of funding for mine action came from international sources (Devlin and Naidoo 2010). Donors no longer consider mine action to be an immediate humanitarian response; rather, it is considered to be part of a broader process that includes conflict prevention, protection, mitigating socio-economic impacts, reintegration (Devlin 2010), humanitarian assistance, and care for survivors (Devlin and Naidoo 2010). Directors of national mine action authorities are responsible for ensuring that mine action activities in their respective countries are implemented in compliance with international law, standards, and policies (GICHD 2007, GICHD and UNMAS 2011). These authorities work in collaboration with other national and international bodies, governments, communities, private companies, and initiatives. They regularly produce summaries of their goals and achievements for distribution to donors and the broader mine action community (UNDP 2011). The operations officers in a mine action authority are part of a small- to large-scale prioritization process. They first must refer to contamination maps at the regional or sub-regional scales to decide where to conduct mine action activities. In a second step, large-scale one-to-one dot maps (also called point symbol maps; cf. Section 2.2.6.1) combined with other elements (e.g., topography, key agricultural land, and infrastructure, as described by Gasser et al. (2011)) are used to decide how to access mined areas. Operations officers are experts in mine action and explosive ordnance disposal and do not necessarily have GIS expertise or experience. Similarly, database administrators do not necessarily have high competency in GIS. Part of their work consists of probing the database for inaccuracy or incompleteness. Spatial data attributes such as coordinates, area type, and area are commonly checked at large scales (1:50‘000 to 5‘000) in coordinated efforts between database administrators and operations.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.2.3.3. Mine Action Data

Mine action data are collected in the field and recorded in the IMSMA relational database management system. IMSMA is an ArcGIS Engine-enabled, self-contained information system that was specifically developed for mine action centers in mine-affected countries (Eriksson 2011). As of 2012, more than 60 centers are using this software for information management. Each country using the system owns their data. IMSMANG is therefore not a global repository of mine action data. National databases are accessible to the GICHD on request for technical support purposes, but national authorities decide whether to share their data with other mine action stakeholders. Most stakeholders struggle to get insight into the contamination data. In particular, donors do not have the access they need to complete, global-scale information about contamination problems. Because each mine-affected country is responsible for collecting data within its territory, the IMSMA allows for customized methods of data collection and entry. The freedom of IMSMA users to choose the data that are collected and the data formats results in very heterogeneous data across the different countries. Nevertheless, this design choice was made to make the system adaptable to the diverse types of information management required in the field of mine action. Furthermore, the capacity of each country to control the quality of the incoming data varies, resulting in a large spread in data quality. The reliability or confidence level of a given record is most commonly indicated by a designated ‗hazard type,‘ which the user assigns to the data either when collecting the data in the field or when entering it into the database. The hazard types defined by UNMAS (2003) are suspected hazardous area (SHA), confirmed hazardous area (CHA), and defined hazardous area (DHA). The definitions of these types are under constant review; for the purpose of this study, we used the UNMAS 2003 definition. A SHA is an area suspected of containing a hazard. An SHA is often identified when a local population reports that a hazard is present, and SHAs typically do not have a known perimeter. Instead, an SHA is represented by point indicating the approximate location or, more rarely, a circle. A CHA is an area identified in a non-technical survey for which the need has been confirmed for further intervention, either through a technical survey or clearance activities. A CHA is typically represented by a polygon; compared to an SHA, a lot more information is available about the area. For instance, information about the suspected type of contamination, potential economic use of the area after clearance, topography, and vegetation is often included. A DHA refers to an area within a CHA that requires full clearance. A DHA is normally identified using original and reliable minefield records. In a typical demining process, the status of a hazardous area evolves from SHA to CHA and then to DHA. The reliability of hazardous areas is considered high for DHA, medium for CHA and low for SHA. In IMSMANG, spatial data are stored as 2-D coordinate pairs that represent polygon vertices, polyline vertices, or points, with an additional attribute recording an estimated or calculated area. In general, the

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

more that is known about a hazard, the more likely it is that the hazard has been recorded as a polygon (Figure 3). Therefore, SHA are most commonly stored as a single point approximation, sometimes with an estimated area as an attribute. The exact extent of the hazardous area remains unknown until after the hazard has been cleared. Most countries' databases contain many more SHA than DHA records.

Figure 3: A DHA stored as a polygon in the IMSMANG MySQL database 1Landmark or reference point: a fixed point of reference located some distance outside the DHA. Landmarks should be easily recognizable features, such as a road junctions or bridges, that can be used to navigate to one or more benchmarks 2Benchmark: a fixed point of reference used to locate a marked area or DHA. Benchmarks are typically located at a short distance outside the DHA 3Start point: the first point of a polygon 4Turning point: a fixed point on the ground that indicates a change in direction along the perimeter of a DHA 5Intermediate point: a point used between turning points that are more than 50 meters apart 6End point: last turning point (typically, same coordinates as start point)

In summary, the degree of completeness of a database and the hazard confidence levels vary significantly among countries. Most contamination data are sensitive. The exact location of mines cannot be revealed to a large audience because civilians or criminals could use this information to steal mines for illicit reuse or to sell on the black market as explosives.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.2.3.4. Requirements of Visualising ERW Hazards

The results of the first user focus group meeting are summarized below. The numerous requirements for the visualization of ERW hazards were sometimes contradictory and overlapping. Although the participants' responses identified a wide spectrum of user needs, groups of users tended to focus their responses at a particular scale (which ranged from global to local) and to agree on their requirements.  Maps should be precise and provide an accurate representation of the nature of the contamination.  At the same time, data confidentiality must be preserved. In terms of data representation, a compromise must be found between data obfuscation and the previous requirement for maps.  Users must be able to control and adjust the representation of the contamination data. For example, users operating at a global scale have different needs from those operating on the national or local level. The objective of using kernel density mapping and the balance between precise maps and data obfuscation vary across the different scales of use. These aspects of data visualization also vary among countries, which makes it impossible to use a standardized solution for each scale. Consequently, the methods must be developed that are flexible enough to let the users select the degree of obfuscation and level of detail that best addresses their needs,  Most mine action stakeholders are not GIS experts. The messages conveyed by the maps should therefore be intuitive to avoid misunderstandings.  It should be possible to combine contamination maps with other types of information (e.g., socio-economic data) to visualize relationships between contamination and other decision- making factors.  The sovereignty of countries should be respected. The method chosen by one country to visualize contamination should not affect the data representation used by neighboring countries.  Adding to the complexity, some country borders are disputed and, in the interest of maintaining neutrality, cannot be shown on maps. Because landmines are more commonly located in these contested areas, the cartographic visualization process must be performed carefully. This reinforces the need for users to be able to control the representation of contamination data on a case-by-case basis. Though secondary to the above requirements from users, it also became evident that the nature of the mine action data and the IMSMA system posed a set of technical challenges. These challenges included the large volume of data; data storage in non-spatial repositories; and a high degree of heterogeneity in the quality, spatial accuracy, reliability, degree of completeness, and spatial distribution of the ERW data.

2.2.4. Experimental Setup

Our test data included data subsets from six databases: Afghanistan, Cyprus, Iraq, Lebanon, South-

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Central Somalia, and Tajikistan. Because our intention was for the cartographic visualization methods analyzed in this paper to be used in any country affected by ERW, the countries were selected to be representative prototype countries based on their diversity in terms of environmental, historical, and cultural characteristics, as well as their predominant demining techniques. All of the selected countries currently use the IMSMA. The sample sizes (Table 1) ranged from less than 100 records for Cyprus to several thousand records for Afghanistan and included either points (and polygon centroids) or polygons. Spatial statistics revealed substantial differences in the geographical distributions of the ERW in the selected countries. In some countries, the data were more randomly dispersed in space (Herzog 2010), while the data clustering was more pronounced in other countries. For example, approximately 90% of the hazards in Cyprus are located in the vicinity of the sparsely populated buffer zone that has divided the Republic of Cyprus into two parts since 1974, following a conflict between the Turkish and Greek Cypriots. In contrast, more than 30% of the Afghan ERW was found in areas where the population density is higher than 100 inhabitants/km2, and 47% was no further than 1 km from a road. The use of heterogeneous data from prototypical countries enables a robust representation of worldwide contamination datasets.

Table 1: Overview of the test datasets Number of Number of Test country points1 polygons2

Afghanistan 6‘644 6‘443

Cyprus 94 89 Iraq 196 195 Lebanon 1‘496 1‘496 South Central 133 3 Somalia Tajikistan 202 1 1 Points and polygon centroids

2.2.5. Visualising Hazards and Mine Hazards: State of the Art

According to Lorz et al. (2010) and others, hazard mapping has become of significant interest for its applications in financial and environmental risk management. The mapping of hazards and disasters (e.g., floods, fires, earthquakes, cyclones, and volcanic eruptions) is widespread, especially in the form of maps or web map services (WMS) on interactive global data platforms (ESA 2011, GEO 2011, Giuliani and Peduzzi 2011, and many others). Some attempts have been made to apply remote sensing and GIS analysis in the field of mine action, mainly focusing on sensors that detect individual ERW through the use of satellite data (Witmer and O‘Loughlin 2009), hyperspectral imaging (Zare et al. 2008, Wong, 2009) or ground-penetrating radar

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

(Havens et al. 2009). Benini (2000) and Benini et al. (2003) coupled landmine data with socio-economic data at the national level to determine clearance priorities, and Riese et al. (2006) described a GIS-based approach to making probabilistic forecasts about the presence of ERW to support decision-making about the allocation of demining resources. In another probabilistic approach, Vistisen (2006) employed Bayesian inference to develop a risk model that quantified the extent to which a given minefield poses a risk to a society. Williams and Dunn (2003) presented an example of the use of GIS in a participatory process during an impact assessment of landmines in selected villages in Cambodia. Andersson and Mitchell (2006) used inverse-distance weighted interpolation to generate population-weighted raster maps for use in the evaluation of mine risk education. More recently, Alegría et al. (2011) used a variety of geostatistical techniques and kernel density estimation (KDE) to analyze and map landmine risk. While the study by Alegría et al. is the most similar to our study, it differs from our work in several important respects. The previous study was restricted to a single scale, rather than using a range of scales from local to global; the ERW data were point data, with no polygon data; and the study was essentially a preliminary study to explore the utility of various analytical tools offered by a particular software package (CrimeStat; Levine 2010) for visualizing ERW risk. Nevertheless, the previous study does clearly highlight the potential for using density-based methods for visualizing ERW, and the methods described are similar to some of the methods introduced below. Maps based on KDE (also called heat maps; Trame and Keßler 2011) are widely used to analyze and visualize spatial distributions of discrete presence or counts data given at point locations. In animal ecology, they are today the preferred methods for estimating and delineating the home range of animals, superior to convex hulls (Katajisto and Moilanen 2006, Wartmann et al. 2010). A selection of recent applications of KDE in geography include delineation of vernacular place names from web documents (Jones et al. 2008), mapping of concentrations of surnames in Britain (Cheshire and Longley 2012), delineation of city centres from topographic map data (Lüscher and Weibel 2013), mapping of environmental risks (Lewis and Bennett 2013), as well as mapping and visualizing social values associated with places (Brown and Weber, 2012, Sherrouse et al. 2011, Van Riper et al. 2012). With regards to the use of cartographic visualization in the field of mine action, only a few maps showing ERW contamination have been published on the web to date (e.g., ICBL 2011b, ITF 2001, Lokey 2001, Rekacewicz 2003). Most of them are difficult to read, are not interactive, and are most likely not up-to- date, based on their dates of publication. ICBL (2011b) provides one worldwide choropleth map that shows four degrees of mine contamination at the country level: very heavy, heavy, low, and none. Two websites (Sasi and Newman 2006, Hennig 2011) show cartograms displaying landmine casualties at the global level, but they do not show contamination by ERW. UNMAS provides monthly updated information about UXO removal and mine risk education in Libya using an interactive web platform (UNMAS 2012). While ERW visualization projects are rare on the web, printed one-to-one dot maps are extensively used by operations and database analysts at the field work level. Choropleth maps are currently in use in

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

several national mine action programs. In particular, the Colombian Programa de Acción Integral Contra las Minas Antipersonal (PAICMA 2012) has published a choropleth showing the number of victims per administrative unit.

2.2.6. Evaluated Visualisation Methods

There are multiple ways to represent a spatial phenomenon graphically. For ERW contamination, the difficulty is finding the appropriate cartographic visualization method because the requirements for visualizing ERW data are numerous. Cartographic textbooks such as Slocum et al. (2009) suggest a wide spectrum of cartographic methods, including point symbolization, choropleth maps and interpolation. Müller et al. (2006) recommend that hazard maps use continua, quantitative and absolute data, and smooth statistical surfaces. Six methods (A, B, C, D, E and F) were examined in this study. They are described in this section and illustrated in Figure 4.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 4: Contamination by ERW: overview of the six tested methods (Afghanistan, 6‘644 ERW)

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Our study should not be considered a comparative study of cartographic visualization methods because the methods in question are not meant to serve the same cartographic purpose. Rather, we aimed to identify the best methods for addressing the requirements of each separate user group.

2.2.6.1. Method A: One-to-one dot maps

In one-to-one dot maps (O‘Sullivan and Unwin 2010), also called point symbol maps (Slocum et al. 2009), each ERW is represented by a point symbol marker. As an example, the coordinates for the 6‘644 ERW areas in Afghanistan, which were originally stored in the IMSMANG repository, are directly displayed in Figure 4. Every marker represents a single point or polygon centroid. This method is based on one-to-one mapping and is an example of pure geovisualization (O‘Sullivan and Unwin 2010, p.65) (i.e., there are as many symbols as represented features).

2.2.6.2. Method B: Proportional symbols

In this method, a contamination value is estimated for each administrative unit by summing the estimated or calculated areas of the ERW located within the unit. Each administrative unit is represented by a symbol whose size corresponds to the level of contamination, with larger symbols indicating a higher level of contamination. Method B is useful for representing absolute count data, such as the number of ERW or the overall contaminated surface in an administrative unit.

2.2.6.3. Method C: Choropleth maps

A choropleth map is a thematic map that shows a generalized depiction of quantitative area distributions (Peterson 1979). In method C, ERW areas are aggregated by administrative unit and colored according to the contamination level, which is normalized by the area of the units. Choropleth mapping is currently the most widely used method for generating maps (O‘Sullivan and Unwin 2010).

2.2.6.4. Method D: KDE applied to points and polygon centroids

K DE has been used in many fields, including crime mapping (Chainey and Ratcliffe 2005), public health (Rushton 2003), home-range analysis (Seaman and Powell 1996), astronomy (Alard 2000), and landscape genetics (Epperson et al. 2010). In ERW applications, KDE can be used to interpolate contamination values in areas where there are no observed features. In commercial GIS platforms such as ArcGIS, the KDE function is only applicable to points and lines, not to polygons. Therefore, ERW areas that were extracted from IMSMANG in the form of polygons were replaced by polygon centroids. A smoothly tapered surface was then fitted over each ERW area, with the

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

highest value located at the point or centroid location. The value diminishes with increasing distance from the point/centroid, following a kernel function (e.g., in ArcGIS, the Epanechnikov kernel described in Equation 1 is applied) that reaches zero at a distance called ―bandwidth‖ (Figure 5). The selection of the bandwidth value and its influence on the rendering of the map is discussed below. Additionally, the points and centroids were weighted by their estimated or calculated area, which was stored in the MySQL database. The output was calculated by summing all kernel surfaces as follows: kde = S [1 – (r/h)2]2 where r < h (1)

0 where r ≥ h w here kde is the estimated density, r is the distance from the original point, h is the bandwidth and S is a scaling function equal to 15/16h for one dimension and to 3/h2 for two dimensions (Silverman 1986):

Figure 5: KDE of a multivariate point dataset

2.2.6.5. Method E: KDE applied to polygons

IMSMANG contains polygonal ERW data with variable shapes and areas. In particular, very long and narrow features (Figure 6) have been registered along country borders after conflicts and along mandatory crossing points such as roads. Method D reduces these polygons to their respective centroids before conducting the KDE procedure. To obtain a closer representation of reality, we propose method E, which involves (1) filling polygons with random points using the algorithm provided by ArcGIS and (2) conducting a KDE on the points. In this method, the KDE is weighted by the area of the polygons, which is calculated from their geometry

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

(coordinates of the vertices). Each random point is assigned a polygon area divided by the number of points in the polygon. The recording precision (i.e., the density of points scattered in the polygons) is called RP. Figure 6 shows the resulting KDE maps for methods D and E. In all KDE computations in our study, a cell size of 250 m was used for the KDE raster map. The cell size was chosen as a function of the target display scale, positional accuracy of the ERW data (accuracy at the meter to decameter scale was rare), requirements for obfuscation of the ERW location information, and practical requirements (Cf. Section 2.2.3.4).

Figure 6: Comparison between methods D and E, both applied to the same dataset along the Cypriot buffer zone, where long polygons are encountered

Due to the high degree of heterogeneity in the spatial distribution of the input datasets, methods D and E were customized. Kernel bandwidth value adjustments and the effect of recording precision (RP) are described in detail in Section 2.2.7 and Section 2.2.8.1, respectively. The effect of the procedure for scattering points in the polygons is evaluated in Section 2.2.8.1.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.2.6.6. Method F: Cartograms

Cartograms are becoming increasingly popular among cartographers and scientists (Dorling 1993, Duczmal et al. 2011, Reveiu 2011, Sun and Li 2010 and others). In method F, the contamination values of the ERW areas are summed for each administrative unit. The geometry of the map is distorted to display the unit areas according to their contamination values. Heavily affected regions are expanded, while other regions are reduced. The cartogram algorithm by Dougenik et al. (1985), which is also used in ArcGIS (―CartogramCreator‖), generates contiguous-area cartograms and preserves administrative neighborhoods, allowing for the administrative units to be compared.

2.2.7. Customising KDE-based methods (D and E): adjusting KDE bandwidth

Described by De Smith et al. (2007, p. 178) as ―often more of an art than a science,‖ bandwidth selection has a significant effect on the visualization and obfuscation of ERW (Herzog 2010). By choosing very small bandwidths, KDE maps tend to resemble to one-to-one dot maps, with small circles around ERW points. Not only are kernel peaks difficult to discern at small scales, but at large scales, the location of the ERW can be inferred to be the centre of the circular patterns of the kernel density map (Figure 7). Conversely, choosing a very large bandwidth could give the impression that small countries are contaminated across their entire geographical extents (e.g., Cyprus in Figure 6) and does not allow for discerning regional differences in contamination (Figure 7a). Moreover, a statistical analysis of the generated raster maps (Figure 7b) revealed that when a larger bandwidth was used, (1) the histogram of the map was more skewed, (2) the maximum density and the standard deviation were smaller, and (3) the kernel surface was smoother.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 7: (a) Influence of kernel bandwidth on visualisation and obfuscation, (b) Logarithmic histograms of KDE in Afghanistan

What, then, is the correct process for determining the appropriate bandwidth? The default bandwidth provided by the ArcGIS Kernel Density tool, for example, is the height or width of the input hazards layer (whichever is shorter) divided by 30. The choice of 30 is arbitrary, has no statistical basis and does not consider the relative distribution of points across the area. Furthermore, the dimensions of the input hazard layers are strongly influenced by outliers. Therefore, we attempted to estimate the correct bandwidth using an approach based on the input point datasets. Bailey and Gatrell (1995) proposed selecting a bandwidth value based on the number of points and the areal extent of the study area. In this method, however, no consideration is given to the spatial relationships between points. Other approaches and their applicability to one-dimensional samples have been described in the literature, most notably the Biased Cross-Validation and Unbiased Cross-Validation (Scott 1992), the Sheather-Jones Plug In (Sheather and Jones 1991, Jones et al. 1996, Loader 1999), and Silverman‘s Rule of Thumb (Härdle et al. 2004). We made an attempt to adapt all of these methods for use with 2-D data. We obtained very large bandwidths in comparison with the country dimensions, up to 160 km in Tajikistan (Sheather-Jones Plug In) and 130 km in Somalia (Unbiased Cross-Validation). This result motivated us to calculate bandwidth as the average distance to the k-th nearest neighbor (ADKNN). Inspired by Williamson et al. (1998), this

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

method is designed to reflect the degree of clustering and the spacing of points, rather than the extent of the study area or the point dataset size. k is derived from Equation 2: k = round [sqrt (n * P)] (2) where sqrt is the square root of a number, n is the number of input ERW and P is a parameter provided to the user for adjusting the level of detail of the map.

2.2.8. Quantitative Evaluation of the Visualisation Methods

To o evaluate the fitness of the six cartographic visualization methods to the requirements stated in Section 2.2.3.4, a quantitative analysis was performed. The main findings of the analysis are presented in Section 2.2.8.1 and Section 2.2.8.2. The results provide an interesting new perspective on these methods and lay the foundation for the qualitative analysis discussed in 2.2.9.

2.2.8.1. Influence of the Sample Density with varying P and RP

Bandwidths were calculated using Equation 2 for each of the test countries and for values of P ranging from 0.1 to 50. The results obtained using method D are plotted in Figure 8a (comparison between countries) and in Figure 8b (comparison between random subsets of the same dataset). The results obtained with method E using the same dataset are plotted in Figure 8c, with points randomly scattered in polygons and a recording precision (RP) between 20 and 200 points/km2. In this method, there is a risk that very small polygons will be omitted. To prevent this from occurring, each polygon is filled with at least one point. In Figure 8c, the largest sample contains 120‘000 points.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 8: Effect of the density of ERW input data on the precision of the output maps, (a) For method D, using data from several test countries, (b) For method D using random subsets of the same dataset, and (c) For method E 1Precision parameter P is from Equation 2 and is used for adjusting the level of detail of the map 2Densities are expressed in 10-3 ERW/km2 3Kernel bandwidths values are expressed in kilometres 4Recording precision (RP) is the density of points scattered in the polygons (number of points/km2)

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 8 clearly demonstrates the effect of sample density on bandwidth. As sample density increased, (1) bandwidth values diminished and curbs stabilized, resulting in improved, smoother maps; and (2) the influence of parameters P and RP on the range of bandwidth values diminished. Nevertheless, users can still adjust the level of detail of the kernel density maps to a certain extent by varying parameters P and RP. This allows them to maintain control over the representation of contamination that they wish to show. To evaluate the effect of the algorithm used for scattering points in polygons, we performed additional tests. (1) Some of the tests shown in Fig. 6c were repeated twice. Two tests with the same point density and the same P and RP values for the same set of polygons resulted in a difference in bandwidth values of less than 4%. (2) We also performed similar tests to the tests in Figure 8c, but this time the ERW polygons were filled with points that were regularly distributed in space. The difference in bandwidth values between the two tests with the same point density, P, and RP for the same set of polygons was less than to 3%. In addition, the difference in bandwidth diminished as input data density increased. Based on the results of these additional tests, the effect of the algorithm used to fill the polygons on the bandwidth is not significant. Additionally, as observed in Figure 8, the bandwidths computed with method D varied greatly among the different countries and consequently show very little adaptability. In contrast, the bandwidths computed with method E were smaller. Therefore, method E is a better fit for the data distributions, as shown in Figure 6.

2.2.8.2. Statistical Analysis of KDE-based Maps

A KDE was performed on the data from each test country, with bandwidths computed using the ADKNN method. The average, maximum, median, standard deviation and third quartile values are summarized in Table 2.

Table 2: Comparison between statistical indicators derived from KDE rasters, representing the six datasets. All values shown in the table are densities of ERW/km2. They were computed on the test data and do not reflect the reality in the field KDE Test country Sample size1 Mean Max Median Standard deviation Third quartile bandwidth2,3 Afghanistan 6‘644 27.02 580 55‘247 0 2‘576 151 Cyprus 94 8.75 310 7‘911 8 868 238 Iraq 196 35.08 1‘341 76‘060 0 5‘862 31 Lebanon 1‘496 4.68 1‘508 98‘585 0 5‘012 8 South-Central 133 31.01 29 2‘981 0 217 0 Somalia Tajikistan 202 23.47 116 2‘609 0 305 23 1Number of points and polygon centroids 2Average distance (km) to k-th nearest neighbour, with k derived using Equation 2 3Precision parameter P equal to 1

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

For each country, the median value was systematically close to zero, and the third quartile value was much lower than the maximum value. This result demonstrates that kernel maps tend to highlight highly affected areas (in other words, they make the contamination problem more visible). Moreover, the third quartile values were relatively low. This result shows that KDE-based methods more effectively preserve areas with low contamination, especially compared to the methods used for maps B and C, which use aggregate data for a given administrative level. The standard deviations were, on average, five times larger than the mean densities, and the number of non-zero values was much higher than in the original vector samples (in which the number of non-zero values was very close to zero). This result demonstrates that the use of KDE methods allows for the general distribution of ERW areas to be displayed without showing precise locations, which was consistent with the requirements of some users. The figures in Table 2 show high variation from one country to another, especially in the ―Mean‖ and ―Max‖ columns, where the lowest value was fifty and thirty-seven times smaller, respectively, than the highest value. These results highlight the risk of exaggerating or understating the contamination problem. Developing a worldwide contamination map will require a number of checks in terms of representation, as we discuss below.

2.2.9. Discussion

In the following section, we discuss the strengths and weaknesses of the six cartographic visualization methods and compare them to the requirements given by each of the four target audiences mentioned in Section 2.2.3. The discussion is based on the quantitative analysis described in Section 2.2.8 and the requirements stated in Section 2.2.3.4. The outcomes of our meetings with experts (Section 2.2.3.1) were also important. In particular, during the second focus group meeting, we showed the participants' paper contamination maps that were created using the cartographic visualization methods described above and the datasets from each country. The attendees were asked to respond individually to the questions listed below.  Based on the paper maps, which cartographic visualization method(s) do you consider to be suitable for mapping ERW hazards?  Which of these method(s) do you use the most?  At which scale(s) does each method fit your cartographic needs? (global / national / sub-national / local)?  Do you use other cartographic visualization methods for mapping ERW besides the methods used in the paper maps?  What are the strengths and weaknesses of each of the six methods?  Which method would best suit the needs of each of the four categories of users? A group discussion was then held. The results of the discussion are summarized below. The discussion

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

below is not structured according to the questions above, but it follows the main points raised by the focus group participants.

2.2.9.1. Ability to make the Contamination problem Visible

The ability to make the contamination problem visible is dependent on the scale of the representation. Methods B, C, D, E and F aggregate the original data at a scale at which the contamination problem is visible. In particular, cartograms have the capacity to enlarge small, heavily contaminated areas. By making all cells an equal size (size = 250 m), the two KDE-based methods provide much more detail than the other methods, induce data fuzziness and ensure a smooth transition between areas of low and high contamination. At smaller scales, method A maps become unreadable due to point symbol overlaps. Similarly, methods B and F become overloaded when the range of contamination values is too wide or when administrative units become too dense. Moreover, with a large number of administrative units, method F distorts administrative neighborhoods. The issue of making the contamination problem visible is related to the question of data confidentiality. With the exception of method A, all of the methods preserve data confidentiality by aggregating the information. They also hide the amount and exact location of map items behind symbols (O‘Sullivan and Unwin 2010). Kernel-based maps (D and E), conversely, do not preserve data confidentiality at large scales (Figure 7a). Another issue is related to the question of country borders. Kernel maps of highly contaminated countries could show contaminated areas erroneously extending into neighboring countries, especially using method D with large bandwidths. Clipping the density raster layers to the boundaries of each country would not be a convenient solution to this problem, given the heterogeneity of distribution patterns across countries. Clipping the layers to the boundaries of each country would make the boundaries discernible and disrupt the continuous display of the data. Generating kernel maps country by country without sharing the information with other countries is one solution, and processing KDE maps using data compiled in national repositories is another alternative. Finally, methods B and C make country boundaries visible, which might be politically problematic.

2.2.9.2. Capacity of Fitting the Original Data Distribution

Most data visualization methods inherently generalize and simplify the original data (Bertin 1977). Methods B, C and F assign one quantity to each administrative unit, and there is an inherent risk that changing the level of administrative unit would affect the message delivered by the map. This effect is known as the Modifiable Areal Unit Problem (MAUP), first described in detail by Openshaw (1984). The two KDE-based methods (D and E) are less sensitive to the MAUP because the units are identically sized grid cells with a resolution of 250 m. Furthermore, by using a much higher density of sample points than

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

D, method E generates smaller bandwidths and is thus a better fit for the original vector data, especially when there are long polygons (Figure 6) located along country borders and roads, where having accurate information about ERW is crucial.

2.2.9.3. Simplicity of the Conveyed Message

In all of the maps (A, B, C, D, E and F), it is naturally understood that higher densities of symbols, larger symbols or darker colors all represent higher values (Gaspar-Escribano and Iturrioz 2011). According to O‘Sullivan and Unwin (2010), proportional symbols induce an underestimation of larger values. Cartograms represent one degree of abstraction beyond choropleth maps (Dorling 1993). Cartograms may be difficult to use for GIS novices, who may consider the maps ―funny-looking‖ (O‘Sullivan and Unwin 2010, p. 79). Reading these maps also requires familiarity with the shape of the administrative units in the area being displayed.

2.2.9.4. Susceptibility to Data Uncertainty

With methods B, C, D and F, each point was weighted by its estimated or calculated area, which is stored as an attribute in IMSMANG. As mentioned earlier, area values and their reliability strongly depend on the hazard type (e.g., SHA or CHA). The accuracy of maps B, C, D and F is heavily dependent on this attribute, and the accuracy of maps using these methods would likely increase if the ERW were to evolve from SHA to CHA or to DHA status during the demining process. In contrast, method E uses the polygon surface area, which is more reliable because it was calculated from vertices' coordinates (see Figure 2). Therefore, the reliability of method E is higher than that of methods B, C, D and F. This result suggests that the maps may be better in countries that have a high capacity for implementing demining operations.

2.2.9.5. Capacity of Combination with other Datasets

All of the proposed contamination maps are composed of a single layer. Each category of end users (Section 2.2.3.2) can combine the layers as needed with other datasets, such as administrative boundaries, socio-economic data, population density, topography, land cover, and logistics. For an example of an ERW map combined with a population density dataset, see Figure 9 (using method C) and Lacroix et al. (2011).

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 9: Possible application of the proposed cartographic visualization methods: multiplication of contamination maps (upper left: method C using Afghanistan data at the district level) by population datasets (upper right: Gridded Population of the World (CIESIN and CIAT 2005)) shows where people are most at risk (bottom)

Similarly, methods B, C, D and E use one visual variable (i.e., value, color, or size), which would make it possible to add a simplified base map (Bertin 1977) or overlay environmental data and would open the door to publication on a web platform.

2.2.9.6. Ease of use and Implementation

Methods A, B and C are easily depicted at any administrative level and easy to generate in any GIS package, even for users with limited ―spatial ability and map experience‖ (Ozimec et al. 2010, p. 94). Though KDE is included in most GIS packages, the algorithm that supports method D is not and was developed from scratch. Given the variation in bandwidth from one country to another (e.g., the discussion in Section 2.2.9.1), method D should be determined at the country level. IMSMANG is a useful tool for doing so because it already holds national databases that are comprised mostly of points. The precision parameter P of Equation 2 can be provided to the users as an input setting if they wish to generate density maps with a higher or lower level of detail. Method E requires processing much more data than method D. Therefore, method E is more demanding in terms of computer performance, and it would be difficult to deploy method E to end users in the dozens

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

of countries that use IMSMANG. Because method E shows a high adaptability to datasets, E is suitable for use at the global level to create a single worldwide map that compiles all national data. In practice, such a project will depend on the capacity of national mine action programs to provide IMSMA data as polygons.

2.2.9.7. Colour Scheme

As mentioned earlier, generating maps of ERW contamination at the global scale requires the development of a universal color ramp. To prevent the under-representation of regions of the world with low contamination and the over-representation of heavily affected areas, we decided to display the natural logarithm of the kernel density raster instead of the raw density values, following Newbury and Bright (1999). This method ensures that the contamination in all regions is measured on a similar scale and reduces the effect of the precision parameters P and RP on the output map (Figure 10). Changing the representation of the data can strongly influence the intended message and ultimately mislead the map reader (Ozimec et al. 2010), especially when applying data classification methods such as natural breaks (Jenks and Caspall 1971), equal intervals, quantiles and geometric intervals to very heterogeneous and skewed kernel distributions. These data classification methods are sensitive to the number of classes (Herzog 2010). They also disrupt the continuous nature of the density rasters and the smooth and fuzzy effect inherent to KDE maps. Therefore, data classification was not used. Instead, we developed a continuous color ramp based on the sample with the widest range of values. Following Müller et al. (2006), with the objective of attracting the reader‘s attention to heavily affected areas, the main colors of the color ramp are yellow, orange and red. To improve the readability of the maps and leave room for background layers (to publish the maps on a web server), transparency was applied to the hazard layer, and a second color ramp, ranging from white to yellow, was used for areas of low contamination (Figure 10). Finally, for reasons of confidentiality, no contamination figures were shown in the legend. Instead, the terms ―low‖, ―medium‖ and ―high‖ were used.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 10: Maps showing contamination by ERW in Afghanistan. The map in the upper left corner corresponds to method A (dots). For the other maps, technique E (KDE on polygonal ERW) was applied in Afghan regions of varying extent and ERW density, and an identical colour ramp was used for each map

Figure 10 juxtaposes the least and most sophisticated of the six cartographic visualization methods and is a good way of highlighting the outcomes of this study.

2.2.9.8. Which Visualisation Methods for whom?

I n this section, we draw conclusions about which of the proposed cartographic visualization methods are the most relevant for each category of actors, as a function of map scale. Users outside the core field of mine action (i.e., donors and the general public) want to compare contamination levels between regions of the world without revealing the exact location of ERW. The message conveyed by the map must be intuitive and unambiguous. For this category of map consumers, method E emerged as the most suitable representation method in the form of a worldwide map combined with a universal color ramp, a basic ordinal legend and metadata in conformance with the OGC standards (OGC 2007a). Publishing this map on a web-GIS portal would help sensitize the users to the global mine contamination problem. Implementing this method requires that national mine action authorities supply

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

polygon data. At the national and sub-national scales, choropleth maps may also be helpful in providing supporting information for this category of users. As mentioned earlier, directors of national mine action programs need maps at the national and regional levels. One purpose of these maps is to support strategic decision making about the use of resources. Another objective is for the directors to be able to present summaries of their achievements to other actors. Methods B, C, D, E and F are likely candidates at these scales (Sections 2.2.9.1 and 2.2.9.6). Method E has a long computation time and would therefore be difficult to use in 60 countries (Section 2.2.9.6). Method F might be difficult for GIS novices to use (Section 2.2.9.3). Methods B and F become overloaded with wide ranges of contamination values or dense administrative units (Section 2.2.9.1). Method C is already in use in some programs, but method D can be used to show more details (Section 2.2.9.1), is less sensitive to the MAUP (Section 2.2.9.2), preserves areas of low contamination (Section 2.2.9.2), stores data at the national level (Section 2.2.9.6), and is likely to encourage the sharing of information (Section 2.2.8.2). Method E handles polygons, which disqualifies it as an option for repositories in which ERW are mostly recorded as points. Based on these criteria, it was decided that method D would be automated and implemented as a standard in the IMSMANG cartographic module. This action should make it easier for directors to disseminate maps inside and outside of the mine action community at the national and regional levels. Methods B, D and F are also likely candidates for helping operations officers make decisions at the regional and sub-regional levels. Methods B and F, however, are not recommended for the reasons described in the previous paragraph. Method A, combined with topographic and logistics layers (Section 2.2.9.5), is the most appropriate at the municipality level, Database administrators and analysts work on the database, upstream of the maps, and aim to reduce uncertainty around data. Method A, combined with layers such as points of interest, is the most suitable at large scales (e.g., at municipality level and beyond)).

2.2.10. Conclusions and future Outlook

This study aimed to develop and assess a range of cartographic techniques that would allow for the visualization of contamination by explosive remnants of war (ERW) at different scales (ranging from the local (municipal) to the global scale) and for different groups of actors involved in mine action. The novelty of this work can be observed from several perspectives. First, for actors in the mine action community, the study defined the requirements of ERW mapping for various user groups working at different scales. Second, the study provides a comprehensive framework and set of tools to visualize ERW contamination data. The systematic comparison and evaluation of six different cartographic visualization methods allowed for each method to be evaluated in terms of its relevance for each of the four categories of mine action stakeholders and at different scales. At the global scale, for example,

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

potential donors can be made aware of the problem of worldwide contamination by ERW and of the regional variation and hot spots of ERW contamination using method E. Third, the paper represents innovation within the scientific community. An extension of the traditional kernel bandwidth selection technique is proposed for KDE-based visualization methods. This technique provides close-to-reality representation of points and polygons at large scales (i.e., sub-national, national, global) for any spatial data distribution, provided that a universal color ramp is used (such a color ramp was proposed, but it may be further improved). Finally, we made these cartographic visualization methods available to the humanitarian demining community. The work presented here also shows potential for further contributions beyond this study. The implementation of the two KDE-based method E could be a first step towards the sharing of geospatial ERW data as a continuously maintained resource that is freely accessible to both the public and private sectors (Ryttersgaard 2001). The data could be shared across institutional, regional, and national borders using compatible technology (Granell et al. 2009), while preserving data non-disclosure agreements and allowing users to control the level of detail they want to share. For several reasons highlighted in this study, our work should encourage national programs to share a maximum amount of their data in future (Barlow 2003), not only for the mine action community but also to promote interdisciplinary scientific work (Köhler et al. 2006). Combining ERW contamination maps with other layers (e.g., population density, strategic infrastructure, points of interest, and development areas) could help decision-makers determine the socio-economic impacts of hazards and prioritize future surveys (Benini et al. 2003, Alegría et al. 2011). Because the reliability of the kernel maps varies with the density of the input data, map consumers should be made aware of the risk of exaggeration or underrepresentation of ERW contamination. In addition, by pointing out the degree of uncertainty in point data and the potential reliability issues of each cartographic visualization method, we as cartographers assume our responsibility to inform end users about the maps‘ limitations (Evans 1997). The main lessons learned during this study are the following. First, map users in the mine action community are more interested in the visual results of the maps than the methods and algorithms with which the maps were developed. Second, users need easy-to-use tools, and the users should be spared the complexity of the algorithms used. Third, our research began three years ago, and it appears that it will take some time before dozens of countries establish the use of methods such as D and E. The successful implementation of these methods will depend on our ability to efficiently demonstrate the benefits of the methods and modify them based on user feedback.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.3. Using Clustering Techniques to improve Visualisation of Contamination

Development of a Novel Unsupervised and Automated Mapping Method based on Clustering and KDE

2.3.1. Introduction

Two extensions of Kernel-based mapping methods have been developed and introduced in the paper entitled ―Methods for Visualising the Explosive Remnants of War‖ (Section 2.2). These methods have some limitations. ADKNN-Points (method D in Figure 4) maps may be very disparate across different countries, which might make maps difficult to compare. ADKNN-Polygons (method E in Figure 4) handles polygons, which limits its applicability to repositories where ERW are mostly recorded as polygons. Both methods require specifying some precision parameters to adjust the level of detail of the contamination map. One the one hand this allows users to keep control over the representation of contamination that will be shown. On the other hand this opens the door to over- or under-estimation of the contamination in their country. To reduce these limitations, we develop a novel mapping method (ADKNN-Clusters) and implement it as an unsupervised and automated function in ArcGIS Desktop. Based on the assumption that mine action data have a certain degree of clustering, this technique goes one step further than the two mentioned previously in terms of fitness to spatial distributions: data are grouped in clusters before being applied a customised KDE. The way we thought and developed this technique, as well as its pros and cons, is described and discussed in Section 2.3. For more clarity, the three KDE-based novel mapping methods, along with their names, are presented in Table 3.

Table 3: Overview of the three novel KDE-based mapping methods developed for this thesis Novel KDE-based Reference in this Processed features Number of KDE mapping method thesis 14 Points and polygon 1 bandwidth / dataset ADKNN-Points Section 2.2 centroids ADKNN-Polygons15 Polygons filled up with 1 bandwidth / dataset Section 2.2 points ADKNN-Clusters Points and polygon 1 distinct bandwidth / cluster Section 2.3 centroids

With the ADKNN-Clusters approach, each cluster has its own kernel bandwidth, calculated as the

14 Method D in Figure 4 15 Method E in Figure 4

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

average distance to k-th nearest neighbour within this cluster. k is proportional to the square root of n and n is the amount of input ERW. The ADKNN approach, which is a common denominator for the three KDE-based techniques, has proved to fit most of requirements stated in this chapter of the thesis: ability to make the problem of contamination visible, simplicity of the message conveyed by the map, preservation of data confidentiality, data obfuscation, capacity of maps to fit the original data to a certain degree, implementability as an automated function and capacity to handle highly heterogeneous data (heterogeneous in type, quality, positional accuracy, reliability and degree of completeness). Section 2.3 is structured as follows. In Section 2.3.2, we describe test data and conduct spatial statistics tests to evaluate their degree of clustering. To explain why these features are not dispersed or arranged randomly in space, we assess dual distribution relation between them and auxiliary data (roads, country borders, population densities, and land cover types). In Section 2.3.3, we expose some general principles of data clustering and describe well-known clustering methods. For GIS implementation, we choose (1) a clustering algorithm developed by Wong et al. (2001), and (2) the Caliński and Harabasz‘s Variance Ratio Criterion (Caliński and Harabasz 1974) as performance measurement. We validate these choices by testing the algorithm on about thirty datasets of various size, shape, density and orientation. The ADKNN-Clusters workflow is exposed in Section 2.3.3.4. In Section 2.3.4, results on different test data are presented for Afghanistan at different scales and are discussed.

2.3.2. Mine action data

2.3.2.1. Test datasets

Test datasets used to develop the ADKNN-Clusters mapping method are presented in Table 4 and described now.

Table 4: Test data used for development of the ADKNN-Clusters mapping method Spatial Dual distribution Implementation Sample Results & References Category Test dataset statistic relation with of the clustering size discussion in this thesis tests auxiliary data algorithm Afghanistan1,2 6‘644 X X X Afg. random subset2 1‘000 X A Sections Cyprus1,2 94 X X X (IMSMANG 2.3.2.2, Iraq1,2 196 X X X national 2.3.2.3 & Lebanon1,2 1‘496 X X X repositories) 2.3.3 Somalia1,2 133 X X X Tajikistan1,2 202 X X X Afghanistan random3 6‘644 X Afghanistan random3 1‘000 X Cyprus random3 94 X Sections B Iraq random3 196 X 2.3.2.3 & (random) Lebanon random3 1‘496 X 2.3.4 Somalia random3 133 X Tajikistan random3 202 X

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

30 self-made point C patterns of various 10 to 370 X Section 2.3.3 (self-made) shape, size, density & points orientation D 2‘496 X (IMSMANG Afghanistan subsets2 946 X Section 2.3.4 subsets) 233 X 1Same test dataset as for the ADKNN-Points method 2Extracted from IMSMANG national repositories in the form of points and polygon centroids 3Points are randomly scattered inside country limits

 Category A: We use six IMSMANG databases. Those are the same data sets than the ones used for the development of ADKNN-Points and ADKNN-Polygons (See Section 2.2). As explained in Section 2.2.4, the selected case study countries are representative of prototypical countries (Afghanistan, Cyprus, Iraq, Lebanon, Somalia and Tajikistan) on account of the relative heterogeneity in terms of environmental, historical and cultural background, as well as predominant demining techniques. Spatial distribution patterns also are heterogeneous, as was mentioned in Section 2.2.4, and country extents range from 9‘250 km2 for Cyprus to 650‘000 km2 for Afghanistan. To ensure robustness of spatial statistics tests, a 1‘000-point random subset is also extracted from the Afghan repository, which makes seven datasets for category A.  Category B: Random points are also scattered across each of the six country extents. On this basis, comparative tests of dual distribution relation with auxiliary data (roads, population densities, land cover types and country borders) can be achieved both on these random points and on the IMSMANG datasets of Category A.  Category C: About thirty artificial datasets of various shape, size, density and orientation are designed for testing the various clustering algorithms and performance measurement processes.  Category D: To discuss cons and pros of ADKNN-Clusters, maps are generated on 2‘496- 946- and 233-point subsets extracted from the IMSMANG Afghan repository, as well as on a shapefile with 1‘000 points randomly scattered inside country limits. When needed, data are re-projected from the World Geodetic System 1984 (WGS 84) to the local Universal Transverse Mercator (UTM).

2.3.2.2. Evaluation of the degree of data clustering

The ADKNN-Clusters method was developed assuming that ERW have not been deployed nor collected in the field in a random or dispersed manner. A number of discussions with mine action experts including GICHD staff have underscored that ERW are likely laid in particular areas (e.g. along country borders and roads, nearby strategic infrastructures, and in locations that were meant to block the access to agriculture or water) and that mine action data are likely to be collected in secure and accessible areas. Various types of constraints (financial and political issues, population settlements, environmental and

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

topographic constraints such as natural barriers, mountains, rivers, and forests etc.) might also have influenced deployment and collection of landmines. As an example, mine action experts recognise that mines in Sudan are mostly found along roads. Alegria et al. (2011) for their part, have put in evidence dual distribution relation between landmine incidents in Colombia and roads connecting population settlements. In this context, we assume that landmines are not arranged in space in a random or dispersed way, and that data repositories hold a certain degree of clustering16. To confirm this assumption, we conduct a series of spatial statistic tests on point patterns presented in the coming pages. The analysis of spatial point distributions has been applied to many disciplines, such as ecology (Rice et al. 2012), epidemiology (Liebman et al. 2012) and biodiversity conservation (Chen et al. 2011). One of the first scientists to follow this type of approach was John Snow in the 1850‘s, who showed a spatial relationship between the clustering of cholera cases around a water pump and the source of the infection (Snow et al. 1965; Gatrell et al. 1996). If spatial statistic concepts and methods are numerous in literature, spatial autocorrelation is particularly interesting in our case, as it is based on the idea that nearby entities share more similarities than entities that are far apart (Tobler's first law of geography; Tobler 1970). In particular, Moran‘s spatial autocorrelation Index, Ripley‘s K function and Kolmogorov-Smirnov test of Complete Spatial Randomness (CSR) are reliable indicators for analyzing point pattern distributions (Alegria et al. 2011, Lieske et al. 2012, Piro et al. 2012). Moran‘s spatial autocorrelation index allows one to evaluate whether a pattern is dispersed (index close to -1), random (rather near 0) or clustered (near +1), and with which confidence this can be affirmed. In commercial GIS platforms such as ArcGIS, calculation of Morans‘ index is assorted with calculation of a Z-score (Esri 2011a) indicating whether the null hypothesis can be rejected or not17. The probability that the null hypothesis was falsely rejected is in turn evaluated from the p-value probability. For instance, if the Z-score is higher than 1.96, then the null hypothesis can be rejected with a p-value probability of being wrong when rejecting it. The ArcGIS Spatial Autocorrelation (Morans I) tool was applied to seven of the IMSMANG test datasets presented in Table 4 (category A). Maximum threshold distances were chosen large enough to surround the range of bandwidth values generated by the ADKNN-Clusters method (see Table 7 below). Results of the Moran analysis are summarised in Table 5.

Table 5: Results of the ArcGIS Spatial Autocorrelation (Morans I) tests Sample Threshold Global Moran’s Test country Z-score p-value size1 distances Index

16 One exception to this rule, however, might be Laos, where hundreds of thousands of bombs dating from successive aerial bombardments were spatially analyzed by Herzog (2010) and proved to be rather randomly distributed. 17 The null hypothesis states that patterns are spatially dispersed

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Afghanistan 6‘644 5 to 75 km 0.14 5.14 0.00

Afghanistan 1‘000 5 to 75 km 0.18 20.27 0.00 (random subset) Cyprus 94 5 to 25 km 0.05 0.05 0.94 Iraq 196 5 to 50 km 0.15 1.70 0.08 Lebanon 1‘496 5 to 25 km 0.02 0.84 0.42 South Central 133 5 to 50 km 0.01 0.32 0.75 Somalia Tajikistan 202 5 to 50 km 0.05 0.20 0.84 1 ERW points and polygon centroids

The Afghan and Iraqi patterns are obviously clustered, with respectively a less-than-1% and a less-than- 8% likelihood that the clustered pattern could be the result of random chance. The Lebanese pattern is somewhat clustered. For Cyprus, South Central Somalia and Tajikistan, patterns are neither dispersed nor clustered according to the Moran index. To refine these conclusions, a spatial cluster test in ArcGIS, based on Ripley‘s K function (Esri 2011a) was applied to the seven datasets. This function can be used to summarise spatial clustering over a range of distances. It compares the observed points with the case of a generated homogeneous Poisson point pattern. Figure 11 shows the difference between the observed and expected values in function of distance. The higher this difference, the more clustered is the pattern likely to be. If the difference is negative, patterns are considered dispersed. Values close to zero correspond to random distributions.

