Modelling projections of international response to sudden-onset disasters - Development of a...

MODELLING PROJECTIONS OF INTERNATIONAL RESPONSE TO

SUDDEN-ONSET DISASTERS

Development of a Numerical Model Using

Central Asian Earthquakes

D. P. Eriksson

December 2006

The work contained within this document has been submitted by the student in partial fulfilment of the requirement of their course and award

ABSTRACT When a sudden-onset natural disaster strikes a developing country, the state of communications and infrastructure in remote areas may be fragile, delaying the start of any regional or international intervention. A delay of even a couple of days (Alexander 2000a:46; Alexander 2002:198; Shakhramanian et al 2000:148) means that certain forms of emergency relief, such as Search And Rescue (SAR) operations in collapsed structures, are no longer beneficial. To improve international relief to disasters in these situations, this study aims to identify steps in the decision process leading up to an international intervention that could benefit from the application of a Decision Support System (DSS). First, user requirements on a DSS are identified through interviews, observation and content-analysis of many different organisations’ internal guidelines. Following this, the DSS options that fulfil the requirements are identified. Fifty-nine earthquake events in central Asia which occurred between 1992 and 2003 are adopted as case studies for this purpose. For each case study, quantitative data on loss, needs and international response have been collected using content- and frequency-analysis of the documentation produced by stakeholders in the international response. The case study data are used to determine which data sources are of benefit to decision makers using each data source’s time of availability and content. Considering the options provided by the identified data sources, a prototype DSS is developed. The prototype builds on the existing Global Disaster Alert and Coordination System (GDACS) to provide a novel type of decision support to potential responders who are located outside the affected country. The intention is to notify decision makers of the occurrence of events that fit the profile of events they have responded to in the past. This could speed up their intelligence-gathering and ultimately provide a faster international response. Using the historical events, ordinal logistic regression is applied to develop a numerical model that produces a projection of the international attention in future events. The study applies the frequency of United Nations Situation Reports as the quantitative indicator of the international attention to past events. The prototype output for a future earthquake is the likelihood of it falling into one of the following categories: (1) marginal international attention; (2) intermediate international attention; or (3) substantial international attention. The accuracy of the prototype proved too low for direct use by practitioners. However, after the development of the prototype, ways to improve the accuracy and to make the prototype applicable to other hazards and geographical regions are suggested.

Keywords: disaster management, decision support systems, humanitarian aid, development assistance, earthquake preparedness, early warning systems, numerical modelling

- ii -

SUPERVISORY TEAM Dr. Graham Marsh Senior Lecturer Centre for Disaster Management Coventry University Prof. Hazel Barrett Head of Department Department of Geography, Environment and Disaster Management Coventry University Prof. Dr. David Alexander Professor of Disaster Management Università degli Studi di Firenze Dr. Tom De Groeve Scientific administrator DG Joint Research Centre European Commission

TECHNICAL ADVISORS Dr. Peter Billing Former Head of Sector for Strategic Planning European Commission Humanitarian Office Mr. Per-Anders Berthlin Senior advisor on overseas operations Swedish Rescue Services Agency

- iii -

ACKNOWLEDGEMENTS I am profoundly grateful to a long list of individuals without whose input and

support this project would never have been started or completed. Grouped in order of their appearance in the life of the research project these persons are: for encouraging me to seek to obtain a research degree, John Flanagan, Matz Wennerström, Benny Ljus, Dr. Aldo Benini and Dr. Dirk Salomons; for giving me the opportunity to do so, Dr. Iain Shepherd; for on-site supervision in Italy, Dr. Delilah Al-Khudhairy; for academic guidance in my first years of research, Prof. Erland Jungert and Prof. Åke Sivertun; for their friendly advice on the perils of PhD research, Dr. Jed Kaplan and Ana-Lisa Vetere; for supporting the field trip to Africa, Christopher Clark, Chuck Conley and Joseph Donahue; for excellent supervisory support in spite of repeated setbacks beyond our control, Prof. David Alexander; for volunteering his time for supervisory support and frequent reviews, Dr. Tom De Groeve; for contact with the ‘real world’, Per-Anders Berthlin; for important material and interviews, Dr. Peter Billing; for leading me into goal in my final year of research Dr. Graham Marsh; for general advice on survival in a British research establishment, Dr. Eleanor Parker; for volunteering to provide pivotal advice on the use of the statistical methods, Prof. Collin Reeves; and, for her comprehensive reviews and proofreading, Prof. Hazel Barrett.

The administrative staff members at Linköping University, the European Commission Joint Research Centre, Cranfield University and Coventry University deserve special thanks for their patience in guiding me through the administrative hoops of multiple transfers and the ground-breaking challenges that I posed them with. This includes Laura Occhetta, Michelle Addison, Ann Daly and Daxa Kachhala.

I have not forgotten the numerous friends that I made throughout the course of this project in Italy, Sweden, Sudan, Spain and the UK. Your continuous support has kept my mind off work and off the prospect of quitting. My friends at the JRC institute for the protection and security of the citizen deserve a special mentioning in this regard: Clementine Burnley, Dominik Brunner, Dirk Buda, Ivano Caravaggi, Dr. Daniele Ehrlich, Martin Jacobson, Sarah Mubareka, Stefan Schneiderbauer, Kenneth Mulligan, Raphaele Magoni, Federica Bocci, Luigi Zanchetta, Jolyon Chesworth, Elena Aresu, Tony Bauna and Dr. Herman Greidanus.

Most importantly, I want to mention my wife for her general support, including endless proofreading and for having endured the life of uncertainty that accompanied this project.

This thesis is dedicated to my friends and colleagues who were injured or killed in the 19th August 2003 Canal Hotel bombing in Baghdad, Iraq.

DANIEL P. ERIKSSON

- iv -

TABLE OF CONTENTS ACKNOWLEDGEMENTS III LIST OF FIGURES VII LIST OF TABLES VIII LIST OF PLATES IX LIST OF ABBREVIATIONS X GLOSSARY XII

1 INTRODUCTION 1

1.1 AIM, QUESTIONS AND OBJECTIVES 3 1.2 DEFINITIONS 3 1.3 BACKGROUND 4 1.4 DOCUMENT STRUCTURE 5

2 INTERNATIONAL RESPONSE TO DISASTERS 7

2.1 DISASTER MANAGEMENT CYCLE 7 2.2 HAZARD, VULNERABILITY AND RISK 9 2.3 INTERNATIONAL DISASTER RELIEF 13 2.4 INITIAL ASSESSMENT OF LOSS AND NEEDS 19 2.5 SUMMARY 21

3 SUPPORTING DECISIONS WITH INFORMATION SYSTEMS 23

3.1 TYPOLOGY 23 3.2 DECISION SUPPORT 25 3.3 USABILITY DESIGN 26 3.4 SUMMARY 27

4 DECISION SUPPORT IN DISASTER RESPONSE 28

4.1 TELE-ASSESSMENT 28 4.1.1 EARLY WARNING 29 4.1.2 LOSS ASSESSMENT 30 4.1.3 NEEDS ASSESSMENT 34 4.1.4 DATA QUALITY 35 4.1.5 USABILITY 36 4.2 EXISTING DECISION SUPPORT SYSTEMS 37 4.2.1 PLANNING AND SCENARIO BUILDING 37 4.2.2 REAL-TIME ALERTS 39 4.2.3 CO-ORDINATION 46 4.2.4 TRENDS 47 4.3 SUMMARY 47

5 RESEARCH PLAN 48

5.1 RESEARCH APPROACH 48 5.1.1 PHILOSOPHY 48

5.1.2 RESEARCH DESIGN 49 5.1.3 METHODS AND SAMPLING 54 5.1.4 COLLABORATIONS AND EXTERNAL INFLUENCES 60 5.1.5 RESEARCH SIGNIFICANCE AND RELEVANCE 62 5.1.6 ETHICAL CONSIDERATIONS 63 5.1.7 ASSUMPTIONS 64 5.1.8 LIMITATIONS 65 5.2 DATA 67 5.2.1 DATA OVERVIEW 67 5.2.2 DATA TYPES 68 5.2.3 DATABASE AND USER INTERFACE 73 5.2.4 QUANTITATIVE DATA SOURCES 74 5.2.5 DATA CLEANING 78 5.2.6 ANALYTICAL DATA CLASSIFICATION 79 5.3 ANALYTICAL METHODS 86 5.3.1 QUALITATIVE DATA ANALYSIS 86 5.3.2 QUANTITATIVE DATA ANALYSIS 86 5.4 METHODOLOGICAL SUMMARY 93

6 EARTHQUAKE: A SUDDEN-ONSET HAZARD 95

6.1 HAZARD ONSET AND COMPLEXITY 95 6.2 MEASURING EARTHQUAKES 95 6.3 MODELLING 100 6.4 IMPACT EFFECTS 101 6.5 EARTHQUAKE ENGINEERING 102 6.6 SUMMARY 102

7 CENTRAL ASIAN REGION 103

7.1 REGION 103 7.1.1 EARTHQUAKE HAZARD 103 7.1.2 VULNERABILITY 107 7.2 NATIONS 108 7.3 SAMPLE EARTHQUAKE EVENTS 113 7.3.1 1997 BOJNOORD, IRAN EARTHQUAKE 113 7.3.2 2002 DAHKLI, AFGHANISTAN/TAJIKISTAN EARTHQUAKE 115 7.4 SUMMARY 117

8 SYSTEMS INVESTIGATION 119

8.1 IMPLEMENTING ORGANISATION 119 8.2 CO-ORDINATING ORGANISATION 126 8.3 FUNDING ORGANISATION 129 8.4 SYSTEMS INVESTIGATION SUMMARY 132

9 SYSTEMS ANALYSIS 133

9.1 ANALYSIS OF ALTERNATIVES 133 9.1.1 A SOURCE EVALUATION FRAMEWORK 134 9.2 DISCUSSION 137 9.2.1 REMOTELY SENSED SEISMIC DATA 138

- vi -

9.2.2 REMOTELY SENSED IMAGERY 138 9.2.3 NUMERICAL MODELS 140 9.3 SYSTEMS ANALYSIS SUMMARY 143

10 SYSTEMS DESIGN AND IMPLEMENTATION 148

10.1 PROBLEM DEFINITION 148 10.2 DATA SELECTION 151 10.3 DATA STANDARDISATION 156 10.3.1 DV CATEGORISATION 157 10.3.2 IV CATEGORISATION 159 10.4 DATA MINING 165 10.4.1 MULTI-VARIABLE ANALYSIS INPUT SELECTION 165 10.4.2 VARIABLE IMPORTANCE ANALYSIS 167 10.4.3 MAIN EFFECTS ANALYSIS 169 10.4.4 MODEL VARIABLE INTERACTION 169 10.5 EVALUATION AND VALIDATION FRAMEWORK 172 10.6 SYSTEMS DESIGN AND IMPLEMENTATION SUMMARY 173

11 EVALUATION 175

11.1 OBJECTIVE 1: USER REQUIREMENTS AND SYSTEM RELEVANCE 175 11.1.1 RELEVANCE OF INTERNATIONAL ALERT SYSTEMS 175 11.1.2 TIMELINESS, ACCURACY AND COMPLETENESS 177 11.1.3 THE SHORTCOMINGS OF EXISTING SYSTEMS 179 11.2 OBJECTIVE 2: QUANTIFYING THE INTERNATIONAL ACTIONS 181 11.2.1 CHALLENGING THE QUANTIFICATIONS AND CATEGORISATIONS 181 11.2.2 PATTERNS IN INTERNATIONAL ACTIONS 183 11.3 OBJECTIVE 3: A PROTOTYPE MODEL 184 11.3.1 UNDER-PREDICTION 185 11.3.2 OVER-PREDICTION 186 11.3.3 WEAKNESSES 188

12 CONCLUSION 190

12.1 AIM AND OBJECTIVES 190 12.1.1 LESSONS LEARNT 191 12.2 FUTURE RESEARCH 192 12.2.1 POTENTIAL MODEL IMPROVEMENTS 193 12.2.2 DATABASE USE FOR OTHER APPLICATIONS 196

13 REFERENCES 199

INDEX 210

14 APPENDICES 211

A-1 CASE STUDY DESCRIPTIVES 211 A-2 MODEL DEVELOPMENT 213 A-3 EXPLORATORY ANALYSIS 214 A-4 INTEREST DATABASE 225

- vii -

LIST OF FIGURES FIGURE 1.1 THESIS CONCEPTUAL OUTLINE.........................................................................................6 FIGURE 2.1 THE DISASTER MANAGEMENT CYCLE................................................................................7 FIGURE 2.2 PRESSURE AND RELEASE MODEL (PAR) .......................................................................10 FIGURE 3.1 DECISION STRUCTURE ACCORDING TO HIERARCHICAL LEVELS ...........................................24 FIGURE 5.1 THE ‘KNOWLEDGE DISCOVERY IN DATABASES’ PROCESS.................................................50 FIGURE 5.2 APPLIED RESEARCH PROCESS MODELS IN RELATION TO THE THESIS OBJECTIVES.................54 FIGURE 5.3 CONCEPTUAL MODEL OF VULNERABILITY DATA ................................................................69 FIGURE 5.4 DISASTER EFFECT CLASSIFICATION ................................................................................70 FIGURE 5.5 ADAPTED MANIFEST CODING.........................................................................................81 FIGURE 5.6 EXCERPT FROM THE RELIEF DATA CLASSIFICATION...........................................................82 FIGURE 5.7 ENVELOPE OF THE SUM OF DEAD AND INJURED IN THE 2002 QUAZVIN, IRAN, EARTHQUAKE 90 FIGURE 6.1 EARTHQUAKE PARAMETERS..........................................................................................96 FIGURE 6.2 ATTENUATION CURVES .................................................................................................98 FIGURE 8.1 SRSA RESPONSE PROCESS....................................................................................... 121 FIGURE 9.1 AVERAGE NUMBER OF DEAD AND INJURED PER ALERT LEVEL ......................................... 141 FIGURE 10.1 CONCEPTUALISATION OF PROPOSED PROGNOSTIC MODEL........................................... 149 FIGURE 10.2 SCATTER-PLOT MATRIX OF OCHA SITREPS, FINANCIAL AID AND HUMAN LOSS............... 150 FIGURE 10.3 SITUATION REPORTS, HUMAN LOSS AND FINANCIAL AID (N=53) .................................. 157 FIGURE 10.4 DISTRIBUTION OF 50KM RADIUS POPULATION IN THE CASE STUDIES ............................ 163 FIGURE 10.5 DISTRIBUTION OF CASES OVER ‘NIGHT’..................................................................... 167 FIGURE 10.6 DISTRIBUTION OF CASES OVER ‘EXPOSED’ ................................................................ 167 FIGURE 10.7 CONCEPTUAL FINAL MODEL..................................................................................... 173 FIGURE 12.1 THE BOWA MODEL................................................................................................ 197 FIGURE 14.1 RELIEF REQUESTS.................................................................................................. 215 FIGURE 14.2 RELIEF REQUEST DISTRIBUTION BY WEALTH .............................................................. 215 FIGURE 14.3 DONATION DESTINATION PER ORIGIN CATEGORY ........................................................ 215 FIGURE 14.4 DONATION ORIGIN PER RECIPIENT............................................................................ 217 FIGURE 14.5 DONATION TYPE DISTRIBUTION PER ORIGIN CATEGORY ............................................... 217 FIGURE 14.6 DONATIONS........................................................................................................... 218 FIGURE 14.7 TIER 2 SHELTER DONATIONS ................................................................................... 218 FIGURE 14.8 INJURY REPORTING ACCURACY................................................................................. 220 FIGURE 14.9 AVERAGE TIME UNTIL FIRST REPORT RELEASE............................................................ 220 FIGURE 14.10 MEDIA PERSEVERANCE PER EVENTS...................................................................... 221 FIGURE 14.11 CORRELATION MATRIX OF MEDIA EXPOSURE............................................................ 221 FIGURE 14.12 MEDIA REPORTING DELAY AND RESPONSE DELAY .................................................... 224 FIGURE 14.13 EARTHQUAKE (SEISMIC) REPORT VIEW ................................................................... 225 FIGURE 14.14 MAIN MENU........................................................................................................ 225 FIGURE 14.15 ADMINISTRATION MENU ....................................................................................... 226 FIGURE 14.16 EVENT POPULATION DISTRIBUTION VIEW ................................................................. 226 FIGURE 14.17 DATABASE EVENT VIEW......................................................................................... 227 FIGURE 14.18 DATA MINING VIEW .............................................................................................. 228 FIGURE 14.19 DATABASE ENTITY-RELATIONSHIP DIAGRAM............................................................ 229

- viii -

LIST OF TABLES TABLE 2.1 EARTHQUAKE-SPECIFIC SOCIAL LEVEL VULNERABILITY INDICATORS......................................12 TABLE 3.1 THE ROLE OF THE INFORMATION SYSTEM PER HIERARCHICAL LEVEL ....................................25 TABLE 4.1 ALERT LEVELS, SCORES AND SEVERITY ............................................................................40 TABLE 4.2 GDACS SUB-FUNCTIONS ...............................................................................................41 TABLE 4.3 QUAKELOSS ALERT PROCESS FOR THE 8TH OCTOBER 2005 EARTHQUAKE IN PAKISTAN.....43 TABLE 4.4 GLOBAL COVERAGE EARTHQUAKE ALERT SYSTEMS............................................................44 TABLE 5.1 EARTHQUAKES STUDIED BY YEAR AND COUNTRY ...............................................................58 TABLE 5.2 CLASSIFICATION OF QUALITATIVE/QUANTITATIVE VERSUS SUBJECTIVE/OBJECTIVE................67 TABLE 5.3 NUMBER OF REPORTS AND ATTRIBUTES PER EVENT ACCORDING TO SOURCE........................74 TABLE 5.4 THE TOP-LEVEL MANIFEST CODES ...................................................................................79 TABLE 5.5 THE RELIEF DATA TAXONOMY .........................................................................................82 TABLE 5.6 NUMERICAL METADATA CATEGORIES ...............................................................................84 TABLE 5.7 PROJECT METHODOLOGICAL OVERVIEW ...........................................................................94 TABLE 6.1 EARTHQUAKE MAGNITUDE MEASUREMENTS .................................................................. 100 TABLE 7.1 COMPARISON OF THE CASE STUDY COUNTRIES .............................................................. 107 TABLE 7.2 BOJNOORD, IRAN, INITIAL DATA.................................................................................... 113 TABLE 7.3 REPORTED IMPACT OVER TIME.................................................................................... 114 TABLE 7.4 REPORTED NEEDS OVER TIME ..................................................................................... 114 TABLE 7.5 REPORTED DISPATCHED RELIEF OVER TIME................................................................... 115 TABLE 7.6 DAHKLI, AFGHANISTAN/TAJIKISTAN, INITIAL DATA.......................................................... 116 TABLE 7.7 REPORTED IMPACT OVER TIME..................................................................................... 116 TABLE 8.1 ROLES IN THE SRSA DECISION PROCESS...................................................................... 120 TABLE 8.2 SRSA INTERVENTION TIMELINE ................................................................................... 124 TABLE 9.1 THE DECISION SEQUENCE IN INTERNATIONAL DISASTER RELIEF........................................ 133 TABLE 9.2 DEFINITION OF APPLIED TERMINOLOGY FOR DATA QUALITY .............................................. 135 TABLE 9.3 DATA AVAILABILITY AND QUALITY OVER TIME.................................................................. 137 TABLE 9.4 PROS AND CONS OF REMOTE SENSING ALTERNATIVES .................................................... 139 TABLE 10.1 CLASSIFICATION OF INDICATORS, ACCORDING TO PURPOSE........................................... 151 TABLE 10.2 SELECTED IVS......................................................................................................... 156 TABLE 10.3 INDICATOR CATEGORISATION ..................................................................................... 158 TABLE 10.4 SUMMARY OF CASE STUDIES PER DV CATEGORIES....................................................... 159 TABLE 10.5 EARTHQUAKE EXPOSURE CATEGORISATION ................................................................. 160 TABLE 10.6 URBAN GROWTH CATEGORISATION............................................................................. 160 TABLE 10.7 OPENNESS CATEGORISATION .................................................................................... 161 TABLE 10.8 VULNERABILITY CATEGORISATION .............................................................................. 161 TABLE 10.9 DATA MINING START VARIABLES ................................................................................ 165 TABLE 10.10 DISTRIBUTION OF EARTHQUAKES OVER NIGHT AND DAY.............................................. 166 TABLE 10.11 FULL MODEL PARAMETER ESTIMATES (CAUCHIT)....................................................... 170 TABLE 10.12 FULL MODEL ORDINAL PREDICTIONS (CAUCHIT) ........................................................ 171 TABLE 10.13 CLASSIFICATION ERRORS........................................................................................ 172 TABLE 14.1 CASE STUDIES AND THE AMOUNT OF LINKED DATA (TWO PAGES) ................................... 211 TABLE 14.2 EXAMPLE USGS LONG EARTHQUAKE NOTIFICATION MESSAGE ...................................... 212 TABLE 14.3 STARTING MODEL PARAMETERS (CAUCHIT)................................................................. 213 TABLE 14.4 FULL MODEL PARAMETER ESTIMATES (CAUCHIT) ......................................................... 213 TABLE 14.5 MEDIA PERSEVERANCE CATEGORIES .......................................................................... 223

- ix -

LIST OF PLATES PLATE 4.1 PAGER GRAPHICAL OUTPUT FOR THE 24TH FEBRUARY 2004 EARTHQUAKE IN MOROCCO.....45 PLATE 4.2 USGS PAGER NUMERICAL OUTPUT FOR THE 24TH FEBRUARY 2004 EARTHQUAKE IN

MOROCCO............................................................................................................................45 PLATE 4.3 QUAKELOSS GRAPHICAL OUTPUT FOR THE 8TH OCTOBER 2005 EARTHQUAKE IN PAKISTAN 45 PLATE 5.1 PROJECTED 50-YEAR MAXIMUM EARTHQUAKE INTENSITY IN CENTRAL ASIA..........................57 PLATE 5.2 WORLDWIDE EARTHQUAKE DISASTER RISK HOTSPOTS.......................................................58 PLATE 5.3 LANDSCAN 2004 RASTER OF GLOBAL POPULATION DISTRIBUTION......................................77 PLATE 5.4 POPULATION DENSITY MAP FOR THE SECOND RUSTAQ EVENT .............................................92 PLATE 7.1 MAP OF CASE STUDY EARTHQUAKE EPICENTRES ............................................................ 105 PLATE 7.2 1997, BOJNOORD, IRAN EARTHQUAKE ........................................................................ 105 PLATE 7.3 2002 DAHKLI, AFGHANISTAN/TAJIKISTAN ................................................................... 106 PLATE 9.1 VIRTUAL OSOCC SCREENSHOT FROM THE OCTOBER 2005 RESPONSE TO THE

PAKISTAN/INDIA EARTHQUAKE............................................................................................. 146 PLATE 9.2 GDACS EMAIL ALERT FOR AN APRIL 2006 EARTHQUAKE IN DR CONGO.......................... 147

LIST OF ABBREVIATIONS Abbreviation 1 Description AFP Agence France-Presse AP Associated Press AVgas Aviation fuel CAP Consolidated Appeal Process CATS Consequence Assessment Tool-Set CRED Centre for Research on the Epidemiology of Disasters DHA United Nations Department of Humanitarian Affairs, (now OCHA) DMA (JRC) Digital Map Archive DSS Decision Support System DV Dependent Variable EC European Commission ECHO European Commission Humanitarian Office EERI Earthquake Engineering Research Institute EM-DAT (CRED) Emergency events Database EMM (JRC) Europe Media Monitoring tool ESB (OCHA) Emergency Services Branch ESRC Extreme Situations Research Centre (Russia) EUSC European Union Satellite Centre EWS Early Warning System FEMA (US) Federal Emergency Management Agency FCSS (ESB) Field Co-ordination Support Section GDACS Global Disaster Alert and Coordination System GDP Gross Domestic Product GIS Geographical Information System GLIDE Global Identifier number GMT Greenwich Mean Time GNA (ECHO) Global Needs Assessment index GPS Global Positioning System HAZUS (MH) (FEMA) Hazards United States – Multi-Hazard version HDI (UNDP) Human Development Index HPI Human Poverty Index IASC (UN) Inter-Agency Standing Committee ICDO International Civil Defence Organisation IDNDR International Decade for Natural Disaster Reduction IFRC International Federation for the Red Cross and Red Crescent societies IHP International Humanitarian Partnership INGO International Non Governmental Organisation INSARAG (UN) International Search And Rescue Advisory Group IS Information Systems ISDR International Strategy for Disaster Reduction IJ (SRSA) international duty officer INTEREST Database for International Earthquakes Loss, Needs & Relief Estimation IV Independent Variable JRC (European Commission Directorate General) Joint Research Centre KDD Knowledge Discovery in Databases Mb Body-wave magnitude MIS Management Information System ML Local magnitude, i.e. Richter magnitude MMI Modified Mercalli Index Ms Surface wave magnitude Mw Moment magnitude NEIC (USGS) National Earthquake Information Centre NGO Non Governmental Organisation OCHA (UN) Office for the Coordination of Humanitarian Affairs OLAP Online Analytical Processing PAGER (USGS) Prompt Assessment of Global Earthquakes for Response PGA Peak Ground Acceleration POET Psychopathology Of Everyday Things

1 For the abbreviations of the statistical variables see Table 10.2 and Table 10.3.

- xi -

Abbreviation 1 Description RADIUS (IDNDR) Risk Assessment Tool for Diagnosis of Urban Areas against Seismic Disasters RC Reinforced Concrete RRM (ECHO) Rapid Reaction Mechanism RWB Reporters Without Borders SAR Search and rescue SIDA Swedish international development cooperation office Sitrep (OCHA) Situation report SMS Short Messaging Service SPSS Statistical Package for Social Sciences SRSA Swedish Rescue Services Agency UN United Nations UNDAC United Nations Disaster Assessment and Coordination UNDP United Nations Development Programme UNICEF United Nations United Nations Children’s (Emergency) Fund UPI United Press International USGS United States Geological Survey VOSOCC (OCHA) Virtual On-Site Operations Coordination Centre VT SRSA duty officer WAPMERR World Agency for Planetary Monitoring and Earthquake Risk Reduction WPFI (RWB) World Press Freedom Index

- xii -

GLOSSARY Cell: In statistical modelling, a group of events defined by the same combination

of all the model characteristics.

Co-linearity: A dependency between two predictor (IV) variables. (Hosmer and Lemeshow 2000:140-141)

Contingency cleaning: The cross-classifying of two variables looking for impossible combinations, e.g. small magnitude earthquakes with great human losses (Neuman 2000:316-317)

Data mining: The use of a data warehouse to identify key factors, patterns and trends in historical data. (O’Brien 1999:274)

Entity-relationship: The concept used in relational databases. Such databases are mapped using entity-relationship diagrams. (O’Brien 1999)

Entry decision: Jargon used by European Commission Humanitarian Office (ECHO) for the decision to engage in a crisis (Billing 2004)

Hypocentre: Also known as the focus, the hypocentre is the point in three dimensions where a seismic fault starts its rupture. (Bolt 2004:354)

Image pair: A set of images of the same area, one taken before an event and one taken after an event of interest. (Al-Khudhairy and Giada 2002)

Informatics: Information science. The collection, classification, storage, retrieval and dissemination of recorded knowledge treated both as a pure and as an applied science (Merriam Webster Collegiate Dictionary, 11th Edition)

Intensity raster: A shake-map showing the spatial distribution of the intensity of the shaking often provided in peak ground acceleration.

Link-function: Also know as the link-model, the link-function converts the categorical variables and model output to a scale from zero to one. (Hosmer and Lemeshow 2000:48)

OLAP: Online analytical processing is the capability of a decision support tool to support interactive examination and manipulation of large amounts of real-time data from many perspectives. (O’Brien 1999)

Ordinal regression: A type of logistic regression in which the DV is expected to be in ordered categories. (Tabachnick and Fidell 2001:542)

Outlier: In statistics, an outlier is a single observation remote from the rest of the data. This can be due to systematic error or faults in the theory that generated the expected values. Outlier points can therefore indicate faulty data, erroneous procedures, or areas where a certain theory might not be valid. (Tabachnick and Fidell 2001)

Pearson residual: An indicator of goodness-of-fit that can be used on a summary level as well as for individual model predictions.

Pseudo-r2: A rough indicator of a model’s fit. In linear regression, the r2 statistic is the proportion of the total variation in the response that is explained by the model (Hosmer and Lemeshow 2000:165). The pseudo-r2 is an attempt to create an equivalent measure for logistic regression

Raster data: Image analysis of often conducted using raster data structures in which the image is treated as an array, or matrix, of values. Each coordinate in the matrix is defined as a pixel or point. For further information see Campbell (2002:102).

Real-time process: Also referred to as an ‘Online process’. This is a process in which data is processed immediately after a transaction occurs. The term ‘Real-time’ pertains to the performance of data processing during the actual time a physical process transpires so that the result of the processing can be used to support the completion of the process. (O’Brien 1999:57)

- xiii -

Relational database: A structure of information elements within a database where information is stored in simple tables. Other tables represent the relations between simple tables. An example would be a table with information on a department being related to a table containing all its employees. (O’Brien 1999:280)

Remote sensing: The harshest definitions of remote sensing see it as the science of telling something about an object without touching it. A narrower definition is that the concept includes all methods of obtaining pictures or other forms of electromagnetic records of the Earth’s surface from a distance and the treatment and processing of that data. (Campbell 2002:6)

Revisit time: In remote sensing the time required by a sensor platform, like an earth orbiting satellite, to return to a specific area. (Campbell 2002:6)

1 INTRODUCTION It is widely accepted that pre-emptive measures in disaster prone regions, such

as causally oriented institutional support for mitigation and preparedness efforts, is

arguably a more cost efficient form of aid compared to traditional palliative post-

disaster relief (Walker 1991; Smillie et al 2003:25). Nevertheless, as shown by Olsen et

al (2003) the media attention given to sudden-onset disasters and the political

incentive to respond to them will continue to create a popular interest and moral

reasons in donor nations to provide immediate help to those suffering (Albala-

Bertrand 1993).

When a natural disaster strikes in a developing country, the undeveloped state

of local information infrastructure in remote areas may delay the start of any

international or regional intervention (Zimmerman 2002). The delay can reach a

point, usually within the first couple of days (Alexander 2000a:46; Alexander 2002:198;

Shakhramanian et al 2000:148), after which certain forms of emergency relief, such as

Search And Rescue (SAR), are no longer beneficial. Walker (1991) questions whether

it ever will be possible for expatriate rescuers to arrive in time. Where SAR is a valid

relief alternative, the number of people saved drops dramatically after only 6-8 hours.

Examples of this dilemma are the Bam earthquake in Iran 2003 in which 1,200

expatriate SAR experts saved 30 people (Mohavedi 2005) and the Armenia earthquake

in 1988 in which 1,800 expatriate SAR experts saved 60 people (UNDRO 1989).

Consequently, if time-sensitive relief is to be dispatched to a far-away location,

the decision to do so has to be taken within hours after the disaster for the relief to

make an impact (Walker 1991). If there is no direct communication to a source with

precise and reliable information on the disaster situation, decision makers will have to

resort to using information from subjective sources, such as the media and local

contacts, for developing an informal needs assessment. Accepting that international

relief will continue in one form or another, the intention of this study is to improve the

support available to decision makers in international relief organisations responding

to disasters. The study does so in a two-pronged approach. First, it investigates how

existing channels of information could be used to provide optimal support to the

decision making process in the responding international organisations, from the

beginning to the end of the emergency phase following a sudden-onset disaster. Then

it identifies a suitable step in the decision process leading up to an international

intervention and develops a prototype Decision Support Systems (DSS) for that step.

Decision makers and practitioners in international relief organisations are formally

and informally interviewed to develop an understanding of how their work can be

supported by DSS. To reduce the complexity of the data collection and analysis,

earthquakes are selected as an archetype of sudden-onset disasters. Fifty-nine

earthquake events in central Asia between 1992 and 2003 have been studied for the

development of the prototype DSS. For each case study, quantitative time-series data

on loss, need and international response is collected using content and frequency

analysis of international organisation documentation such as situation reports and

inter-agency co-ordination reports. Although the initial intent was to collect data for

all earthquakes in developing countries, the high level of detail of the data required to

be collected restricted the research to a case study region – central Asia. The central

Asian region was selected for its relative high earthquake risk. In the development of

a prototype DSS, the study applies the frequency of United Nations Office for the

Coordination of Humanitarian Affairs (OCHA) Situation Reports (sitreps) as a

quantitative indicator of the international attention given to an event. By adopting the

case studies as a reference set, ordinal regression is used to develop a model that

predicts the international attention. This prognostic model predicts the likelihood of a

subsequent international intervention falling into one of three categories of

international attention: marginal international attention, intermediately sized

international attention, or substantial international attention. The purpose of the

model is to probe the feasibility of developing models that circumvent the current

paradigm in DSS for international relief - loss estimation - and increasing the

relevance of the resulting alerts to the international relief community. When the

research project started, several loss estimation tools for global use were in

development. At the time, the tools had still not achieved functionality to operate

without human oversight as the earthquakes took place. This functionality has,

however, become more common in the last couple of years. With the emergence of

these loss-based tools, it is important to reach beyond them to identify potential future

solutions in the use of DSS for international relief to sudden-onset disasters. This

research project represents a probe into one of these future solutions.

1.1 Aim, Questions and Objectives The aim of this research is to improve international relief to sudden-onset

disasters by identifying novel ways of supporting the decision process surrounding it.

The pre-empting research question is to explore under what circumstances

decision support would be beneficial to the international relief effort and whether

existing systems are adequate. With the existing gap and relevance of the decision

support determined, the subsequent question is how to develop a decision support

system (DSS) fulfilling the identified requirements.

The research objectives are:

1. To establish a set of user requirements, including thresholds for timeliness,

accuracy and notification content; and to determine the relevance of a DSS

for use in the initial phase of international relief to sudden-onset disasters.

2. To collect, to structure and analyse the data required to develop a DSS

fulfilling the identified requirements.

3. To develop and evaluate a prototype DSS.

1.2 Definitions These definitions will be elaborated further in the document, but to introduce

the reader to the approach of the project, they are presented in brief here. First, the

decision to engage in a crisis is termed by European Commission Humanitarian Office

(ECHO) as the “entry decision” (Billing 2004). This term was adopted both because

ECHO activities were central to the research and because ECHO terminology is

widely used among implementing organisations partly on account of ECHO’s

position as one of the world’s largest donors. An event is defined as a strike of a

hazard. An event becomes a disaster when the resulting loss generates a need for relief

that exceeds the national resilience, which leads to a requirement for international relief.

Loss and impact are used interchangeably to refer to the total damage that a hazard

causes on an affected society as a result of a disaster, e.g. the loss of life, structures, or

financial means. Need is defined as the quantitative requirement of assistance.

International need is consequently the need that cannot be covered by local, national,

or regional assets. The applied definition of national resilience is that of the Journal of

Prehospital and Disaster Medicine:

Pliability, flexibility, or elasticity to absorb the event. […] As resiliency increases, so does the absorbing capacity of the society and/or the environment. Resiliency is the inverse of vulnerability (in Thywissen 2006:23)

This definition is adopted because it puts resilience in contrast with

vulnerability and thereby facilitates quantitative analysis of the two characteristics. In

addition to being the inverse of resilience, vulnerability is defined in line with

International Federation of the Red Cross (IFRC) (1999) and Wisner et al (2004:11) as

being:

The characteristics of a person or group in terms of their capacity to anticipate, cope with, resist and recover from the impact of a natural or man-made hazards

Vulnerability is also accepted as a spatially and socially dependent characteristic

in accordance with the use of Schneiderbauer and Ehrlich (2005). In their 2005 study

Schneiderbauer and Ehrlich analyse vulnerability on a set of social levels reaching

from individual to a cultural community.

International attention is defined as the size of donated relief and media coverage

provided to a disaster by the international community. Although there will be

attempts in this thesis to quantify this attribute, it is inherently qualitative.

Finally, the main categories of considered actors and potential users are defined

as being part of either: implementing organisations, co-ordinating organisations, or funding

organisations. Funding organisations provide funding for implementation and co-

ordination of a relief mission. Implementing organisations do the field work on-site,

e.g. food distribution or medical support. The co-ordinating organisation can either

facilitate information exchange or actively guide the efforts of the implementing

organisations through, for instance, the provision of advice to the funding

organisations.

1.3 Background When this research project started there were no operational tools providing

predictions of the consequences of sudden-onset disaster as they happened. Several

tools have, however, become operational over the last couple of years. These tools are

centred on the prediction of human losses. The uncertainty of the data available in the

immediate aftermath of a sudden-onset disaster gives the prediction of the human

losses a wide spread. This complex output can reduce its relevance to the users for

which it is intended. Some tools are not automated but use human experts to analyse

incoming data following disasters. This allows for a more accurate alert, but it delays

the delivery of the alert and it is also costly.

This project will investigate how the international decision makers in the

immediate aftermath of sudden-onset disasters can best be supported. Which

decision requires support and how should it be supported? The intention is to

introduce a novel way of looking at alerting by distancing the research from the

current paradigm of human loss prediction. This project will attempt to predict which

events will receive international assistance rather than which events that will result in

high death-tolls.

A more accurate alert system has the potential to improve the international

relief, both in terms of speed and content. An earlier alert would allow for more time

to collect additional information from on-site representatives and other time-

consuming channels. More information, if accurate and relevant, leads to a better

informed entry decision and better use of resources.

The European Commission provided a grant to this research project with

interest to improve its financial responses to sudden-onset disasters. At the start of

the research, the European Commission had initiated the development of a prototype

alert system for this purpose, based on loss assessment models. Before the completion

of that prototype, the alerting was made through a duty officer who was tasked with

watching the news and determining, based on the media coverage, whether to fund

relief missions in the area. The purpose of this research project was to enhance the

support provided by the prototype tool in development. The researcher had at the

time just completed a two-year project in Kosovo involving the development of

decision support systems for humanitarian de-mining operations and there was a

potential of synergy between the past project and that suggested by the European

Commission.

1.4 Document structure Figure 1.1 presents a conceptual outline of the thesis structure. The Introduction

leads into the two separate chapters: ‘2. International Response to Disasters’ and ‘3.

Supporting Decisions with Information Systems’. These two chapters introduce the

reader to the current theory in international relief, particularly following sudden-

onset disasters, and in the use of information systems for decision support. After this

briefing, the specific domain of ‘Decision support in Disaster Response’ is presented

in Chapter 4. Here, the theories from the preceding two chapters are combined and

state-of-the-art decision support tools in disaster management are reviewed.

Source: Author

Figure 1.1 Thesis conceptual outline

At this stage the thesis has provided a foundation necessary for the presentation

of Chapter 5, the ‘Research Plan’. In the research plan chapter a structure is prepared

for the development of the prototype model. As part of the chapter, the central Asian

region is selected for a case study and earthquakes are chosen as archetypes of

sudden-onset hazards in general. Consequently, the theory of earthquakes is

presented in the Chapter 6, ‘Earthquake: A Sudden-onset hazard‘ and the case study

area is presented in the Chapter 7, the ‘Central Asian Region’. In the Research Plan,

the two main adopted substructures are selected and described: the Information

Systems (IS) development cycle and the Knowledge Discovery in Databases (KDD)

process. The IS development cycle is a cyclical structure containing four stages:

Systems Investigation, Systems Analysis, Systems Design and Systems

Implementation. Although much iteration of these elements was made they are laid

out sequentially in the thesis. The KDD process was applied as part of the Systems

Implementation stage.

The IS development cycle is exited to the Prototype evaluation in Chapter 11.

The prototype is evaluated as part of the IS development cycle, but the evaluation is

deepened here and the results are linked to the thesis aim and objectives. Based on

the shortcomings of the model and the lesson learnt in its development, the potential

direction of future research is presented as part of the chapter. Finally, the Conclusion

discusses the main findings of the research and summarises the results of the project.

2 INTERNATIONAL RESPONSE TO DISASTERS This chapter presents current theories on the disaster response, the quantitative

constituents of disasters, the role of the international community in disaster response

and methods for assessing losses and needs following disasters. The purpose is to

present an analytical framework and to probe the literature for the relevance of the

aim of this research project.

2.1 Disaster management cycle Alexander (2002:5-6) presents one of many views on the disaster management

cycle - a model central to disaster management studies. The model (see Figure 2.1;

Alexander 2002:6), describes the cyclical approach that should be applied for

successful management of recurring disasters.

Source: Alexander 2002:6

Figure 2.1 The disaster management cycle

The model offers a framework for planning disaster management tasks. It is not

necessarily an accurate depiction of how disaster management projects are being

implemented in reality, particularly in the developing world where assets are lacking

and governments are weak (Twigg 2004:64). Other models, like the ones of Sundnes

and Birnbaum (1999) and Albala-Bertrand (1993), have been proposed for the task of

disaster management within specific domains. These models provide additional

attention and detail to a limited part of the disaster management task and cannot be

seen as being in competition with Figure 2.1 but rather complementing it. In the case

of Sundnes and Birnbaum (1999) additional phases for health disturbance assessment

and post response health assessment are added to improve the model relevance to

health practitioners. Similarly, the model of Albala-Bertrand (1993:12-13) is

developed for the purpose of analysis of causality, as well as the relation between

disaster management concepts in the domain of economics.

There is relative consensus in the literature with regards to the purpose and

content of each of the disaster management phases, represented by the sections of the

middle ring in Figure 2.1. However, Sundnes and Birnbaum (1999) as well as Walker

(1991) highlight the disparity between theory and practice when it comes to the “cost-

benefit” of actions. Walter (2004:11) as well as Walker (1991) point to the pre-disaster

phases as being the time during which invested efforts will generate the greatest

benefit. Nevertheless, post-disaster aid has long been favoured by funding and

implementing organisations (Walker 1991; Twigg 2004).

The mitigation phase covers two groups of activities: (1) prevention measures

aiming to avoid exposure to hazards altogether and, (2) mitigating measures aiming

to reduce the impact of hazards should they strike by structural, i.e. engineering

solutions and, non-structural means (Alexander 2002:9). In the preparedness phase,

following the mitigation phase, the focus is on activities taken in advance to increase

the effectiveness of an eventual response. This includes the development of operating

procedures such as evacuation plans and the development of tools like Early Warning

Systems (EWS) (IFRC 1995). Mitigating and preparedness measures require long-term

pro-active commitments from the involved actors (Twigg 2004:105). However,

projects tend to focus on “short-term outputs, rather than long-term outcomes”

(Walter 2004:108). Consequently, the bulk of aid is reactive to post-disaster situations.

The activities in the response phase have their emphasis on the prevention of

further losses by life preservation and provision of basic subsistence needs such as

healthcare, food and shelter (Alexander 2002:5). According to Albala-Bertrand (1993)

private, public and international actors have separate motivations for responding to

disasters. The underlying reasons for international interventions, Albala-Bertrand

(1993:153) writes, can be “put into a broad utilitarian framework (political and

economic) “. This statement will be examined further in section 2.3.

In the recovery phase the purpose is to bring the affected area back to its

previous state through reconstruction and restoration of damaged structures. The

recovery phase presents the start of a “window of opportunity” (Alexander 2002:8-9)

in which there is greater acceptance among the population for the implementation of

mitigation measures that normally would have been seen as unpleasant. The start of

the mitigation activities closes the disaster management circle.

2.2 Hazard, vulnerability and risk In her excellent comparative glossary Thywissen (2006) lists a plethora of

definitions of risk, hazard, vulnerability, exposure and additional concepts central to

disaster management. The concepts are essential to the understanding of the

mechanisms of disasters. Alexander (2000a:7) defines a hazard as “an extreme

geophysical event that is capable of causing a disaster”. Alexander continues to

classify hazards according to the degree in which human actions play a causal role.

The spectrum goes from social hazards, like crowd stampedes, where both the hazard

and its outcome are totally dependent on the presence of humans, technological

hazards, such as industrial accidents, through to natural hazards. It is important to

realise that human involvement is central for a hazard to develop into a disaster even

in the case of natural hazards (Hewitt 1983). Without human presence there would be

no disaster. The definitions of risk provided by Thywissen (2006) converge on risk as

a probability. This includes Alexander, who defines risk as:

The probability, that a particular level of loss will be sustained by a given series of elements as a result of a given level of hazard impact (2000a:7)

The terms ‘risk’ and ‘exposure’ are related. Peduzzi et al (2002:5) define

“physical exposure” as the product of the population at risk and the frequency and

severity of a given hazard. The process of risk assessment and the role of

vulnerability and hazard is clarified by the Pressure And Release model (PAR)

developed by Wisner et al (2004:51). In Figure 2.2, the pressures are depicted inside

the arrows on the left of the disasters and the release being represented by the hazard.

Wisner et al (2004:11) define vulnerability as “the characteristics of a person or group

and their situation that influence their capacity to anticipate, cope with, resist, and

recover from the impact of a natural hazard”. In Figure 2.2 Wisner et al (2004:51)

make clear that risk is a product of vulnerability and hazard.

- 10 -

Source: Wisner et al (2004:51)

Figure 2.2 Pressure And Release model (PAR)

Wisner et al (2004:49) point out that by removing the hazard or reducing the

vulnerability to a theoretical level of zero, the risk is removed. The access-based

approach, on which the PAR model of vulnerability is based, is only one of many

models of the relations between hazard, vulnerability and risk. In a comparative

study of vulnerability models, Vatsa and Krimgold (2000) contrasted the access-based

approach against an “asset-based approach” propounded by Swift (1989) amongst

others. Their findings included that both modelling approaches see poverty as a core

cause of vulnerability. If general development assistance and efforts of mitigation and

preparedness are lacking, the disaster management cycle only represents a model of

utopia. The reality in developing countries is a vicious circle (Alexander 2000a:13)

where each disaster increases the vulnerability of the affected people. Poverty is by

definition a situation in which the individual has limited assets (Sen 1999). Similarly,

poverty limits access to power, structures and resources; i.e. the root causes listed in

Figure 2.2. Considering the central role of poverty, it is clear that short-term disaster

relief alone will never solve the problem with excessive vulnerability (Walker 1991;

Twigg 2004). It may in fact increase vulnerability in that a vulnerable country

becomes reliant on a donor nation for preparedness efforts (Glantz 2003). In relation

to this, Wisner et al mention that:

lack of understanding [of the causes of vulnerability is] likely to result in policy makers and decision takers, restricted by the scarce resources at their disposal, addressing immediate pressures and unsafe conditions while neglecting both the social causes of vulnerability as well as the more distant root causes (2004:61)

- 11 -

This situation is reflected by the model developed by Alexander (2000a:4)

showing the disparity in the distribution of resources over the post-disaster activities

in responses to events in developing versus industrialised countries.

The earthquake-hazard The earthquake hazard is consistently classified as a natural hazard though it is

clear that “destruction is always dependent upon the presence and character of

human settlement and land uses” (Hewitt 1997:197). Wisner et al (2004:276) divide

earthquake-specific vulnerability into ex ante and ex post vulnerability. Ex ante

vulnerability refers to the situation that exists before the strike of a hazard. Ex post

vulnerability is related to secondary and tertiary impact in that it relates to “what

happens after the initial shock and in the process of recovery” (Wisner et al 2004:276).

The ex post vulnerability can be increased by a set of deleterious factors that may

follow a disaster. Examples of these include bad weather or food insecurity that on

their own could have been absorbed by the affected society.

In his article titled “Issues in the definition and delineation of disasters and

disaster areas”, Porfiriev (1998) attempts to define what constitutes an ‘affected area’.

He concludes that there is no single definition. Instead, he claims that it varies

depending on the ‘values’ of the user. When concentrating on defining the affected

area for a single earthquake event the physical exposure is more tangible and its

distribution over an area can be estimated using a range of factors such as magnitude

and hypocentral depth (Hewitt 1997:220; Yuan 2003). The estimation of the physical

exposure caused by earthquakes will be examined from an earthquake engineering

perspective in section 5.2.

Table 2.1 (Schneiderbauer and Ehrlich 2004:32) provides a list of indicators of

hazard-specific vulnerability coupled with their individual relevance in the

vulnerability estimation process. It does so for each ‘social level’, stretching in five

seamless steps from the individual, to the administrative community, to the country,

to region and to the cultural community.

- 12 -

Table 2.1 Earthquake-specific social level vulnerability indicators

Social level Parameter Indicator Relevance Individual and

Household Quality of and age of building

Building construction date. High

Availability and enforcement

of building code incorporating seismic resistance.

Main building material. High Urban growth. High Size of building Number of floors. High

Number of families per residential building.

Location of building Terrain (e.g. slope, gradient). High Hygiene Access to drinking water. Medium Quality of sewage system. Medium

Administrative community

Preparedness Fraction of earthquake resistant buildings

Country Availability and enforcement

of building code incorporating seismic resistance.

Region Vaccination Fraction of population vaccinated.

Medium

Legal requirements of vaccination.

Medium

Cultural community Source: Schneiderbauer and Ehrlich 2004:32

Even though macroscopic, many of these indicators reflect the state of the built

environment. Examples include average number of floors and the average number of

inhabitants per dwelling. Wisner et al (2004:277) categorise the determinants of

vulnerability to earthquakes as: the location of the earthquake, the temporal

characteristics of the earthquake, the characteristics of the buildings and the protective

measures. These determinants agree with those listed in Table 2.1. The data which

can be expected to be available on these determinants in a developing context is of far

lower quality than what can be expected in a developed country (Albala-Bertrand

1993:39). There are, however, possibilities to use proxy indicators of vulnerability.

Hewitt (1997:215) writes that vulnerability towards earthquakes in developing areas

“tends to reflect the more or less local ‘building culture’” which he defines as the

“available construction material and their costs, economic activity, social and political

organisation, the history and modern transformations of construction technique, and

customary or fashion preferences”.

The urban growth rate as an indicator of earthquake vulnerability has been

highlighted by Schneiderbauer and Ehrlich (2004). A theory is that fast urban growth

- 13 -

result in lower quality buildings and reduced efficiency of mitigation measures. This

subject will be re-examined through the point of view of seismologists and earthquake

engineers in Chapter 4.

2.3 International disaster relief There is no universal definition of what constitutes international disaster relief.

Smillie and Minear (2003:19) point to this lack of a common definition and

recommend the development of common terminology as a step to support more

objective relief policies. Albala-Bertrand defines ‘disaster response’ as “a wide array

of endogenous and exogenous reactions, measures, and policies to counteract,

mitigate, and prevent disaster impacts and their effects” (1993:20). This would, in

effect, cover all actions available in the disaster management cycle. However, Albala-

Bertrand (1993) sees ‘disaster relief’ as the whole set of responses aimed at

containment of indirect effects on people. In other words, he sees ‘disaster relief’ as a

subset of ‘disaster response’. Endogenous responses, according to Albala-Bertrand,

are channelled through society’s “inbuilt mechanisms” (1993:21). Other authors refer

to these mechanisms as society’s ‘coping capacity’ or ‘resilience’ (Schneiderbauer and

Ehrlich 2004; Thywissen 2006). Exogenous responses are channelled through

mechanisms that “bypass in-built frameworks [and] shift initiatives away from

regular actors” (Albala-Bertrand 1993:22). Albala-Bertrand (1993) argues that

international relief following sudden-onset disasters normally is exogenous and

focused on the effects of the disaster, as opposed to the cause. This makes the key

proponents of success in international responses different to those identified in

domestic responses by Fischer (1998:89-94) i.e.: co-ordination, designated roles and an

institutional existence.

An emerging tool for co-ordination of disaster relief funding is the Consolidated

Appeal Process (CAP) (Tsui 2003:39). The CAP is most commonly used in protracted

emergencies, to seek funding for recovery phase operations (Smillie and Minear

2003:21-24). Tsui (2003:39) does, however, mention that in cases when there is an

open CAP for a country that is subsequently struck by a disaster, a revised version of

the CAP is usually issued. He describes the CAP as a process in which:

[...] national, regional, and international relief organizations jointly develop a common humanitarian programming, strategic planning, and resources mobilization document, which is regularly reviewed and revised. (2000:39)

- 14 -

Smillie and Minear (2003:21-24) criticise the current use of the CAP and claim

that it leads to “cherry picking” of projects by the donors and a de facto exclusion of

small organisations from the funding appeal process. Considering that the CAP is not

the main financial instrument in the initial international response to disasters, it will

not be investigated further here.

Ebersole (1995) looks at ethical and legal issues in disaster relief and presents a

set of criteria for appropriate humanitarian assistance. He recommends that

humanitarian assistance should follow the principles of humanity, impartiality,

neutrality, independence and empowerment (Ebersole 1995:16). This translates into a

recommendation for humanitarian aid to focus on “human suffering” with the relief

being free from discrimination and guided solely by the needs of those suffering

without any attachments to “political, military or other interests” (Ebersole 1995:16).

The pitfalls of disaster relief Alexander (2000a:84) discusses the benefits and dangers of international short-

term relief and its role as a geo-political tool. He claims that the decision makers in

funding organisations are “forced by scarcity of funds to be discriminating in their

donations, but one never knows what will be next, and hence how great the needs

will be in the next disaster.”(2000a:84) In other words, the international community

has limited resources and cannot get involved in every disaster. If continuously

adopting a reactive approach, the cause of the disaster will never be resolved. As

Kent puts it, it is a question of whether to “cure or cover” (1987:20). The decision

maker in the relief organisation hence has to determine to which disasters to respond,

how and with which purpose. However, the purpose is not limited to whether to cure

or cover. Albala-Bertrand (1993) argues that humanitarian aid is a more powerful

geo-political instrument than its military counterpart. Aid can be focused on

countries with which the donor wants to improve relations or “be withheld in order to

bring retribution upon citizens of uncompliant [sic] nations” (Alexander 2000a:85).

Absolute need is hence not necessarily what governs the nature of the international

relief. However, Alexander continues to state that the relief that is supplied in direct

relation to a sudden-onset disaster commonly “is sufficiently limited in size and

divested of strategic connotations to be relatively free of constraints on its allocation”

(2000a:85).

- 15 -

Another danger is the public notion that some relief, independent of type, is

better than nothing at all. Both Kent (1987:12-21) and Alexander (2000a:87) highlight

that not all types of aid are helpful. Both authors agree in that inappropriate or

redundant aid reduces the efficiency of the overall response because it absorbs

logistical assets and manpower that is diverted from dealing with more pertinent task

such as storing and distributing urgently required relief.

Media, politics and disaster relief Public image, fund raising potential, peer group prestige and ultimately the ability to respond are now more dependant than ever […] on whether one’s actions are seen on TV. (Walker 1991)

Olsen et al (2003) present a hypothesis that the magnitude of the humanitarian

aid resulting from a disaster is governed by three factors: the intensity of the media

coverage, political interest in the affected area and the presence of international NGOs

(INGO) in the disaster area. They conclude that media influence is not as strong as

commonly conceived and that the most important factor is the political interest in the

affected area. Kent supports them in their conclusion:

Geo-politics, we are often told, is one reason for the unpredictability of humanitarian intervention. Of course, politics at any level of human activity is a crucial factor, and it certainly is in the case of disaster relief. (1987:176)

Others, like Benthall (1993:221), point to cases where the media has been pivotal

to the emergence of international relief. The current selective approach of western

media can lead to ‘forgotten disasters’ (Holm 2002; Wisner et al 2004:28-29). This

occurs when less photogenic disasters, usually slow-onset prolonged events, fall out

of the media limelight and are likely to remain in the fringes of international attention

until their situation is dramatically worsened such as in Somalia and North Korea in

the late 1990s (Jeffreys 2002). Smillie and Minear support the conclusions of Olsen et

al (2003) and proceed to provide twelve recommendations that may rectify the

situation, one of which is “Less politicized humanitarian financing” (2003:15).

Although Smillie and Minear (2003:15) acknowledge that “political pressures on the

humanitarian delivery chain are unavoidable” they propose that joint studies that

“demonstrate the humanitarian cost of politicized choices” could be used as a tool for

increasing the objectivity of relief. Albala-Bertrand (1993:141) presents a series of

arguments on what motivates the various relief actors. As previously mentioned, he

sees the motivation of the international actors as being largely utilitarian. However,

he proceeds to argue that there are exceptions to this rule. Bi-lateral aid can be

- 16 -

influenced by powerful political lobbies in the donor country and this can “explain

some of their short-term motivation to disaster relief in the absence of other more

permanent reasons (e.g. economic, strategic, political)” (Albala-Bertrand 1993:153).

This effect, he claims, can be reduced if the aid agency is multi-lateral, as long as no

single actor has unduly strong influence.

International Search And Rescue (SAR) Coburn and Spence (2002:104) define SAR as the rescue process of determining

“the location and rescue of victims trapped in collapsed reinforced concrete

structures”. In the international arena this type of aid became commonplace in the

1980s (Coburn and Spence 2002). The literature (Coburn and Spence 2002:105; Walker

1991) agrees on both the limited contributions provided by international SAR, as well

as SAR’s importance as a public gesture of sympathy. Research has, however,

suggested ways in which the effect of international SAR can be increased. Walker

(1991:18-19) states seven criteria that international SAR missions must fulfil in order to

be effective in life-saving:

1. They must possess skills and equipment to locate entrapped individuals.

2. They must possess skills and equipment for stabilizing victims before

handing them over to the medical authorities.

3. They must possess skills and equipment to extricate trapped individuals

from collapsed buildings.

4. “In order to apply the above criteria successfully there must be live

victims for them to attend to. Therefore the international relief must

arrive on site no later than 48 hours after the disaster strikes and

preferably within 12 hours.” (Walker 1991:19, emphasis added) This

view is supported by Coburn and Spence who states that “A significant

improvement in the live recovery rate of international SAR teams could

be achieved by speeding up their time of arrival on the disaster site.”

(2002:106)

5. They should be self-contained in terms of food, water, logistics,

accommodation etc. in order to reduce stress on local authorities. There

are, however, examples of where well organised units with the intention

to be autonomous fail to be so. For instance, in the 2003 Bam, Iran,

- 17 -

earthquake wooden baulks were needed to shore up the tunnels into

collapsed buildings. Wood is scarce in the desert area around Bam and

international agencies failed to take this into account when launching

their response (personal communication with David Alexander, June

2006).

6. They should have the ability to speak, communicate and co-ordinate

with local administrators and have an understanding of how local

systems of authority function.

7. The teams should only be sent to earthquakes that have resulted in a

type of impact to which the team’s expertise is beneficial. Walker

mentions that “teams are only likely to be useful where multi-storied

precast concrete buildings have collapsed leaving voids where people

may be trapped”(1991:19). This means that earthquakes in areas without

such structures are not likely to benefit from SAR relief.

In a study of the use of SAR assets in the international relief missions following

earthquakes, Walker (1991) found that, at the time, only a minority of the international

teams had the equipment and training necessary to locate and extricate trapped

victims. He was even more sceptical of whether the level of medical skills possessed

by teams was appropriate. This is likely to have change since the study of Walker, but

there are indications from more recent publications that similar problems still exist.

Coburn and Spence (2002:108) see it as a requirement that international SAR missions

are accompanied with appropriate medical expertise and equipment. They suggest

that the international community can be of help in the provision of specialised

hospital equipment and skills required to treat injuries typical to earthquakes. To a

degree the IFRC is contradicting Coburn and Spence (2003) when they state that:

Local medical practitioners are better able to respond to immediate needs and the local health system is far better adapted to common local problems than any expatriate team (1993:22)

The above citation in isolation does not mention the role of the pre- and post-

state of the local and national medical capabilities, which is of central importance

when judging the requirement and potential impact of external relief (Coburn and

Spence 2003; Darcy and Hofmann 2004). Still, the IFRC quote can be interpreted as

- 18 -

stating that the pre-disaster training and outfitting of medical assets in high risk areas

is a more cost-efficient form of aid.

When Walker (1991) conducted his study he found that the main causes behind

the low efficiency of international SAR due to lack of co-ordination with local

authorities and potentially the inappropriate selection of earthquakes to which SAR

assets are dispatched. Since then, the emergence of OCHA as an actor in the co-

ordination of the international relief is likely to have improved the situation.

Nevertheless, it is relevant to analyse the problems that the international SAR

missions were faced with in their early days. With regards to excessive response

times, the Office for U.S. Foreign Disaster Assistance (OFDA) found in a study

conducted in 1987 that the main temporal bottle-necks in the delivery of SAR-based

relief following earthquakes were (OFDA 1987 in Walker 1991):

1. The host countries’ delay in issuing a state of emergency.

2. OFDA not making an immediate decision to deploy SAR assets.

3. SAR teams not being close to an appropriate ‘departure site’.

4. SAR teams requiring “a great deal of time” to get equipment, etc. ready.

5. Delays in arranging logistics.

6. Lack of internal co-ordination in the dispatching process.

7. Long travel times to the rescue site potentially necessitating rest periods

for the SAR team before the start of work.

The consequence of the delay in the international response has shown itself on

numerous occasions. A commonly mentioned example is the 1988 earthquake in

Armenia where more than one thousand international SAR and medical professionals

arrived in country, some as late as more than a week after the event, only to extract 62

persons (UNDRO 1989). Based on the numerous failures of international SAR relief

Walker suggests that the relatively large financial assets required to mount a single

SAR mission “… could save more lives if directed at community preparedness, which

might include the training of local search and rescue capacity” (1991:27) and that:

The international solidarity which we would wish to express can be better achieved through a long term relationship with vulnerable communities rather than a three week mission during an emergency. (Walker 1991:27)

- 19 -

2.4 Initial assessment of loss and needs Walker, […] a policy planner for the International Federation of the Red Cross and Red Crescent Societies, […] extols the art of relief, which ‘is to make hard decisions under pressure and with minimal information’ (Benini 1997:351)

Smart (2005) defines the decision situation that Walker describes in the above

citation by Benini as the “knowledge-intensive tasks in humanitarian aid”. Smart

(2005) divides these tasks into: situational assessment, needs assessment, relief

planning and future vulnerability planning. Focus in this study lies on situational and

needs assessment where loss assessment is part of the situational assessment. In a

different typology of the same subject Kent (1987:136) divides the communication

surrounding international relief in three pragmatic phases according to the purpose of

the communication in each phase: (1) “Has a disaster occurred?“, (2) “Assessing the

disaster.” and (3) “Responding to a disaster.”

Kent (1987) is supported by Darcy and Hofmann (2003:7) and Currion (2003) in

his opinion that it is in his first phase, in the immediate aftermath of a sudden-onset

disaster, that the scarcity of baseline data is causing most problems for the decision

makers. To make matters worse, phase one is the time when data and information on

loss and needs are most relevant to the organisations potentially providing relief (De

Ville De Goyet 1993; Comfort et al 2004). To deal with this problem, large

organisations commonly have internal policy guidelines for assessment; see for

instance USAID (1994) and IFRC (1999). Research on ‘needs assessment’ processes

has been ongoing for several decades (Kent 1987:23) and there are plenty of models

such as McConnan (2000), ADPC (2000) and Darcy and Hofmann (2003). For

example, the Asian Disaster Preparedness Centre (ADPC) has proposed a model with

a set of “planning factors” for the estimation of needs, this includes “X Search and

Rescue teams per Y missing people” and “X litres of water per person for Y days”

(2000:5).

In her recommendations, McConnan (2000) approaches ‘initial assessment’ in

the context of complex and slow onset disasters. This can be deduced from the level

of detail that McConnan (2000) expects in the ‘initial assessment’. An example is that

the profile of the affected population should include: “Demographic profile (by

gender, age, social grouping)” and an account of which “Assets people have brought

with them” (McConnan 2000). In the ‘initial assessment’ McConnan (2000:179-184)

sees the aim as providing decision makers with an “understanding of the emergency

- 20 -

situation and a clear analysis of people’s needs for shelter, clothing and household

items”. In a list of fourteen ‘guidance notes’ for the making of an initial assessment,

McConnan (2000) mentions: the assessment of infrastructure, the outline of physical

geography and the use of early warning information. This level of detail is clearly

impossible to achieve in sudden-onset disaster, even if a significant amount of the

data are collected ex ante. An alternative is to estimate the approximate medical needs

following the sudden-onset disasters as discussed by Coburn and Spence (2002:122).

However, even with the support of formal guidelines, agencies often rely on their

experience to estimate need and plan a response (Darcy and Hofmann 2003:7). Apart

from taking time and absorbing resources, an additional drawback with detailed

assessments made in the field is that of ‘assessment fatigue’ in the affected population

(Benini et al 2005; Keen and Rile 1996). When the assessed population is subjected to

repeated uncoordinated assessments from individual relief agencies without seeing

anything to their benefit coming out of the process they lose confidence in expatriate

staff (Keen and Ryle 1996). Co-ordination and cooperation is hence of importance not

only in programme implementation, but also in assessments (Benini et al 2005). Before

a field assessment can be made, the remote decision makers have to judge whether to

make an assessment at all. Darcy and Hofmann (2003) propose four overarching

questions that the remotely placed decision makers are confronted with in the initial

stage following a potential disaster:

1. Whether to intervene

2. The nature and scale of an appropriate intervention

3. The optimal prioritisation and allocation of resources

4. Programme design and planning

These questions are all related to the humanitarian need in a disaster zone. Just

as the term ‘humanitarian aid’ is not defined, Darcy and Hofmann (2003:5) point out

that there is no common definition of ‘humanitarian need’. They choose to use the

term ‘humanitarian need’ “to describe the need for (a particular form of) relief

assistance or some other form of humanitarian intervention”(2003:5). They go on to

set general criteria for good assessment practice (2003:45). They claim that the

following six criteria are relevant to all types of humanitarian assessment and not only

in situations following sudden-onset disasters (2003:45):

- 21 -

1. Timeliness: providing information and analysis in time to inform key

decision makers.

2. Relevance: providing the information and analysis most relevant to those

decisions.

3. Coverage: providing a level of detail on par with the scale of the problem.

4. Continuity: providing information throughout the course of a crisis.

5. Validity: using methods that can be expected to lead to a sound conclusion.

6. Transparency: being explicit about assumptions made, methods used and

information

These criteria are similar to those applied by for instance Vereign (1998) in the

domain of data quality analysis in remote sensing, which will be reviewed in section

5.3.2. When compared, it is clear that McConnan’s (2000) guidance notes and Darcy

and Hofmann’s (2003) criteria carry a similar message but are targeted at different

audiences: field practitioners in the former case and policy makers in the latter case.

As an example, McConnan (2000) supports Darcy and Hofmann’s (2003) criterion on

the pivotal role of transparency in initial assessments following sudden-onset

disasters. Darcy and Hofmann argue that humanitarian assessment in that context

“depend as much on assumption, estimate and prediction as they do on observed

fact” (2003:8). It is hence essential to openly present the assumptions made.

2.5 Summary This chapter has examined the theory in the international disaster response

domain and has probed the relevance of this research project as a precursor to the

deeper relevance study in Chapter 8. The disaster management cycle was introduced

to provide a framework for identifying the various phases in which the international

community is acting and within which decisions can be supported. Existing research

indicated that international activities in the response phase of the disaster

management cycle are bound to be ineffective. Nevertheless, research shows that the

international interventions in these phases will continue. Therefore, it makes sense to

invest resources to improve the international actions in the response phase. The

factors that drive international relief were shown to include media coverage,

international presence and the political relation between the affected country and the

donor.

- 22 -

Being a common type of international response following earthquakes, SAR

relief was specifically reviewed. Literature was consulted for the general keys to

success and for the most common obstacles preventing success in the implementation

of international SAR aid.

The concepts of hazard, vulnerability and risk were presented for the purpose of

providing a way of grouping and identifying quantifiable aspects of disasters that

may serve as a basis for a DSS.

Finally, existing guidelines for loss and needs estimations were examined for the

purpose of establishing the processes and types of decisions that the decision makers

are faced with. This adds detail to the disaster management cycle and enables more

targeted analysis of the actions taken by international community immediately after a

potential disaster.

- 23 -

3 SUPPORTING DECISIONS WITH INFORMATION SYSTEMS In this study, Management Information Systems (MIS) are tested as the primary

solution to the identified problems in the international response to sudden-onset

disasters. This chapter serves to introduce the fundamental concepts of MIS. It

provides an analytical framework for the discussion on which types of users can be

supported and with which kinds of MIS they should be supported. Some views on

the pitfalls in the development of DSS and generic information systems are also

presented.

3.1 Typology MIS have been used in the commercial industry since the advent of information

technology (O’Brien 1999). Even though profit-based industry was the first to adopt

the use of MIS, its potential use in non-profit organisations has been discussed for

some time (Wallace and Balogh 1985). Organisational focus on financial profit is not a

prerequisite for these systems to be beneficial. As an example, Wisniewski (1997)

presents a set of case studies where quantitative decision support methods have

successfully been applied to governmental and non-profit activities.

MIS are distinct from regular information systems in that they are used to

analyse other information systems used for operational activities in the organisation.

Examples of operational data in a furniture selling business are stocks, supplier

orders, customer orders etc. These data can be used by a MIS to support management

decisions. O’Brien (1999) defines analytical databases as databases consisting of

summarised data and information extracted from operational databases with the

purpose of supplying decision makers in the organisation with the most needed data

and information. Analytical databases are often multidimensional and complex to the

extent that a software interface is required to query, interpret and present its contents

to a user in an understandable format (O’Brien 1999). Two types of such systems are

DSS and Online Analytical Processing (OLAP) systems.

OLAP systems work in real-time, i.e. they process requests from users using live

data, delivering the output to the user without delay. OLAP systems are central to the

structural support method outlined by Kersten (2000), in which data and information

are digested with the aim to provide a quick and intuitive overview of vast amounts

of data. O’Brien (1999:460) classifies the most common analytical operations in OLAP

- 24 -

as: ‘Consolidation’, ‘Drill-down’ and ‘Slicing and Dicing’. Consolidation is the

grouping of information into coherent sets, e.g. cities into provinces and drill-down is

its opposite. Slicing and dicing gives the ability to look at information from different

angles and contrasting types, e.g. analysing the sales trend of a product in a set of

regions over time.

O’Brien (1999:61) distinguishes between DSS and ‘expert systems’ based on their

role in the host organisation. He defines expert systems as being systems aiming to

replace the human involvement, often by applying artificial intelligence technology to

automate a decision (O’Brien 1999:63). This characteristic separates expert systems

from both DSS and OLAP systems. The automation of decisions requires structure

and this makes expert systems most powerful in the well structured environment

surrounding operational management (see Figure 3.1) (O’Brien 1999:456).

Source: O’Brien (1999:456)

Figure 3.1 Decision structure according to hierarchical levels

Examples of such systems include those for diagnosing illnesses or financial

planning systems. DSS, on the other hand, does not aim to replace the human

decision. Instead it supports the decision process by providing an interactive tool that

provides the decision maker with “analytical modelling, simulation, data retrieval,

and information presentation capabilities” on an ad hoc basis (O’Brien 1999:61).

Although O’Brien (1999) sees MIS as being a tool separate to expert systems and DSS,

a more common view (see for instance Kersten et al 2000:40) is of MIS as a collective

term for the science of developing and maintaining expert systems, DSS and similar

systems.

- 25 -

O’Brien (1999:518) discusses the use of IT to “break barriers” in processes, the

most frequently targeted barrier being the ‘time barrier’. The common name for this

activity is ‘just-in-time-something’, e.g. just-in-time-inventory (O’Brien 1999).

Although just-in-time actions in the industry are facilitated by comprehensive IT

projects (O’Brien 1999), similar effects could be achieved in the humanitarian relief

sector by the appropriate application of DSS (Kersten 2000; Zschau and Küppers

2003). Although not explicitly stated in literature, DSS used in the preparedness and

response phases in emergency management are often referred to as Early Warning

Systems (EWS) (see for instance Zschau and Küppers 2003).

3.2 Decision support Andersen and Gottschalk (2001) list the typical questions that managers at

various organisational levels in a generic commercial organisation are faced with (see

Table 3.1; Andersen and Gottschalk 2001) and what the purpose of an information

system is on each organisational level.

Table 3.1 The role of the information system per hierarchical level

Level Questions Purpose Strategic management

What kind of business? Which products? Which markets?

Control (decision) benefit

Tactical management

Given business, what kinds of resources are needed and how are they best developed?

Operational management

Given business and resources, how are they best utilised?

Administration How to do these functions in the best way?

Rationalisation (automation) benefit

Operations How to make the products in the best way?

Rationalisation and market benefit

Source: Andersen and Gottschalk 2001

Table 3.1 highlights the intentions behind the implementation of information

systems at the various organisational levels. On the managerial levels, the aim is to

increase the control of the organisation through improved decisions. On the

operational and administrative levels the main benefit of information systems is

rationalisation (Andersen and Gottschalk 2001).

O’Brien (1999:457) sees a strategic DSS as a system supporting decisions relating

to long term goals in uncertain and ever-changing context (see Figure 3.1). By

confronting the requirements on strategic DSS suggested by O’Brien (1999) and

- 26 -

Andersen and Gottschalk (2000) with the underlying decision support methodologies

suggested by Kersten (2000) it is clear that a strategic DSS must take advantage of all

methodologies to be successful. Kersten (2000:47) suggest that DSS are built on using

monadic, structural and contextual methodologies. A DSS that focus on projecting

data in a more intuitive and readable structure is defined by him as a monadic system.