Results of tests for spatial cluster (based on Ripley's K function)

Afghan subset

Afghanistan

1 80 Iraq

Tajikistan

Lebanon Clustered Cyprus

South Central Somalia Random0

0 20 40 60 Difference between

Dispersed observed and expected values

-80 Distance analysis1

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 11: Results of the Multi-Distance Spatial Cluster Analysis (Ripley‘s K function) tests 1 All values are expressed in kilometers

Ripley and Moran tests go in the same direction: all countries show clustering, especially Afghanistan and Iraq, and Tajikistan to a lower degree. In Cyprus, South-Central Somalia and Lebanon, point patterns are clustered until 17 km, 28km and 37 km respectively. These thresholds are larger than the bandwidth values produced by the ADKNN-Clusters (see Table 7 below). The curve obtained with a subset of the Afghan IMSMANG repository is close to the one obtained with the entire database, which proves the robustness of our tests. A Kolmogorov-Smirnov test of CSR was also performed on the Afghan and Cypriot datasets (Herzog 2010) and D-values were evaluated. A D-value of 0.47 is obtained for Afghanistan and 0.18 for Cyprus18. Poisson distribution D-values equal to 0.25 and 0.22 were also calculated, confirming that the patterns do not follow a Poisson distribution.

2.3.2.3. Dual distribution relation between mine action and auxiliary data

To explain the spatial clustering of point patterns, we searched for dual distribution relation between the six IMSMANG repositories (Table 4 – Category A) and auxiliary data such as transportation network (OpenStreetMap contributors 2012), population densities LandScanTM (ORNL 2008), land cover types GlobCover (ESA 2010) and administrative boundaries (GADM 2012). Results are spectacular. They are presented below. In Cyprus, 90% of recorded hazards are concentrated along the 350 km2 buffer zone that partitions the island into two parts since 1974 after a conflict between Turkish and Greek Cypriots. Clear patterns of mine repartition along conflict lines are also perceptible in the case of Lebanon. It is especially visible in relation to the wars with Israel, who occupied parts of the country from March 1978 to July 2000 (Rolland 2003). More than 50% percent of the mines are concentrated in a thousand meters distance from the border with Israel. Another 2.5 percent of the landmines are concentrated in the area separating Lebanon from the Israeli-occupied Syrian Golan, including the Lebanon-claimed Shebaa farms (Da Lage 2011). The nearness to transportation network was examined in Afghanistan, Iraq, Somalia and Tajikistan, by counting the amount of ERW found within concentric buffers around major roads. Similar tests were achieved on patterns with same sample sizes, composed of points randomly scattered inside countries‘ limits (see Table 4 – Category B). A logarithmic tendency function was fitted over each of the two curves (Figure 12).

18 A value of 1 indicates a regularity of the point pattern and a value of zero indicates a complete randomness

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Amount of ERW recorded along roads 1 100%

90%

80%

70% 2

60%

50% IMSMA-NG datasets Random datasets 40%

Amount of ERW of ERW Amount LOG(IMSMA-NG datasets) 30% LOG(Random datasets) 20%

10%

0% 0 2 4 6 8 10 12 14 16 18 20 Distance to roads (km)

Figure 12: Hazards versus distance to roads 1In Afghanistan, Iraq, South-Central Somalia and Tajikistan 2Overall = 7‘175 points and polygon centroids

47% of recorded ERW are no further than 1 km from a road, versus 19% for random distributions. Below 2 km, the amount of IMSMANG ERW is between 2 and 3 times higher than the amount of random points. R2 values are equal to 0.93 for IMSMANG datasets and 0.98 for random points. 17.4% of hazards recorded in the six countries (versus 12.4% of the random points) are located in zones with more than 50% of croplands or at least 50% of urban areas (categories 11-14-20-190 of GlobCover). In Afghanistan, some 30% of recorded landmines are found in areas with a population density greater than 100 persons / km2, and about 20% where it is higher than 200 persons / km2. A set of geo-statistical tests conducted in Sections 2.3.2.2 and 2.3.2.3 have shown that ERW are not recorded and deployed in space in a random or dispersed manner. Tests have also highlighted a substantial heterogeneity in the degree of clustering of the data. In some countries, hazard information has been collected mainly inside and/or along objects with different shape, size and orientation, such as frontiers, roads, population settlements, croplands. To fit point patterns at best, contamination maps should then be able to display clusters of various shape, size, density, orientation and degree of clustering, and at various scales. Section 2.3.3 describes the choice of the clustering algorithm.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.3.3. The clustering algorithm

2.3.3.1. Requirements

Cluster analysis consists of grouping points for solving different types of classification (Rohlf 1974), typological and taxonomical problems (Caliński and Harabasz 1974) and finds innumerable applications spanning many topics, including image analysis (Chang et al. 2010), genetics (Krishna and Murty 1999), bioinformatics (Miller et al. 2008) and exploration seismology (Aminzadeh and Chatterjee 1984). Clustering algorithms are meant to find underlying structures in datasets, i.e. to divide patterns in dissimilar groups that can be treated as single patterns with specific characteristics. In the case of mine action data, the challenge is to implement a robust, reliable and efficient algorithm able to address numerous requirements. Some of these requirements are conclusions of Section 2.3.2.3 (see below), others are recurrent in literature:  Catch clusters of large variability in shape, size, density, orientation and degree of clustering (Section 2.3.2.3).  Capture clusters at various scales (Section 2.3.2.3).  Work unsupervised (Wong et al. 2001, Ma and Chow 2004). In the case of mine action data repositories, the number of clusters and the location of their centroids are a priori unknown.  Be immune to the order of input patterns (Xu and Wunsch 2005).  Work in the two spatial dimensions, but also support 3D or even higher dimensionality. Even though IMSMANG repositories hold XY pairs of coordinates, the use of other IMSMA numerical and ordinal attributes as third and further dimensions (e.g. elevation, clearance priority) is not excluded in the future.  Detect outliers and noise (Dave 1991, Chiu et al. 2001).  Handle newly occurring data (Duda and Hart 1973, Xu and Wunsch 2005). In some countries like Kosovo, Nicaragua and Zambia, a couple of edits are done in IMSMANG each year. In other countries like Afghanistan it is likely dozens of thousands. The clustering algorithm must be able either (1) to deal with new data without relearning from the scratch, or (2) to provide good performances for regenerating new contamination maps on request.  Be easy to implement (Euntai Kim et al. 1997, Zhuang et al. 1998).  Be validated, in order to determine the optimal number of clusters and to separate the signal in the data from noisy structure (Lange et al. 2004). This stability-based validation process bases on the optimisation of a performance index (Davies and Bouldin 1979) and stops the algorithms when a ―stopping rule‖ (Milligan and Cooper 1985) is respected, e.g. when the performance index starts to grow or reaches its first local maximum. To ensure the efficiency and effectiveness of clustering algorithms, performance measurement has to be tested on arbitrary shapes and provide satisfying visual results (Xu and Wunsch 2005).

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.3.3.2. Choosing the clustering algorithm

Hundreds of clustering algorithms have been published, and the challenge for us was to select and implement one that could fit a maximum of the requirements stated in the previous section. On the basis of 300 scientific references, Xu and Wunsch have presented in 2005 a survey of the most common clustering techniques, including hierarchical clustering (e.g. Unweighted Pair Group Method with Arithmetic Mean (UPGMA)), squared error-based methods (e.g. K-means), mixture density functions (e.g. expectation-maximisation (EM)), fuzzy clustering (e.g. fuzzy C-means (FCM)), graph- theory, combinatorial search techniques, neural networks-based clustering, random sampling, sequential clustering and similarity measurement. The authors have put in evidence their pros and cons and have listed a number of major research trends and issues. On this account, and with regards to our requirements, some of these clustering techniques were irremediably disqualified. If we consider classical hierarchical clustering for instance, the most common criticisms are (1) the lack of robustness of the algorithms and (2) their computational complexity. Hierarchical clustering algorithms are at least O(n2) – n is the sample size – which implies that processing 5‘000 ERW points would involve several dozens of millions of computing operations. Another example is the well-known K-means. Because it minimises the distance to the cluster‘s centroid, it is limited to finding spherical clusters and not arbitrarily shaped ones. Furthermore, K-means forces outliers to belong to a cluster and thus distorts the cluster shapes. In the case of FCM, the initial number of clusters has to be predetermined, which does not address out needs. With EM, results are sensitive to initialisation. Apart from the above mentioned clustering techniques, we investigated other methods extensively used within the scientific community (Potts 1952, Sokal and Michener 1958, Cuevas et al. 2001, Ma and Show 2004, Chang et al. 2004 and others). The main strength of Potts model appears when working with raster data, which is not our case. The algorithm is also quite complicated to adapt as it requires knowledge of other algorithms such as the Swendsen-Wang and Wolff algorithms, as well as to perform Markov-Chain-Monte-Carlo-simulations. Similarly, the methods that were investigated proved to be difficult to implement, either because of (1) the difficulty to find available source code or (2) the complexity of the underlying theory. The algorithm that we selected was proposed by Wong et al. (2001). Based on replacement of movable vectors and on regulation of a similarity measure, it gathers the input points into several clusters such that the points within a cluster are more similar than outside this cluster. The different steps of the process are illustrated with blue in Figure 13: (1) At first iteration (q = 1), one of the input points is chosen as a ―reference vector‖ and a convergence variable is set. (2) A similarity matrix helps to find ―comparison vectors‖, i.e. points that have a high similarity to the reference vector. (3) The reference vector is then replaced by the average of the comparison vectors. (4) Steps (2) and (3) are repeated until all points are seen as reference vectors. In this configuration,

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

all the replaced vectors tend toward their cluster centre.

(5) At this stage, Vd values (d is the d-th cluster) are computed by comparing the inter-cluster

distances with the intra-cluster distances. Each Vd is a ratio. At numerator, the squared distance between d-th cluster centre and the cluster centre that is closest. At denominator, the quadratic mean

within-distance to d-th cluster centre. The performance index is calculated as the average Vd over all clusters. (6) Steps (1) to (5) are iterated while the convergence variable decreases. At each of the iterations, the number of clusters decreases and the performance index is recalculated. The algorithm stops when the performance index grows for the first time, i.e. when it gets higher at q-th iteration than at (q-1)-th iteration.

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 13: Workflow of the ADKNN-Clusters mapping method

Wong‘s algorithm was experienced on six of the IMSMANG samples (Table 4 - Category A: Afghanistan, Cyprus, Iraq, Lebanon, South-Central Somalia and Tajikistan) as well as about thirty self-made datasets of various size, shape, density, orientation and degree of clustering (Table 4 - Category C). Some of the results are presented in Figure 14. The number of clusters that were found by the algorithm is in maroon and the number of ―expected‖ clusters is in violet. By ―expected‖ we mean clearly ―consistent with the inherent structure of the real data set‖ (Wong et al. 2001, p 440). Wong‘s algorithm has many advantages. First, it works unsupervised. All values come from the underlying structure of the dataset. There is no need either to predetermine number of clusters or cluster centres during the initialisation phase. Second, no similarity measure has to be predetermined according to the shapes of the data. Third, it is applicable for multi-dimensional data. Fourth, it is well-adapted to data that change over time (according to the authors). Fifth, it required writing less than 500 lines of Python code, developed from scratch. Sixth, it is able to detect clusters of various size, shape, density, orientation and degree of clustering (Figure 14), especially in obvious pattern and when clusters are clearly isolated. Nevertheless, weaknesses can be found in the following situations. With small cluster separations the algorithm is permissive: it finds fewer clusters and stops later than one would ―expect‖ (Figure 14c and Figure 14j). Another weakness of this method is that its ability to detect small clusters and noise is perfectible. Samples shown in Figure 14h and Figure 14i both contain 62 points and have similar cluster separations. The only difference between them is the small central group of points that was moved to South-East. In Figure 14h however, the algorithm finds the ―expected‖ clusters while in Figure 14i, the three Southern groups (including the moved group) form one cluster where three would be ―expected‖. It seems that the moved group is attracted by the two neighbouring groups, whose size is larger. Comparison can be done with Newton force, which is proportional to the mass of each point and to the square of the inverse distance. In our case, the number of points in each cluster is likely to influence the attractiveness of large clusters on small ones. By reducing the number of points in the Southern group (Figure 14g), the algorithm now finds the ―expected‖ clusters. A third weakness of this method it that with more complicated patterns (Figure 14m) or with IMSMANG data, Wong‘s algorithm often stops at first iterations and catches more clusters than one would ―expect‖. Additional tests were performed to refine the analysis: Wong‘s algorithm was re-executed on the same 30 test datasets, as well as on the six IMSMANG countries, but this time the stopping rule was removed. Results were promising: in most situations there was systematically one iteration for which the ―expected‖ clusters (number and centroids) were found19. This proves that he algorithm is working fine although there is room for improving the stability-based validation process. Having the objects already

19 In some situations however, it was not possible to "expect" a set of clusters (e.g. with random points)

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

clustered, we should replace the validation process with another that fits our requirements. The choice of a new validation process could potentially affect the performance index, the stopping rule, or both.

2.3.3.3. Choosing the stability-based validation process

A set of performance measurements that could replace the one developed by Wong have been analyzed. These performance measurements all come from an overview presented by Milligan and Cooper (1985). The most extensively used in the scientific community are the C-Index (Dalrymple-Alford 1970, Hubert and Levin 1976), Duda and Hart‘s criterion (Duda and Hart 1973), Caliński and Harabasz‘s dendrite method (Caliński and Harabasz 1974), Goodman and Kruskal‘s Gamma index adapted by Baker and Hubert (1975), the Mojena stopping rule (Mojena 1977), the Davies-Bouldin index (Davies and Bouldin 1979), the Point-Biserial coefficient of correlation (Milligan 1980, Milligan 1981) and the Cubic Clustering Criterion (Ray 1982, Sarle 1983). To mitigate the drawbacks encountered with Wong‘s performance measurement, we decided to replace it with (1) a performance index that considers all clusters (and not only the closest one), and with (2) a stopping rule that corresponds to a local maximum of the index (rather than to a comparatively increase). These two requirements motivated us to choose the Variance Ratio Criterion (VRC). This performance criterion was developed by Caliński and Harabasz (1974) with bacteriological, anthropometric and plant breeding data. The principle of this criterion is to both minimise heterogeneity between clusters and maximise homogeneity within clusters. The VRC is calculated using Equation 3: BGSS n  Nq VRC  (3) q 1 WGSS where BGSS = Between Groups Sum of Squared distances, WGSS = Within Group Sum of Squared distances, q = current iteration, n = sample size and Nq = number of clusters at q-th iteration. Caliński and Harabasz suggest choosing the number of clusters for which the VRC reaches the first local maximum. In line with Euntai Kim et al. (1997) and Zhuang et al. (1998), Calinski and Harabasz‘s dendrite method is easy to implement in a GIS such as ArcGIS. It also benefits from an excellent academic reputation (1‘105 citations in Google Scholar as of May 2012). Wong+VRC was applied to the 30 datasets and to the six IMSMANG countries. Some of the results are shown in Figure 14. They are in red colour and can be compared to Wong‘s (in maroon) and to what we are ―expecting‖ (in violet).

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 14: Comparative clustering tests on point patterns with different stability-based validation processes. ―Expected‖ numbers of clusters are in blue, results with Wong in maroon results with Wong combined with the VRC in red

Results of Wong+VRC are much better than results of Wong. In most cases, the combination of Wong+VRC gives the ―expected‖ results. Wong‘s clustering algorithm provides a relevant list of clusters (number + centroids) and the VRC calculates which element of the list to keep. Results with IMSMANG

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

data are also significantly improved, especially considering scale change. This will be discussed in more details in Section 2.3.4.1. Nevertheless, some situations are still difficult to handle, in particular with complicated patterns (Figure 14o) and with non-clustered samples (Figure 14p). Similarly, nothing guarantees that Wong+VRC will provide good results in all situations. We should keep in mind that ―no clustering algorithm can be universally used to solve all problems‖ (Xu and Wunsch 2005, p.672).

2.3.3.4. ADKNN-Clusters mapping method: methodology and workflow

The general workflow is presented in Figure 13 and comprises the following steps:  Data preparation (green) Input rows are points and polygon centroids extracted from IMSMANG to shapefiles or Esri Geodatabase feature classes. Datasets from distinct countries (e.g. neighbouring) may be merged into one dataset. Duplicates shall be identified and removed if needed.  Clustering algorithm (blue) IMSMANG data are clustered by Wong‘s algorithm. Any iteration of the algorithm provides a distinct set of clusters and cluster centroids. The number of clusters at iteration q+1 is lower than at iteration q.  Performance measurement (red) Caliński and Harabasz‘s VRC is estimated at any iteration. The preferred number of clusters is determined when the performance index reaches its first local maximum.  KDE (green) A distinct KDE is applied to each cluster. ERW are weighted by an attribute of their calculated or estimated area. Kernel bandwidth is equal to within-cluster ADKNN.  Raster sum All density rasters are summed into one final Kernel map.

2.3.4. Results and discussion

To discuss pros and cons of the ADKNN-Clusters mapping method, a similar quantitative analysis as the one conducted for ADKNN-Points (see Section 2.2) is presented below. For both visualisation methods, KDE maps are produced from subsets of IMSMANG repositories in different countries, and from a 1‘000- point random pattern on the Afghan extent. Some statistical indicators of KDE distributions are provided in Table 6. Statistics about number and density of points and KDE bandwidth are also produced at cluster level (Table 7). The added value of ADKNN-Clusters in comparison to ADKNN-Points is emphasised, and the sensitiveness of the ADKNN-Clusters mapping maps to parameters such as scale, outliers and spatial randomness of point patterns is examined, and illustrated in Figure 15 and Figure 16.

100

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Table 6: Comparison between statistical indicators derived from KDE rasters, representing three different subsets of the IMSMANG Afghan dataset. All values shown in the table are densities of ERW/km2. They were computed on test data and are not meant to reflect reality in the field Sample KDE Standard Third Mapping method Mean Max Median size1 bandwidth2,3 deviation quartile 2‘496 ADKNN-Clusters 1 by cluster 410 99‘021 0 2,546 75 2‘496 ADKNN-Points 47.03 405 20‘636 9 1,363 230 946 ADKNN-Clusters 1 by cluster 508 99‘004 0 3,463 49 946 ADKNN-Points 48.87 500 19‘246 18 1,619 222 233 ADKNN-Clusters 1 by cluster 4,088 830‘820 0 33,901 0 233 ADKNN-Points 5.54 4,036 228‘664 0 18,353 91 1Number of points and polygon centroids 2Average distance (km) to k-th nearest neighbour, k being proportional to the square root of the sample size (ADKNN-Points) or the square root of the cluster size (ADKNN-Clusters) 3 Precision parameter P = 2

Table 7: Comparative statistics between density rasters produced with ADKNN-Points and ADKNN- Clusters in Afghanistan

Number Number of points / cluster KDE bandwidth4,5 Sample Point 1 Mapping method of Standard 3 Standard size 2 Min Max density Min Max Mean clusters deviation deviation 2‘496 ADKNN-Points 1 2‘496 2‘496 0 4.28 47.03 47.03 47.03 0 2‘496 ADKNN-Clusters 17 2 904 225 10.32 4.26 71.03 29.42 18.18 946 ADKNN-Points 1 946 946 0 2.71 48.87 48.87 48.87 0 946 ADKNN-Clusters 13 2 315 103 12.07 4.26 53.31 25.2 13.56 233 ADKNN-Points 1 233 233 0 92 5.54 5.54 5.54 0 233 ADKNN-Clusters 5 9 119 47 1‘470.6 2.56 10.99 4.29 3.91 1‘0006 ADKNN-Points 1 1‘000 1‘000 0 - 107.47 107.47 107.47 0 1‘0006 ADKNN-Clusters 7 121 216 43 - 59.51 72.5 67.38 4.78 1Number of points and polygon centroids 2With ADKNN-Points, each sample can be considered as a single cluster 3 Density of the sample size (ADKNN-Points) or average of densities calculated for each cluster (ADKNN-Clusters). Densities are expressed in 10-3 ERW/km2 4Average distance (km) to k-th nearest neighbour, k being proportional to the square root of the sample size (ADKNN-Points) or the square root of the cluster size (ADKNN-Clusters) 5 Precision parameter P = 2 6 Random points

2.3.4.1. Pros of the ADKNN-Clusters mapping method

 Is able to show the contamination problem without showing the ERW’s exact locations From Table 6 it is possible to emphasise similar map characteristics than from Table 1: capacity of preserving areas with low contamination (low third quartiles), aptitude to enlighten highly affected areas (low median values, third quartiles much lower than maximum values) and ability to display the general ERW distribution without giving precise locations (high standard

101

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

deviations compared to mean values, high number of non-zero values). These characteristics are even more pronounced with ADKNN-Clusters than with ADKNN-Points (Table 6): third quartiles are 4 times lower (or even null), standard deviations twice larger and maximum values 4.5 times greater.  Provides a close-to-reality representation of data Bandwidths values generated by ADKNN-Clusters are between 1.3 and 1.9 times smaller than the ones produced by ADKNN-Points (Table 7), as ADKNN-Clusters is calculated on the basis of a smaller sample size (cluster versus overall dataset). Smaller bandwidths result in sharper pictures of contamination (Figure 15a-b-c versus Figure 15d-f). The ADKNN-Clusters algorithm is able to capture a large range of cluster sizes (Table 7), shapes and orientations (Figure 15a, Figure 15b and Figure 15c), thus to better represent reality than ADKNN-Points. Similarly, local contamination specificities are displayed with ADKNN-Clusters (Figure 15a) and not with ADKNN-Points (Figure 15d). By zooming to district level (Figure 15c and Figure 15f) this difference in representing reality is even more evident: the area shown in Figure 15f seems to be highly contaminated on a 60-km width, though it is obviously not the case in the field.  Is adaptable to scale If we look successively at Figure 15a, Figure 15b and Figure 15c, adaptability of the ADKNN- Clusters mapping method to change of scale is evident. The number of clusters decreases progressively and slightly with bigger scale (17 clusters at 1:20‘000‘000, 13 clusters at 1:7‘500‘000 and 5 clusters at 1:1‘250‘000) while the size of contamination spots remains uniform, keeping map reading comparable and consistent from one scale to another.  Is unsupervised It has been demonstrated in 2.2.9.1 that the influence of parameters P and RP20 on ADKNN- based maps significantly decreases when the density of processed ERW grows. The independency of the ADKNN approach on precision parameters is made effective with clustering, as densities are much higher (between 2.5 and 16 times) inside the clusters than for overall distributions (Table 7).  Is able to show contamination along country borders, without showing the borders A sensitive aspect of the mapping process concerns frontiers between mine-affected countries. As mentioned in Section 2.3.2.3, many landmines have been deployed during conflicts related to sovereignty over border areas. Some of these borders are still sources of conflicts and cannot be shown on maps, thus mapping of contamination in these areas has to be processed cautiously. With the ADKNN-Points method, which will be deployed along with IMSMANG releases for national mine action authorities, kernel maps may largely spill over neighbouring countries, as shown in Figure 15f. Prima facie, a convenient solution to this issue would be to clip the

20 Precision parameters for adjusting the level of detail of maps

102

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

computed density rasters to the extent of each country. But with heterogeneous pattern distributions, ADKNN-Points may generate very different kernels values either side of frontiers, which would make them discernible. With the ADKNN-Clusters method, clusters either side of frontiers can be caught, thus overflow of kernel maps over neighbouring countries can be avoided keeping borders indiscernible.

103

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

104

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 15: Density rasters in Afghanistan. (a), (b), and (c): ADKNN-Clusters method applied to IMSMANG subsets at different scales; (d) and (f): ADKNN-Points method applied at national level; (e): ADKNN-Clusters method applied to random points

2.3.4.2. Limitations of the ADKNN-Clusters mapping method

 Looses its singularity at the global level Datasets from non-neighbouring countries (Cyprus, Iraq, Tajikistan and Somalia) were processed with ADKNN-Clusters. The algorithm produces as many clusters as there are distinct countries (Figure 16a) and generates similar results as if the four national data repositories had been processed separately with ADKNN-Points. This test puts in evidence a weakness of ADKNN-Clusters: the method is taken over by ADKNN-Points when processing data on non- contiguous extents. In particular, ADKNN-Clusters looses its singularity at the global level. Nevertheless, this method can be of great interest at supra-national level for mapping of datasets on contiguous extents, for example the Asian South-eastern sub-continent (Cambodia, Lao, Thailand and Vietnam) or Central Asia (Afghanistan, Azerbaijan, Georgia, Iran, Iraq and Tajikistan).  Is sensitive to outlier points When combined with the VRC, Wong‘s algorithm is able to detect small clusters and outliers. Nevertheless, calculating the ADKNN of a cluster composed of a single point (like in Figure 16b) is absurd. However, outlier ERW data should not be set aside by kernel maps, and a way has to be found to assign them a bandwidth value. We chose this value as the ADKNN of the entire distribution. The consequence of this is that outliers may be represented by large kernel spots, which may significantly mislead the map consumer. The grey kernel spot in Figure 16b just seems like a numerical artefact in the middle of the map, with no environmental, geographical and cultural singularity. Overlaying it with auxiliary data such as land cover of the Terrestrial Regions of the World (Olson et al. 2001), slopes (Jarvis et al. 2008), temperatures of the WorldClim database (Hijmans et al. 2005) and with ethno-linguistic data does not put in evidence distinct natural, environmental or cultural assemblages: (1) 29.2% of the grey circle corresponds to Central Afghan mountains xeric woodlands and 57.1% of it to Registan-North Pakistan sandy desert; (2) slopes are very disparate and range from 0° to 72.2°, as well as (3) January maximum temperatures that oscillate between -5.1°C and 15.1°C; (4) three different ethnic groups speaking four distinct languages cohabit, the Pasthuns, the Tajiks and the Balochs.

105

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 16: Weaknesses of ADKNN-Clusters mapping method: (a) When applied to data from non-neighbouring countries, ADKNN-Clusters generates one cluster per country and gets similar as ADKNN-Points, and (b) Big pollution in the light grey circle is due to one outlier

 Becomes meaningless if input data are not clustered The ADKNN-Clusters method was applied to a 946-point IMSMANG subset (Figure 15b) and to a 1‘000-point random dataset, on the same area (Figure 15e). Table 7 shows comparative statistics about number of points and KDE bandwidth at cluster level. With clustered IMSMA data, bandwidth values range from 4.26 to 53.31 km (standard deviation = 13.56 km) and the algorithm finds 13 clusters with sizes comprised between 2 and 315 points (standard deviation = 103). With random data, landmines are distributed in 7 clusters with sizes comprised between 121 and 216 ERW (standard deviation = 43) and the range of bandwidth values is much smaller (between 59.51 and 72.5 km, standard deviation = 4.78 km). As a result, the map in Figure 15e is quite uniform and gives the impression that no clustering algorithm has been applied. With random data, ADKNN-Clusters gets meaningless, which may severely limit its applicability to repositories where the landmine distribution is not clustered (e.g. Laos and Cambodia).  Does not provide good computing performances Processing 2‘496 points takes around one hour21. Since dozens of thousands of rows have been registered in IMSMANG, it would be more comfortable for users if performances of the clustering algorithm were improved. This would make it easier to regenerate up-to-date maps as IMSMANG repositories are updated. For doing so, migration of the source code from Python to Visual Basic .Net is recommended.

21 This test was preformed on an Intel® Core ™ i7 CPU Q720 @1.60GHz with 6GB of memory

106

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.3.5. Conclusion

In Section 2.3 we presented ADKNN-Clusters, a novel unsupervised mapping method, based on data clustering and KDE that we developed as an automated ArcGIS function. Adaptable to scale changes, ADKNN-Clusters is able to show the problem of contamination by ERW while preserving data confidentiality. With ADKNN-Clusters, pictures of the contamination are sharper and closer to reality than with other visualisation methods (e.g. choropleth maps and ADKNN-Points). Local contamination areas are also better reproduced and their shapes are respected. The aptitude of ADKNN-Clusters to process data either side of a border without showing the frontier, and to prevent overflow effect of a country over another by reducing kernel bandwidth values, is another of its advantages. ADKNN-Clusters is applicable for a wide range of scales (subcontinent, national and sub-national). It does not require specifying parameters and works for any spatial data distribution. Some limitations were noticed. First, the method looses its singularity if ERW are processed on non- contiguous countries. As a consequence, a global overview of contamination is not an option with ADKNN-Clusters. Second, it is sensitive to outlier records. Third, the method becomes meaningless with non-clustered datasets. Finally, performances of the clustering algorithm could be improved. Based on the discussion of Section 2.2.9, we see ADKNN-Clusters as a suitable visualisation method for two categories of humanitarian demining actors: (1) For users outside the core mine action domain (donors and the general public) to have an overview of the degree of contamination in regions of the world, at sub-continental and national level. (2) For operations officers to decide where to plan demining operations.

107

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.4. Determining the impact

Based on: Mapping Populations at Risk of ERW

Pierre Lacroixa,b, Jonas Herzogc, Daniel Erikssond a University of Geneva, Institute for Environmental Sciences, enviroSPACE Lab., Battelle – Building D, 7 route de Drize, CH-1227 Carouge; b University of Geneva, Forel Institute, 10 route de Suisse, CH-1290 Versoix; c Joint Mine Action Coordination Centre, 111 Palm City, Tripoli, Libya; d Geneva International Centre for Humanitarian Demining, 7bis, avenue de la Paix, P.O. Box 1300, CH-1211 Geneva 1

2.4.1. Abstract

Having precise, available data on recorded explosive remnants of war hazards does not necessarily represent the big picture concerning the contamination distribution in a country. However, when available datasets are evaluated with population-density data, heavy concentrations of ERW hazards are more easily detectable. This article examines a few of the many tasks that can be achieved by analyzing ERW hazard data and by combining it with other information.

2.4.2. Introduction

The Geneva International Centre for Humanitarian Demining is collaborating with the University of Geneva to explore the feasibility of creating worldwide visualisations of the density of ERW contamination, such as anti-personnel landmines and cluster munitions. Using ERW hazard data collected with the Information Management System for Mine Action and other relevant data, the Server for Explosive Remnants of War Information Systems project helps visualise large-scale contamination.

2.4.3. Objectives of the paper

This paper‘s primary objective is to explore the feasibility of visualising at-risk populations through maps that combine ERW hazards and population data. The interpretation of the SERWIS maps can highlight the highest densities of ERW hazards in populated areas. This also allows further tests and analyses to help determine the impacts of hazards on the population. Throughout this paper, we used the following formula: Population at risk of ERW = Presence of population × Presence of hazards. We applied this method to data from Afghanistan provided by the Mine Action Coordination Centre of Afghanistan (MACCA 2010). However, the project is potentially open to all countries that desire to share their information with the mine-action community.

108

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.4.4. Estimating the population density

Population data can be found in the form of global datasets, such as the LandScan™ global population database (ORNL (Oak Ridge National Laboratory) 2008) and the Gridded Population of the World (GPW) (CIESIN (Centre for International Earth Science Information Network) et al. 2005).

2.4.4.1. LandScan Global population database

The LandScan global population data is a common dataset used by numerous governmental and nongovernmental organisations. Based on remote-sensing data, it is regarded as the most accurate and reliable population-distribution model. Various other data are included to estimate the population density, such as information about population movement, census counts, land cover, infrastructure, terrain, etc. LandScan population data is available for the entire world as a grid with a resolution of 1 kilometre, where each pixel represents one kilometre and every cell has a value for the estimated population. Figure 17 and Figure 18 show the LandScan population dataset on global and national levels.

Figure 17: The LandScan global population dataset

109

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 18: LandScan population dataset for Afghanistan

2.4.4.2. The Gridded Population of the World

The GPW dataset is based on census data and provides population data on a global level. It is calculated based on two inputs: administrative boundary data and population estimates associated with those administrative units. Grid cells are assigned a portion of the total population for the administrative unit they fall within, dependent on the proportion of the area of the administrative unit that the grid cell takes up. This allows transforming administrative boundary data from their native units into raster cells, the smallest unit of measurement within raster data, which is a means of representing spatial data. As a consequence of this, the precision of the dataset depends heavily on the availability of data at a high level of precision. This means that countries with many small administrative units (such as municipalities) have more accurate data than those with few high level units (such as provinces). As an example, Afghanistan has an area of more than 600‘000 square kilometres (231‘661 square miles) and is divided into 328 administrative units at the second level (called districts). These districts (Hijmans 2012) have an average area of almost 2‘000 sq. km. (772 sq. mi.) and an estimated 65‘000 inhabitants. The population density is shown in Figure 19.

110

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 19: GPW population density in Afghanistan

Where there is no data available for a given year, the values are estimated by alternate known values from previous years. Using these inputs, estimates for years between 1990 and 2015 were made for all countries. The last edition (v3) of GPW was developed between 2003 and 2005. For each country, the population estimates for the available years were produced based on the 2000 census. Unlike LandScan, GPW does not aim at modelling time-of-day distribution by analyzing various factors (e.g. land cover, elevation and satellite imagery), but it indicates a residential picture of the population over a five- to 10-year period.

2.4.4.3. LandScan and GPW compared

The LandScan and GPWv3 datasets are among the most widely used datasets for representing population data worldwide. Both of them have their advantages and disadvantages (Table 8).

Table 8: Comparison of LandScan to GPW Characteristics LandScan GPW Data from 1990 to 2015, a new Very recent data available (from update will be available in early Version 2008) 2012

111

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Data is updated annually Datasets for every 5 years Update frequency

Available online for free for Availability educational purpose and United Available online for free Nations Humanitarian efforts22 Available in three different formats Format Available as Esri grid (ASCII, BIL and Esri grid) The precision and the reliability of The data are aggregated from many the data depends on the Source different sources and information administrative units and on the authorities that make the census Depending on the administrative High (30 arc seconds, Resolution units, but generally lower (2.5 arc approximately 1 km) minutes, approximately 5 km) All contaminated countries are All contaminated countries are Extent covered covered

Both datasets are complementary: In GPW, each administrative unit is assigned the same value of population, while LandScan will help at a higher resolution (down to 1 km). Moreover, LandScan represents a daily average population count while GPW indicates a residential picture of population over a five- to 10-year period.

2.4.5. Estimating the ERW hazard density

With large amounts of information on recorded AP landmines and other ERW, a way to visualise the respective densities was needed.

22 In the 2008 version of LandScan

112

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 20: Point data for Afghanistan, every marker represents a hazard (point or centre of a polygon). Source: MACCA

In Figure 20, the map becomes unreadable with an increasing amount of data. In some areas, an excess of records renders the map unreadable due to record overlap. Using spatial statistics, estimating contamination densities by conducting a Kernel Density Estimation in ArcGIS™ is possible. The principle of this tool is that for each cell, a density value is calculated that depends on the number of points and their proximity to one another. The closer the points are to one another, the higher the values. Additionally, each point is weighted with a value according to the hazard area. If they are more distant than a pre-defined radius (also called Kernel size or bandwidth), they are not taken into account at all. The values are then coloured from yellow (low values) to red (high values). White areas have no known hazards.

113

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 21: Density map showing estimated contamination in Afghanistan calculated using the ArcGIS Kernel density tool. Source: MACCA

The result is a map showing the densities of the input points (see Figure 21). While large areas have a greater impact than small ones, these are either hazards or, if they are polygons, their centres. Looking at Figure 21, it is easy to distinguish the high concentrations of observed record and other regularities. The KDE approach is used in many other domains, such as crime mapping, estimating the home range of animals, etc. Another advantage of this method is that the resulting density maps are smoothed. This means that geometric shapes (the neighbouring cells within the bandwidth) with estimated contamination values replace the original hazards and their area. Thus, this does not allow for the drawing of conclusions about the exact locations of hazards records, which can be sensitive information. This map‘s output helps to visualise the problem and can also be used for further analyses. This representation gives an overview of the overall hazard distribution, but is unable to identify the exact locations.

2.4.6. Combining hazard density maps with population data

GIS now provides various tool sets and innovative methods to combine different datasets into a new dataset. The methodology described below was used to combine the hazard probability with each of the population datasets.

114

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.4.6.1. Combination with LandScan population data

In order to make the two datasets comparable, the LandScan data (with an original resolution of 1 km.) was processed into a lower resolution. For the estimation measurement of at-risk people, the density map was multiplied by the LandScan population values, cell by cell.

2.4.6.2. Combination with GPW population data

As with the LandScan population data, the two input datasets had to be harmonised. In this case, the original hazards were added by district, while including their areas. Having population data on the same level of detail, namely on district-level, made it possible to multiply the values district by district and come up with a reasonable result. With this methodology, creating density maps was not necessary—only vector data (points and polygons) were processed.

2.4.7. Results and discussion

Using data from the two different sources, the maps (Figure 22 and Figure 23) presented here suggest where most people are at risk.

Figure 22: Populations at risk of ERW hazards in Afghanistan, calculated with the LandScan dataset

115

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 23: Populations at risk of mines in Afghanistan, calculated with the Gridded Population of the World (GPW) dataset, at second administrative unit level

Figure 22 shows detailed information about populations at risk, at national level and with a resolution of several square kilometres. It is possible to have a rapid overview of provinces presenting high, densely populated areas coupled with important hazard probability, such as in the Kabul province. Further investigations regarding ways to represent population and ERW density in the same map should be targeted. While this did not fall within the scope of this study, there needs to be a way to easily distinguish the contribution of each one of them, with elements such as two-dimensional colour ramps. Figure 23 shows the result of the combination with GPW data. For every district, one value is estimated to represent the overall situation in the district. The information is reduced to the district level (there are 328 Afghan districts), which produces the effect of biasing the interpretation of the results: People in entire districts seem to be at no risk, despite the reality that they might be. The symbology applied to the output data may largely influence the map interpretation. The choice to multiply the input datasets to model the exposure leads to a high percentage of zero values. Zero values occur if there is either no population or no contamination, thus it is easily readable and interpretable. Moreover, both maps give a first overview of the risk situation at a national level. In addition to harmonising the data resolution to be combined, the methods chosen can be applied to any

116

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

contaminated country in the world and simply automated with standard ArcGIS desktop tools. Furthermore, computing performances proved effective: Interpolating the 6,000 test hazards on the entire Afghan area took only minutes. The comparison of the two resulting maps suggests that both input population datasets can be used depending on the scale at which the population exposure is displayed. Moreover, they could be combined with other layers, such as infrastructure, land use, resources, etc.

2.4.8. Conclusion

This work is a first approach for modelling and estimating people at risk of ERW hazards. It proposes a new research approach including spatial analysis, applicable for entire countries, with results at the district level or lower. It is possible to indicate in which districts or in which populated areas ERW hazards are present and are threatening people, and therefore probably affecting their daily lives. Also, the model will be influenced by the assumption that more hazards are recorded where population density is higher. A further step will aim to incorporate the confidence of each recorded ERW area, in order to more accurately reflect the reality on the ground. For instance, minefields are usually associated with a higher confidence than dangerous areas. Hence, this should be included in the hazard-density estimation by readjusting the methodology: the Kernel radius should be smaller for the former than for the latter. Future project steps will evaluate the risk on populations (taking into account the population vulnerability), the impact on populations (with socio-economical considerations) and the impact on infrastructures. Datasets other than LandScan and GPW, and other GIS and statistical methods could be used and should be investigated.

117

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

2.5. Highlights of Chapter 2

Mine action data are collected and maintained at national level in non-spatial repositories. Databases are characterised by a large amount of data and a heterogeneous degree of completeness. ERW data are critically heterogeneous in the type, quality, positional accuracy, reliability, and spatial distribution pattern. Spatial distribution patterns are however not random and show a low to high degree of clustering. In certain countries, landmines are mostly concentrated along country borders, roads, or in dense populated areas. This high heterogeneity is one of the main challenges that have to be faced if we want to visualise ERW data. So far, only few attempts have been made to visualise such data. If hazard mapping is widespread, especially in the form of maps or web map services on global data platforms, only few maps showing contamination by ERW and their impacts on populations have been published on the internet. Very few of them are of good quality, interactive and up-to-date. To fill this gap, we discussed in Chapter 2 of this thesis how cartography could meet the needs of humanitarian demining actors for cartography. To that end, we defined the requirements for ERW visualisation of different user groups and at different scales. We evaluated seven visualisation methods and matched each of them against the needs of any of four categories of users at the scale at which they have to operate. Three of the seven visualisation methods were selected for their capacity to obfuscate the data. These methods are: ADKNN-Points, ADKNN-Polygons and ADKNN-Clusters. They are based on KDE. With these methods, ERW areas are extracted from national repositories in the form of points or polygons. Polygons are either replaced by their centroid, or filled up with points before being applied a KDE. When designing KDE-based maps, selecting the KDE bandwidth has a significant effect on visualisation, fuzziness and obfuscation: the larger the bandwidth, the smaller is the kernel peak, the more skewed is the histogram of the map and the smoother is the kernel surface. In order to integrate this issue and to best fit the ERW geospatial distribution patterns, traditional KDE was extended by customising the KDE bandwidth as follows: it is calculated as the average distance to k-th nearest neighbour (ADKNN), k being proportional to the square root of the number of ERW in the distribution. In addition to this, we also provide users with a way of adjusting the level of detail of the kernel map through user-defined parameters, to let them keep control over the representation of contamination that they want to display. These three visualisation methods have strengths and weaknesses. They are summarised below, based on the research presented in Chapter 2.  Strengths of the three KDE-based mapping methods The strengths of the three KDE-based visualisation methods are listed hereafter. (1) The three methods are able to show the problem of contamination by ERW in a comprehensive way. (2) They have the capacity to preserve data confidentiality at low scales, e.g. below 1:50‘000, in particular along country borders, where the information is in certain cases very sensitive. (3) The

118

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

maps produced with these three methods have the capacity to fit highly heterogeneous distribution patterns, especially when clustering points and scattering polygons with points. (4) The three methods are less sensitive to the MAUP than other well-known cartographic visualization methods (e.g. one-to-one dot maps, choropleth maps, cartograms). (5) Unlike other maps (e.g. cartograms, graduated or proportional symbols), KDE maps do not get overloaded in case of wide range of contamination values or when administrative units become dense. (6) ADKNN-Clusters is so to say unsupervised because it consists of grouping the ERW data in clusters before applying the customised KDE to them. (7) ADKNN-Clusters is also able to show contamination either side of a frontier, without showing the frontier. (8) The three visualisation methods can be applied to a wide range of scales and to a large spectrum of users, as shown in Table 9. Note that information provided in this table consists of recommendations and guidelines, and can be adjusted or extended to specific users‘ needs. For this reason, we have deliberately separated the columns in this table with dashed lines rather than with solid lines.

Table 9: Recommendations: which cartographic visualization methods are most suitable for which category of humanitarian demining actors? Municipality Target audience Global level1 Sub-continental level National level3 Sub-national level level4 Users outside the core ADKNN-Polygons ADKNN-Polygons mine action domain ADKNN-Clusters2 Directors of national ADKNN-Points ADKNN-Points mine action authorities Choropleth maps Choropleth maps ADKNN-Points ADKNN-Points One-to-one Operations officers ADKNN-Clusters (ADKNN-Clusters) dot maps Database One-to-one administrators dot maps 1Worldwide to sub-national scales 2The mapped area should however be composed of contiguous areas 3ADKNN-Points is planned to be implemented as an IMSMANG cartographic function 41:50‘000 to 1:5‘000  Weaknesses of the three KDE-based mapping methods The three KDE-based visualisation methods expose weaknesses. ADKNN-Points cannot be implemented as a single worldwide map. This method is more suitable at national level. ADKNN-Polygons handles polygons, which makes it more accurate but limits the number of potential candidate repositories. Both methods require specifying user-defined parameters. The third one, ADKNN-Clusters, is unsupervised and adaptable to scale change but still depends on the degree of clustering of the input patterns and on the robustness of the clustering algorithm. When processing data on non-contiguous extents, ADKNN-Clusters calculates one cluster for each country and results are similar to the ones provided by the ADKNN-Points method. ADKNN-Clusters method also presents weaknesses at the global and local scale, and computing

119

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

performances needs improving. Finally, the three ADKNN-based methods are sensitive to outliers. The main lesson learnt from the research presented in this chapter of the thesis is that it is not possible to find one universal mapping method, applicable to any dataset, at any scale and useful to any humanitarian demining actor. It appears that clustering algorithms and cartographic visualization methods are perfectible and do not address at the same time all requirements stated in this chapter of the thesis. The degree of understanding and acceptance of maps is also dependent on the visual effect that is produced, and on who the map consumer is. The solution for him/her is probably to use a combination of several of these cartographic visualization methods. The results of our research, then, can be quite frustrating. At the same time, however, it is promising as GICHD has expressed the wish to implement one of the visualisation methods in the existing IMSMA technology. As a matter of fact, it was decided to integrate ADKNN-Points in the IMSMANG cartographic module as raster generator functionality (Figure 24). This customised, flexible, easy-to-use and performing23 add-on will allow users to generate maps in a standardised way and in a well-known format (e.g. TIFF). The raster generator will make it easier for mine-affected countries to disseminate maps inside and outside the mine action community. For several reasons that will be exposed in Chapter 4, dissemination through web services (WMS/WCS) is recommended.

23 Processing 6‘000 points on a 600‘000 km2 extent with a 600 m output cell size takes minutes. This test was preformed on an Intel® Core ™ i7 CPU Q720 @1.60GHz with 6GB of memory.

120

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

Figure 24 : Raster generator GUI

Integration of the Raster generator into the SERWIS platform will be described more in detail in Section 4.5. The choice of the colour ramp also is crucial and has a large influence on the interpretation of maps. There is a serious risk to over- or under-estimate the picture of contamination of certain regions of the world. In the case of ERW contamination maps, two solutions were envisioned to reduce this risk:  Colour scheme

121

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

To ensure that all regions are measured on a similar scale, a universal colour ramp was developed, and the natural logarithm of the kernel density was mapped rather than the density itself. To preserve the smooth nature and the fuzzy effect of the density rasters, the colour ramp was made continuous. To attract map consumers‘ attention on the highly mine- affected regions, the main colours of the colour ramp were: white (no contamination or very low), yellow (low), orange (medium) and red (high). Transparency was applied to let room for overlaying with other datasets. To guarantee non-disclosure and data confidentiality, legends were made ordinal and do not show figures about contamination. Instead, the terms ―low‖, ―medium‖ and ―high‖ are used.  Metadata It is envisaged to integrate, as a further expansion of the raster generator, a function that would automatically export the information entered by users through the raster generator GUI. These metadata would inform the map reader on how and when the raster has been produced. With this metadata generator, we want to prevent the map reader from comparing maps based on datasets with different characteristics, e.g. a map showing ongoing hazards as of 2012 and a map showing hazards that have been completed in 2011. Another promising perspective of our research is that the three KDE-based contamination density maps (ADKNN-Points, ADKNN-Polygons and ADKNN-Clusters) are composed of a single layer. Combining them with other layers (e.g. topography, urban areas, strategic infrastructure, land cover, development areas, population densities and others) is then easy and may help decision-makers to prioritise surveys and to determine socio-economic impacts of hazards. In particular, combining raster density maps with population data makes it is possible to highlight areas where populations are most at risk of ERW. The model that was introduced in Section 2.4 assumes that: Populations at risk of ERW = Presence of population × Presence of hazards. This approach can be applied at a large range of scales (from global to sub-national) and resolutions (1 km with LandScan data versus the administrative unit with GPW). The choice of LandScan and GPW as input data sets for this model is motivated by the fact that the two datasets are complementary: the former represents a daily average population count while the latter indicates a residential picture of population over a five- to 10-year period. Another reason for the choice of these two datasets is that they can be downloaded in well-known formats and readily extracted on the desired extent before combination with density rasters in an automated process. The outputs of our model are rasters showing populations at risk of ERW. These rasters are meant to be published on the same platform than the ERW contamination density maps (SERWIS). As for contamination maps, WMS/WCS are recommended. A weakness of our model, however, is that it is not possible to distinguish the contribution of populations and that of ERW contamination. White pixels represent areas with either no ERW contamination or null population density. Our research also deserves deeper analysis if we want to assess and to display population vulnerability. To our knowledge no peer-reviewed publication has been done on vulnerability to ERW. The concept of

122

Chapter 2: To what extent can GIS improve visualisation of contamination and its impact on population?

vulnerability has however been widely analysed in literature, especially in environmental sciences. Vulnerability is defined by Blaikie et al. (p. 9) as the capacity of a person or a group to ―anticipate, cope with, resist and recover from the impact‖ of an event. The Intergovernmental Panel on Climate Change (IPCC) characterises vulnerability to climate change as a function of exposure, sensitivity and adaptive capacity (IPCC 2007). Eriksson (2006, p. 4) points out that vulnerability is also accepted as a spatially and socially dependent characteristic, at social levels reaching from individual to a cultural community. Many references to GIS-based vulnerability assessment can be found in environmental sciences related literature, e.g. Navulur and Engel (1998: groundwater vulnerability), Sanyal and Lu (2005: flood vulnerability) and Giuliani and Peduzzi (2011: global data risk platform). As for what concerns this PhD thesis, the research that we conducted has combined ERW density with population density to show the impact (blue boxes in Figure 25). Further research direction should then consider integrating in the analysis factors such as age, gender, demining capacity, mine risk education, and other socio-economic variables, and combining them with at-risk population data to produce rasters of population vulnerability (red boxes in Figure 25).