He sees ‘structural’ methodologies as being used for structured tactical and

operational level decisions, i.e. O’Brien’s (1999) ‘expert systems’. Kersten’s (2000)

contextual methodologies include methods that “aids to structure decision problems,

estimate probability distributions, analyze risk and check for consistency of the

decision maker’s reasoning” (Kersten 2000:46). For the specific case of decision

support in humanitarian assistance Tsui sees the best practice to be centred on the

developer’s commitment to:

Define user needs and utilise data sets and formats that directly support decision-making at the field and headquarter levels. Identifying user groups, conduct user requirement analysis, inventory information resources and define core information products based on user input. (2002:14)

3.3 Usability design The science of usability comes from the domain of engineering. It, and its

synonym ‘user friendliness’, has been used in human computer interaction (HCI)

since the 1970s (Faulkner 2000:6). Norman’s (1998:188) theory of the Psychopathology

of Everyday Things (POET) includes seven principles for transforming difficult tasks

into simple ones. The first rule is for the designers of artefacts to “use both

knowledge in the world and knowledge in the head” (Norman 1998:189). Faulkner

(2000:190) elaborates on this to state that the knowledge necessary for completing a

given task using an artefact should be available in the real world. According to

Norman (1998:164) one of the problems caused by inappropriate design is that of

‘selective attention’ of the user. He gives the example of people evacuating a building

on fire who push against the emergency exit, harder and harder, failing to realise that

the door opens by pulling. When a prominent problem exists in the user’s

environment, the user tends to focus on that, to the cost of reduced attention to other

factors (Norman 1998). This is obviously of great concern in the development of tools

for disaster management. Norman (1998) recommends that the designer take this into

account by adopting ‘forcing functions’ that prevents a user from operating the

artefact wrongly. Emergency doors, for instance, should open outwards. Similar

- 27 -

consideration has to be taken in the design of software tools in general (Norman

1998:177) and DSS in particular (Kersten 2000:43).

Norman (1998:172) warns of the “Two deadly Temptations for the Designer”

that are directly counterproductive to systems with the purpose of providing decision

makers with digested information, like Kersten’s (2000) monadic DSS. Norman

names the temptations as “Creeping featurism” and the “Worshipping of false

images” (1998:172). The worshipping of false images is caused by the urge of the

designer to introduce complexity as a means of showing the user the technical

sophistication of the artefact. Creeping featurism is caused by the designers urge to

make the user’s life easier by adding features to the artefact. With each added feature

the complexity of the artefact is increased exponentially, which lowers its usability

(Norman 1998:172). Norman (1998:172) suggests that the best way to prevent this

situation is by being very restrictive in adding functionality. If that cannot be done he

recommends that the features are organised through modularisation. Modularisation

can be achieved by tools like drill-down and data-slicing (O’Brien 1999).

3.4 Summary This chapter laid out the alternatives for the development of a DSS. It is

particularly relevant to the formulation of the user requirements. The identification of

the types of users in this study in Chapter 8 relates back to this chapter for

identification of suitable support methods.

The models and methods presented here were developed for application in

commercial organisations. It is, however, likely that the typology of decisions and

systems are going to be similar in a non-profit environment. Furthermore, the section

on usability and the POET model provide an important reminder that the system will

be used by humans. Usability is as pivotal in the successful development of

information systems, as it is in the development of other everyday artefacts. Existing

systems used in immediate disaster relief is analysed for usability, among other

aspects, in Chapter 9.

- 28 -

4 DECISION SUPPORT IN DISASTER RESPONSE The quantitative phase of this research project requires data from the case

studies to be collected and stored in taxonomies. The first section examines existing

processes for estimating loss and needs incurred by disasters, without visiting the

disaster site. In the second section the current state-of-the-art in decision support for

international response is reviewed.

4.1 Tele-assessment Tele-assessment in international relief is here defined as a set of methods used

for estimating the characteristics of a potential disaster at a distance. This combines

disaster science with the science of management information systems. Although

hazard data are fundamental, Wyss (2004b) writes that even when complete and

accurate, hazard data alone is not sufficient to judge whether an event will require

international intervention. For an assessment of loss or need there is hence a

requirement to combine hazard data with indicators of vulnerability.

Furthermore, disasters are spatial in their nature (Alexander 1993:25) and data

collected on them will hence be spatial. This in turn is reflected in the analysis of the

data (Alexander 2002:18; Coburn and Spence 2002:97). Researchers in disaster

management were pioneering in the use of Geographical Information Systems (GIS)

as a means to study the disaster phenomena in its entirety (Johnson 1995). Today, GIS

is applied in most areas of disaster management (see for instance Oosterom et al 2005

or Bankoff et al 2004). Although there are success stories of the use of GIS in disaster

mitigation and preparedness efforts (POST 2005), Zerger and Smith (2003) show that

the practical difficulties increase significantly when applying decision support in the

post-disaster phases. These phases require real-time analysis of data (Beroggi and

Wallace 1995) that fundamentally change the requirements on the systems. In their

case study of an introduction of a disaster DSS in a city council, Zerger and Smith

(2003) encountered organisational problems in the implementation that are more

common in major organisational changes (see for instance Eriksson and Stanojlovic

2000). Zerger and Smith (2003) also found that in real-time systems the user

requirements on temporal resolution exceeded that of spatial resolution; which they

claim is the opposite from the requirements in pre-disaster system. This leads to

systems attempting to provide decision support as early as possible.

- 29 -

4.1.1 Early warning Early warning and Early Warning Systems (EWS) are frequently used terms in

disaster management; exemplified by the United Nations’ sponsored series of

international conferences on EWS (see for instance Zschau and Küppers 2003). The

official United Nations definition (ISDR 2004) is:

The provision of timely and effective information, through identifying institutions, that allow individuals exposed to a hazard to take action to avoid or reduce their risk and prepare for effective response

Twigg (2003) presents the “early warning process” in a way that can be

interpreted to be in competition with the idea of the disaster management cycle.

Twigg’s (2003) early warning process starts with “Evaluation/forecasting

(observation and prediction)” leading into “Warning/dissemination” and ending in a

response implementation. He consequently sees early warning as something that

extends beyond the tool used to produce the warning. Seibold (2003) adopts a more

generic stance and sees early warning as the art of estimating and communicating

In earthquakes, early warning following a tremor is limited in time and scope by

the 8 kilometre per second theoretical maximum speed of the seismic waves (Seibold

2003). Here, early warning is hence not used to refer to warnings taking place before

the hazard has started. The long list of “early warning projects” gathered by Zschau

and Küppers (2003) give an indication of the wide interpretation of the concept.

Judging by the emphasis given to early warning for earthquakes in their publication,

it is safe to assume that Zschau and Küppers (2003) see it as a genuine subject and not

as part of what Coburn and Spence (2002:77) see as: the yet to be scientifically

accepted domain of earthquake prediction.

EWS development guidelines Glantz considers that a generic EWS should provide information on five central

W’s:

What is happening with respect to the hazard(s) of concern? Why is this a threat in the first place […]? When is it likely to impact […]. Where are the regions most at risk? Who are the people most at risk, i.e. who needs to be warned? (2004:17)

Glantz is supported by King (2005) who sees the role of the senior decision

maker in the early post-disaster phase, that King defines as the ‘situational awareness’

phase, as being to find answers to the following questions:

- 30 -

• What is the latest/current humanitarian situation in the country?

• What are the most recent severity indicators?

• Who are the affected populations? How many are there and where are they

located?

• What are the conditions and humanitarian needs of the affected populations?

• What is the assessment of damage to infrastructure?

• What is the latest/current security situation in the affected areas of the

country?

An additional aspect in which there is relative agreement in the scientific

community is the fundamental importance of transparency in early warning systems.

Both Darcy and Hofmann (2003) and King (2005) see the declaration of assumptions

pivotal to initial assessment. Glantz (2004:20) takes a similar stance to transparency

although with regards to early warning. He claims that because it is not feasible to

provide early warning without making assumptions, one should ascertain that ones

assumptions are openly stated, although he admits that transparency is not a clear-cut

issue.

Government and EWS managers might want to keep uncertain EWS output

internal to avoid false alarms and potential “cry-wolf”-effect (Atwood and Major

1998). Glantz (2004:41) points out that the output of a EWS is not only received by its

intended end users; it is also used as input to other systems and processes. He defines

this phenomenon as the early warning cascade (2004:41). False alarms can hence

result in a domino effect if the hosts of the EWS are not aware of any such cascades

starting with their system. Alexander summarises the characteristics of a successful

warning system as: “flexibility and a marriage of technical and social expertise”

(2002:147).

4.1.2 Loss assessment Existing methods for early warning in earthquakes focus on either losses or

needs (Zschau and Küppers 2003). Most methods are created for use in developed

countries with large amounts of baseline data. For instance, Tralli (2000) developed

and tested the suitability of a method, called the Early Post-Earthquake Damage

Assessment Tool (EPEDAT), using a range of ground based sensors in an urban

- 31 -

setting in a developed context. This approach is costly and requires a degree of

knowledge of where an earthquake is likely to strike in order to pre-position the

sensors (Bolt 2004:113); which makes it unsuitable for use in developing countries.

Remotely sensed imagery Showalter (2001) presents the progression of remote sensing in the disaster

management domain since the 1970s. The use of remote sensing has become more

common in disaster management as prices have gone down, delivery times improved

and most significantly as the resolution of the sensors has increased (Showalter 2001).

Using airborne or space-born platforms, it is possible to acquire images of a disaster

area. When performing loss assessment using remotely sensed imagery, the two main

methods entail the use of a post-event image only or the use of an image pair

consisting of pre- and post- event images (Eguchi et al 2003). According to Eguchi et al

(2003), regardless of which method is applied, partial damage and damage to the

vertical parts on structures can seldom be detected. They continue to show that, even

with expert input, only totally collapsed structures that are not hidden in shadows can

be accurately detected.

Al-Khudhairy et al (2002b) apply a semi-automated method for detecting

severely damaged structures in a post-event image and concluded that even though

the method is feasible the commission errors, i.e. the number of sound structures

classified as damaged, are considerable2. However, one important conclusion of their

study (2002b) is that automated damage detection is more accurate when applied in

rural areas where structures are relatively isolated. This was confirmed in a later

study focusing on applying their method in the built environment (Al-Khudhairy et al

2003), but the commission errors were still considered too high.

Al-Khudhairy et al (2002b) showed that the use of image pairs results in higher

accuracy than with a post-event image alone. However, Al-Khudhairy and Giada

(2002) found that a major difficulty lies in finding a pre-event image that is compatible

with the acquired post-event image. In their study, this was particularly true in rural

areas and developing countries because image archives seldom contain images of

such regions. They also showed that considerable amounts of precious time elapse

from when a disaster has occurred to when a post-event image (including image 2 Using selective object oriented image classification to detect severely damaged or collapsed structures in a rural environment Al-Khudhairy et al (2002b) found that the omission errors were 0-25% and the commission errors 14-92%

- 32 -

acquisition, reception, processing and delivery) is ready to be used by the decision

maker. In their case studies they showed that, not including analysis, the delivery

time in an optimal case is 48 hours but more realistically three or four days depending

on revisit time of the platform and the metrological conditions in the area. A

subsequent field study by Kandeh et al (2005) confirmed that situation and showed

that the infrastructure in a developing country can further increase the time required

for the map product to reach those who need it.

Remote sensing may be successfully applied in the pre-disaster phases to

support mitigation, preparedness and response. For instance, remote sensing has

been applied to estimate building stock over large areas using radar imagery

(Shakhramanian et al 2000:137; Chung et al 2003; Brzev et al 2001). Even though the

application of remote sensing for initial loss assessment has many drawbacks, the

sensors and methods are constantly improving and the reliable detection of damage to

complex structures such as bridges and roads will soon be possible (Eguchi et al 2003).

Numerical modelling Scientific literature contains a plethora of attempts at modelling the impact of

earthquakes numerically. The main challenge in creating and applying models in real

time in a developing context is the lack of baseline data. Shakhramanian et al (2000)

have solved this issue through the adoption of a proprietary, somewhat secretive,

database of building qualities for the majority of cities in the world. Their baseline is a

main contributor to the development of what now are several newly spawned tools

for global loss and needs estimation (e.g. Wyss 2004a). In a separate numerical study,

Gutiérrez et al (2005) analysed the feasibility of applying Principal Component

Analysis (PCA) to determine which quantitative factors have the greatest influence on

the mortality in earthquakes. Their conclusion is that:

The highest mortalities are correlated with poorly developed, rural and semi-rural areas, whereas highly developed urban centres are the least vulnerable. (2005:22)

In their analysis, Gutiérrez et al (2005) included earthquakes in both developed

and developing countries. With “highly developed urban centres” (2005:22) they are

referring to urban centres in rich, i.e. highly developed, countries. Dense urban

centres in poor countries, and particularly the rapid process of urbanisation in those

countries, point to higher mortality – the opposite of the situation in rich countries.

The speed of urbanisation is hence an indication of the vulnerability to earthquakes in

poor countries.

- 33 -

Aleskerov et al (2005) developed a scenario-building model for the estimation of

structural damage, human losses and resulting need for external aid. Their model is

not intended for real time use and their baseline data were collected using

questionnaires sent out to district and sub-district government officials in Turkey.

With the exception of the baseline data collection, their methodology has got the

potential to be applicable world-wide and could be altered for real-time use. In the

development Aleskerov et al (2005) apply statistical cluster analysis to buildings

according to characteristics like use, age, predominant construction material and

number of stories. The model’s predictive output is a percentage of casualties for each

cluster or cell, e.g. three storey reinforced masonry structures built in the 1940s. With

knowledge of the number of occupants it is then possible to calculate the number of

casualties; as well as the number of individuals that will need shelter. The model of

Aleskerov et al (2005) is an example of when categorisation of low quality data can

enable useful analysis. The cluster based qualitative building data that forms the base

for their research is likely to be similar to the proprietary data controlled by

Shakhramanian et al (2000). Both these models show that it is possible to develop

prognostic systems with relatively rough data on the affected area. Badal et al (2004:1)

test an interesting model of “the relationship of the macroseismic intensity to the

earthquake economic loss in percentage of the wealth” in an effort to predict the

human as well as economic impact of events. The economic loss is measured by Badal

et al (2004) in the context of what they define as “social wealth”. The social wealth is

quantified using a function involving the national Gross Domestic Product (GDP) and

the cell population (see Equation 1). In their study, they used a grid of 4600 x 3500

metres; the finest available for their study region in Spain.

.Re)/( gionalcell GDPpulationregionalpotioncellpopulaGDP ×=

Equation 1 Badal et al‘s (2004:6) function for social wealth distribution

Something that the model of Badal et al (2004) have in common with all the

numerical loss estimation models is that they include proxy indicators on the quality

and characteristics of the built environment in the affected area, which, with the

exception of Shakhramanian et al’s (2000) proprietary database, is not available on a

global level. Another factor is the diurnal effect on human loss. Logically, losses

should be greater during night when people are asleep and take longer to react

(Alexander 2000b). However, the difference in vulnerability of buildings occupied

during day and night has to taken into consideration (Coburn and Spence 2002:104).

- 34 -

In developing countries the living accommodation may be more resistant to

earthquakes than the reinforced concrete structures used by, for instance, schools and

industry.

4.1.3 Needs assessment The extension of loss assessment into needs assessment does not seem to come

naturally for the research community. Lamontagne (2005) does not even mention the

possibility of providing such information in his survey of what information is useful

for inclusion in alerts to decision makers. As the estimated loss forms the basis for the

calculation of the amount of need (Shakhramanian et al 2000:146-160; McConnan

2000:180-187), the reports including data on need will not come available before the

reports on loss. Patterns of injury and need among the affected population have been

investigated by several groups (see for instance Alexander 1984; Coburn and Spence

2002:118). Calculations of quantified needs are often based on output from such

models, e.g. estimations of the number of homeless individuals are translated into a

need for shelter (Aleskerov et al 2005; Shakhramanian et al 2000:146-160).

The model for shelter needs prognosis developed by Aleskerov et al (2005) uses

a series of assumptions for reaching a number of persons who will be unwilling to

return to their original accommodation. This approach has potential in domestic

disasters. If one assumes complete knowledge of the loss in a given disaster, the

quantitative need is a function of the initial needs subtracted by the amount of aid

received in the area and the amount on its way there. In an ideal situation, by

knowing the actual losses sustained by the affected population and its coping

capacity, it would be possible to estimate the absolute needs before any external relief

process is initiated. Shakhramanian et al (2000) developed a needs estimation model

using this logic. However, de Ville de Goyet (1993) argues, that if the relief is not co-

ordinated and well-structured with regards to information sharing, the ability to

correctly estimate the actual need in the disaster area at an exact point in time

diminishes as relief arrives in the disaster area. Additional criticism of the logic

applied by Shakhramanian et al (2000) is that it assumes knowledge of local coping

capacity and centres on quantitative needs. In international relief scenarios the

estimation of need is much more complex.

Wijkman and Timberlake (1984) write that for experts in political sciences,

similar losses or physical effects in two separate countries with different economic

- 35 -

and institutional conditions could have very different implications. They continue to

write that an event that could pass relatively unperceived in a large country could

mean a catastrophe in a small one due to the differential absorption capacity of each

of the involved social systems. Similar damage in rich and poor countries have more

serious social implications in the poor countries, where the under privileged social

groups are usually the most affected. This highlights the requirement to model a

country’s ability to absorb disasters and deal with them in an appropriate manner

without external help.

4.1.4 Data quality Inadequate data quality is a major obstacle in disaster research (Alexander

2000a:36-39; Fischer 1998:37-87; Stallings 2002). The situation is not better for the

practitioners. Tsui writes that:

Just as the uncoordinated arrival of relief supplies can clog a country’s logistics and distribution system, the onslaught of unwanted, inappropriate, and unpackaged information can impede decision-making and rapid response to an emergency. (2003:50)

In his thesis of trade-off between information certainty and operational

effectiveness, Benini (1997) shows that complete certainty is difficult to achieve in the

implementation of an effective humanitarian intervention. Uncertainty is an integral

part of humanitarian operations in response to disasters. In relation to this, Keen and

Ryle state that:

The nature of contemporary disasters in Africa […] militates against the rapid collection of […] data. By the same token, reliable base-line statistics that predate the crisis are seldom available. Parties to conflict may attempt to manipulate information about the populations under their control; and relief agencies, in the rush for funding, may promulgate statistics that owe more to guesswork and imagination than to research. (1996:328)

To add uncertainty, the data in sudden-onset disasters change quickly. The

needs resulting from a sudden-onset disaster are not static. As the priorities change in

the disaster zone, so does the need for external relief. When relief items arrive, the

needs change, which increases the relevance of co-ordination (Tsui 2003; Dykstra

2003). A common problem with relief and needs data identified by de Ville de Goyet

(1993) is the inadequate use of technical specifications of dispatched and received

relief. As an example de Ville de Goyet (1993:170) mention a stereotypical report

concerning the reception of “a plane load from Country X with 15 tons of medical

supplies, food, tents and blankets”. De Ville de Goyet concludes that such coarse

- 36 -

statements are “totally insufficient for disaster management purposes” (1993:170).

Alexander (1995) gives a similar example from the 1988 Armenian earthquake where

pharmaceutical relief was labelled in foreign languages or not labelled at all; resulting

in two thirds of the relief having to be destroyed. There are no indications in

literature that these shortcomings of the information flow have changed. Only by

accepting uncertainty and incorporating tools for dealing with it in the information

systems can the situation be managed (Comfort et al 2004). As a conclusion of his

article on information uncertainty in humanitarian aid Benini writes that “Only he

who does nothing is certain” (1997:352).

This discussion has highlighted the requirement for the research project to deal

with uncertainty in the collected data. For this purpose, an analytic framework for

data quality is presented in section 5.3.2.

4.1.5 Usability Finally, information managers, practitioners and decision-makers should know and understand technology’s limits. Technology is a means to an end, and not an end in itself (Tsui 2002:20)

Walker (1991) outlines the factors required for making international SAR

response to an earthquake disaster cost-efficient; one being quicker responses.

Logically, early warning could enable teams to reach the areas where they are most

needed, sooner. However, as eluded to by Tsui (2002) in the above quote, early

warnings, particularly those based on approximate tele-assessments, can be a

disadvantage as well as an advantage. If an EWS is accurate, timely and able to

inform its users, the benefits are obvious. However, Glantz (2003:29) suggests that

over-reliance on inexact warning systems can increase community vulnerability. The

users come to expect timely warnings and reduce their readiness. Additionally,

competition between systems may cause ambiguous warnings and confusion among

the user community (Glantz 2003:33).

The message containing the tele-assessment must be, in the words of Glantz,

“designed for the special needs of specific users” (2003:30). For early warning

messages intended for decision makers in emergency services Lamontagne (2005)

shows that expanded warning messages combined with the use of maps in the

communication of the message is beneficial. Lamontagne mentions that:

- 37 -

A current misconception is that emergency managers always understand the meaning of the information sent by seismologists. This is not always the case, especially [in areas], where decades can separate damaging events (2005:396)

Simplistic warning messages, containing only seismic data, such as magnitude

and epicentre location, is hence seldom sufficient for emergency manager making a

decision about the response. The message must be targeted at chosen user to have

maximum effect. A system allows for this by including customisable functionality

that allows individual users to adopt the output to their specific needs. When

incorporating such functionality the tele-assessment system can be seen as a decision

support system.

4.2 Existing Decision Support Systems The development of decision support in disaster management is not a novel

concept. There are many systems in use by organisations worldwide; the most

common ones are reviewed here. The DSS are presented grouped according to their

main functionality.

4.2.1 Planning and Scenario building It is accepted that the planning and scenario-building systems are not in direct

competition to the research being conducted as part of this thesis. They are targeted

at different users and decisions. However, they are all central to the development of

DSS for disaster management over the last two decades and as such they provide

relevant knowledge of the flora of existing systems and their differences. These

systems are used in the pre-disaster phases by governments, industry and

international organisations. Most such systems have limited geographic scope, like

that developed by Mehrotra et al (2003), with baseline data requirements that make

their application impossible in a developing country context.

HAZUS and CATS Several US government agencies have independently produced computer based

tools for supporting emergency management. The tools are not intended to be

applied outside of the USA without major alteration. Before a recent amalgamation of

the major software tools, the US Federal Emergency Management Agency (FEMA)

endorsed two tools for disaster management. For natural hazards within the USA,

FEMA promoted the use of HAZUS993. The Consequence Assessment Tool Set

3 Now HAZUS-MH (Multihazard)

- 38 -

(CATS) was recommended for loss assessment of technological hazards, hurricanes,

as well as for earthquakes occurring outside the USA. Whitman et al (1997) describe

the loss estimation methodology in HAZUS99 developed by FEMA in partnership

with the National Institute of Building Sciences (NIBS). Using GIS technology,

HAZUS99 allows users to compute estimates of damage and losses that could result

from a hazard. HAZUS99 in its standard edition did not support Online Analytical

Processing (OLAP), which means that it could not provide support in real-time to

emergency responders (Schneider and Drury 1999). Research has been made in this

area and the HAZUS project is soon likely to offer OLAP functionality for events in

the USA (Kircher 2003). To support FEMA's mitigation and emergency preparedness

efforts, the version replacing HAZUS99, HAZUS-MH, has been expanded into a

multi-hazard (MH) methodology with modules for estimating potential losses from

wind (hurricanes, thunderstorms, tornadoes, extra tropical cyclones and hail) and

flood (riverine and coastal) hazards.

Swiatek and Kaul (1999) present the CATS as a powerful combination of tools

for assessing the consequences of technological and natural disasters to population,

resources and infrastructure. Developed under the guidance of the US Defence Threat

Reduction Agency (DTRA) and the FEMA, CATS provides assistance in emergency

managers' training, exercises, contingency planning, logistical planning and

calculating requirements for humanitarian aid. CATS contains models that predict

the damage and assesses the consequences associated with that damage as a result of

a technological or natural hazard (Swiatek and Kaul 1999). The natural hazard

portion of CATS provides for the calculation of damage and consequence from

earthquakes and hurricanes. The earthquake model is a collection of software

programmes that models the severity and the geographical extent of the damage due

to the primary earthquake hazard of ground shaking as well as to the collateral

hazards of ground failure, tsunami and fire following an earthquake. The

consequence of a damaging earthquake is assessed in terms of the facilities,

infrastructure and population at risk.

HEWSWeb The Interagency Standing Committee (IASC), a mechanism organising key UN

and non-UN humanitarian partners (IASC 2006), is the host of the Humanitarian

Early Warning Service (HEWSWeb) (HEWS 2006). HEWSWeb is an information focal

point with focus on slow-onset natural hazards. An expansion into human-made and

- 39 -

sudden-onset disaster is planned for 2006. The site is based on the early warning

information produced by a variety of specialized agencies and institutions.

HEWSWeb does currently not generate its own information.

RADIUS The Risk Assessment Tool for Diagnosis of Urban Areas against Seismic

Disasters (RADIUS) was developed as part of the International Strategy for Disaster

Reduction (ISDR) and the International Decade for Natural Disaster Reduction

(IDNDR). Okazaki (2000) presents the project as an initiative to reduce earthquake

disasters in nine case studies of cities through support to mitigating and preparedness

efforts. The RADIUS tool is raster-based GIS built as a plug-in to Microsoft Excel.

Each cell in a spreadsheet represents a pixel in the raster (Okazaki 2000:32). Okazaki

writes that:

The tool requires only simple input data and will provide visual results with user-friendly process [sic.] with help and instruction documents. (2000:31)

Although this might be true, the simplicity of the tool limits its scope to coarse

pre-disaster risk assessments. Furthermore, the spatial data prerequisites, e.g. soil

types, lifelines facility distribution, limits the possibility of using the tool for tele-

assessment.

4.2.2 Real-time alerts There are several real-time, OLAP systems, for decision support in earthquake

response in operation. The systems are reviewed here with the aim of clarifying how

others have approached the task of providing alerts following earthquakes.

Global Disaster Alert and Coordination System The Global Disaster Alert and Coordination System (GDACS) grew out of the

Digital Map Archive (DMA) alert tool which was developed by Dr. Tom De Groeve

and Dr. Daniele Ehrlich at the Joint Research Centre of the European Commission.

The original application, called the ‘DMA Earthquake Alert Tool’ aimed to provide

the decision maker with “systematic, reliable, and objective estimate of the affected

population […] within hours after the event” (De Groeve and Ehrlich 2002:4). The

research on the subject has since progressed to include other hazards and different

types of prognostic output. The original loss estimation model was based on the

population density, the country vulnerability and the magnitude of the earthquake.

After requests from the users, a decision was taken to introduce a qualitative output

from the prognostic model. The development of the qualitative output is part of this

- 40 -

PhD research project and the research grant provided by the European Commission

was for research in support of the GDACS tool. The sequence of generations of the

GDACS tool developed by De Groeve and Ehrlich (2002) was evaluated by the

researcher while stationed at the JRC and published as an internal report: De Groeve

and Eriksson (2005). In its current version the tool provides a three-tier alert

following an earthquake (see Table 4.1) (De Groeve and Eriksson 2005:7). The

functionality of this sequence of models is presented below. The logic and

motivations behind the choice of methods and numerical cut-offs of De Groeve and

Ehrlich (2002) are not analysed herein because the research did not have such insight

into the development of the early models later evaluated by himself and De Groeve in

Table 4.1 Alert levels, scores and severity

Alert Level Alert Score Severity Red (3) >2 High

Orange (2) >1 and <=2 Medium Green (1) <=1 Low

Source: De Groeve and Eriksson 2005:7

The underlying algorithm for this system has undergone several

transformations, but is based on loss estimation. The first three generations of the

algorithm are described in the evaluation by De Groeve and Eriksson (2005). In the

most recent generation the qualitative severity is estimated using a set of functions

(described in Equation 2 and Table 4.2) each producing a quantitative output between

zero and three (see Table 4.1; De Groeve and Eriksson 2005:7).

cba VPMT ××=

Equation 2 GDACS alert level function4

4 De Groeve and Eriksson (2005:7)

- 41 -

Table 4.2 GDACS sub-functions

Indicator Function Quantitative alert level (T) T = alert level Earthquake magnitude (M) )0;5.4max( −= trueMM

Vulnerability (V) V = the Global Needs Assessment index if available, otherwise the default vulnerability of 0.6

Population (P) )0);

80000(max(log 100

Weights (a, b, c) a = 1, b = 0.5 and c = 1.5 Source: De Groeve and Eriksson 2005:7

In the current model, the affected population is calculated for a 100 km radius.

Only in cases where the population exceeds 80 000 will the function result in an alert.

A logarithm is applied to quantify the population approximately between zero and

three (see Table 4.2). De Groeve and Ehrlich (2002) determined the weights in Table

4.2 through calibration against past events that had resulted in an international

financial response. The resulting draft score T* in Equation 2 is then converted to a

final alert score through a set of filters: Red alerts are limited to earthquakes with a

magnitude above 6; the final score of an intermediate depth earthquake is reduced by

1 and the final score of deep earthquakes is set to zero, thereby effectively ignoring the

magnitude, population and vulnerability completely. The depth is classified as

shallow (up to 100 km), intermediate (up to 300 km) and deep. This results in a final

alert value that is translated into an alert on Table 4.1 and broadcasted to the

registered users via SMS and e-mail.

PAGER The USGS tool for Prompt Assessment of Global Earthquakes for Response

(PAGER) was launched in 2005 (Earle et al 2005). In summary, Earle et al describe

PAGER as a system that will:

distribute alarms via pager, mobile phone, and e-mail that will include a concise estimate of the earthquake s impact. The alarms will also report the earthquake location, magnitude, and depth, an estimate of the number of people exposed to varying levels of shaking, a description of the region’s vulnerability, and a measure of confidence in the system’s impact assessment. Associated maps of shaking level, population density, and susceptibility to landslides will be posted on the Internet. This information will be available within minutes of the determination of the earthquakes location and magnitude. (2005:1)

The PAGER system is the most recent addition to the set of alert systems that

offer global coverage. It builds on the existing USGS Earthquake Notification System

(ENS). Like the ‘Russian family’ of systems, described below, PAGER provides an

- 42 -

output with an intensity raster with the difference that PAGER only produce radial

attenuations for earthquakes outside the USA. With the intensity raster PAGER

estimates the number of persons in pixels expected to experience each level of

intensity (see Plate 4.1, Earle et al 2005, on page 45). Added to the maps is a chart

showing the ‘population exposure’ to the event (see Plate 4.2, Earle et al 2005, on page

45). This is one step short of the ‘Russian family’ which approximates the number of

casualties based on the vulnerability of the buildings in nearby urban areas.

The ‘Russian family’ A set of alert systems with an unclear common origin in the Soviet civil defence

institution, which today is the Ministry of Russian Federation for Civil Defence,

Emergencies and Elimination of the Consequences of Natural Disasters (EMERCOM),

has been enhanced by at least two separate organisations: the Extreme Situations

Research Centre (ESRC) and the World Agency of Planetary Monitoring and

Earthquake Risk Reduction (WAPMERR). Due to outright polemics between the two

organisations, concerning among other issues the ownership of the baseline dataset,

there is a relatively large amount of secrecy surrounding the baseline data and

functioning of the system. According to Wyss (2004a) the tool was originally

developed by staff members of the ESRC in Moscow. The publication authored by

Shakhramanian et al in 2000 was made as part of the original development of this

system now referred to as ‘Extremum’ by the ESRC. The system incorporates a

database of the world's population distribution coupled with categorised

characteristics of the built environment. It is claimed that the structural data are

categorised in a similar way as that applied by Aleskerov et al (2005) in their study of

a Turkish city. However, Aleskerov et al (2005) collected their data through detailed

interviews and questionnaires, whereas Extremum incorporates that data aggregated

on a city-level. The spatial data are stored in point format for 1.2 million ‘populated

places’ all over the world (Frolova 2006). The estimations made by the system include

the spatial distribution of human losses and structural impact classified in five

categories. The baseline database is fundamental to the tool’s ability to calculate the

structural as well as human losses incurred by earthquakes. The estimations can be

done in real-time on cases as they occur or beforehand on scenarios (Wyss 2004a).

One version of the tool is now hosted by WAPMERR. The organisation was

created in 2001 in Geneva, Switzerland, “as a non-profit organization for the purposes

of reducing risk due to disasters and for rescue planning after disasters” (Wyss 2006).

- 43 -

WAPMERR is marketing the tool under the name ‘QUAKELOSS’. As the name

reveals, QUAKELOSS focus on providing loss estimations for earthquakes. Plate 4.3

on page 45 gives an example output from the QUAKELOSS system made in real-time

following the devastating earthquake in Pakistan. The QUAKELOSS system relies

heavily on traditional channels of communication. Table 4.3 shows the information

flow following the 2005 Pakistan earthquake between Wyss and a Swiss international

SAR organisation.

Table 4.3 QUAKELOSS alert process for the 8th October 2005 earthquake in Pakistan

Message 1: Telephone call Date: Sat, 8 Oct 2005 04:20 GMT From: Max Wyss Subject: earthquake in Pakistan "A very serious disaster has occurred in Northern Pakistan" Message 2: E-mail Date: Sat, 8 Oct 2005 04:32 GMT From: Max Wyss Subject: earthquake in Pakistan An earthquake with the following parameters has occurred: 08Oct2005 03:50:38.6 34.4N 73.5E 10 M =7.6 M*NEI PAKISTAN A large shallow quake in this location is a serious disaster. We estimate that thousands of fatalities may have occurred and the injured may be 10,000 or more. Message 3: E-mail Date: Sat, 8 Oct 2005 04:40 From: Max Wyss Subject: Pakistan earthquake "The cities most affected in today’s earthquake in Pakistan are: Baffa and Abbottabad". Message 4: E-mail Date: Sat, 8 Oct 2005 04:52 From: Max Wyss Subject: Pakistan earthquake "The attached map shows the average damage in the settlements in N. Pakistan due to today’s earthquake as estimated by QUAKELOSS." [see Plate 4.3]

Source: Wyss 2006

Wyss applies predictions with wide intervals as a means to deal with

uncertainty. An example of this is a real-time prediction of human loss following an

earthquake in Iran (Wyss 2006, accessed February 2006) to be between 410 and 20 000

people. Wyss (2004a) claims that his loss estimates are “acceptable” in 92% of the

cases. His testing methods and his definition of “acceptable” will be examined in

section 9.2.3.

Comparison The system outputs and the methods applied to arrive at the output differ from

one another, but the systems are either fully automatic or require the involvement of a

human expert. The characteristics of the tools are summarised in Table 4.4. These

alert tools will be examined further in section 9.2.

- 44 -

Table 4.4 Global coverage earthquake alert systems

Tool Approach Inherent Baseline Data Output Disaster Alert Tool (JRC)

Simple spatial arithmetic

Demographics Alert level

Russian family Expert enhanced spatial analysis

Demographics, Building quality

Building loss; Injured and dead; Intensity field

PAGER Spatial arithmetic using shake-map

Demographics Affected population with shaking intensity

Adapted from De Groeve and Ehrlich (2002), Earle et al (2005), Shakhramanian et al 2000

- 45 -

Source: Earle et al 2005

Plate 4.1 PAGER graphical output for the 24th February 2004 earthquake in Morocco

Plate 4.2 USGS PAGER numerical output for the 24th February 2004 earthquake in

Morocco

Source: Wyss 2006, accessed April 2006

Plate 4.3 QUAKELOSS graphical output for the 8th October 2005 earthquake in Pakistan5

5 The colour scale gives the expected damage state (intensity) with blue resulting in light damage and black in “total collapse”. The size of the circles indicates settlement size. No legend is available for the settlement sizes.

- 46 -

4.2.3 Co-ordination Coordination is a commonly discussed subject confused by the various assumptions about its meaning. To some it implies the sharing of information; to others coordination implies centralised decision-making. The implication is that a common understanding must exist between the parties involved (Kent 1987:161)

The importance of effective co-ordination in international relief is mentioned

repeatedly in literature (see Benini 1998; Comfort et al 2004, Dalton et al 2003:34pp,

Walker 1991, 1995; Zimmerman 2002). Consequently, there is a need for decision

support to co-ordination. As implied in the quotation from Kent (1987) above, the

term ‘co-ordination’ is ambiguous. Depending on the organisation of the relief

mission, the co-ordination and its DSS, can be focused on information sharing or

centralised control of resources. Benini (1998) argues that both types of co-ordination

in international relief to sudden-onset disasters are possible to achieve without one

organisation taking the official lead. This is not to say that a laissez-faire situation is

preferable. On the contrary, Tsui (2003) sees the appointment of a response

figurehead as important for efficiency, at least within the UN domain. The potential

of a leaderless situation does, however, mean that DSS cannot be developed with a

single user in focus. The issue of co-ordination is not only one between individual

organisations; it is also relevant within large organisations. In their development of

the Interactive, Intelligent, Spatial Information System (IISIS), a prototype DSS for

disaster management for use in public organisations, Comfort et al (2004) study

internal co-ordination processes. Comfort et al (2004) conclude that a DSS aiming at

enhancing internal co-ordination needs to have the ability to accommodate for

changing requirements on information amalgamation as the system output moves up

in the decision hierarchy. O’Brien (1999) and Kersten (2000) support this stance in

their generic recommendations for the development of high-level DSS.

On the international earthquake relief scene the only operational inter-

organisational co-ordination system is the Virtual On-Site Operations Co-ordination

Centre (VOSOCC). The VOSOCC is similar to HEWSWeb in that it enables “real-time

exchange of practical information related to emergency response” (OCHA 2006).

More on the VOSOCC and co-ordination efforts of its host organisation, the OCHA, is

presented in section 8.2.