Population density Presence of ERW

Populations at risk of Adaptive capacity ERW

Population vulnerability

Figure 25 : Assessing population vulnerability to ERW. The blue boxes represent the current research. The red boxes are possible further research directions

123

Chapter 3. What are the contributions and limits of GIS for improving decision-making in mine action?

Contributing research papers  Lacroix P., Escobar R. (2012). 5D: a GIS-based approach for Determining and Displaying a Degree of operational Difficulty of Demining. The Journal of ERW and Mine Action, 16(3). Available from: http://maic.jmu.edu/journal/16.3/rd/lacroix.htm  Lacroix P., Santiago H., Ray N. MASCOT: Multi-criteria Analytical SCOring Tool for ArcGIS Desktop. Submitted to the International Journal of Information Technology and Decision Making  Lacroix P., De Roulet P., Escobar R., Cottray O. (2013). NAMA: A GIS-based Network Analysis Approach for Mine Action, Accepted for publication by the Journal of ERW and Mine Action

124

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.1. Introduction

In chapter 2 of this thesis, we made an attempt to answer the question how ERW contamination maps can help mine action stakeholders make informed decisions, for example to set clearance priorities and address humanitarian, economic and financial issues. Seven visualisation methods were applied to ERW data and matched against the needs of four categories of stakeholders (database administrators, operations officers, directors of national mine action authorities and donors) as well as four levels of scale (global, national, sub-national and local). The proposed visualisation methods have in common to generate single layer contamination maps. This provides a smart visual support to users who want to make informed decisions, but we hope to show that GIS support for mine action can go way beyond ERW visualisation to support decision-making. In this chapter, we assume that bringing other data than ERW data in the analysis and combining them through GIS-based models such as multi-criteria analysis tools can significantly improve decision-making in the humanitarian demining field. We also consider that environmental, geographic and socio-economic conditions, as well as human activity, have substantial influence on demining activities. For example, the use of demining tools may be limited by dense or high vegetation, ground softness, high degrees of slope, huge temperature gradients, presence of wetlands and distance to paved roads (GICHD 2009b). On the other hand, human activity nearby hazardous areas may facilitate demining activities, as well as the development of infrastructures (e.g. bridges, tunnels, modern roads) improving the access to contaminated areas. Most of these environmental, geographic, human and socio-economic factors are common geospatial data that can be found on the Internet for the entire world. Such data were presented in the introduction as ―auxiliary data‖, in contrast with data from the core mine action domain, which are mostly stored in IMSMANG. The interest for combining IMSMANG data with auxiliary data has been shown by Lacroix et al. (2011) who overlaid landmine contamination maps with population density rasters in Afghanistan to highlight where most populations are at risk of ERW. This interest for combining IMSMA with auxiliary data is corroborated by the high dual spatial correlation that was calculated between them (see Section 2.3.2.3). With this in mind, we explore in Chapter 3 of this thesis novel ways of combining diverse geospatial indicators with IMSMA data through GIS-based models, in particular decision-support tools. Such tools, commonly known as ―Spatial Decision Support Systems‖ (SDSSs) find innumerable applications spanning many fields, including agriculture, business, energy, fire protection, land suitability analysis, transportation, utility and water resource management. SDSSs have been praised by numerous scientists for their capacity of processing many datasets at the same time (Yildirim and Yomralioglu 2011),

125

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

handling with complex structures (Goodchild and Kemp 1990), providing spatial statistics functions (Zamorano et al. 2008) at the desired geographic scale (Knezic and Mladineo 2006), making decision- making visual and interactive (Goodchild and Palladino 1995), reducing decision-time and providing consistency and accuracy to decision-making (Wellar 1990, Crossland et al. 1995, and others). The work that we present in Chapter 3 is likely to bring new light to the mine action community, as few studies have been published on the use of SDSSs in humanitarian demining. One of the few available references is Knezic and Mladineo (2006), who demonstrated the benefits of a hierarchic GIS-based DSS (Decision-Support System) in the setting of humanitarian demining priorities and in the financial distribution process. More recently, Mladineo (2012)) developed a GIS-based multi-criteria analysis Web application for setting priorities in humanitarian demining. With this application, users can combine layers through predefined scenarios at the county24 or municipality level. Output results are presented in an intuitive way using simple ranking (PROMETHEE method: See Brans et al. (1986)). The application clearly highlights the potential of multi-criteria analysis for the setting of priorities in the mine action domain. However, three elements could curb its deployment in countries with low GIS knowledge or financial resources: (1) it takes an ArcGIS Server administrator to prepare data, study areas and scenarios, (2) software requirement is cost-demanding (one ArcGIS server licence), (3) the weighting and scoring processes are not accessible to end-users, which reduces substantially the flexibility of the application. In Chapter 3 of this thesis, we introduce three GIS-based analytical approaches and implement them as GIS tools. These tools are then applied to a set of mine action and auxiliary data. With regard to the requirements mentioned in the previous paragraph, we put effort on developing tools that are simple to use, flexible and customised to users‘ needs. We also integrate a number of difficulties and constraints that are outlined below:  To facilitate their use in dozens of mine-affected countries, the tools developed should be integrated with existing platforms and take into account financial capacity of countries. The ArcGIS desktop is a good candidate for it as mentioned in Chapter 1.  The tools should be participatory in order to strengthen cross-sector collaboration between sub- national governments, sector ministries, statutory bodies and mine action authorities, as recommended by the GICHD (2009a). Likewise, outputs of the tools should be understandable, e.g. readily mappable.  Given the lack of computer performance resources in some countries, and the large amount of features to process, good computing performances are systematically sought.  The tools should include the multi-scale approach (see Table 9), with a focus on the national and sub-national levels for priority-setting (Mülli and Paterson 2012). Chapter 3 is structured as follows. Section 3.2 is based on a paper entitled ―5D: a GIS-based approach for Determining and Displaying a Degree of operational Difficulty of Demining‖ that was published in the

24 "County" and not "country"

126

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Journal of ERW and Mine Action. In this paper, we present an analytical method for assessing and visualising in the form of a map, a degree of operational clearance difficulty. Realistic terrain raster indicators are weighted and summed up into a new raster. This raster is classified into four ordinal categories of demining difficulty (―low‖; ―medium‖; ―high‖ and ―extreme‖), from which macro-statistics can be computed. On this basis, it is possible to determine the demining capacity of a country and the surface of land that can be cleared in a given area, with a given demining technique and a given level of difficulty. In Section 3.3, we present an analytical method based on the use of network transportation analysis GIS tools in mine action. The model is called ―NAMA‖: Network Analysis for Mine Action and was accepted for publication by the Journal of ERW and Mine Action. The case study that was conducted is mainly focusing on assistance to victims. More broadly, this type of analysis has potential for other applications provided that input datasets be refined and analysis extended to other case studies, as will be discussed below. In section 3.4 we present MASCOT (Multi-criteria Analytical SCOring Tool), an SDSS based on spatial analysis. With MASCOT, input features (typically, landmines) receive a score in function of their Euclidian distance to real-world scoring objects (e.g. distance to roads, schools and medical centres, cropland surface etc.). A typical application of MASCOT is the setting of clearance priorities. A paper on MASCOT was submitted to the International Journal of Information Technology and Decision-Making. Throughout Chapter 3, we demonstrate that the overarching benefit of using 5D, MASCOT and network analysis GIS methods over non-spatial approaches is to reduce decision-time and costs and beyond, to save more lives. The main contributions of the tools to the mine action community are summarised and discussed in Section 3.5, as well as challenges remaining for the future. In particular we discuss in 3.5 the conditions that could make easier for the mine action community to adopt these tools as standards in their everyday work. Even though the tools and methodologies presented in Chapter 3 of this thesis were developed in the context of mine action, effort was constantly put on making them as flexible and generic as possible, as a bridge between humanitarian demining actors and a broad community of scientists and experts operating in various disciplines and at various geographical scales.

127

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.2. Choosing the Right Technique

Based on: 5D: a GIS-based approach for Determining and Displaying a Degree of operational Difficulty of Demining

Pierre Lacroixa,b, Rocío Escobara a University of Geneva, Institute for Environmental Sciences, enviroSPACE Lab., Battelle – Building D, 7 route de Drize, CH-1227 Carouge; b University of Geneva, Forel Institute, 10 route de Suisse, CH-1290 Versoix

3.2.1. Abstract

Clearance operations are highly dependent on environmental, geographic and socioeconomic conditions. These conditions make demining easier, more difficult or nearly impossible. This article proposes an analytical method called 5D (Determining and Displaying a Degree of Operational Difficulty of Demining), which classifies degrees of difficulty as low, medium, high or extreme. Different factors such as land cover and slope, distance to points of interest (POI), distance to roads, hydrology and potentially others, are combined on an output map. On this basis, macro statistics can be computed for each degree of difficulty and provided to decision-makers and operators. The model is applicable to any country or any region, at the national and sub-national scales and for any demining method.

3.2.2. Introduction

The Geneva International Centre for Humanitarian Demining is collaborating with the University of Geneva to explore the feasibility of displaying the impact of explosive remnants of war in contaminated countries through maps, without revealing the ERW‘s exact locations. This project, Server for Explosive Remnants of War Information Systems, also aims to develop Geographical Information System tools and methods to identify where populations are most at risk. In addition, SERWIS endeavours to determine and display the degree of operational difficulty of demining (5D) on account of realistic and measurable terrain criteria, such as land cover, slope, distance to sensitive points of interest, distance to roads, hydrology, etc. By combining such geospatial datasets into a multi-criteria process at the macro level, this project is meant to refine the evaluation of a country or region‘s demining capacity and help improve demining efficiency. Results provided by the model can act as a good starting point for operational teams that wish to prepare their intervention in the field. Decision-makers can use the model for determining the order in which contaminated areas are to be cleared and which tools should be used.

128

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.2.3. Objectives

Thanks to the human, financial and technological support of international organisations, an area of 52 km2 was cleared in Mozambique between 2002 and 2007, using 15 demining machines. Since 2005, the number of international collaborators and donors has declined, which has decreased Mozambique‘s demining capacity. In late 2008, the overall mine-affected surface remained at an estimated 10 km2, while the demining capacity was estimated at 2 km2 per year. According to these figures, clearance of all mine- affected areas would take approximately five years. This raises a number of challenges. How can this duration be reduced? Which method (mechanical, dog detection or manual) would be most suitable for a given area, and what would be the level of operational difficulty for a given type of machine? As a hypothesis for our model, we assume that demining is strongly dependent on geographic, environmental and socioeconomic conditions. Some of them, such as severe gradients and dense and/or high vegetation, may limit the use of certain demining tools. For example, hill climbing capacity of demining machines is limited to a certain degree of slope. Tiller performance is reduced among dense vegetation and larger tree trunks and is highly dependent on ground softness, rock content and distance to paved roads. Human activity may also influence use of clearance machines. For example, human activity may facilitate mechanical demining, such as the development of roads and bridges providing better access to hazardous areas. When using animal detection methods, complicating factors include terrain, humidity, slope and scent contamination. All of these factors are also likely to affect the degree of difficulty in employing manual clearance methods, although to a lesser extent. Geographical data that can act as direct or indirect indicator of the degree of difficulty are available for most of these factors. In this paper we will focus on mechanical demining, but does not prevent a future focus on other tools or methods. For each tool, developing a model of operational difficulty requires involving both geographers and experts on the tool in question. This enables the identification of appropriate layers of geographical data and the individual role of those layers in the model. For instance, a geographic layer on the ferro-magnetic qualities of the soil might be a good input into a model indicating the difficulty of using metal detectors, but that same layer is likely not useful when estimating the difficulty to use animal detection. Only an expert on manual demining can determine which layers that a geographer proposes are relevant for manual demining. These models are also likely to depend on the local environment. The factors that make manual demining difficult in one country, is likely not exactly the same in another country. The primary objective of this article is to present an analytical method – a map – for the evaluation and visualisation of the degree of operational difficulty for demining contaminated areas. By weighting various datasets, a new dataset is created and classified into four ordinal categories of demining difficulty: low, medium, high and extreme. From this dataset, macro statistics can be obtained and used in a first step. This first step aims to determine the percentage of land that may be cleared in a region or a country, with a given technique and a specific level of operational difficulty. The percentage of surface deemed

129

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

extreme to demine is also estimated. In a second step, the interpretation of information regarding operational difficulty may contribute to improving decision-making to better target clearance operations in the field. This method is applicable for demining with machines, animals or human beings. A model was developed in ArcGIS, inputted with datasets obtained from different sources and applied to the entirety of Mozambique. This case study focuses on mechanical demining, on the basis of a fictive machine with medium class characteristics (length with flail approximately 4.7 m; weight approximately 8T; working capacity approximately 860 m2/hr in topsoil, 900 m2/hr in sand, 840 m2/hr in gravel) commonly used in many countries. The model does not aim to estimate financial cost, hence the use of the term ―operational difficulty‖. A cost assessment would require data collection and analysis on a local level, while the 5D model holds national and regional relevance. For the same reason, the model does not attempt to calculate physical risk.

3.2.4. Inputs

The model contains seven input layers, which can be found on the Internet in the form of free global datasets. These layers include land cover, slope, points of interest, roads, rivers, lakes and national boundaries. These datasets are described below.

3.2.4.1. GlobCover database

GlobCover (ESA 2010) is a global land cover map available for two periods, December 2004 to June 2006 and January to December 2009. Data is missing for only 1% of total land area. GlobCover has been used in many fields of work (e.g., crop mapping, assessment of global forest cover and estimations of biomass burning emissions) and is easy to apply to a country like Mozambique. In the present case, this dataset was used to identify human activity such as farming and urban settlement. GlobCover is freely available online for any non-commercial use at a 300-m resolution (Figure 26 and Figure 27) in a raster format. Each pixel represents a 300-m x 300-m cell and holds a value indicating the category of land cover found at the position where it is located (see Figure 26). For instance, category 14 corresponds to rain-fed croplands, category 140 to sparse vegetation and category 200 to bare areas (Figure 27 and Figure 28). The data is in Tagged Image File Format (TIFF), and the spatial reference is the World Geodetic System 1984 (WGS 1984).

130

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 26: The GlobCover dataset over Africa. Data source: European Space Agency GlobCover project, led by MEDIAS-France

131

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 27: GlobCover dataset over Mozambique

3.2.4.2. The slope dataset

The slope dataset (Figure 28) was obtained from the digital elevation model (DEM) provided by the NASA Shuttle Radar Topography Mission (SRTM) (Jarvis et al. 2008). The DEM data is available in

132

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

raster format, with a 3 arcsecond resolution (90 m approximately), where each pixel contains an elevation value. The DEM can be obtained online in the form of a 5°x5° tile mosaic (1° is approximately 110 km), and its use is restricted to non-commercial redistribution. It is provided in the WGS 1984 coordinate system. For an easier download, using the ―Topo View‖ interface is recommended. Slopes are derived from the DEM. Each pixel contains a slope value in degrees or percentages.

Figure 28: The slope dataset for Mozambique

133

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.2.4.3. OpenStreetMap

Composed of different datasets – infrastructure, water, forest cover, points of interest, administrative boundaries – this database, OpenStreetMap, is distributed under an open content license (OpenStreetMap contributors 2012). Data is available at the global level in vector format and in WGS 1984. It can be downloaded by country. It was developed on the basis of government and commercial data sources and benefited from the contribution of volunteers worldwide. From this database, the case study on Mozambique uses the POI and roads layers. The POI layer stores information about the location of different features such as airports, train stations, schools, hospitals, post offices, shops, telephone boxes, car parks, etc. In the mine action framework, POI are likely to restrict demining activities, since they represent crowded locations or areas frequented by civilians. The roads layer is found as a line shapefile and contains various categories of roads, from footway to primary roads. Unlike POI, roads are likely to facilitate activities, since they increase the access of demining resources to hazardous areas.

3.2.4.4. HydroSHEDS

HydroSHEDS data (Lehner et al. 2008) is a hydrological dataset derived from the Shuttle Radar Topography Mission. This dataset includes vector and raster data such as river networks, watershed boundaries, drainage directions and flow accumulation. The HydroSHEDS dataset covers almost the entire globe, but it requires a manual download region by region. It can be used non-commercially. For this case study, the river network was used, provided in the form of river lines stored in shapefiles. The data resolution is 15 arcseconds, approximately 500 m.

134

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.2.4.5. GLWD

The Global Lakes and Wetlands Database (GLWD) (Lehner and Döll 2004) was developed on the basis of seven digital maps and attribute datasets for lakes and wetlands. The Conservation Science Programme of the World Wildlife Fund (WWF) publishes it globally. Three different datasets can be used, depending on the level of detail required: large lakes and reservoirs (GLWD-1), smaller water bodies (GLWD-2) and wetlands (GLWD-3). For this case study, a combination of level one and level two was used to include lakes with an area > 50 km2, reservoirs with a storage capacity > 0.5 cu km and smaller water bodies with a surface > 0.1 km2. All these datasets are provided in vector format (polygons) and for typical scales of use ranging from 1:1,000,000 to 1:3,000,000. The GLWD can be used for non- commercial, scientific, conservation and educational purposes. All the databases presented above are available at the global level for free in high resolution in the WGS 1984 coordinate system and with a low percentage of missing values (less than 2%). Formats may vary from one dataset to another, but they are all well-known formats (e.g., shapefile, TIFF, etc.), readable by many GIS. Table 10 summarises the main characteristics of these databases.

Table 10: Main characteristics of the input datasets Slope database Characteristics GlobCover OpenStreetMap HydroSHEDS GLWD CIAT_CGIAR Version V2.2 (2009) V4 (2008) - - - Update ~ 3 years ~ 2 years Continuously - - frequency Available online for Available online for Available online for Available online non-commercial Available online for Availability any non-commercial non-commercial for non- scientific, free use purposes commercial use conservation and educational purposes ArcInfo ASCII and Format TIFF Shapefile Line shapefile Polygon shapefile GeoTiff Derived from the Hydrological Developed on the basis Developed on the Medium Resolution Derived from the dataset derived of government and basis of seven digital Imaging digital elevation model from the SRTM. commercial data maps and attribute Spectrometer (DEM) provided by Developed by Source sources and the datasets by the (MERIS) on board the NASA Shuttle WWF‘s contribution of University of Kassel the European Spatial Radar Topography Conservation volunteers around the (Germany) and Agency‘s Envisat Mission (SRTM) Science world WWF platform Programme For typical scales of 15 arc seconds 3 arc seconds use from Resolution 300m (Approximately (Approximately 90m) 1:1‘000‘000 and 500m) 1:3‘000‘000 All contaminated All contaminated All contaminated All contaminated countries are All contaminated countries are covered. countries are covered. Extent countries are covered. Can be countries are Can be downloaded by Can be downloaded by covered downloaded covered tiles of 5° x 5° country region by region

135

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.2.5. The Expert Model

As shown in Figure 29, the model is composed of (1) input data, (2) tools, (3) outputs and (4) parameters.

Figure 29: Overview of the model ―Operational difficulty of demining‖. A scalable PDF is available from: http://maic.jmu.edu/journal/16.3/rd/lacroix.htm

Input data include Mozambique‘s administrative limits and the six layers described above: land cover, slope, POI, roads, rivers and lakes. A blue oval symbolises each input data in Figure 29. Orange rectangles represent the model tools in Figure 29. Each rectangle corresponds to a particular step in the model workflow, e.g. extraction on a given area, conversion from vector to raster, raster reclassification, weighting and generation of the final map. Input data are first extracted on the entirety of Mozambique. A conversion tool is then used to transform the four input vector layers (POI, roads, rivers and lakes) to raster layers for further cell-by-cell analysis. During this conversion, a 200-m resolution is applied to recognise the original data precision (Table 10) while keeping the model performing at macro scale. Given that they represent quantitative or qualitative factors not in the same units, the six raster layers need placement on a similar ordinal scale. For this reason, they are reclassified to four categories that are meant to represent the four degrees of operational difficulty (Table 11). To do this, each pixel is assigned a value from 0 to 3 (Table 12). The reclassified layers are weighted and combined to a new ―Operational Difficulty‖ raster. Weights are expressed in percentages (e.g., 20% or 30%: see Table 13). The higher the weight, the higher the influence of the layer is on the degree of operational difficulty.

Table 11: Degree of operational difficulty of demining Degree of operational Category difficulty of demining 3 Low

136

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

2 Medium 1 High 0 Extreme

Table 12: Classification of the input layers in four categories of operational difficulty Description Degree of difficulty GlobCover 11 Post-flooding or irrigated croplands (or aquatic) High 1 14 Rain-fed croplands High 1 20 Mosaic cropland (50-70%) / vegetation (grassland/shrubland/forest) (20-50%) Medium 2 30 Mosaic vegetation (grassland/shrubland/forest) (50-70%) / cropland (20-50%) High 1 40 Closed to open (>15%) broadleaved evergreen or semi-deciduous forest (>5m) Extreme 0 50 Closed (>40%) broadleaved deciduous forest (>5m) Extreme 0 60 Open (15-40%) broadleaved deciduous forest/woodland (>5m) Extreme 0 70 Closed (>40%) needleleaved evergreen forest (>5m) Extreme 0 90 Open (15-40%) needleleaved deciduous or evergreen forest (>5m) Extreme 0 100 Closed to open (>15%) mixed broadleaved and needleleaved forest (>5m) Extreme 0 110 Mosaic forest or shrubland (50-70%) / grassland (20-50%) High 1 120 Mosaic grassland (50-70%) / forest or shrubland (20-50%) Medium 2 Closed to open (>15%) (broadleaved or needleleaved, evergreen or deciduous) 130 Medium 2 shrubland (<5m) Closed to open (>15%) herbaceous vegetation (grassland, savannas or 140 Low 3 lichens/mosses) 150 Sparse (<15%) vegetation Low 3 Closed to open (>15%) broadleaved forest regularly flooded (semi-permanently or 160 Extreme 0 temporarily) - Fresh or brackish water Closed (>40%) broadleaved forest or shrubland permanently flooded - Saline or 170 Extreme 0 brackish water Closed to open (>15%) grassland or woody vegetation on regularly flooded or 180 Extreme 0 waterlogged soil - Fresh, brackish or saline water 190 Artificial surfaces and associated areas (Urban areas >50%) Extreme 0 200 Bare areas Low 3 210 Water bodies Extreme 0 220 Permanent snow and ice Extreme 0 230 No data (burnt areas, clouds, …) Extreme 0 Slope 0° - 30° Low 3 30° - 35° High 1 > 35° Extreme 0 Roads Sites located < 1km away from a road Low 3 Sites located > 1 km away from a road High 1 Points of interest (POI) POI Extreme 0 Sites not considered as POI Low 3 Rivers Inside the river Extreme 0 Land Low 3 Lakes Inside lakes Extreme 0 Land Low 3

137

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Table 13: Weighting of the input layers. Weights that are provided in this table are fictive and shall not reflect reality Layer1 Global level1 Land cover 30% Slope 30% Roads 20% Points of interest 20% 1Lakes and rivers do not appear in this table as their influence on the model is binary: demining with machines is impossible inside water and possible outside

Outputs of the model are the green ovals in Figure 29 and correspond to data generated by the execution of model tools, including the final map on the extreme right of the model. The final map (Figure 30) is generated by reclassifying the ―Operational Difficulty‖ raster on a scale ranging from 0 to 3 and is composed of 200-m x 200-m pixels, where each is assigned a value representing an ordinal degree of operational difficulty of demining: low, medium, high or extreme. Areas where demining is set as extreme, hold the value 0 and are coloured in dark brown (e.g., lakes, rivers, dense vegetation, high degree of slope, etc.). Areas where demining is considered very difficult are coloured in brown and assigned the value 1. A value of 2 indicates medium difficulty (in orange) and a value of 3 indicates low difficulty (e.g., buffers around roads in yellow). In Figure 29, model parameters can be identified by the letter P above a blue or a green oval, offering the user the option of specifying their value before running the model. Administrative limits are placed into parameters, because the model is meant to be applied to any country and region in the world. Environmental, geographical and socioeconomic factors (land cover, slope, POI, roads and hydrology) are applied using parameters as well, because they may influence operational difficulty of demining in different ways for different study areas while using different demining techniques. It is possible to add further parameters to the model: other factors (e.g. human settlements, temperature gradients, conflict zones etc.), the weights of Table 12, the weights of Table 13 and so on. The underlying complexity of the workflow (Figure 29) is hidden from the users (e.g. decision-makers and operations) who only interact with the system through this set of parameters (Figure 31).

138

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 30: Output raster representing the operational difficulty of demining in Mozambique. The presented results are for a ―fictive (demining) machine with medium class characteristics and commonly used in many countries

139

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 31: Parameters provided to users at the execution of the model

3.2.6. Benefits of the Model

The model is a powerful tool that can calculate in 30 minutes an operational difficulty layer of the entirety of Mozambique (about 800,000 km2), with a 200-m resolution. In addition, the model is flexible, user- friendly and does not require advanced GIS skills from its users. It holds national and regional relevance and is potentially applicable to any mine-affected country. Since environmental, geographical and socioeconomic conditions vary from one country to another, the input data, the area of study and the weights can be set as the model‘s parameters. Other parameters (e.g. human settlements, temperature gradients, soil types and characteristics, elevation, conflict zones, etc.) can be added as inputs according to data availability and user needs.

140

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

The main output of the model is a map. With it, users can have an overview of the situation in their area of work at a glance. The map can also be overlaid with other information, such as hazardous areas, population densities, internally displaced populations, etc. Zonal statistics can easily be derived from the output raster map for each degree of operational difficulty. For example, the overall surface with a low degree of difficulty is directly read into the output raster. This kind of information may be significant for decision-makers and operators, especially in financial terms. With further work, in fact, this model opens the possibility to estimate the financial implications of their operational choices.

3.2.7. Conclusion

The 5D model is a first approach for modelling an operational difficulty of demining at a macro level. The model was developed in ArcGIS Desktop, which is readily available in most mine-affected countries. Users interact with the model via an intuitive and graphical interface by using a set of parameters that can be modified each time the program runs, especially the area of study and input factors. Even if the workflow may seem complex, using the model does not require intensive GIS skills. The resulting map is a good starting point for decision-makers and operators to refine their evaluation of the degree of operational difficulty and improve efficiency in their work. However, this tool is intended as a guide, and real world political or economic factors may lead to or prevent demining activities in a way that may not agree with the tool. In addition, deminers should be aware that modification of one parameter could affect the outputs of the model significantly.

141

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.3. Calculating the Shortest Path

Based on: NAMA: A GIS-based Network Analysis approach for Mine Action

Pierre Lacroixa,b,c, Pablo De Rouletd, Rocio Escobara, Olivier Cottrayd a University of Geneva, Institute for Environmental Sciences, enviroSPACE Lab., Battelle – Building D, 7 route de Drize, CH-1227 Carouge; b University of Geneva, Forel Institute, 10 route de Suisse, CH- 1290 Versoix; c United Nations Environment Programme, Division of Early Warning and Assessment, Global Resource Information Database – Geneva, International Environment House, 11 chemin des Anémones, CH-1219 Châtelaine; d Geneva International Centre for Humanitarian Demining, 7bis, avenue de la Paix, P.O. Box 1300, CH-1211 Geneva 1

3.3.1. Abstract

The time needed to travel through a network of roads can be of decisive importance in the context of mine action. In this research we present NAMA (Network Analysis for Mine Action), a method based on GIS network analysis tools that can be advantageously used for many applications supporting strategic planning and decision-making in the context of mine action, and beyond in the wider context of humanitarian activities (e.g. food delivery to impoverished people and rescue operations). We illustrate one of these possible applications with a case study to identify the best location for building a new medical facility to improve medical care to mine victims. The study area is in Colombia but NAMA can be applied to any region or any country, at national and regional geographical scales.

3.3.2. Introduction

Treatment of ERW victims represents a key issue for mine-affected countries, and its importance is highlighted in the Mine Ban Treaty. The fact that accidents often occur in remote areas, away from medical facilities and in impoverished communities where road infrastructure is often poor or non- existent, makes medical assistance especially difficult. This poses a challenge for victim assistance, including the long process of physical rehabilitation. In this context, the quality of roads and the number and accessibility of medical centres play an important role in the quality of medical support that victims are provided with. As noted by the World Health Organisation, physical rehabilitation plan ―that requires a poor person living in a rural area to travel frequently to the city is likely to fail‖ (WHO 2010). Within this framework, GIS-based network analysis tools can have an important role to play in improving victim assistance programmes by assessing the accessibility to medical facilities by victims of ERW. We

142

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

start with the assumption that many mine victims suffer accidents from location close to where they live and work, and use the place of accidents as the most precise available information on the location of their livelihoods. Among the possible GIS analysis of an area contaminated with mines, it is possible to assess the ability of affected communities to reach medical facilities and propose solutions for better accessibility (Verjee 2010). This research is a part of a collaboration involving the GICHD, the University of Geneva and the United Nations Office for the Coordination of Humanitarian Affairs (UN OCHA) in Colombia.

3.3.3. Objectives

This research aims to show and discuss the potential of network analysis tools for mine action. As we will show below, an assessment of the accessibility of facilities can be done using relevant tools and data. The example used here relate to long-term victim assistance showing how network analysis can help the dispatching of mine victims to different health centres. Our analysis will not deal in depth with the specifics of victim assistance and medical care, as it is beyond the scope of this paper and the area of competence of its writers, but more modestly to propose one possible application of network analysis to support mine action. We hope and believe that this paper will interest medical professionals, but want to stress that the method may be applied for any issue where the circulation of people and/or goods on a road network is needed. We will here identify the best potential location for a medical facility according to the time needed to bring a victim to a hospital. The intermediary results from this study are statistics and maps indicating (1) which accidents are covered by which medical centres within a given amount of time, and (2) which human settlement is the most accessible to the highest number of ERW victims. The data needed to apply the Network Analysis for Mine Action (NAMA) approach on a region or a country consists in GIS vector layers representing roads, settlements, facilities, such as hospitals and other medical centres, and accidents.

3.3.4. The ArcGIS Network Analyst extension

ArcGIS Network Analyst extension allows transportation network analysis, based on a polyline layer representing roads. The principle of the extension is to find routes, from a set of origin points to a set of destination points. The extension allows several operations, including least-cost routing, finding closest facilities, determining service areas and building origin-destination (OD) matrices. All these functions calculate routes by minimising various criteria, which may include metric distances along roads, but also travel time or fuel consumption. In particular, the OD matrix that was used in our approach, calculates the shortest travel time between pairs of origin and destination points. By default, the OD matrix solver will search all possible destinations. It is however possible either to specify a maximum time to reach a

143

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

destination (this time is called ―cut-off time‖ value), or to indicate a maximum number of locations to find, if, for example, the user only wants to find the closest two potential facilities.

3.3.5. Inputs of the Model

3.3.5.1. Study Area

The study area for this network analysis study is a 130'000 km2 (about 50‘000 sq. mi.) area overlapping the three Colombian provinces of Caldas, Cordoba and Antioquia. This region was chosen for two main reasons. First, we could benefit from relevant data sets, notably road network provided by UN OCHA and accident data provided by PAICMA. Second, it accounts for a quarter of all of the almost 10'000 ERW victims in Colombia for the last twenty years.

3.3.5.2. The Road Network

The starting point of a network analysis is the road dataset. It is an interconnected set of lines, representing geographic features through which a vehicle can move. This set of lines must be topologically clean to function, i.e. properly connected and without duplicate features. The network used for this study is made of about 30‘000 road segments of a total of 65'250 kilometres, covering the study area. The attribute table of the road network stores the length of each road segment. The table can be edited and customised to store other types of elements, notably, road interdictions, height or weight restrictions, speed limit, estimated speed or road surface type and condition. Any of these may be used as impedances to travel. The NAMA model attempts to minimise travel time calculation on the basis of the length of the roads and the time spent on the road, which strongly depends on its quality and on the categories of roads. Primary and secondary roads are given the speed of 70 km/h, lower quality roads and rural path are respectively given the speed of 50 and 20 km/h. Fictive paths were specifically created for this study to connect both accidents and hospitals to the road network. The size of these segments averages 370 meters between the medical facilities and the road network, and 1.95 kilometres between the accidents and the road network. It was assumed that these segments represent small paths where maximum possible speed is limited to 5km/h. Speed and distances allows users to model travel time for each road segment of the network. This speed and time estimate has been modelled specifically for the Colombian context, but it is of course customizable to other countries and regions, where the quality of the roads will vary.

3.3.5.3. Accidents

The second datasets used for the NAMA model comprises a set of 566 accident data, recorded between 1992 and 2010. The bulk of the accidents was recorded as occurring between 2001 and 2009. This

144

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

accident layer was provided by the Colombian national demining programme PAICMA, from their IMSMANG database.

3.3.5.4. Medical Centres

A layer of medical centres representing facilities that could implement a mine victims rehabilitation programe was downloaded from OpenStreetMap, a free open- and crowd-sourced map service (OpenStreetMap contributors 2012). OpenStreetMap is developed worldwide on a voluntary basis, from both governmental and private data sources. The layer of medical centres contains 14 clinics or hospitals in various cities of the study area.

3.3.5.5. Human Settlements

This layer containing 179 cities and towns for the study area is used as a final input to the model. It was extracted from an OpenStreetMap layer of thousands of localities across Colombia. Characteristics and origins of all input data are summarised in Table 14.

Table 14: Main characteristics of the input datasets Characteristics Road Network Accidents Medical Centres Human Settlements Period covered: - - Version August 2012 1992 to 2010 Update frequency - Continuously Continuously Continuously Transmitted by Available online Available online for Available online for Availability PAICMA to for free free non-commercial use GICHD Extracted as Shapefile Shapefile Data format Shapefile shapefile from IMSMANG Source UN OCHA PAICMA OpenStreetMap OpenStreetMap Geometry type Polyline Point Point Point Colombia All contaminated All contaminated Antioquia, Caldas, Extent countries are countries are covered Cordoba covered

3.3.6. Workflow

The purpose of our analysis is to identify the best potential location for a medical facility according to the time needed to bring a victim to a hospital. This analysis is divided into five steps:  Step 1: we determine which accident victims can be brought in a reasonable time to an existing hospital, using an OD matrix calculation and a 2-hour cut-off value (Figure 32). This travel time allows a person to go and come back from treatment in the same day. Note that this value could

145

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

be reduced or augmented, following a definition of needs. Origin features are the 566 accidents while destination features are the 14 medical centres.  Step 2: not all the accidents processed in Step 1 are taken in charge by the 14 existing hospitals. Therefore, as a second step, we select the accidents that are not covered and we use them as origin features for a second OD matrix calculation, with the human settlements as destination features (Figure 33). Here again, a 2-hour cut-off value is applied.  Step 3: summarising the results of the last OD matrix calculation gives the number of accidents covered by each human settlement. On this account, it is possible to define the location for the implementation of a new rehabilitation centre, to which a maximum of victims could be brought in less than two hours. The town of ―Ficticia‖ appears to be the best place to locate a medical facility with rehabilitation capacity.  Step 4: we implement this new facility and add it to the hospitals layer. To evaluate the benefits of implementing this new facility in the town of ―Ficticia‖, we recalculate the same OD matrix as in Step 1: origin features are the 566 accident locations and destination features are the 15 hospitals. This time the hospitals layer is composed of the 14 original ones plus the one in ―Ficticia‖ (See Figure 34 and Table 15).  Step 5: Steps 1 to 4 are repeated without a cut-off time value to provide comparative statistics (See Table 16 and Table 17).

146

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 32: Step 1 of the workflow: OD matrix showing the accidents that are within 2 hours of roads travel of existing medical facilities

147

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 33: Step 2 of the workflow: OD matrix with a 2-hour cut-off value. Origin features are the accidents that are not covered by existing facilities. Destination features are the human settlements

148

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 34: Step 4 of the workflow: OD matrix with a 2-hour cut-off value. Origin features are all accident locations. Destination features are the 14 original hospitals plus the new medical centre in ―Ficticia‖

149

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.3.7. The Model

Steps 1-2-3 of the workflow described above are illustrated in Figure 35. This workflow was implemented within the ArcGIS ModelBuilder programming environment. It is composed of (1) input data, (2) tools, (3) output, and (4) parameters. The input data (in blue) is composed of the vector data described above: roads, accidents, medical centres and human settlements. The tools (in orange) make up particular steps of the model, e.g. calculation of OD matrices and counting operations to determine the destinations that cover the highest number of accidents. Outputs (in green) correspond to data and statistics that are generated by the execution of the model. They are materialised in Figure 32-Figure 33-Figure 34 and in Table 15-Table 16. The parameters, identified in the model with the letter ―P‖ will allow users from other countries to apply this model to their own data. Parameters can be entered with the window shown in Figure 36. The cut-off time value has also been set as a parameter to give potential users more flexibility and adapt to the specificities of their country. It is possible to add further parameters to the model, e.g. study area and output coordinate system. Even though the model presented in Figure 35 might seem complicated, the user will interact with it using only a few predefined parameters (See Figure 36).

Figure 35: Overview of the NAMA model: Steps 1-2-3

150

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 36: Parameters provided to users at the execution of the model

3.3.8. Case Study Results

As shown in Table 15, it appears that only 75 of the 566 accidents (13%) processed in Step 1 are covered by the existing hospitals with a 2-hour cut-off time value. Eighty-three accidents (16%) are located in areas less than two hours of road travel from the town called ―Ficticia‖. This means that if a new facility is implanted in this particular place (Step 4), the number of accidents located at less than two hours travel time to a medical facility will grow from 75 to 158 and that this facility would treat more victims than all other medical centres combined. This result points to factors beyond the scope of this study, such as capacity constraints on the facilities. With the new situation, 29 % of the victims of all accidents could be treated within 2 hours, versus 13% for Step 1. These results are corroborated by comparison between Figure 32 and Figure 33, which highlights at a glance the benefits of implementing a new facility to serve a larger number of victims.

Table 15: Number of accidents covered: comparison with (Step 1) / without (Step 4) the newly implemented hospital in ―Ficticia‖. A 2-hour cut-off time is applied Input accident locations Accidents covered Accidents uncovered

Existing hospitals 566 75 491 (Step 1) Existing hospitals + 566 158 408 Ficticia (Step 4)

Table 16 also provides interesting results in the case where all accidents reach a medical facility, i.e. without imposing a cut-off time value (Step 5). Implementing a new hospital in the town of ―Ficticia‖

151

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

significantly reduces the average charge per hospital (93 accidents taken in charge versus 111) and the median charge per hospital (89 versus 149).

Table 16: Mean and median charge per medical centre (Step 5): comparison with / without the newly implemented hospital in ―Ficticia‖. No cut-off time value is applied Mean charge per Median charge per medical centre medical centre Existing hospitals 111 149 (Step 5) Existing hospitals + 93 89 Ficticia (Step 5)

In Step 5, we examine and compare the average and median travel times and distances in the two possible cases (with / without the newly implemented hospital in ―Ficticia‖) without cut-off value. Results show a strong reduction of time spent as well as distance covered with a new facility (See Table 17). The mean time decreases from 4h to 3h15, the median time from 4h to 3h, the mean distance is reduced by 23% and the median distance by 35%.

Table 17: Statistics about the travel time and distance covered (Step 5). Comparison with / without the newly implemented hospital in ―Ficticia‖. No cut-off time value is applied Mean time Median time Mean distance Median (hours) (hours) (km) distance (km) Existing hospitals 4 4 165‘120 159‘750 (Step 5) Existing hospitals + 3h15 3 126‘520 104‘330 Ficticia (Step 5)

3.3.9. Perspectives for NAMA

Within the simplified parameters of this study, the NAMA model analysis has allowed identification of the best potential location for a new medical facility to treat ERW victims in the shortest time and to reduce the average charge of the existing facilities. Intermediary results of the model are (1) the selection of accidents that can already be treated within a predefined amount of time, and (2) the identification of which human settlement is the most accessible to the highest number of ERW victims. We can also integrate local specificities, such as time and distance to provide guidance for optimizing the efficiency of a rehabilitation centre. The reproducibility of the model gives it strong potential to be used at both regional and national scales. The model is user-friendly, flexible and does not require advanced GIS skills from users. The input data being in a well known GIS format is also a help for its use in different contexts. The fact that users interact with the model through an intuitive and graphical interface, can encourage the use of such GIS methods by humanitarian demining actors. NAMA is a fast tool, able to process 30‘000 road segments (corresponding to 65‘250 kilometres of network) in a few minutes (e.g.: in this study the processing took six minutes). In terms of strategic planning, an analysis of this type can

152

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

strongly help the decision-making process of national and regional authorities to improve the conditions of the treatment of the victims of ERW. However, an important limitation to the use of a network analysis-based analytical model is the dependence on the availability of good quality data, especially for the road network. The high data quality required to be able to use the network makes data preparation a long process. In the case of the NAMA study, we estimate that around 50% of the work was dedicated to data preparation, mainly the road network. This means that reproducing the experience elsewhere highly depends on data quality. Another interesting perspective with the NAMA approach is to divide the set of facilities into different categories, with regard to the degree of medical care that is provided, e.g. a medical centre that can provide some trauma care versus a health clinic that has no capability to address trauma cases but could provide some emergency first aid. This would allow NAMA to be used not only for long-term rehabilitation but could benefit including emergency treatment, and may be used more widely than for the case of ERW victims in a long-term development strategy. However, the difficulty of finding such data is similar as the above-mentioned rarity of good quality road network data. The analysis presented in this research shows the results in optimal conditions, where it is possible to circulate on all roads at any time. Difficulties may however arise while using roads that may be temporarily or permanently blocked due to a variety of reasons, such as maintenance work, security issues, flooding or the presence of ERW or Mines other than Anti-Personnel Mines (MOTAPM). To take into account these potential difficulties, it is easy with the NAMA model to simulate road barriers, where one or several roads are blocked to circulation. The possibility to simulate barriers also gives network analysis a strong potential for the specific process of road clearance. The survey procedures in road release process being divided into road segments and the possibility to insert the information gather into attribute tables makes it especially fitting for network analysis with vector data (GICHD 2008a). A network analysis can then be used both to find detours when roads are dangerous, but also to prioritize the segments that must be cleared first by analyzing which blockage provoke the strongest disruptions in the overall network. Beyond the questions of victim rehabilitation and road clearance priority setting, we believe that NAMA might benefit strategic and operational decision-making in the broader context of humanitarian activities and more widely in other topics. If we substitute accidents and medical facilities with other types of origins and destinations in the geoprocessing analysis, NAMA can be used for many other applications for mine action and the humanitarian community as a whole, such as rescue operations in case of catastrophic events (e.g. conflict or natural disaster), routing of the victims of road accidents to the closest medical facility, and food delivery from warehouses to delivery points.

153

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.4. Setting Priorities

Based on: MASCOT: Multi-criteria Analytical SCOring Tool for ArcGIS Desktop

Pierre Lacroixa,b,d, Helder Santiago.c, Nicolas Raya,b,d a University of Geneva, Institute for Environmental Sciences, enviroSPACE Lab., Battelle – Building D, 7 route de Drize, CH-1227 Carouge; b University of Geneva, Forel Institute, 10 route de Suisse, CH-1290 Versoix; Independent Consultant, 7 rue de Dizerens, CH-1205 Genève; d United Nations Environment Programme, Division of Early Warning and Assessment, Global Resource Information Database – Geneva, International Environment House, 11 chemin des Anémones, CH-1219 Châtelaine Preprint of an article submitted for consideration in the International Journal of Information Technology and Decision Making © 2013 [copyright World Scientific Publishing Company] [http://www.worldscientific.com/loi/ijitdm]

3.4.1. Abstract

MASCOT (Multicriteria Analytical SCOring Tool) is a help decision tool based on spatial analysis. The principle of MASCOT is to score items (points, lines, and polygons) in function of their Euclidian distance to other data (points, lines, polygons, rasters). Born in the context of humanitarian demining, it was created through a partnership between the Geneva International Centre for Humanitarian Demining (GICHD) and the University of Geneva. Originally developed for the mine action community, with the purpose of prioritizing clearance of mine hazards in function of their distance to a set of predefined and measurable factors (e.g. environmental, human, economic, etc), its flexibility and interdisciplinary nature does not limit its use to this community: MASCOT can efficiently be used in many other application fields. The tool has been released for free download to be integrated with ArcGIS Desktop 9.3.1 and will be maintained in the future. MASCOT makes it possible to achieve a complete workflow – which may include data preparation, grouping of factors by thematic, weighting, scoring, post-processing and decision-making – without closing ArcGIS. To achieve the weighting process, the Analytic Hierarchy Process (AHP) has been integrated in MASCOT.

Keywords: AHP, ArcGIS Desktop, Euclidian distance, priority, scoring, SDSS

3.4.2. Introduction

Spatial Decision Support Systems (SDSS) are used in many topics, e.g. agriculture, business, energy, fire protection, land suitability analysis, transportation, utility and water resource management. Good references on these topics can be found in Malczewski (1999) and Sugumaran and Degroote (2010). A shorter overview of Geographic Information Systems (GIS) multi-criteria analysis was presented by

154

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Karnatak et al. (2007) and an overview with a focus on environmental Decision Support Systems (DSS) by Matthies et al. (2007). Numerous articles have been published, where a spatial support system has been applied to find best locations for various topics: landfill evaluation in Southern Spain (Zamorano et al. 2008), solid waste management (Nakakawa 2006), determination of emergency shelter sites (Kar and Hodgson 2008), water quality management (Assaf and Saadeh 2008), vineyard sites identification in Southern England (Foss et al. 2010), pipeline route selection (Yildirim and Yomralioglu 2011), forest conservation (Valente and Vettorazzi 2008), tourist advice (Dye and Shaw 2007), wind farm site selection in Lebanon (Bazzi and Fares 2008) and many others. Geoffrion was one of the first to suggest, in 1983, that DSS should help users explore the solution space (Geoffrion 1983). Since that time, the amount of geospatial data has been constantly increasing and spatial models have become increasingly complex. Malczewski (1997) has pointed out that the outcomes or consequences of most decision alternatives vary in space, and scientists operating in many fields have praised the following benefits of having spatial functionalities in DSS:  Increasing accessibility to spatial models (Karnatak et al. 2007) and handling with complex structures (Goodchild and Kemp 1990).  Processing many geospatial datasets at the same time (Yildirim and Yomralioglu 2011).  Providing spatial statistics functions (Zamorano et al. 2008).  Providing consistency and accuracy to decision-making, and reducing decision-time (Wellar 1990, Crossland et al. 1995, Zamorano et al. 2008).  Providing aggregated information at the desired geographical scale (Knezic and Mladineo 2006),  Making decision-making visual and interactive (Goodchild and Palladino 1995). In the humanitarian demining field, Knezic and Mladineo (2006) also demonstrate the benefits of a hierarchic GIS-based DSS in the setting of humanitarian demining priorities and in the money distribution process.

3.4.3. Why developing MASCOT, and for whom?

Apart from Knezic and Mladineo (2006) few examples of the use of DSS or SDSSs in humanitarian demining have been published. By processing mostly non-geospatial information, the scoring processes described below do not invoke spatial relationships between features. Socioeconomic impacts have been described for local examples like Kosovo (Benini 2000, iMMAP 2002). The objective of such surveys was to classify and score affected communities in relation to the severity of socio-economic impacts (broad nature of munitions, resources and facilities blocked, number of recent victims, etc) caused by landmines and Unexploded Ordnances (UXO)25. The Kosovo survey resulted in national strategic plans, with priority to clearing areas with high scores. Additional spatial

25 UXO: ―bombs, shells, mortars, grenades and the like that have been used but which have failed to detonate as intended‖ (GICHD 2010b, p. 13)

155

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

analysis was conducted with the intention to include environmental factors in the definition of priorities – e.g. overlay with slope or vegetation datasets – but this was done as a further process, out of the scoring tool. De Leener and Pastijn (2002) have used multi-criteria analysis for selecting sensor combination in the framework of landmine detection. Goslin (2003) proposed a standardized task impact assessment (TIA) tool to help mine action operators prioritize their implementation tasks in the field, and Lisica (2003) a qualitative priority-setting model by crossing risk indicators with socio-economic impacts. More recently, a Resource Planning Tool (GICHD 2010a) was integrated in the Information Management System for Mine Action – Next Generation (IMSMANG; Eriksson 2011), an ArcGIS Engine-based suite coupled with a MySQL Relational Database Management System (RDBMS). Specifically developed by the Geneva International Centre for Humanitarian Demining (GICHD), IMSMANG allows the mine action centers in mine-affected countries to record and report the information relative to mine action. As of 2013, sixty mine-affected countries – representing 1‘200 computers and 3‘000 users – are equipped with it. The IMSMANG Resource Planning Tool helps measuring a likelihood of hazards affecting household income resources and prioritizing according to predefined strategy goals. With it, factors such as mine hazards, victims, mine risk education, agricultural and infrastructures blockage can be read from MySQL columns (―attributes‖) as inputs of the scoring process. In all the mine action-related examples presented above, spatial relationships between objects (e.g. distance, inclusion, overlap, etc.) do not influence the results of the scoring analysis, which is essentially based on qualitative or quantitative information stored in non-geospatial attributes. Within this framework, the GICHD decided to develop MASCOT (Multicriteria Analytical SCOring Tool), in collaboration with the University of Geneva. This tool should find challenging applications in the mine action field, like helping determine priority clearance of mine hazards in function of their distance to a set of predefined and measurable factors (e.g. environmental, human, economic etc). For example, coupling food security areas, livelihood areas, key economic sectors and key infrastructures with demographic data within a scoring process would help prioritize strategic zones and nodes that have spillover effects beyond the nearby households and the immediately neighboring villages. Then, expanding the number of factors and the range area from local to higher level might contribute to calculating economic effects in the form of a return on investment assessment over the area to be demined. The major aims of MASCOT are: (1) to open humanitarian demining task prioritization processes to new geospatial factors and complex data models, (2) to include the multi-scale approach, (3) to bring even more flexibility, accuracy and consistency to decision-making processes, and (4) to reduce decision-time. The overarching benefit of using MASCOT over non-spatial approaches is to save more lives. Even though MASCOT was born in the framework of mine action, we also wished to provide a generic tool that could be beneficial to many others topics, a tool that would connect humanitarian demining actors in a broad community of scientists and experts operating in various disciplines and at various scales, and characterized by highly heterogeneous needs, data, resources and GIS knowledge. Typically,

156

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

MASCOT will be implemented by a group composed of several collaborators (Matthies et al. 2007): one system analyst, one or several stakeholders who are experts in their discipline and one or several managers/decision-makers (Malczewski 1997). The system analyst or ―geomatician‖ would run the tool and post-process the outputs, while the other ones would decide how to feed it (data, thematics, analysis distances and weights), take relevant decisions and decide whether further analysis is needed (Goodchild and Kemp 1990).This set of collaborators, who might in certain cases be a single person, should have an in-depth understanding of the tool‘s assumptions and shortcomings (Matthies et al. 2007). In Section 3.4.4, we explain how we thought and designed the technical specifications of MASCOT, list the characteristics that were set as requirements, and explain why the tool was developed in combination with the AHP in the ArcGIS Desktop framework. In Section 3.4.5 we describe how to use MASCOT through a typical scoring workflow, as well as describe in details the inputs, outputs, parameters, and main functions of the tool, in particular the weighting process. In Section 3.4.6 we discuss what makes the novelty of MASCOT, what are its limitations and perspectives, and we propose improvements for future versions.

3.4.4. Design of MASCOT

3.4.4.1. Requirements

Literature review helped us figure out which characteristics actually define a ―good‖ SDSS. The technical specifications of MASCOT were designed on this account. Characteristics that seemed relevant to us are listed below:  GIS integration.  Ability to help decision-making and to validate users‘ choices.  Ability to allow users to supervise the analysis.  Reduction of users‘ effort and decision-time.  User-friendliness, intuitiveness and flexibility.  Interoperability and interdisciplinarity.  Multiscalability.  Ability to fit users‘ needs. In the acronym ―SDSS‖, the first ―S‖ stands for ―Spatial‖. Thereby, an SDSS should naturally be coupled with other modules supporting GIS functionalities, e.g. a GIS based RDBMS and a Display generator (Armstrong and Densham 1990). Uran and Janssen (2003) pointed out the necessity for SDSSs to support spatial analysis and evaluation of model outputs. Similarly, a careful and adequate preparation of spatial data (e.g. data classification, clustering, aggregation, extraction, filtering) before processing them through an SDSS is crucial to obtain good performances and results. Decision support tools, whether they include spatial components or not, are participatory tools that are

157

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

meant to assist users in making choices and validate these choices (Goodchild and Kemp 1990).This particularly makes sense when decision-making is made jointly by several actors having different preferences with regard to evaluation criteria (Matthies et al. 2007).These actors will necessarily have to agree on a compromise solution (Pohekar and Ramachandran 2004). The quality of this agreement often relies on the capacity of tools to propose functions for assessing different alternatives and scenarios (Geoffrion 1983, Assaf and Saadeh 2008). An SDSS should let enough room for users to supervise all the links of the processing chain. Goodchild and Kemp (1990) assert that the role of decision-makers is essential in defining problems, achieving analysis and determining whether further runs of the tool are needed. Reducing decision-time and effort is also a major concern for users (Erden and Coskun 2010, Rigopoulos et al. 2010, Yildirim and Yomralioglu 2011). At stake, the necessity of processing an ever-increasing amount of geospatial data and the need for increasing effectiveness of a decision process and minimizing costs. Flexibility and reusability are steps in this direction (Sangtani and Serpen 2010, Rahman et al. 2012), as well as ease of use and intuitiveness. The significance of having user-friendly, intuitive and flexible tools sounds like a leitmotiv in literature. A good reference is Bazzi and Fares (2008) who implemented a multi-criteria procedure within the existing ArcGIS graphical user interface (GUI) and designed a specific Visual Basic interface. For Geoffrion (1983) and others, decision support tools should have a powerful, easy to use and flexible interface. The need for flexibility does not only concern the interfacing but also tool capacities, in particular capacity to deal with uncertainty. Malczewski (1997) and Chen et al. (2011) stressed the uncertainty that often surrounds decisions and the need for multi-criteria analysis tools to integrate it. SDSSs are often asked to process multiple criteria, conflicting objects (Eldrandaly et al. 2003, Pohekar and Ramachandran 2004), qualitative and quantitative criteria and both objective and subjective information (Javalgi and Jain 1988, Ray and Burgman 2006) often coming from interdisciplinary sources (Matthies et al. 2007). Within the GIS environment, this implies that a ―good‖ SDSS accepts the most common GIS data formats as inputs and generates interoperable outputs. When it comes to GIS-based tools, a multi-scale approach is important to support decision-making at various geographic and administrative levels (Knezic and Mladineo 2006, Bazzi and Fares 2008). The ability for a SDSS to fit with users‘ needs is another major concern. Matthies et al. (2007) and others point out the importance of involving end users in the development and implementation of help-decision tools. One should also keep in mind that users‘ needs evolve and that help-decision tools should be able to follow this evolution (Geoffrion 1983). As described all along the next sections of this paper, a strong effort was put to conform to the above- mentioned requirements when developing MASCOT. In particular, the choice of the ArcGIS Desktop platform is explained in detail in Section 3.4.4.2.

158

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.4.4.2. ArcGIS Desktop Integration

In this section, we expose the reasons why the ArcGIS Desktop platform was chosen for the integration of MASCOT. The tool was initially developed to address a need of the humanitarian demining community. Data related to mine action are recorded in the field with GPS tools and registered in IMSMANG. This application is using a non-spatial database engine (MySQL) but spatial data are stored as 2D coordinate pairs and can then easily be spatialized to feed a GIS. To take the full step to GIS analysis, most mine- affected countries chose the Esri suite in connection with the IMSMANG server. As of 2012, 1‘300 ArcGIS licenses have been provided to the mine action community. If MASCOT was originally developed for the mine action community, we want this tool to be generalist, multidisciplinary, generic and usable by a maximum number of scientists. The ArcGIS Desktop platform is a good candidate for its integration as Esri counts 350‘000 clients operating in many disciplines in dozens of countries (Esri 2011b). Another reason in favor of this integration is the richness of the ArcGIS Desktop platform in terms of functions for preparing and post-processing data, e.g. data conversion to the appropriate format, re- projection to the appropriate coordinate system, classification, extraction to the required extent, aggregation at the desired geographical or administrative level, removal of duplicates, application of attribute-based or spatial filters, overlay operations (Esri 2009a), mapping, computation of spatial statistics, export to open standard formats and graph generation. There are two different ways of accessing these functions. (1) From the ArcGIS ArcToolbox GUI. Functions can be customized in the Python language, documented and combined in a predefined order in the ModelBuilder visual programming environment (Esri 2009a), along with parameters provided for end users to test and evaluate separate alternatives. Processes may even be automated and their execution planned. (2) From the ArcMap GUI. Processes and interface can be customized and enriched in Visual Basic .Net almost without limit. The ModelBuilder application presents shortcomings that are prohibitive with regards to requirements of Section 3.4.4.1: (1) data are processed and tools are executed in a predefined order, (2) interface is hardly customizable, (3) basic functionalities such as weighted overlay only work with raster data, (4) reporting functionalities are not included, and (5) performances with Python are much lower than with Visual Basic .Net, especially when processing rasters. For all the reasons exposed in this section, we decided to develop MASCOT in the ArcGIS Desktop platform. The underlying programming language is Visual Basic .Net.

3.4.4.3. Integration of the AHP in MASCOT

In line with requirements of Section 3.4.4.1, MASCOT holds a module for assisting users during the weighting process. By default, weights can be entered from the direct inquiry of experts/decision-makers, but the tool proposes a second classification method based on users‘ preferences and on mathematical

159

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

algorithms minimizing misclassifications (Rigopoulos et al. 2010). This classification method was chosen among a set of well-known techniques, which are described and discussed below: pair-wise comparison, ranking, rating and trade-off analysis. The Analytic Hierarchy Process (AHP) has been extensively studied and used since the 1970s. It is a very popular method, applicable to extremely different cases: energy, conflict management, transportation planning, operational research, production, human resources and many others. For example, Pohekar and Ramachandran (2004) estimate that 20% of energy decision-making applications use the AHP. Spatial implementations of this technique were made in several GIS like IDRISI (Eastman and Jiang 1995), ArcInfo GRID (Veitch and Bowyer 1996, Hill et al. 2005, Yanhua et al. 2009) and ArcView 3 (Malczewski 2006). An attempt to integrate it in ArcGIS Desktop 9.1 was made by Marinoni (2009) but the tool, called AHP 1.1, did not support vector datasets and was not upgraded to ArcGIS 9.2 or further versions. The principle of the AHP is to consider the relative importance of criteria in obtaining the best alternative regarding identified objectives. Criteria and their relative importance are put in a pair-wise comparison table (see example provided in Table 18) in conformance with Saaty‘s fundamental scale table (Table 19; Saaty 1990).

Table 18: AHP: scale of relative importance for pair-wise comparison Criterion Criterion 1 Criterion 2 … Criterion N Criterion1 1 0.33 … 0.2 Criterion2 3 1 … 0.62 … … … 1 … Criterion N 5 1.6 … 1

Table 19: Pair-wise intensity value description Intensity of importance on an Definition Explanation absolute scale Two activities contribute equally to the 1 Equal importance objective Moderate importance of one Experience and judgment slightly favour one 3 over another activity over another Experience and judgment strongly favour one 5 Essential or strong importance activity over another An activity is strongly favoured and its 7 Very strong importance dominance demonstrated in practice The evidence favouring one activity is of the 9 Extreme importance highest possible order of affirmation 2 – 4 – 6 – 8 Intermediate values When compromise is needed

Let A be the pair-wise comparison table (Table 18) and N the number of criteria. A is a square N- dimensional matrix whose elements Aij are integers chosen between 1 and 9 (Table 19) and follow Equation 4:

160

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

1 i  N,j  N, Aij  . (4) Aji

If i-th criterion is Aij times more important than j-th criterion, then j-th criterion is Aij times less important

(Aji times more important) than i-th criterion. All values in the diagonal are equal to 1. The AHP finds the normalized weighting vector W (also called ―principal eigenvector of A‖ or ―priority vector‖), for which A

× W = λmax × W, where λmax, the ―maximum latent root‖, reflects the coherence of the matrix. λmax is given

by det(A - λmax × I) = 0 where I is the identity matrix, exclusively composed of values equal to 1, and Wi =

(Wi1, …, Win) the weight of i-th criterion. The larger the values in Wi, the more important the element in its

level. Each Wi is given by Equation 5 and Equation 6: Mi Wi  . N (5) Mk k1 where:

N Mi  N  Aij. (6) j1

To validate these weights a Consistency Index (CI) can be calculated by CI = (λ max - N) / (N-1). The lower CI, the more consistent users‘ choices are. The global consistency of the matrix can be evaluated through the Consistency Ratio CR = CI / RI where RI, the Random consistency Index, measures the degree of randomness of the rating matrix and is given by Table 20 (Saaty 1980). If CR is larger than 10%, users should revise their choices.

Table 20: Random consistency indices for different number of criteria Number of 3 4 5 6 7 8 9 10 11 12 13 14 15 criteria Random Consistency 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.49 1.51 1.48 1.56 1.57 1.59 Index (RI)

Other well-known classification methods than the AHP were investigated. Their strengths and weaknesses, discussed by Malczewski (1999), Karnatak et al. (2007), Banai-Kashani (1990), Jankowski (1995), Kleindorfer et al. (1993), Pitz and McKillip (1984), Schoemaker and Waid (1982) and Siddiqui et al. (1996) are summarized in Table 21.

Table 21: Strengths and weaknesses of major weighting methods. N is the number of criteria Pair-wise Feature comparison (e.g. Ranking Rating Trade-off analysis AHP) Number of N (N-1) / 2 N N < N judgments Response scale Ratio Ordinal Interval Interval Hierarchical Yes Possible Possible Yes

161

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Underlying Statistical/heuristic None None Axiomatic/deductive theory Ease of use Easy Very easy Very easy Difficult Trustworthiness High Low High Medium Precision Quite precise Approximations Not precise Quite precise Weights can be Weights can be Weights can be Use in a GIS Flexible. Allows for imported from a imported from a imported from environment multiscalability spreadsheet spreadsheet Logical Decisions

The AHP has lots of advantages, which are in line with the requirements of Section 3.4.4.1. Participatory, mainly based on users‘ choices, the AHP recognizes their knowledge and expertise (Karnatak et al. 2007) and gives prominence to supervising. Users can revise their choices if the calculation of the consistency ratio suggests some inconsistency or irrelevance. At the time of entering weights in the pair-wise comparison table, users‘ effort is not that very high: N (N – 1) / 2 values are required (e.g. 21 for 7 criteria). A two-by-two criteria comparison also proves to be intuitive, as a good complement to the direct inquiry method. Given its flexible nature, the AHP is able to deal with both tangible and non-tangible aspects of a decision (Rao and Davim 2006) and allows hierarchisation of a problem and multi-scalability, which is particularly useful in the GIS environment. The underlying algorithm is easy to implement and performing: calculating weights from a 7 × 7 pair-wise comparison table is immediate. Finally, the AHP can be applied to many topics and does not exist yet in ArcGIS Desktop 9.3 and 10. Within this framework, we decided to integrate the AHP in MASCOT (Figure 37).

(a)

(b)

Figure 37: Integration of the AHP in MASCOT. (a) Weighting of each road category with the AHP. (b) Absolute weights and consistency ratio

3.4.5. MASCOT

This section describes how to use MASCOT. During the development phase, a strong effort was put to conform to requirements of Section 3.4.4.1.

3.4.5.1. Principle of the Tool

Most of the spatial scoring tools that we found in literature (e.g. AHP 1.1, the overlay tools in ArcGIS

162

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

(Esri 2009a)) perform a cell-by-cell weighted sum of input rasters through local functions (Figure 38). MASCOT goes one step beyond as it offers a scoring process based on Euclidian distance between features (Figure 38: focal functions) and as inputs can be raster or vector data.

Figure 38: Scoring vector items with MASCOT in function of nearby vector and raster objects/categories

The principle of MASCOT is to score items with regard to their distance to a set of predefined scoring objects. These objects may be stored either in raster layers (e.g. a raster with land cover types and a raster with population density ranges) or in vector layers (e.g. a shapefile (SHP) with different categories of public roads). Each range or category (= each ―sub-criterion‖) is weighted, and a distinct analysis distance is set for each layer (one layer = one criterion). Each item is then scored by counting how many scoring objects of each type are located within the corresponding buffer, and by summing the weighting contributions of all scoring features. Weighting and scoring processes, as well as the way distance analysis operates depending on the layer type (vector, integer raster, or float raster) are described in detail below.

3.4.5.2. Workflow

In conformance with Karnatak et al. (2007) a typical MASCOT workflow comprises a number of steps that are ordered intuitively and logically (Figure 39): input from experts and professionals, data preparation, grouping of the scoring layers by thematic (one thematic = one criteria group: See Figure 40), weighting, scoring, post-processing, and spatial decision analysis. Users do not necessarily have to close ArcGIS to go through all these steps.

163

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 39: MASCOT workflow. It is possible to achieve the entire workflow without closing ArcGIS Desktop

To reduce computing time as well as interaction with the machine, it is recommended that users prepare carefully their data before using MASCOT. Examples of data preparation are provided in Table 22.

Table 22: Recommendations for data preparation Data preparation Objectives Note Reducing the number of classes in the input dataset Refer to Figure 41b and Figure Data reclassification Converting float rasters to integer rasters 41c Putting the input data on a similar ordinal scale Preventing from scoring each edge of road, Feature aggregation (e.g. roads by thus from overestimating the impact of the category) road criterion Reducing the amount of features to be Attribute or spatial filters processed Refer to Figure 41a and Figure Enlarging the range of attributes that are Joins 41b, where the Label field is candidates for weighting read in an external Excel file - Projection of all scoring layers to - Ensuring a spatial consistency between - This is mandatory same coordinate system layers

164

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

- Privileging of projected coordinate - Ensuring that analysis distances are systems (e.g. UTM) over geographic expressed in meters or feet, and not in coordinate systems (e.g. WGS 1984) decimal degrees - Privileging of equal-distance - Avoiding biases in the use of the analysis projections to other types distance

In conformance with 3.4.4.1 and because it is impossible to know in advance the needs of potential users, MASCOT allows one to integrate a large number of vector and raster formats supported by ArcGIS. Shapefiles, Access geodatabases (MDB) and File geodatabases (GDB) are the supported formats for items to be scored. Points, lines and polygons are scored in separate processes, because in these formats they are stored into separate datasets. Scoring datasets – namely, the criteria – may be vector (shapefiles and personal geodatabases) or raster (any format readable by ArcGIS: TIFF, JPEG, IMG, raster dataset in a geodatabase, etc.). Scoring rasters may be of two different types: (1) integer rasters, with categories, such as GlobCover (ESA 2010) and (2) continuous rasters, with float values, e.g. digital elevation models such as SRTM (Jarvis et al. 2008). Data with various resolutions can be processed at the same time, but users should be aware that the quality and precision of the results will be highly affected by the data with lowest resolution. Once the data are prepared, the user creates criteria groups, where each group represent a distinct thematic (e.g. environment, topography, infrastructures). He/she allocates the scoring layers to each group (Figure 40 and Table 23). The number of criteria and groups is unlimited, and criteria and groups can be added or removed at any time during the process. Each scoring layer holds a number of categories (e.g. land cover types on Figure 41a and b or slices (e.g. population density ranges on Figure 41c).

Figure 40: MASCOT tree. Criteria (vector and raster layers) are grouped by thematic in criteria groups

Table 23: Typical MASCOT use case Hierarchical Object GIS terminology Example (Distance analysis) level Shapefile with points or (excl.) Items to be scored ERW polygons Vector or raster layer with an 1. Places (750m) Scoring layer Criterion attribute for weighting 2. Land cover (1.5km)

165

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3. Population densities (1km) Category or range of values 1. Schools (750m) Scoring category or Sub-criterion stored in an attribute in the 2. Irrigated cropland (1.5km) slice scoring layer attribute table 3. Less than 1 person/km2 (1km) 1. Economic and social structures Several layers Several layers Criteria group 2. Economic and social structures Layer group 3. Populations

166

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

(a)

(b)

(c)

167

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 41: Weighting categories and slices. (a) Each land cover category is a sub-criterion. (b) Weighting land cover categories by direct input. (c) Weighting slices of population density by direct input

Weighting is performed at three distinct levels:  Sub-criteria: within each criterion, an attribute is selected for assigning a weight to each category (vectors and integer rasters: see Figure 41a and Figure 41b) or each slice (float rasters: Figure 41c). The sum of all weights must be equal to 100%. The sums are displayed in percentages for reasons of user-friendliness. The sub-sub-criteria level is not implemented, because GIS layers do not reach this level of detail.  Criteria: each criterion receives a weight within its group, the sum of all weights being equal to 100%.  Criteria groups are also weighted, the sum of all weights being equal to 100%. To keep the weighting process simple, criteria groups cannot contain other criteria groups. At each of these three levels, weighting can be performed either by direct input or through the AHP. With more than seven sub-criteria, criteria or groups, using the AHP is possible but not recommended (Saaty 1990). MASCOT is limited to fifteen. AHP can be followed by the calculation of a Consistency Ratio CR. If CR is larger than 10%, a message pops up, but the analysis is not stopped. Slices (e.g. for slopes or population density) cannot be weighted through the AHP: direct input has to be used (Figure 41c). If the user wants to exclude a sub-criterion, a criterion or a group from the analysis, he/she just keeps a null weight. In parallel, MASCOT offers a high flexibility, given that the number of slices, criteria and criteria groups is modifiable on the fly. To guide users, weighting is performed following a precise workflow: at sub-criteria level (Figure 37 and Figure 42a) then at criteria level (Figure 42b) then for criteria groups (Figure 42c). Weighting cannot be performed at a level if it has not been achieved at inferior level.

168

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

(a)

(b)

(c)

Figure 42: Weighting process. At each step, weighting can be done by direct input or through the AHP. (a) First step: weighting at sub-criteria level. (b) Second step: weighting at criteria level. (c) Third step: weighting criteria groups

As for weighting, the scoring process is performed through a hierarchical workflow (Figure 42). Each item (point, line or polygon) is scored as follows: MASCOT first inventories scoring features that are found within a predefined analysis distance threshold. This distance is set by the user and is unique for each criterion. When it comes to scoring mine hazards with regard to their potential impact on nearby populations or infrastructures, a strong request from the GICHD was to be able to assign scores as a decreasing function of the distance. If the scoring layer is a

169

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

vector, features that touch a scored item are assigned the maximal weight, and this weight diminishes linearly until the limits of analysis distance, where it is equal to zero (Figure 43). If a polygon is being scored, the analysis distance is considered equal to zero inside the polygon (including its perimeter), but objects that are found inside the polygon contribute to the scoring process. If the scoring layer is a raster, the analysis distance operates differently for performance reasons: the cells contained in the radius of analysis are processed without regard to their distance to the central item.

Figure 43: Weighting as a decreasing function of distance to the scored item (vector layers only)

A scoring object of vector type contributes up to its weight (e.g. 20% for hospitals on Figure 42a) inversely weighted by the distance to the item to be scored. For rasters, spatial statistics are computed within the analysis distance. A float raster contributes up to the weight that was assigned to the slice containing the mean cell value (for population densities), the sum of cell values (for population counts), the minimum cell value or the maximum cell value. An integer raster contributes up to a weight corresponding to the average weight of the cells. This contribution is then multiplied by the weight of the criterion within its group (e.g. 30% for places on Figure 42b). The resulting score is then multiplied by the weight of the group (e.g. 40% for Economic and Social Structures on Figure 42c). Finally, the scores obtained from all vector and raster contributions are summed up. To characterize the outputs of a scoring process, a few keywords separate: performances, interoperability, reusability, minimization of users‘ effort, coupling with GIS modules. They are in line with requirements of Section 3.4.4.1. One scoring process generates one output File Geodatabase to which the scored items are copied. The File Geodatabase format was preferred to other ArcGIS native formats for several reasons (Childs 2009), notably: (1) in SHP field names are limited to 13 characters, (2) SHP and MDB storage capacity does not exceed 2 GB, (3) GDB provides higher performances than the others and (4) this format is accessible to non-Esri applications (Esri 2011c). The output GDB feature class holds the original attributes, plus one attribute with the score calculated from each criterion and one attribute holding the overall score (Figure 44).

170

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Figure 44: Scored items feature class, with one specific score for each criterion plus the overall score

After scoring, the output layer is automatically added to the ArcGIS map, for post-processing (e.g. computation of statistics on the basis of the output scores, mapping (Figure 45), geoprocessing, spatial analysis, etc.) and decision-making.

Figure 45: Application of MASCOT to mine action: scoring ERW to determine areas with high clearance priority

Each run also produces one XML log file with metadata describing the inputs, parameters, weights, weighting methods, scores and outputs related to the process. This file is updated dynamically as criteria or groups are excluded during the analysis, or new ones are added.

171

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

All along the weighting process, weights can be saved to tables. These tables can be reused in other processes, to spare time and to minimize users‘ effort. To keep track of any scoring process, the output File Geodatabase and XML log file are saved to a folder with a name referring to the date and hour of execution.

3.4.5.3. User Help

In line with requirements of Section 3.4.4.1, effort was done on supporting and assisting users all along the scoring process. As it is integrated with ArcGIS Desktop, MASCOT benefits from the ArcGIS help (Esri 2009a). A tooltip accompanies each button. More than 40 warnings and error messages are available. MASCOT is also delivered with a complete user guide and a set of test data including vector features (ERW, hydrology, roads, railways, administrative boundaries, contours, etc), integer and float rasters with different resolutions (LandScan 2008 TM (ORNL 2008), GlobCover).

3.4.6. Perspectives

MASCOT is currently compatible with ArcGIS Desktop 9.3.1, because most mine-affected countries are using this release of the software. A new version of the tool for ArcGIS 10 is envisaged for the future. The fact that MASCOT shows up as a toolbar in the ArcGIS interface will make easier integration of new functions. The development of this new version will provide the opportunity to increase the current performances and capabilities of the tool. The feasibility of developing the following functions is under study and allows one to glimpse interesting perspectives, still in line with requirements stated in this paper. First, automation of the scoring process is envisaged by looping over comparable alternatives/scenarios within a single run of MASCOT. Each alternative would correspond to a distinct set of criteria, groups and weights. Second, the possibility of scoring from an XML log file is investigated. This would reduce interaction with the machine, open the door to process automation, improve reusability capacity and increase the accessibility of MASCOT to non-Esri users. Third, the possibility to recover from an XML log file at any point of a process is also envisaged. This would bring more comfort to users in case of system or software crash. As a fourth improvement to the actual model, enrichment of distance functions is envisaged. If performances are acceptable, scoring as a decreasing function of distance could be applied to raster datasets as it is now the case for vector objects. Other models than a first-degree polynomial could be provided in version 10, to enlarge the applicability of the tool to a broader community of scientists. As an example, the magnitude of a seismic waveform is inversely proportional to the logarithm of the epicentral distance (Odaka et al. 2003). Fifth, the possibility of using cost path rasters as input of MASCOT is under study. Cost rasters find useful applications in many disciplines, e.g. 3D analysis (Lee and Stucky 1998, Pingel 2010), ecology (Liu et al. 2007, Pinto and Keitt 2009, Rayfield et al. 2010), landscape genetics (Ray 2005), animal species motion (Foltete et al. 2008), mobility and routing (Boothby and Dummer 2003, Suvinen 2006, Davies and Whyatt 2009, Hiratsuka et al. 2010). For

172

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

what concerns MASCOT, using least cost distance instead of the traditional Euclidian distance between scoring features and features to be scored could bring a significant value added to the analysis, by providing a more realistic representation of the world, in particular complex systems of connectivity (Pinto and Keitt 2009, Greenberg et al. 2011). In the mine action field, complex real-time operations such as medical evacuation and planned operations such as routing of demining machines to contaminated areas would be supported in a more efficient manner. For example, the presence of tunnels and bridges might be included in the paths (Yu et al. 2003), and real travel time or real travel distance might take into account the influence of land-use and land cover (Pinto and Keitt 2009), topography and natural barriers. Finally, evolution of MASCOT to a web enabled application is envisaged, as a gateway for users to improve facilitation of decision-making and access to the system in a convenient and efficient manner (Karnatak et al. 2007). The Esri suite is a good candidate for this, through its server products.

3.4.7. Conclusion

In this paper, we presented MASCOT, a new help decision tool based on spatial analysis and multi- criteria approach and integrated into ArcGIS Desktop. MASCOT was originally developed for the humanitarian demining community, to answer a need for determining areas with clearance priority. We see in this tool a strong potential for humanitarian demining actors to increase efficiency in their work, and beyond to save more lives. Due to its flexible and interdisciplinary nature, MASCOT can integrate a large variety of scenarios, in many other topics than mine action, at any geographical scale (global, national, regional, local), using it does not require high skills in GIS, and it is potentially accessible to hundreds of thousands of ArcGIS users around the world (Esri 2011b). The major novelty of this tool is to score vector items in function of their Euclidian distance to a set of vector or raster objects. Where most SDSSs perform a local cell-by-cell mathematical combination of raster data, MASCOT takes into account the relative distance between scored and scoring objects. For vector objects, scores even decrease linearly with distance, which opens the door to a more realistic and accurate modeling of potential impacts of features of the real world upon others. To assist users during the weighting process, the AHP was integrated in MASCOT – thus in ArcGIS Desktop – along with a module for evaluating the consistency of the pair-wise comparison matrix. Weighting may be applied through the AHP but also by direct input, and at different levels: criteria (GIS vector or raster layers), sub-criteria (categories or slices in a GIS layer), and criteria groups (thematic). Each alternative/scenario corresponds to a separate run of MASCOT with a set of criteria, criteria groups and weights. During the development phase of the tool, stress was put on guiding users, minimizing their efforts, giving them the chance to log, backup and reuse inputs, parameters and outputs of the scoring workflow. MASCOT is entirely integrated in ArcGIS, so that it is possible for users to achieve a complete workflow, from data preparation to decision making through grouping of criteria by thematic, weighting, scoring, post-processing and mapping, without closing the GIS.

173

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

Users should however be aware that MASCOT is an SDSS, but it does not replace their choices. It is up to them to prepare data, to decide which qualitative and quantitative criteria should be taken into account – and how – to group them by thematic, to put them on a similar ordinal scale (even though they may correspond to different units), to assign weights and to take decisions inside or outside the GIS. Users should also be aware that a lack of available data would be a major issue in the decision-making process. The sixty mine-affected countries in which the GICHD will distribute MASCOT represent an incredible application field in real conditions, on the account of a highly heterogeneous environmental, geographical, historical and political background. Thus, and to ensure that the tool evolves as users‘ needs evolve, mine action users‘ feedback will be considered carefully in the future. The MASCOT package, including installation setup, user guide and test data, is also downloadable for free at http://www.unige.ch/sig/outils/MASCOT.html, making it possible for scientists from other communities than humanitarian demining to provide their feedback and to suggest improvements. The tool requires the ArcGIS Spatial Analyst extension.

174

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

3.5. Highlights of Chapter 3

If GIS-based DSS find useful applications in many fields, their use is still very limited among the mine action community. To contribute to filling this gap, we developed three GIS-based approaches within the framework of this PhD: 5D, NAMA and MASCOT. Developed to address specific users‘ needs, they were implemented in ArcGIS with deployment in sixty mine-affected countries as further objective. 5D (Determining and Displaying a Degree of operational Difficulty of Demining) is an analytical method for assessing and visualising a degree of clearance complexity on the basis of quantifiable terrain criteria (e.g., slope, vegetation, ground softness, precipitations, temperatures etc.). For any type of mechanical platform (e.g. flail, tiller etc.), users can enter realistic and measurable limits for these criteria (e.g. high difficulty for slopes between 20° and 30°, low difficulty for slopes below 20°, medium difficulty for urban areas, high difficulty for sandy soil combined with high precipitations). The criteria are weighted (e.g. 20% for slopes, 10% for ground softness) and summed up into a raster layer. From this layer, macro- statistics can be computed to evaluate the demining capacity of a country, to decide which generic tool should be deployed in a region and to set clearance priorities. The output raster also provides a visual overview of the ordinal difficulty of demining in the study area, which is intended to support operations officers to prepare their intervention in the field and provide funders with an accurate picture of the cost of clearance operations. So far, the model has been calibrated from datasets to the extent of Afghanistan, Cyprus and Mozambique, but it is meant to be reproducible in any mine-affected country at the national and sub-national levels, NAMA (Network Analysis for Mine Action) is an analytical method that shows the potential of using transportation network analysis GIS tools in mine action. A case study was done that helps to determine suitable location for building a new medical facility to improve healthcare to ERW victims. NAMA intends to minimise travel times between incidents and facilities, and possible extensions of the model include minimisation of travel distance and influence of route closures on the generated paths. The model has been developed based on datasets on the extent of three provinces in Colombia (around 130‘000 km2). It can be reproduced in any country and at scales ranging from the national to the sub-national level. Although the study described in this PhD thesis is mainly focusing on victim assistance, NAMA could find useful applications in other topics such as food delivery management, planning of in-field demining operations, and prioritisation of road clearance. MASCOT (Multi-criteria Analytical SCOring Tool) is a scoring tool based on spatial analysis. The principle of this tool is to score vector items (e.g. mine hazards) in function of their Euclidean distance to a set of predefined factors (e.g. land use categories, human settlements, urban areas, distance to roads and infrastructures, presence of medical facilities etc.). Scores decrease linearly with distance, providing users with a close-to-reality representation of potential impacts of features of the real world upon each other. To

175

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

assist users during the process of weighting criteria, the AHP was integrated with MASCOT. The tool might find useful applications in the mine action community to help (1) directors of national mine action authorities determine areas with high surveying or clearance priority and (2) donors take appropriate decisions about which areas to fund in priority and (3) operations officers select suitable sites for setting up of a base camp in demining campaigns. An overview of the three approaches is provided in Table 24.

Table 24: Overview of the three GIS-based tools developed for helping decision-making in mine action Reference GIS-based Inputs from Combination of Typical use case in mine Outputs GUI in this tool users the input data action thesis Raster: each - Evaluating the demining pixel has one ArcMap - Vector data1 Rasters are capacity of a country ordinal degree of (ModelBuilder 5D - Raster data applied a cell-by- - Selecting the right Section 3.2 demining programming - Weighting2 cell weighted sum demining technique for difficulty environment) deployment in a region

- Determining suitable Vector: origin- location for building a new Minimisation of destination ArcMap medical facility to improve Vector data (road time or distance matrix paths (ModelBuilder NAMA medical care to victims of Section 3.3 network) between incidents between programming ERW and facilities locations and environment) - Prioritisation of road facilities clearance - Vector data (= - Items are - Prioritising areas for items to be weighted ArcMap (Visual surveying or clearance scored) - Items are scored Vector: each Basic Dot Net - Selecting suitable sites MASCOT - Vector & raster in function of scored item has Section 3.4 programming for setting up of a base data (= scoring their Euclidian one score environment) camp in demining objects) distance4 to campaigns - Weighting2,3 scoring objects 1Vector data are converted to rasters before being scored 2Each input layer is a ―factor‖ or ―criterion‖ and is assigned a weight 3Each category in the input layer is a ―sub-criterion‖ and is assigned a weight 4 For vector layers, scores decrease linearly with distance

We see numerous reasons why 5D, NAMA and MASCOT might help humanitarian demining actors improve decision-making in their everyday work. These reasons are exposed below:  Integration of new geospatial factors and complex data models The capacity of the three tools to integrate data from various sources and domains opens demining processes (e.g. task prioritisation, demining capacity assessment etc.) to new geospatial factors and complex data models. With these models, network connectivity, soil characteristics and other factors that have significant influence on demining activities, are best taken into account. Similarly, environmental, geographical, socio-economic and cultural specificities of mine-affected regions are well integrated, bringing significant accuracy and consistency to decision-making.

176

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

 Flexibility and multi-scalability Users interact with the tools through a set of parameters (e.g. study area, criteria and weights). These parameters can be modified at each execution of the tools, making it possible to test various scenarios, possibly at different scales26 before making decision.  Participatory platforms Integrating complex data models supposes that users dedicate significant time and efforts to data preparation, integration, combination and processing, and to the interpretation of results. As a response to this, we see 5D and MASCOT as real participatory platforms around which stakeholders from diverse domains of competence and with complementary skills (e.g. data quality analysts, GIS experts, operations officers, representatives of national mine action authorities, elected officials etc.) can gather their efforts with a common goal. Quality and accuracy of final decisions might greatly benefit from this pooling of expertise.  Centralisation of workflows The three models presented in Chapter 3 were designed and developed with the intention of letting users focus on modelling and decision-making, rather than on software handling. For this reason, they were entirely integrated with the same GIS platform – ArcGIS Desktop– and using them does not require switching from one technology to another. It is also possible for users to achieve a complete workflow, from data preparation to the final decision-making process, using the same map document file.  User-friendliness To reduce the time spent on software handling, effort was put on providing simple and intuitive GUIs and detailed guidance. A typical MASCOT workflow comprises a number of steps that are ordered intuitively and logically. The MASCOT GUI contains about one hundred tool tips, warnings and error messages that are meant to assist users during the scoring process. User guide and test data are also included in the MASCOT installation package. For their part, 5D and NAMA benefit from the ArcGIS web help and their GUI is reduced to a unique dialog box.  Integration of the AHP The AHP was integrated in MASCOT (1) to help users decide how they want to weight the different criteria and (2) to point out possible inconsistencies while weighting. If some inconsistent weighting is applied, the algorithm will suggest users to revise their choices before continuing the scoring process.  Generation of ready-to-map outputs 5D generates a raster depicting degrees of demining difficulty, NAMA determines routes between incidents and facilities and MASCOT creates a vector layer with scores. These outputs

26 5D and NAMA are relevant at the national and sub-national levels while MASCOT will typically be used at scales ranging from the global to the local level

177

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

can directly be mapped to provide stakeholders with a visual overview of the situation in their area of decision-making.  Connections with topics outside the core mine action The above-mentioned outputs are composed of a single layer. This layer might be (1) overlaid with other information (e.g. administrative boundaries, base maps and geo-referencing information) to improve decision-making and (2) combined with other type of information (e.g. socio-economic data) to visualise the spatial relations with topics outside the core mine action field.  Satisfactory computing performances Following the objective of integrating complex data models and processing them at small scales, effort was put on developing performing tools, and good performances were actually obtained. For example, it takes 30 minutes to calculate the operational difficulty layer on the entirety of Mozambique (800‘000 km2-area), on the basis of seven criteria and with a 200-m resolution. NAMA can process more than 25‘000 road features in a few minutes. For the reasons exposed above, we see in 5D, NAMA and MASCOT a great potential for helping humanitarian demining actors improve decision-making in their everyday work. More generally, because their ease of use, their flexibility and their generic nature, NAMA and MASCOT might as well be beneficial to a broad community of scientists and experts, operating in various domains and at various scales and characterised by highly heterogeneous needs, data, resources and GIS knowledge. 5D, NAMA and MASCOT should now meet their audience. To increase the chances that the tools capacities will be best exploited in the medium and long run, we suggest:  Conducting proof of concept studies This is an essential step towards possible acceptance of 5D, NAMA and MASCOT as standards tools in users‘ everyday work. We suggest conducting proof of concept studies in pilot countries with different GIS capacity, for example one country with low GIS capacity, one with medium GIS capacity and one with high GIS capacity. Prototypical countries with different geographical, environmental and socio-economic background should be selected.  Maintaining the tools 5D, NAMA and MASCOT should be maintained with future versions of ArcGIS. The models were integrated with ArcGIS because most mine action users have chosen this software. The development of new versions of the tools will provide the opportunity to improve the existing functions and to develop new ones. As for MASCOT, we already envisage to improve reusability capacity (e.g. by offering the possibility of looping over comparable scenarios within a single run), to reduce the interaction between the user and the machine (e.g. by scoring from an XML file) and to provide more realistic models (e.g. by integrating least cost distance functions as a complement to the traditional Euclidian distance functions).  Optimising GIS data access

178

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

5D, NAMA and MASCOT were designed to be flexible and capable of integrating a large variety of scenarios at different levels of scale (5D and NAMA are relevant at the national and sub-national levels while MASCOT can be used at scales ranging from the global to the local level). The models were calibrated from a set of data representing up to twenty vector and raster layers over four countries, as summarised in Table 25. Some of them can be downloaded for free on the Internet at the global level.

Table 25: Test data used to develop 5D, NAMA and MASCOT GIS-based Test data Source Study area tool Ground softness FAO soils1 Hydrology HydroSHEDS1 Hydrology GLWD1 Afghanistan 5D Land cover GlobCover1 Cyprus Points of interest OpenStreetMap1 Mozambique Roads OpenStreetMap1 Slope SRTM1 Programa de Acción Integral contra ERW accidents las Minas Antipersonal (PAICMA) NAMA Roads network UN OCHA Colombia3 Medical centres OpenStreetMap Human settlements OpenStreetMap Roads UN OCHA Mine Action Coordination Centre of Afghanistan (MACCA) ERW accidents Programa de Acción Integral contra las Minas Antipersonal (PAICMA) Internally Displaced UN OCHA Persons (IDPs) Land cover GlobCover1 Afghanistan MASCOT Medical Centres OpenStreetMap Colombia Population density LandScan2 Roads OpenStreetMap1 National Roads Institute of Roads Colombia Schools UN OCHA Slope SRTM1 1 Freely downloadable on the Internet at the global level 2 LandScan 2008 is free of charge for educational purposes. LandScan 2010 is payable 3 3 provinces: Antioquia, Caldas and Cordoba

Other datasets than the ones listed in Table 25 are good candidates as inputs for 5D, NAMA and MASCOT. For example, weekly or daily precipitation heights coupled with soil water storage capacity are likely to influence preparation of a mechanical demining operation in the field. Factors such as scent contamination and temperature gradients might influence animal detection. Conflict zones could be integrated into a MASCOT model for prioritising areas to be cleared or surveyed. Do such data exist in the study area? If so, do mine action users know where to find

179

Chapter 3: What are the contributions and limits of GIS for improving decision-making in mine action?

these data? Are they available on the Internet? Should they be requested on a case by case basis from governmental organisations? From local mapping agencies? From partners? Are they free of charge? Are they ready-to-use at the desired scale and extent?  Reducing the time spent on data preparation Data preparation is a significant chain link in a typical 5D, NAMA or MASCOT workflow. Examples of data preparation are: extraction to the appropriate format and to the desired extent, conversion to the required geometry type (point / line / polygon), re-projection to the appropriate coordinate system, aggregation at the desired geographical or administrative level, removal of duplicates, application of attribute-based or spatial filters, and correction of topological errors. Depending on data quality, preparing large amounts of data might in some cases presents a major constraint for users with limited time or with limited GIS expertise. We experimented this during the development of NAMA: more than 50% of the time spent on the project was devoted to finding and preparing the data. Reducing this ratio would allow to focus more on GIS analysis and decision-making.

Through the rest of this thesis, thematic geospatial data with high resolution and quality will be called ―best available data‖. Such data might be downloadable from the Internet, or be obtained from governmental authorities, mapping organisations or partners. In some cases, they would require additional work for an optimal use. We assume that making users gain in GIS capacity can significantly reduce the time spent on finding and/or improving quality of ―best available data‖. The next chapter of this thesis is entitled ―How to best build GIS capacity in mine action?‖ In this chapter, we attempt to provide solutions for improving access to GIS information and technologies for non-experts users, and for expanding their GIS knowledge. On the basis of the needs described in the introduction (see Section 1.4), we (1) develop an ArcGIS extension to help users optimise data and map preparation for their GIS workflows, (2) publish an E-learning solution on GIS for humanitarian demining, and (3) develop methodologies and tools for sharing geospatial information with the mine action community and more widely with the GIS community as a whole in a standardised, comprehensive and efficient way.

180

Chapter 4. How to best build GIS capacity in mine action?

Contributing research papers

 Lacroix P., de Roulet P., Ray N. (2014). Simplified Toolbar to Accelerate Repeated Tasks (START) for ArcGIS: Optimising Workflows in Humanitarian Demining. Accepted for publication by the International Journal of Applied Geospatial Research  Giuliani G., Dubois A., Lacroix P. (2013). Towards OGC Web Feature and Coverage Services performance testing: towards efficient access to geospatial data. Accepted for publication by the Journal of Spatial Information. Available from: http://www.josis.org/index.php/josis/article/view/112

181

Chapter 4: How to best build GIS capacity in mine action?

4.1. Introduction

As an introduction to Chapter 4, we analyse the requirements for best building GIS capacity in the mine action community. First, an optimal use of GIS models such as 5D, NAMA and MASCOT requires improving the access by users to ready-to-use geospatial data and reducing the time spent on data preparation. Users do not necessarily know where to find ―best available‖27 auxiliary28 data. If they do, nothing guarantees that these data can be used out of the box. In some cases, they would require additional work for an optimal use, for example correction of topological errors or aggregation to the desired geographical or administrative level. Second, the design of maps from in-field data often requires complex GIS workflows including, for instance, data extraction from IMSMANG, conversion of coordinates (or distances and bearings) stored in Excel sheets to different types of geometry (points, lines or polygons), re-projection to the appropriate coordinate system (e.g. from WGS 84 to UTM), clipping to the desired extent and integration of meaningful cartographic elements (e.g. scale bar, title, caption and base maps). Optimising GIS workflows might allow users to gain significant efficiency, as might increasing users GIS expertise. Third, a successful use of the visualisation methods introduced in Chapter 2 requires providing the technology that can support easy dissemination of maps for users inside and outside the core mine action (e.g. donors, NGOs and private demining companies). It also depends on the will of national mine action authorities to show contamination data in an ―open sharing spirit‖ (Masser 2005) and without under- or over-estimating the picture of contamination in their countries. Another requirement for best building GIS capacity is to encourage standardisation of GIS processes. An appropriate reading and interpretation of geospatial mine action information by users inside and outside the humanitarian demining community requires designing understandable maps, based on standard cartographic rules. It also requires supplying comprehensive metadata. This includes how and when the data was collected, its type (e.g. AP, AV, BAC etc.), its status (e.g. active or closed), some information on data confidence (estimated or calculated ERW area), the sample size, a title, an abstract, keywords, contact details and coordinate system. Standardising this information might allow users from dozens of countries talk a common language and readily understand each other‘s work. Finally, geospatial information should be quickly accessible. Hundreds of maps might potentially be displayed at the same time over dozens of mine-affected countries. Potential map consumers include the directors of national mine action authorities, in-field workers (e.g. operations officers in a mine action

27 ―Best available‖ were introduced in Section 3.5 as geospatial data with high resolution and quality, e.g. thematic geospatial data that can be downloaded for free on the web or GIS data maintained by local mapping agencies 28 ―Auxiliary data‖ were defined in the introduction as data that do not belong too the core mine action domain, e.g. roads, land use types, digital elevation models, population settlements, climate data etc.

182

Chapter 4: How to best build GIS capacity in mine action?

programme and first responders of emergency situations), the private contractors, governmental and non- governmental organisations, academic institutions, international donors and the general public. In extreme situations (e.g. during a conflict) hundreds or thousands of these users might want to visualise and/or to retrieve information simultaneously. Spatial Data Infrastructures (SDIs) can partly or totally respond to most of these requirements. SDIs are collaborative platforms capable of delivering geospatial information from many different sources to the widest possible group of potential users (Coleman et al. 1997), regrouping governmental, non- governmental, commercial, non-profit and academic institutions as well as individuals. SDIs have become increasingly popular over the past years. Well-known examples of SDIs at the global and continental scale are the Global Earth Observation System of Systems (GEOSS29) (Geo Secretariat 2005), the United Nations Spatial Data Infrastructure (UNSDI) (Henricksen 2007) and the Infrastructure for Spatial Information in the European Community (INSPIRE) (European Commission 2007a). Such platforms are seeking to relate data providers to public and private audience with the goal of enhancing and improving decision-making (Giuliani 2011). SDIs are based on the concept of collaboration and partnership, and are meant to facilitate production, management and integration of consistent and readily accessible data (Samadi Alinia and Delavara 2011, Rajabifard and Williamson 2011). SDIs make it possible to reduce the time spent on acquiring and maintaining data (Mansourian et al. 2006) and easier to handle vast amount of data, as heterogeneous as it may be, at various scales, from the global to the local level (Crompvoets and Bregt 2004, Ezigbalike 2004). Following the Open Geospatial Consortium (OGC30) interoperable standards31 allows users with limited GIS capacity to exchange geospatial information and fosters bypass of institutional arrangements and policies (e.g. copyright) that may restrict access to data and metadata (Giuliani 2011). Finally, SDIs contribute to increasing efficiency in users‘ everyday work, and beyond, to reducing costs (Nebert 2007, Henricksen 2007). With this in mind, we propose the development of SERWIS, a SDI for mine action to address the needs of the mine action community for GIS information. SERWIS was introduced in Chapter 1 as a global geospatial platform that would allow users inside and outside the core mine action domain to access, share and/or maintain geospatial information as well as GIS methods and tools. The content of SERWIS will be described in detail in Section 4.5. The two primary goals of SERWIS are, in one hand to serve data and maps showing mine action information while preserving data confidentiality, and in the other hand to improve the access of humanitarian demining actors to geospatial data. To help these users efficiently prepare and process data and produce maps, SERWIS will also provide a free access to various types of customised GIS material, including tools like START (Simplified Toolbar to Accelerate Repeated Tasks) and the ―GIS for Humanitarian Mine Action‖ online course. START is an ArcGIS

29 http://www.earthobservations.org/geoss.shtml 30 http://www.opengeospatial.org 31 Key concepts like Web Map Service (WMS), Web Feature Service (WFS), Web Coverage Service (WCS), ISO 19115 and ISO 19139 will be defined in Section 4.5.1

183

Chapter 4: How to best build GIS capacity in mine action?