- 47 -

4.2.4 Trends The current trends in early warning is towards multi-hazard monitoring

(Westervelt and Shapiro 2000; Zschau and Küppers 2003) and early warning cascades

(Glantz 2003). This does not mean that a single system would handle all warnings,

but that systems would interchange data and output information in order to improve

their own output (Comfort et al 2004). There is also a general trend in the academic

community towards the support of disaster mitigation, which has been know for quite

some time (Walkers 1991) to be a more cost-efficient alternative to response measures.

This is, however, not always reflected among funding organisations. Although he

admits that initial funding for the creation of EWS often is available, Glantz (2003)

mentions that funding for the equally important long-term maintenance of EWS is

much scarcer.

In the specific case of DSS for international earthquake relief, Wyss (2004) sees

the optimal trend as being towards: improvement of the ‘last mile’ alerting6,

improvement of hypocentral depth estimations, increased use of image remote

sensing for loss assessment, faster delivery of loss estimates, improvement of global

spatial data on building stock, improvement of global spatial data on soil

characteristics and development of regionally dependent attenuation functions for the

representation of earthquakes.

4.3 Summary In the early stages following a potential disaster, a range of methods of tele-

assessment are available for the provision of support to the remotely located decision

maker. The methods are intended to provide early warning to the relevant decision

makers of events resulting in excessive losses or needs for external relief. Some of the

methods are currently used in active DSS. The focus of this study is on DSS that, in

real-time, alerts remotely located decision makers of sudden-onset events potentially

requiring their attention. This research project builds on the achievements of a system

providing such functionality, the GDACS. A DSS successful in this task must

incorporate methods for maintaining accuracy and usability of the output in spite of

insufficient and uncertain input data.

6 In the case of public alerts, this is the task of reaching all individuals in danger and having them perform actions to reduce their exposure to an oncoming hazard.

- 48 -

5 RESEARCH PLAN

5.1 Research Approach This chapter explains the overall organisation of the study and motivates the

choice of research methods.

5.1.1 Philosophy The research project is heavily influenced by the methods developed by De

Groeve and Ehrlich (2002). The main reason for this is that the research funding was

provided to build upon their results which were made with positivistic modelling. In

discussion with the funding organisation it was decided that the research should have

two sequential steps. The first step was to investigate the requirements posed on the

alert system by the users and to determine how these requirements currently are

fulfilled; the first objective of the thesis. The second step was to develop a prototype

tool that better targets these requirements of the user community; the second objective

of the thesis. The development would benefit in the achievements of the GDACS tool

in the search for a novel method and concept of alerting. A quantitative positivistic

approach to the modelling of the processes surrounding the international responses to

earthquakes is necessary for the development of predictive models.

In contrast to the predictive model development, the determination of the user

requirements in the first step represents a superficial ethnography of the international

relief community that is developed with interpretive intentions using qualitative

methods. Even though quantitative and qualitative methods are used, the

overarching methodology is not triangulation as described by Blaikie (2000:262) or

Tashakkori and Teddlie (1998).

In the creation of a prototype tool, the main intention is to develop the model

using inductive methodology; creating a theoretical model through observation of

activities in reality. The intention is to probe the suitability of using a model created

in this way to predict future actions of the international community.

One section of the research assumes what can be seen as a normative stance.

Behavioural patterns are identified, analysed and discussed with the purpose of

evaluating the appropriateness of the patterns; both in terms of their practical

suitability for inclusion in a prognostic model and for their morality. This analysis

cannot be made without a certain degree of subjectivity and judgment as to what is

- 49 -

practical and what is morally just. Should it be evident that international attention is

predominantly given to events where there is no need for external support, it would

not be suitable to provide alerts in future events based on the past.

Epistemologically, this research project was born out of computer science. Over

its life it has gone through a series of metamorphoses where subjects like remote

sensing, information systems ontology, earthquake engineering, seismology and

statistics each have played central roles. This process has formed what has become a

truly epistemologically fragmented project much in line with what is common in

disaster management research (see for instance Alexander 2000a:35). Repeated

attempts were made to focus on one science, but such limitations consistently

prevented the aim of the thesis to be achieved. The final approach is to use numerical

modelling informed by the sciences of seismology, earthquake engineering and socio-

economics.

5.1.2 Research design The research followed a non-linear path (Neuman 2000:124) often associated

with qualitative research. An investigation of the relevance of the research in the first

year of study concluded that the project was not heading in the right direction if the

end goal was to produce results that could be implemented in a user organisation.

Consequently, the research changed paths in its second year, with a new focus that

required revisiting several phases of the research plan; thus creating an iterative

process. Although the problem and the potential solutions were determined by the

start of the second year of research, the iterations continued well into the third year.

Nevertheless, the research is presented in a sequence.

Information systems development cycle The adopted research structure is based on the information systems (IS)

development cycle. O’Brien (1999:92) describes the IS development cycle as a process

for solving problems in organisations by applying solutions with an IS component;

which is what this study aims to do. O’Brien divides the process into four stages: (1)

Systems Investigation, (2) Systems Analysis, (3) Systems Design and (4) Systems

Implementation. In the systems investigation stage he sees the task of the investigator

as being to determine if a problem exists and to establish whether it is possible to

solve the problem with the resources at hand, i.e. the first objective of this study. In

the systems analysis stage, more detailed requirements of the functions and output of

- 50 -

the proposed solution are gathered from the potential users and from the context in

which the proposed system will work. The systems design stage determines which

exact data, information and tools are required to provide the output and functionality

requested by the users. Together, the systems analysis and systems design stages

cover the second objective of the study. Finally, the systems implementation stage is

where the proposed solution is developed and tested, i.e. the third objective of the

study.

Knowledge Discovery in Databases process Although the recommendations provided by O’Brien (1999:92) provide an all-

encompassing structure to the research project, they are crude with regards to the

specific domain of the research project. To remediate this, a process model better

targeted at the development of a DSS is adopted. The Knowledge Discovery in

Databases (KDD) of Mahadevan et al (2000) is a method for the development of DSS

and expert systems, created with applications in the domains of sustainable

development and international relief in mind. As such it fits the needs of this research

project well and its structure is adopted to provide additional support, particularly in

the implementation stage of the IS development cycle.

The KDD process relies heavily on the developer’s knowledge of the studied

problem domain in order to extrapolate models from databases. As indicated by

Figure 5.1 (Mahadevan et al 2000:345), the KDD process is both interactive and

iterative to a greater extent than it is sequential.

Source: Mahadevan et al (2000:345)

Figure 5.1 The ‘Knowledge Discovery in Databases’ process

Mahadevan et al’s model start with a problem definition phase where the intention

is to “obtain an understanding of the application domain, specify the expected

outcomes of the process (user goals and expectations) and define the domain […]

knowledge that might be needed” (2000:346). The subsequent data selection process is

the phase in which Mahadevan et al (2000:346) claim that the domain knowledge of

- 51 -

the developer has the greatest importance. Knowledge of the domain helps to prevent

the inclusion of pairs of variables with false correlations. The purpose is to select

appropriate elements, both in terms of variables and samples, from the database. The

data selected for further processing might be incomplete or unsuitable for processing

in its raw format. The data standardisation phase in Figure 5.1 targets these issues and

assists with the provision of a model with greater explanatory power. Problematic

variables are transformed to enable data mining. The data mining phase is the heart of

the process where the knowledge is discovered. Resulting models are then tested in

the model evaluation & testing phase.

The relation between the IS development cycle and the KDD process in the

research design is described below using the IS development cycle as a

superstructure.

5.1.2.1 System investigation stage The purpose of the system investigation stage is to determine the problem in

existing information flow surrounding the international response to sudden-onset

disasters; equal to the purpose of the problem definition phase in the KDD-process.

Due to the number of stakeholders in the research project, the problem

definition was not clear from the start. Instead, problems in the decision process of

the international relief efforts were identified through informal interviews with

practitioners. As the goal to please all stakeholders was abandoned at the end of the

first year of research, the complexity of defining a common problem was reduced.

Nevertheless, even with an identified user that the research project could focus on, the

purpose and application domain of the tool remained open. Questions that remained

open included which decision in the decision process should be supported, which

types of natural hazards should be covered, which geographical area should be

considered and how should the decision be supported?

As part of the system investigation stage, the existing workflows in potential

user organisations were clarified through interviews with organisation members,

participatory observation and content analysis of official documentation. The analysis

of the international relief workflow can be likened to a shallow ethnography of

decision maker context. The investigation mainly unveiled how the organisations are

supposed to operate in that both the formal and informal interviews provide an

indication of the organisation member’s perception of the workflow. The workflow

- 52 -

includes standards for when and how a response to a disaster should be mounted.

These delimiters are defined by Billing and Sieber (2003) as entry decision “triggers”.

5.1.2.2 Systems analysis stage The purpose of the systems analysis stage is to find potential solutions to the

problems identified in the preceding stage and match the solutions against the

requirements of the potential users. This phase is included in the problem definition

phase of the KDD-process.

The proposed solution should fit its organisation and not vice versa (O’Brien

1999:630). Attempts to use the introduction of new information system as a means for

forcing change in its organisational context are rarely successful and should be

avoided (Eriksson and Stanojlovic 2000). A range of solutions were explored for all

identified problems. The systems analysis is merely a preliminary investigation of

what was feasible to achieve with the tools and resources at hand. Over the first year

of research, these solutions consistently included tools based on information systems

ontology for the interconnection of heterogeneous databases and remote sensing of

natural hazard impact on the urban environment. Existing decision support methods

were scrutinized based on their benefits and disadvantages to the decision maker.

Common limitations included the timeliness of decision support if applied in a live

situation, for instance the real-time availability of remote sensing imagery, the benefit

of the expected output on the decision process, the cost in time to develop a tool and

in financial terms to run it and the availability of the data required to develop and test

a live version of the tool. As the system analysis matured the use of numerical

modelling took over as the central method.

5.1.2.3 Systems design stage With a defined problem and a set of prototype solutions determined, the project

proceeded to select methods for developing and testing prototype numerical models.

This stage is covered by the data selection and data standardisation phases of the

KDD-process.

When the project reached this stage it was clear in which decision that the

project outcome should support the user. The requirements on the tool by the user

were also known. However, due to the explorative nature of the research project, it

still was not clear what baseline data would be required to develop a model that could

provide the requested output. The patterns in the data were unknown, though there

- 53 -

were some expectations on logical relationships for instance between the volume of

the international relief and the number of dead and injured. The subsequent systems

implementation was planned to include a data-mining component that would require

a large number of variables to be collected. To avoid restrictions in the model

development that would follow, the data collection had to be wide and inclusive. To

enable such detailed case studies, the project could no longer have a global scope. It

would be impossible to collect all data on all international interventions to natural

hazards; therefore, a geographic region and a natural hazard were selected for a

focused study. At this stage the research philosophy changed to become positivistic.

With knowledge of the end goal and resources at hand, the research process

correspondingly transformed to become structured and sequential.

5.1.2.4 Systems implementation stage It is in the systems development stage that the KDD-process provides added

support to the research. The data mining phase is the heart of the KDD process. Here

the selected data are analysed using methodologies that are categorised by

Mahadevan et al (2000) as being: Predictive, Descriptive, or Prescriptive. Predictive

modelling is used to develop models that predict the future behaviour of some entity,

which is the intention of the research project. The detailed substructure of the KDD

process adopted in this stage is presented in section 5.3.2 ‘Quantitative data analysis’.

5.1.2.5 Summary Figure 5.2 summarises all the applied process structures and their position in

the overall research process. The three levels do not duplicate each other. Instead,

they provide added structure to the research process. The most detailed adopted

process is that of Hosmer and Lemeshow (2000), which will be presented in section

5.3.2.

- 54 -

Source: Author

Figure 5.2 Applied research process models in relation to the thesis objectives

5.1.3 Methods and sampling Formal and informal interviews as well as participant observation were applied

to target the first objective of the thesis: to clarify the user requirements of the

international relief community. The output from those interviews constitute the sole

instance of primary data in this study. Case studies and content analysis were applied

to target the second objective, but became central in the third objective. The case

studies were all historical events and content analysis targeted documentation and

running records relating to those events. The research on historical cases recorded in

archived documents can be seen as a case of historical-comparative research (Neuman

2000:397). Considering this, although not recognised with a separate heading below,

the research used historical-comparative methods. In the later part of the model

development, once a foundation of data had been created, quantitative analytical

methods were applied on the data to search for patterns. Those methods are

presented separately in section 0.

Semi-structured telephone interviews were conducted with two individuals. As

part of the first thesis objective, the purpose of the interviews was to ascertain the

expectations and requirements on decision support posed by its potential users. The

sampling of those interviewed was purposive in that both interviewees were involved

in a decision process that could benefit from an alert system. In other words, the

individuals were selected because they, in their positions, are faced with the decision

on whether or not to respond to international disasters and, if so, to what extent.

- 55 -

Those interviewed were Mr. Per-Anders Berthlin, Senior Advisor on Overseas

Operations at the Swedish Rescue Services Agency (SRSA) and Mr. Fidel Suarez,

international emergency manager with the Spanish rescue service’s canine unit.

Further agencies, including the German Techniches Hilfswerke (THW) and the British

Fire Services Search and Rescue team, were contacted with requests for interviews,

but they did not respond.

Several follow-up interviews were made with Mr. Berthlin over the phone as

well as in meetings. Informal interviews took place with additional users at

practitioner conferences and meetings as part of the participant observation. The

interviews were conducted using the time-line (Thomas et al 1998:140-142) of pre-

disaster and emergency events to provide a loose structure (see Figure 2.1 on page 7).

As mentioned above, the interviews with Mr. Suarez were limited in time due to them

having to be performed through the use of interpreter. Mr. Suarez’s position as an

operational manager made him a secondary user of alert tools and therefore not well

informed on the user requirements and workflow surrounding the international

interventions.

Although only two persons were formally interviewed, numerous additional

encounters that helped guide the development of a set of user requirement were made

under less formal circumstances. This less formal approach, which can be seen as a

participant observation or as informal interviews, was used in the first and second

objectives to investigate how the relevant decision makers in the European

Commission, the United Nations and the Swedish Rescue Services Agency (SRSA)

undertook their daily work and to collect data on events that occurred during the

study. The observation took many forms; meeting notes were saved and

correspondence between DG JRC and the users of the GDACS tool were saved as a

source of user requirements. The researcher took part in a series of conferences on the

subject of disaster alert systems where stakeholders, including practitioners in the

international relief community, were present. These events were: the 2002 United

Nations Office for Outer Space Affairs (UNOOSA) conference on the use of space

technology for disaster management in Africa, the 2003 Wilton Park conference on

improving the international relief to disasters, the United States Geological Service

(USGS) conference on earthquake alert tools in Boulder 2004, the 2004 UN/European

Commission user conference on the GDACS, the GDACS follow up technical

- 56 -

conference in 2005 and the 2006 conference on Information Systems for Crises

Response and Management (ISCRAM).

During the above meetings, to provide insight in a co-ordinating organisation,

contact was made with Mr. Thomas Peter of the FCSS and Mr. Craig Duncan of the

OCHA. Additional information on this co-ordinating body was gathered from

organisation documentation available on Reliefweb.

The European Commission is studied as an archetype of a funding organisation,

i.e. a donor. The study of the workflow in the EC, surrounding international relief to

sudden-onset disasters, is based on discussions with Dr. Peter Billing, a former head

of the sector for strategic planning within the ECHO organisation. These meetings

occurred through the period 2002 to 2005. Additional information has been extracted

from official documentation supplied by Dr. Billing and sourced through public

channels.

When combined, the interviews and participant observation provided a

substantial sample of the international earthquake relief community.

In addition, two longer observations were made. The first was a one-month

sojourn with the European Union Satellite Centre (EUSC) in Madrid. In Europe, the

EUSC is on the forefront of applied remote sensing imagery in political decision

support. The secondment with them provided knowledge of the advantages and

limitations of remote sensing as a source of information for decision making. It also

provided understanding of the requirements faced by an organisation running a

decision support system that is in use non-stop. The second observation was a one

month visit to the Sudan and Kenya. OCHA was starting a new branch in the Nuba

Mountains in the south of Sudan which included the set-up of field-based spatial

decision support systems. The stay was too short for the impact of these systems to be

evaluated. As the research project distanced itself from the use of remote sensing, the

benefits of this field visit became unclear. Nevertheless, the field studies can be seen

as having had a pivotal role in that they provided knowledge that changed the course

of the research project.

Post event case studies The central Asian region’s high seismicity (Lomnitz 1974:243; Khan 1991:65-66)

and the high vulnerability to earthquakes of its vernacular housing and infrastructure,

especially in poor areas as described by Coburn and Spence (2002:210-211), make it a

- 57 -

suitable choice for a case study. The map of Berz and Siebert (2004) in Plate 5.1

confirms the earthquake risks in the central Asian region and they are supported by

Dilley’s (2004) analysis of worldwide earthquake risk (see Plate 5.2 on page 58). The

risk is evident from the high frequency of strong earthquakes. The high frequency is

of benefit to the research project because it provides for a larger sample of events to be

used. In light of the risk experienced by the central Asian region combined with the

high number of historical events, the region was chosen for the case studies.

However, data on the historical events proved to be scarce before 1993, likely caused

due to the absence of the UN Department of Humanitarian Affairs. Post-1993, the

data availability was gradually increased with the introduction of the Internet.

Consequently, the case studies for this research are based on 59 earthquake events

that occurred in the central Asian region in the period from 1993 to 2005. The events

are listed in Table 5.1 and depicted on Plate 7.1 on page 105.

Source: Berz and Siebert 2004

Plate 5.1 Projected 50-year maximum earthquake intensity in central Asia7

7 In MMI, Red =>IX, Grey<=V

- 58 -

Source: Dilley 2004

Plate 5.2 Worldwide earthquake disaster risk hotspots

With the exception of the 2001 Gujarat earthquake, the host country is

determined by the location of the epicentre even in cases where the major impact was

in a different country. The central Asian countries considered in the study are

Afghanistan, Iran, Kazakhstan, Kyrgyzstan, Pakistan, Turkmenistan and Uzbekistan.

The Xinjang Uygur and Xizang/Tibet provinces of China are also included. The

Gujarat earthquake had its epicentre in India, outside the case study region. But data

showed that it caused considerable damage in Pakistan, which led to the decision to

include it in the study.

Table 5.1 Earthquakes studied by year and country

Country/Year

Afghanistan 0 1 0 1 0 2 1 1 2 3 1 1 0 13 China8 0 0 0 1 2 0 0 0 0 0 1 0 0 4 Iran 1 3 0 0 3 4 3 2 1 5 5 1 2 30 Kazakhstan 0 0 0 0 0 0 0 0 0 0 1 0 0 1 Kyrgyzstan 0 0 0 0 1 0 0 0 0 0 0 0 0 1 Pakistan 0 1 0 0 1 0 0 0 19 2 0 1 0 6 Tajikistan 0 0 0 0 0 0 0 1 0 2 0 0 0 3 Turkmenistan 0 0 0 0 0 0 0 1 0 0 0 0 0 1 Uzbekistan 0 0 0 0 0 0 0 0 0 0 0 0 0 010 Annual Sum 1 5 011 2 7 6 4 5 4 12 8 3 2 59

Source: Author

8 Only Xinjang Uygur and Xizang/Tibet provinces. 9 This epicentre was in Gujarat in India. 10 The 2003 earthquake in Kazakhstan had an impact on Uzbekistan, but no events of interest with epicentres in Uzbekistan were identified during the period. 11 No events of interest during 1995.

- 59 -

The cases were selected with the aim of including all earthquakes in the given

area and period for which non-seismic data existed in any one source. It is not a

sample for the region and period, but a complete population of events. Care has been

taken in order not to overlook any events. All events that have been mentioned in any

one of the non-seismic sources has been included. Seismic records provided a

reference in the initial identification of events and search of non-seismic data. The

data collection process started with a seismic event, consisting of a series of

earthquakes, first being identified in the NEIC database. Location and time data from

the NEIC were then used for querying all sources for data that could be linked to the

seismic event; such as losses, needs and dispatched relief material. The NEIC

database contains an abundance of seismic events. Most earthquakes that occur are of

low magnitude (Coburn and Spence 2002:19) which is reflected in the consulted

seismic databases. None of the earthquakes with magnitude in the NEIC database

below 4.5 resulted in any non-seismic records. Consequently, there are many low-

magnitude seismic events that could not be linked with any data on loss, needs, or

relief. Although no lower limit of the earthquake magnitude was set for inclusion in

the study, the earthquakes for which no data were found, apart from the seismic

characteristics, were not included. Non-seismic data linked to the seismic data were

deemed necessary in order to enable a fruitful analysis of the data. The assumption

here is that the lack of non-seismic data point either to there being no impact of the

event or that local assets successfully dealt with the situation. Seismic data on its own

could potentially be used to detect events that should have received attention but did

not. However, without primary earthquake impact data, there are few grounds for

conducting such analysis.

An additional challenge was to separate intertwined events when collecting

data. For events close in time or space it was sometimes impossible to judge to which

one out of two or more events that data referred. In other words, the determination of

when an event ceases and when a new event starts, in both time and space, posed a

challenge. The adopted solution was the use of a combination of the Global Identifier

number (GLIDE) (GLIDE 2006), the Emergency events database (EM-DAT) managed

by the Centre for Research on the Epidemiology of Disasters (CRED) and the indirect

grouping of events into series of sitreps made by OCHA on their Reliefweb website

(Reliefweb 2006). These sources issue identifier numbers for events. These numbers

were used to guide the decision as to which event to allocate certain data.

- 60 -

Content analysis The majority of the time invested in the research project was spent on content

analysis. The data collection process was largely made up by content analysis of

reports issued by local, national and international organisations involved in the

response to each of the 59 case studies. The collected reports were analysed and

classified, as described in detail in section 5.2, for frequency using manifest coding

(Neuman 2000:294). The content analysis provided the primary data for the

subsequent quantitative analysis.

The content analysis was made before it was certain what kind of support that

the decision makers should be provided with. The reasoning behind this was that the

quantitative data availability for the case studies would affect the range of models that

could be developed. Consequently, data were collected on a broad front, influenced

only by literature presented in section 2.3 and the existing DSS presented in section

The majority of the analysed reports were textual and quantitative data had to

be manually extracted from the text. Not all text could be fully digitised in this

manner. Due to the amount of collected media reports, those texts could not be

digitised beyond the earmarking with meta-data on source, release time and related

event. The referencing of reports to a specific point in time was complicated by the

reports not consistently including time of release and time zone. This was true for

media reports as well as reports from relief organisations and the UN. The problem

prevented the analysis of time sequence on an hourly resolution for most case study

events. Furthermore, news media reports, weather reports and concurrent event

reports were not fully analysed for content frequency due to the large volume of

reports. Instead, the reports were stored in the INTEREST database and linked to the

relevant earthquake event. This allows for the database to be queried for how many

reports that are linked to one event, but the contents of the individual reports of the

above-mentioned type cannot be quantitatively analysed.

5.1.4 Collaborations and external influences This project was born out of an idea developed by the researcher while working

with geographical DSS for the United Nations in the Balkans. The DSS used in

humanitarian de-mining have their roots in military applications (Kreger 2002). Their

military origins have provided a head start in terms of funding and prototype

- 61 -

applications development that have not been available in other areas of humanitarian

aid, such as Urban SAR (USAR) or refugee camp management. The researcher’s idea

was consequently to pursue research that promoted the use of DSS in a new domain

of humanitarian operations.

In competition with other research proposals the Joint Research Centre (JRC) of

the European Commission granted funding to the researcher for a three-year doctoral

research project on the subject of ‘decision support to senior decision makers in the

international relief to sudden-onset natural hazards’. An operational system, called

the Digital Map Archive (DMA) earthquake alert tool, later renamed to the Global

Disaster Alert and Coordination System (GDACS), was in development at the JRC (De

Groeve and Ehrlich 2002). The intent was for the doctoral research project to build on

that tool by increasing its usability in terms of accuracy, functionality and geographic

scope. The main area of collaboration was the development of user requirements.

Prototypes and live alert systems were tested by the GDACS development team and

this doctoral research project benefited from lessons learnt. The doctoral research

project and the GDACS project complement each other in that the generalist

macroscopic GDACS project can benefit from the results of the hazard and region

specific doctoral project and vice versa.

For the period during which the doctoral project received funding from the

JRC, it was based at the JRC headquarters in Ispra in northern Italy. The co-location

of the research project within the JRC enabled the observation of potential users and

provided the benefit of being embedded in the decision procedure that was the centre

of the research.

Academically, the research project started out in the domain of computer

science at Linköping University in Sweden. The first year focused on the

investigation of the potential areas of application and user requirements of a DSS of

this kind. The research ideas at this stage were centred on the use of remote sensing

and formal ontology as means of quickly providing decision makers with relevant

information following a potential disaster in a distant location. Two important field

trips were made at this stage. The first trip was a one month secondment to the

European Union Satellite Centre (EUSC) in Madrid. The second trip was a six week

observation of the use of DSS and early warning systems by the United Nations in

Sudan and the horn of Africa. At the end of year one, the conclusion of the user

- 62 -

requirement study and the evaluation of the remote sensing and information systems

ontology was that the stakeholders in the research project had colliding interests.

Radical change was required to bring the goals of the various stakeholders inline with

what was possible to achieve within the scope of the doctoral research project. As will

be discussed in this document, remote sensing is not the silver bullet to the problems

faced by the decision makers. Instead, a multi-disciplinary solution, in which

information technology only represents one out of an array of tools, would have to be

adopted for the study to provide a useful and tangible piece of research.

Consequently, the academic institution for the project was changed from the

Computer Science Department at Linköping University to the Disaster Management

Department at Cranfield University, UK.

The second and third years went into data collection and processing of case

studies. Several aspects of the research were presented and discussed at conferences.

Papers were presented at the European Seismological Commission annual conference

in Potsdam in 2004, the United States Geological Survey (USGS) conference on

international earthquake alerting in Boulder, Colorado in 2004, the United Nations

sponsored conference on early warning (EWCIII) held in Bonn in 2006 and the

international conference on Information Systems for Crises Response And

Management (ISCRAM) in Newark, New Jersey in 2006.

At the end of the third year the Disaster Management Department at Cranfield

University was closed and the project was faced with a second transfer. This time it

moved to the Disaster Management Department at Coventry University. The move to

Coventry also marked the end of the funding provided by the JRC, which meant that

the research project could move in its entirety to Coventry.

5.1.5 Research significance and relevance Part of the research project has been to investigate the relevance of initial alert

tools in the international relief process. A common opinion voiced in meetings and

conferences that the researcher took part in is that initial alert tools are too

approximate and complex to provide any useful input to the response decision and

that the effort of developing and maintaining an alert tool would be saved simply by

phoning people in the affected area and asking them of the need for international

assistance. These opinions can easily be debunked and evidence of the relevance of

an initial alert tool will be provided as part of this project.

- 63 -

In addition, the work leading up to the model will be enabling for future

research. The collected data can act as a foundation for studies of data accuracy,

international organisation co-ordination and international relief efficiency and

effectiveness. The qualitative study of the international relief process can help to

target future studies in the domain of international relief. The database and data

analysis software package developed as part of the study (see appendix A-4) can be

used for information flow studies following earthquakes in other regions, or for other

hazards. Over all, the study provides both the scientific community and practitioners

with new knowledge.

5.1.6 Ethical considerations The research has ethical considerations. First, some of the information received

in discussions with the relief community members has been provided off-the-record.

This information provides an interesting understanding of the current operating

procedures of organisations involved in international relief on all levels. However,

because the disclosure of this information could hurt the informants, it can not be

included. The research is forward-looking and not dependant on describing any

shortcomings in current operational procedures to which the disclosed information

relates. To avoid any mistakes, the individuals concerned have been contacted to

confirm that the information provided is correct and that it can be freely shared.

Second, participant observation was not one of the original methods of the

project. It was at a late stage that it became clear that the inclusion of the researcher in

the studied procedures had provided knowledge that otherwise would have been

missed. It was, however, never made clear to the members of the organisation that

the researcher’s experience would be included as part of the study. It is therefore not

ethical to include the actions of individuals without their prior consent. Such consent

has been requested and provided for the statements made in this thesis.

Third, the most important ethical aspect is the potential effect of the outcome of

the research on the actions of the international relief community. The development of

a model that predicts the actions of the international community could

unintentionally impose a status quo. If the international community paid too much

attention to the model output, they would continue responding to the types of events

and situations that they always have responded to without learning or adapting to a

changing environment. Even if the model engine is updated with up-to-date indicator

- 64 -

data, a status quo would create a dangerous situation in which a country or situation

that is not expected by the model to receive international attention will become more

vulnerable by the virtue of being seen as less vulnerable by the model. This

conundrum is avoidable if the users of an alert system use the system as a decision

support system that, rather than directing human decisions, supports them. In this

way the final decision is based on expert judgement that takes into consideration the

changes in the international relief environment. This approach also prevents the

decision support to result in increased vulnerability due to over-reliance on it in the

operations as described by Glantz (2003). A malfunctioning system should not

prevent the operations of its host organisation.

5.1.7 Assumptions The main assumptions in the analysis relates to the validity and accuracy of the

applied indicators. Some of the indicators of the characteristics of the case study

countries are static in time whereas the events are dispersed between 1992 and 2005.

This is the case for the 2005 GNA average, the 2004 World Press Freedom Index

(WPFI) and the 2003 Landscan. The assumption here is that the relative difference

among the case study countries has not changed much over the studied period. An

exception to this assumption is Afghanistan that, in particular with reference to press

freedom, has changed dramatically since the end of the Taliban regime in 2001

(Brossel 2002). A similar assumption is in place where national level indicators, such

as GDP, are applied on local events. Here the motivation for the assumption is

twofold. First, there simply are no complete sub-national data available to replace

national level data. Second, the national level indicator still gives an indication of

national resilience, i.e. the national ability to absorb the effects of an event.

In the development of a prognostic model of international attention, the most

central assumption is that the international attention can be quantified by the number

of sitreps issued by the OCHA. There are, or at least were, no strict guidelines for

when the duty officers in OCHA should release the reports or when to wait for more

information before releasing. It is therefore possible that whoever assembled the

sitreps could have had a ‘bad day’ or just missed a minor event and thus gave a

wrong impression of international attention. This problem is solved by categorising

the events according to the number of issued sitreps. The categorisation gives room

for occasional erratic behaviour of the official issuing the reports

- 65 -

The usefulness and relevance of the research project output relies on the user

not having instant access to analysed remotely sensed imagery. Currently, these

kinds of data are only available to the military. If politicians and humanitarian aid

workers had access to geo-stationary sub-metre resolution satellite imagery, there

would be less need for impact estimations. The impact could be detected from the

imagery and the international need could be estimated based on the impact. Such

satellite technology is still science fiction in the civilian domain and is likely to remain

so for at least a decade until the use of micro-satellites has reached maturity.

5.1.8 Limitations General limitations to the study include finances and time. Access to endless

resources would have enabled deeper analysis of more variables in more case studies

and thus increased the chances of finding patterns that could be implemented in a

DSS. Two field-trips were made in the initial phase of the project, but as the project

reached the analysis stage it became evident that the trips should have been better

targeted. On-site information gathering during or immediately after an international

relief mission to an earthquake could have enabled the project to develop more

precise models of current international relief requirements. Even if the case study

countries had been known at an earlier stage it would have been difficult to realise a

field-trip to an unfolding event in central Asia. The main limiting factors are the

reaction time, travel cost, language and the security situation in some of the studied

areas. In addition, in the cases where the United Nations was not co-ordinating the

international relief, it would have been difficult or impossible, to get insight into

national government activities without high-level contacts (Personal communication

with Dr. Nina Frolova, September 2004).

Data availability and quality In general, limited access to national records in the studied countries due to

language and in some cases non-existence of long term archives, increased the

project’s dependency on data published on the Internet. It also limited the project to

the study of events occurring after the 1980s. The increase of data available from the

early 1990s can be attributed to the introduction of the Internet, the restructuring of

the DHA and the granting of independence from the USSR of several of the case study

countries.

- 66 -

Unreliability or lack of accuracy is a problem when working with disaster data.

The unreliability issue is complex and relates to several aspects of the data; ranging

from definitions to data storage. The complexity and unreliability of disaster data has

been discussed by Albala-Bertrand (1993), Alexander (2000a:36-39; 1985), Fischer

(1998:37-87) amongst others. Correspondingly, all the data in this document suffer

from quality concerns. The study includes no primary data and must consequently

rely on secondary sources to provide an accurate and comprehensive data-set.

Inadequate resolution of the data limits the scope of the analysis. The digitised

reports are often lacking in detail both in terms of time and in reference to extracted

quantitative data. There are ways to ameliorate the impact of these issues on the

analysis, as will be presented in Chapters 8, 9 and 10.

Sample and Population of events In the analysis of case studies, the main limitations are the population size, i.e.

the number of case studies and the data availability for each case study. The intention

in this study was, however, not to use inferential statistics on a sample. Instead, the

central Asian region is chosen as a detailed case study for which all earthquakes that

left any international paper trail between 1992 and 2005 are included, regardless of

magnitude. The project is therefore a complete census of the earthquake events of

that region and period. This means that the data will be characteristic of the region

and consequently, the resulting predictive models cannot be expected to work on

future cases occurring outside of the studied region. It is also important to note that

events that did not leave an international paper trail are not included. This includes

low-magnitude earthquakes and some events occurring in the early 1990s when the

region was under the control of the USSR and before modern documenting routines

were in place. The limitations posed by the number of case studies become very clear

when the data are analysed in niche areas. An example is an analysis of the relation

between the donation of hospital tents and reported number of injured. This filter

would narrow down the number of cases to just one or two and make it difficult to

use the result for predicting the use of hospital tents in future events. A counter

argument is that the international community’s approach to dealing with earthquakes

in the developing world has changed continuously over the study period. Seeing that

the behaviour changed with time, one could argue that a greater window of time

would not improve the research results.

- 67 -

Only data that were deemed relevant for the final path of the research were

collected for events that took place after the initial systems design stage had been

completed in early 2003. With the exploratory analysis of suitable indicators being

completed, there was no requirement for the time-consuming wide and deep

collection of data. Events that are affected by this decision include the December 2003

earthquake in Bam, Iran and the five case studies following it in 2004 and 2005 (see

Table 14.1 in the appendix for a description of the cases).

5.2 Data This subchapter describes in detail how the case study data were collected and

prepared for analysis. It builds on the manifest coding presented in Table 5.4 in

section 5.2.6.

5.2.1 Data overview Although secondary quantitative data makes up the majority of the data used in

this project, the qualitative primary data collected through interviews and observation

in order to develop the user requirements plays an equally important role. The

classification into qualitative and quantitative should not be confused with

classifications into objective or subjective. Qualitative data can be objective and

quantitative data can be subjective. These atypical combinations are standard in this

thesis. Table 5.2 provides examples of how the collected data falls into the various

combinations.

Table 5.2 Classification of Qualitative/Quantitative versus Subjective/Objective

Qualitative Quantitative Objective Some of the data provided by Berthlin in his

interview is objective. For instance, the purview of the decision makers.

GDP, population, urban growth and earthquake magnitude are all examples of objective quantitative measures. They can, however, be seen as subjective because of, for instance, the method or baseline data used to produce the measurements.

Subjective The deadline for the provision of decision support provided by Berthlin in his interview is not objective. It is likely that other stakeholders would give a slightly different estimate.

Much of the data used in the development of the predictive model fits in this category. Although the measurements themselves can be objective, their application as proxy indicators may make them subjective. Examples in include the Vulnerability, the use of earthquake frequency as an indicator of Exposure, or the use of OCHA Situation Reports as an indicator of international attention.

Source: Author

The division of data into objective or subjective is not as clear cut as the division

into qualitative or quantitative. For instance, although the frequency of earthquakes

- 68 -

that a country has experienced over the last twenty years is an objective and

quantitative measurement, its use as an indicator of exposure makes it subjective.

The classification of collected data into primary or secondary, adds yet another

dimension. In this it is, however, only the data collected through formal and informal

interviews for the purpose of developing the user requirements that are primary. All

the data used to develop the final model are secondary.

5.2.2 Data types The purpose of this section is to review some existing disaster data structures

that influenced the data collection in the research case studies. The sought data

structures include both taxonomies of domain concepts as well as the relations

between information entities. The initial intent was to map existing research on formal

ontology for information systems (Guarino 1998) in the disaster management domain.

Existing research on the subject did, however, prove scarce, which increased the risk

of abusing the term ‘information systems ontology’ in the way Guarino describes it:

In some cases, the term “ontology” is just a fancy name denoting the result of familiar activities like conceptual analysis and domain modelling (1998:3).

No source contained reference to all the studies categories of data. Disaster data

are instead presented here in the six distinct categories for which individual

references were found: vulnerability data, loss data, relief data, needs data and

contextual data. These sections will form the basis for the later development of

taxonomies for this research project.

Because the hazard of interest is earthquakes there is no need to review existing

taxonomies of hazards in general. Instead, characteristics of earthquakes are

presented in section 5.3.