Desktop extension developed for optimising GIS workflows in humanitarian demining. ―GIS for Humanitarian Mine Action‖ is an E-learning course specifically designed for creating maps supporting land release efforts. A paper on START was accepted for publication by the International Journal of Applied Geospatial Research (see Section 4.2) while ―GIS for Humanitarian Mine Action‖ was published in 2011 on the Esri website (see Section 4.3). To facilitate the design by users of comprehensive cartographic products by users, we update, in parallel to the SERWIS project, the existing collection of mine action symbols (GICHD 2005). This collection integrates new topics and methodologies developed over the past years. We propose this revised version as a standard and common visual language for IMSMANG and ArcGIS users (see Section 4.4). Finally, the successful implementation of the SERWIS platform highly depends on its capacity to provide responsive web services to ensure an efficient access to data and maps. In order to track latencies, bottlenecks and errors that might negatively impact the overall quality of data visualisation and retrieval through web services, we propose in Section 4.6 an approach for measuring performance of web services. This approach comes along with a set of recommendations for data providers to improve the quality of their services. It is based on the benchmark of a set of vector and raster data characteristics including geometry type, number of features, data resolution, complexity and number of attributes, field indexation and formats for both inputs and outputs. The benchmark simulates two representative groups of potential users of SERWIS: (1) one organisation with less than 16 concurrent requests, and (2) a set of 1‘600 users retrieving data simultaneously to simulate, for example, a rush to the web service that might occur after a catastrophic event. This work, entitled ―Testing OGC Web Feature and Coverage Services performance: towards efficient delivery of geospatial data‖, was published in the Journal of Spatial Information Science. The key ideas of Chapter 4 are summarised in Section 4.7.

184

Chapter 4: How to best build GIS capacity in mine action?

4.2. Optimising GIS workflows

Based on: Simplified Toolbar to Accelerate Repeated Tasks (START) for ArcGIS: Optimising Workflows in Humanitarian Demining

Pierre Lacroixa,b,d, Pablo de Rouletc, Nicolas Raya,b,d a University of Geneva, Institute for Environmental Sciences, enviroSPACE Lab., Battelle – Building D, 7 route de Drize, CH-1227 Carouge; b University of Geneva, Forel Institute, 10 route de Suisse, CH-1290 Versoix; c Geneva International Centre for Humanitarian Demining, 7bis, avenue de la Paix, P.O. Box 1300, CH-1211 Geneva 1; d United Nations Environment Programme, Division of Early Warning and Assessment, Global Resource Information Database – Geneva, International Environment House, 11 chemin des Anémones, CH-1219 Châtelaine This paper appears in ‗Simplified Toolbar to Accelerate Repeated Tasks (START) for ArcGIS: Optimising Workflows in Humanitarian Demining‘ edited/authored by the International Journal of Applied Geospatial Research. Copyright 2012, IGI Global, www.igi-global.com. Posted by permission of the publisher.

4.2.1. Abstract

In this paper we present START (Simplified Toolbar to Accelerate Repeated Tasks), a new, freely downloadable ArcGIS extension designed for non-expert GIS users. START was developed jointly by the Geneva International Centre for Humanitarian Demining and the University of Geneva to support frequent workflows relating to mine action. START brings together a series of basic ArcGIS tools in one toolbar and provides new geoprocessing, geometry and database management functions. The toolbar operates as a bridge between non-spatial repositories (e.g. MySQL and Excel) and GIS. It also connects mine action professionals recording data in the field to GIS experts and improves data interoperability between GIS professionals working in different disciplines. Originally created to help humanitarian demining actors optimise GIS workflows and be more efficient in their everyday work, the toolbar might also benefit scientists operating in other fields.

Keywords: Geographic Information Systems, ArcGIS, Efficiency, Interoperability, Humanitarian Demining

4.2.2. Introduction

According to the International Campaign to Ban Landmines (ICBL), 72 states and 7 non-internationally recognised territories were confirmed or suspected to be mine-affected as of August 2011 (ICBL 2011a). Within the framework of mine action, the Geneva International Centre for Humanitarian Demining, a

185

Chapter 4: How to best build GIS capacity in mine action?

non-profit foundation established by Switzerland and several other countries in 1998, strives to eliminate Explosive Remnants of War (ERW) and reduce their humanitarian impact. In cooperation with its partners, the GICHD provides capacity-development support to national and local authorities in affected countries to ―efficiently plan, coordinate, implement and monitor safe mine action programmes‖ (GICHD 2012). In addition, the GICHD supports the implementation of relevant instruments of international law, such as the anti-personnel Mine Ban Treaty. Countries that have signed the treaty are obliged to collect, analyze and report spatial data on mine action. This data allows the directors of national mine action authorities to provide an overview of their work to the global mine action community and to donors, and enables operations officers to prioritise and access the areas to clear (Benini et al. 2003, Yvinec and Renaissance 2005). Each year, hundreds of thousands of records are collected in the field, many of them with GPS and mobile digital tools (Dunbar 2010). This data is registered in a RDBMS called ―Information Management System for Mine Action – Next Generation‖ (IMSMANG) (GICHD 2011), an ArcGISTM Engine-enabled self-contained information system. The data is stored in MySQL format in the form of 2D coordinate pairs. As of 2012, 3‘500 users have been trained to use IMSMANG in more than sixty mine-affected countries. To perform spatially-explicit analysis, most of these actors are using ArcGIS in conjunction with an IMSMANG data server. Users‘ needs include performing common GIS tasks such as data extraction from non-spatial repositories, conversion to GIS formats, georeferencing, and visualisation. These tasks are typically repeated many times given the amount of data and updates. Users do not necessarily have GIS expertise or experience, and computer literacy is sometimes limited, as well as financial resources. Therefore, these users should be provided with simple, comprehensive, readily accessible and free add-ins. While ArcGIS is one of the most powerful and comprehensive platforms for managing and analyzing geographic information (Hilton 2007), its complexity tends to discourage new users. However, the scripting capacity of ArcGIS permits the development of toolbars (or toolboxes) that can extend ArcGIS capacities and/or greatly facilitate the use of existing functionalities. Examples include toolboxes related to particular topics such as topography (Dilts 2010), CAD (Kuehne 2005), and marine geospatial analysis (Roberts 2009). More generalist toolboxes also exist, such as ET Spatial Techniques (Tchoukanski 2009), XTools (Data East 2012), the ArcGIS Workflow Manager extension (Esri 2012) and the Geospatial Modelling Environment suite (Spatial Ecology LLC 2012), which replaces Hawth‘s Tools (Beyer 2004) with version 10 of ArcGIS. ET SpatialTechniques proposes one hundred functions located in seven different toolbars, but not all tools are free. XTools and ArcGIS Workflow Manager are commercial products. The Geospatial Modelling Environment suite is free but requires advanced GIS skills. To address the specific needs of the mine action community, we developed START (Figure 46), which brings together a series of existing and new ArcGIS tools that are frequently used in typical workflows. START allows efficient access to the most commonly used tools, provides a series of functionalities for spatializing tables and managing vector and raster geospatial data, as well as making it possible to easily

186

Chapter 4: How to best build GIS capacity in mine action?

design maps. The toolbar aims both to assist novice users of GIS technologies and to enhance the productivity and comfort of GIS professionals. In what follows we start by presenting the toolbar and its functionalities, as well some thoughts on the way we have gathered user feedback. We then present a possible workflow for the mine action community, and finally discuss how START may benefit users in other fields.

4.2.3. The Toolbar

Figure 46: START and its seven drop-down menus. Tools that have no direct equivalent in ArcToolbox are indicated with a star (*) symbol

START is composed of seven drop-down menus (Figure 46). Five of them (Zoom, View, Insert, Graphics, and Print) compile existing ArcMap tools that are usually found in six different standard toolbars. The gathering of tools simplifies and speeds up viewing, layout design and printout tasks. These tools are placed from left to right on START, echoing the procedure and steps of viewing data, making a map, and printing the layout. Simple layout templates are also provided to help with map design. The Tools menu (Figure 46) of the toolbar presents fifteen tools. Six of them are related to projection and clipping, and they call directly the existing ArcToolbox equivalents. The other nine tools help to manage data in any format readable by ArcGIS: vector and raster data, CAD formats, Excel files and other tabular information. Outputs of these tools can be shapefiles (SHP), geodatabase feature classes (e.g. MDB, GDB, and ArcSDE), TIFF, ESRI GRID or TXT files. These nine tools are new in the sense that there is no direct equivalent in ArcToolbox: (1) some of them were developed from scratch, (2) some are customisations of existing tools to users‘ needs, and (3) some combine several tools into one. These nine tools were originally designed to assist demining planners in the management of geographical information. Some of these tools help users to calculate points, lines and surfaces from either GPS coordinates, or distances and bearings of a set of points (GICHD 2008b, GICHD 2010b). In humanitarian demining, these points, lines and surfaces typically represent accidents, victims, or minefields. Two tools help to extract information from a point layer and write it in the attribute table of the dataset: Extracting

187

Chapter 4: How to best build GIS capacity in mine action?

XY Fields calculates point coordinates in any spatial reference, while Extracting Distances and Bearings calculates the distances and bearings between the points of a layer. Extracting XY Fields goes one step further than the standard ArcGIS equivalent, as it (1) generates a field with the projection name and (2) has a more explicit and detailed help function. In line with the idea of simplifying a typical workflow, two tools are specifically designed to transform Excel files (or any tabulated file accepted by ArcGIS, such as TXT and CSV files) into point layers: xls File Coordinates to Points generates a point layer from an Excel sheet with X/Longitude and Y/Latitude information columns, and xls File Distances and Bearings to Points does a similar task from a table that stores distances and bearings between consecutive points. Three other tools manage the conversion of polygon or polyline vertices into points, as well as the conversion of points into vertices. The Points to Polylines and Points to Polygons tools are designed to generate linear and polygonal objects from a series of ordered points. The Polygons or Polylines to Points tool enables the reverse function, namely transforming vertices into a point dataset. This tool is more complete than the standard ArcGIS equivalent as it generates fields storing the original feature‘s identifier, the number of the point, XY coordinates and the name of the projection, allowing the reverse operation if needed. Points to Polylines also processes features with an identifier for the line part. Points to Polygons does not have an equivalent in ArcToolbox. The Elevation to Slope tool produces slope rasters from Digital Elevation Models (DEM). This tool is crucial for mine action planning, e.g. for determining the demining technique that will be used on a terrain. Slope rasters are also regularly used for other types of risk analysis, such as predicting which areas are most likely to be affected by landslides (Chau et al. 2004). To avoid confusing non-expert users, Elevation to Slope is a simplified version of the existing ArcToolbox equivalent: (1) it only accepts DEM in World Geodetic System 1984 (WGS 84), such as Shuttle Radar Topography Mission data (Jarvis et al. 2008), and (2) the calculation of the Z factor parameter (the slope correction factor used when XY ground units are different from Z units) has been hardcoded using the average latitude of the input DEM. Finally, the MySQL Connection tool is included to convert lists of X/Longitude and Y/Latitude coordinates stored in MySQL databases into point, polyline or polygon shapefiles, through an ODBC connection. This tool was originally developed to provide a bridge between non-spatial data recorded in IMSMANG and ArcGIS. However, it can be used in other fields that require connecting MySQL data to GIS.

4.2.4. Tool Improvement and User Help

The tools in START require valid inputs (e.g. in terms of projection, format) to ensure valid outputs. To facilitate identification and understanding of the potential problems relating to inputs, an important effort was directed towards the development of the help functions. Warning and error messages, as well as interactive tool help, assist users throughout the various functions. A fifty-page user guide describes

188

Chapter 4: How to best build GIS capacity in mine action?

several workflow scenarios for novice users, in addition to detailed explanations on how to use each tool of START. The improvements of the user interactions with the tools and the development of the help functions were made possible through a two-day end-user workshop. START was tested on different versions of Windows, with self-made data and IMSMANG data. A dozen mine action experts with varying levels of GIS expertise and representing different domains of competences in mine action (information management, national programme management, operations, and database administration) were involved in the workshop. We see this panel of experts as representative of the future community of START users. Particular attention was paid to their feedback and their perception of the toolbar: appropriateness for their specific needs, ease of use, computing performances, quality of the help functions, and user interface. These users were also asked to suggest improvements and propose new functions and tools. Following this survey, planned improvements include a function to calculate security perimeters based on convex hulls. In addition, the xls Distance and Bearings to Points tool, today limited to decimal degrees (DD) inputs, will soon support degrees and decimal minutes (DM), degrees minutes seconds (DMS) and Military Grid Reference System (MGRS) formats.

4.2.5. Mine Action Scenario

START supports commonly used workflows for mine action. We illustrate in Figure 47 one possible workflow in which mine action data with different extents and different coordinate systems (WGS 84 and Universal Transverse Mercator (UTM)) are combined with a slope raster to determine appropriate demining techniques. The input data of the workflow (Box A in Figure 47) include a DEM, a MySQL view of ERW polygons and an Excel sheet containing the distances and bearings of ERW polygons. In the data preparation phase (Box B in Figure 47) the DEM is clipped with the Clip Raster tool to fit to the extent of a study zone. Slopes are then computed with the Elevation To Slope tool. The MySQL view of polygon data is transformed into a polygon layer using the MySQL Connection tool. In parallel, the xls File Distances and Bearings To Points tool is used to transform the Excel data into a point layer that is projected to WGS 84 with the Project Vector tool. With the Points To Polygons tool, these points are converted to polygons. The Layout view mode is used for mapping (Box C in Figure 47) while other drop- down menus, such as Graphics, Insert and Print, are employed for design and printing. At this point, a standard GIS workflow for mine action is completed. Further steps might include using the maps to support additional GIS analysis or to facilitate the decision-making process. The output is designed to help national mine action authorities establish the complexity of clearance operations. This can help professionals working in this field to choose appropriate demining techniques and identify areas with high clearance priority.

189

Chapter 4: How to best build GIS capacity in mine action?

Figure 47: A possible workflow for humanitarian demining actors using START. The large boxes (A, B, and C) represent major steps in the workflow. The grey boxes are input or output of the various sub- analyses. START tools used in this workflow are indicated between the boxes with their respectives name and icon. See main text for explanations

4.2.6. Alternative use of START outside the Mine Action Community

START was originally created to support frequent workflow scenarios relating to mine action, but the toolbar can advantageously be used in other fields. One example is in hydrology and watershed modeling, where the toolbar was shown to facilitate the data set up before its integration into a hydrological model. START can accelerate the extraction of XY coordinates of hydrological and meteorological stations, the obtaining of the slopes from a DEM file, and the definition of layers‘

190

Chapter 4: How to best build GIS capacity in mine action?

projection (e.g. river network, stations, soil types and slopes). Further GIS analysis does not necessarily imply the use of START, which might occur during the phase of mapping. Emergency mapping is another example where quick evidence-based decisions are required from first response teams. Immediately following catastrophic events such as conflict or natural disaster, the infrastructure required to assist people in need is often disrupted, damaged or completely destroyed. First response Information Management personnel are asked by non-governmental organisations (NGOs) to collect all kinds of geospatial information from the field, such as water points, boundaries of refugee camps, delimitation of the different ethnic groups, transportation networks and utilities (Cova 1999). This information is typically recorded with GPS or mobile devices and stored as point coordinates in tabular files (Savopol and Armenakis 2002). With START, these points can be rapidly converted into lines (in the case of networks or piping) and polygons (for boundaries and delimitations) before being mapped. By gathering the required tools in a single toolbar and by offering interactive and detailed tool help, START can significantly reduce the time spent on software handling and help stakeholders to rapidly build maps, as well as keep them up-to-date within rapidly evolving crisis situations. A third example is in ecology where field scientists typically record the location of samples, including single point observations, field surveys as polygons, transects as polylines, in Excel files or databases (Funk et al. 1999). START can help these users to rapidly prepare their data and map them in conjunction with background information from either raster files (e.g. DEM, land cover) or vector files (e.g. road network, species home range). More generally, START can be useful for any mapping task that involves measurement with devices such as GPS, theodolites and laser rangefinders and storage of point coordinates in tabular files such as Excel or MySQL. Clip, projection, and layout are also universal GIS operations that can be accessed directly from the drop-down menus, enabling the rapid transfer of information to decision-makers.

4.2.7. Conclusion

START aims to serve as a bridge between professionals recording data in the field and GIS experts. The toolbar improves access to GIS technologies for non-expert users and data interoperability between GIS professionals working in different fields. The toolbar is currently being used by several institutions and the initial feedback from users indicates that START is able to respond to important needs in the field of GIS for humanitarian demining. The development of future versions of START will provide the opportunity to improve its performance and to create new tools, which we hope will be of benefit to the mine action community and ultimately assist with clearance in mine-affected countries. START can be downloaded for free at: http://www.unige.ch/sig/outils/StartToolbar.html, including installation setup, user guide tutorial and sample data. Software requirements: ArcGIS Desktop 10.1.x or ArcGIS 9.3. Maintenance of the toolbar is planned for the future. Elevation to Slope is the only tool of START that requires a commercial license (Spatial

191

Chapter 4: How to best build GIS capacity in mine action?

Analyst or 3D Analyst) in addition to the ArcGIS Desktop license. MySQL 5.5 with ODBC driver is necessary for using the MySQL Connection tool.

192

Chapter 4: How to best build GIS capacity in mine action?

4.3. E-learning

Collaboration between the Geneva International Centre for Humanitarian Demining, Esri Inc., the University of Kansas and the University of Geneva. GIS for Humanitarian Mine Action. Online training available from: http://training.esri.com/gateway/index.cfm?fa=catalog.webCourseDetail&courseid=2065

Maps perform a critical role for the efficiency and the safety of humanitarian demining activities. Ambiguous or confusing symbols may lead to misunderstanding of maps resulting in severe consequences for civilians or on demining personnel (GICHD 2005). Designing easy to understand maps does not only require using unambiguous symbols, but also relevant data, appropriate coordinate system and meaningful layout. ―GIS for Humanitarian Mine Action‖ is a free web course that teaches the basics for creating maps to support land release efforts. The course was developed with the purpose of building GIS capacity of organisations involved in mine action through training of individuals. Based on ArcGIS Desktop, it targets individuals involved in humanitarian demining who need to work with GIS maps in the field. It may also sensitise users outside the core mine action domain to the problem of contamination by landmines. The course is composed of 8 modules of 3 hours each. It alternates theoretical lessons with practical exercises, on the basis of mine action-oriented datasets. After completing the course, users will be able to (1) explore geospatial data in the most common GIS formats, (2) learn the fundamental concepts of satellite imagery, (3) load, create and use standard or customised cartographic symbols, (4) design a map layout and use a map template, (5) manage coordinate systems, (6) georeference maps and digitise data, (7) process vector data and (8) process raster data. Each of the 8 modules is followed by an exam with a dozen question marks. Students must answer correctly to 80% of the questions to pass. This course can be a first step for novice GIS users to acquire a basic knowledge in GIS, and a good way for more advanced users to focus on one of the 8 topics described above. It was published on the Esri website in July 2011. As of 2013, about 1‘300 people have enrolled in this course with about 500 people completing the entire course. The course should be maintained with future versions of ArcGIS.

193

Chapter 4: How to best build GIS capacity in mine action?

4.4. Standardising Symbology

4.4.1. Objectives of this research

One major work in the framework of the assessing contribution of GIS to the mine action field is the development and revision of its current cartographic symbology. The ambition of this research is to develop a unique and harmonised set of symbols that would be as universal as possible and intuitive for any demining professional or civilian in any, or most cultures. Three reasons justify this choice. First, the aim of this research is that all users of cartographic material in mine action use the same visual language. A common symbology aims at ensuring a concise and consistent method for marking georeferenced features for information exchange between organisations. Among other advantages, it can be an important time-saver if the different actors do not have to decode and recode the symbols when sharing maps. Second, the needs for a common symbology is made especially clear as ERW contamination affects more than sixty countries in five continents, while dozens of international organisations and thousands of potential users are involved in demining, A common symbology follows the general goal of mine action, to bring actors together in symbiosis, rather than working only on their own side. Third, maps are a form of visual communication, a language for describing spatial relationships. Cartography makes the use of visual resources such as colour and shape for recording and communicating information about the location and spatial characteristics of our environment, society and culture (Cleveland 1994). In the framework of the geographical information management for humanitarian demining, cartography and GIS perform a critical role for the safety and efficiency of mine action operations. Symbology is a key element for the correct interpretation and understanding of the landmine hazards categories and mine action processes. Therefore, ambiguous or confusing symbols may lead to map misinterpretation resulting in severe consequences, including serious injury or death of demining personnel or civilians (GICHD 2005). For this reason, the success of the cartographic communication should entail an accurate translation to map users of the real-world features represented by symbols (Béguin and Pumain 2003). In practice this means on one hand enhancing the quality of the visual suggestions conveyed by the symbols and on the other integrating new operational concepts and processes in mine action. Following some root principles of cartographic representation, this symbology analysis reconsiders and examines the existing IMSMANG symbology (GICHD 2005) by taking in consideration visual and semiotic standards as well as users‘ needs. In this sense, some symbols have been transformed and created incorporating chromatic perception, interactions between colours and most frequent signs associations, and in some cases words. These modifications and new symbols have been designed and integrated into an Esri style file that is technically ready to use in ArcGIS. The creation of this style will allow its use in IMSMANG. The new

194

Chapter 4: How to best build GIS capacity in mine action?

symbology has been validated with an online survey toward the mine action community and presented in a poster for the 14th International Meeting of National Mine Action Programme Directors and UN Advisors. These symbols represent a new step towards standardisation of the mine action symbology. A key challenge for this work of harmonisation is, as pointed by Gerber et al. (1990, p. 96) that ―no graphical symbol can be assumed to be universal‖. It means that along relying to the broadest and most cross-cultural semiotic studies and conventions all changes had to be coupled with consulting concerned professionals familiar with ERW-contaminated countries.

4.4.2. Loose standards for cartography in mine action

Since 2001, the IMAS define several aspects of humanitarian demining, in terms of requirements and/or best practice, ranging from the themes of Land Release to Mine Risk Education. However, there is no official standard defined for cartography by the IMAS. The first ―Cartographic Recommendations for Humanitarian Demining Map Symbols in the Information Management System for Mine Action‖, were developed by the GICHD and the University of Kansas in 2005 compiling different international standard and led to a major change in the IMSMA library of symbols. IMSMA being the most widely used database system for mine action, its embedded symbology can be seen as a de facto standard. Accessory elements, such as airstrips and hospitals, and military symbology are respectively inspired by ISO and North Atlantic Treaty Organisation (NATO) standards (GICHD 2005). But symbology specific to mine action often has to be created from scratch, sometimes inspired by the IMAS. As stated above the IMAS do not define any cartographic standard for humanitarian demining, but some symbols are inspired by the marking standards of the IMAS, including all those related to Hazards (Kostelnick et al. 2008, UNMAS 2003, 08.40). The main transformation conducted with the 2005 Cartographic Recommendations for Humanitarian Demining Map Symbols in IMSMA is the change of most of the icons from acronym-based symbols to figurative icons. It led, for example, to the change from the acronym ―DA‖ (Dangerous Area) to a skull and crossed bones circumscribed in a triangle for the Hazard sign. The replacement of letters by a figurative icon represented an important step in the standardisation of the symbology, removing almost all non-Latin scripts from IMSMA symbology. The UneXploded Ordnance symbol is the only exception, as it is still represented by the acronym ―UXO‖, as no alternative could be found. It also leans on the IMAS recommendations for Hazard signs, as the IMAS advise to warn from mined areas with a skull within a red triangle around the minefield. This choice stems from the fact that red is identified to danger in most cultures32. As several studies on human reactions towards colours as warning codes show, red colour has the strongest power in suggesting danger, especially associated with black, while blue and green colours suggest safety to most subjects (Chapanis 1994). A sign of danger coupled with a red colour not only presents a redundant effect, but increases the perception of the concept of danger (Braun et al. 1995).

32 Although IMAS advises adaptations in the case of countries where the red symbology changes

195

Chapter 4: How to best build GIS capacity in mine action?

Explicitness in the symbology of warning is also generally advised and is believed to help enhance prudence in the general public (Laughery et al. 1993).

4.4.3. The 2011 recommendations

4.4.3.1. Boundaries of our work

In the course of 2011, the previous recommendations have been followed, revised and updated in close collaboration with demining operations specialists, due to the necessity of integrating and elaborating a new symbology that could broadly communicate new elements that need to be integrated in maps. This updated symbology considers the existing inventory of symbols for: Hazard, Hazard Reduction, Land Release, Assistance to Victims and Task, adding in some cases different levels of complexity. As all cartographic fields, mine action uses point, line and surface elements. However, the point element is particularly prominent in humanitarian demining, including for surfaces, for which a point symbol in the centroid of the corresponding polygon is used to identify clearly and instantly the type of area represented. This allows for example to keep track of hazardous areas even at small scales (e.g. country level) where some surfaces would not be visible otherwise. This prominence of punctual symbols has led the choice to start the symbols updates by revising the points. Therefore, all the symbology discussed here will refer to points, whose desired characteristics are to be ―accurately comprehended and easily distinguished from the background map‖ (Clarke 1989).

4.4.3.2. Methodology

The methodology to build the symbology follows the visual variables described by the cartographer and theorist Jacques Bertin. In his seminal work, Bertin defined seven visual variables used in cartography: size, value, colour, texture, grain, orientation and shape (Bertin 1977). Table 26 shows the different variables and their properties and adequateness to display the different type of information in a map.

Table 26: Bertin‘s visual variables and their properties

Data Relationship Visual variables Difference Order Quantity Association Differentiation Size ++ ++ Value ++ + Colour ++ ++ Texture + ++ Grain + ++

196

Chapter 4: How to best build GIS capacity in mine action?

Orientation + ++ Shape + ++ The sign ―++‖ indicates which variables are the most adequate for which type of information. One ―+‖ signify that the variable can be used to a less extent

Four of those variables are used to build the IMSMA point symbols: shape, colour, value and size. Because the size results in change of the dimensions of a symbol, it is used within a category of objects to show their relative importance. The colour variable is used to differentiate both between categories of objects, and within a category of object. It is importantly used to indicate degrees of danger/safety. Shape is used both as an abstract geometric object and as an icon. The value variable is rarely used but notes a quantitative difference in a process. The symbols are created with the help of ArcGIS Style Manager, which facilitates updates and the creation of new items. Since 2005, the structure of the humanitarian demining cartographic symbols is relatively complex and can, for most of the symbols be divided into five elements: the outer shape of the object, the outline, the colour, an inside icon and its size. Those are used to differentiate categories of objects and to differentiate between objects within a category. Outer shape and icon relate both to the shape variable, but in a different fashion, the outer shape is geometric and abstract, whereas the icon can be image-related, concept-related or abstract. As noted by Forrest and Castner (1985), the colour has the strongest visual power, followed by size and then by shape. However, icons convey in turn a strong evocation power, especially if they are explicit enough to allow the user not to need to refer to a legend (Morrison and Forrest 1995). In the same time we can note that a key constraint of icons is their size, which forces strong simplifications. Thus the Victim icon represents, for example a person stepping on a mine. Hazard icons are also explicit either showing a skull and crossed bones for its generic version or the specific type of ERW concerned, such as anti-personnel or anti-vehicle mine. As shown by Wogalter et al. (1998) the skull is the most powerful shape to convey the idea of danger to the public on warning signs. The size variable is used since 2006 to denote the importance of the object represented. Its most important use in the mine action symbology is for the victim category, where three sizes exists to denote whether there is an individual victim, a small group (2 to 4) or a greater group (more than 5) of victims (see Figure 48).

Figure 48: Use of the size variable to identify a number of victims

197

Chapter 4: How to best build GIS capacity in mine action?

4.4.3.3. Achievements

The update of the symbology set of cartographic symbols concerned a sixty symbols, mostly the inside icon of the symbols. It responded partly to the multiplication of the tasks categories in mine action. As mentioned above, the mine action symbology is made of simple geometric shapes representing types of objects, such as hazards and victims. Within these shapes, small icons are used to represent explicitly the type of item concerned. Those are either image-related or concept-related, representing for example an explosion to signify an accident (see Figure 49).

Figure 49: Generic symbols for ten categories of objects

To distinguish between objects within a category, the colour variable was an important tool, as well as the shape. As shown in Figure 50, red is used for danger and the paradoxical green hazard is used to signify that a clearance task was finished, or ―closed‖. Variance in the outer shape is used to signify that a hazard is being cleared, or ―transitional‖ (also alternatively called ―ongoing‖). The grey colour is used as a neutral colour for hazard that have the status ―cancelled‖, considered safe area after a non-technical survey.

Figure 50: The hazard by status sub-category

An important projected change in the symbology relates to the inside icon of the symbols. Some elements were updated to make them fit better with the intended representation. The Location symbol, previously represented by a human character icon was, for example, replaced with a pin inside, instead of someone doing something (Figure 51). It was thought that the concept of location would be better evoked with a concept-related and simple to identify image, than with a human figure.

Figure 51: Changes in the Location symbol

198

Chapter 4: How to best build GIS capacity in mine action?

Research was done to represent new items such as sub-munition, which do not have a specific symbol so far. It is represented in the new recommendations as an icon made from four small triangles. In this regard the new sub-munition icon is a mix of an abstract shape (the triangle) with an image related one, with the multiplication of items, imaging the spatial dispersion typical of this type of munition (Figure 52).

Figure 52: The new sub-munition icon

An important aspect of the symbology update consisted in the updating of the task symbols. The general category of Task is represented by a simplified notepad as its outer shape. Inside icons and colours are used to note specification within this category. Figure 53 shows how colour and value were used to distinguish between the different objects of the category ―Task by Status‖. Green colour is used to indicate that a task is completed, or partly completed to signify safety. On the other hand, the colours orange and yellow indicate that Tasks are ―issued‖ or ―ongoing‖, signalling possible danger. Other cases are shown using values ranging from white to dark grey, as this variable works well to show qualitative difference of order.

Figure 53: Colour and value are used in the Task by Status category to differentiate between objects

4.4.4. Validation process

4.4.4.1. Online survey

Despite disagreements due to the lack of proper IMAS definition for some process, there is already a positive return from several mine action experts. In order to ensure that symbols fit with the users' needs an online survey of users was set to collect their impressions and critiques. To give the survey the widest possible relevance, and ensure that the new symbols can be understood in different cultural environments, fifteen demining professionals polled came from the IMSMA users group, which encompasses specialists from all continents. The results showed that a large majority of those polled found the new symbols ―easy and clear to understand‖ in every case.

199

Chapter 4: How to best build GIS capacity in mine action?

4.4.4.2. Waiting for the IMAS definitions

The current updated symbology has been designed and is now ready to use. However, its use in mine- affected countries is currently blocked. Although most of the symbology fits to a certain consensus among mine action professionals, some new items still do not. This results partly from disagreements over the readability of some icons contained in the symbols, but also because some IMAS definitions have not yet been set. One such example is the issue of tasks and whether they should be called ―transitional‖ or ―ongoing‖. But as long as the definition of the real world element, whether it is a task, a hazard reduction or other is not clearly defined, it will not be possible to design its corresponding cartographic symbol. However, the establishment of the IMAS definitions adapted to new practices for land release in mine action, already stored in drafts and set for review, should clarify the current situation and prepare their implementation into IMSMA.

4.4.5. Perspectives of our research

This updated symbology can already be used in ArcGIS, and as mentioned above only needs validation. This work on point symbols was the most important landmark change since the watershed of 2005 report, when most IMSMA symbols were thoroughly changed. This research for an update of the symbology only concerned the point symbols, as they represented the most urgent need. However, polygon and line symbology could certainly gain in terms of readability with a new update of the library. Polygon symbology is defined for all hazardous areas, such as SHAs or CHAs. Its main characteristics are that the fill colour must be partly transparent in order for the base map to be visible. This stems from the important use that is made by demining professionals of natural features to identify and mark mined areas. Polylines typically represent roads or routes, either as a hazard or as a process. These series of recommendations would make a new improvement for the cartographic standards of mine action. However, they can still be improved and changed in the future if new updates are needed to respond better to the needs of a common visual language for mine action. They are bound to progress and be transformed along with the practice and challenges of mine action. Many users of IMSMANG and GIS in mine action will gain skills in the years to come, which will help countries to customise their own cartography, for example in the case where colours might convey different meaning in their cultures.

4.4.6. Conclusion

The issue of symbology is essential to provide a common language for the end users of cartography in mine action. Such a common language may help communication in the whole chain of mine action professionals, and consequently facilitate demining and enhance safety for deminers. A new symbology also has its role to play in terms of building capacity for IMSMA and GIS users: instead of each country or organisation using its own symbology, a library can help both harmonise and level up capacities

200

Chapter 4: How to best build GIS capacity in mine action?

between countries and users. A ready-to-use symbology can help users to learn fast and efficiently how to produce a map. Finally, a complete symbology signifies also a gain of time for users that they can invest in other GIS tasks.

201

Chapter 4: How to best build GIS capacity in mine action?

4.5. Sharing Data, Maps, Technologies and Processes

4.5.1. Definitions

In this paragraph, we give an overview of basic concepts underlying SDIs, with a focus on web services and metadata standards. The following definitions will help the reader to go through Section 4.5. These definitions are based on the OGC specifications33.

4.5.1.1. Web Map Service (WMS)34

The OpenGIS® Web Map Service Interface Standard provides a simple HyperText Transfer Protocol (HTTP) interface for requesting geo-registered map images from one or more geospatial databases. A WMS request defines the geographic layer(s) and area of interest to be processed. The response to the request is one or more geo-registered map images (returned as JPEG, PNG, GIF etc.) that can be displayed in a browser application. The interface also supports the possibility to specify whether the returned images should be transparent so that layers from multiple servers can be combined or not. WMS is used for mapping purposes and does not give access to the data itself. Here is an example of WMS URL with a GetMap request: http:///arcgis/services/MapServer/WMSServer?&BBOX=42.36252,23.32867,48.3625 2,26.32867&STYLES=&REQUEST=GetMap&VERSION=1.1.1&LAYERS=SERWIS:NetworkAnalysis& WIDTH=600&HEIGHT=300&SRS=EPSG:4326&FORMAT=image/png This request returns to a client a map of selected geospatial layers, on a 6x3 decimal degrees rectangle in PNG format and in WGS 84 coordinate system. The GetMap operation requires various parameters. BBOX allows users to specify the coordinates of the bounding box indicating minimum X, minimum Y, maximum X and maximum Y. In the example above, STYLES has an empty value, which means that the default style provided by the data custodian will be applied. REQUEST with the value ―GetMap‖ is for displaying a map in raster format. VERSION specifies the version of the specification. LAYERS lists the names of layers separated by a comma. In the example above, a single layer has been selected. WIDTH and HEIGHT specify the dimensions of the returned coverage in pixels. SRS identifies the spatial reference system. FORMAT allows users to specify the format of the returned coverage.

33 http://www.opengeospatial.org 34 http://www.opengeospatial.org/standards/wms

202

Chapter 4: How to best build GIS capacity in mine action?

4.5.1.2. Web Feature Service (WFS)35

This OGC standard defines an interface for retrieving and updating geospatial features encoded in Geography Markup Language (GML). Unlike WMS, WFS provides direct access to attributes and geometry of vector data. Similar to WMS, a URL invokes a WFS interface for performing diverse operations relative to data manipulation. There are two types of WFS services: (1) basic WFS for retrieving and/or querying features, and (2) transactional WFS for creating, deleting or updating features. Here is an example of basic WFS URL with a GetFeature request: http:///arcgis/services/MapServer/WFSServer?&BBOX=42.36252,21.32867,47.3625 2,26.32867&STYLES=&REQUEST=GetFeature&VERSION=1.0.0&TYPENAME=SERWIS:AdminLimits &SRS=EPSG:4326 This request returns features representing administrative boundaries in WGS 84 on a 5x5 decimal degrees extent. It is possible to constrain the query spatially and/or non-spatially. The syntax and parameters of the URL are similar to those used in a WMS request. BBOX allows users to specify the coordinates of the bounding box following minimum X, minimum Y, maximum X and maximum Y. In the example above, STYLES has an empty value, which means that the default style provided by the data custodian will be applied. REQUEST with the value ―GetFeature‖ is for retrieving vector data. VERSION is the version of the specification. TYPENAME specifies the name of a single selected vector dataset. SRS identify the spatial reference system.

4.5.1.3. Web Coverage Service (WCS)36

The OGC Web Coverage Service supports electronic retrieval of geospatial data as "coverages" – that is, digital geospatial information representing space/time-varying phenomena. In other words, a WCS provides access to raster datasets in forms that are useful for client-side rendering, as input into scientific models, and for other clients, e.g. DEM and remote sensing imagery. As WMS and WFS service instances, a WCS allows clients to choose portions of a server's information contents based on spatial constraints and other query criteria. Here is an example of WCS URL with a GetCoverage request: http:///arcgis/services/MapServer/WCSServer?&BBOX=42.36252,21.32867,47.3625 2,26.32867 &STYLES=&REQUEST=GetCoverage&VERSION=1.1.1&COVERAGE=SERWIS:Contamination&WID TH=500&HEIGHT=500&CRS=EPSG:4326&FORMAT=Tiff This request returns to a client a 5x5 decimal degrees tile in Tiff format and in WGS 84 coordinate system. The syntax and parameters of the URL are similar to those used in a WMS request. BBOX allows users to specify the coordinates of the bounding box following minimum X, minimum Y, maximum X

35 http://www.opengeospatial.org/standards/wfs 36 http://www.opengeospatial.org/standards/wcs

203

Chapter 4: How to best build GIS capacity in mine action?

and maximum Y. In the example above, STYLES has an empty value, which means that the default style provided by the data custodian will be applied. REQUEST with the value ―GetCoverage‖ is for retrieving raster data. VERSION specifies the version of the specification. COVERAGE specifies the name of a single selected raster dataset. WIDTH and HEIGHT specify the dimensions of the returned coverage in pixels. CRS identifies the coordinate reference system. FORMAT allows users to specify the format of the returned coverage. If the request is sent from a web browser, users can download the returned coverage. If it is sent through a Desktop client (e.g. ArcGIS Desktop), the returned coverage can be used directly into the map document.

4.5.1.4. ISO 1911537

This standard defines the schema required for describing geographic information and services. It provides information about the identification, the extent, the quality, the spatial and temporal schema, spatial reference, and distribution of digital geographic data. ISO 19115:2003 is applicable (1) to the cataloguing of datasets, clearing-house activities, and the full description of datasets, and (2) to geographic datasets, dataset series, and individual geographic features and feature properties. ISO 19115:2003 defines mandatory and conditional metadata sections, metadata entities, and metadata elements. It also specifies the minimum set of metadata required to serve the full range of metadata applications (data discovery, determining data fitness for use, data access, data transfer, and use of digital data). It defines optional metadata elements – to allow for a more extensive standard description of geographic data, if required – as well as a method for extending metadata to fit specialised needs. Though ISO 19115:2003 is applicable to digital data, its principles can be extended to many other forms of geographic data such as maps, charts, and textual documents as well as non-geographic data.

4.5.1.5. ISO 1913938

Complementary to ISO 19115, this standard identifies and defines the architecture patterns for service interfaces used for geographic information, defines its relationship to the Open Systems Environment model, and presents a geographic services taxonomy and a list of examples of geographic services placed in the services taxonomy. It also prescribes how to create a platform-neutral service specification, how to derive conformant platform-specific service specifications, and provides guidelines for the selection and specification of geographic services from both platform-neutral and platform-specific perspectives. Concretely, ISO 19119 specifies a form and a content for the XML document describing the capabilities of a web service.

37 http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=26020 38 http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=39890

204

Chapter 4: How to best build GIS capacity in mine action?

4.5.2. SERWIS services

The technical specifications of SERWIS were designed on the account of the requirements analysed in Section 1.4. They are illustrated in Figure 54. SERWIS relies on a simple architecture and has been designed as a user-friendly and flexible platform, as recommended by Giuliani (2011) and others. SERWIS proposes the following:  A geoportal for mine action For countries that feel comfortable with sharing their information related to mine action, SERWIS can serve density rasters showing contamination by ERW (see Section 2.2). To encourage the sharing of information while preserving data confidentiality, we provide users with customised mapping tools based on the principle of data obfuscation (see Section 2.2). SERWIS can also show maps of populations at risk of ERW (see Section 2.4), rasters holding a degree of operational clearance difficulty (see Section 3.2) and potentially other data and maps in relation to demining activities, based on vector and/or raster data. More generally, mine- affected countries will be able to share the outputs of their GIS analysis (e.g. the outputs of a MASCOT, a NAMA or a 5D process) through web services hosted by SERWIS. To ensure interoperability across platforms (Bernard and Craglia 2005), especially for users that are not equipped with the Esri suite, the possibility of querying and extracting these maps on the desired extent will be given through web services (WMS, WFS, WCS). This compliancy with OGC standards will also ensure an efficient access to spatially distributed data (Bernard and Craglia 2005) by users outside the mine action community (e.g. NGOs operating in other humanitarian fields), and contribute to reducing data duplication (Giuliani 2011). Developed on the ArcGIS Server technology, SERWIS will make it possible for users to document their maps and data with ISO 19115/19139 metadata to facilitate comprehension and reusability of geospatial information.  Links to “best available” auxiliary data SERWIS will provide links to geospatial platforms where ―best available‖ auxiliary data can be downloaded for free (GlobCover, HydroSHEDS, SRTM, OpenStreetMap). SERWIS will also furnish a list of links to existing online web services (e.g. ArcGIS online, Bing maps). This may contribute to developing capacity in mine action GIS, facilitate development of collaboration with partners outside the core mine action domain and open the door to the integration of new activities.  Hosting of mine action data For national mine action programmes that have limited GIS capacity or resources, SERWIS can provide data hosting facilities. Likewise, the platform may host and serve mine action data of closed mine action programmes.  Access to GIS teaching material

205

Chapter 4: How to best build GIS capacity in mine action?

SERWIS will furnish a link to the ―GIS for Humanitarian Mine Action‖ online course described in Section 4.3.  Access to GIS tools for analyzing geospatial data SERWIS will supply links for downloading the tools that were introduced previously in this thesis (5D, NAMA, MASCOT and START). Although 5D and NAMA will first be implemented as desktop applications in the mine action countries, SERWIS might as well serve them as web processing services in the future.

4.5.3. From IMSMANG to the web

One major strength of SERWIS is the integration of customised cartographic39 and GIS functions40 that provide quick and easy transfer of information from national repositories to web map and data services. These functions were designed to allow directors of national mine action authorities showing up-to-date information about contamination and ongoing demining activity in their country. The whole process from data extraction to web publishing, including data preparation, obfuscation, geo-processing and map design is presented in Figure 54, as part of the SERWIS architecture.

39 The Raster Generator cartographic module introduced in Section 2.5 40 START, the ArcGIS extension introduced in Section 4.2

206

Chapter 4: How to best build GIS capacity in mine action?

Figure 54 : SERWIS architecture, based on ArcGIS Server

The IMSMA user (Box A in Figure 54) is based in a country. He/she manages the IMSMANG repository (Box C in Figure 54). The repository is composed of tables, geodatabase feature classes and non- geographic or geographic SQL views. The geographic views are created with graphical database management tools such as Navicat and FlySpeed SQL Query. They support data extraction from IMSMANG for GIS analysis and mapping. In particular, the Raster Generator cartographic module introduced in Section 2.5 will allow the IMSMA user to generate density rasters showing contamination (Box B in Figure 54). Contamination density rasters created with the Raster Generator can be produced on request or at a frequency that depends on updates made in the IMSMANG database. The MySQL Connection function of START (see Section 4.2) queries IMSMANG tables and generates vector data (e.g. shapefiles) from views typically representing hazardous areas, accidents locations and victims. These core mine action geospatial vector and raster data can be combined with auxiliary data for deeper GIS analysis. To assist these operations, SERWIS will supply links to global auxiliary datasets (e.g.

207

Chapter 4: How to best build GIS capacity in mine action?

GlobCover, HydroSHEDS, OpenStreetMap) and existing online services (e.g. ArcGIS online, Bing Maps) illustrated by Box F in Figure 54. From vector data, raster data and online services, the GIS user (Box D in Figure 54) performs GIS analysis and designs maps. Examples of GIS analysis are 5D, NAMA and MASCOT processes while START may help to accelerate data preparation and map design. In some countries, the GIS user and the IMSMANG user may be a single person. Vector data, raster data and maps (Box E in Figure 54) can be transferred via a secure File Transfer Protocol (FTP: Box G in Figure 54) to the GIS server administrator (Box H in Figure 54) based in Geneva, who copies them on the Web server and exposes them through web services. The ArcGIS Server Geoportal extension (Box I in Figure 54) allows the GIS server administrator to define user groups with specific privileges. In particular, a group of users with advanced GIS skills can be created with the privilege of publishing web services without requiring assistance from the GIS server administrator. In Figure 54, this option is materialized by the arrow between Box E and Box I. End users (Box J in Figure 54) need an Internet connection to access web services. Maps will be queried through WMS and data retrieved through WFS (vector) and WCS (raster). The final client might be a web browser, a desktop application such as ArcMap or another open source or proprietary licensed software. For reasons of confidentiality, the access to each service can be restricted to a specific user group. On the other hand, data providers can make web services accessible to anyone who has an Internet connection. As a further expansion of the raster generator, we suggest integrating a function that would automatically export the information that users enter from the raster generator GUI. These metadata are meant to inform the map reader on how and when the raster was produced. They would include data type (e.g. AP, AV, BAC etc.), status of the data (e.g. active or closed), information on data confidence (estimated or calculated ERW area), information on how and when the data was collected, sample size, contact details, coordinate system, creation date of the map etc. All metadata would be exported to a XML file and would be available from providers of contamination density maps upon request.

208

Chapter 4: How to best build GIS capacity in mine action?

4.6. Improving the Quality of Web Services

Based on: Testing OGC Web Feature and Coverage Services performance: towards efficient delivery of geospatial data

Grégory Giuliania,b,c, Alain Duboisd,e, Pierre Lacroixa,b,c a University of Geneva, Institute for Environmental Sciences, enviroSPACE Lab., Battelle – Building D, 7 route de Drize, CH-1227 Carouge; b University of Geneva, Forel Institute, 10 route de Suisse, CH-1290 Versoix; c United Nations Environment Programme, Division of Early Warning and Assessment, Global Resource Information Database – Geneva, International Environment House, 11 chemin des Anémones, CH-1219 Châtelaine; d University of Geneva, Institute of Environmental Sciences, Human Ecology Group, Battelle – Building D, 7 route de Drize, CH-1227 Carouge; e University of Applied Sciences Western Switzerland, HEPIA, 4 rue de la Prairie, CH-1202 Genève

4.6.1. Abstract

OGC Web Feature Service (WFS) and Web Coverage Service (WCS) specifications allow interoperable access to distributed geospatial data made available through Spatial Data Infrastructures (SDIs). To ensure that a service is sufficiently responsive to fulfil users‘ expectations and requirements, performance of services must be measured and monitored to track latencies, bottlenecks and errors that may negatively influence its overall quality. Despite the importance of data retrieval and access, little research has been published on this topic and mostly concentrates on the usability of services when integrating distributed data sources. Considering these issues, this paper (1) extends and validates the FOSS4G approach to measure the server-side performance of different WFS and WCS services provided by various software implementations and (2) provides some guidance to data providers looking to improve the quality of their services. Our results show that performance of tested implementations is generally satisfactory and memory tuning/data and storage optimisation are essential factors to handle increased efficiency and reliability of services.

Keywords: Quality of Service, performance, WFS, WCS, benchmarking

4.6.2. Introduction

Facilitating access and making geospatial data interoperable are recognised as important factors in determining the future success of Spatial Data Infrastructures (SDI) (Rajabifard and Williamson 2001,

209

Chapter 4: How to best build GIS capacity in mine action?

Bernard and Craglia 2005, Masser 2005, Boes and Pavlova 2008). Indeed, geospatial data are extensively used in various domains such as environmental monitoring, disaster management and decision-making processes, requiring seamless integration in GIS and sharing of data across organisations and providers (Sahin and Gumusay 2008). Geospatial data can be a shared resource maintained continuously and made accessible for different purposes in both the public and private sectors (Ryttersgaard 2001). SDIs aim to support sharing and exchange of geospatial data across institutional, regional, and/or national borders, mostly through Service Oriented Architecture (SOA) principles (Simonis and Sliwinski 2005, Granell et al. 2009). SOA is defined as an architectural approach where standardised interfaces give access to functionalities as a set of independent, interoperable, loosely-coupled distributed services that can be reused. The rapid evolution of web-based communication technology allows for easier access to distributed geospatial data sources and related services, and results in an increasing use of geospatial data in many areas (Paul and Gosh 2006). To improve interoperability within the GIS community, the Open Geospatial Consortium (OGC) introduced various standards specifications covering data sharing, retrieval, processing, content, visualisation and cataloguing (Brauner et al. 2009, Bulatovic et al. 2010). These standards are: Web Map Service (WMS) (OGC 2006a), Web Feature Service (WFS) (OGC 2005), Web Coverage Service (WCS) (OGC 2006b), Catalogue Service for the Web (CSW) (OGC 2007c), Web Processing Service (WPS) (OGC 2007b) and Web Coverage Processing Service (WCPS). They are built around web services technologies. A web service is a stateless application containing a collection of operations, exposed as a function, that is accessible through the web and that users can discover and invoke (Sahin and Gumusay 2008). In other words, interoperable web services aim to provide users just the functionality they need independently of the computing platform (e.g. Operating System, Data Management System) and programming language. Moreover, the composability and reusability of standard components offered by web services allow for building applications specific to the needs of domain and/or community of users, overcoming disadvantages and inflexibilities of monolithic GIS (Simonis and Sliwinski 2005). OGC web services (OWS) are based on eXtended Markup Language (XML) to encode calls and HyperText Transfer Protocol (HTTP) for communicating. Several studies (Booz and Allen 2005, European Commission 2006, Craglia and Campagna 2009) have shown that projects that have adopted and implemented geospatial interoperability standards saved around 25% of their time, compared to those who rely on proprietary standards. These reports also showed that using geospatial interoperability standards lowers the transaction costs for sharing data and information. The fact that exchange of data and information is performed on standardised interfaces enhances flexibility and adaptability of projects over time. These benefits provided by geospatial interoperability standards probably contribute to their growing success. The INSPIRE State of Play (Vandenbroucke 2005) highlights that more and more European countries are focusing their attention on interoperability issues.

210

Chapter 4: How to best build GIS capacity in mine action?

View services appear to be very well developed and download services are recently beginning to emerge. Consequently, we can assume that WMS (OGC 2006a, p 85) is now well accepted among different communities to make their data viewable in an interoperable manner, but this is not yet the case for download using WFS (OGC 2005, p 131) and WCS (OGC 2006b, p 143) standards. Only in a few cases have data been made available through these standards (Da Silva et al. 2004, Christensen et al. 2006, Khoumeri and Benslimane 2007, Bermudez 2009, Bruniecki et al. 2010). Currently SDIs are mainly concerned with data retrieval, data processing and data visualisation (Baranski 2008, Schaeffer 2008), allowing data discoverability using Catalogue Service for the Web (CSW) (OGC 2007c, p 218), retrieval with WFS and WCS, processing and analysis through WPS (OGC 2007b, p 87) and WCPS, and visualisation through WMS. Initiatives at the regional and global scale such as the Infrastructure for Spatial Information in the European Community (INSPIRE) (European Commission 2007a) and the Global Earth Observation System of Systems (GEOSS) (Geo secretariat 2005) are good examples of what can be done in terms of geospatial interoperability standards. These initiatives are seeking to relate environmental data providers to the widest possible audience, with the objective to enhance and improve decision-making. In these two examples, as in many others, OWS are key enablers providing interoperable access to data in an efficient and timely manner. However, having interoperable access to data is only a first step towards data integration. To satisfy users‘ expectations, it is obviously also required to have services of sufficient quality, especially in terms of performance (O‘Dea et al. 2011). To ensure that a service is sufficiently responsive to fulfil users‘ needs and requirements, performance of a given service must be measured and monitored. This includes tracking latencies, bottlenecks and errors that may negatively influence its overall quality. Despite the importance of data retrieval and access, little research has been conducted on benchmarking and evaluating the quality of WFS and WCS services (Tu et al. 2004, Simonis and Sliwinski 2005, Zhang 2005, Lance et al. 2006, Vanmeulebrouk et al. 2009). These studies mostly concentrate on the usability of OGC Web Services (OWS) when integrating distributed data sources, but provide neither a framework nor guidance to measure the performance of data services. Moreover, the only published papers specifically on this topic examine the WMS (Horak et al. 2009) and the WFS (Bauer 2012) specifications, to our knowledge. Consequently, a framework to assess the usability and performance of download services through a set of quantitative measures is required that allows for the quantification, repeatability, comparability and understandability of results. Based on these considerations the aims of this paper are:  To extend the FOSS4G approach for measuring the server-side performance of different WFS and WCS services provided by two widely-used software implementations,  To provide some guidance to data providers aiming at improving the quality of their services,  To encourage other users to contribute to completing this benchmark by setting up other scenarios and performing other tests (e.g. client-side, different parameters). All developed material (benchmark scripts, data, and procedures) is freely available by contacting the authors.

211

Chapter 4: How to best build GIS capacity in mine action?

4.6.3. Geospatial data interoperability

Data accessibility, availability and compatibility are among the most frequent difficulties encountered while preparing Strategic Environmental Assessments (SEA) and Environment Impact Assessments (EIA) in Europe (Bernard and Craglia 2005). Moreover, it is estimated that up to 50% of users‘ time is spent in data discovery and transformation to make them compatible and harmonised (Ma et al. 2005, Wei et al. 2009). These authors emphasise various reasons leading to such problems:  Geospatial data are voluminous.  Geospatial data are geographically distributed (e.g. various data collector/providers around the World).  Geospatial data are heterogeneous (e.g. formats, schemas, coordinate systems).  Geospatial data are complex (e.g. geometries, relationships, attributes).  Institutional arrangements and policies (e.g. copyright) are lacking or restricting access to (meta)data. All these factors may influence the way data providers store, publish and deliver geospatial data. Hence, making data interoperable and improving the quality of this interoperability can potentially enhance the above-mentioned situation, allowing data users to spend more time in data analysis than in data discovery, which would enable more people to benefit from using geospatial data. To be fully interoperable a system must be syntactically, semantically and schematically interoperable. However, OGC specifications concentrate mostly on syntactical interoperability allowing exchanging data with other components or systems. To reach semantic and schematic interoperability, both components of a system must agree on a common reference model providing the possibility to interpret accurately, unambiguously and meaningfully the information exchanged. These levels of interoperability are important issues that have been extensively studied (Fu et al. 2005, Lacasta et al. 2007, Lutz et al. 2009, Zhao et al. 2009). Hence, to enable effective and syntactically interoperable access to geospatial data, OGC WFS (for vector data) and WCS (for raster data) specifications are essential components and a prerequisite for testing performance of different software implementations. The OGC WFS specification (OGC 2005) defines an interface for accessing feature-based geospatial data, commonly known as vector (e.g. rivers, country borders and cities), encoded in Geography Markup Language (GML) (Sahin and Gumusay 2008, Wei et al. 2009). A WFS interface is invoked through a URL and can perform a certain number of operations (e.g. retrieving, creating/deleting/updating) allowing a client to handle data (Sayar et al. 2005). The OGC WCS specification (OGC 2006b) defines a web interface allowing a client to access raster data sets. A raster data set represents a matrix of cells in continuous space organised in rows and columns where each cells contains a value. Thus WCS services give access to different types of space and/or temporal-varying data such as a Digital Elevation Model (DEM) or remote sensing imagery. WCS

212

Chapter 4: How to best build GIS capacity in mine action?

delivers raw data and does not have transactional capabilities (Sahin and Gumusay 2008, Wei et al. 2009).

4.6.4. Quality of Service (QoS)

Satisfaction of users is a major objective of any service provider. However, even if a system is fully interoperable (e.g., syntax, semantics, schema), this satisfaction would not necessarily be guaranteed. Evaluating and predicting users‘ satisfaction is a complex task because quality can be based on quantitative and qualitative measurements (Lance et al. 2006). The overall performance of the service is dependent on the combination of server (e.g., data service), network and client (e.g., computer sending a request) components. In particular, the overall performance will be affected by the slowest component that needs to be identified first. Quality can be interpreted both as a measure that represents accessibility/performance and also as a perceived quality of a service (Simonis and Sliwinski 2005). Despite the importance of qualitative perceptions (e.g. good description, easy access to a service) it is undeniable that unresponsiveness or slowness of a service will negatively affect users‘ satisfaction (Menasce 2002). With the expected increasing diffusion of OWS, Quality of Service (QoS) will be an important factor to distinguish between reliable services and others. Therefore, QoS can be referred as a set of measurable attributes related to the individual behaviour of a web service to that guarantees sufficient availability and performance of a service (Simonis and Sliwinski 2005). This excludes all qualitative perceptions and concentrates on technical aspects that are easily measurable and replicable under different conditions/environments. A detailed overview of testing methods both on the server and client sides is given in (Horak et al. 2009). These authors recognise the importance of testing to evaluate characteristics that identify measurable/quantifiable parameters. The Infrastructure for Spatial Information in the European Community (INSPIRE) is a legal framework (that entered into force in May 2007 and will be fully operational by 2019) for the establishment of a European Union SDI. The Directive provides five sets of Implementing Rules (IR) that set out how the various elements of the system (metadata, data sharing, data specification, network services, monitoring and reporting) will operate, and ensures that spatial data infrastructures of the Member States are compatible and usable in a Community and transboundary context (Bernard et al. 2005). To ensure sufficient availability and accessibility to data, the European Commission (European Commission 2009, European Commission 2010) has adopted regulations on Network services and specifically about download services. Annex I of the regulation on download Services sets out Quality of Service criteria and corresponding values that any services must achieve:  Performance Refers to how fast a request can be completed. The response time for sending the initial response shall be a maximum of 30 seconds in a normal situation (e.g. periods out of peak load) and shall maintain a sustained response greater than 0.5 Megabytes (or 500 spatial objects) per second.

213

Chapter 4: How to best build GIS capacity in mine action?

 Capacity Is the limit of simultaneous requests that a service can handle with guaranteed performance (Horak et al. 2009). For a download service, the minimum number of served simultaneous services shall be 10 requests per second. The number of requests processed in parallel may be limited to 50.  Availability Measures the probability that a network service is available and shall be 99% at any given time. These quantitative measurements are important as they allow easy quantification, repeatability, comparability and understandability (Mendes and Mosley 2006, Horak et al. 2009). QoS criteria can be measured either on the server-side (e.g. number of visits, session duration and average time per page) or the client-side (e.g. load testing, capacity testing). In general, the tested service is considered as a black box receiving requests and providing responses without considering software implementation at all.

4.6.5. Methodology of testing

Currently, even if regional initiatives such as INSPIRE are highlighting the importance of QoS there is no commonly agreed framework to measure performance of download services. Consequently, it is important to measure performance with approaches that (European Commission 2007a):  Are sufficiently generic to be independent of infrastructure and application design.  Are independent on the communication (e.g. transport network) between the service and client.  Are based on request-response pairs and avoid complex transactions. Our proposed approach is based on the one developed by the Free and Open Source for Geospatial (FOSS4G) Community41 to test WMS services. The aim is to extend this open approach to WFS and WCS services by testing various OWS implementations on a common set of geospatial data located on the same platform. Obviously, to fulfil the requirements mentioned previously, it is important that these tests are based on realistic usage scenarios. For that purpose, load testing (also known as stress testing) is generally performed simulating multiple concurrent queries to a service. This approach allows analyzing the behaviour (e.g. response to unusual loads) of a service under various conditions that are beyond normal usage patterns (Horak et al. 2005). In our case, WMS testing was achieved through the use of the GetMap query (i.e., it returns to a client a raster map of selected layers. Parameters include: map extent, coordinate system, width and height of the map and image format), with the unique purpose of calibrating our resulting curves with those of FOSS4G42 2009 shootout. Regarding WFS and WCS testing, the use of the GetFeature (i.e., returns a GML result set with geometry and feature attributes) and GetCoverage (i.e., returns a response to a client

41 http://wiki.osgeo.org/wiki/FOSS4G_Benchmark 42 http://wiki.osgeo.org/wiki/Benchmarking_2009

214

Chapter 4: How to best build GIS capacity in mine action?

that either contains or references the requested coverage. If the request is sent from a web browser, users can download the returned coverage) queries placed the emphasis on actual access to the data. Our general methodology was to take one case as a basic case, and to vary a given set of parameters related to data characteristics (geometry type, data resolution, complexity and number of attributes, field indexation, input and output formats) as well as computer memory, one by one. Due to their small impact on WFS robustness (Bauer 2012), complexity and completeness of metadata were not tested. Random screen-sized requests were performed, on commonly used scales and extents corresponding to the bounding box of various countries. Concerning the WCS, the proposed methodology focuses on 2- dimensional data delivery. However, the WCS specification defines an interface to serve N-dimensional coverages that are common and important when going beyond maps into 3D time series or voxel data (e.g., 4D/5D climate data). To represent realistic usage scenarios, two groups of users were simulated: (1) a small team (e.g. people working within the same laboratory) with a workload varying from 1 to 16 concurrent requests, and (2) a larger set of 150 users simultaneously retrieving data such as in an emergency situation (e.g. after Haiti or Japan earthquakes). Each test was performed three times but only the last measure, considered as the most stable one, was retained.

4.6.5.1. Technical architecture and software

One of the aims of the proposed approach is to provide some guidance to service providers who are looking to share their data in an interoperable manner. In particular, vast amount of geospatial data are available within institutions (e.g., universities, research centres) that do not have means and/or do not want to develop complex infrastructures. Consequently, the proposed architecture to serve OWS must be simple, reflecting these conditions and showing that with little effort it is feasible to provide download services of sufficient quality. Therefore, the measurements will reflect performance of the server-side component of the architecture. The technical architecture (Figure 55) is based on a three-tiered model: (1) a data layer (2) a service layer and (3) a client layer.

215

Chapter 4: How to best build GIS capacity in mine action?

Figure 55: Architecture used for testing different OWS implementations

In the data layer, vector and raster data can be stored directly either into a database or in file directories. PostGIS 1.5.143 and ArcSDE 9.3.144 were used enabling PostgreSQL 8.4.445 Database Management System (DBMS) to support geospatial data. Interoperable access to data is provided through the service layer using GeoServer 2.1.146 and ArcGIS Server 9.3.147 that are able to manage different data sources such as PostGIS, ArcSDE or files in folders. Even though there are a lot of available solutions to publish WFS and WCS services (e.g. MapServer, Deegree) we decided to focus on ArcGIS Server and GeoServer because they are widely used in the GIS community and allow to test ―native‖ solutions (e.g. ArcGIS Server/ArcSDE, GeoServer/PostGIS) as well as ―crossed‖ solutions (e.g. ArcGIS Server/PostGIS, GeoServer/ArcSDE). The client requesting services were simulated using JMeter48, a Java open source software designed to test servers, networks, and services performance under various load conditions on static and dynamic resources (e.g. files, servlets, scripts, databases and queries, FTP servers). JMeter was originally designed to test web applications but has since expanded to other testing functions using plugins. Despite the fact that this is a desktop application, it is important to note that JMeter is not a browser (e.g. it does not execute Javascript nor does HTML rendering). Its usage is very simple, requiring only the URL querying a specific service. Users can design their testing plan using variables, counters, logs or whatever parameter to be tested. A JMeter test then corresponds to a script that can be easily executed and maintained through the JMeter GUI. Once tests have been executed, performance logs can be accessed

43 http://postgis.refractions.net/ 44 http://www.esri.com/software/arcgis/arcsde/ 45 http://www.postgresql.org/ 46 http://www.geoserver.org 47 http://www.esri.com/software/arcgis/arcgisserver/ 48 http://jakarta.apache.org/jmeter/

216

Chapter 4: How to best build GIS capacity in mine action?

through text files providing various types of information such as number of requests, number of threads, or execution time. This information can be used to draw results as graphs or tables. JMeter scripts provided by the FOSS4G WMS 2009 benchmark have been used and extended/adapted to test our WFS and WCS instances. Components and characteristics of the test environment are summarised in Table 27.

Table 27: Characteristics of the test environment. All network interfaces between computers were based on 1GB LAN connections Operating RAM Computer Processor Hard disk Software system memory 2 x Quad Core Suse Linux Database Xeon E5420 Enterprise 4 x 1TB 7200 PostgresSQL 8.3.0, Management 2.5Ghz, 2x6MB 8 GB Server 10, 32 rpm, SATA ArcSDE 9.3.1 System Cache, 1333MHz bits FSB ArcGIS Server 9.3.1 Windows 1 x Quad Core 2 x 1TB, .NET enterprise SP2, server 2003, Xeon X3323, Geoservices 7200rpm, Geoserver 2.1.1 for enterprise 2.5GHz, 2x6MB 4 GB server Near Line windows, MS4W 3.0 edition, 32 bits, Cache, 1333MHz SAS, including Mapserver SP2 FSB 5.6.3 Client Dual Core E6600 1 x 148GB, requesting Ubuntu 10.04 2.4 GHz, 4MB 2 GB JMeter 2.3.4 7200rpm geoservices Cache

4.6.5.2. Testing scenarios

Testing scenarios were developed to simulate user‘s behavior when downloading data through WFS and WCS services. They are based on:  Common vector and raster input and output formats: shapefile, Esri geodatabase, ArcSDE, GML, TIFF, GeoTIFF, and JPEG.  Popular datasets, such as the Blue Marble New Generation (BMNG) (Stöckli et al. 2005) and the Global Lakes and Wetlands Database – Level 2 (GLWD-2) (Lehner and Döll 2004) were chosen for their potential use in many fields and their availability at global scale in WGS 84 spatial reference.  Widely used platforms implementing OWS: ArcGIS Server and GeoServer. Various conditions were defined according to (1) our daily experience using OWS and (2) past experiments made during the development of the PREVIEW geoportal (Giuliani and Peduzzi 2011). The different testing scenarios developed to test WFS services are summarised in Table 28 and those concerning WCS in Table 29. Each scenario is tested both under ArcGIS Server and GeoServer platforms.

Table 28: Summary of the WFS testing scenarios. Case #1 is the ―base case‖ Case study Tested Input format Input geometry Input attributes

217

Chapter 4: How to best build GIS capacity in mine action?

for WFS parameter testing 1 None ArcSDE Polygons Base attributes 2 Input format Shapefile Same as case #1 Same as case #1 ESRI File 3 Input format Same as case #1 Same as case #1 Geodatabase 4 Input geometry Same as case #1 Polylines Same as case #1 5 Input geometry Same as case #1 Points Same as case #1 Complexity and 6 number of Same as case #1 Same as case #1 No attributes attributes Additional attributes Complexity and - text 255 7 number of Same as case #1 Same as case #1 - five eight-byte attributes double-precision columns Attribute Indexation of the id 8 Same as case #1 Same as case #1 indexation field

Further tests were also performed on memory configuration by varying the allowed memory and by comparing performance of each of the three runs of case #1 with varying memory. An ArcGIS Server endpoint is for WFS: http:///arcgis/services//MapServer/WFSServer? And for WCS: http:///arcgis/services//MapServer/WCSServer? While a GeoServer endpoint corresponds to http:///geoserver/ows? As an example, the following WFS request executed under ArcGIS Server allows one to retrieve all rivers of Switzerland (i.e. ―cntry_name‖ field equal to ―Switzerland‖) present in the dataset ―WFS_GDB:river‖. This query uses the OGC filter encoding specification capabilities (OGC 2005) to extract data based on their attributes: service=WFS&styles=&request=GetFeature&version=1.0.0&srs=EPSG:4326&typename=WFS_GDB:r iver&Filter=cntry_nameSwitzerland In GeoServer, the same request shows differences (e.g., ―ogc‖ term mandatory in ArcGIS) in the syntax for the expression of the filter, which already highlights a potential problem in terms of interoperability and universality of semantic: service=WFS&styles=&request=GetFeature&version=1.0.0&srs=EPSG:4326&typename=owsbenchma rk:AQ.SDE.river&filter=cntry_name Switzerland

Table 29: Summary of the WCS testing scenarios. Case #1 is the ―base case‖ Case study Tested for WCS Input format Output format Output resolution parameter testing

218

Chapter 4: How to best build GIS capacity in mine action?

Tiled images in 0.008 Decimal Degrees 1 None TIFF TIFF (DD) ESRI File 2 Input format Geodatabase Same as case #1 Same as case #1 Raster Catalog Case #1 files 3 Input format served by ESRI Same as case #1 Same as case #1 Image Server 4 Output format Same as case #1 GeoTIFF Same as case #1 5 Output format Same as case #1 JPEG Same as case #1 Half of the base case 6 Amount of data Same as case #1 Same as case #1 resolution Case #1 resolution divided 7 Amount of data Same as case #1 Same as case #1 by 4

Concerning the WCS request, the base case corresponds to the following parameters: SERVICE=WCS&VERSION=1.0.0&REQUEST=GetCoverage&COVERAGE=1&CRS=EPSG:4326&BB OX=0, 0,40,20&WIDTH=9600&HEIGHT=4800&FORMAT=tiff

4.6.5.3. Datasets

To fit the GIS community‘s common users‘ practices, we have selected widely used data sets with a global extent and a large number of features (e.g. attributes, resolution, formats). Therefore vector data sets (used for WFS tests) were extracted from:  The GLWD-2 for Polygon data set. This very popular dataset, used by numerous governmental and non-governmental organisations, contains about 250‘000 natural lakes and man-made reservoirs, and 8‘500 rivers – respectively 4‘000‘000 and 350‘000 vertices. By default, the polygonal layer includes a dozen textual and numeric attributes.  The European Space Agency (ESA-ESRIN) World Fires Atlas Program (ATSR) as a Point data set. This represents an estimation of fires events around the world between 1997 and 2008, and is available in a compiled form on the PREVIEW Global Risk Data Platform (Giuliani and Peduzzi 2011). This data set is composed of around 1‘090‘000 points. While for the WCS testing the raster dataset we selected as a base case was the BMNG. This dataset is provided into 240 GeoTiff tiles (e.g. separate images) of 12 MB and 15 DD each, spatially referenced on the WGS 84 ellipsoid, with a 0.008 DD pixel size, for a total of 2.87GB of data. The BMNG is traditionally used not only in education and research, but also often as a base map, and has a relatively high spatial resolution.

4.6.6. Results

The tests performed with JMeter consist of sending requests for data with random geographical extent to the different services and measuring the response time. The performance indicator is expressed in data

219

Chapter 4: How to best build GIS capacity in mine action?

served per second in the case of WFS and in map images served per second in the case of WCS. Different load conditions ranging from 1 to 40 simultaneous threads allow evaluating the server response. Vector data sets were used for testing WMS and WFS, and raster data sets for WCS.

4.6.6.1. WMS

First, it is important to validate the proposed architecture by comparing results obtained in the FOSS4G Benchmark with the system used in the experiment. For that, the same data set storing edge polylines (roads, rivers) for Texas (shapefile with spatial index) was used. The performance tests show that the results for shapefiles from FOSS4G WMS benchmark 2009 are similar for Mapserver CGI and FCGI on the Linux platform and for Mapserver CGI and FCGI on Windows49 (Figure 56, Table 30). This means that the infrastructure used for the benchmark is comparable in terms of performance.

Table 30. WMS results comparison. FOSS4G benchmark 2009 in italic. CGI: Common Gateway Interface, SHP: shapefile, FCGI: Fast CGI, MS4W: MapServer for Windows. [map image /s] Mapserver MapServer MS4W CGI MS4W FCGI Requests Threads CGI SHP FCGI SHP SHP SHP 1 100 5.0 8.2 3.6 5.7 10 200 11.0 17.1 12.2 17.5 20 400 11.0 16.8 11.0 16.9 40 800 11.5 17.6 11.5 17.3

49Package Mapserver for Windows - MS4W 3.0 http://www.maptools.org/ms4w/ containing Apache 2.2.15, Mapserver 5.6.3

220

Chapter 4: How to best build GIS capacity in mine action?

Figure 56: WMS results under various conditions (OWS implementation and file formats)

4.6.6.2. WFS

WFS 1.1.0 services have been tested through different scenarios on an ArcGIS Server and Geoserver including different data sets (e.g. points and polygons), stored in different formats: ArcSDE, Shapefile, ESRI File geodatabase and PostgreSQL/PostGIS. The fire points data set (in shapefile format) containing about 1‘090‘000 features for a size of 99 MB is about four times slower than the lake data set (shapefile of 117 MB) containing about 250‘000 polygons and 4‘000‘000 vertices. This means that the relationship between the storage size of a data set and the performance is not straightforward. ―The bigger in file size the slower‖ is not always true, as performance also depends on the number of features. On ArcGIS Server, using the data set ―Lake polygon‖, performance was only slightly affected by the storage format, with a small advantage for the ESRI file geodatabase format (Figure 57, Table 31).

Table 31: WFS ArcGIS Server tests results. Refer to Table 30 for acronyms [Data /s] SDE Fires SHP Lake SDE Lake FGDB Lake Requests Threads point polygon polygon polygon

221

Chapter 4: How to best build GIS capacity in mine action?

1 100 0.5 1.7 1.8 1.8 2 200 0.8 3.8 4.4 4.5 4 400 1.0 4.9 4.1 5.1 8 800 1.1 4.7 4.5 5.3 16 1‘600 1.1 5.0 4.6 5.3

Figure 57: WFS ArcGIS Server tests results

Compared to ArcGIS Server, Geoserver WFS runs are about three times faster. The polygon data set representing lakes is more efficient when stored in PostGIS. A small decrease in performance is observed when the number of simultaneous threads exceeds 4, but remains very good (Figure 58 and Figure 59, Table 32). This phenomenon may come from a disk access bottleneck. An increasing number of service errors have been observed when the simultaneous threads increase.

Table 32: WFS Geoserver tests results [Data /s] SDE Fires SHP Lake SDE Lake PostGIS Lake Requests Threads point polygon polygon polygon 1 100 1.4 6.5 5.1 8.2 2 200 2.5 12.1 9.8 14.7 4 400 3.2 13.5 14.0 17.0

222

Chapter 4: How to best build GIS capacity in mine action?

8 800 3.5 13.2 14.8 16.9 16 1‘600 3.3 12.6 13.9 16.0

Figure 58: WFS Geoserver test results

223

Chapter 4: How to best build GIS capacity in mine action?

Figure 59: Comparison of best result obtained on each software implementation: GeoServer and ArcGIS Server

4.6.6.2.1. Comparison on memory

Comparative tests on memory show that the performance of WFS depends on memory configuration. The variation is significant with ArcGIS Server: by allowing 4 instances instead of 2 (which is the default value), performance gains are about 35% with 4 or more simultaneous threads. Because each separate process uses the same amount of RAM, giving more RAM allow more processes to run independently. With GeoServer, memory configuration shows ambiguous results on WFS performance. Adding 25% of memory, results in gains are smaller than the variation between the different runs of measures. However, GeoServer appears more stable in term of throughput. The three runs of each of the previous tests on memory generate comparable throughputs. Variations in performance range from 6% to 9%. First runs are slower than the two others with an increasing number of simultaneous requests.

224

Chapter 4: How to best build GIS capacity in mine action?

4.6.6.3. WCS

The performance of WCS 1.1.1 is tightly linked to the size and resolution of the data set. The Blue Marble 8 bits colour image was stored in low-resolution 3‘600x1‘800 pixels and high-resolution. Medium resolution was produced by directly resampling the high-resolution image through the WCS instance and requesting a smaller number of pixels in height and width. Different storage options have been tested for the high-resolution image consisting of (1) one big geotiff image 45‘688x22‘509 pixels (2.87 GB), and (2) 240 tiles of geotiff images 1876x1876 pixels (10 MB each tile). Setting up a responsive WCS instance involves efficiently accessing the raster data set. On the ArcGIS Server platform, the WCS is faster when the data set is stored in ESRI file geodatabase raster than in flat Tif file (Figure 60, Table 33). This is true for low-resolution, while in medium-resolution the file geodatabase is exactly the same as the flat Tif file. In high-resolution the performance of the two storage formats is about the same. One hypothesis to explain this pattern is that in case of heavy load, the server is limited by other factors such as CPU or memory rather than data access. The overall best result is obtained with ArcGIS Image Server with tif tiled in high resolution. This result is explained by the overview images created automatically when the ArcGIS Image Service is built. These overview images speed up the service for the WCS requests that cover a large area at small scale.

Table 33: WCS Geoserver (GS) and ArcGIS Server (AGS) tests results. An ArcGIS Server Image Service (IS) has also been tested. Tests have been executed on Linux and Windows (win) operating systems. Lr: Low-resolution, Mr: Medium-resolution, Hr: High-resolution

Data (tif format) Tif tiled

AGS AGS- [Images AGS AGS AGS AGS GS GSwin AGS GSwin GS hr IS /s] Requests lr lr mr mr mr mr hr hr lr tif fgdb hr Threads tif fgdb tif fgdb tif geotif tif geotif tif geotif 1 100 1.0 1.5 2.2 0.3 0.3 0.8 0.01 0.08 0.10 0.9 0.11 2 200 0.9 1.3 2.3 0.2 0.2 0.2 0.01 0.07 0.10 1.5 0.02 4 400 0.9 1.3 1.8 0.3 0.3 0.1 -- 0.11 0.10 1.7 0.04 8 800 0.9 1.4 1.5 0.3 0.3 -- -- 0.14 0.10 1.8 0.02 16 1‘600 0.9 1.4 1.7 0.4 0.4 -- -- 0.23 0.20 1.8 --

225

Chapter 4: How to best build GIS capacity in mine action?

Figure 60: WCS ArcGIS Server tests results

Geoserver has been tested on Linux with Tomcat and on Windows with Jetty. Tomcat and Jetty are open- source HTTP servers and Java servlet containers. These two applications provide a pure Java web server environment as well as services such as load balancing, data services and security. It seems that GeoServer performs faster on Linux. When the load on the server increases in terms of simultaneous threads and with higher resolution, the number of service errors increases and the throughput decreases to very small values. It takes more than one minute to produce each image when the WCS server is overloaded (Figure 61).

226

Chapter 4: How to best build GIS capacity in mine action?

Figure 61: WCS Geoserver tests results

4.6.7. Discussion

Geospatial data are valuable in various domains particularly in multi-disciplinary research activities that require integrating different data sets from different sources with different formats. Therefore, having these data readily available and easily accessible is a key requirement. OGC standards and especially WFS and WCS are of high interest because they can increase interoperability of heterogeneous data sets. Even though the two tested solutions (GeoServer and ArcGIS Server) show substantial differences in performance, our aim is not to focus on the debate of proprietary versus open source licensed software. Instead we wish to give a first insight concerning server-side WFS and WCS implementations of two of the most widely used solutions within the GI community. We also wish to contribute to: (1) improving the quality of these services independently on the platform used, (2) discussing the possibilities to test performance of these services, and finally (3) sharing our testing scripts and data used with the scientific community (e.g. FOSS4G and other testers) to improve the proposed framework of testing. The proposed extended approach can be considered as relevant because results obtained from calibration of our architecture against WMS services show similar values as those of the FOSS4G Community. Additionally, the stability of our results is also positive element to consider meanings that what has been measured are differences in performance caused by the various tested parameters. Our tests have highlighted that globally these different implementations can provide fairly good performance directly ―out of the box‖, regarding the vast amount of data (up to 4‘000‘000 rows), the

227

Chapter 4: How to best build GIS capacity in mine action?

retrieval extent (worldwide in some cases) and the high number of simultaneous requests (up to 1‘600). However they clearly show that memory is a critical factor to control and that many elements can influence the response of a given service. Therefore, simplifying the instance (e.g. turning off extraneous services or options), configuring suitable memory parameters (e.g. image rendering, type of data served) and managing the number of requests (e.g. limiting the number of concurrent requests that could prevent timely responses, workload manager) are factors that can improve the overall quality of web services (Esri 2010, OpenGeo team 2010). Various other factors may affect performance, in particular if implementations use containers such as Java Virtual Machine (JVM)50 or Jetty51. Tuning the different options provided by these containers may significantly improve performance of the proposed services, as well as increasing network speed (Bauer 2012). Serving large amounts of data efficiently requires optimising data and storage (Esri 2010, OpenGeo team 2010). Our results show that ESRI file geodatabase appears a suitable format, both for vector and raster data, as it provides good performance compared to flat files or ArcSDE. Knowing that FGDB could potentially become a common and cross-platform format with the recent release of an open Application Programming Interface (API)52opens interesting perspectives for stakeholders. Native solutions like ArcGIS Server/ArcSDE and GeoServer/PostGIS also give globally good results and may be reliable solutions to efficiently serve data using WFS and WCS standards. When storing vector data in databases, indexing geometry and attributes significantly improves performance by accelerating data delivery. This is also particularly important because the relationship between storage size of the data set and overall performance of the download service (e.g. number of features, attributes) is not linear. Regarding raster data, building pyramids, caches and overviews can improve the performance of the proposed service. These operations will decrease the amount of data sent when panning or zooming by tiling and downsampling an image to a standardised size and creating overviews as separate image files in a hierarchy (Esri 2009b). In the same direction, on-the-fly re-projection can be CPU intensive and storing data in the most frequently requested projection may also improve performance. Finally, symbology can also influence the performance of services. In this study we have decided to not test this parameter, because (1) this is not the most sensitive parameter (compared to memory) and (2) lots of factors need to be tested to achieve a reliable symbology (e.g. classes, forms, texture and size). In a more general context, WFS and WCS specifications are suitable standards to share and access data in an interoperable manner. However, even if data and storage are tuned and optimised, some bottlenecks may appear when transferring large volumes of data (e.g. geographical extent, attributes, features, or resolution). Indeed, WFS uses GML encoding (Peng and Zhang 2004, OGC 2007d, Amirian and Alesheikh 2008) to serve data. Due to its verbose nature, transferring large amounts of data might be problematic and lead to high latencies and low performance. Similarly WCS is very sensitive to data

50 http://www.java.com/en/ 51 http://jetty.codehaus.org/jetty/ 52 http://resources.arcgis.com/content/geodatabases/10.0/file-gdb-api

228

Chapter 4: How to best build GIS capacity in mine action?

resolution. Consequently, these standards are more suited to share local medium-resolution data than global high-resolution data because of the limitation cause by file formats, file size and network bandwidth. Additionally, differences in OGC specifications have been noted in the various implementations tested (e.g. filter encoding, parameters). This may lead to interoperability problems, especially if clients do not implement the different flavours of these specifications, and may limit seamless data integration capabilities. Although this was not the purpose of this study, we noted differences between clients (e.g. ArcGIS, QGIS, uDig) when accessing a layer provided by a given service. This can be potentially caused by differences in ways of reading data in the various client software implementations. This should be further investigated as it may influence the way users perceive the quality of a service. Concerning the versions of WFS and WCS, we tested the available implementation of these standards in ArcGIS Server and GeoServer, respectively 1.1.1 and 1.1.0. However these standards are not the current versions specified by the OGC. WFS and WCS are now both available in version 2.0 (OGC 2010a, OGC 2010b). This is a current limitation, as data providers may be dependent on the version implemented in the selected software. Even in the most recent version of ArcGIS Server (10.1) WCS 2.0 is not currently implemented. This limitation can be tackled by testing other software implementations such as RasDaMan (http://www.rasdaman.com) or Deegree (http://www.deegree.org) thanks to the open approach proposed in our work (by providing access to our testing scripts and data sets). This can allow users to test more recent standard versions. In this study, we have focused our attention on the performance criteria, which is the most critical factor to test when assessing Quality of Service. As stated by the European Commission (European Commission 2009, European Commission 2010), the two other factors are capacity and availability. Further research should then consider measuring these two factors. Capacity is the limit of simultaneous requests that a service can handle with guaranteed performance. Measuring capacity is possible with the two tested implementations, as they manage threads with queuing functionality. After a certain number of threads, the throughput is stable and the service seems to wait until a worker is free and then handles the request. Availability (i.e. the probability that a network service is available) is more difficult to measure as it is strongly related to the architecture of a system. Having multiple instances, load balancers and high availability routers will help to ensure reliable availability of a service. Monitoring and measuring the status of these components requires further investigation. Finally, this extend approach can be considered a starting point for many other tests to broaden and cover all aspects of QoS for data services. With such an open approach, interesting investigations can be conducted for example on: (1) on-the-fly reprojections; (2) symbology, (3) CPU, (4) new version of software and/or implementations (e.g., Deegree, MapServer, RasDaMan), (5) different architectural style of services (e.g., Representational State Transfer REST), (6) WCS serving N-dimensional coverages (e.g., 3D time series, voxel data such as 4D/5D climate data), (7) clients (e.g. ArcGIS Desktop, QGIS, uDig) and network components.

229

Chapter 4: How to best build GIS capacity in mine action?

4.6.8. Conclusion

Spatial Data Infrastructures seek to facilitate the access and integration of geospatial data coming from various sources. To achieve this objective, systems must be interoperable. OGC specifications are key enablers providing interoperable access to data in an efficient and timely manner. Syntactical, semantic and schematic operability are not enough to ensure that a given service is sufficiently responsive to fulfil users' expectations and requirements. Performance of a given service must also be measured and monitored to track latencies, bottlenecks and errors that may negatively influence its overall quality. The objectives of this study were to (1) extend the FOSS4G approach to measure the performance of different WFS and WCS implementations, (2) provide some guidance to data providers looking to improve the quality of their services, and (3) share our testing scripts and test data to the community (e.g. FOSS4G and others) to improve the proposed framework of testing. Our work has shown that overall server-side performance of the tested implementations is globally satisfactory already ―out-of-the-box‖ (e.g., without tuning different server-side parameters). This can be interpreted as a positive sign for data providers that may potentially be reluctant (due to too much complexity) to publish their data using OWS with these software. However, our tests have shown that to achieve reliable services, tuning memory is an essential and critical factor even if memory gets more and more optimised with default software installation. Additionally, optimising data (e.g., attributes indexation and reduction, projection) and storage (e.g., File GDB for flat file, PostGIS for database, Image Server for Raster) are factors that can easily increase efficiency of services. Some differences have been highlighted regarding the various implementations of WFS and WCS specifications. This can potentially limit data integration if clients do not implement the different flavours of these specifications. Finally, by nature these specifications are not well suited to transfer large volumes of data and the current specifications are more appropriate to share local medium-resolution data than global high-resolution data. This can be a potential issue in the near future, especially given the ever-increasing volume of high- resolution data available.

230

Chapter 4: How to best build GIS capacity in mine action?

4.7. Highlights of Chapter 4

Our strategy for best building GIS capacity in mine action includes: improving the access of mine action stakeholders to geospatial data (mine action data and auxiliary data) and to maps, providing them with tools for optimising data preparation and GIS workflows, facilitating integration of new complex geospatial factors in their everyday work, increasing their GIS expertise and encouraging the production and sharing of understandable maps related to mine action. From this perspective, we propose the development of SERWIS, a SDI for mine action. This geoportal will make it possible for users to:  Share and access data and maps related to mine action The primary goal of SERWIS is to give access to mine action maps, such as density rasters showing contamination by ERW, populations at risk of landmines and degrees of operational difficulty in land clearance while preserving data confidentiality. The geoportal may as well expose other information related to demining activities, including data stored in IMSMANG and socio-economic data. For visualisation/retrieval by users that are not equipped with the Esri suite (typically the international donors), and for a better access to data and maps, SERWIS can distribute them through web services. We see SERWIS as a promising framework for users to access and expose data and maps in an open spirit. However, the fact that the participation of users to this initiative will be done on a voluntary basis may be a major limitation to its success, in the sense that only a few data providers might want to take the first step.  Access ready-to-use GIS data from many different sources To reduce the time spent on finding or preparing ―auxiliary‖ data, the SERWIS platform will supply links to global datasets and existing online web services representing many different topics, such as land use, terrain characteristics, digital elevation models, hydrology, population density, transportation networks, or other relevant information.  Optimise GIS workflows and integrate new complex geospatial factors From SERWIS, users will be able to download START, 5D, NAMA and MASCOT. START has been presented in Section 4.2 and was specifically developed to help non-experts GIS users optimise data and map preparation for their workflows. 5D, NAMA and MASCOT will facilitate the integration and automation of new complex geospatial data models in their everyday work. The tools were designed to be easy-to-use and come along with detailed and custom help functions.  Gain in GIS expertise

231

Chapter 4: How to best build GIS capacity in mine action?

As a complement to these tools, an online course has been developed and published on the Esri website53. Users can acquire in a few days the basics of ArcGIS (e.g. how to choose unambiguous symbols, appropriate coordinate system and meaningful map layout). SERWIS will provide a link to this course.  Follow standards For the design of comprehensive cartographic products, we invite users to follow the suggestions provided in Section 4.4. The collection of cartographic symbols that is described there was established based on universal rules of semiology and in relation to the ISO standards. This collection of symbols represents a step towards standardisation of mine action symbols in IMSMANG and in ArcGIS. With it, users will talk a common and familiar visual language for perceiving and describing geospatial information. Finally, for several reasons exposed in Section 4.5 (among which interoperability across platforms and quick access to data and metadata), we incite users to publish data and maps through OGC-compliant web services (WMS, WFS and WCS) and to associate their data and map services with ISO metadata.  Be connected to stakeholders inside and outside the core mine action domain Some of the START tools operate as a bridge between non-spatial repositories (e.g. IMSMANG, MySQL and Excel) and GIS. Even though these tools were initially developed for connecting mine action operators recording data in the field to GIS experts, their use is likely to improve data interoperability between GIS professionals working in different disciplines. In the same vein, the Esri online training was originally designed for professionals working in the humanitarian demining field, but it is open to anyone who has an Internet access. It may (hopefully) sensitise users outside the core mine action domain to the problem of contamination by landmines. Likewise, by providing links to auxiliary data and to collaborative platforms and tools, SERWIS should strengthen cross-sector collaboration between humanitarian demining actors as recommended by GICHD (2009a), and facilitate connection with experts from other user communities. More widely, we see in SERWIS a good opportunity to encourage development of partnerships between mine action users and stakeholders operating in various domains, organisations and countries, at scales ranging from the global to the local level. To increase our chances of ―best‖ building GIS capacity, the material made accessible from SERWIS was developed in close collaboration with end users. Along with several publications, presentations, workshops and/or surveys were organised to get users‘ suggestions or feedback. The technical specifications of the IMSMANG Raster Generator were discussed during numerous meetings with GICHD experts in information management, and during two focus group meetings regrouping about twenty users from different domains of expertise in mine action and in GIS54. 5D, NAMA, MASCOT and START

53 The course can be completed at: http://training.esri.com/gateway/index.cfm?fa=catalog.webCourseDetail&CourseID=2065 54 These focus groups have been described in detail in Section 2.2

232

Chapter 4: How to best build GIS capacity in mine action?

were developed and tested using data provided by seven national mine action programmes. The model underlying 5D was presented in the Journal of ERW and Mine Action while MASCOT and START were submitted and published, respectively, in international peer-reviewed journals specialised in GIS. Between 2011 and 2012, MASCOT was presented to a twenty organisations involved in humanitarian demining and/or in GIS (e.g. national mine action authorities, information managers, academic institutions etc.). A communication was also made during the plenary session of the GIS for the United Nations conference held in Geneva in April 201255. The improvements of the user interactions with the START tools and the development of the help functions were made possible through a two-day end-user workshop involving a dozen users from information management, national programme management, operations and database management. For their part, the suggestions for mine action cartographic standards were validated through a survey towards the mine action community and a poster was presented at the 14th International Meeting of National Mine Action Directors an UN Advisors held in Geneva in March 2011. To satisfy the users‘ expectations, serving data and maps in conformance with standards and in an interoperable way is not necessarily enough. Because of (1) the vast amount of mine action data potentially be served by SERWIS, (2) the complexity and variety of GIS features, symbols and attributes, and (3) the high number of potential users retrieving data simultaneously from SERWIS, it is also required to have responsive web services. In order to track latencies, bottlenecks and errors that might negatively impact the overall quality of data visualisation and retrieval through web services, we benchmarked performance of web services with two different software implementations (ArcGIS Server and GeoServer) and with varying data characteristics (geometry type, data resolution, complexity and number of attributes, field indexation, input and output formats) and computer memory configuration. Two representative groups of potential users were simulated: (1) one organisation with 1 to 16 concurrent requests, and (2) a larger set of 1‘600 users retrieving data simultaneously, simulating a rush to the web service that might occur after a catastrophic event. Global raster datasets with low to high resolution as well as vector data sets with up to 4‘000‘000 rows were served. From this benchmark, the following results and best practices can be listed:  Performance of WFS instances is acceptable with ArcGIS Server. With GeoServer they are about three times faster.  With ArcGIS Server, performance of WMS is faster for File GDB than for SHP (five times) and ArcSDE (1.3 times).  With ArcGIS Server, performance of WFS is slightly faster for File GDB than for SHP and ArcSDE.  WFS performance increases when the amount of data decreases, but the relationship is not straightforward and depends on the number of features.

55 http://www.gisfortheun.com/

233

Chapter 4: How to best build GIS capacity in mine action?

 WFS performance is faster with field indexation than without.  It is recommended to design maps with all layers in the same coordinate system (e.g. WGS 1984 or UTM).  It is recommended to reduce the number of attributes before exposing vector data.  Complexity and completeness of metadata do not have significant impact on WFS performance.  With ArcGIS Server, the performance of WFS can be improved by allowing more instances.  The performance of WCS is tightly bounded to the size and resolution of the data set. WCS is more suited to serve local medium-resolution data than global high-resolution data.  With ArcGIS Server, images served as TIFF or File GDB show similar performance. Image Server is about ten times faster in high resolution.  The number of service errors grows with an increasing number of simultaneous threads.

234

Chapter 5. Conclusion

235

Chapter 5: Conclusion

5.1. Conclusion: To what extent does GIS contribute to efficient mine action?

This research aimed to answer the question how GIS can contribute to efficient mine action. It has shown the potentially significant input of GIS to the humanitarian field. It is however far from being exhaustive and other topics than the ones addressed here could have been explored, for example 3D analysis, imagery, mobile solutions, and geodetics. In this last section, we summarise conclusions and recommendations from each chapter. We attempt to answer to what extent this thesis addresses each of the three research questions and we draw perspectives for future directions of research.

5.1.1. To what extent can GIS improve visualisation of contamination and its impact on population?

The first objective of this research was to explore to what extent GIS can meet the needs for making the problem of contamination by ERW visible. To that end, we explored a set of seven visualisation methods and systematically evaluated their fitness to four categories of humanitarian demining stakeholders (donors and the general public, directors of national mine action authorities, operations officers and database administrators) and at four geographical scales, ranging from the municipal to the global level. To conduct our research, we integrated numerous constraints and requirements. First, few papers have been published on the use of GIS in the humanitarian field. Second, the amount of data varies significantly across different countries. Only little amount of polygons are stored in national IMSMANG repositories. Third, ERW data are highly heterogeneous in their type, quality, positional accuracy, reliability and spatial distribution pattern. Fourth, contamination maps should be easy to produce for data providers and understandable by map consumers. In parallel, non-disclosure has to be guaranteed. Therefore, a compromise had to be found between data obfuscation and the providing of close-to-reality representations. Finally, mapping the contamination either side of country borders needs to be processed carefully and should consider not making frontiers discernible. We demonstrated that GIS can meet users’ needs and requirements for making the problem of ERW contamination visible. This cannot be achieved through the use of a unique cartographic visualisation method but requires several methods56. Table 9 provides recommendations on which visualisation method should be used by which category of users and at which scale. In this table we proposed to each category of users at least one solution for visualising ERW at the scale at which they operate. Two of the proposed solutions are based on established visualisation methods (one-to-one dot maps and choropleth maps) while three of them (ADKNN-Points, ADKNN-Polygons and ADKNN-

56 Working hypothesis H1.1 is thus refuted

236

Chapter 5: Conclusion

Clusters) are extensions of traditional KDE-based mapping. To best fit users‘ needs, key parameters were customized to enhance cartographic capability of the traditional KDE. Each of the three KDE-based methods has strengths and weaknesses that are revisited below. ADKNN-Points and ADKNN-Polygons make it possible for users to adjust the level of detail of the kernel map through user-defined parameters. This way, end users control and adjust the representation of contamination. ADKNN-Polygons and ADKNN-Clusters provide close-to-reality representations of ERW data. ADKNN-Polygons is the most promising for publication of a worldwide map displaying contamination by ERW. However, this method handles polygons, which so far limits its deployment among the mine action community. ADKNN-Clusters is unsupervised and adaptable to scale, but it has some major limitations. First, it highly depends on the degree of clustering of the data, which varies significantly across different countries57. Therefore we encourage users to ensure that data are clustered before they use ADKNN- Clusters. Relevant techniques for evaluating the degree of data clustering are provided in Section 2.3.2.2. Second, when processing data from non-contiguous extents, ADKNN-Clusters calculates one cluster for each country, thus the added value of this method is limited compared to ADKNN-Points. Third, computing performances of ADKNN-Clusters need to be improved, e.g. by migrating the Python code to VB.Net. Fourth, nothing guarantees that the clustering algorithm and the stability-based validation process on which ADKNN-Clusters builds will provide suitable representations of ERW data with any point pattern. As discussed above, ―no clustering algorithm can be universally used to solve all problems‖ (Xu and Wunsch 2005, p.672). ADKNN-Points holds national and sub-national relevance. We developed a prototype raster generator based on this visualisation method (see Figure 24). The GICHD envisages integrating it with IMSMANG. With this cartographic function, users in 60 mine affected countries will be able to choose which type of data they want to display and to specify parameters for adjusting the level of detail for their maps of contamination density. We showed that a serious risk with so much flexibility provided to users is that some of them might under- or over-represent the contamination in their country58. To keep control over this possible drift, we envision two solutions. The first one is to ensure that the contamination values of all regions are measured on a similar scale. To that end, we mapped the natural logarithm of the kernel density instead of the density itself. The second one is to integrate, as a further expansion of the raster generator, a function that would automatically export the information that users enter from the raster generator GUI. These metadata would inform the map reader on how and when the raster was produced (e.g. hazard type and status) and would be available from providers of contamination density maps upon request. With this metadata generator, we want to prevent the map reader from

57 Hypothesis H1.2 is thus partly confirmed 58 Hypothesis H1.3 is thus confirmed

237

Chapter 5: Conclusion

comparing maps based on datasets with different characteristics, e.g. two different maps showing hazards with different status and date of clearance. ADKNN-Points, ADKNN-Polygons and ADKNN-Clusters are sensitive to outlier ERW data. The presence of outliers may be due to incompleteness or inaccuracy of the IMSMA database. To improve probation of inaccuracy, we recommend the use of geospatial data analysis tools similar to the ones provided by START for extracting coordinates, converting to GIS formats, managing coordinate systems, clipping to the desired extent and quickly mapping. ADKNN-Points and ADKNN-Clusters are sensitive to the ERW confidence, since the calculation of kernels is weighted by an IMSMA attribute storing a calculated or an estimated ERW surface. Further research should then consider evaluating in more detail the reliability of this attribute. Since ADKNN-Points, ADKNN-Polygons and ADKNN-Clusters generate single-layer outputs, it was possible to generate maps showing populations at risk of ERW by multiplying contamination density maps with population density rasters. This was performed at two different resolutions (1-km resolution and the administrative unit). This model is a first approach towards mapping the impacts of landmines on populations that still has some limitations. First, it is influenced by the fact that more hazards are recorded in more populated areas. Second, it is not possible with our model to distinguish the contribution of ERW contamination to that of population density. For example, white pixels represent areas that are either not populated or not contaminated (see Lacroix et al. 2011). Further investigations regarding ways of using two-dimensional colour ramps could be targeted. Despite these limitations, this model opens promising perspectives for assessing and mapping population vulnerability59. As shown in Section 2.3.5, such analysis could be performed by integrating the capacity of populations to adapt to impacts. Further directions of research should then consider combining at-risk population rasters with factors such as age, gender, landmine risk education, demining capacity and other socio-economic variables.