Vulnerability data Both ex ante and ex post vulnerability data are characterised by indirect proxy

measurements. These aspects are hard to quantify and seldom exist with a spatial

resolution down to a settlement level. The dynamic pressures and unsafe conditions

listed by Wisner et al (2004:51) include some examples of proxy indicators of

vulnerability, such as rapid population change and increased arms expenditure (see

Figure 2.2, page 10). In Table 2.1 Schneiderbauer and Ehrlich (2004) investigated how

vulnerability relates to different natural hazards and identified sets of indicators that

can be collected before a disaster strikes to form an estimate of the vulnerability of the

- 69 -

affected population. Their parameters for earthquake hazards are: Quality of and age

of building, Size of building, Location of building, Preparedness, Hygiene and

Vaccination. Both Wisner et al (2004:277) and Schneiderbauer and Ehrlich (2004) point

out that data on the micro-geography of the affected settlements, i.e. location of

building, provide an important indication of the local earthquake vulnerability.

However, such baseline data are rarely available for developing countries (Currion

2003). In lieu of a better alternative, composite proxy indicators of macroscopic ex ante

vulnerability are often applied (Badal et al 2005; Albala-Bertrand 1993). An example

of this is the Global Needs Assessment (GNA) index (Billing and Siber 2003)

developed by the European Commission Humanitarian Office (ECHO), which were

discussed in section 5.2.4.

Wisner et al do not suggest any indicators of ex post earthquake vulnerability but

describe it as vulnerability relating to “what happens after the initial shock and in the

process of recovery”(2004:276). Together, the writings of Schneiderbauer and Ehrlich

(2004), Albala-Bertrand (1993) and Alexander (2000a) indicate measurable aspects of

ex post vulnerability to include characteristics of secondary disasters, harsh weather,

food insecurity and unemployment. Figure 5.3 combines put the definitions of

Schneiderbauer and Ehrlich (2004) and Albala-Bertrand (1993) into a hierarchy. Not

all this ex post vulnerability data can be collected beforehand and it is hence more time

sensitive than ex ante data.

Source: Adapted from Wisner et al (2004:277); Schneiderbauer and Ehrlich (2004)

Figure 5.3 Conceptual model of vulnerability data

Loss data Loss and impact are used interchangeably in the literature. Although both

terms refer to the negative result of a disaster, ‘impact’ and ‘effect’ (Alexander 2000a;

- 70 -

Albala-Bertrand 1993) tend to describe the overarching qualitative outcome whereas

‘loss’ tends to describe the quantitative outcome exemplified by Wyss (2004b).

Albala-Bertrand (1993:12) provides a classification of disaster effects, outlined in

Figure 5.4 (Albala-Bertrand 1993:12). He divides the effects into ‘direct’ and ‘indirect’,

with several sub categories (not all are shown in Figure 5.4). The direct effects are

arguably the most objective and most commonly reported effects. Data in this

category include number of dead or injured persons as well as number of collapsed or

damaged structures (Albala-Bertrand 1993). In contrast, the indirect effects have a

qualitative character. In this category, Albala-Bertrand (1993) includes household

condition; general health and nutrition; the state of the economic circuit and public,

i.e. government, activities.

Source: Adapted from Albala-Bertrand (1993:12)

Figure 5.4 Disaster effect classification

In his analysis of the challenges in acquiring accurate and timely disaster impact

data Alexander (2000a:36) mentions the challenges caused by low reliability. There

are a range of intangible effects of disasters that are less readily quantified. This is

true for Albala-Bertrand’s indirect as well as the direct effects. Some measures that

involve a degree of subjectivity might never be settled. An example of this is the

number of injured (Alexander 2000a:37). Deaths can occur during the impact of a

disaster, though the disaster was not the cause. Alexander (2000a:36) writes:

[…] no category would seem more absolute than death, yet it is not so clear. If death occurs as a direct and immediate consequence of the disaster, there is no particular problem. But then there are indirect causes, such as disease, accident or secondary disaster (2000:36).

- 71 -

An example is heart attacks, which are an increasingly common cause of deaths

in earthquakes (Kario et al 2005). These heart-attacks might have occurred without the

added strain of the earthquake. If so, they would have been part of what Alexander

(2000a:37) defines as the ‘background mortality’. In this context it is important to

point out the difference between mortality and fatality. The 2005 Merriam-Webster

dictionary defines mortality as “the proportion of deaths to population”, i.e. a fraction

usually per 100 000 inhabitants. The fatality is the absolute number of killed people

and the ‘case fatality rate’ is the fraction of those who are injured for which the

injuries prove fatal.

Albala-Bertrand (1993:40) points out that a prerequisite for accurate data on loss

is “a set of pre-disaster information for the disaster area and appropriate post-disaster

methods for observation and enquiry”. According to Stallings (2002:52), such pre-

disaster information is a luxury that cannot be taken for granted in developing

countries. There is also a temporal dimension of the uncertainty. Alexander

(2000a:37) writes that “not all disaster-related deaths occur immediately [as the]

disaster strikes”. Loss data are hence not static, even in the case of sudden-onset

disasters. Both the real figures and the attempted measures of those figures vary

independently until a final figure for the two has been agreed upon.

Needs data Literature on ‘needs’ and particularly on the informatics of needs are

surprisingly rare. There seems to be a quiet consensus on needs being self

explanatory. In their article “Disasters: what are the needs?” Tailhades and Toole

(1991) approach the subject of post-disaster needs as health professionals. Although

they provide an extensive list of relevant data for loss and the general disaster context

their only mention of needs data are the very diffuse question of the “nature and

quantity of key emergency supplies needed from outside” (1991:21). This exemplifies

the trivialisation of the rather complex issue of needs data.

McConnan (2000) outlines categories of needs: ‘Water Supply and Sanitation’,

‘Nutrition and Food aid’, ‘Shelter’ and ‘Health services’. An additional category is

provided by Darcy and Hofmann (2003), ‘Protection’. With greater detail, Coburn

and Spence (2002:104) as well as Shakhramanian (2000) list specific types of needs that

arise after earthquake disasters. Similar assumptions can be made for other types of

disasters (Schneiderbauer and Ehrlich 2004). These models do, however, not

- 72 -

recommend typologies or standard units of scale for the reported data, but merely list

items that might be needed.

Relief data Literature contains several classifications of relief data. Alexander (2000a:86)

classifies relief resources in three categories: goods, services and cash. The relief of

‘goods’ include intangible forms of relief like logistics, energy, or communication

medium. Alexander (2002:73-79) provides a detailed list and classification of

international relief supplied in past disasters. Although he describes how the data

were collected, it is not clear how he developed the classification. Albala-Bertrand

(1993:31) differentiates services as being either ‘technical’ or ‘labour’. He defines

technical services as experts such as managers or scientists, whereas labour is defined

as a numerous workforce of, for instance, volunteers.

Smillie and Minear (2003:20) classify relief, with emphasis on financial relief as:

earmarked/unearmarked and bi-lateral/multi-lateral. Earmarking is used by bi-

lateral donors to specify “the geographic or sectoral areas in which a multi-lateral

agency or NGO can spend its contribution” (Smillie and Minear 2003:20). Bi-lateral

aid is aid channelled directly from the donor to the beneficiary, be that a host nation,

an NGO, or an independent agency. Multi-lateral aid is predominantly channelled

through the United Nations.

Contextual data Kersten (2000:41) divides contextual data into two groups according to its

purpose in the decision support process. His model-oriented data are those data aimed

directly at informing the decision maker.

Tailhades and Toole (1991) list model-oriented data with importance for the

health professionals’ response. The list includes the type and normal standard of local

communications, infrastructure, health-services, power, water and sanitation systems.

McConnan (2000:180) provides an even more extensive list, including the effect of the

disaster on particularly vulnerable groups. In addition, she recommends the

development of a demographical profile (by age, gender and social grouping) of the

affected population coupled with data on traditional lifestyle including architecture

and means of support and coping-strategies.

Kersten (2000:41) labels his second group data-oriented data. He defines the

purpose of data-oriented data as being an input to models that in turn produce

- 73 -

information that is more relevant or more intuitive to the user. This group contains

secondary indicators of characteristics, i.e. proxy indicators, which are suspected to

have an effect on a sought after quality, for example building vulnerability.

5.2.3 Database and User interface When the data were collected, all data were entered into a custom built

relational database management system (RDBMS) named the Database for

International Earthquakes Loss, Needs & Relief Estimation (INTEREST). The

platform had to be developed to facilitate both data entry and data analysis. Due to

the number of variables, the database uses a MySQL back-end with eleven tables and

a front-end developed in Microsoft Access. The research project started in the domain

of computer science and this legacy is evident in a complex and largely over-

normalized database structure (see the Entity-Relationship (ER) diagram in Figure

14.19 in the appendix). The database structure was developed with flexibility and

ability to store taxonomic data as a priority. Although this approach facilitated the

original research interest in information ontology, it was not optimal for the final

purpose of the database. The high level of normalisation poses a considerable

challenge in the extraction of data. Before being analysed, the data output had to be

thoroughly controlled for errors caused by mistakes in the database querying.

The dedicated user interface (see screen-shots in Figure 14.13 to Figure 14.18 in

the appendix) was developed to ease the data entry of the vast amount of data. An

example of the data entry accelerating functionality is the automated earthquake data

extraction function using a web connection to the NEIC to gather seismic data related

to an event based on entered spatio-temporal data. The initial data mining iterations

were made within the database application with a software module developed for the

purpose. The model allowed for time-series analysis and analysis of reporting style

and frequency. The final statistical analysis was made using the SPSS v14 software

package.

As data were entered into the database, a range of additional attributes were

stored for administrative purposes. These include the time of entry into the database

and method of entry (i.e. keyed in, scanned with optical character recognition, or

imported).

- 74 -

5.2.4 Quantitative data sources Table 5.3 lists the queried sources together with the number of events and

amount of information that they provided (see Table 5.4 for definitions). The

reference dataset for events in central Asia is based on information derived from the

CRED EM-DAT, the OCHA, the world’s news media, seismological institutions,

national and international NGOs and scientific institutions (see for instance EERI 2003

and Kaji 1998). For each case study, indicators were gathered, including the changes

of every indicator by each source over time.

Table 5.3 Number of reports and attributes per event according to source

Source Events covered

Reports Attributes

Centre for Research on the Epidemiology of Disasters (CRED)

53 54 176

Global Disaster Alert and Coordination System (GDACS)12

44 44 88

UN Office for the Coordination of Humanitarian Affairs (OCHA)

19 66 822

Reuters 15 79 23 UN Department of Humanitarian Affairs (DHA) 12 33 664 Agence France-Press (AFP) 11 67 11 United Press International (UPI) 8 23 3 Associated Press (AP) 8 10 2 Unknown 6 11 0 International Federation of the Red Cross/ Crescent (IFRC)

6 6 62

UN Integrated Regional Information Network (IRIN) 5 7 17 United Stated Geological Survey (USGS) 3 3 6 Earthquake Engineering Research Institute (EERI) 2 2 3 Local Media 2 2 2 European Commission Humanitarian Office (ECHO) 2 2 0 Intl. Committee of the Red Cross/Crescent (ICRC) 1 3 28 Christian World Services (CWS) 1 1 9 Middle-East Council of Churches (MECC) 1 1 5 Action Churches Together (ACT) 1 1 5 World Food Programme (WFP) 1 1 1 UN Children’s Fund (UNICEF) 1 1 0 UN Centre for Regional Development (UNCRD) 1 1 0 Other academia 1 1 0 British Broadcasting Corporation (BBC) 1 1 0

Source: Author, INTEREST Database

Seismic data The NEIC of the USGS was the sole source for seismic data. The NEIC database

is one of the few to offer global coverage. Selected data from the NEIC database were 12 As well as the predecessor: the Digital Map Archive (DMA)

- 75 -

manually imported to the INTEREST database. The selection criterion was that a

seismic event had to have its epicentre within a radius of 100km from the highest

magnitude earthquake that could be related to a location in a non-seismic report; i.e.

the closest city. The end result is a series of earthquakes linked to an event in the

INTEREST database. Although a series of earthquakes are categorised by

seismologists into foreshocks, main shock and aftershocks (Lomnitz 1974) the

database referred to the series as an earthquake ‘event’. The recorded attributes for

each earthquake are magnitude, hypocentral depth, time and epicentre. The time of

impact is consistently reported in Greenwich Mean Time (GMT). These attributes

were selected because they all are available to a decision maker in the moments

following an earthquake, potentially with the exception of hypocentral depth. This

timely availability is important if a model is to work in real-time in the future.

The earthquake is modelled spatially using a fixed radius of 50 km. A

shakemap would provide a more accurate model of the shaking (Hewitt 1997).

However, the shape is individual to each case and the calculation requires data on the

local geology (Bolt 2004) that currently is not available on a suitable resolution for the

studied region. Other developers, like Wyss (2004a) and Yuan (2003), have produced

spatial models for intensity. Those models are, however, either not for use without

expert input or not suited for use in developing countries. Consequently, until

accurate shakemaps can be provided in real-time for earthquakes worldwide, the only

solution is to use a fixed radius. The USGS and NEIC are closing in on finding a

solution to this problem. At the time of writing, the PAGER project took advantage of

raster attenuation models although not with global coverage (Earle et al 2005).

Media data Several studies, for instance Benthall (1995:36-42) and Olsen et al (2003), have

argued that media influence international relief community actions. Consequently,

the potential media influence has to be taken into account in the construction of an

accurate model of international relief community behaviour. With this reasoning, a

library of media reports for the period of interest was created. For events that took

place before 1999, the main media source used in the study is a manually assembled

database with English articles released by AFP, AP, Reuters and UPI. For events

occurring in 1999 and onwards, the Internet has served as the main source of news

articles because by that time it had matured and contained a broad selection of

archived material. To ascertain a more complete sample of media articles, the

- 76 -

European Media Monitoring (EMM) tool (Best et al 2005) was consulted. The EMM

stores and categorises news articles across all the European languages and news

agencies (EMM 2006). Due it being a recent tool, the EMM could only provide articles

from 2003 an onwards, making its benefit to the project limited.

After manual filtering and categorisation of the identified articles a total of 10

800 articles were stored in the INTEREST database. Only in a fraction of those articles

can the contained information be directly linked to the case studies. All the articles

are, however, related to world events concurrent to the case studies. The purpose of

the articles that are not directly linked to any of the case studies is to provide

information on the contextual situation of the case studies. For instance, this could be

information on concurrent natural disasters in other parts of the world, or other major

events potentially overshadowing a case study event in the news. Mitchell et al (1984)

argued that concurrent events are important factors in the estimation of the event

significance to the international community. The limited international response to the

1994 Mazar-I-Sharif, Afghanistan earthquake is likely to be related to a concurrent

landfall of a cyclone in Bangladesh, which may have diverted world attention. Olsen

et al (2003) indicate that such contextual elements could affect the resulting

international relief.

The attributes stored for each media report are: official source, release time,

release time zone, release location, case study link (if present), article heading and the

article itself. Although all of these attributes can be stored, not all data were supplied

with the original report in a majority of the cases. The most commonly missing data

are the release location and time zone.

Socioeconomic data For each case study, the spatial model of the earthquake was used to extract

data using the ESRI ArcView software package. The extracted data are nearby

settlements, the population and population density calculated using the Landscan

raster (see Plate 5.3). The Landscan dataset is an example of a readily available source

of spatial data. It is a population density raster with global coverage developed at the

Oak Ridge National Laboratory (ORNL) in the USA. The resolution is 30”x30” (arc

seconds) which at the equator is roughly one kilometre square (Bhaduri et al 2002).

The dataset was developed using variable resolution and adaptive proxy indicators of

local population distribution. Two editions have been released, one for 2002 and one

- 77 -

for 2004. All the Landscan data have been developed using approximate methods

and care should be taken in the use of the data. Furthermore, for the use of Landscan

in case studies occurring in the early 1990s the data are even more approximate

because the population model is derived from data that are a decade more recent.

Source: ORNL 2006

Plate 5.3 Landscan 2004 raster of global population distribution

On a national level, the GNA of the European Commission Humanitarian Office

(ECHO) was used as an indication of vulnerability. ECHO developed this composite

indicator of generic need of external assistance for use on a national level in 130

developing countries (Billing and Siber 2003). Their approach is not substantially

different from that of Badal et al (2004). Billing and Siber (2003) take nine normalised

indicators grouped into four categories of intended proxy indication:

• overall situation: Human Development Index (HDI); Human Poverty Index

(HPI);

• exposure to major disasters: natural disaster risk based on CRED EM-DAT

data; conflict prevalence based on the conflict barometer maintained by the

Heidelberg Institute for International Conflict (HIIK);

• humanitarian effects of population movements (the number of hosted

refugees based on United Nations High Commissioner for Refugees

(UNHCR) data; the highest estimate of Internally Displaced Persons (IDP)

based on data from UNHCR, the Norwegian Refugee Council and the US

Committee for Refugees) and;

- 78 -

• situation of children using the UNICEF child (<5 years) malnutrition data

and the UNICEF child mortality data.

The final indicator, donor contributions, constitutes a fifth group on its own.

Every country was given a score by Billing and Siber (2003) for each of the above

groups. For each group the top 25 percent were given a score of three, the mid 50

percent given a score of two and the lower 25 percent given a score of one. If an

indicator was missing, a score of zero was given. The total average for all groups for

each country is hence between zero and three, where three is the most in need. Billing

and Siber claim that the overall average then may “serve as a priority list for

humanitarian assistance” (2003:9).

In addition to the GNA, the World Press Freedom Index (WPFI), produced by

Reporters Without Borders (RWB), was applied. The WPFI gives an indication of the

relative press freedom in a country. This index could also be used as a proxy

indicator of a country embracing western democratic values. With an assumption

that donor countries are more willing to allocate funds to countries with western style

governance, this indicator could be important to the model development. Albala-

Bertrand (1993) has indicated that international relations and political agendas play an

important role in the allocation of funds and thus the interest provided to events.

National urban growth rate has been recommended as a macroscopic indicator

of earthquake vulnerability by Schneiderbauer and Ehrlich (2004). The urban growth

rate indicator applied in this study is an estimate of the relative increase of urban

areas on a national level between 2000 and 2005 made by HABITAT (2003, accessed

January 2006). The figure is an approximation and since the completion of this study

it has been updated by HABITAT. The new estimate includes a significant increase of

the urban growth rate in China. The researcher was made aware of the update in the

very last stage of the research was hence unable to re-run the analysis with the new

figures.

5.2.5 Data cleaning The massive amounts of data that had to be taken into account resulted in errors

in each of the stages leading up to the data being accurately stored in the database.

The coding standard of the data changed over the year as the focus of the research

project became clearer. These factors introduced several sources of error in the data

collection process. Consequently, to safeguard against errors the database went

- 79 -

through three iterations of thorough data cleaning. The data were checked with

contingency cleaning (Neuman 2000:314-317) procedures as well as through random

sampling followed by the verification of stored data. Even after these formal steps,

data cleaning continued; as discrepancies were encountered in the analysis, the data

were corrected and the complete database was then searched for similar errors. When

the data analysis started, there were no signs of discrepancies in the stored data.

5.2.6 Analytical Data Classification The conceptual top-level manifest codes used in the content analysis, see Table

5.4, did not provide sufficient detail for the use of statistical methods in Chapter 10.

The manifest codes were hence divided up further before being stored in the

INTEREST database.

Table 5.4 The top-level manifest codes

Code Definition Event The central entity to which all other terms are directly or indirectly linked. Each

case study is one event. Report A report is a set of information relating to an event. A report is a document that

originates from one source, though its attributes may have other sources. A report can only be linked to one event. It can contain many attributes.

Attribute An attribute can be textual or numeric and is a part of a report. It can only be linked to one report. An attribute has a source that does not have to correspond to the source of the report.

Source: Author

The need for increased level of detail created a major challenge while

populating the database. A strategy for the consistent interpretation and storage of

the data were needed. The task of data collection is divided into four subtasks for

which a strategy is required:

1) The scavenging of information on complex disasters in order to find

homogeneous events and to identify the reports belonging to these

unique events.

2) The identification of the atomic pieces of quantitative data in each report

and the linking of each datum with its original source and release time.

3) The retention of often loosely defined units and attributes of the

quantitative meta-data while still making analysis possible.

4) The clarification of the quantitative data meta-data, e.g. if an

organisation reports that “1 000 blankets have been dispatched” is that to

- 80 -

be interpreted as the number of blankets sent by them in (1) total or (2)

since their last report or (3) as all the blankets sent to the affected country

by all actors?

The main hurdle in the first subtask is to define and identify individual events.

Ultimately the decision whether to see an event as an entity on its own or as a part of

another event is a subjective arbitrary decision. However, if an event is significantly

separate geographically from its potential parent, or if the subsidiary event causes

new needs to arise it is more likely that it will be entered as a unique event. In their

sitreps, OCHA often indicates the nearest airport to the disaster site, this information

was used to support a decision that an event is new because new logistical routes

have were set up. The most pragmatic approach, however, is to use GLIDE numbers,

EM-DAT entries and OCHA sitrep issue patterns as the references. The sitreps are the

most realistic of these methods because they are issued by practitioners with field

experience of what constitutes a new event in terms of the mobilisation of a relief

effort. With the event determined it is less challenging to identify the reports and

attributes as described in subtasks two and three in the list above. The main concern

with reports was that not all of them would be found; especially those that were not

available on the Internet, as would be the case with most bi-lateral actions and local

government documentation. This issue cannot be solved easily, however, the time

span for the study was chosen with this problem in mind. OCHA and other

organisations have comprehensive online catalogues of reports starting from the early

1990s; due to the creation of the UN Department of Humanitarian affairs (DHA), the

granting of independence from the USSR of several of the case study countries and

the emergence of the Internet. It would have been impossible to obtain enough

information to provide an accurate account of what happened if the study had

stretched further back in time.

Subtask number two concerns the requirement to extract each piece of data in a

text to allow for frequency analysis of the reports. It was clear from the outset of the

research project that most reports contained a collection of sub-reports created by

sources that were not the same as the source of the report. The solution was to divide

each report into one sub-report per source. Each quantifiable statement in every sub-

report was then defined as an attribute. Figure 5.5 presents the hierarchy and lists the

characteristics stored on each hierarchical level.

- 81 -

Source: Author

Figure 5.5 Adapted manifest coding

Subtask number three concerns the requirement to include the units of

quantitative data to allow for numerical analysis. All units were retained in the

database as they were reported. For example, if 100 families were reported homeless;

it was stored as two values in the database ‘100’ and ‘families’ and not converted to an

approximate number of persons. The decision to store data this way makes the

analysis more complex but it does not introduce additional inaccuracy. Loosely

defined attributes were the standard used in the reports and these posed a much

greater challenge. The plethora of units that have been recorded made analysis

difficult. Units like “donkey load”, “caravan load” and “congregation” are examples

of this. Three stereotypical examples of these are:

A. “shelter and water supply needed for 5 villages for two months”,

B. “an additional 25mT of relief items and 4 relief teams to the value of

75kUSD have been sent to the area”,

C. “the livelihoods of people in three regions have been destroyed”

The solution to enable analysis of fuzzy attributes was to create a relief-data

taxonomy. The taxonomy consists of four separate classifications for data on Loss,

Needs, Response and Situation. The ‘situation’ classification is mainly used for

academic and media reports, e.g. weather and seismic characteristics, which do not fit

in the other categories. Each classification contains up to four levels, called tiers (see

Figure 5.6 and Table 5.5).

- 82 -

Source: Author

Figure 5.6 Excerpt from the relief data classification

In Figure 5.6 the first tier is complete, but the sub-classes are not shown for the

top-classes with dotted outlines. As data were entered into the database, new tiers

were created on a needs basis. In essence, as entity types were found in reports they

were put in the taxonomy.

Table 5.5 The Relief data taxonomy

Tier 1 Tier 2 Tier3 Tier 4 Equipment Lanterns Excavation Financial Food Cooking equipment Stoves Water Containers Jerry cans Fuel Coal AvGAS Wood Health Medical Supplies Vaccine Medical Services Hygiene Water purification Soap Laundry soap Human Resources SAR Logistics Air transport Ground transport Ambulances Water tankers Command, Control, &

Communications

Shelter Tents Large (i.e. Rubbhall) Blankets Clothing Shoes Plastic sheeting Generators Ground sheets Climate control Heaters Tarpaulins

Source: Author

- 83 -

The depth at which an attribute is stored in its classification tree reflects the

fuzziness of the attribute. Exact attributes are stored deep in the classification while

inexact attributes are stored in the top. Each attribute was entered in this way, often

resulting in several entries for a single sentence of analysed text. Analysing the three

stereotypical examples above, A and B contain two attributes and C one. In example

B the report states that ‘relief items’ and ‘relief teams’ have been sent to the area. It

does in other words contain two attributes. The locations of the two attributes in

example B in the hierarchy are: for the first attribute at the top as generic relief and for

the second attribute under ‘Human Resources’. In the database the attributes of the

above examples, not including meta-data, are stored in the following manner (top

level; deeper levels when required; quantity; unit):

A. Need; Shelter; 5; villages

Need; Water&Sanitation; Water; 5; villages

B. Relief; 25; mT

Relief; Human Resource; 4; teams

C. Loss; Human; Affected; 3; regions

In the database, each location in the taxonomy was coded e.g. 59 refers to ‘Relief;

Financial; Unearmarked; Cash’. Some information had to be kept in textual format as

a comment in the report that the attribute is part of or as a comment in the attribute

itself. This includes the financial value of the relief in example B above and the length

of the need in example A.

The last challenge presented in subtask four in the list above is that of

identifying the meta-data of the attributes. Consider the following additional

examples:

D. “15 families are homeless and not 150 families as previously reported”

E. “the government of France has donated 150kFFR which results in the total

of donations now exceeding 500kUSD”

F. “as mentioned in the previous report the government of Spain has

dispatched 5 dog teams, but they have not arrived yet”

G. “shelter is needed but SAR assets are not required”

- 84 -

To enable statistic analysis of the data a meta-data tag had to be added to the

attributes. Six categories for numerical data were created for this purpose (see Table

Table 5.6 Numerical metadata categories

Number categoryNumber category description Absolute Referring to the overall current event situationAccumulative The absolute for a single organisation Correction A correction of previously reported data Increment The difference from the last report. Non-Quantified Textual data Reiteration Data identical to a previous report.

Source: Author

An Absolute number is a number referring to the theatre wide situation e.g. 200

persons have been injured. Together with the increments it is the most common

category. In the analysis one cannot sum absolute numbers i.e. if the IFRC reports

that 200 persons have been injured and the local government reports that 550 persons

have been injured it would not be a correct conclusion that 750 persons are injured.

Accumulative numbers are Absolutes for a specific organisation and are only

used for relief data for instance: “up until today we have dispatched 10 SAR teams

and donated 50kUSD”.

Correction is used when previous data were incorrectly reported due to incorrect

translation or typos made by the reporting agency, not when the data itself was

incorrect. For instance, a number is seen as a correction if first report indicates that

“15 villages are damaged”, but the second report from the same source says that the

first report should have read that “15 villagers were injured”. If an organisation first

reports that it has sent 5 SAR teams but in later reports it becomes clear that only one

team reached the affected area it is not seen as a correction. Such differences are

reflected in the absolute numbers.

Increment shows the difference since the last report of the reporting organisation.

For instance, “we have sent one additional 50kW generator” or “we need an

additional 20mT of AV gas” or “the local hospital reports that an additional 20

persons have died”.

Non-quantified attributes are qualitative attributes that cannot be converted into

a quantitative equivalent. The intention was that all attributes should be quantitative.

However, in their reports organisations quite often do not provide numerical data

- 85 -

that could be beneficial to this study. Instead, they use qualitative information. It

could be that a report indicates that “shelter is needed” or that “water purification

equipment has been sent”. When all data were entered in the database, it became

clear that the ‘non-quantified’ type was the most common type.

Reiteration is when a report reiterates something from a previous report. For

instance, on the Monday the source reports that “Greece has sent 20mT of clothing” in

a second report on Wednesday it still mentions that “Greece has sent 20mT of

clothing”. This is probably not to be interpreted as if Greece has sent an additional

load of clothing, an increment, but as a reiteration of the information.

Using this terminology and classification, the above-mentioned examples would

be stored in the database as:

D. Loss;Human;Homeless;15;families – correction ( with reference to the first

report)

E. Relief;Financial;150;kFFR – increment

Relief;Financial;500;kUSD – absolute

F. Relief;Human Resource;SAR;Dogs;5;teams – reiteration

G. Need;Shelter;empty;empty;empty - Non-quantified

Need;Human Resources;SAR;0;empty; - absolute

- 86 -

5.3 Analytical methods This subchapter presents the analytical methods that are applied on the case

studies in Chapters 8, 9 and 10.

5.3.1 Qualitative data analysis The main task for the applied qualitative methods is to provide a set of user

requirements on a decision support system. Using the output from the formal and

informal interviews, observations and the analysis of organisational documentation,

time sequence analysis (Neuman 2000:433-434) is applied to develop a schema of the

current sequence of events in the international relief process and the allocation of time

by the decision makers to the various stages. In other words, the formal interviews

were conducted along a time-line to clarify the chain of events leading up to a

decision on whether to respond to a potential disaster. This approach forms the basis

for the structure of Chapter 8.

5.3.2 Quantitative data analysis Logistic regression allows for the prediction of a discrete outcome from a set of

variables that can be continuous, discrete, dichotomous, or a mix. A Dependent

Variable (DV) is predicted using a set of Independent Variables (IV). This is the

intention of the prototype model of this study. The international response to an

earthquake, the DV, is predicted using a set of indicators, the IVs. Logistic regression

is hence a suitable method to use in the search of patterns in the case study data.

Logistic regression is a relatively flexible tool in that it does not set requirements

on the IVs in terms of distribution or equal variance within groups. In addition,

unlike multiple-regression, logistic regression does not produce results below zero or

above one. Instead, using a link-function the result is transposed onto a probability

score on a logistic curve (Le 1998:116). Ordinal regression is a case of multi-nominal

regression where the response categories on the DV have an inherent meaningful

order. For each analysed case the probability of it falling into each one of the ordinal

categories on the DV is calculated. Hosmer and Lemeshow (2000:288) mention

opinions (i.e. strongly disagree, disagree, agree, strongly agree) and severity of

disease (i.e. none, some, severe) as common examples of ordinal outcomes on the DV.

- 87 -

Combined, logistic regression and ordinal regression gives ordinal logistic regression.

Ordinal logistic regression enables the output of the model to be put on a colour scale,

as previously done by De Groeve and Ehrlich (2002) in their replication of a traffic

light to represent the severity of an emergency. This way of conveying the

information follows Norman’s advice to “use both knowledge in the world and

knowledge in the head”(1998:189) to increase usability. Furthermore, the logistic

element of each classification provides a probability that a future event of a given

characteristic is classified in each of the ordinal categories. This reduces the

requirement of using complex numbers to convey uncertainty in the data, as

previously made by Wyss (2002a). Ordinal logistic regression is hence a suitable

method for the third thesis objective to develop a model to give a prognosis of the

actions of the international community on an ordinal scale.

Analysis process Starting with the selection of DV and IVs in Chapter 10, the statistical analysis in

this thesis adopts Mahadevan et al’s (2000) KDD process (see Figure 5.1 on page 50

and Figure 5.2 on page 54). An appropriate DV is searched for in the first phase of the

process, the problem definition. In the data selection phase the search is for IVs with

logical relation to the DV. In the data standardisation phase the selected variables are

cleaned and scrutinised for their appropriateness for use with the chosen statistical

method. This includes the selection of the representation of the variables in terms of

categorisation and data type. The actual application of the statistical method takes

place in the data mining phase. Although the KDD process gives an overview of the

model development, it is too coarse to be of help in the actual application of the

statistical method in the Data Mining phase. For this purpose the thesis adopts a

detailed model-building path in section 10.4. The path is developed by Hosmer and

Lemeshow (2000:91) and it includes five phases specifically developed for use with

ordinal logistic regression:

1. Uni-variable analysis

2. Multi-variable analysis input selection

3. Variable importance analysis

4. Main effects analysis

5. Model variable interaction

- 88 -

The process did not need to be adapted for use in this study. The uni-variable

analysis is a careful isolated examination of each candidate variable. Abnormalities

targeted in the search include categorical variables with empty cells, extreme values

and unexpected distributions on individual variables.

In the subsequent phase, the multi-variable analysis input selection, the previously

scrutinised variables are selected based on their suitability for analysis. One

limitation in the single-variable phase is that it ignores the possibility that a collection

of variables, each of which is weakly associated with the outcome, can be become an

important predictor of outcome when taken together (Hosmer and Lemeshow

2000:95). If that is expected, the variable should be included in the model even if it

has little effect on the DV. Empty cells occur when there are no instances of events for

a specific combination of values on the IVs. If the number of empty cells is too high,

logistic regression is unlikely to produce useful results. The solution recommended

by Tabachnick and Fidell (2001) is to categorise continuous variables, collapse

categories, or delete variables. Complete separations by dichotomous IVs are an

additional challenge in logistic regression. A complete separation is when one value

of an IV completely separates a value on the DV, thus making the other IVs

superfluous in the prediction of that value on the DV. This situation is commonly a

result of the sample being too small rather than the IVs miraculously being able to

exactly predict all outcomes (Tabachnick and Fidell 2001). Although the IVs can be re-

categorised or split up, the best solution is to expand the sample of events.

In the variable importance analysis, the complete model produced in the preceding

step is analysed for the importance of individual variables. The output of this phase is

the final set of variables. In the fourth phase, the main effects analysis, the relation

between IVs and the DV is critically analysed. Up until this stage the relation is

assumed to be linear. The function that defines the relation is referred to as the link-

function and the options are defined in the statistical package (SPSS 2003). In the final

stage, the model variable interaction, the interaction among the IVs is analysed for co-

linearity which might be detrimental to the model. Co-linear pairs can be replaced by

interaction variables when it makes sense based on domain knowledge. Hosmer and

Lemeshow (2000) refer to the model at this stage as the “preliminary final model”.

The model still requires testing and fit analysis to be adopted and become a final

- 89 -

model. Hosmer and Lemeshow summarise the numerical hurdles in the model

development in the following way:

In general, the numerical problems of a zero cell count [i.e. empty cells], complete separation, and collinearity, are manifested by extraordinarily large estimated standard errors and sometimes by a large estimated coefficient […]. New users […] are especially cautioned to look at their results carefully for evidence of numerical problems (2000:141).

With this advice in mind, the process of Hosmer and Lemeshow (2000) is

applied in the two phases of systems design and systems implementation in Chapter

Model testing methods A common way of testing a predictive model is to leave out a subset of events,

called test set, from the model development to be used for testing the preliminary

final model for accuracy (Mahadevan et al 2000). However, a major challenge in this

study is that the number of events will be insufficient if a set of events, large enough

to substantiate a test set, is left out from the model development. An alternative

available in logistic regression is the classification table of the observed and predicted

outcomes. The classification table is used to determine the effects of IVs and the

general characteristics of the model. The classification table provides a complete

overview of the accuracy of the model in a way that a summary indicator can not.

However, sometimes, like in the comparison of rough models, it is preferable to

measure fit with such a unitary statistic.

Summary comparison of ordinal logistic regression models is a complex task. A

rough indicator of the model’s fit is the pseudo-r2. In linear regression, the r2 statistic is

the proportion of the total variation in the response that is explained by the model

(Hosmer and Lemeshow 2000:165). The pseudo-r2 is an attempt to create an equivalent

measure for logistic regression. The Nagelkerke pseudo-r2 is used in this study, which

like the regular r2, goes from zero to one (SPSS 2003). A model that explains all

variation of the DV scores one. Although the pseudo-r2 gives a rough indication of the

predictive power of the model, it should not be given the same level of credibility as

regular r2 statistics in linear regression. The pseudo-r2 is useful when comparing

models but not particularly informative when determining whether a model is ‘good’

in general (Hosmer and Lemeshow 2000:164). Alternatively, the Pearson residual or the

Chi-square deviance (Hosmer and Lemeshow 2000:145) can be used as an indicator of

goodness-of-fit on a summary level as well as for individual values. These summary

- 90 -

comparisons and a classification table are all used in the model evaluation in section

As always in statistics there is a danger when drawing a conclusion of causality.

Even if a set of variables correctly predicts the outcome, this does not have to mean

that the IVs cause the outcome.

Time sequence analysis The real and reported changes over time following the impact of an event were

recorded for all quantitative data. This was done to allow for a time sequence analysis

in support of the analysis of the qualitative data collected in the interviews and

observation. When compared to the final agreed value of the attribute it is possible to

determine how the accuracy of the reported values changed over time for the

attributes. Figure 5.7 shows how the difference between the reported minimum and

maximum values of the number of injured reported in the 2002 Quazvin, Iran,

earthquake, changes over time. Similar graphs were produced for all case study

events for all the attributes in the four taxonomies.

Days after Event

Source: Author; INTEREST database

Figure 5.7 Envelope of the sum of dead and injured in the 2002 Quazvin, Iran, earthquake

The frequency of reporting over time was also studied. For instance, the count

of reported attributes containing data on dispatched relief was analysed over time

following each event. This gave a rough indication of the start, finish and crescendo

of the international relief.