5.1.2. What are the contributions and limits of GIS for improving decision-making in mine action?

The second objective of this research was to investigate how GIS can help decision-making in mine action. To conduct our research, we showed that decision-making in mine action is unique60 and we identified which conditions contribute to make its uniqueness. These conditions are the following:  Decision-making in mine action is a participatory process that can involve different categories of users operating at different geographical levels (from the field to the national level) and in very different topics (e.g. a representative of a local governmental organisation, an expert in mechanical demining, and a GIS officer).

59 Working hypothesis H1.4 is thus confirmed 60 Working hypothesis H2.1 is thus confirmed

238

Chapter 5: Conclusion

 Demining capacity is very heterogeneous across countries.  The distance to hazardous areas is an important factor in humanitarian demining.  The human factor is important since people‘s health and lives are at stake.  There are huge financial costs at stake. The cost of clearance is variable depending on the area and the demining technique, but can be estimated to a minimum of $1‘000‘000 per square kilometre.  The fact that the status of a hazardous area evolves throughout the clearance process also contributes to the uniqueness of decision-making. As described in Section 2.2.3.3, the status of a hazardous area evolves from ‗suspected‘ to ‗confirmed‘ and then to ‗dangerous‘. The reliability and the size of the data also evolve along with the successive status. To conduct our research, we also demonstrated that the mine action community needs simple and flexible GIS-based tools61. In parallel, we showed that environmental and socio-economic conditions can have significant influence on humanitarian demining activity62. From this perspective, we proposed 5D, NAMA and MASCOT, three GIS analytical models based on combination of IMSMANG data with geospatial auxiliary data. We highlighted the great potential of these models for supporting and improving decision-making. 5D classifies degrees of operational difficulty of demining into four ordinal categories (low, medium, high or extreme) with regard to environmental factors and human activity (e.g. temperatures, precipitations, vegetation, rock content, slope, distance to roads, infrastructure etc.). The output of the model is a raster dataset. Macro-statistics can be derived from this output in a first step to determine the percentage of land that may be cleared in a given area, with a given technique (machine, human or animal) and with a specific level of operational difficulty. The percentage of land deemed extreme is also assessed. In a second step, the interpretation of geospatial and statistical information regarding operational difficulty may help decision-makers to better target in-field clearance operations. 5D is not meant to estimate financial cost nor physical risk. Such assessments would require data collection and analysis on a local level, while 5D was designed for macro analysis. With further work, 5D opens the possibility to estimate the financial implications of users’ operational choices. Based on transportation network analysis GIS tools, NAMA intends to minimise travel time (or potentially other attributes such as travel distance and fuel consumption) from a set of origin features to a set of destination features. In particular, a case study was conducted for victim assistance with the purpose of determining best potential location for a new health facility. MASCOT is a SDSS. It was designed to help users set clearance priority, building on multi-criteria analysis. The main specificity of MASCOT compared to other multi-criteria analysis systems lies in that the input features (typically, ERW) are scored with regard to their Euclidian distance to real-world scoring objects (typically, schools, markets or agricultural areas). MASCOT is also able to process vector

61 Working hypothesis H2.2 is thus confirmed 62 Working hypothesis H2.3 is thus confirmed

239

Chapter 5: Conclusion

and raster data in the same workflow and to integrate a large variety of scenarios. The AHP was integrated to assist users through the weighting process and to point out possible inconsistencies while weighting. With regards to the abovementioned specificities of decision-making in mine action, 5D, NAMA and MASCOT were designed to be flexible63. As such, 5D and NAMA are meant to be used at the national and sub-national scales while MASCOT can be used at scales ranging from the global to the local level. The three models have been prototyped and can potentially be experimented in any country/region of the world, provided that users adapt them to national/sub-national specificities. The models may typically be implemented by a group of collaborators rather than by one individual. For example, an expert in manual demining will determine which of the layers that a geomatician proposes are relevant for assessing the operational difficulty in his/her area of work. From this perspective, we see 5D, NAMA and MASCOT as participatory platforms around which different experts might gather their competences to make the final decision more accurate. In conformance with users‘ needs, we developed simple tools64. To let users focus on analysis and decision-making rather than on software handling, effort was put on providing user-friendly and intuitive GUIs and detailed guidance. In particular, the underlying complexity of workflows was hidden to users who only interact with the models through a set of parameters. To reduce software handling and to avoid switching from one technology to another, the models were entirely integrated with ArcGIS Desktop, which is the standard GIS used by sixty national mine action authorities. The main output of each model is a single-layer dataset that can be readily combined and overlaid with other geospatial information, allowing for example to analyze and visualise relation with topics outside the core mine action domain. Effort was also put on providing fast tools. We showed that the humanitarian demining community is not yet ready to integrate these tools as standards in their everyday work65. We also identified a number of steps towards this possible integration. As a first step, we described the models in research papers and presented them to various organisations involved in humanitarian demining. As a second step we propose to conduct proof of concept studies in mine affected countries. We suggest selecting programmes with different GIS capacity (e.g. one programme with low GIS capacity, one with medium GIS capacity and one with high GIS capacity) in countries with different geographical, environmental and socio-economic conditions. As a third step we propose to make 5D, NAMA and MASCOT accessible from a geoportal specifically designed for mine action, where users could also access “best available data”. A fourth step would be to maintain the tools with future versions of ArcGIS. Maintaining 5D and NAMA should demand minimal effort since their design in ArcGIS ModelBuilder did not require writing any line of code. The development of new versions may provide the opportunity to increase the tools current capabilities of the

63 In line with working hypothesis H2.1 64 In line with working hypothesis H2.2 65 Working hypothesis H2.4 is thus refuted

240

Chapter 5: Conclusion

tools. In particular we suggest enriching NAMA and MASCOT with least-cost path functions. Such functions may bring more accuracy to the analysis, for example by extending network analysis to the distance covered outside of the roads and by integrating the real distance between features rather than the traditional straight line distance. Other suggestions for future directions of research concern possible extensions of MASCOT towards automation of the scoring process (e.g. by looping over comparative scenarios) and reduction of the interaction with machine (e.g. by scoring from an XML file). If the 5D model is specific to the determination of an operational difficulty layer, MASCOT and NAMA are more generic. MASCOT was originally designed to improve clearance priority setting, but its flexibility and its interdisciplinary nature make it usable for other topics, not necessarily related to mine action. In theory, MASCOT allows the scoring of any set of vector features in function of their distance to any vector or raster data set. Likewise, NAMA was originally developed to improve victim assistance but it is not limited to this topic. NAMA is able to integrate scenarios based on road disruption in case of natural or man-made events (e.g. flooding, security issues or the presence of ERW or MOTAPM) and has a great potential for further applications, in the humanitarian demining field (e.g. prioritisation of road clearance) and in other fields (e.g. optimisation of food delivery to impoverished communities). From this perspective, MASCOT and NAMA should find interesting applications in topics outside the core mine action domain and should provide the opportunity to strengthen existing collaborations and to set up new ones.

5.1.3. How to best build GIS capacity in mine action?

The third objective of this research was to define a strategy for best building GIS capacity in mine action. Recognizing (1) the lack of key structuring elements to improve their use of GIS66 and (2) the needs for improving data access and sharing67, we developed SERWIS, a geoportal for mine action. The main purposes of SERWIS are to facilitate the access of users to geospatial data, GIS tools, maps and teaching material and to encourage the standardisation of cartographic and GIS processes within the humanitarian demining community. Beyond this, we see in SERWIS a good opportunity to connect mine action users with stakeholders working in the humanitarian demining field as well as in other topics. First, SERWIS will provide access to START – a free ArcGIS toolbar that allows enhancing preparation of geospatial data for further GIS analysis and map design in the field of mine action. The novelty of START is to compile at the same location a series of existing ArcGIS tools that are commonly used in mine action workflows as well as new tools specifically developed to address mine action professionals‘ needs in GIS. Some of these tools operate as bridges between IMSMANG and the GIS, while others contribute to accelerating the transfer of geospatial data between in-field workers and decision-makers. To help users handle the toolbar, we put an important effort on (re)developing the help

66 Working hypothesis H3.1 is thus confirmed 67 Working hypothesis H3.2 is thus confirmed

241

Chapter 5: Conclusion

functions. Planned improvements of START include the development of new functions (e.g. calculation of security perimeters based on convex hulls) and the support of DM, DMS and MGRS formats. Even though START was created to facilitate building GIS capacity in mine action, the toolbar could advantageously be used in other topics. We illustrated possible alternative use of START in hydrology and watershed modelling, in emergency mapping and in ecology. In terms of visualisation, SERWIS will be the final link in the whole process from obfuscating the IMSMANG data by KDE to their publication through web services. From this perspective, SERWIS will allow quick and easy publishing of IMSMANG contamination data and maps. But SERWIS will not be limited to it. Our ambition is to provide users with a dedicated spatial data infrastructure for accessing and sharing maps, data, models, tools and teaching material related to demining activity. First, SERWIS will provide links to auxiliary data with good resolution and quality, in order to reduce time spent on finding and preparing data for GIS analysis and mapping. Second, SERWIS will furnish links for downloading the tools introduced in Chapter 3 (5D, NAMA and MASCOT), with the intention of opening traditional mine action workflows to new geospatial factors and complex data models. Third, SERWIS will give users the opportunity to publish data and maps through OGC-compliant web services (WMS, WFS and WCS). This should facilitate access and sharing of information, notably to users who are not equipped with the Esri suite. Such infrastructure should also be a step towards standardisation of cartographic and GIS information and processes. We strongly suggest that data and maps be associated with ISO-compliant metadata to provide relevant information about what is being displayed (e.g. how and when the data was collected). Fourth, SERWIS will supply a link to ―GIS for Humanitarian Mine Action‖ a free web course that teaches the fundamentals of map creation to support land-release efforts. Completing the course only requires an Internet connection and a few days of work. The primary goal of this course is to build the GIS capacity of organisations involved in mine action through training of individuals. Beyond this, this E- learning solution might as well help users to communicate better and to understand each other‘s work through the use of maps. The teaching material includes mine action-oriented datasets and scenarios that can be reused for other purposes (e.g. demos, exercise books, workshops, and user guides). The course is not restricted to users from the humanitarian demining community but was also designed to sensitise users outside the core mine action domain to the problem of contamination by landmines. As for 5D, NAMA, MASCOT and START, we recommend maintaining the course with future versions of ArcGIS. We also suggest adapting it to non-English speakers (e.g. the seven francophone programmes and the five Spanish speaking programmes) by providing it in other languages. As a contribution to standardisation of visual communication within the mine action community, we revised in a fifth step the IMSMA collection of cartographic symbols dating of 2005. The revised version integrates new topics and methodologies since that time and is bound to progress and be transformed along with the practice and challenge of mine action. Such a common language may help

242

Chapter 5: Conclusion

communication in the whole chain of mine action professionals, thus facilitate demining and enhance safety for demining personnel. A ready-to-use symbology may also help users learn fast and efficiently how to produce a map and consequently gain time. Further directions of research should integrate the definition of standards for the land release process by the IMAS. As of 2013, it is under discussion. Since our work focussed on point features further research should also consider revising symbology for line and polygon features. There are hundreds, if not thousands of potential users of SERWIS. Those are the mine action users presented earlier (the international donors, the general public, the directors of national mine action authorities, the operations officers), their partners (private contractors, governmental and non- governmental organisations, academic institutions) and more broadly the enlarged scientific community. Hundreds of maps can potentially be displayed at the same time, showing a vast amount of data with varying characteristics and symbol complexity. To ensure an efficient access to these maps and data, it is required to have responsive web services. In order to identify potential bottlenecks that might negatively affect the quality of web services, we measured the performance of web map and data services with different software implementation, data characteristics, server memory configuration and workload. The results of this research are a set of best practices and recommendations for data providers to improve the quality of their services:  Software implementation: the ArcGIS Server solution provides acceptable performance but GeoServer is faster. In particular, WFS are three times faster.  Format: with ArcGIS Server, performance of WMS and WFS is faster for File GDB than for shapefile and ArcSDE. Images served as TIFF and File GDB provide similar performance, while Image Server is about ten times faster in high resolution.  Scale: we recommend not to serve high-resolution rasters at large scales (e.g. globally).  Coordinate system: we recommend to project all layers to the same coordinate system (e.g. WGS 1984 or UTM zone) before publishing maps.  Attributes: we recommend to index the attributes and geometry and to reduce the number of attributes before publishing vector data.  Metadata: the volume and complexity of metadata do not have a negative influence on the overall performance of web services.  Server configuration and management: web services are memory-consuming. In particular, the relationship between the performance of WFS and the allocated memory is straightforward. An increasing number of service errors has also been observed when the simultaneous threads rise up. We encourage users to complete this research by setting up other scenarios and performing other tests, e.g. measuring overhead due to symbology. Looking at Figure 2, SERWIS appears as the keystone of our research. The geoportal is a central element that supports all the models and tools presented in this thesis. Most of them have been prototyped and are

243

Chapter 5: Conclusion

technically operational. They are in GICHD‘s hands. A successful use of these models and tools now depends on a series of important conditions68. They are discussed below. One of these conditions is the will of users to share data, maps and knowledge with the rest of the mine action community and more widely, with the enlarged scientific community. Users should be aware that SERWIS will not show to other users what they do not want to show (e.g. the exact location of ERW or country limits). There are two reasons for this. First, the use of KDE-based mapping methods allows obfuscating the information and preserving non-disclosure. Second, SERWIS makes it possible to restrict the access to data/map services to a specific user group on request from data providers. Another condition is the will of mine action information managers to maintain the tools and technologies taking into consideration users’ feedbacks carefully. Another difficulty to software development lays in the geographical distance and dispersion of all main actors. As pointed out by Herbsleb et al. (2001) distance is a key obstacle to software developments and their use. Some of the difficulties rising from the distance can be mitigated through the use of technologies that are increasingly frequently used and popular, such as Skype, and tools like SurveyMonkey that allow getting feedbacks from users. They do not, however, totally replace direct contact and such occasions are rare opportunities to discuss directly and in depth the issues at stake. A fourth condition is to be patient. The fact that most mine-affected countries are poor and have a low level of higher education graduates signifies an important limit to the overall level of competence in the use of technologies (Altbach and Knight 2007). The question of technology transfers and upwards levelling is on the one hand a long process, because the capacity of a country in a field or another does not change overnight, while on the other hand the end users' needs and skills must constantly be kept in mind (Carver 2007). Any technology transfer requires time and support from the providers to be effective (Krugman 1979), while the mine contamination problem is an urgent one. Another major challenge in the production of software was highlighted by Conradi et al. (2002) and relates to the dichotomy and contradictory aspects of creativity and discipline. Interactions between these two mechanics envelope the whole process of software development and both appear as a necessary condition for the successful creation of effective applications. The implementation of projects does not only depend on the working chart of the organisation. It is also strongly influenced by the role of interpersonal relations both inside and outside of the organisation. Superimposed conflicts that may arise in an organisation have the potential of severely breaking the scientific process, by rendering the passage of information more difficult and by putting barriers to the implementation of products. In fact, while conflicts may be a forced pass to common understanding, it also can be an important obstacle to the flow of communication and affect the whole process of a project by severely slowing it down (Alter 1993). Finally, the success of implementation of the models and tools presented in this thesis will highly depend on the conviction of non-technical users that this research is valuable and can concretely contribute to

68 All these conditions corroborate working hypothesis H3.3

244

Chapter 5: Conclusion

improve efficiency in mine action. To make one step toward this, an issue brief tentatively entitled ―GIS for Mine Action‖, based on this PhD thesis and directed towards strategic stakeholders, in-field operators and mine action programme managers, is being written at the moment, in collaboration between GICHD and University of Geneva.

245

Chapter 5: Conclusion

5.2. Contributions of our work

The contributions of our work are summarized below. Within the framework of this PhD, we were afforded numerous opportunities:

5.2.1. To provide mine action users with a new framework to make better decision

 To define the requirements of ERW mapping for various user groups working at different scales.  To provide a comprehensive framework to visualize ERW contamination data.  To define the specificities of decision making in the field of mine action.  To provide a GIS framework in line with these specificities to help users make better decision.  To identify users‘ needs in terms of GIS capacity.  To conduct ten research projects and to highlight the perspectives and limits of these projects for future implementation in sixty mine affected countries.

5.2.2. To develop new tools and new material

 To extend the traditional kernel bandwidth selection technique for KDE-based visualization methods. Our technique provides close-to-reality representation of points and polygons at large scales (i.e., sub-national, national, global) for any spatial data distribution, provided that a universal color ramp is used (such a color ramp was proposed, but it may be further improved).  To propose a novel unsupervised cartographic visualisation method based on clustering and kernel density estimation.  To develop four new, freely ArcGIS Desktop extensions including a spatial decision support system and a toolbar designed for non experts GIS users. The novelty of the former is to allow scoring features based on their Euclidian distance to other features. The novelty of the latter is to bring together a series of basic ArcGIS tools in one toolbar and to provide new geoprocessing, geometry and database management functions to support frequent workflows relating to mine action. This toolbar also operates as a bridge between non-spatial repositories (e.g. IMSMANG and Excel) and GIS.  To integrate the Analytic Hierarchy Process (AHP) in ArcGIS Desktop.  To draft the technical specifications of a new cartographic module for IMSMANG including development of an ArcGIS Engine prototype. The specificity of this module is to provide quick and easy transfer of information from national repositories to web map and data services. These functions were designed to allow directors of national mine action authorities showing up-to-

246

Chapter 5: Conclusion

date information about contamination and ongoing demining activity in their country. They include the whole process from data extraction to web publishing including data obfuscation.  To draft the technical specifications of a SDI for mine action including configuration of a geoportal.  To participate to the drafting of an E-learning course on GIS. As of 2013, about 1‘300 users enrolled in this course. It is difficult to know whether they belong to the mine action community or to other user communities.  To revise and update about sixty IMSMANG cartographic symbols.  To write about 20‘000 lines of code in six different programming languages: ArcGIS API for Flex, Python, R, SQL, VBA, and VB.Net. Most of the methodologies and models presented in this thesis have been prototyped. Seven national mine action databases on four continents were accessed for calibrating our models: Afghanistan, Colombia, Cyprus, Iraq, Lebanon, South-Central Somalia, and Tajikistan. The prototypes are now waiting for evaluation by end-users.

5.2.3. To communicate on our research

 To have three scientific papers accepted for publication in international peer reviewed journals (2 as main author).  To submit another scientific paper to an international peer reviewed journal.  To publish two papers in the Journal of ERW and Mine Action, which is the ―longest continuous source of information on ERW and mine action in the world‖ (CISR 2012). This journal has received about 170‘000 unique visitors in 2010-2011 and is read in more than 150 countries.  To have another paper accepted for publication in the Journal of ERW and Mine Action.  To design and present a poster at the ―GIS for the United Nations‖ conference.  To design and present a poster at the ―14th International Meeting of National Mine Action Program Directors and United Nations Advisors‖.

247

Acronyms

248

5D Determining and Displaying a Degree of operational Difficulty of Demining ADKNN Average Distance to K-th Nearest Neighbour AHP Analytic Hierarchy Process AMAE Albanian Mine Action Executive AP Anti-Personnel mine API Application Programming Interface APOPO Anti-Persoonsmijnen Ontmijnende Product Ontwikkeling AT Anti-Tank mine ATS Area Targeted for Survey ATSR Along-Track Scanning Radiometer AV Anti-Vehicle mine BAC Battle Area Clearance BGSS Between Groups Sum of Squared distances BIL Band Interleaved by Line BMNG Blue Marble Next Generation BW kernel BandWidth CAD Computer-Aided Design CCM Convention on Cluster Munition CCW Convention on certain Conventional Weapons CGIAR Consultative Group on International Agricultural Research CHA Confirmed Hazardous Area CI Consistency Index CIAT Centro Internacional de Agricultura Tropical CIDHG Centre International de Déminage Humanitaire de Genève CIESIN Centre for International Earth Science Information Network CISR Centre for International Stabilisation and Recovery CL Community Liaison CR Consistency Ratio CSR Complete Spatial Randomness CSW Catalog Service for the Web DBF DataBase File DBMS DataBase Management System DD / DMS / DM Decimal Degrees / Degrees Minutes Seconds / Degrees decimal Minutes DEM Digital Elevation Model DHA Defined Hazardous Area DMA Directorate of Mine Action DSS Decision Support System EIA Environment Impact Assessment EM Expectation-Maximisation ERW Explosive Remnant of War ESA European Space Agency Esri Environmental systems research institute FAO Food and Agriculture Organisation FCM Fuzzy C-Means algorithm FOSS4G Free and Open Source Software for Geospatial FTP File Transfer Protocol GADM Global ADMinistrative areas GDB file GeoDataBase GEO Group on Earth Observations GEOSS Global Earth Observation System of Systems GICHD Geneva International Centre for Humanitarian Demining

249

GIF Graphics Interchange Format GIS Geographic Information System GLWD Global Lakes and Wetlands Database GML Geographic Mark-up Language GPS Global Positioning System GPW Gridded Population of the World GUI Graphical User Interface HTML Hyper-Text Mark-up Language HTTP HyperText Transfer Protocol ICBL International Campaign to Ban Landmines ICRC International Committee of the Red Cross IDP Internally Displaced Person IGAC Intituto Geográfico Agustín Codazzi IMAS International Mine Action Standards IMG ERDAS IMaGine iMMAP information Management and Mine Action Programmes IMSMANG Information Management System for Mine Action – Next Generation IND Instituto Nacional de Desminagem INSA INgenería y Servicios Aeroespaciales S.A. INSPIRE INfrastructure for SPatial InfoRmation in the European community IPCC International Panel on Climate Change IR Implementing Rules ISO International Organisation for Standardisation ITF International Trust Fund JPEG Joint Photographic Experts Group JVM Java Virtual Machine KDE Kernel Density Estimation LIS Landmine Impact Survey LMAC Lebanon Mine Action Centre LOG Neperian LOGarithm MACC Mine Action Centre in Cyprus MACCA Mine Action Coordination Centre of Afghanistan MASCOT Multi-criteria Analytical SCOring Tool MAUP Modifiable Areal Unit Problem MDB Microsoft DataBase MERIS MEdium Resolution Imaging Spectrometer MGRS Military Grid Reference System MOTAPM Mines Other Than Anti-Personnel Mines MRE Mine Risk Education NAMA Network Analysis for Mine Action NATO North Atlantic Treaty Organisation NGO Non-Governmental Organisation NTS Non-Technical Survey ODBC Open DataBase Connectivity OGC Open Geospatial Consortium ORNL Oak Ridge National Laboratory OWS OGC Web Services PAICMA Programa de Acción Integral contra las Minas Antipersonal PI Performance Index PNG Portable Network Graphics POI Point Of Interest

250

QC Quality Control QoS Quality of Service RDBMS Relational DataBase Management System RI Random consistency Index SADA Space Assets for Demining Assistance SDE Spatial Database Engine SDI Spatial Data Infrastructure SDSS Spatial Decision Support System SEA Strategic Environmental Assessment SERWIS Server for Explosive Remnants of War Information Systems SHA Suspected Hazardous Area SHP Shapefile SOA Service Oriented Architecture SQL Structured Query Language SRTM Shuttle Radar Topography Mission START Simplified Toolbar to Accelerate Repeated Tasks TIA Task Impact Assessment TIFF Tagged Image File Format TMAC Tajikistan Mine Action Centre TS Technical Survey UNDP United Nations Development Programme UNEP United Nations Environment Programme UNITAR United Nations Institute for Training And Research UNMAS United Nations Mine Action Service UN OCHA United Nations for the Coordination of Humanitarian Affairs UPGMA Unweighted Pair-Group Method with Arithmetic mean UTM Universal Transverse Mercator UXO UneXploded Ordnance VB Visual Basic VRC Variance Ratio Criterion WCS Web Coverage Service WCPS Web Coverage Processing Service WFS Web Feature Service WGSS Within Group Sum of Squared distances WGS 84 World Geodetic System 1984 WHO World Health Organisation WMS Web Map Service WPS Web Processing Service WWF World Wildlife Fund XML eXtensible Mark-up Language

251

References

252

References

Alard, C., 2000. Image subtraction using a space-varying kernel. Astronomy & Astrophysics Supplement Series 144, 363–370.

Alegría, A.C., Sahli, H., Zimányi, E., 2011. Application of density analysis for landmine risk mapping, in: 2011 IEEE International Conference on Spatial Data Mining and Geographical Knowledge Services (ICSDM). IEEE, Fuzhou, China, pp. 223–228.

Altbach, P.G., Knight, J., 2007. The Internationalization of Higher Education: Motivations and Realities. Journal of Studies in International Education 11, 290–305.

Alter, N., 1993. Innovation et organisation: Deux légitimités en concurrence. Revue Française de Sociologie 34, 175.

Aminzadeh, F., Chatterjee, S., 1984. Applications of clustering in exploration seismology. Geoexploration 23, 147 – 159.

Amirian, P., Alesheikh, A.A., 2008. Implementation of a Geospatial Web service Using Web Services Technologies and Native XML Databases. Middle-East Journal of Scientific Research 3, 36–48.

Andersson, N., Mitchell, S., 2006. Epidemiological geomatics in evaluation of mine risk education in Afghanistan: introducing population weighted raster maps. International Journal of Health Geographics 5.

APOPO, NPA, DDG, DCA, MAG, HI, GICHD, 2012. Proposal And Justification For Amending An Existing International Mine Action Standard, Developing A New Standard, Or A New Technical Note For Mine Action [WWW Document]. URL http://www.mineactionstandards.org/fileadmin/user_upload/MAS/documents/review- board/others/Proposal_for_Review_of_MDD_IMAS_09_40_to_44.pdf (accessed 3.18.13).

Armstrong, M.P., Densham, P.J., 1990. Database organization alternatives for spatial decision support systems. International Journal of Geographical Information Systems 3, 3–20.

Assaf, H., Saadeh, M., 2008. Assessing water quality management options in the Upper Litani Basin, Lebanon, using an integrated GIS-based decision support system. Environmental Modelling & Software 23, 1327–1337.

Backeljau, T., De Bruyn, L., De Wolf, H., Jordaens, K., Van Dongen, S., Winnepennincks, B., 1996. Multiple UPGMA and neighbor-joining trees and the performance of some computer packages. Molecular biology and evolution 13, 309–313.

Bailey, T.C., Gatrell, A.C., 1995. Interactive spatial data analysis. Longman Scientific & Technical, Essex, England.

Baker, F.B., Hubert, L.J., 1975. Measuring the power of hierarchical cluster analysis. Journal of the American Statistical Association 70, 31–38.

Banai-Kashani, R., 1990. Probabilistic dimension of the AHP approach to input-output analysis: A note. Mathematical and Computer Modelling 14, 1161–1163.

Baranski, B., 2008. Grid computing enabled Web Processing Service. Presented at the Proceedings of the 6th Geographic Information Days, IFGI Prints, pp. 243–256.

Barlow, D., 2003. Plays nicely with others: Some thoughts on issues raised at the 6th International Meeting of Mine Action Directors, Geneva, March 17–20, 2003. International Journal of Mine Action 7.

253

References

Bauer, J.R., 2012. Assessing the Robustness of Web Feature Services Necessary to Satisfy the Requirements of Coastal Management Applications (Master Thesis). University of Wisconsin.

Bazzi, A.M., Fares, D.A., 2008. GIS-based wind farm site selection in Lebanon, in: IEEE International Conference on Electro/Information Technology. Ames, Iowa, USA, pp. 201–204.

Béguin, M., 2003. La représentation des données géographiques: statistique et cartographie, 2e ed, Cursus. A. Colin, Paris.

Benini, A., 2000. A comparison between the ―classic‖ Level-1 Socio-Economic Impact Survey and the emergency survey in Kosovo. Towards Harmonized Information Systems for Mine Action in South East Europe, ISPRA, Italy.

Benini, A.A., Conley, C.E., Shdeed, R., Spurway, K., Yarmoshuk, M., 2003. Integration of different data bodies for humanitarian decision support: An example from mine action. Disasters 27, 288–304.

Bermudez, L., 2009. Web feature service (WFS) and sensor observation service (SOS) comparison to publish time series data, in: Cook, T., Forrest, D., Bogden, P., Galvarino, C., Creager, G., Graybeal, J. (Eds.), Presented at the International Symposium on Collaborative Technologies and Systems, CTS‘09, Washington DC, pp. 36–43.

Bernard, L., Craglia, M., 2005. SDI - From Spatial Data Infrastructure to Service Driven Infrastructure. Presented at the Research Workshop on Cross-Learning Between Spatial Data Infrastructures and Information Infrastructures, Enshede, The Netherlands.

Bernard, L., Kanellopoulos, I., Annoni, A., Smits, P., 2005. The European geoportal - one step towards the establishment of a European Spatial Data Infrastructure. Computers, environment and urban systems 29, 15–31.

Bertin, J., 1977. La graphique et le traitement graphique de l‘information. Flammarion, Paris.

Beyer, H.L., 2004. Hawth‘s Analysis Tools for ArcGIS [WWW Document]. URL http://www.spatialecology.com/htools (accessed 6.20.12).

Bezdek, J.C., Ehrlich, R., Full, W., 1984. FCM: The fuzzy c-means clustering algorithm. Computers & Geosciences 10, 191–203.

Bilukha, O.O., Brennan, M., Woodruff, B.A., 2003. Death and injury from landmines and unexploded ordnance in Afghanistan. JAMA: The journal of the American Medical Association 290, 650– 653.

Björk, K., 2012. Ridding the World of Landmines: The Governance of Mine Action. Brown Walker Press.

Blaikie, P., Cannon, T., Davis, I., Wisner, B., 1994. At Risk: Natural hazards, People‘s vulnerability, and disasters. London, Routledge.

Boes, U., Pavlova, R., 2008. Is there a Future for Spatial Data Infrastructures? Presented at the Proceedings GI-Days, Muenster, Germany, pp. 305–314.

Boothby, J., Dummer, T.J.B., 2003. Facilitating mobility? The role of GIS. Geography 88, 300–311.

Booz, Allen, Hamilton, 2005. Geospatial Interoperability Return on Investment. NASA, Geospatial Interoperability Office.

254

References

Borrie, J., 2009. Unacceptable Harm: A History of How the Treaty to Ban Cluster Munitions Was Won. United Nations Publications UNIDIR.

Brans, J.P., Vincke, P., Mareschal, B., 1986. How to select and how to rank projects: The Promethee method. European Journal of Operational Research 24, 228 – 238.

Braun, C.C., Glusker, S.A., Holt, R.S., Silver, N.C., 1995. Adding Consequence Information to Product Instructions: Changes in Hazard Perceptions, in: Proceedings of the Human Factors and Ergonomics Society Annual Meeting. SAGE Publications, pp. 346–349.

Brauner, J., Foerster, T., Schaeffer, B., Baranski, B., 2009. Towards a research agenda for geoprocessing services. Presented at the 12th AGILE International Conference on Geographic Information Science, Hanover, Germany.

Brown, B., Weber, D., 2012. Measuring change in place values using public participation GIS (PPGIS). Applied Geography 34, 316–324.

Bruniecki, K., Kulawiak, M., Moszyn, 2010. Concept of web service for real-time satellite imagery dissemination. Presented at the 2010 2nd International Conference on Information Technology (ICIT), pp. 149–152.

Bulatovic, V., Ninkov, T., Susic, Z., 2010. Open Geospatial Consortium Web Service in Complex Distribution Systems. Geodetski list 64, 13–29.

Caliński, T., Harabasz, J., 1974. A dendrite method for cluster analysis. Communications in Statistics - Theory and Methods 3, 1–27.

Carver, J.C., Kendall, R.P., Squires, S.E., Post, D.E., 2007. Software Development Environments for Scientific and Engineering Software: A Series of Case Studies, in: ICSE ‘07 Proceedings of the 29th International Conference on Software Engineering. pp. 550–559.

Chainey, S., Ratcliffe, J., 2005. GIS and crime mapping. John Wiley & Sons, London.

Chang, D.X., Zhang, X.D., Zheng, C.W., Zhang, D.M., 2010. A robust dynamic niching genetic algorithm with niche migration for automatic clustering problem. Pattern Recognition 43, 1346– 1360.

Chapanis, A., 1994. Hazards associated with three signal words and four colours on warning signs. Ergonomics 37, 265–275.

Chau, K., Sze, Y., Fung, M., Wong, W., Fong, E., Chan, L., 2004. Landslide hazard analysis for Hong Kong using landslide inventory and GIS. Computers and Geosciences 30, 429–443.

Chen, H., Wood, M.D., Linstead, C., Maltby, E., 2011a. Uncertainty analysis in a GIS-based multi- criteria analysis tool for river catchment management. Environmental Modelling & Software 26, 395–405.

Chen, J., Liu, J., Yang, X., Wang, Y., Yu, X., 2011b. The structure and spatial patterns of three desert shrub communities in the western Ordos Plateau: Implications for biodiversity conservation. Journal of Food Agriculture & Environment 9, 714–722.

Cheshire, J.A., Longley, P.A., 2012. Identifying spatial concentrations of surnames. International Journal of Geographical Information Science 26, 309–325.

255

References

Childs, C., 2009. The top nine reasons to use a file geodatabase [WWW Document]. URL http://www.esri.com/news/arcuser/0309/files/9reasons.pdf (accessed 3.30.12).

Chiu, T., Fang, D., Chen, J., Wang, Y., Jeris, C., 2001. A robust and scalable clustering algorithm for mixed type attributes in large database environment, in: Proceedings of the 7th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ‘01. ACM, New York, NY, USA, pp. 263–268.

Christensen, F., Bernard, L., Kanellopoulos, I., Nogueras-Iso, Peedell, S., Schade, S., Thorne, C., 2006. Building service oriented applications on top of a spatial data infrastructure‚ a forest fire assessment example, in: Proceedings of the AGILE-Shaping the Future of Geographic Information Science in Europe. Visegrad, Hungary.

CIESIN, Columbia University, CIAT, 2005. Gridded Population of the World Version 3 (GPWv3): Population Density Grids [WWW Document]. URL http://sedac.ciesin.columbia.edu/gpw (accessed 6.8.10).

CISR, 2012. The Journal of ERW and Mine Action [WWW Document]. URL http://maic.jmu.edu/journal/index/ (accessed 10.25.12).

Clarke, L.M., 1989. An experimental investigation of the communicative efficiency of point symbols on tourist maps. The Cartographic Journal 26, 105–110.

Cleveland, W.S., 1994. The elements of graphing data. AT&T Bell Laboratories.

Coleman, D.J., McLaughlin, J.D., Nichols, S., 1997. Building a Spatial Data Infrastructure, in: 64-th Permanent Congress Meeting of the Fédération Internationale Des Géomètres (FIG). Singapore, pp. 89–104.

Cömert, C., 2004. Web services and National Spatial Data Infrastructure (NSDI). Presented at the XXth ISPRS Congree, Istanbul, pp. 12–23.

Conradi, H., Fuggetta, A., 2002. Improving software process improvement. Software, IEEE 19, 92–99.

Cova, T.J., 1999. GIS in emergency management. Geographical information systems 2, 845–858.

Craglia, M., Campagna, M., 2009. Advanced Regional Spatial Data Infrastructures in Europe. J.R. Center, Ispra.

Crompvoets, J., Bregt, A., 2004. World Status of National Spatial Data Clearinghouses. Journal of the Urban and Regional Information Systems Association 15, 43–50.

Crossland, M.D., Wynne, B.E., Perkins, W.C., 1995. Spatial decision support systems: An overview of technology and a test of efficacy. Decision Support Systems 14, 219–235.

Cuevas, A., Febrero, M., Fraiman, R., 2001. Cluster analysis: a further approach based on density estimation. Computational Statistics and Data Analysis 36, 441 – 459.

Da Lage, O., 2011. RFI - Liban - Chebaa : les fermes de la discorde [WWW Document]. Radio France International. URL http://www.rfi.fr/Actufr/Articles/006/article_2721.asp (accessed 4.19.12).

Da Silva, J., Times, V.C., Fidalgo, R., Barros, R., 2004. Towards a Web Service for Geographic and Multidimensional Processing, in: Proceedings of the Brazilian Symposium on GeoInformatics. Campos do Jordao, Brazil.

256

References

Dalrymple-Alford, E.C., 1970. Measurement of clustering in free recall. Psychological Bulletin 74, 32– 34.

Data East, 2012. XTools Pro Extension or ArcGIS [WWW Document]. URL http://www.xtoolspro.com/ (accessed 9.26.12).

Dave, R.N., 1991. Characterization and detection of noise in clustering. Pattern Recognition Letters 12, 657 – 664.

Davies, D.L., Bouldin, D.W., 1979. A cluster separation measure. IEEE Transactions on Pattern Analysis and Machine Intelligence 224–227.

Davies, G., Whyatt, J.D., 2009. A least-cost approach to personal exposure reduction. Transactions in GIS 13, 229–246.

De Leener, I., Pastijn, H., 2002. Selecting land mine detection strategies by means of outranking MCDM techniques. European Journal of Operational Research 139, 327–338.

De Smith, M.J., Goodchild, M.F., Longley, P., 2007. Geospatial analysis: a comprehensive guide to principles, techniques and software tools. Troubador Publishing.

Delhay, S., Idrissa, M., Lacroix, V., 2005. PARADIS: GIS Tools for Humanitarian Demining, in: Proceedings of the 2nd International ISCRAM Conference. pp. 213–219.

Devlin, J., 2010. Mine Action funding: Trends, modalities and future prospects. Results of a survey of donor countries carried out in May-June 2010 [WWW Document]. URL http://www.gichd.org/fileadmin/pdf/ma_development/LMAD-Funding-Report-Nov2010.pdf (accessed 12.20.12).

Devlin, J., Naidoo, S., 2010. Mine-action Funding: GICHD Survey of Donor Countries. The Journal of ERW and Mine Action 14, 29–32.

Di Luzio, M., Srinivasan, R., Arnold, J.G., 2004. A GIS-Coupled Hydrological Model System for the Watershed Assessment of Agricultural Nonpoint and Point Sources of Pollution. Transactions in GIS 8, 113–136.

Dilts, T., 2010. Topography Tools for ArcGIS (9.3, 9.2, 9.1/9.0) [WWW Document]. URL http://arcscripts.esri.com/details.asp?dbid=15996 (accessed 11.16.12).

Dorling, D., 1993. Map design for census mapping. The Cartographic Journal 30, 167–183.

Dougenik, J.A., Chrisman, N.R., Niemeyer, D.R., 1985. An algorithm to construct continuous area cartograms. The Professional Geographer 37, 75–81.

Duczmal, L.H., Moreira, G.J.P., Burgarelli, D., Takahashi, R.H.C., Magalhaes, F.C.O., Bodevan, E.C., 2011. Voronoi distance based prospective space-time scans for point data sets: a dengue fever cluster analysis in a southeast Brazilian town. International Journal of Health Geographics 10, 29.

Duda, R.O., Hart, P.E., 1973. Pattern classification and scene analysis. John Wiley and Sons.

Dunbar, M.D., 2010. Mobile Geographic Information Systems (GIS) for Humanitarian Demining (Unpublished doctoral dissertation). University of Kansas.

257

References

Dye, A.S., Shaw, S.L., 2007. A GIS-based spatial decision support system for tourists of Great Smoky Mountains National Park. Journal of Retailing and Consumer Services 14, 269–278.

Eastman, J.R., Jiang, H., 1995. Fuzzy measures in multi-criteria evaluation, in: Proceedings of the 2nd International Symposium on Spatial Accuracy Assessment in Natural Resources and Environmental Studies. Fort Collins, Colorado, pp. 527 – 534.

Eldrandaly, K., Eldin, N., Sui, D., 2003. A COM-based Spatial Decision Support System for Industrial Site Selection. Journal of Geographic Information and Decision Analysis 7, 72–92.

Epperson, B.K., McRae, B.H., Scribner, K., Cushman, S.A., Rosenberg, M.S., Fortin, M.J., James, P., Dale, M.R., 2010. Utility of computer simulations in landscape genetics. Molecular Ecology 19, 3549–3564.

Erden, T., Coskun, M.Z., 2010. Multi-criteria site selection for fire services: the interaction with analytic hierarchy process and geographic information systems. Natural Hazards and Earth System Sciences 10, 2127–2134.

Eriksson, D., 2006. Modelling Projections of International Response to sudden-onset Disasters. Development of a Numerical Model Using Central Asian Earthquakes (PhD Thesis). University of Coventry.

Eriksson, D., 2008. Total Quality Management in Mine Action [WWW Document]. Journal of Mine Action. URL http://maic.jmu.edu/journal/12.1/notes/eriksson/eriksson.htm (accessed 10.22.12).

Eriksson, D., 2011. Information-management Activities at the GICHD. The Journal of ERW and Mine Action 15.

ESA, 2010. European Space Agency Ionia GlobCover Portal [WWW Document]. URL http://www.esa.int/esaEO/SEMGSY2IU7E_index_0.html (accessed 11.8.11).

ESA, 2011. ATSR - World Fire Atlas [WWW Document]. URL http://due.esrin.esa.int/wfa/ (accessed 12.16.11).

ESA, 2012. SADA (Space Assets for Demining Assistance) [WWW Document]. URL http://iap.esa.int/projects/security/space-assets-for-demining-main-page (accessed 9.18.12).

Esri, 2009a. ArcGIS Desktop Help 9.3, including 9.3.1 [WWW Document]. URL http://webhelp.esri.com/arcgisdesktop/9.3/index.cfm?TopicName=welcome (accessed 3.30.12).

Esri, 2009b. Services de carte hautes performances: cas d‘utilisation avec les données CORINE Land Cover. Esri, Redlands, California.

Esri, 2010. Performance and Throughput Tips for ArcGIS Server 9.3.1 Cached Map Services and the Apache HTTP Server. Esri, Redlands, California.

Esri, 2011a. ArcGIS Desktop Help 10.1 [WWW Document]. URL http://resources.arcgis.com/en/help/main/10.1/ (accessed 10.23.12).

Esri, 2011b. Esri Info | Our Qualifications [WWW Document]. URL http://www.esri.com/about- esri/about/qualifications.html (accessed 3.30.12).

Esri, 2011c. File Geodatabase API [WWW Document]. URL http://resources.arcgis.com/content/geodatabases/10.0/file-gdb-api (accessed 3.30.11).

258

References

Esri, 2012. ArcGIS Workflow Manager [WWW Document]. URL http://www.esri.com/software/arcgis/extensions/arcgis-workflow-manager (accessed 10.17.12).

Euntai Kim, Minkee Park, Seunghwan Ji, Mignon Park, 1997. A new approach to fuzzy modeling. Fuzzy Systems, IEEE Transactions on 5, 328–337.

European Commission, 2006. Assessing the impacts of Spatial Data Infrastructures. J.R. Center, Ispra: 1- 61.

European Commission, 2007a. INSPIRE Network Services Performance Guidelines. J.R. Center, Ispra: 1- 22.

European Commission, 2007b. Directive 2007/2/EC of the European Parliament and the Council of 14 March 2007 establishing an Infrastructure for Spatial Information in the European Community (INSPIRE). European Union: 14, Brussels.

European Commission, 2009. COMMISSION REGULATION (EC) No 976/2009 of 19 October 2009 implementing Directive 2007/2/EC of the European Parliament and of the Council as regards the Network Services. European Union: 11-20, Brussels.

European Commission, 2010. COMMISSION REGULATION (EU) No 1088/2010 of 23 November 2010 amending Regulation (EC) No 976/2009 as regards download services and transformation services. European Union: 1-10, Brussels.

Evans, B.J., 1997. Dynamic display of spatial data-reliability: Does it benefit the user? Computers & Geosciences (special issue on Exploratory Cartographic Visualization) 23, 409–422.

Ezigbalike, D., 2004. SDI - Africa: An Implementation Guide [WWW Document]. URL http://geoinfo.uneca.org/sdiafrica/default1.htm (accessed 7.30.12).

Foltete, J.C., Berthier, K., Cosson, J.F., 2008. Cost distance defined by a topological function of landscape. Ecological Modelling 210, 104–114.

Forrest, D., Castner, H., 1985. The Design and Perception of Point Symbols for Tourist Maps. The Cartographic Journal 22, 11–19.

Foss, C., Morris, D., Burnside, N.G., Ravenscroft, N., 2010. Champagne comes to England: assessing the potential of GIS in the identification of prime vinyard sites in south east England. Findings in Built and Rural Environments (FiBRE).

Fu, G., Jones, C.B., Abdelmoty, A.I., 2005. Building a Geographical Ontology for Intelligent Spatial Search on the Web, in: Proceedings of IASTED International Conference on Databases and Applications. Innsbrück, Austria.

Funk, V.A., Zermoglio, M.F., Nasir, N., 1999. Testing the use of specimen collection data and GIS in biodiversity exploration and conservation decision making in Guyana. Biodiversity and Conservation 8, 727–751.

Gaspar-Escribano, J.M., Iturrioz, T., 2011. Communicating earthquake risk: mapped parameters and cartographic representation. Natural Hazards and Earth System Sciences 11, 359–366.

Gasser, R., Knezevic, G., Carrier, M., 2011. Mine Risk Management by Mapping. The Journal of ERW and Mine Action 15, 46–49.

259

References

Gath, I., Geva, A.B., 2002. Unsupervised optimal fuzzy clustering. IEEE Transactions on Pattern Analysis and Machine Intelligence 11, 773–780.

Gatrell, A.C., Bailey, T.C., Diggle, P.J., Rowlingson, B.S., 1996. Spatial point pattern analysis and its application in geographical epidemiology. Transactions of the Institute of British Geographers 256–274.

GEO, 2005. Global Earth Observation System of Systems 10-Year Implementation Plan Reference Document. GEO.

GEO, 2011. GEO-Portal - Map Viewer [WWW Document]. URL http://www.geoportal.org/web/guest/geo_map_viewer (accessed 12.16.11).

Geoffrion, A.M., 1983. Can OR/MS evolve fast enough? INTERFACES 13, 10–25.

Gerber, R., Burden, P., Stanton, G., 1990. Development of public information symbols for tourism and recreational mapping. The Cartographic Journal 27, 92–103.

GICHD, 2005. Cartographic Recommendations for Humanitarian Demining Map Symbols in the Information Management System for Mine Action (IMSMA) [WWW Document]. URL http://www.gichd.org/fileadmin/pdf/IMSMA/IMSMA-Symbology-FinalReport.pdf (accessed 3.18.13).

GICHD, 2007. National Mine Action Plan for Completion - Fulfilling Obligations under Article 5 of the Antipersonnel Mine Ban Treaty 2008-2011 [WWW Document]. URL http://www.gichd.org/fileadmin/pdf/ma_development/nma-strat/NMAS-Mauritania-2007-2011- en.pdf (accessed 12.21.11).

GICHD, 2008a. A Guide to Road Clearance [WWW Document]. URL http://www.gichd.org/fileadmin/pdf/publications/Road-Clearance-2008.pdf (accessed 10.10.12).

GICHD, 2008b. A Guide to Marking and Fencing. GICHD, Geneva, Switzerland.

GICHD, 2009a. Linking Mine Action and Development - States Affected by Mines | ERW [WWW Document]. URL http://www.gichd.org/fileadmin/pdf/ma_development/Guidelines/Guidelines- LMAD-States-Nov2009.pdf (accessed 5.29.12).

GICHD, 2009b. A handbook of mechanical demining [WWW Document]. URL http://www.gichd.org/lima/reports-publications/detail/publications/a-handbook-of-mechanical- demining/ (accessed 10.25.12).

GICHD, 2010a. Resource Planning Tool for Humanitarian Demining and Mine Action [WWW Document]. URL http://proceedings.esri.com/library/userconf/proc09/uc/abstracts/a1827.html (accessed 3.18.13).

GICHD, 2010b. Guide to Mine Action, 4th ed. GICHD, Geneva, Switzerland.

GICHD, 2011. IMSMA Software [WWW Document]. URL http://www.gichd.org/main-menu/IMSMA (accessed 11.16.12).

GICHD, 2012. Overview and Strategic Chart [WWW Document]. URL http://www.gichd.org/ about- gichd/overview-and-strategic-chart (accessed 10.25.12).

Giuliani, G., 2011. Spatial Data Infrastructures for Environmental Sciences (PhD Thesis). University of Geneva, Geneva.

260

References

Giuliani, G., Peduzzi, P., 2011. The preview global risk data platform: a geoportal to serve and share global data on risk to natural hazards. Natural Hazards and Earth System Sciences 11, 53 – 66.

Goodchild, M.F., Kemp, K.K., 1990. NCGIA Core Curriculum in GIS. National Center for Geographic Information and analysis, University of California, Santa Barbara CA.

Goodchild, M.F., Palladino, S.D., 1995. Geographic information systems as a tool in science and technology education. Speculations in Science and Technology 18, 278–286.

Goslin, B., 2003. Making analytical tools operational: task impact assessment. Third World Quarterly 24, 923–938.

Granell, C., Diaz, L., Gould, M., 2009. Distributed geospatial processing services. Encyclopedia of Information Science and Technology 1186–1193.

Greenberg, J.A., Rueda, C., Hestir, E.L., Santos, M.J., Ustin, S.L., 2011. Least cost distance analysis for spatial interpolation. Computers & Geosciences 37, 272–276.

Grujic, Z., 2011. The Bosnia and Herzegovina Mine Action Information System. The Journal of ERW and Mine Action 15.

Härdle, W., Müller, M., Sperlich, S., Werwatz, A., 2004. Nonparametric and semiparametric models. Springer Verlag.

Havens, T.C., Stone, K., Keller, J.M., Ho, K.C., 2009. Sensor-fused detection of explosive hazards. SPIE 7303.

Helander, E., Programme des Nations Unies pour le développement, 1993. Prejudice and dignity: an introduction to community-based rehabilitation. United Nations Development Programme, New York.

Hennig, B., 2011. Views of the World | worldmapping beyond mere description [WWW Document]. URL http://www.viewsoftheworld.net/ (accessed 12.16.11).

Henricksen, B., 2007. UNSDI Compendium: A UNSDI Vision, Implementation Strategy and Reference Architecture. United Nations Geographic Information Working Group.

Herbsleb, J.D., Mockus, A., Finholt, T.A., Grinter, R.E., 2001. An empirical study of global software development: distance and speed, in: Proceedings of the 23rd International Conference on Software Engineering, ICSE ‘01. IEEE Computer Society, Washington, DC, USA, pp. 81–90.

Herzog, J., 2010. Mapping mine hazard. Estimating the global threat by explosive remnants of war (Master Thesis). University of Zürich.

Hijmans, R., 2012. GADM (Database of Global Administrative Areas) [WWW Document]. URL http://www.gadm.org (accessed 4.25.12).

Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology 25, 1965–1978.

Hill, M.J., Braaten, R., Veitch, S.M., Lees, B.G., Sharma, S., 2005. Multi-criteria decision analysis in spatial decision support: the ASSESS analytic hierarchy process and the role of quantitative methods and spatially explicit analysis. Environmental Modelling & Software 20, 955–976.

Hilton, B.N., 2007. Emerging Spatial Information Systems and Applications. IGI Global, Hershey, PA.

261

References

Hiratsuka, Y., Katoh, F., Konishi, K., Shin, S., 2010. A design method for minimum cost path of flying probe in-circuit testers, in: Proceedings of SICE Annual Conference. IEEE, Taipei, Taiwan, pp. 2933–2936.

Horak, J., Ardielli, J., Horakova, B., 2009. Testing of Web Map Services. Presented at the GSDI 11, Rotterdam, The Netherlands, pp. 1–25.

Hubert, L.J., Levin, J.R., 1976. A general statistical framework for assessing categorical clustering in free recall. Psychological Bulletin 83, 1072–1080.

ICBL, 2011a. Landmine Monitor Report 2011. Mines Action Canada, Canada.

ICBL, 2011b. Mine Contamination as of August 2011 [WWW Document]. URL http://www.the- monitor.org/cmm/2011/images/2011_Monitor_Mine_Contamination_full.jpg (accessed 11.26.12).

ICRC, 2012. Cartographie et action humanitaire : la réflexion et l‘expérience du CICR [WWW Document]. URL http://humanitaire.revues.org/index1305.html iMMAP, 2002. Report: Decision support for mine action [WWW Document]. URL http://www.gichd.org/fileadmin/pdf/evaluations/database/Kosovo/Decision_Support_For_Mine_ Action-Kosovo.pdf

IPCC. 2007. Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, eds. M.L. Parry, O.F. Canziani, J.P. Palutikof, P.J. van der Linden and C.E. Hanson. Cambridge , UK : Cambridge University Press, 976 pp.

ITF, 2001. Regional Map of areas with known contamination of mines and UXO [WWW Document]. URL http://www.see-demining.org/website/itf_new3/viewer.htm (accessed 12.21.11).

Jankowski, P., 1995. Integrating geographical information systems and multiple criteria decision making methods. International Journal of Geographical Information Systems 9, 251–273.

Jarvis, A., Reuter, H.I., Nelson, A., Guevara, E., 2008. Hole-filled seamless SRTM data V4, International Center for Tropical Agriculture (CIAT) [WWW Document]. URL http://srtm.csi.cgiar.org (accessed 3.30.12).

Javalgi, R.G., Jain, H.K., 1988. Integrating multiple criteria decision making models into the decision support system framework for marketing decisions. Naval research logistics 35, 575–596.

Jenks, G.F., Caspall, F.C., 1971. Error on choroplethic maps: Definition, measurement, reduction. Annals of the Association of American Geographers 61, 217–244.

Jones, M.C., Marron, J.S., Sheather, S.J., 1996. A brief survey of bandwidth selection for density estimation. Journal of the American Statistical Association 91, 401 – 407.

Jones, C.B., Purves, R.S., Clough, P.D., Joho, H., 2008. Modelling vague places with knowledge from the Web. International Journal of Geographical Information Science. 22, 1045–1065.

Kar, B., Hodgson, M.E., 2008. A GIS-Based Model to Determine Site Suitability of Emergency Evacuation Shelters. Transactions in GIS 12, 227–248.

Karnatak, H.C., Saran, S., Bhatia, K., Roy, P.S., 2007. Multicriteria spatial decision analysis in web GIS environment. Geoinformatica 11, 407–429.

262

References

Katajisto, J., Moilanen, A., 2006. Kernel-based Home Range Method for Data with Irregular Sampling Intervals. Ecological Modelling 194, 405–413.

Khoumeri, E.H., Benslimane, D., 2007. Web GIS for multiple representation data. Presented at the 2nd International Conference on Digital Information Management. ICDIM ‘07, pp. 188–192.

Kleindorfer, P.R., Kunreuther, H.C., Schoemaker, P.J.H., 1993. Decision sciences: An integrative perspective. Cambridge University Press, Cambridge.

Knezic, S., Mladineo, N., 2006. GIS-based DSS for priority setting in humanitarian mine-action. International Journal of Geographical Information Science 20, 565–588.

Köhler, P., Müller, M., Sanders, M., Wächter, J., 2006. Data management and GIS in the Center for Disaster Management and Risk Reduction Technology (CEDIM): from integrated spatial data to the mapping of risk. Natural Hazards and Earth System Sciences 6, 621–628.

Kostelnick, J., Dobson, J., Egbert, S., Dunbar, M., 2008. Cartographic Symbols for Humanitarian Demining. The Cartographic Journal 45, 18–31.

Krishna, K., Murty, M.N., 1999. Genetic K-means algorithm. IEEE Transactions on Systems Man and Cybernetics Part B-Cybernetics 29, 433–439.

Krugman, P., 1979. A model of innovation, technology transfer, and the world distribution of income. The Journal of Political Economy 253–266.

Kuehne, D., 2005. A CAD Geoprocessing Sample Toolbox for ArcGIS 9.0 [ESRI] [WWW Document]. URL http://arcscripts.esri.com/details.asp?dbid=13639 (accessed 11.16.12).

Lacasta, J., Nogueras-Iso, J., Bejar, R., Muro-Medrano, P.R., Zarazaga-Soria, F.J., 2007. A web ontology service to facilitate interoperability within a Spatial Data Infrastructure: Applicability to discovery. Data & Knowledge Engineering 63, 947–971.

Lacroix, P., 2012. How GIS contributes to Efficient Mine Action. Map gallery, GIS for the United Nations conference. Map gallery.

Lacroix, P., Eriksson, D., 2011. How can GIS benefit demining activity. Geoworld Magazine 25.

Lacroix, P., Herzog, J., Eriksson, D., 2011. Mapping populations at risk of ERW. The Journal of ERW and Mine Action 15.

Lacroix, V., Acheroy, M., Wolff, E., 2002. PARADIS: A prototype for assisting rational activities in humanitarian demining using images from satellites. The Journal of ERW and Mine Action 6.

Lance, K., Georgiadou, Y., Bregt, A.K., 2006. Understanding how and why practitioners evaluate SDI performance. International Journal of Spatial Data Infrastructures Research 1, 65–104.

Lange, T., Roth, V., Braun, M.L., Buhmann, J.M., 2004. Stability-based Validation of Clustering Solutions. Neural Computation 16, 1299–1323.

Laughery, K.R., Vaubel, K.P., Young, S.L., Brelsford Jr., J.W., Rowe, A.L., 1993. Explicitness of consequence information in warnings. Safety Science 16, 597–613.

Lee, J., Stucky, D., 1998. On applying viewshed analysis for determining least-cost paths on Digital Elevation Models. International Journal of Geographical Information Science 12, 891–905.

263

References

Lehmann, A., Overton, J.M.C., Leathwick, J.R., 2003. GRASP: generalized regression analysis and spatial prediction. Ecological modelling 160, 165–183.

Lehner, B., Döll, P., 2004. Development and validation of a global database of lakes, reservoirs and wetlands. Journal of Hydrology 296, 1–22.

Lehner, G., Verdin, K., Jarvis, A., 2008. New global hydrography derived from spaceborne elevation data. Eos, Transactions, AGU 89, 93–94.

Levine, N., 2010. CrimeStat: A spatial statistics program for the analysis of crime incident locations. Ned Levine & Associates and the National Institute of Justice, Houston, USA.

Lewis, T., Bennett, S., 2013. The juxtaposition and spatial disconnect of environmental justice declarations and actual risk: A new method and its application to New York State. Applied Geography 39, 57–66.

Liebman, K.A., Stoddard, S.T., Morrison, A.C., Rocha, C., Minnick, S., Sihuincha, M., Russell, K.L., Olson, J.G., Blair, P.J., Watts, D.M., Kochel, T., Scott, T.W., 2012. Spatial Dimensions of Dengue Virus Transmission across Interepidemic and Epidemic Periods in Iquitos, Peru (1999- 2003). PLOS Neglected Tropical Diseases 6.

Lieske, S.N., McLeod, D.M., Coupal, R.H., Srivastava, S.K., 2012. Determining the relationship between urban form and the costs of public services. Environment and Planning B-Planning & Design 39, 155–173.

Lisica, D., 2003. Bosnia and Herzegovina mine problem: Priority setting. Journal of Mine Action 7.2.

Liu, W., Chen, D., Scott, N.A., 2007. Effects of cell sizes on resistance surfaces in GIS-based cost distance modeling for landscape analyses, in: Proceedings SPIE 6754 K7540. SPIE-INT SOC OPTICAL ENGINEERING, Nanjing, China.

Lloyd, S., 1982. Least squares quantization in PCM. IEEE Transactions on Information Theory 28, 129– 137.

Loader, C.R., 1999. Bandwidth selection: Classical or plug-in? Annals of Statistics 27, 415–438.

Lokey, J., 2001. Global Focus on Landmines in Afghanistan [WWW Document]. URL http://maic.jmu.edu/journal/5.3/features/joe_lokey/joe_lokey2.htm (accessed 11.29.11).

Lorz, C., Fürst, C., Galic, Z., Matijasic, D., Podrazky, V., Potocic, N., Simoncic, P., Strauch, M., Vacik, H., Makeschin, F., 2010. GIS - based probability assessment of natural hazards in forested landscapes of central and South-Eastern Europe. Environmental Management 2010, 920 – 930.

Lüsher, P., Weibel, R., 2013. Exploiting Empirical Knowledge for Automatic Delineation of City Centres from Large-scale Topographic Databases. Computers, Environment and Urban Systems 37, 18– 34.

Lutz, M., Sprado, J., Klien, E., Schubert, C., Christ, I., 2009. Overcoming semantic heterogeneity in spatial data infrastructures. Computers & Geosciences 35, 739–752.

Ma, E.W.M., Chow, T.W.S., 2004. A new shifting grid clustering algorithm. Pattern Recognition 37, 503 – 514.

264

References

Ma, X., Pan, Q.H., Li, M.L., 2005. Integration and share of spatial data based on Web Service, in: Proceedings of PDCAT 2005: Sixth International Conference on Parallel and Distributed Computing, Applications and Technologies. Dalian, China, pp. 328–332.

MACCA, 2010. Mine Action Coordination Centre of Afghanistan [WWW Document]. URL http://www.macca.org.af (accessed 4.25.12).

MacQueen, J., 1965. Some methods for classification and analysis of multivariate observations, in: Proceedings of the 5th Berkeley Symposium on Mathematical Statistics and Probability. pp. 281–297.

Malczewski, J., 1997. Spatial Decision Support Systems. NCGIA Core Curriculum in GIScience.

Malczewski, J., 1999. GIS and Multicriteria Decision Analysis. John Wiley and Sons, New York.

Malczewski, J., 2006. Integrating multicriteria analysis and geographic information systems: the ordered weighted averaging (OWA) approach. International Journal of Environmental Technology and Management 6, 7–19.

Mansourian, A., Rajabifard, A., Valadan Zoej, M.J., Williamson, I., 2006. Using SDI and web-based system to facilitate disaster management. Computers & Geosciences 32, 303–315.

Marinoni, O., 2009. AHP 1.1 - Decision support tool for ArcGIS [WWW Document]. URL http://arcscripts.esri.com/details.asp?dbid=13764 (accessed 11.7.11).

Maslen, S., 2001. Anti-Personnel Mines Under Humanitarian Law: A View from the Vanishing Point. Intersentia nv.

Masser, I., 2005. The Future of Spatial Data Infrastructures. Presented at the ISPRS Workshop on Service and Application of Spatial Data Infrastructure, Hangzhou, China.

Matthies, M., Giupponi, C., Ostendorf, B., 2007. Environmental decision support systems: Current issues, methods and tools. Environmental Modelling & Software 22, 123–127.

McEachren, A.M., Robinson, A., Hopper, S., Gardner, S., Murray, R., Gahegan, M., Hetzler, E., 2005. Visualizing geospatial information uncertainty. What we know and what we need to know. Cartographic and Geographic Information Science 32, 139–160.

Menasce, D.A., 2002. Load testing of Web sites. Internet Computing, IEEE 6, 70–74.

Mendes, E., Mosley, N., 2006. Web Engineering. Springer.

Mentor Softwar, 2012. MGRS Conversion Utility [WWW Document]. URL http://mgrs-conversion- utility.software.informer.com/ (accessed 11.23.12).

Miller, D., Wang, Y., Kesidis, G., 2008. Emergent unsupervised clustering paradigms with potential application to bioinformatics. Front Biosci 13, 677–90.

Milligan, G.W., 1980. An examination of the effect of six types of error perturbation on fifteen clustering algorithms. Psychometrika 45, 325–342.

Milligan, G.W., 1981. A review of Monte Carlo tests of cluster analysis. Multivariate Behavioral Research 16, 379–407.

265

References

Milligan, G.W., Cooper, M.C., 1985. An examination of procedures for determining the number of clusters in a data set. Psychometrika 50, 159–179.

Mladineo, N., 2012. GIS-based Multi-Criteria Analysis of priority selection in humanitarian demining [WWW Document]. URL http://161.53.165.103/WebGIS/Labs/MINES/#/Home (accessed 7.3.12).

Mojena, R., 1977. Hierarchical grouping methods and stopping rules: An evaluation. The Computer Journal 20, 359–363.

Morrison, C., Forrest, D., 1995. A study of point symbol design for computer based large scale tourist mapping. The Cartographic Journal 32, 126–136.

Müller, M., Vorogushyn, S., Maier, P., Thieken, A.H., Petrow, T., Kron, A., Büchele, B., Wächter, J., 2006. CEDIM Risk Explorer-a map server solution in the project ―Risk map Germany‖. Natural Hazards and Earth System Sciences 6, 711–720.

Mülli, A.S., Paterson, T., 2012. Priority-setting in Mine Action: getting more Value for the Investment. The Journal of ERW and Mine Action 16.

Nakakawa, A., 2006. A spatial decision support tool for landfill site selection for municipal solid waste management (Master Thesis).

Navulur, K. C. S., Engel B.A., 1998. Groundwater vulnerability assessment to non-point source nitrate pollution on a regional scale using GIS.Transactions of the ASAE 41(6), 1671–1678.

Nebert, D.D., 2004. Developing Spatial Data Infrastructure: the SDI Cookbook [WWW Document]. URL http://sdi.abudhabi.ae/Sites/SDI/Content/AR/PDF/sdi- cookbook,property%3Dpdf,bereich%3Dsdi,sprache%3Dar,rwb%3Dtrue.pdf (accessed 11.2.12).

Newbury, D.E., Bright, D.S., 1999. Logarithmic 3-band color encoding: Robust method for display and comparison of compositional maps in electron probe X-ray microanalysis. Microscopy and Microanalysis 5, 333–343.

O‘Dea, E., Haddad, T., Dunne, D., Walsh, K., 2011. Coastal Web Atlas Features, in: Wright, D., Dwyer, E., Cummins, V. (Eds.), Coastal Informatics: Web Atlas Design and Implementation. Information Science Reference, IGI Global, pp. 12–32.

O‘Shea, M.T., 2012. Between the map and the reality : some fundamental myths of Kurdish nationalism. Peuples méditerranéens 165–183.

O‘Sullivan, D., Unwin, D.J., 2010. Geographic Information Analysis, 2nd. edition. ed. John Wiley & Sons, New Jersey.

Odaka, T., Ashiya, K., Tsukada, S., Sato, S., Ohtake, K., Nozaka, D., 2003. A New Method of Quickly Estimating Epicentral Distance and Magnitude from a Single Seismic Record. Bulletin of the Seismological Society of America 93, 526–532.

OGC, 2005. Web Feature Service Implementation Specification. OGC: 131.

OGC, 2006a. OpenGIS Web Map Server Implementation Specification. OGC: 85.

OGC, 2006b. Web Coverage Service (WCS) Implementation Specification.

266

References

OGC, 2007a. OpenGIS® Catalogue service Implementation Specification 2.0.2 - ISO Metadata Application Profile [WWW Document]. URL http://www.opengeospatial.org/standards/cat (accessed 11.11.12).

OGC, 2007b. OpenGIS Geography Markup Language (GML) Encoding Standard. OGC: 1-437.

OGC, 2007c. OpenGIS Web Processing Service. OGC: 87.

OGC, 2007d. OpenGIS Catalogue Services Specification. OGC: 218.

OGC, 2010a. OGC® WCS 2.0 Interface Standard - Core.

OGC, 2010b. OpenGIS Web Feature Service 2.0 Interface Standard.

Olson, D.M., Dinerstein, E., Wikramanayake, E.D., Burgess, N.D., Powell, G.V.N., Underwood, E.C., D‘Amico, J.A., Itoua, I., Strand, H.E., Morrison, J.C., Loucks, C.J., Allnutt, T.F., Ricketts, T.H., Kura, Y., Lamoreux, J.F., Wettengel, W.W., Hedao, P., Kassem, K.R., 2001. Terrestrial Ecoregions of the World. BioScience 51, 933–938.

OpenGeo team, 2010. GeoServer in Production. OpenGeo, New-York.

Openshaw, S., 1984. The modifiable areal unit problem. Geo Books, Norwich.

OpenStreetMap contributors, 2012. OpenStreetMap [WWW Document]. URL http://www.openstreetmap.org (accessed 8.31.12).

ORNL, 2008. LandscanTM Global Population Database [WWW Document]. URL http://www.ornl.gov/sci/landscan/ (accessed 4.3.12).

Ozimec, A.M., Natter, M., Reutterer, T., 2010. Geographical Information Systems-based marketing decisions: Effects of alternative visualizations on decision quality. Journal of Marketing 74, 94– 110.

PAICMA, 2012. PAICMA Website [WWW Document]. URL http://www.accioncontraminas.gov.co (accessed 8.31.12).

Paul, M., Ghosh, S.K., 2006. An approach for service oriented discovery and retrieval of spatial data, in: SOSE ‘06: Proceedings of the 2006 International Workshop on Service-oriented Software Engineering. ACM, pp. 88–94.

Peng, Z.-R., Zhang, C., 2004. The roles of geography markup language (GML), scalable vector graphics (SVG), and Web feature service (WFS) specifications in the development of Internet geographic information systems (GIS). Journal of Geographical Systems 6, 95–116.

Peterson, M.P., 1979. An evaluation of unclassed crossed-line choropleth mapping. Cartography and Geographic Information Science 6, 21–37.

Pingel, T.J., 2010. Modeling slope as a contributor to route selection in mountainous areas. Cartography and Geographic Information Science 37, 137–148.

Pinto, N., Keitt, T.H., 2009. Beyond the least-cost path: evaluating corridor redundancy using a graph- theoretic approach. Landscape Ecology 24, 253–266.

Piro, P., Nock, R., Nielsen, F., Barlaud, M., 2012. Leveraging k-NN for generic classification boosting. Neurocomputing 80, 3–9.

267

References

Pitz, G.F., McKillip, J., 1984. Decision analysis for program evaluators. Sage Publications, Beverly Hills.

Pohekar, S.D., Ramachandran, M., 2004. Application of multi-criteria decision making to sustainable energy planning - A review. Renewable and Sustainable Energy Reviews 8, 365–381.

Potts, R., 1952. Some Generalized Order-Disorder Transformations. Mathematical Proceedings 48, 106– 109.

Rahman, M.A., Rusteberg, B., Gogu, R.C., Lobo Ferreira, J.P., Sauter, M., 2012. A new spatial multi- criteria decision support tool for site selection for implementation of managed aquifer recharge. Journal of Environmental Management 99, 61–75.

Rajabifard, A., Williamson, I.P., 2001. Spatial Data Infrastructures: Concept, SDI Hierarchy and Future directions. Presented at the Geomatics‘80, p. 10.

Raleigh, C., Linke, A., Hegre, H., Karlsen, J., 2010. Introducing ACLED: An Armed Conflict Location and Event Dataset Special Data Feature. Journal of Peace Research 47, 651–660.

Rao, R.V., Davim, J.P., 2006. A decision-making framework model for material selection using a combined multiple attribute decision-making method. The International Journal of Advanced Manufacturing Technology 35, 751–760.

Ray, A.A., 1982. SAS user‘s guide: basics. SAS Institute.

Ray, N., 2005. PATHMATRIX: a geographical information system tool to compute effective distances among samples. Molecular Ecology Notes 5, 177–180.

Ray, N., Burgman, M.A., 2006. Subjective uncertainties in habitat suitability maps. Ecological Modelling 195, 172–186.

Rayfield, B., Fortin, M.J., Fall, A., 2010. The sensitivity of least-cost habitat graphs to relative cost surface values. Landscape Ecology 25, 519–532.

Rekacewicz, P., 2003. Land mines in the Balkans, UNEP/GRID-Arendal Maps and Graphics Library [WWW Document]. URL http://maps.grida.no/go/graphic/land_mines_in_the_balkans (accessed 11.29.11).

Reveiu, A., 2011. Techniques for representation of regional clusters in geographical information systems. Informatica Economica 15, 129–139.

Rice, J.M., Halpern, C.B., Antos, J.A., Jones, J.A., 2012. Spatio-temporal patterns of tree establishment are indicative of biotic interactions during early invasion of a montane meadow. Plant Ecology 213, 555–568.

Riese, S.R., Brown, D.E., Haimes, Y.Y., 2006. Estimating the probability of landmine contamination. Military Operations Research 11, 49–61.

Rigopoulos, G., Askounis, D.T., Metaxiotis, K., 2010. NeXCLass: A Decision Support System for non- ordered Multicriteria Classification. International Journal of Information Technology and Decision Making 9, 53–79.

Roberts, J., 2009. Marine Geospatial Ecology Tools 0.7 [WWW Document]. URL http://arcscripts.esri.com/details.asp?dbid=16498 (accessed 11.16.12).

Rolland, J.C., 2003. Lebanon: current issues and background. Nova Publishers.

268

References

Rushton, G., 2003. Public health, GIS, and spatial analytic tools. Annual Review of Public Health 24, 43– 56.

Ryttersgaard, J., 2001a. Spatial data infrastructure: Developing trends and challenges. Presented at the Second Meeting of the Committee on Development Information, 4-7 September 2001, Addis Ababa.

Ryttersgaard, J., 2001b. Spatial Data Infrastructure: Developing Trends and Challenges. Presented at the International Conference on Spatial Information for Sustainable Development.

Saaty, T.L., 1980. The Analytic Hierarchy Process. McGraw-Hill, New York.

Saaty, T.L., 1990. How to make a decision: The Analytic Hierarchy Process. European Journal of Operational Research 48, 9–26.

Sahin, K., Gumusay, M.U., 2008. Service oriented architecture (SOA) based web services for geographic information systems. Presented at the 21st ISPRS Congress, Beijing, China, pp. 625–630.

Samadi Alinia, H., Delavara, M.R., 2009. Applications of Spatial Data Infrastructures in Disaster Management. Presented at the GSDI-11, Rotterdam, The Netherlands.

Sangtani, J., Serpen, G., 2010. Automated composition of Web service workflow - A novel QoS-enabled multi-criteria cost search algorithm, in: ENASE. Proceedings of the 5th International Conference on Evaluation of Novel Approaches to Software Engineering Location. Athens, Grece.

Sanyal, Joy, Lu X.X., 2005. Remote sensing and GIS‐based flood vulnerability assessment of human settlements: a case study of Gangetic West Bengal, India. Hydrological Processes 19(18), 3699– 3716. Sarle, W.S., 1983. Cubic clustering criterion. SAS Institute, Cary, NC.

Sasi, Newman, M., 2006. Worldmapper: The world as you‘ve never seen it before [WWW Document]. URL http://www.worldmapper.org/display.php?selected=290 (accessed 12.16.11).

Savopol, F., Armenakis, C., 2002. Merging of Heterogeneous Data for Emergency Mapping: Data Integration or Data Fusion? International Archives of Photogrammetry Remote Sensing and Spatial Information Sciences 34, 668–674.

Sayar, A., Pierce, M., Fox, G.C., 2005. OGC Compatible Geographical Information Systems Web Services ( No. TR610). School of Informatics and Computing, Bloomington.

Schaeffer, B., 2008. Towards a Transactional Web Processing Service (WPS-T), in: Proceedings of the GI-Days. Muenster, Germany.

Schoemaker, P.J.H., Waid, C.C., 1982. An experimental comparison of different approaches to determining weights in additive utility models. Management Science 28, 182–196.

Scott, D.W., 1992. Multivariate density estimation: theory, practice, and visualization. John Wiley & Sons, United States.

Seaman, D.E., Powell, R.A., 1996. An evaluation of the accuracy of kernel density estimators for home range analysis. Ecology 77, 2075 – 2085.

Sheather, S.J., Jones, M.C., 1991. A reliable data-based bandwidth selection method for kernel density estimation. Journal of the Royal Statistical Society. Series B (Methodological) 53, 683–690.

269

References

Sherrouse, B.C., Clement, J.M., Semmens, D.J., 2011. A GIS application for assessing, mapping, and quantifying the social values of ecosystem services. Applied Geography 31, 748–760.

Siddiqui, M.Z., Everett, J.W., Vieux, B.E., 1996. Landfill siting using geographic information systems: A demonstration. Journal of environmental engineering 122, 515–523.

Silverman, B.W., 1986. Density estimation for statistics and data analysis. Vol 26 of Monographs on statistics and applied probability. Chapman & Hall, London.

Simonis, I., Sliwinski, A., 2005. Quality of Service in a Global SDI, in: Proceedings of the 8th International Conference for Global Spatial Data Infrastructure.

Slocum, T.A., McMaster, R.B., Kessler, F.C., Howard, H.H., 2009. Thematic cartography and geovisualization, 3rd ed. Prentice Hall, United States.

Sneath, P.H.., Sokal, R.R., 1973. Numerical taxonomy. The principles and practice of numerical classification.

Snow, J., Frost, W.H., Richardson, S.B., 1965. Snow on cholera, Hafner. ed. New-York.

Sokal, R.R., Michener, C.D., 1958. A statistical method for evaluating systematic relationships. Univ. Kans. Sci. Bull. 38, 1409–1438.

Spatial Ecology LLC, 2012. Geospatial Modelling Environment [WWW Document]. URL http://www.spatialecology.com/gme/index.htm (accessed 11.16.12).

Steinbach, M., Karypis, G., Kumar, V., 2000. A comparison of document clustering techniques, in: KDD Workshop on Text Mining. pp. 525–526.

Stöckli, R., Vermote, E., Saleous, N., Simmon, R., Herring, D., 2005. The Blue Marble Next Generation - A true color earth dataset including seasonal dynamics from MODIS. NASA Earth Observatory.

Sugumaran, R., Degroote, J., 2010. Spatial Decision Support Systems: Principles and Practices, 1st ed. CRC Press, Boca Raton, FL.

Sun, H., Li, Z., 2010. Effectiveness of cartogram for the representation of spatial data. The Cartographic Journal 47, 12–21.

Suvinen, A., 2006. A GIS-based simulation model for terrain tractability. Journal of Terramechanics 43, 427–449.

Taylor, S., 2002. Moving Forward: Recommendations for a Landmine Victim Data Collection and Management System. Landmines in Africa.

Tchoukanski, I., 2009. ET Geowizards LT [WWW Document]. URL http://arcscripts.esri.com/details.asp?dbid=11903 (accessed 11.16.12).

Tobler, W.R., 1970. A computer movie simulating urban growth in the Detroit region. Economic Geography 46, 234–240.

Trame, J., Keßler, C., 2011. Exploring the lineage of volunteered geographic information with heat maps. In: Proceedings of the Geoviz 2011 Conference, Hamburg, Germany.

270

References

Tu, S., Flanagin, M., Wu, Y., Abdelguerfi, M., Normand, E., Mahadevan, V., 2004. Design strategies to improve performance of GIS Web services, in: Proceedings of the International Conference on Information Technology: Coding and Computing (ITCC). IEEE, pp. 444–448 Vol.2.

UNDP, 2011. UNDP Sudan | Project: Mine Action Capacity Building and Programme Development [WWW Document]. URL http://www.sd.undp.org/projects/mine.htm (accessed 12.21.11).

UNMAS, 2003. Glossary of mine action terms, definitions and abbreviations [WWW Document]. URL http://www.mineactionstandards.org/international-standards/imas-in-english/list-of-imas/ (accessed 8.24.11).

UNMAS, 2010. E-MINE - Mine Clearance [WWW Document]. URL http://www.mineaction.org/overview.asp?o=16 (accessed 5.30.12).

UNMAS, 2011. E-MINE - What Is Mine Action? [WWW Document]. URL http://www.mineaction.org/section.asp?s=what_is_mine_action (accessed 12.16.11).

UNMAS, 2012. Mines on the Map [WWW Document]. URL http://www.mineactionmap.org/index.php?page=map (accessed 9.20.12).

Uran, O., Janssen, R., 2003. Why are spatial decision support systems not used? Some experiences from the Netherlands. Computers, Environment and Urban Systems 27, 511–526.

Valente, R.O.A., Vettorazzi, C.A., 2008. Definition of priority areas for forest conservation through the ordered weighted averaging method. Forest Ecology and Management 256, 1408–1417.

Vandenbroucke, D., 2005. Spatial Data Infrastructures in Europe: State of play Spring 2005 (Summary report of a study commissioned by the EC, EUROSTAT & DGENV).

Vanmeulebrouk, B., Bulens, J., Krause, A., De Groot, H., 2009. OGC standards in daily practice: gaps and difficulties found in their use, in: Proceedings GSDI-11 Building SDI Bridges to Address Global Challenges. Rotterdam, The Netherlands.

Van Riper, C.J., Kyle, G.T., Sutton, S.G., Barnes, M., Sherrouse, B.C., 2012. Mapping outdoor recreationists‘ perceived social values for ecosystem services at Hinchinbrook Island National Park, Australia. Applied Geography 35, 164–173.

Veitch, S.M., Bowyer, J.K., 1996. ASSESS: A system for Selecting Suitable Sites, in: Raster Imagery in Geographic Information Systems. OnWord Press, Santa Fe, p. 495.

Verjee, F., 2010. GIS Tutorial for Humanitarian Assistance, Pap/DVD. ed. ESRI Press.

Vistisen, J.B., 2006. Risk assessments of minefields in humanitarian mine action - A Bayesian Approach (PhD Thesis). Technical University of Denmark, Lyngby, Denmark.

Wartmann, F., Purves, R.S., Van Schaik, C., 2010. Modelling ranging behaviour of female orang-utans: a case study in Tuanan, Central Kalimantan, Indonesia. Primates 51, 119–130.

Wei, Y., Santhana-Vannan, S.K., Cook, R.B., 2009. Discover, visualize, and deliver geospatial data through OGC standards-based WebGIS system, in: 17th International Conference on Geoinformatics. pp. 1–6.

Wellar, B., 1990. Science, Applications, Coherence and GIS: Seizing the Moment, in: GIS/LIS ‘90 Proceedings. Anaheim, California.

271

References

WHO, 2010. Health Component, in: Community-based Rehabilitation: CBR Guidelines. World Health Organization, Geneva, Switzerland, p. 79.

Williams, C., Dunn, C.E., 2003. GIS in participatory research: Assessing the impact of landmines on communities in north-west Cambodia. Transactions in GIS 7, 393–410.

Williamson, D., McLafferty, S., Goldsmith, V., McGuire, P., Mollenkopf, J., 1998. Smoothing crime incident data: New methods for determining the bandwidth in kernel estimation, in: 18th Annual Esri User Conference, 23-31 July 1998. San Diego, California.

Witmer, F.D.W., O‘Loughlin, J., 2009. Satellite Data Methods and Application in the Evaluation of War Outcomes: Abandoned Agricultural Land in Bosnia-Herzegovina After the 1992-1995 Conflict. Annals of the Association of American Geographers 99, 1033–1044.

Wogalter, M.S., Kalsher, M.J., Frederick, L.J., Magurno, A.B., Brewster, B.M., 1998. Hazard level perceptions of warning components and configurations. International Journal of Cognitive Ergonomics 123–143.

Wong, C.-C., Chen, C.-C., Su, M.-C., 2001. A novel algorithm for data clustering. Pattern Recognition 34, 425–442.

Wong, G., 2009. Snapshot hypersectral imaging and practical applications. Journal of Physics: Conference Series 178, 1–5.

Xu, R., Wunsch, D., 2005. Survey of clustering algorithms. IEEE Transactions on neural networks 16, 645–678.

Yanhua, C., Qingjie, Z., Shilin, L., 2009. GIS-Based Multi-criteria Evaluation of Land Use Suitability for Disaster Prevention in Tangshan City, in: Proceedings of the International Forum on Information Technology and Applications. Chengdu, China, pp. 680–683.

Yildirim, V., Yomralioglu, T., 2011. NABUCCO pipeline route selection through Turkey comparison of a GIS-based approach to a traditional route selection approach. OIL GAS European Magazine 37, 20–24.

Yu, C., Lee, J., Munro-Stasiuk, M.J., 2003. Extensions to least-cost path algorithms for roadway planning. International Journal of Geographical Information Science 17, 361–376.

Yvinec, Y., Renaissance, A.S.B.L., 2005. A validated method to help area reduction in mine action with remote sensing data, in: Proceedings of the 4th International Symposium on Image and Signal Processing and Analysis (ISPA 2005). Zagreb, Croatia.

Zamorano, M., Molero, E., Hurtado, A., Grindlay, A., Ramos, A., 2008. Evaluation of a municipal landfill site in Southern Spain with GIS-aided methodology. Journal of hazardous materials 160, 473–481.

Zare, A., Bolton, J., Gader, P., Schatten, M., 2008. Vegetation mapping for landmine detection using long-wave hyperspectral imagery. IEEE Transactions on Geoscience and Remote Sensing 46, 172 – 178.

Zhang, C., 2005. The Roles of Web Feature and Web Map Services in Real-time Geospatial Data Sharing for Time-critical Applications. Cartography and Geographic Information Science 32, 269–283.

Zhao, P.S., Di, L.P., Yu, G.N., Yue, P., Wei, Y.X., Yang, W.L., 2009. Semantic Web-based geospatial knowledge transformation. Computers & Geosciences 35, 798–808.

272

References

Zhuang, Y., Rui, Y., Huang, T.S., Mehrotra, S., 1998. Adaptive key frame extraction using unsupervised clustering, in: 1998 International Conference on Image Processing. pp. Vol.1, 866–870.

273

List of Figures

274

List of Figures

Figure 1: Mine contamination as of 2011 (source: ICBL 2011b) ...... 23 Figure 2: Contributions of GIS to efficient mine action. Overview of the three research areas of this PhD thesis and the GIS research projects that were conducted ...... 54 Figure 3: A DHA stored as a polygon in the IMSMANG MySQL database ...... 63 Figure 4: Contamination by ERW: overview of the six tested methods (Afghanistan, 6‘644 ERW) ...... 68 Figure 5: KDE of a multivariate point dataset ...... 70 Figure 6: Comparison between methods D and E, both applied to the same dataset along the Cypriot buffer zone, where long polygons are encountered ...... 71 Figure 7: (a) Influence of kernel bandwidth on visualisation and obfuscation, (b) Logarithmic histograms of KDE in Afghanistan ...... 74 Figure 8: Effect of the density of ERW input data on the precision of the output maps, (a) For method D, using data from several test countries, (b) For method D using random subsets of the same dataset, and (c) For method E ...... 76 Figure 9: Possible application of the proposed cartographic visualization methods: multiplication of contamination maps (upper left: method C using Afghanistan data at the district level) by population datasets (upper right: Gridded Population of the World (CIESIN and CIAT 2005)) shows where people are most at risk (bottom)...... 81 Figure 10: Maps showing contamination by ERW in Afghanistan. The map in the upper left corner corresponds to method A (dots). For the other maps, technique E (KDE on polygonal ERW) was applied in Afghan regions of varying extent and ERW density, and an identical colour ramp was used for each map ...... 83 Figure 11: Results of the Multi-Distance Spatial Cluster Analysis (Ripley‘s K function) tests ...... 91 Figure 12: Hazards versus distance to roads...... 92 Figure 13: Workflow of the ADKNN-Clusters mapping method ...... 97 Figure 14: Comparative clustering tests on point patterns with different stability-based validation processes. ―Expected‖ numbers of clusters are in blue, results with Wong in maroon results with Wong combined with the VRC in red ...... 99 Figure 15: Density rasters in Afghanistan. (a), (b), and (c): ADKNN-Clusters method applied to IMSMANG subsets at different scales; (d) and (f): ADKNN-Points method applied at national level; (e): ADKNN-Clusters method applied to random points ...... 105 Figure 16: Weaknesses of ADKNN-Clusters mapping method: (a) When applied to data from non- neighbouring countries, ADKNN-Clusters generates one cluster per country and gets similar as ADKNN- Points, and (b) Big pollution in the light grey circle is due to one outlier ...... 106 Figure 17: The LandScan global population dataset ...... 109 Figure 18: LandScan population dataset for Afghanistan ...... 110 Figure 19: GPW population density in Afghanistan ...... 111 Figure 20: Point data for Afghanistan, every marker represents a hazard (point or centre of a polygon). Source: MACCA ...... 113 Figure 21: Density map showing estimated contamination in Afghanistan calculated using the ArcGIS Kernel density tool. Source: MACCA ...... 114 Figure 22: Populations at risk of ERW hazards in Afghanistan, calculated with the LandScan dataset .. 115 Figure 23: Populations at risk of mines in Afghanistan, calculated with the Gridded Population of the World (GPW) dataset, at second administrative unit level ...... 116

275

List of Figures

Figure 24 : Raster generator GUI ...... 121 Figure 25 : Assessing population vulnerability to ERW. The blue boxes represent the current research. The red boxes are possible further research directions ...... 123 Figure 26: The GlobCover dataset...... 131 Figure 27: GlobCover dataset for Mozambique ...... 132 Figure 28: The slope dataset for Mozambique ...... 133 Figure 29: Overview of the model ―Operational difficulty of demining‖. A scalable PDF is available from: http://maic.jmu.edu/journal/16.3/rd/lacroix.htm ...... 136 Figure 30: Output raster representing the operational difficulty of demining in Mozambique. The presented results are for a ―fictive (demining) machine with medium class characteristics and commonly used in many countries ...... 139 Figure 31: Parameters provided to users at the execution of the model ...... 140 Figure 32: Step 1 of the workflow: OD matrix showing the accidents that are within 2 hours of roads travel of existing medical facilities ...... 147 Figure 33: Step 2 of the workflow: OD matrix with a 2-hour cut-off value. Origin features are the accidents that are not covered by existing facilities. Destination features are the human settlements ..... 148 Figure 34: Step 4 of the workflow: OD matrix with a 2-hour cut-off value. Origin features are all accident locations. Destination features are the 14 original hospitals plus the new medical centre in ―Ficticia‖ ... 149 Figure 35: Overview of the NAMA model: Steps 1-2-3 ...... 150 Figure 36: Parameters provided to users at the execution of the model ...... 151 Figure 37: Integration of the AHP in MASCOT. (a) Weighting of each road category with the AHP. (b) Absolute weights and consistency ratio ...... 162 Figure 38: Scoring vector items with MASCOT in function of nearby vector and raster objects/categories ...... 163 Figure 39: MASCOT workflow. It is possible to achieve the entire workflow without closing ArcGIS Desktop...... 164 Figure 40: MASCOT tree. Criteria (vector and raster layers) are grouped by thematic in criteria groups ...... 165 Figure 41: Weighting categories and slices. (a) Each land cover category is a sub-criterion. (b) Weighting land cover categories by direct input. (c) Weighting slices of population density by direct input ...... 168 Figure 42: Weighting process. At each step, weighting can be done by direct input or through the AHP. (a) First step: weighting at sub-criteria level. (b) Second step: weighting at criteria level. (c) Third step: weighting criteria groups ...... 169 Figure 43: Weighting as a decreasing function of distance to the scored item (vector layers only) ...... 170 Figure 44: Scored items feature class, with one specific score for each criterion plus the overall score . 171 Figure 45: Application of MASCOT to mine action: scoring ERW to determine areas with high clearance priority ...... 171 Figure 46: START and its seven drop-down menus. Tools that have no direct equivalent in ArcToolbox are indicated with a star (*) symbol ...... 187 Figure 47: A possible workflow for humanitarian demining actors using START. The large boxes (A, B, and C) represent major steps in the workflow. The grey boxes are input or output of the various sub- analyses. START tools used in this workflow are indicated between the boxes with their respectives name and icon. See main text for explanations ...... 190

276

List of Figures

Figure 48: Use of the size variable to identify a number of victims ...... 197 Figure 49: Generic symbols for ten categories of objects...... 198 Figure 50: The hazard by status sub-category ...... 198 Figure 51: Changes in the Location symbol ...... 198 Figure 52: The new sub-munition icon ...... 199 Figure 53: Colour and value are used in the Task by Status category to differentiate between objects ... 199 Figure 54 : SERWIS architecture, based on ArcGIS Server ...... 207 Figure 55: Architecture used for testing different OWS implementations ...... 216 Figure 56: WMS results under various conditions (OWS implementation and file formats) ...... 221 Figure 57: WFS ArcGIS Server tests results ...... 222 Figure 58: WFS Geoserver test results ...... 223 Figure 59: Comparison of best result obtained on each software implementation: GeoServer and ArcGIS Server ...... 224 Figure 60: WCS ArcGIS Server tests results ...... 226 Figure 61: WCS Geoserver tests results ...... 227

277

List of Tables

278

List of Tables

Table 1: Overview of the test datasets ...... 65 Table 2: Comparison between statistical indicators derived from KDE rasters, representing the six datasets. All values shown in the table are densities of ERW/km2. They were computed on the test data and do not reflect the reality in the field ...... 77 Table 3: Overview of the three novel KDE-based mapping methods developed for this thesis ...... 86 Table 4: Test data used for development of the ADKNN-Clusters mapping method ...... 87 Table 5: Results of the ArcGIS Spatial Autocorrelation (Morans I) tests ...... 89 Table 6: Comparison between statistical indicators derived from KDE rasters, representing three different subsets of the IMSMANG Afghan dataset. All values shown in the table are densities of ERW/km2. They were computed on test data and are not meant to reflect reality in the field...... 101 Table 7: Comparative statistics between density rasters produced with ADKNN-Points and ADKNN- Clusters in Afghanistan ...... 101 Table 8: Comparison of LandScan to GPW ...... 111 Table 9: Recommendations: which mapping methods are most suitable for which category of humanitarian demining actors? ...... 119 Table 10: Main characteristics of the input datasets ...... 135 Table 11: Degree of operational difficulty of demining ...... 136 Table 12: Classification of the input layers in four categories of operational difficulty ...... 137 Table 13: Weighting of the input layers. Weights that are provided in this table are fictive and shall not reflect reality...... 138 Table 14: Main characteristics of the input datasets ...... 145 Table 15: Number of accidents covered: comparison with (Step 1) / without (Step 4) the newly implemented hospital in ―Ficticia‖. A 2-hour cut-off time is applied ...... 151 Table 16: Mean and median charge per medical centre (Step 5): comparison with / without the newly implemented hospital in ―Ficticia‖. No cut-off time value is applied ...... 152 Table 17: Statistics about the travel time and distance covered (Step 5). Comparison with / without the newly implemented hospital in ―Ficticia‖. No cut-off time value is applied ...... 152 Table 18: AHP: scale of relative importance for pair-wise comparison ...... 160 Table 19: Pair-wise intensity value description ...... 160 Table 20: Random consistency indices for different number of criteria ...... 161 Table 21: Strengths and weaknesses of major weighting methods. N is the number of criteria ...... 161 Table 22: Recommendations for data preparation ...... 164 Table 23: Typical MASCOT use case ...... 165 Table 24: Overview of the three GIS-based tools developed for helping decision-making in mine action ...... 176 Table 25: Test data used to develop 5D, NAMA and MASCOT ...... 179 Table 26: Bertin‘s visual variables and their properties ...... 196 Table 27: Characteristics of the test environment. All network interfaces between computers were based on 1GB LAN connections ...... 217 Table 28: Summary of the WFS testing scenarios. Case #1 is the ―base case‖ ...... 217

279

List of Tables

Table 29: Summary of the WCS testing scenarios. Case #1 is the ―base case‖ ...... 218 Table 30. WMS results comparison. FOSS4G benchmark 2009 in italic. CGI: Common Gateway Interface, SHP: shapefile, FCGI: Fast CGI, MS4W: MapServer for Windows...... 220 Table 31: WFS ArcGIS Server tests results. Refer to Table 30 for acronyms ...... 221 Table 32: WFS Geoserver tests results ...... 222 Table 33: WCS Geoserver (GS) and ArcGIS Server (AGS) tests results. An ArcGIS Server Image Service (IS) has also been tested. Tests have been executed on Linux and Windows (win) operating systems. Lr: Low-resolution, Mr: Medium-resolution, Hr: High-resolution ...... 225

280