Data quality Analysis of data quality will be necessary to evaluate the information sources

available to decision makers following disasters. This section presents the analytical

framework for that analysis.

- 91 -

O’Brien defines data as “objective measurements of the attributes (the

characteristics) of entities (such as people, places, things and events)” (1999:46). This

definition can be contested as we have seen in section 5.2.1 that data can be subjective.

Information, on the other hand, O’Brien defines as data converted into a “meaningful

and useful context for specific end users”. The subject of defining and determining

data quality is central to the science of remote sensing (Laurini and Thompson 1999;

Campbell 2002). The definitions developed in that domain are open to be used in

other applications as well. Three characteristics of data quality defined by Vereign

(1998) are Accuracy, Resolution and Completeness. Each of these aspects is in turn

grouped according to spatial, temporal and thematic quality.

In general conversation accuracy is often equated to quality, which is a very

simplistic view. Vereign (1998) relates accuracy to the lack of errors in the data; i.e.

the difference between the stored value and the measured reality. In statistical terms,

Campbell (2002:383) sees accuracy as measurements with low bias and low variability.

The comparison between stored value and measured reality can be non-trivial,

particularly when the measured reality is complex, subjective, impractical to observe

or plainly unobservable (Vereign 1998). Spatial accuracy of non-point data

exemplifies that accuracy in itself is a complex measurement. For instance, a stored

polygon can be of an accurate shape, but in the incorrect geographical location or

scale. Vereign (1998) does not see temporal accuracy being connected to temporal

metadata, i.e. entry date or database ‘up-to-dateness’, but the accuracy of the temporal

attributes. For example, a dataset claiming to depict the spatial vulnerability in 2001

is likely to have low temporal accuracy if it is constructed using data from previous

years. Thematic accuracy concerns the quantitative and qualitative attributes in the

database. In qualitative thematic accuracy, Campbell (2002:392) defines the possible

classification errors as either omission errors or commission errors. Omissions are cases

where an observation has not been allocated to its correct class, whereas commissions

are the cases where an observation is allocated to an incorrect class.

The resolution of data, according to Vereign (1998), is the amount of detail that it

contains. Campbell (2002:272) likens resolution to the ability of a sensor to capture

data on an object. Resolution is linked with accuracy as well as with what Laurini and

Thompson (1999:300) define as ‘precision’. In other words the resolution is the

density of measurements, while the accuracy is the consistency between reality and

- 92 -

the stored value and the precision is the exactness of each measurement. Higher

resolution is not always preferable. Low resolution can simplify the analysis process

for certain applications (Laurini and Thompson 1999).

Vereign (1998) defines completeness as the lack of omission errors on a database-

level; i.e. whether all desired aspects of an object are stored. The level of completeness

depends on the intended use for the data. Vereign (1998) points out that data that are

complete for solving one task might not be so for a different task.

Geographical Information Systems Each event in the database has been analysed using GIS. The outcome of the

GIS analysis is a population density map (see Plate 5.4) and an approximation of the

population size near the epicentre of each event. For all events, independent of

seismic characteristics, a circle with a 50 kilometre radius was used to extract the

population size.

Source: Landscan; INTEREST Database; Author

Plate 5.4 Population density map for the second Rustaq event13

As part of the GIS analysis, an attempt was made to geo-reference the media

reports stored in the database. The aim was to test if the concept of “information

black-holes” mentioned by Mr. Berthlin in an interview could be applied to give

13 The blue circle is the 50 kilometre radius around the earthquake epicentre. The turquoise stars are previous earthquakes in the area. The raster colour represents the pixel population: grey is areas with less than 10 people per square kilometre, maroon areas are densely populated urban areas.

- 93 -

estimations of the impact magnitude of an event. The information black-hole concept

was mentioned in interviews as being one of the indicators currently used to provide

rough estimations of the impact. After a disaster, the impact on local infrastructure

may prevent the flow of information out of the disaster area; thus creating an

information black-hole. The size of this black-hole could be an interesting attribute for

analysis in relation to the estimation of the severity of the event.

5.4 Methodological summary Sections 5.1, 5.2 and 5.3 reflect a project incorporating a mosaic of methods and

scientific domains. The best summary of the project is made through a presentation of

the main influences in the various stages of the research process. Figure 5.1 gives an

overview of this process, but Table 5.7 summarises all concepts of methodological

concern.

The first thesis objective – to establish a set of user requirements and to

determine the relevance of DSS in the international response to disaster – is covered

by the systems investigation and systems analysis stages of the IS development cycle.

In these stages, interviews were used to clarify current processes in the studied

organisations and to determine their requirements on a DSS as well as their perceived

relevance of such a system. As a complement to the interviews, several meetings with

stakeholders were attended by the researcher. Organisational documentation such as

guidelines and planning materials were consulted when possible. The second thesis

objective – to collect, to structure and analyse data for the development of a DSS

prototype – is achieved as part of the systems design stage. Content analysis is

applied to a wide range of documentation surrounding the international response.

The data generated in the content analysis are entered in a database developed for the

purpose: the INTEREST database. The problem with limited data accuracy is solved

through standardisation and aggregation of data. The third objective – to develop

and test a prototype DSS – is hampered by the limited number of case study events.

The prototype is developed through the use of ordinal logistic regression in the SPSS

software package. Because the alternatives for testing are limited by the number of

case study events, the final model will be analysed using the classification table of its

output.

- 94 -

Table 5.7 Project methodological overview

System Investigation Systems Analysis System Design System

Implementation

Problem Definition

Data selection & Standardisation Data Mining Model Evaluation

Persons Org. docs. Persons

Int. Org reports, sitreps, media, governments

INTEREST database

Interviews Interviews Content Analysis, GIS

Ordinal Logistic Regression, Hosmer & Lemeshow

(2000)

Critical analysis

Timeline analysis,

Data Quality analysis

SPSS Classification table

Limited access to interviewees Data quality Small Sample Small Sample

Participant observation Aggregation,

Standardisation Aggregation Use of

Classification table

Source: Author

- 95 -

6 EARTHQUAKE: A SUDDEN-ONSET HAZARD The purpose of this chapter is to give an overview of the domains of seismology

and earthquake engineering and to link them to the field of tele-assessment of need.

6.1 Hazard onset and complexity The temporal hazard parameter of Tobin and Montz (1997) include the speed of

onset, defined by them as the ‘warning period’. This is the time from a reliable

prediction of the near impact of a disaster to the time of its actual impact. Common

classes used for describing the onset speed are: creeping, slow, rapid onset and

sudden-onset hazards (Alexander 1993; Twigg 2004; Quarantelli 1998). Most authors,

like Alexander (1993:8), only distinguish between slow and sudden-onset hazards. In

reality the scale is finer (Albala-Bertrand 1993:11), but there is no consensus on the

definitions beyond the dichotomous. Alexander (2002:141) lists earthquakes,

tornadoes and flash floods as examples of hazards that allow for a very short warning

period: called sudden-onset hazards. On the other side of the scale he writes that

tsunamis, cyclones and drought usually can be predicted in ample time for the

potentially affected population to be able to take suitable action. This does, however,

not mean that they are predicted in an appropriate time-frame.

It is not always possible to identify the hazards that caused a disaster. In a

developing context, disasters are seldom sequential and independently identifiable.

This situation is referred to as a complex disaster. In the words of Kent (1987:6) a

complex disaster “is one where one disaster agent exposes vulnerabilities which open

the way for the impact of other disaster agents.” Alexander (2000a:214) argues that in

complex disasters “natural disasters are merely punctuating events in a constant

stream of misfortunes: normality is a disaster, peace and security are seemingly

unattainable goals”. Alexander (2000a) is supported by Albala-Bertrand (1993) in his

claims that the causes of disasters are harder to confront when sudden-onset disasters

occur as part of complex disasters. Albala-Bertrand (1993) sees the reason for this as

being the central role played by military and political opportunism and diplomatic

self-interest.

6.2 Measuring earthquakes Initial seismic data lack hypocentral depth, but there will be estimates of the

epicentral location and magnitude of the event (Woodward et al 1997). The epicentre,

- 96 -

a point on the surface of the earth, is a useful indicator of where an earthquake took

place (see Figure 6.1). However, it is not an accurate spatial representation of the

seismic event (Hewitt 1997:220). The seismic waves that constitute an earthquake

start in a focus (Bolt 2004:39), or hypocentre, at some depth underneath the epicentre.

They then spread gradually in three dimensions along the fault plane as it ruptures

(Bolt 2004:101).

Source: Guevara (1989) in Lagorio (1990:41)

Figure 6.1 Earthquake parameters

In big earthquakes, the fault rupture may exceed 1 000 km, which was the case

in the December 2004 Sumatra earthquake (NEIC 2004). The length of the fault can be

estimated using statistical relationships among the characteristics of the earthquake

(Bonilla et al 1984). However, although the rupture starts in the hypocentre, the fault

can spread in any direction from it, i.e. the hypocentre may be in the end of the fault

as well as anywhere along the fault. The exact location of the focus is much harder to

determine than the location of the epicentre (Sambridge et al 2003). If an earthquake is

of high magnitude, one can expect that its reported characteristics are accurate and

that the approximate focal depth will be known within an hour after the event

(Woodward et al 1997) depending on the distance to the epicentre from the sensors. In

the case of very strong earthquakes Woodward et al (1997) indicate that the initial

reports of characteristics are commonly underestimates. Shallow focus earthquakes

- 97 -

are arbitrarily defined by Bolt (2004:39) as those occurring within 70km of the earth

surface, while intermediate focus earthquakes reach 300km, and deep foci are deeper

with some exceeding 700km. The majority of the destructive earthquakes are shallow-

focus (Bolt 2004:41). The tectonically active regions of the earth have been mapped

together with the active fault lines where earthquakes are likely to occur (Bolt

2004:53). Seismologists have a fair knowledge of the depths of typical earthquakes

occurring along these faults (Sambridge et al 2003).

Each quake is built up of a wave train of individual types of waves with certain

characteristics (Keller and Pinter 2002:19). These waves travel through earth at

different speeds and are hence useful in locating the hypocentre. They also tell

something about the characteristics of the earthquake, which is important because the

various types of waves cause ground motion, in direction, amplitude and frequency,

that affect buildings differently (Keller and Pinter 2002:23). The wave train should not

be confused with foreshocks and aftershocks (Bolt 2004:41) which are earthquakes in

their own right.

The size of earthquakes can be measured in several ways, with each measure

having its specified purposed. Bolt (2004:158) presents the most common views on

earthquake intensity, earthquake magnitude and the, in earthquake engineering,

central measures of peak ground movement.

Intensity of shaking If the area hit by an earthquake can be accessed it is possible to produce an

intensity map based on an on-site survey of the damage (Bolt 2004). The commonly

used scale for this is the Modified Mercalli Intensity scale (MMI). In the survey, all

buildings or homogenously affected areas are allocated to one of twelve categories on

the MMI scale (I-XII), based on observation of the effects of the earthquake on

buildings, ground and people; referred to as a macroseismic scale (Bolt 1998:159-167).

The increasing levels of the scale range from almost imperceptible shaking to

complete destruction. Although the observation is guided by the damage descriptions

of the intensity levels, it is still a subjective assessment, particularly when basing it on

the accounts of the affected residents. Because the surveyed effects depend on,

amongst other aspects, the distance from the epicentre and the nature of the ground,

one earthquake will have many MMI values. An intensity scale cannot be used to

compare the size of earthquakes occurring in different parts of the world because it

- 98 -

depends on demographics and structural qualities (Bolt 2004:164). For instance MMI

is attuned to North American building characteristics. Other scales, like the MSK and

MCS, have been developed for use in Europe. No such widely accepted scales exist

for the developing countries.

Bolt (2004:271-273) shows that the MMI scale can be roughly correlated with

Peak Ground Acceleration (PGA). For example, MMI VII corresponds to a PGA of

between approximately 0.1g and 0.29g; MMI IX corresponds to PGA exceeding 0.50g.

Figure 6.2 (Bolt 2004) shows an attenuation functions for the PGA in relation to the

distance from the source of the shaking.

Source: Bolt 2004

Figure 6.2 Attenuation curves

By using networks of seismographs and GPS receivers Tralli (2000) develops

maps displaying the PGA over an affected area. Estimated intensity maps like these

are referred to as shake maps (Bolt 2004:161). Shake maps can be converted into a loss

estimate by combining them with spatial data on building quality and demographics

(Tralli 2000; Earle et al 2003, 2005). An example of a system built on this approach is

presented in section 4.2.2 (see Plate 4.1, page 45).

Peak Ground Acceleration (PGA) The Richter magnitude does not take the wave frequency or duration of the

seismic event in to consideration, both factors which are important in estimating the

resulting damage (Coburn and Spence 2002:267). The peak ground motion is the

collective name of a set of measurements relating to the movement characteristics of

the ground during an earthquake. These characteristics are the PGA, the Peak

Ground Velocity and the Peak Ground Displacement. The PGA is a measurement

- 99 -

commonly used in building design and as a reference in building codes (Lagorio

1990:28). The PGA is the parameter most often associated with the severity of ground

motion. However, the PGA on its own is not necessarily a good measure for the

damage potential of an earthquake because the acceleration can be very short-lived

(Coburn and Spence 2002:267). The accelerometers required to measure peak ground

motion have to be located relatively close to the source (Bolt 2004:113). PGA data are

hence not available for all earthquakes on the globe and it is particularly rare for

earthquakes in poor countries and in areas where earthquakes are not anticipated.

Magnitude As an alternative to the location-dependent intensity scales, seismographs are

used to measure the physical parameters of earthquakes. Unlike the strong motion

sensors used for measuring the PGA, the seismographs do not have to be placed very

close to the seismic source (Keller and Pinter 2002:42). The output from the

seismographs can be interpreted using a range of methods optimized for different

types of earthquakes. This results in a range of different units. Based on Bolt

(2004:158), Lagorio (1990:13), Coburn and Spence (2002:16-26) and Keller and Pinter

(2002:16-20), Table 6.1 lists the most common of these units. For tele-seismic

measurements the first report can be expected to be provided in Ms or mb (Bolt

2004:167; Menke and Levin 2005). Menke and Levin (2005) show that the delay of

more than four hours in the reporting of a sufficiently accurate moment magnitude is

one of the impediments that prevented timely tsunami warnings to be disseminated

following the Sumatra earthquake in 2004.

- 100 -

Table 6.1 Earthquake magnitude measurements

Magnitude Name

Description

ML Richter/Local Developed in 1935 by C.F Richter. Initially intended for use on the US west coast. The scale can be applied to moderate size earthquakes (3<ML<7). It is no longer used in the scientific domain, but often referred to by the media.

Ms Surface wave A tele-seismic measure, saturating at magnitude 8.3, for which no depth corrections are applied. Ms is hence not computed for depths greater than 50 km.

mb Body wave A tele-seismic measure developed specifically to treat deep-focus (50km<) earthquakes. Saturates at 6.2. Based on the P-wave amplitude.

Me Energy The logarithm of the amount of energy, measured in Ergs, which is radiated from the hypocentre in the form of seismic waves.

Mw Moment The seismic moment is the most precise and comprehensive measure of earthquake size. It saturates at about 8.5, but can be manually calculated for bigger events if special care is taken.

Source: Bolt (2004:158); Coburn and Spence (2002:16-26); Keller and Pinter (2002:16-20); Lagorio (1990:13)

6.3 Modelling The shaking produced by an earthquake does not result in uniform levels and

types of shaking at all locations. Although attenuation implies dissipation of energy

with the distance from the source of the shaking (Coburn and Spence 2002:246), this

does not mean that the shaking always gets weaker with the distance from the source.

The local geology around the fault affects the strength and direction of the seismic

waves emanating from it (Bolt 2004). Material amplification of the shaking of the

surface waves can result from the waves entering softer and wetter ground (Keller

and Pinter 2002:21) producing local effects that can be disastrous. The local

topography also affects the level of shaking (Bolt 2004:22; Yuan 2003) (see Figure 6.1;

Lagorio 1990:41).

Hewitt (1997:220) discusses the danger of approaching the task of estimating

physical exposure space using simplistic methods such as plots with earthquake

impact being represented by circles. He writes “damage patterns are rarely, if ever, of

this radial kind, and are poorly predicted by the radial attenuation or dissipation of

the seismic energy” (Hewitt 1997:220). Non-radial models are already in use in

several impact-estimation tools (see for instance Wyss 2004a; Shakhramanian 2000).

An approximate non-radial representation of shaking is not difficult to achieve. For

example, a prototype was developed by Yuan as part of his postgraduate thesis

- 101 -

(2003). However, as pointed out by Yuan (2003), the challenge lies in the automatic

production of an accurate attenuation in real-time following an earthquake. No

literature has been found indicating that a fully automated non-radial function for this

is in use outside the US and Japan. Non-radial representations depend on the

proximity of existing faults as well as a range of factors in the local geography.

Although Wyss (2004a) claims to incorporate the effects on fault proximity in his

model, it is not clear how he does so and, most importantly, to what extent that

process requires human expert input.

Although it is not the focus of this research project, the progress in the efforts of

the scientific community to predict earthquakes is worth an overview. Coburn and

Spence (2002:73) mention probabilistic seismic hazard assessment (PSHA) as a tool for

long term prediction. The PSHA analyse historical patterns of earthquakes to estimate

a return period and character of future earthquakes. The PSHA is very approximate

and can only support the long-term planning on a regional level. Coburn and Spence

(2002:77) refer to short-term earthquake prediction as “an illusory goal” and

summarise the prospect of short-term earthquake prediction in the below statement.

Despite half a century of work on short-term earthquake prediction, the prevailing mood among scientists is rather pessimistic. To date no reliable and widely accepted precursors have been found. […] Of the many short-term predictions of earthquakes that have been made, none […] have been both precise enough to lead to public action and subsequently proved correct. Claims for success tend to rest on the prediction of events expressed in a rather imprecise way. (2002:77)

6.4 Impact effects Earthquakes may lead to secondary and tertiary effects. Damage to structures

such as dams, dangerous industries, nuclear installations etc. can significantly amplify

the impact of the event (Albala-Bertrand 1993:14). The local effect of wet and sandy

soils can result in liquefaction that topples buildings (Keller and Pinter 2002:33). The

tsunami phenomena entered the limelight of loss estimation modelling following the

2004 Sumatra earthquake. Earthquakes on the sea floor or landslides entering a body

of water may cause disastrous waves and salt water inundation of coastal areas

(Papathoma 2003). Landslides triggered by earthquakes disrupt or destroy

infrastructure and can in their own right be significant cause of mortality (De Groeve

and Ehrlich 2002).

The most common secondary effect of earthquakes in urban areas is fire

(Davidson 1997). An earthquake may initiate many fires simultaneously and may

- 102 -

reduce the capability of the fire services through the disruption of water supplies and

general infrastructure destruction. Attempts at estimating the impact of an

earthquake (De Groeve and Ehrlich 2002; Schneiderbauer and Ehrlich 2004) show that

it is important to include the potential of possible consequential effects.

6.5 Earthquake engineering Earthquake engineering is the science of making structures better prepared for

earthquakes. Coburn and Spence (2002:263-265) write that the methods used in the

construction of a building gives a very good indication of how well it will resist

ground shaking resulting from earthquakes. For instance, a building made out of

unreinforced masonry is more vulnerable than a timber frame building (Lagorio

1990:144-158). All buildings have types of ground shaking to which they are

particularly vulnerable. Lagorio (1990:159-192) presents ways to strengthen the

buildings before an earthquake and to approximate the damage that the structure will

suffer in an earthquake. The number of floors in a building in combination with the

construction material gives an indication of which type of shaking that the building

will be most sensitive to (Wyss 2004b). Tall buildings are more vulnerable to low

frequency shaking and small buildings to high frequency (Bolt 2004:175). The high

frequencies generated by earthquakes tend to die off quicker with distance than the

low frequencies (Bolt 2004:175). As exemplified by the 1985 Mexico City earthquake,

it is hence possible that tall buildings several hundred kilometres away are affected

where small buildings are not (Coburn and Spence 2002:267; Keller and Pinter

2002:21). The Mexico City earthquake also showed that the different swaying of tall

buildings in urban areas may cause them to collide.

6.6 Summary This chapter has provided standards for measuring the shaking caused by

earthquakes (section 6.2) and has probed literature for recommendations made in

relation to the spatial modelling of earthquakes (sections 6.3 & 6.5). Deaths in

earthquakes are mainly caused by collapsing structures; therefore, to estimate the

losses caused by an earthquake, it is important to represent the level of shaking and

the characteristics of the affected structures in the development of loss assessment

models. The terminology and concepts presented in this chapter will be of particular

relevance in Chapter 10.

- 103 -

7 CENTRAL ASIAN REGION The purpose of this chapter is to provide an outline of the situation in the case

study countries. In order to benefit from the knowledge discovery process in the

analysis, it is important to understand how the countries compare to each other and

what their context is.

7.1 Region The main characteristic that ties the case study countries together is an aspect of

their physical geography. The meeting of the Alpide and Altai ranges as defined by

Lomnitz (1974:244) forms an area of high seismic activity centred on the Pamir

Mountains (see Plate 7.1 on page 105). The area stretches from the mountainous

region ending in southern Kazakhstan and in the south it reaches the Iranian coast.

Longitudinally the area starts with the two western-most Chinese autonomous

regions of Xinjiang Uygur and Tibet. In the west the area of interest stretches to the

Caspian Sea and the Iran-Iraq border. The studied region encapsulates Afghanistan,

part of China, Iran, Kazakhstan, Kyrgyzstan, Pakistan, Tajikistan, Turkmenistan and

Uzbekistan.

Political tension among the countries in central Asia still exists from the

revolutions in 1917 and 1991 (Mohammadi and Elitesh 2000). Immediately following

the 1991 break-up of the USSR a concern of the west was the threat of the spread of

radical Islamic regimes into the newly independent states (Shaw 1995). The west

feared that central Asia would follow the disastrous case of the break-up of the Balkan

states in the 1990s (Rumer et al 2000). When a reasonable degree of stability had been

achieved, the west capitalised on the potential of the natural resources and economic

opportunities left behind by the USSR (Rumer et al 2000). The region is rich in oil, but

the landlocked countries are dependent on each other for transport to the world.

Investments in the three main oil producing states in central Asia, Kazakhstan,

Turkmenistan and Uzbekistan is projected to make the region a significant global

actor by 2010 (Mohammadi and Elitesh 2000).

7.1.1 Earthquake hazard IFRC (1993:84) shows that although the central Asian region is not a region

where earthquakes are the greatest hazard facing inhabitants, it was the region with

the highest earthquake fatality in the 1980s. The collision of the Indian and Eurasian

- 104 -

plates creates a complex system of seismic activity (see Plate 5.1 and Plate 7.1). In

1974 Lomnitz (1974:248) divided the region into seismically coherent sub regions for a

better overview of the seismic characteristics which still are relevant.

The Iran – Caspian Sea region: A strip the width of Iran stretching from the

Iranian coast in the south, going north through western Afghanistan, ending in

Turkmenistan and in the south-western parts of Uzbekistan. Large earthquakes are

according Lomnitz (1974) not common in this area. However, shallow earthquakes

with magnitudes between 6.5 and 7.5 occur around the edges of the Iranian plateau

and on the ranges between Afghanistan and the Caucasus.

The Pamir – Balkash region: A strip between the Aral Sea and Lake Balkash

starting in the Pamir Mountains in Tajikistan, north of the Himalayas, going north

covering Kyrgyzstan, eastern Uzbekistan and southern Kazakhstan. The region is

small, but highly seismic. The Pamir Mountains are considered the structural knot of

the Alpide and Altaid ranges and the northern Pamir Mountains in Tajikistan are the

source of the greatest shallow earthquakes in the region. The region also includes the

Hindu Kush range that contains a well-known concentration of intermediate depth

earthquakes that can be of high magnitude. The great earthquakes in the Xingjian

region in China are the result of the extension of the Pamir range.

The Pakistan – Afghanistan region: An area covering eastern Afghanistan, the

whole of Pakistan, continuing south into India. This region has a relatively low

seismicity. The majority of the shallow earthquakes occur along the border between

Pakistan and Afghanistan. Additional active ranges enter Afghanistan from the

Hindu Kush.

The Tibetan – Chinese region: An area covering all parts of China excluding

Xingjian, Xizang and Mongolia. Earthquakes are relatively infrequent in this region.

However, the ones that occur tend to be very destructive. .

- 105 -

Source: Author; GIS Analysis, NEIC 2006

Plate 7.1 Map of case study earthquake epicentres14

Source: Author; GIS Analysis, Landscan data, NEIC 2006

Plate 7.2 1997, Bojnoord, Iran earthquake15

14 The tectonic plates are represented by the dotted line. The red stars represent the earthquake epicentres from the NEIC (2006) database. 15 Light pink pixels contain less than 5 people (roughly equal to 5 people per square kilometre), dark red pixels contain up to 450 people and dark grey pixels contain more than 450 people. The blue circle indicates the 50 kilometre radius of the event.

- 106 -

Source: Author; GIS Analysis, Landscan data, NEIC 2006

Plate 7.3 2002 Dahkli, Afghanistan/Tajikistan16

16 Light pink pixels contain less than 5 people (roughly equal to 5 people per square kilometre), dark red pixels contain up to 450 people and dark grey pixels contain more than 450 people. The blue circle indicates the 50 kilometre radius of the event.

- 107 -

7.1.2 Vulnerability Indicators for the case study countries for some of the characteristics previously

presented in Table 2.1 and Figure 2.2 are combined in Table 7.1.

Table 7.1 Comparison of the case study countries

Country HDI17 GDP17 Index

GDP/Capita18 Pop. growth rate %19

Urban pop. growth rate

Pop/km2

Afghanistan 0.346 - 822 - 4.88 49 China21 0.745 0.64 5’640 1.2 2.94 - Iran 0.732 0.70 6’690 2.6 1.23 42 Kazakhstan 0.766 0.68 5’870 0.3 0.82 5.5 Kyrgyzstan 0.701 0.46 1’620 1.6 1.81 26 Pakistan 0.497 0.49 1’940 2.4 4.17 207 Tajikistan 0.671 0.38 980 2.2 2.81 51 Turkmenistan 0.752 0.63 4’300 2.4 2.46 10 Uzbekistan 0.70 0.47 1’670 2.3 2.71 60

Source: See footnotes; GDP and Population density from UNDP 2004

According to Bloom et al (2002), apart from Afghanistan and Pakistan, the

central Asian countries fare pretty well in terms of quality of living when compared to

the rest of Asia. The vulnerability to earthquakes is, however, high in the whole

region.

More than one-half of all residential buildings in the Central Asian capitals would likely collapse or be damaged beyond repair if exposed to an MSK IX level of shaking. This means that a severe earthquake near a capital would cause, in addition to the deaths and injuries already mentioned, tremendous physical destruction of the city, with consequent inconvenience and economic disruption. (Geohazards 1996:1)

China, Iran and Pakistan are not included in the 1996 Geohazards document but

the situation in those countries cannot be expected to be much better. The

construction methods vary between rural and urban areas. There are several recent

examples, like the 2003 Bam earthquake, where the predominant construction

material, adobe (a type of mudbrick), was very sensitive to seismic effects. Soviet era

style reinforced concrete (RC) structures are common in urban areas in the former

Soviet republics (Geohazards 2006). Out of the six Soviet designs of multi-storey RC

buildings that were used in the former republics, only one is designed with

earthquake resistance in mind and it only saw limited use in areas with high

17 From UNDP Human Development Report 2004 18 From UNDP Human Development Report 2004, USD, Purchase Power Parity 19 1975-2002, from UNDP Human Development report 2004 20 2000-2005 estimate by UN HABITAT (2003, accessed January 2006) 21 Data is for the whole country. See chapter 7.2 for regional differences.

- 108 -

earthquake risk (Geohazards 2006). With reference to Kazakhstan, Kyrgyzstan,

Uzbekistan and Tajikistan, Bolt states that:

The present economic condition is such that adequate resources to increase the seismic resistance of many of the multiple-story unreinforced buildings are not likely to be available for decades. (2004:44)

The presence of vulnerable RC structures are of interest to this study because

they offer more time for trapped victims to survive and thus increase the efficiency of

international response efforts (Walker 1991). The richer and more exposed countries,

i.e. Iran and Pakistan, perform better than their neighbours in terms of enforcement of

seismic resistant buildings codes. A recent initiative to strengthen the bonds between

the former Soviet republics in central Asia in their efforts of disaster management is

the Central Asian Seismic Risk Initiative (CASRI) started in 2006 (CASRI 2006).

Although still in its infancy, the initiative is aimed at improving the mitigation and

preparedness efforts of the participating countries.

7.2 Nations

Afghanistan Afghanistan is an extremely poor, landlocked country, highly dependent on

farming and livestock-raising. The US-led military intervention in October 2001

marked the most recent phase in the country’s civil war (IFRC 2003). According to the

CIA (2006), during the 10-year Soviet military occupation one-third of the population

fled the country. Afghanistan is far from ethnically homogenous. The dominant

ethnic group are the Pashtuns, but there are more than 20 other distinct ethnic groups

speaking more than 30 languages. Due to the history of foreign involvement the

current borders of the country split ethnic and linguistic groups (Arney 1990). The

heterogeneity and lack of a common cultural identity is an obstacle in the creation of

any institutions, including those for disaster management (UNAMA 2003). The

international community has been accused of dealing with Afghanistan in “confused

and contradictory manner”, particularly during the Taliban-era (Leader 2001). The

majority of the population continues to suffer from poverty exacerbated by military

operations and political uncertainties (CIA 2006). Afghanistan has been on the

agendas of the international NGOs for decades (Nicholds and Borton 1994). The

country still hosts a plethora of international relief organisations. The Afghanistan

Information Management System (AIMS) (2004) records show the number of active

NGOs as of March 2004 to be in excess of 500. Afghanistan has an active Red Crescent

- 109 -

Society with 1 200 full time staff (IFRC 2003) supported by the national Red Cross and

Crescent Societies of many western countries. Nevertheless, the pre-disaster projects

in Afghanistan are limited to the preparedness phase of the disaster management

cycle (IFRC 2003). Benini (1998) gives a picture of the relief to the 1998 Rustaq,

Afghanistan, earthquake in which the majority of the effort seems to have been

international and ad-hoc. Local capacity has since been built up with support of the

United Nations Assistance Mission in Afghanistan (UNAMA) with an official policy

on disaster management being available (UNAMA 2003), but the country remains

prone to earthquakes.

China Only the two north-western autonomous regions Xingjian Uygur, also called

East Turkistan; and Xizang, also called Tibet, are included in the study. Both regions

cover a large area, Xingjian being the largest administrative area in China. Together

their area constitute almost a third of the China. Compared to the coastal Chinese

provinces they are both under-developed. As an example, infant mortality of the

inland provinces overall is almost twice that of the coastal provinces (Renard 2002).

Among all the Chinese regions GDP ranking Xinjiang climbed from 18th in 1978 to 12th

in 1995 whilst Tibet fell from 8th to 28th in the same period (Renard 2002). With the

economic development and acceleration of urbanisation, earthquake disasters in

Xinjiang could result in greater economic losses and bigger social catastrophe.

Information on the disaster management efforts in the Xingjiang Uygur and

Xizang provinces is vague. Official government sources claim that, as a country,

China is well prepared to deal with the aftermaths of earthquakes (Xinhua 2003).

Nearly 600 000 people were killed by earthquakes in China in the last 100 years,

accounting for 50 percent of the global earthquakes fatalities (Renard 2002). This

includes a very modest estimate of the death-toll caused in the 1976 Tangshan

earthquake some sources estimate at causing 650 000 deaths (Albala-Bertrand 1993).

Iran Iran has still to break away from the legacy of seeing the ex-Soviet states as

being part of the west (Mohammadi and Elitesh 2000). Mohammadi and Elitesh

(2000) point out that Iran lacked a coherent foreign policy towards the new central

Asian states, which resulted in fragile political relations. Relations have been better in

the past, Iran shares a long common history with the central Asian countries.

International relations have gradually worsened over the last couple of years, marked

- 110 -

by the 2005 election of the religiously conservative president Mahmoud Ahmadinejad

(Recknagel 2005) and the subsequent revealing of Iran’s apparent desire to develop

nuclear weapons (Broad and Sanger 2006). Its tarnished international relations aside,

Iran has consistently been the richest country in the region, which is reflected in its

domestic disaster management programmes (Ghafory-Ashtiany 1999). Disaster risk

reduction programmes in Iran are supported by major donors including the European

Union and the United Nations (UNESCO 2003). In addition to activities run by the

government, the Iranian Red Crescent are implementing preparedness programmes

(IFRC 2003). According to the International Civil Defence Organisation (ICDO) (2002)

Iran posses a well trained and funded civil defence organisation. Iran has experienced

more than 130 major earthquakes with a magnitude of 7.5 or more in the past

centuries. In the 20th century, 20 large earthquakes claimed more than 140 000 lives,

destroyed many cities and villages and caused extensive economic damage (UNESCO

2003).

Kazakhstan Kazakhstan is the former Soviet republic with the largest landmass, excluding

Russia. It possesses enormous oil and coal reserves as well as plentiful supplies of

other minerals and metals (Shaw 1995). Kazakhstan has enjoyed economic growth

since the late 1990s and has built up a well funded, equipped and trained civil

protection corps (ICDO 2002). Only the south-eastern border region of the country is

exposed to earthquakes (Suslov 1961:532; Lomnitz 1974). The low level of earthquake

hazard exposure combined with relative high level of preparedness arguably makes

Kazakhstan the country with the lowest earthquake risk in the study.

Kyrgyzstan When it became independent from the USSR, Kyrgyzstan was a poor,

mountainous country with a predominantly agricultural economy (Shaw 1995). It still

is one of the least developed countries in the region (UN 2003). Kyrgyzstan has,

however, distinguished itself by adopting relatively liberal economic policies. It was

consequently the first Commonwealth of Independent States (CIS) country to be

accepted into the World Trade Organisation (WTO). The improvement of the

economic situation in the country at the turn of the millennium convinced major

donors like ECHO to stop funding core humanitarian operations in 2000 (Taylor

2003).

- 111 -

In their 2003 Common Country Assessment, the United Nations see the greatest

risk as being posed by sociological hazards caused by “inequalities between regions

and communities and by unresolved border issues between neighbouring countries”

(UN 2003:5). The whole country is, however, also prone to earthquakes and

landslides (UN 2003). A functioning disaster management organisation is in place

(IRIN 2003), but it is reliant on foreign support.

Pakistan In the CIA World Fact book (2006), Pakistan is described as an impoverished

and underdeveloped country that has suffered from “decades of internal political

disputes, low levels of foreign investment, and a costly, ongoing confrontation with

neighbouring India”. As is the case with several of the case study countries, the

population in Pakistan is growing at a very rapid rate (HABITAT 2003).

Simultaneously, its rate of urbanisation is the highest in the region (HABITAT 2003)

(see Table 7.1). The country is also home to the world’s largest refugee population

with over 3 million people from Afghanistan (IFRC 2003).

The Pakistan Red Crescent Society, with 1 000 full-time staff, has an official

auxiliary role in the domestic disaster response capacity. The international funding of

mitigation and preparedness in Pakistan are very limited compared to the other

countries in the region (ECHO 2002). Earthquakes occur in all parts of Pakistan. They

are, however, more common in the west and the north on the border with

Afghanistan (Khan 1991). Both these areas are part of what is called the western

highlands. Their prevalence in these regions is unfortunate because international

relief has long been hampered by lack of security and inadequate infrastructure

(Nicholds and Borton 1994).

Tajikistan Tajikistan had one of the lowest per capita GDPs among the former Soviet

Republics (Shaw 1995). The civil war (1992-97) severely damaged the already weak

economic infrastructure. Even though the CIA (2006) claims the 60 percent of its

people continue to live in poverty, Tajikistan has experienced steady economic growth

since 1997. According to Taylor (2003) Tajikistan remained a classic example of a

‘forgotten crisis’ throughout the 1990s; humanitarian needs were almost unknown to

the rest of the world due to lack of media coverage. Thankfully Tajikistan is no longer

a ‘forgotten crisis’, although there are still forgotten needs (Taylor 2003). Earthquakes

are common in Tajikistan, but they are nevertheless not considered to pose a great

- 112 -

risk. Other hazards like drought, floods and famines have been frequent in recent

times preventing the disaster management institutions, like the national Red Crescent

Society, to adopt a proactive approach in their efforts (IFRC 2003).

Turkmenistan Turkmenistan is largely a desert country with intensive agriculture in irrigated

oases and large gas and oil resources. With an authoritarian ex-Communist regime in

power and a tribally based social structure, Turkmenistan has taken a cautious

approach to economic reform, intending to use gas and cotton sales to sustain its

economy (CIA 2006). Overall prospects in the near future are discouraging because of

widespread internal poverty and the burden of foreign debt. Turkmenistan's

economic statistics are state secrets and GDP and other figures are subject to wide

margins of error (CIA 2006). There are no signs in the literature of any organised

disaster management activities in the country.

Lomnitz (1974) describes the south east of the country as seismically active with

earthquakes that rarely are shallow or strong. However, coupled with the poverty

levels and lack of preparedness, the impact on society can be severe (IFRC 2003).

Uzbekistan Uzbekistan is the most populous state in central Asia (IFRC 2003). It was one of

the poorest areas of the USSR with more than 60 percent of its population living in

densely populated rural communities (Shaw 1995). Uzbekistan is now the world's

third largest cotton exporter, a major producer of gold and natural gas and a

regionally significant producer of chemicals and machinery (CIA 2006). The scarcity

of water in Uzbekistan, a landlocked country consisting of around 85 percent desert or

semi-desert, may in the long run become a source of tension both between and within

states (Rumer 2000).

Water-related hazards continuously claim lives (IFRC 2003) and earthquakes are

consequently not the greatest risk that the country is faced with. Earthquakes are

most frequent in the eastern-most parts of the country, though there are examples of

relatively strong intra-plate earthquakes in the central parts of the country (Lomnitz

1974). The national Red Cross and Crescent Society is working in close collaboration

with the government and international donors to provide support in all phases of the

disaster management cycle (IFRC 2003). No earthquakes with their epicentre in

- 113 -

Uzbekistan were identified in this study. There are, however, case study earthquakes

with their epicentre outside Uzbekistan that have impacted the country.

7.3 Sample earthquake events Two case study events stereotypical of intermediate international attention

events are presented here. The events were selected to show intricacy of identifying

the resulting losses, needs and international relief.

7.3.1 1997 Bojnoord, Iran earthquake The 1997 Bojnoord earthquake affected north-eastern Iran, bordering to

Turkmenistan (see Plate 7.2 on page 105). The cities of Shirvan, Ghochan, Bojnoord,

Esfarain, Sabzavar, Neishapor, Mashhad, Gondbad and Minoodasht were impacted.

Snow and below zero temperatures hampered logistics throughout the relief mission.

Two dozen villages were cut off due to damaged roads caused by subsequent

landslides and snowfall. The data on the event was gathered from DHA Geneva/UN

OCHA, Reuters, Cable News Network (CNN), United Press International (UPI),

Christian World Services (CWS), the International Federation of the Red

Cross/Crescent (IFRC) and the Earthquake Engineering Research Institute (EERI).

The event is stored in CRED EM-DAT (id: 19970017) with its data sources for the

event being Lloyds, Swiss Re and AFP.

Table 7.2 Bojnoord, Iran, initial data

Date 4 February 1997 Time 10:37 GMT = 14:07 Local Latitude 37.39N Longitude 57.35E Magnitude/Max Intensity

6.1 Richter / VIII MMI (6.5Mw NEIC)

GDACS Alert level Orange 50km population 362 007 Characteristics Three major earthquakes measuring 5.4, 6.1 and 4.0 on the

Richter scale. Several hundred aftershocks recorded. Hypocentral depth unknown.

Source: OCHA 1997, NEIC 2006, GDACS 2006a, Landscan data, GIS analysis

The final toll as reported in June by the International Institute of Earthquake

Engineering and Seismology (IIEES) was 88 casualties and “considerable damage” in

173 villages (Tatar 1997:1). The level of damage in those villages equated to an

intensity of VIII on the Modified Mercalli Intensity scale. According to the IIEES

report, few lifeline structures were located within the area affected by the quake. A

petrochemical facility had to close down temporarily after the quake, but did not

- 114 -

sustain any damage. Some steel and concrete bridges were within 30-40 km of the

epicentre, but none of them received any observable damage. The damage was

concentrated in 50 villages, 15 of which were totally destroyed. The first OCHA

report (1997:1) concluded “Final damage/loss is expected to exceed the above”.

Substantial attention was given by international media including CNN, Reuters and

UPI. The reported impact over time is summarised in Table 7.3.

Table 7.3 Reported impact over time

Damage 5 February 7 February 20 February 1 June Casualties 57-72 79 88 88 Hospitalised 498 Injured 160-200 360 1 450 Houses Damaged 2 400 -

2 800 11 000

Houses Destroyed 2 800 5 500 Villages Damaged 29 49 173 173 Villages Destroyed 21-73 14

Source: OCHA 1997

The UN disaster assessment team was dispatched from Mashad, Iraq, on 5th

February 1997 (OCHA 1997). Their findings were presented in the DHA report of 7th

February 1997 which included the preferred types of assistance (see Table 7.4). In the

DHA report of 20th February 1997, the government of Iran announced that they were

ready to receive aid of the types specified in Table 7.4.

Table 7.4 Reported needs over time

7 February 20 February Cash Cash Tents Medical Tents w/ equipment

Blankets Tents New Warm Clothes Blankets

Rice Cooking Oil

Pulses Other Foods

Source: OCHA 1997

The Red Crescent Society (RCS) of Iran established four operational task forces

in the area. In total, the RCS operation included 165 relief workers divided in 15

teams. Table 7.5 contains consolidated information on all aid that was reported to

have been sent to the disaster zone. It is not clear if the reported data are incremental.

- 115 -

Table 7.5 Reported dispatched relief over time

Response 5 February 7 February 20 February Cash (kUSD) 20 550 Tents (pcs) 850 2 200 11 500 Medical Equipment Yes Blankets Yes 5 800 38 700 New Warm Clothes 91mT* 6 000 pcs Rice Cooking Oil Pulses Other Foods (mT) 91* 40 175 Household Utilities (mT) 91* Detergents & Soap (mT) 18 Plastic Sheeting (mT) 50 Heaters Yes 740 4 150 Heavy Duty Machines Yes Ambulances Yes Fuel (mT) Yes 14 Helicopters 2 3 External Relief Workers 165 2 000

* = Combined shipment with unspecified division of contents Source: OCHA 1997

Bojnoord event summary The information base for this event is good. Iran experienced a string of

earthquakes during the spring of 1997, of which this event is the first. The Orange

alert level issued by GDACS is correct in this case. The impact does not justify an

immediate and unquestioned international intervention. Secondary effects, media

coverage, and political agendas might, however, influence the requirement for

attention.

7.3.2 2002 Dahkli, Afghanistan/Tajikistan earthquake The epicentre of the main earthquake for this event was in Tajikistan, 25km from

the border with Afghanistan (see Plate 7.3). This event is just a small part of a

complex emergency situation. War, food shortage, landslides, floods, disease,

malnutrition and other factors made the condition very acute. A second major

earthquake occurred three weeks after this one. The emergency phase for this event

did not end, but continued into the second event.

The description of the event is based on several reports from NGOs and relief

organisations available on the Reliefweb website (OCHA 2006). The event is stored in

CRED EM-DAT with ID 20020122 for Afghanistan and ID 20020127 for Tajikistan.

- 116 -

Table 7.6 Dahkli, Afghanistan/Tajikistan, initial data

Date 3 March 2002 Time 12:08 GMT = 17:08 Local Latitude 38.543N (36.5N NEIC) Longitude 70.424E Magnitude/Max Intensity

7.2 Richter (7.4 Mw NEIC)

DMA/GDACS Alert level Red 50km population 45 535 Characteristics Depth >200km. Preceded by another deep earthquake and

followed by a string of both deep and shallow earthquakes with magnitudes between 4 and 5 Mb.

Source: OCHA 2002, NEIC 2006, GDACS 2006a, GIS analysis

The initial damage reports from the urban areas indicated that the overall

damage was limited. However, in the aftermath it became clear that a remote village

had been hit by a large landslide (OCHA 2002). The OCHA situation reports outline

how the landslide was started when a “huge” limestone rock face fell off a

mountainside and became pulverised. The landslide went through a village and

stopped in a river. As a consequence the river flooded and caused the residents of

villages upstream to evacuate their homes and the villagers downstream lost their

supply of water. During the rescue operations the remaining part of the rock face was

unstable and likely to fall down. The landslide had dumped 30 000 cubic metres of

material in the river, out of which 15 000 cubic metres needed to be removed for the

flow to return to normal. The water rose 3 metres per day until a hole in the dam was

made. The landslide blocked some of the main feeder roads in the area. These roads

received further damage as they became flooded. To add to the complexity of the

operation, the area was contaminated with landmines.

Table 7.7 Reported impact over time

Damage 4 March 5 March 7 March Casualties 57 157 Injured Persons 150 165 169 Homeless Persons Affected Families Houses Damaged 100 125 Houses Destroyed 32 672 925 Villages Damaged Villages Destroyed 1 Killed Livestock 500

Source: OCHA 2002

Tajikistan suffered few injuries and no deaths, but 470 houses, 30 schools and 30

medical facilities were damaged. An additional landslide of 10 million cubic metres

in Tajikistan threatened to block the Vakhsh River which would have had catastrophic

- 117 -

consequences for the whole country, but the slide did not enter the river. The

majority of the relief items that were needed seem to have been available inside the

country. It was the river blockage that gave rise to some specific requests such as:

explosives, geology expertise, construction expertise, survey engineers, high capacity

water pumps, steel pipes, concrete culvert pipes and heavy diggers. All of the

reported responses were made by organisations, mainly NGOs, which were stationed

in the country at the time of the earthquake. No records of international donations

were found.

Response 4 March 5 March 7 March Tents 1,000 Yes 600 Blankets 30,000 Yes 3,800 Water containers Yes Hygiene Supplies Yes Medical Supplies (mT) Yes Clothing Yes Heavy vehicles 10 12 19 Food Yes Helicopters 2

Source: OCHA 2002

The event received limited coverage in international media. One article each

from AP and AFP were identified. These reports speculated in the death toll being

100 to 150 persons.

Dahkli event summary What made the event serious was the extreme vulnerability of the already

existing complex disaster in the region. The landslide made the rescue and

intervention even more precarious. The event did, however, not cause additional aid

to flow from other nations. The reason for this is probably that the region had

received plenty of aid as a response to other disasters at the time. Since there was no

specific international response to the earthquake, it could be argued that the event

should not be given a Red alert. However, if previous aid and political agendas are

set aside, the Red alert issued by the GDACS is well justified.

7.4 Summary The countries in central Asia are both prone to and vulnerable to earthquakes.

Earthquakes are most frequent in the southern Iranian provinces and on the border

between Pakistan and Afghanistan. Tajikistan, Kyrgyzstan and the western Chinese

provinces also experience earthquakes, but on a less regular basis.

- 118 -

The region is economically poor, with the possible exception of Iran, and bi-

lateral aid between the countries in the region can be expected to be limited due to

political tensions and poverty. The IFRC plays a major role in the disaster

management activities in most of the case study countries. Information on the level of

earthquake preparedness in the countries is limited. Kyrgyzstan, Kazakhstan and

possibly Uzbekistan seem to occupy the centre in terms of earthquake preparedness.

Data for Afghanistan is scarce, but the HDI, the low GDP and the rate of urbanisation

point to it being the poorest and most vulnerable of the case study countries, followed

by Tajikistan. On the opposite end of the spectrum, Iran is best prepared for the

earthquake hazard. Pakistan, although poor and highly exposed to earthquakes, is

well prepared for an earthquake emergency. However, as with all the case study

countries, the implementation and enforcement of earthquake-resistant building

codes and other mitigation measures is very limited. Information on disaster

management activities in China, and in the two case study provinces in particular, is

unclear. Official sources claim a high level of preparedness, but the vast and poor

provinces in the west are likely to be overlooked.

- 119 -

8 SYSTEMS INVESTIGATION Mahadevan et al (2000) refer to this task as the Problem Definition in their KDD

process (see Figure 5.1). This phase allows the identification of precarious situations

where decision makers lack appropriate data or where the wait for suitable data can

cause temporal bottlenecks in the decision process. It is in those situations that a DSS

could make the greatest beneficial impact (O’Brien 1999:95-97).

8.1 Implementing organisation This section is based on a series of interviews with Mr. Per-Anders Berthlin, the

Swedish Rescue Services Agency (SRSA) Senior Advisor on Overseas Operations, a

tactical decision maker (Figure 3.1), and with Mr. Fidel Suarez, manager of the

Spanish rescue services’ canine unit, an operational decision maker. The SRSA is the

Swedish government agency responsible for domestic emergency management. The

agency is also implementing relief missions of short-term emergency response

character on the international scene. In the case of earthquakes the relief most

commonly takes the shape of Search And Rescue (SAR), but it can occasionally

involve components of medical aid, water access, shelter, etc. Mr. Berthlin is

responsible for managing the Swedish international SAR assets and in that role he is

making non-political decisions in all regards to international SAR missions.

The entry decision process in SRSA was mapped based on the information from

the interviews and meetings with Mr. Berthlin. The mapped process stretches from

the occurrence of an event to the point when a decision of whether to intervene is

made. The actors involved in the decision process according to Berthlin are listed in

Table 8.1. In the table Berthlin himself has the role of a SAR response domain expert.

The following summary of the SRSA organisation and response processes is based on

the series of interviews with Mr. Berthlin, supported by preliminary analysis of the

data as well as observations made at practitioner conferences.

- 120 -

Table 8.1 Roles in the SRSA decision process

Role Responsibility SOS Alarmering AB Receive and disseminate urgent requests for assistance. International department desk officer (IJ)

SRSA internal logistics and co-ordination.

SRSA duty officer (VT) Central to the process. He briefs the director general and acts as a point of contact throughout the process.

Domain experts International co-ordination and decision support development.

Senior decision makers Maintain all domestic contacts Director General Formal ultimate responsibility.

Source: Personal communication with Mr. Berthlin

The SRSA used to have an in-house news monitoring department that alerted

the duty officer in case of a potential disaster. Berthlin defined a disaster as an event

that required “a swift response” from his organisation, thus excluding slow-onset and

protracted events. Due to reorganisation in early 2006 and expansion of the

organisation, this process has been completely changed. First, in the process outlined

in Figure 8.1, in the case of requests for assistance coming from the affected country or

a co-ordinating body, the alert is managed by an external company, ‘SOS Alarmering

AB’22. Berthlin mentioned that they have been tasked by the SRSA to act as the initial

point of contact and to activate the decision process by alerting relevant staff at SRSA.

Alternatively, before any external request for assistance has been made, other alerts of

phenomenological nature, e.g. seismological reports, go directly to the relevant SRSA

domain experts who analyse the data and activate the decision process if deemed

necessary. Berthlin stated that “At this stage ‘necessary’ is anything that could require

assistance”. Secondly, ‘SOS Alarmering’ or the domain expert activates the decision

process by alerting the SRSA Duty Officer (VT) and the International Department

Desk Officer (IJ). VT and IJ then discuss the situation and decide if there is cause to

proceed to activate the next step of the decision chain. According to Berthlin, at this

stage the process almost always continues to the subsequent step. Third, VT and IJ

issue an internal alert that goes out to the relevant senior decision makers on

department (avdelning) and unit (enhet) level.

22 Swedish for ‘SOS Alarm-raising Incorporated’; hereinafter ‘SOS Alarmering’

- 121 -

Source: Personal communication with Mr. Berthlin

Figure 8.1 SRSA response process

If the relevant domain experts were not involved in the initial alerting, they are

contacted at this stage. VJ, IJ, the senior decision makers and the domain expert then

critically analyse and discuss the pertinent questions:

• Is there a need for a response from the SRSA?

• Does SRSA have the ability to respond in terms of skills and resources?

If the above three steps results in a decision to proceed with an intervention,

Berthlin claims that the steps outlined below will follow and occur in parallel.

Domestic governmental contacts are made by the SRSA senior decision makers. The

contacted institutions are:

- 122 -

• The Ministry of Foreign Affairs (Utrikesdepartementet):

Berthlin describes their role to provide foreign policy input and to establish

contact with in-country sources in the affected country that could provide

additional information on the situation.

• The Swedish International Development cooperation Agency (Styrelsen för

Internationellt Utvecklingssamarbete - SIDA):

According to Berthlin, SIDA provides input on the potential side-effects of an

intervention with regards to development policy impact. Furthermore, the

relief budget to which any response is debited is managed by SIDA and their

approval is essential. SIDA also provides contacts with in-country information

sources.

• The Department of Defence (Försvarsdepartementet):

In the interviews Berthlin made clear that although that SRSA resides under

the Department of Defence, the role of the Department of Defence is very

limited in international emergencies. Nevertheless, Berthlin said that the

director general of the Department of Defence is the formal decision maker

with the ultimate decision whether to respond. This power has, however,

been delegated for emergencies and the Department of Defence approval is

only necessary for non-emergency, planned, interventions.

In parallel with the above domestic activities, Berthlin mentioned that IJ and the

Domain experts conduct a comprehensive search for additional information by

contacting a range of international agencies and identified sources in the affected

country. These include the OCHA, the Virtual OSOCC, the International Search And

Rescue Group (INSARAG), the European Commission Monitoring and Information

Centre (MIC) and the North Atlantic Treaty Organisation (NATO). Other responding

countries are contacted so as to avoid duplication of efforts. Berthlin said that SRSA

has a close relationship with a set of countries that are among the most frequent and

experienced SAR responders. This was apparent to the researcher at various

conferences. The attendant nations and practitioner representatives were the same

and all the representatives and managers within Europe knew each other well.

Berthlin summarised the core group of countries as SRSA partners in the International

Humanitarian Partnership (IHP), which includes Belgium, Denmark, the Netherlands,

Norway and the United Kingdom. Standards for equipment, communications etc.

- 123 -

have been developed within the IHP to allow improved co-ordination and

cooperation. In case of a SAR response, Berthlin pointed out that a set of additional

countries that usually provide significant international assistance were contacted.

These include: Estonia, Germany, Switzerland and the United States of America. This

fits with the researcher’s observations at practitioner conferences as well as in the

collected data on international responses. As the final step Berthlin said that any

agencies in the region of the affected country are contacted.

The third step occurring in parallel with the domestic and external

arrangements is the internal intervention preparation initiated by the IJ. Berthlin

described the process as the relevant administrators within SRSA being mobilised so

that they in turn can mobilise the intervention assets. Each administrator has an area

of responsibility: logistics, personnel, equipment, communications and healthcare.

Berthlin elaborated on the role of the healthcare administrator. Although all

healthcare equipment except medicines is pre-packaged there is still some co-

ordination required. Depending on the disaster impact and the geographical region

the contents of the healthcare package might need alteration.

The fourth parallel activity is the activation and briefing of the SRSA media

relations department. Berthlin said that they are continuously supplied with

information on the planned intervention activities for dissemination to the Swedish

and international media.

Intervention timeline According to Berthlin it is the intention of SRSA to have assets airborne within a

maximum of ten hours following a request for their assistance. In the case of SAR, the

policy is for the decision to intervene to be taken within six hours of receiving the

request for assistance. Berthlin stated that if the decision to intervene takes more than

six hours the SAR teams are likely to arrive in the affected area too late to have a

significant impact on the rescue work. Nevertheless, Berthlin admitted that SAR

missions are sometimes launched after well beyond six hours of deliberations.

However, in those cases Berthlin emphasised that the decision to intervene is based

purely on political priorities in the foreign policy domain of the Ministry of Foreign

affairs and not on analysis of the potential benefit of the response to those directly in

- 124 -

Berthlin claimed that the most common temporal bottleneck in the process

following the decision to intervene is the sourcing of a suitable aircraft. This seldom

takes less than six hours and Berthlin’s intervention plan for the SRSA was developed

with this bottleneck in mind. Table 8.2 presents the process described by Berthlin.

The loading of the airplane is scheduled to start six hours after the decision to

intervene has been taken, to be synchronised with the sourcing of an aircraft.

Table 8.2 SRSA intervention timeline

Actor Action Required time (h)

Time after event (h)

Affected nation or Co-ordinating body

Issue a request for aid or an alert. 1 1

Take decision to intervene. 1-6 2-7 Internal alarm 1 3-8 Activation of staff 1 4-9 Mobilisation of staff and transport to collection point

4 8-13

Loading of airplane. 3 11-16 Source: Personal communication with Berthlin

Equipment preparations When queried for the process of determining the composition of the relief

package, Berthlin answered that the SRSA uses a standard set of equipment packages.

These kits are packed in shipping containers ready to be loaded on to an airplane.

Although the contents of the containers are not changed between an entry decision

and the dispatch of a relief mission, Berthlin said that the composition was evaluated

after every intervention. The researcher targeted this statement with several

questions relating to the suitability and logic of a policy not to change the composition

before the dispatch of the relief. Berthlin stated that the kit in use had been developed

“based on more than twenty years of field experience and will not be changed unless

there are some specific feedbacks from units of other countries that have arrived at the

scene early”. He used the Bam earthquake as an example of a situation where heavy

rescue equipment was deemed not to be required by those arriving first on the scene.

Berthlin clarified that the contents of the kits are revised after the completion of

interventions based on indications from the response teams of something missing.

Although, Berthlin summarised: “the kits still have to be assembled to fit your

average type of intervention”.

- 125 -

Alert tool user requirements Both Mr. Suarez and Mr. Berthlin were questioned on the user requirements of

an alert tool. Neither one of them had used DSS in previous interventions. They had,

however, used a simple alert system based on a system broadcasting an alert over

pagers when a NEIC notifies of an earthquake with a magnitude exceeding five (ML,

Mw, Mb or Ms). Berthlin mentioned that in cases were he is uncertain whether an

event is a disaster, he would look for an information ‘black-hole’. He defines an

information black-hole as areas from which no reports emanate. The size of the black-

hole can also be used as crude indication of the geographical spread of an event.

Although this is a method for supporting his decisions it can not be seen as a DSS.

Suarez highlighted the importance of timeliness of the alert. His opinion was

that inexact information is a part of life for decision makers in this domain and that

they consequently know how to benefit from such information and have to accept

false positive alerts. He also saw the lack of automated processes for the response as a

hindrance in the interventions that he had taken part in. Berthlin was also questioned

on what he expected from a DSS in terms of timeliness, content, quality, as well as the

role of the DSS in the decision process. Regarding the user requirement on the

timeliness of an alert message Berthlin stated that:

Considering that it often takes one hour or more for the alert or request to come through traditional channels, any alert that is provided before that point in time is of potential benefit. The alert will be of no use after more than six or seven hours following the event. (Berthlin personal communication December 2005)

Suarez stated that the usefulness of the alert is higher the sooner that it is

received, but that it will be of no use after the first 24 hours following and event. It is

obvious that the alert is more helpful the sooner it is received by the user, but

information content and quality is also important. When questioned what the

minimum level of information and accuracy that is expected from an early request or

non-phenomenological alert Berthlin, without hesitation, gave the following points, in

order of importance:

1. Knowledge of the level of clarity that the reporting agency has of the

situation. This mainly consists of metadata on information quality. Berthlin

provided the following example questions: Has the reported information

been confirmed by on-site sources? Is there any information coming from the

field whatsoever? What assumptions have been made?

- 126 -

2. Estimation of the level of need.

3. Estimation of the type of need.

4. Loss assessment and information on regional and local response efforts.

Suarez also stated knowledge of accuracy of the alert as an important factor.

Berthlin sees the above information coming from a co-ordinating body like OCHA or

the VOSOCC. He claims that the affected nations commonly take too long a time to

disseminate requests for international aid for the requests to be helpful in the process.

When asked by the researcher whether this was all that he saw relevant, Berthlin

continued to mention a second type of alert that is tied to the nature of the hazard.

These are the near real-time alerts sent to the domain experts at SRSA. These include

phenomenological data such as meteorological reports or seismological reports. To be

useful, Berthlin mentioned that these reports have to be interpreted by domain

experts before they can be included in the material provided to the decision maker.

He clarified that he sees the role of such alerts to serve as an extra source of warning

that can either start a process of collection of additional information or “support a

theory as to whether an event require a response from the SRSA”. He continued to

state that these types of alerts have to be received within one hour to be useful. In

addition, of the tools that he has seen he said that he knew that the usefulness of the

hazard data that is dramatically improved if it is coupled with demographic data.

Berthlin concluded that “when these two types of data are combined in a timely alert

it will enable the domain expert to identify cases in which it is certain that there will

be no need for our assistance”. In uncertain cases he could see the alert as being

useful to trigger the intelligence gathering process. No matter what, Berthlin finished,

“the alert has to be with us within an hour, to be of use”.

8.2 Co-ordinating organisation Within the UN some of the main institutions concerned with earthquake

disaster management are the Disaster Management Training Programme (UNDMTP),

the International Strategy for Disaster Reduction (UNISDR), the Development

Programme (UNDP), the Human Settlements Programme (UNHABITAT) and the

OCHA. Of the UN institutions, OCHA is the one with the greatest involvement in

sudden-onset disaster response and the development of supportive tools and

- 127 -

methods. The role of OCHA with regards to information sharing is best summed up

by the United Nations Disaster Assessment and Coordination (UNDAC) handbook:

OCHA is the principal organization through which information on the humanitarian situation is gathered and analyzed. OCHA is also, therefore, responsible for regularly - communicating the results of the analysis to interested parties such as emergency responders, donors and the media, in the form of regular situation reports and briefings. (UNDAC 2000:B3.3)

As mentioned in this quotation, the main communication medium used by

OCHA is the ‘Situation Reports’ (sitreps). The sitreps are based on reports provided

by other organisations. Government bodies and in-country international

organisations provide reports in which they give their loss assessment and estimation

of needs. In emergencies it is the aim of OCHA to release a daily sitrep (UNDAC

2000), though this is governed by the intensity of the information flow and indirectly

by the speed which the emergency is developing. Slower onset disasters and disasters

with little international interest, i.e. forgotten crises, generally have fewer sitreps

written about them. There have, however, been cases where low intensity in the

information flow or high uncertainty in the information has affected the frequency in

which sitreps are issued. In one of the case studies, the 1994 Mazar-I-Sharif,

Afghanistan earthquake, the limited international presence combined with the

attention of international media being absorbed by a concurring natural disaster in

Bangladesh, are likely to have resulted in a reduced number of sitreps.

Within OCHA, the Emergency Services Branch (ESB)23 is the main body

involved in the response to sudden-onset disasters. The ESB in Geneva maintains a

non-stop duty officer system to be prepared to take emergency calls and to alert the

international community of an unfolding event (UNDAC 2000). For loss assessment,

needs assessment and co-ordination of the international response the ESB has set up

its Field Coordination Support Section (FCSS). According to the FCSS website its

main purpose is to:

develop, prepare and maintain stand-by capacity for rapid deployment to sudden-onset emergencies in order to support the authorities of the affected country and the United Nations Resident Coordinator in carrying out rapid assessment of priority needs and in coordinating international relief on-site. (OCHA 2006)

The 2000 UNDAC handbook mentions the On-Site Operations Coordination

Centres (OSOCC) as one of the tools that the FCSS use to achieve this goal. The

23 formerly the Disaster Response Branch

- 128 -

methods of the FCSS have, however, progressed since the publication of the UNDAC

handbook. Currently, the FCSS have the following assets at their disposal:

• The UNDAC team, which can establish an OSOCC on request.

• The International Search and Rescue Advisory Group (INSARAG)

• The Virtual On-Site Operations Coordination Centre (VOSOCC)

• OCHA's stand-by partners

The UNDAC teams are stand-by teams of disaster management professionals

who are nominated and funded by member governments, OCHA, UNDP and

operational humanitarian United Nations Agencies such as the World Food

Programme (WFP), UNICEF and the World Health Organisation (WHO). The OCHA

website describes their role as:

Upon request of a disaster-stricken country, the UNDAC team can be deployed within hours to carry out rapid assessment of priority needs and to support national authorities and the United Nations Resident Coordinator to coordinate international relief on-site. (OCHA 2006)

The UNDAC team is also responsible for collecting on-site information on the

situation to be disseminated to the international community through the sitrep

created by OCHA (UNDAC 2000:D5.1). The UNDAC team is expected to supply

OCHA with input to the sitrep on a daily basis. According to the UNDAC Field

Handbook (2000:E2) once in place in the affected country the UNDAC team will

perform an initial assessment in the following order: a general situation assessment

including estimation of losses, needs assessment and an in-depth sectoral assessment.

In developing countries lacking domestic expertise the UNDAC initial assessment is

commonly the first formal assessment that becomes available to the international

community.

INSARAG is a global network of more than 80 countries and disaster response

organisations involved in Urban Search And Rescue (USAR). INSARAG includes

earthquake-prone countries as well as organisations and countries that are providing

relief. A central task for INSARAG is to establish standards for international USAR

teams and to develop procedures for international co-ordination in earthquake

response. As part of this effort, INSARAG has developed the VOSOCC - an on-line

information exchange and co-ordination tool. The VOSOCC is primarily focused on

supporting the co-ordination of emergencies requiring a SAR response. The website

- 129 -

provides a notice board on which organisations can interchange relevant textual

information on needed and dispatched relief. During 2006, VOSOCC will migrate to a

new platform and join forces with GDACS to provide near real-time disaster alerts. A

screenshot of the beta-version of the new system is shown on Plate 9.1 (page 146). As

of yet the VOSOCC only provides alerts via email or SMS as new threads are posted

by users on the notice board. An additional asset that is not part of the OCHA

organisation is the ‘Resident Co-ordinator’. As the official representative of the UN

Secretary General, the Resident Co-ordinator leads the permanent UN country team,

the ‘resident co-ordinator system’, in day-to-day development operations. In the

words of the UN Administrative Committee on Coordination (ACC)

The resident coordinator system aims at improving the efficiency and effectiveness of operational activities at the field level, through a coordinated multidisciplinary approach to the needs of recipient countries under the leadership of the resident coordinator. (ACC 1995:1)

It should be stressed that co-ordination of sudden-onset emergencies is not the

main purpose of the resident co-ordinator system. This is suggested by the ACC when

they state that “The resident coordinator should normally coordinate the

humanitarian assistance of the United Nations system at the country level” (ACC

1995:5). Sudden-onset extreme events could fall out of what can be considered

‘normal’ and the capacity and expertise of the resident co-ordinator system might not

be well suited for such operations. In relation to the domain of responsibility, the

official aim as stated by the ACC is that the system should be targeted at achieving “a

better co-ordination of operational activities for development” (ACC 1995:1).

Although appropriate disaster relief is part of sustainable development, the ACC

statement does imply longer term operations in the phases of recovery, mitigation and

preparedness.

8.3 Funding organisation The European Commission’s main tool in the response to sudden-onset natural

or human-made disasters is the Rapid Reaction Mechanism (RRM), formally

described in European council regulation 381/2001 (OJEC 2001). The RRM was

created in 2001 as a mean for the European Commission to rapidly respond with

financial grants to projects that work to ameliorate the negative effects in the

aftermath of disasters in countries outside the European Union (OJEC 2001). The total

budget for 2005 was €30 million (RELEX 2006). The regulation specifies a range of

- 130 -

requirements needed to be fulfilled by both the disaster and the recipient. Under no

circumstances may the RRM be used to fund projects that otherwise could be funded

by conventional ECHO budget lines. The response must be of such an urgent nature

that the regular funding process is unable to react in due time. The funded project

must be limited in scope and time and regulations stress the importance of co-

ordination with other organisations responding to the event. There are internal policy

guidelines in ECHO (Billing 2004:8-11) that outline the triggers that are used for the

entry decision. The categories of considered information can be summarised as:

• In-situ assessments: In-country experts with the ability to communicate loss

and needs estimates to the ECHO office in Brussels provide an important

input to the decision process. “Their [the in-situ ECHO experts’] assessment,

complemented by assessments undertaken by partners and sitreps of

international organisations, EC Delegations and NGOs present in the affected

area should be used to define the level of needs…”(Billing 2004:9).

• Affected government actions: A declaration of a state of emergency by the

affected nation can be a sign of need, though Billing points out that there are

cases where this need is not genuine or when genuine need does not result in a

state of emergency and the legal definition of a state of emergency and

conditions under which it can be declared vary from one country to another.

“… a government may not be willing to declare a state of emergency (e.g. in

the case of armed conflict) even if one part of the population is under serious

threat or suffering. In other cases a country might be tempted to declare a state

of emergency simply to attract foreign assistance“(Billing 2004:9). In most

cases a request for international relief is required before any response is

mounted. “Calls for international assistance would normally be broadcast by

the national government. In the case of weak states, or failed states […] the

request for international assistance may come from the ICRC or another

international organisation present” (Billing 2004:9).

• Proxy assessment of vulnerability: “ECHO´s GNA may be appropriate

instruments to gauge vulnerability as they reflect lack of resources to face

hazards, assuming that the higher the degree of development in a given

country, the higher the capacity of that country's people to deal with

humanitarian suffering” (Billing 2004:9). The level of disaster preparedness of

- 131 -

the affected country gives an indication of the degree to which the event can

be dealt with internally. According to Billing it is likely that countries with

frequent disasters are more likely to require international assistance. “It is

important to gauge the level of an affected population’s organisational

capacity to carry out effective disaster preparedness and response

programmes” (Billing 2004:9).

The above points help to develop an estimation of the need for international

relief. The entry decision is also affected by an additional set of contextual factors that

are not related to the need.

• Availability of funds

• Coverage by other donors: “If the needs have been covered by other donors,

ECHO may decide not to intervene at all or to intervene on a small scale

focusing on unmet or forgotten needs” (Billing 2004:10).

• Absorption capacity of recipient community: The level of support will depend

on the availability of the present partners’ ability to implement activities to the

extent of allocated funds. When ECHO is already active in the area, and if the

new operation is small and limited in scope, ECHO can envisage funding

more easily and rapidly because the project can fit within a larger operation

that only needs to be slightly adjusted to the situation. “The speed and level of

intervention can be assumed to be higher if ECHO has previous experience in

that country” (Billing 2004:11).

• Intervention cost/benefit ratio: Billing mentions that negative side-effects of an

intervention, most commonly of a political or environmental nature, should

guide the entry decision. The political impact is particularly relevant in

situations where there is an ongoing conflict in the affected area.

• Access and security: It may be impossible to access certain affected

populations due to restrictions in movement imposed on humanitarian

agencies by governments or warring factions. Billing sees that even if access is

authorised, the security of implementing partners may be so precarious as to

render an intervention unfeasible.

- 132 -

8.4 Systems investigation summary The systems investigation has presented the workflow in the studied

organisations. The user requirements on alert systems were identified. Although

these requirements were relatively clear in the implementing organisation, this was

not the case in the co-ordinating and funding organisations. For timeliness, the

implementing organisations required an alert to be received by then within an hour,

so that an entry decision could be taken within six hours. The level of accuracy

required was not important to the implementing organisation as long as the level of

accuracy was known. The required content of the notification will depend on which

decision that it will support, which will be elaborated upon in the systems analysis

chapter. These requirements are forwarded to the systems analysis stage where

alternative solutions fulfilling these requirements will be sought.

- 133 -

9 SYSTEMS ANALYSIS The purpose of the system analysis stage is to evaluate various alternatives for

supporting the users. Like the systems investigation it is hence part of the first

objective of the thesis. Instead of supplying decision makers with all available

information, it is important to identify which types of data are required to provide

useful decision support in a timely manner (Currion 2003).

9.1 Analysis of alternatives To give structure to this task and to summarise the questions requiring answers,

Table 9.1 shows the decision sequence based on the combination of practitioner

interviews, observations and the previously presented theories from domains of

development and disaster management (Darcy and Hofmann 2004, Glantz 2004, Kent

1984) and information management (Kersten 1999, O’Brien 1999:456, Smart 2005 and

Andersen and Gottschalk 2001). The emphasis of this research project is on the initial

tasks in Table 9.1: hazard alert, loss assessment and needs assessment. The tactical

decisions are the decisions related to the “entry decision” (see section 1.2).

Table 9.1 The decision sequence in international disaster relief

Time Phase Task Decision-maker

Question/Decision

Hazard alert Phenomena Experts

What is the nature and location of the hazard?

Loss assessment

What is the humanitarian impact?

Is international relief required? What is the optimal nature of the relief?

Disaster impact

Needs assessment

Tactical

What is the optimal scale of the relief? Which are the affected areas? How should the aid be prioritised between the areas? How should the aid be delivered?

Response Co-ordination Operational

When is the emergency phase over? What were the lessons learned from the response?

Recovery/ Mitigation

Policy creation

Strategic

Should policies with regards to response, recovery, mitigation and prevention be changed?

Source: Author; Figure 2.1, Table 3.1, personal communication with Berthlin

The initial tasks are related to the first question stated by Kent (1987:136): Has a

disaster occurred? But what is a disaster? The philosophical constituents of a disaster

are something that has been analysed in depth in other studies (Quarantelli 1998). In

- 134 -

the practitioner interviews Berthlin defined a disaster as an event that requires swift

intervention from his organisation. This question is closely linked to the subsequent

tactical decisions. A clearer definition of ‘disaster’ is required to answer whether a

disaster has occurred.

Following the intention of this project to encompass sudden-onset disasters in

general, the subsequent tactical level decisions are better suited for analysis. This

builds on the assumption that tactical decision support, particularly needs assessment,

is independent of the hazard type. This is not true for the first of the tactical

questions, the loss assessment, which is hazard-dependent (Whitman et al 2004;

Shakhramanian et al 2000). There are, however, signs that the subsequent tactical

decisions can be made hazard independent (Olsen et al 2003; Albala-Bertrand

1993:141). Is decision support in the tactical tasks desired by the decision makers?

The relevance of tactical decision support is reflected in the interviews as well as in

the literature. For instance, in the interviews, Berthlin stated that phenomena data are

more helpful in the decision process when combined with socio-economic data. The

focus is on the first decision in the needs assessment task: Is international relief

required?

9.1.1 A source evaluation framework Which sources and types of information are best suited for supporting the

question on whether international relief is required? To answer this question the

INTEREST database was analysed for the sequence of information that was made

available by sources following the 59 case studies.

The time of availability following the disaster impact for each information type

was determined through content analysis of the collected reports. In the majority of

the case studies no evidence could be found of remote sensing or loss assessment

models having been applied by decision makers24. There were no indications of

decision makers having used more advanced DSS operationally. There were,

however, cases where DSS was used for research purposes. An example is the 2001

Badin/Gujarat earthquake. In all those events, the DSS were only tested and thus not

used for making operational decisions. The systems presented in section 4.2.2 were

24 In the interviews, Berthlin mentioned the use of pager alerts that were issued based on a magnitude threshold as soon as the USGS provided a report on an event. This is a hazard alert and not a loss assessment.

- 135 -

tested on the case studies after the INTEREST database had been completed. The

GDACS system was running in real-time for testing during 2003 and it was examined

for both timeliness and accuracy while the more recent QUAKELOSS system only

could be tested for accuracy. The PAGER system was left out of the test due to lack of

access to this recently developed tool. Because remote sensing was not used in the

case studies, the research of Al-Khudhairy and Giada (2002) was used to provide

input on timeliness and information content of remotely sensed imagery.

The timeliness and suitability of the information types is analysed by identifying

the extremes in the case studies, i.e. the fastest and the slowest time of availability of

an information type in the case studies. In the examination this is coupled with

measures of quality of the supplied data and information based on the analytical

framework presented in section 5.3.2. The applied definitions of the information

quality used in the examination are presented in Table 9.2.

Table 9.2 Definition of applied terminology for data quality

Quality Definition None Low Intermediate High Accuracy Percentage of the studied cases

where the reported value, when taking into account the reported confidence interval, did correspond to the final value.

N/A <60% 61-80% >80%

Completeness Percentage of the studied cases where the accumulated information was sufficient to determine: -Disaster: Whether to intervene -Need: The nature and scale of an appropriate intervention

0% <60% 61-80% >80%

Source: Author

Information quality is analysed based on accuracy and completeness. The

accuracy indicates how well the data collected at a certain stage following an event

corresponds to the reality intended to be measured, for instance how well the first

reported hypocentral depth corresponds to the final depth. The completeness of

information is a measure of how well the accumulated information covers the

information needed by a decision maker to make a fully informed decision. The

initial needs assessment questions proposed by Darcy and Hofmann (2003) are used

as references for completeness. The first question, labelled ‘Disaster’ in Table 9.2,

queries whether the decision maker is able to determine if international relief is

required. The second question, labelled ‘Need’, queries whether the decision maker

- 136 -

can determine “The nature and scale of an appropriate intervention” (Darcy and

Hofmann 2003:6). The determination of completeness is a subjective task. Arguably,

only the decision maker can determine if he or she felt fully informed at a certain time

following an event with regards to a specific question. For the qualitative

information, like situational accounts, the accuracy and completeness were hence

determined based on input from the practitioner interviews. The collated set of

information groups and their corresponding timeliness, accuracy and completeness

are listed in Table 9.3.

Based on the data stored in the INTEREST database in the case studies, the chain

of events outlined in Table 9.3 is the following: When the earthquake strikes the

affected population will be the first to notice the effects of the event. Shortly after,

seismological institutions will record seismic data. Mass media and local government

will receive initial information from the affected population. Occasionally, large

organisations have permanent on-site representatives that dispatch situation reports

to their employers. To minimise the delay and increase the objectivity of the

information in the early stages, the international organisations may refer to one of

several existing techniques for conducting formal loss and needs assessments

remotely. In the last stage, data from satellite platforms becomes available. Academic

and esoteric reports like EERI (2003), Kaji (1998) and IFRC (1993; 1995) combine

information from all sources into summarised final reports.

Considerations For several information types the case study data were insufficient to pinpoint

the time of availability. In addition, even when a report contained meta-data on when

it was produced, in no case does meta-data indicate when the decision maker received

it. For instance, reports from the media rarely contain more time-related meta-data

than the date of release. In such cases, Table 9.3 indicates only the unit of time within

which data were made available to decision makers. Furthermore, the data and

information produced by the sources in Table 9.3 are not uniquely divisible. The later

in time after an event that a source releases data and information the more data from

preceding sources tend to be included. For instance, academic studies, which are

among the last to appear, include data from all sources. Some sources use the output

of preceding sources to provide value-added information. Marked with ‘red italics‘ in

the table are those sources that fully depend upon baseline data and information from

- 137 -

preceding sources. For those sources, the time-frame of availability is provided as an

increment to the time required for acquiring the data that the source is based upon.

Table 9.3 Data availability and Quality over time

Time of availability

Data Quality

Completeness Data Source

Min Max Accuracy Disaster? Needs?

Epicentre, Magnitude, Time

Remotely sensed seismic data, NEIC.

Seconds Minutes Intermediate Low None

Depth and improved Epicentre and Magnitude

Remotely sensed seismic data, NEIC.

Minutes Hours High

Affected population size estimate

Numerical models + minutes + hours Low Low

Human loss and Structural loss estimates

Numerical models with expert input

+ minutes + hours Intermediate Intermediate

Situational accounts On-site representatives

Minutes Hours High High Intermediate

Textual eye-witness accounts

Media Minutes Days Low

Injured; dead; homeless; buildings and/or villages damaged or destroyed.

Loss assessment by host government

Hours 13 days Intermediate

On-site loss assessment by Co-ordinating body

3 days 4 days High

List of needed relief items and expertise.

Host government appeal

Hours 16 days Intermediate High

On-site needs assessment by Co-ordinating body

3 days 4 days High

List of dispatched material and shortfalls

Co-ordinating body 1 day 6 days High

Post disaster maps for navigational purposes

Remotely sensed optical imagery

2 days Weeks High

Post disaster maps with estimated structural damage

Expert interpreted Remotely sensed optical and radar imagery

+Hours +Weeks Intermediate

Building damage type and cause

Structural survey Weeks Months High

Academic reports Weeks ∞ High

Source: Author; INTEREST Database

9.2 Discussion What are the implications of the results presented in Table 9.3? What

information and sources can potentially support the decision on whether international

relief is required? Table 9.3 contained three groups of sources no dependent on

presence in the affected area: remotely sensed seismic data, remotely sensed imagery

and numerical models. The suitability and timeliness of the information of these

groups are discussed here.

- 138 -

9.2.1 Remotely sensed seismic data Although Berthlin mentioned that his organisation, as well as others, is using

alert systems based only on seismic data, he also admitted that this produced a large

amount of false warnings. Earthquakes can occur in the ocean, in uninhabited areas

or in developed countries that area resilient. All such events would result in false

warnings. The strength of remotely sensed seismic data is its speed. Depending on

the location of the earthquake, initial data will be available in less than an hour,

sometime in less than a minute. Nevertheless, seismic data on its own is insufficient

for providing decision support in the question whether international relief will be

required.

9.2.2 Remotely sensed imagery In which ways can remotely sensed imagery be of help in the decision process in

Table 9.1? An assumption in this analysis is that because this study focuses on

disasters in areas with poor infrastructure, the only source of remotely sensed

imagery are sensors on space-born satellite platforms (Al-Khudhairy and Giada 2002).

Al-Khudhairy et al (2002a) showed that although other platforms such as airplanes

and helicopters have to be hired and sent to the affected area, this can be very costly

with respect to both time and money. In addition to being faster and cheaper,

satellites have an advantage in that they circumvent the unwillingness of some states

to have their territory examined by airborne means. Based on the projects presented

in section 4.1.2 (Eguchi et al 2003, Al-Khudhairy et al 2003, Mehrotra et al 2003) it is

clear that there are two main uses for remotely sensed imagery in the response phase

following a sudden-onset disaster.

• Navigation: in case of insufficient access to up-to-date maps, remotely sensed

images can help rescue organisations navigate their way through the disaster

area (Altan 2005; Al-Khudhairy and Giada 2002).

• Loss assessment: using manual and automated methods, the images can be

analysed in order to detect where damage has been inflicted and to what

extent (Al-Khudhairy et al 2002b; Eguchi et al 2003).

When replacing the use of a map, optical images are better suited than radar

imagery, as the former are easier for an inexperienced user to comprehend (Campbell

2002:209-241). However, optical remote sensing requires daylight and the absence of

clouds (Campbell 2002:157-171). These limitations can cause a delay in the delivery of

- 139 -

the image (Al-Khudhairy and Giada 2002). If the image is to be used solely for

navigation or as a pre-event reference image Al-Khudhairy and Giada (2002) showed

that it often is possible to find copies in the archives of the image providers that can be

delivered without delay. The main weakness of non-optical, i.e. radar, images is that

they require expert interpretation before being used for any purpose and that the

resolution is lower than that of many optical sensors (Campbell 2002:209-241; Al-

Khudhairy and Giada 2002). The main benefit of radar is that it functions in all light

conditions and regardless of the presence of clouds or smoke. However, in their case

study Al-Khudhairy and Giada (2002) showed that expert interpretation can delay the

delivery of an optical or radar image with several days, which in a real scenario could

render the output information useless. The pros and cons of these methods of remote

sensing are summarised in Table 9.4.

Table 9.4 Pros and cons of remote sensing alternatives

Pro Con Radar Works in darkness, through

smoke and clouds. Coarse resolution. Requires expert interpretation. Loss assessment not feasible.

Optical Single image

Can replace a map for navigation purposes.

Does not allow for loss assessment. Requires day-light and line-of-sight.

Image pair Provides indication of where damages have been made to structures.

Takes time to acquire and requires processing for loss assessment. Does not show damage to vertical parts of structures.

Will remotely sensed imagery answer whether international relief is required? Remote sensing imagery analysed with automated loss assessment models

provide an estimate of the humanitarian impact; answering the first of the tactical

questions in Table 9.1. However, as discussed in section 4.1.2, automated loss

assessment requires an image pair for the output to be accurate and those take time to

acquire, which is reflected in Table 9.3.

From Table 9.3 it is clear that a decision maker will seldom be able to take

advantage of remotely sensed imagery in the immediate aftermath of a disaster,

including the decision if international relief is required. This is mainly due to the

amount of time required to acquire and interpret an image pair (Al-Khudhairy and

Giada 2002). In the interviews, Berthlin sets a requirement of six hours for the

availability of loss estimations. For the decision maker to benefit from remotely

sensed imagery, one currently has to resort to using a pre-event image of high-

- 140 -

resolution, if available (Al-Khudhairy et al 2002a). Such an image will only be useful

for navigation purposes and possibly to provide an overview of settlements located in

remote areas (Al-Khudhairy et al 2002b). In cases where existing mapping is

inadequate, remotely sensed imagery can be useful in supporting the operational

decision makers in logistic tasks.

However, remote sensing carries potential to the relief effort if the international

response is protracted. If a decision to respond is not taken within a couple of days, it

will be feasible to consider the use of analysed image pairs (Al-Khudhairy and Giada

2002; Al-Khudhairy et al 2003). An example of a situation where a decision can take

some time is a case with a widespread affected area or a case with damage to local

infrastructure that inhibits the ability of launching reconnaissance efforts on the

ground. In these cases, the time required to process and analyse the images is

preferable to the number of days that would be required to reach all the areas by land.

It is important to remember that even under optimal conditions, remote sensing can at

best only assist in navigation or in approximate loss estimation; on-site detailed needs

assessment will have to be conducted as an input to the operational decision making

procedure. This assessment can, however, be better targeted if it is prioritised to areas

expected to have experienced severe losses based on the remote sensing loss

assessment.

9.2.3 Numerical models If remotely sensed imagery is not useful for supporting tactical decisions in a

typical sudden-onset disaster the alternative solution for the remotely located decision

maker is to make the most out of on-site sources combined with numerical loss and

needs assessment models. The numerical models presented in section 4.2.2 are

analysed in further depth here with regards to their timeliness in Table 9.3 and to

their content. The PAGER system was left out of the analysis due to lack of access to

this recently developed tool.

Global Disaster Alert And Coordination System The GDACS alert is delivered either in an SMS or in an email. This includes a

link to an online report developed based on O’Brien’s (1999) methods for information

presentation: consolidation, drill-down and slicing and dicing. The alert is provided

in a qualitative ‘level’ that portrays the seriousness of the event (De Groeve and

Eriksson 2005) in three degrees: Red, Orange and Green (see example Plate 9.2 on

- 141 -

). For the analysed case studies in which a GDACS alert had been issued, the

alerts were issued within 30 minutes of the NEIC release of earthquake data.

The accuracy of the tool is harder to measure due to it not being clear what the

output ‘alert level’ should be compared to. Seriousness is a subjective measure. In

Figure 9.1 the alert levels issued by GDACS for the case studies are juxtaposed with

the sum of human losses. Although the variation in the data is great, it is clear that

events with higher human losses are more likely to be classified with a higher alert

level. De Groeve and Eriksson (2005) analyse the accuracy of the tool deeper. Their

report is clear in stating that the tool is not a quantitative loss assessment tool, but a

qualitative alert tool. However, what speaks against them is their use of quantitative

loss data for validation of the model.

Error Bars s how Mean +/- 1.0 SD

Bars show Means

Alert Level

f Kill

94 1340 1892

Figure 9.1 Average number of dead and injured per alert level

Using loss data for validation De Groeve and Eriksson (2005) find the tool to be

correct in 65 percent of their test cases. In 18 percent of the cases, events observed to

be serious were incorrectly classified as green. This type of classification error,

omission error, is the most serious classification error because it delays the decision

that it is to support. These errors could be the result of the use of data on loss for

calibration of a tool that does not claim to predict losses.

The tool does not supply the user with a level of confidence in the issued alert.

Berthlin saw an indication of “the level of clarity […] in the situation” as an important

output of a tool. An advantage is that GDACS does live up to requirement of

transparency set out by Darcy and Hofmann (2003), King (2005) and Glantz (2004).

The tool is well documented (see for instance De Groeve and Ehrlich 2003; De Groeve

and Eriksson 2005) with the underlying methods and baseline data being declared.

- 142 -

The relationship between the researcher’s unit, and the potential user

organisation, was very close throughout the three years that the researcher was based

at the JRC. The JRC provided support in the development of information

management tools for use in the ECHO. In the development of an alert tool for

earthquakes (see De Groeve and Ehrlich 2002) it proved difficult to please the end

users of the system. After the launch of the first real-time alert system in 2003 it

quickly became apparent that the tolerance of false warnings, particularly commission

errors, was unexpectedly low among the users. Disgruntled users deactivated the

SMS alert service on their duty phones following alerts at night or alerts of events

which were not deemed relevant. This seemed to take place immediately or after a

few errors by the system without feedback to the development team. This situation

was worsened by the attempt to develop loss estimation functionality for the alert

tool. Due to the lack of detailed socio-economic data for the developing countries the

estimations were very approximate (De Groeve and Eriksson 2005). The attempts to

convey the uncertainty using estimations in ranges made the output more complex

and less user-friendly. The system in use at the time did not take the national

vulnerability into account, which caused alerts to be issued for serious earthquakes in

for instance Japan. The solution was a continuous improvement of the accuracy of the

alert tool, an effort which this research project was part of. Added functionality for

limiting the times of day for when the alerts were to be broadcasted gave individual

users the option of postponing alerts being issued off hours. As the GDACS

improved, it started to suffer from Norman’s (1998) “creeping featurism” syndrome

(see section 3.3) with many small pieces of added functionality gradually reducing

usability. Seeing that GDACS is developed by a scientific organisation, this problem

could be caused by a case of Norman’s (1998) “worshipping of false images”, which is

often seen in techno-centric development. GDACS is developed in cooperation with

FCSS, which is a practitioner co-ordinating organisation. This cooperation should

influence the end result to the better and help prevent a reduction in usability.

QUAKELOSS The QUAKELOSS tool does not offer an interactive user interface and relies on

manual telephone calls and static email output including delivering a map (see Plate

4.3 on page 45) and quantitative human and structural loss estimates (see Table 4.3)

(Wyss 2005). The first output delivered by QUAKELOSS is enhanced by phenomena

experts and converted into qualitative loss estimates before being delivered by phone

- 143 -

or email (see Table 4.3). This requires that phenomena experts are available on stand-

by for the generation of the alert. The use of telephone calls for the alert limits the

number of possible users, though it is a preferable form of communication. The heavy

reliance on human involvement requires large investment to set up and maintain the

organisation. Furthermore, an expansion into other hazards would dramatically

increase the required body of staff on call. In his first 2004 report Wyss tests the

accuracy of the model. In contrast to the GDACS model, the QUAKELOSS model is a

quantitative loss assessment model. Determination of accuracy is hence more

straight-forward. Wyss classify events according to the number of killed as either

major (more than 1 000 people), small to moderate (200 to 999 people), or as no

disaster (less than 200 people). In his testing Wyss define the levels of accuracy in his

prognosis as:

(1) Correct estimates are defined as those for which the reported number of fatalities lies within the formal two standard deviation range[…](2) For earthquake disasters with fewer than 200 fatalities, immediate international rescue assistance is almost never needed. Therefore, estimates for which the minimum or maximum lies within fewer than 200 fatalities of the reported numbers are judged to be acceptable. (3) For major disasters, exact numbers of expected fatalities are not needed. Therefore, estimates for which the range of calculated values lies within a factor of two from the reported fatalities are classified as acceptable. For extreme disasters, this latter rule may be relaxed to accept any estimate exceeding 2000 fatalities as correct, if the reported number is larger, regardless of how large it is because with an estimate of 2000 the rescue agency will have to mobilize in any case. (2004a:7)

With this definition Wyss estimates his tool to be correct in predicting 71 percent

of the major events and 58 percent of the small to moderate events. QUAKELOSS

does include an indication of the level of uncertainty. All fatality figures are provided

in ranges with a minimum and maximum that conveys the level of certainty that the

tool provides. With regard to its transparency, although Wyss (2004a:8) provides a

“Brief Summary of the Method for Calculating Losses”, the baseline data and the

applied methods remain a mystery. The relation between QUAKELOSS and its

Russian predecessor developed by Shakhramanian et al (2000) is also unclear. The

tool can hence not be seen as being transparent.

9.3 Systems analysis summary In his interview, Berthlin mentioned a limit or 6 hours within which a decision

to deploy a SAR asset has to be taken. According to Table 9.3, within this time-limit

the potential sources of support include numerical models. Other sources that

possibly could deliver information include media, government and country

- 144 -

representatives. However, the accuracy of reports by media and local government are

questionable (Billing 2003; Fischer 1998). In the first hours after a disaster it is

unrealistic to expect that a country representative, in a context of damaged and

originally imperfect infrastructure, will be in a position to have an overview of the

situation and be able to communicate this to a decision maker. Although they are

unlikely to be fast enough to provide an alert, these sources can be useful for calling

off a disaster alert made by a numerical model. If the initial sources, i.e. the numerical

models, indicate it being possible that an unfolding event is a disaster, whilst the mass

media, government officials and ambassadors claim that little or no damage has

occurred, one could conclude that the on-site sources are more likely to be correct. If

on the contrary there are direct indications from the on-site sources of damage or if

one of Berthlin’s ‘information black-holes’ arises, there is good reason to put response

resources on high alert while the investigation continues.

The sources becoming available, between numerical model output and the

satellite-based assessments can therefore assist in excluding non-disaster events. It is

important to be clear that although information sources provide output of differing

quality, a source providing higher information accuracy is not necessarily better than

sources of inferior accuracy. The purpose of the faster sources can be to alert the more

exact and time consuming sources of an event that perhaps could be a disaster, i.e.

Glantz’s (2003) cascading alerts. The output of the alert systems can be improved

using human expertise. However, using human experts requires time and, in the end,

what governs quality of the output of the models is the readiness to invest time and

human expertise to refine the quality of the input and output data.

To be effective, a decision on whether to intervene and how to intervene in the

aftermath of a disaster, an “entry” decision, has to be taken within a very limited

amount of time. For the most time sensitive forms of relief, like SAR, interviews have

given an indication that the decision to mobilise has to be taken within six hours.

Through discussions with both relief organisations and funding organisations it was

clear that within this window of time it is rare that the decision maker has access to an

accurate source with the complete information required for a decision. It also became

clear that the most valued sources for estimating the requirement of an international

response are in-country contacts such as country representatives paired with

information from relief networks, i.e. the OCHA Virtual OSOCC. If no direct

- 145 -

communication with a source that has correct and indisputable information on the

disaster situation is possible, an “entry” decision has to be taken based on the

incomplete and inaccurate information at hand. Model-based DSS are hence of

importance for “entry” decisions in that they provide an early alert that enables other

sources to provide more refined information. Human experts can improve the output

of the models, but this will be at the cost of time.

Remotely sensed imagery will only be useful for the tactical decision maker if

(a.) the time required to make the analysed material available to the decision maker is

reduced to a matter of hours; or (b.) the area of interest is so remote or widespread

that the time required for on-site reports exceeds that of acquiring and interpreting

remotely sensed imagery.

- 146 -

Source: VOSOCC 2006

Plate 9.1 Virtual OSOCC screenshot from the October 2005 response to the Pakistan/India earthquake

- 147 -

Source: GDACS 2006b

Plate 9.2 GDACS email alert for an April 2006 earthquake in DR Congo

- 148 -

10 SYSTEMS DESIGN AND IMPLEMENTATION The systems design stage is when the data are collected and prepared for the

development of the application. In the systems implementation stage the application

is developed. This chapter adopts its headings from the KDD process (see Figure 5.1)

created by Mahadevan et al (2000). The preceding chapter covered the tasks outlined

by the ‘problem definition’ phase in the KDD process. The first phase covered here

closes the problem definition.

10.1 Problem definition In the preceding chapter it was made clear that automated alert models can be

beneficial to the international relief process, but that the users dislike output that is

overly complex or of unknown accuracy. The loss and needs estimates are, however,

complex due to the necessary inclusion of confidence intervals to communicate the

information certainty requested by some users in the interviews and observations.

Furthermore, the lack of transparency in existing systems has been shown to be a

major concern amongst practitioners (Darcy and Hofmann 2003; King 2005; Glantz

2004). The question to be supported identified in section 9.1 was whether

international relief will be required. A potential solution to the problem of complexity

in the alert is to create an indicator that communicates the probability of an

international response. Such an indicator will only occupy one dimension. As such it

will be easy to comprehend and useful for automated alert systems. Figure 10.1

illustrates the logic applied in such a prognostic model. The arrows on top represent

the process in a conventional model and the lower arrow representing the novel

model. Instead of attempting to predict the incurred losses and the subsequent need

for international relief, the suggested model aims to predict the international

response.

- 149 -

Source: Author

Figure 10.1 Conceptualisation of proposed prognostic model

Although the model output can be a probability, the data used for developing

the prognostic model have to be certain. The attribute used for developing the model

has to be a crisp quantitative representation of the size of the international response.

By identifying the characteristics of case studies that have received large international

response it will be possible to predict which future events that will receive

international attention. However, the measure of ‘size’ of an international

intervention is not clear. The pragmatic measurements of ‘size’ are the financial cost

of the implemented international relief and possibly the amount of dispatched aid

material. Alexander warns of the dangers in attempting to classify disasters according

their ‘size’:

Many attempts have been made to quantify disasters, and to invent classifications and taxonomies. I must admit that it is a lure to which I am far from insensitive. Yet most disaster taxonomies are either facile or inoperable. What should they be based upon? Numbers of deaths and injuries? The dollar value of damage? The sum of total human misery? No combination of factors is without snags. (2000a:192)

None of the data mentioned by Alexander is complete for the case studies. The

reported currencies and the tagging of the financial aid, e.g. earmarking, for spending

on donor country services only, made a straight comparison awkward. Clear

financial donation data are available for most large scale interventions. This could of

course be interpreted to mean that the international community only respond to

certain events and when doing so responding on a large scale. Other data, like the

amount of dispatched relief material pointed to this interpretation being false.

Subsequent attempts at quantifying the international response included the analysis

of individual types of donated material e.g. donated tents, blankets, field hospitals.

Once again, however, these indicators proved to contradict each other.

- 150 -

Frequency analysis Neuman (2000:294) suggests that frequency analysis is used when data collected

through content analysis is unsuitable for stand-alone analysis. Adopting this

approach as a proxy indicator of ‘size’, the reporting frequency of all reporting

sources was tested against the absolute sums of financial aid and common item and

service donations. The frequency was analysed on each reporting level: attribute,

report and event (see Table 5.4 for definitions). This approach provided promising

results. The frequency of the OCHA situation report, the sitrep, proved to show

overall relationship to the majority of the absolute-figure indicators of ‘size’ initially

tested. Figure 10.2 shows a matrix of scatter-plots displaying the relation between

sitreps and financial aid and human loss for the cases with an international financial

response.

1000080006000400020000

FinAid

70006000500040003000200010000

LossTotal

70006000500040003000200010000

LossTotal

Figure 10.2 Scatter-plot matrix of OCHA sitreps, Financial aid and Human loss

Having established the frequency of sitreps as a candidate indicator of size to be

used as a Dependent Variable (DV), the research could continue to the data selection

phase in Mahadevan et al’s (2000) KDD process to identify the Independent Variables

(IV) used to predict the outcome on the DV. For the sake of clarity, the DV, i.e. the

frequency of sitreps, is also referred to as ‘International Attention’ from here on.

- 151 -

10.2 Data selection The selection of IVs is a sensitive task. It is clear that in reality international

attention does not depend on a handful of characteristics. On the contrary, the list of

factors can be made very long. However, if too many IVs are included in the model

development it creates empty cells. An empty cell is a combination of attributes that is

not represented by an event in the studied sample. Empty cells reduce the power of

the statistical analysis. The selection of data in the development and evaluation of the

GDACS earthquake alert tool by De Groeve and Eriksson (2005) provided an

important input in the selection of data in this model development. Existing models

and tools are fairly consistent and similar to the GDACS model in their selection of

indicators. To create an overview of the common data, Table 10.1 use Schneiderbauer

and Ehrlich (2004) as a basis to classify data according to its purpose in the analysed

models. The table is not exhaustive; it only includes the most common indicators

mentioned in the literature of models and tools (see Wyss 2004b; Shakhramanian et al

2000; Badal et al 2004; De Groeve and Eriksson 2005).

Table 10.1 Classification of indicators, according to purpose

Purpose Indicator Primary proxy Secondary proxy Seismic character

Intensity Magnitude (ML) Hypocentral depth (km)

Population density, Time of impact, Weather

Local geology Slope, land use

Exposure

Building quality GDP, Urban growth, average number of floors.

Vulnerability

Resilience GDP, access to vital resources25

Impact ƒ(Seismic character, Vulnerability) Needs ƒ(Impact, Resilience)

Source: Badal et al 2004; De Groeve and Eriksson 2005; Schneiderbauer and Ehrlich 2004

The indicators to be used in the model development were selected after much

iteration, testing the wide range of absolute and frequency data collected for all the

case studies. As promising indicators were identified they often turned out to be

incomplete, resulting in all the case studies having to be revisited in an attempt to

achieve a complete sample. The challenges in selecting the indicators were: 25 Also known as lifeline e.g. water, food, health services, heating, communications (Wisner et al 2004).

- 152 -

• That only a very limited set of indicators can be used in order to avoid empty

cells (Le 1984) due to the limited size of the case study population.

• That, if used to predict high-level international attention, the indicators should

not separate Afghanistan and Iran.

• That the indicator had to have a logical relationship to the resulting

international attention based on the researchers domain expertise (Mahadevan

et al 2000).

• That the indicators would be available in the immediate aftermath of future

earthquakes in order to make real-time use of the model possible (Beroggi and

Wallace 1995).

In the selection process, the researcher’s experience from the domain was

complemented with results published in the literature and the logic applied in the

existing models (Table 10.1). The prediction of the international response requires

that conventional models of loss and needs are altered or bypassed in accordance with

Figure 10.1. The probable indicators of international response size that could facilitate

a model bypassing the estimations of loss and needs have been suggested by the

literature reviewed in section 2.3, by interviews and by observations, to be:

• Media coverage: Olsen et al (2003) as well as Benthall (1993) suggest that the

media have an important role in affecting the size of the international response

to disasters.

• Political interest: The standing of the host country on the global political arena

is considered by several authors as being an important or even the single most

important factor that currently governs the international response (Olsen et al

2003; Darcy and Hofmann 2003; Dalton et al 2003; Smillie and Minear 2003;

Leader 2000). Donor countries are prone to use disaster relief as a political tool

(Albala-Bertrand 1993). Berthlin highlighted the role of political interest in

events for which a response is initiated after the first six hours.

• International presence: The presence of international organisations as the

disaster strikes tend to give the event increased exposure to the world.

Although this was only considered by Olsen et al (2003) for slow onset and

complex events, it could be of relevance for sudden-onset disasters,

particularly in a complex context. Billing pointed out that the presence of a

- 153 -

competent implementing partner in the affected area changes the inclination to

allocate funding to the event positively.

• Acceptance of aid: Although not substantiated in literature, the discussions

with both Billing and Berthlin gave that the host country’s inclination to

request external aid is a pre-empting factor as to whether there will be any

international response.

The challenge of finding suitable indicators is immense. It is further

complicated by the researcher’s desire to include a maximum of the data which were

collected at high cost of time and effort. The solution was to adopt an inclusive

approach and include an excess of indicators in the statistical analysis. The inclusive

approach allows for a small set of inappropriate indicators to be filtered out in a later

stage of the analysis in accordance with Hosmer and Lemeshow’s (2000:91) model

development process.

Vulnerability and International presence Direct data on international community presence is available for recent years

(Durch 2004), particularly after OCHA started to monitor International NGO (INGO)

presence through their in-country Humanitarian Information Centres (HIC).

However, as the direct data are not available for the historical case studies, a proxy

indicator with complete coverage is required. Assuming that international

development assistance organisations focus their work in vulnerable countries, an

indicator of vulnerability can be used to indicate the presence of international

organisations in an area. This is partially supported by the analysis of Darcy and

Hofmann (2003) who discuss the priorities of the INGO community. Based on this

assumption, to reduce the number of indicators in the model development and thus

the likelihood of empty cells, the composite GNA indicator was used for country

vulnerability as well as international presence. As presented in section 5.2.4, the

ECHO calculates the GNA index for the 130 poorest developing countries (Billing and

Siber 2003). All countries included in this study have a GNA index. The adoption of

the GNA as an indicator of vulnerability makes clear that the researcher is attempting

to separate the vulnerable from the almost as vulnerable. The alternative use of an

economic measurement, like Gross Domestic Product (GDP), would only represent

one facet of vulnerability.

- 154 -

Media coverage, political interest and acceptance of aid In open societies a government’s reluctance to ask for external aid in times of

disaster, independent of the aid relevance, tends to result in public outrage. A recent

example of this is the aftermath of the sinking of the Russian Kursk submarine.

Countries not inclined to request aid in times of need are hence likely to be governed

by despotic regimes with tight control of all facets of society, including media. On the

assumption that political interest of western countries is greater in democratic

countries and that media presence is greater in countries respecting democratic

values, an indicator of press freedom could be used to indicate media coverage,

political interest and acceptance of aid. This is admittedly a paramount set of

assumptions. A generic indicator of global political interest in a single country is

coarse. The status of individual bi-lateral relations varies. Furthermore, international

aid can be used as a tool to democratise despotic regimes and hence increase with

reduced press freedom. Nevertheless, it has to be accepted that the analysis will not

have high resolution due to the uncertainty in the involved data. The ‘World Press

Freedom Index’ (WPFI) is a qualitative indicator of press freedom developed by

‘Reporters Without Borders’ (RWB). The WPFI is based on questionnaires completed

by “local journalists or foreign reporters based in a country, researchers, jurists,

regional specialists…” (RSF 2006). Zero on the score is a society with complete press

freedom. Most countries score between 1 and 100. The included measure is estimated

by RWB for 2005.

Indicators of loss In accordance with Figure 10.1 the estimation of loss is not the focus of this

study. The emphasis should lie on the use of non-loss indicators to predict the

resulting international attention. However, the losses have a logical relation to the

resulting international attention. Until the existing loss assessment tools, like those

presented in section 4.2.2, are working in real-time and providing reliable output, it is

futile to build a model reliant on their output in order to function. Consequently, the

expected losses have to be represented in the model development using data that

currently is available following earthquakes.

The conventional indicators of loss in earthquakes are shown in Table 10.1.

These indicators represent the seismic character of the earthquake combined with

indicators of earthquake-specific vulnerability. The indicators applied here are the

magnitude, hypocentral depth and the population within 50 km of the epicentre. The

- 155 -

urban growth rate is used as an indicator of earthquake vulnerability based on the

findings by Schneiderbauer and Ehrlich (2004). The effect of high urban growth on

rural areas is, however, unclear. Logically, it could affect the vulnerability in either

direction. Reduced rural population means more space for selecting a safe site for a

dwelling, it potentially also increases the availability of materials for construction and

reduce the need for multi-storey structures (Wisner et al 2004:292-303). However,

with urbanisation primarily affecting the younger generations (USAID 2005), the rural

areas could be left with reduced human capacity for community response.

The national level of earthquake preparedness, defined as the affected nation’s

ability to deal with the response to a domestic earthquake disaster, could prove to be

an important predictor of international attention. Preparedness is, however, hard to

measure partly because it is an abstract attribute but also due to the lack of useful

proxy data relating to civil protection and civil defence organisation and spending.

There are alternative proxy indicators such as data on international financial support

to mitigation and preparedness projects, but that would not serve as a good indicator

for the richer countries that do not receive such aid. Membership of the International

Civil Defence Organisation (ICDO) (see Table 10.5) is a very rough indicator of the

existence of a civil defence structure. The correctness of using ICDO membership as a

proxy indicator of preparedness is highly questionable.

An appropriate interim proxy indicator is the exposure of each country to the

occurrence of earthquakes. It is important to point out that preparedness and

exposure are fundamentally different. A high experience of earthquakes might

reduce a country’s preparedness due to fatigue of both domestic and international

resources. Correspondingly, Schneiderbauer and Ehrlich (2004:18) argue that the

development process and, indirectly, the preparedness level, are negatively affected

by each disaster. To measure earthquake exposure for each case study country, the

corresponding frequency of earthquakes stored in the NEIC database since 1980 is

used as an indicator. This will provide an indication of how seasoned the local

population are to earthquakes. The geographical size of a country and the presence of

active faults obviously affect the number of earthquakes that it experiences. It is

assumed that the overall national earthquake exposure is related to the mental

preparedness of the local population.

- 156 -

Generic natural hazard exposure is part of the composite GNA indicator. This

does not result it duplication in the model. The earthquake-specific exposure is

potentially important and its influence is plausible to be important in the estimation of

the aftermath of an earthquake. The above indicators were combined into the final list

of indicators in Table 10.2.

Table 10.2 Selected IVs

Attribute Description Mag The magnitude of the main shock as first reported by the NEIC. Depth The hypocentral depth of the main shock in kilometres. 50kmPop The total population living within 50 kilometres from the epicentre

calculated using the 2003 ORNL Landscan population density raster. EQPrev Earthquake prevalence. The number of earthquake stored in the NEIC

database that the country has experienced since 1980. LocalTime The local time of day of the main shock. GNA The 2004 GNA score. UGrowth National urban growth 2000-2005 as estimated by UN HABITAT. WPFI The 2004 country-level WPFI.

Source: Author

According to the analysis of the temporal availability of data source in section

9.1, the bottlenecks among the selected indicators are the hypocentral depth and the

earthquake magnitude, both which generally becomes available within an hour after

an earthquake. This delay is within the six hour limit of delivery of the alert, as set

out in the interview with Berthlin. It is hence possible to produce an output using

real-time input from future events as they occur. To facilitate their use in the

statistical software package the indicators are abbreviated. In summary, based on

Table 10.1 the purpose of each selected indicator is the following: Mag and Depth

represent the seismic character; 50kmPop, LocalTime and UGrowth represent overall

exposure; EQPrev, GNAAvg and WPFI represent resilience.

10.3 Data standardisation To absorb some of the inaccuracies in the data mentioned above, the variables

are categorised into meaningful categories. The categorisation of the DV in three

levels provides a user friendly cognitive mapping (Norman 1998) to a traffic light

similar to De Groeve and Ehrlich (2002) in GDACS. The model will therefore attempt

to predict international attention in one of three categories, low, medium or high. An

ordinal categorical output requires the use of ordinal regression. A negative aspect of

this is increased sensitivity to any empty cells. Furthermore, the use of scale IVs

instead of categorical or ordinal parameters creates many cells, which due to a small

- 157 -

population of events, results in empty cells. It is hence seldom advantageous to

include more than one scale variable in studies involving categorical IV or DVs (SPSS

2003).

10.3.1 DV categorisation Although a dichotomous categorisation of the DV would reduce the number of

cells, it does not reflect the empirical knowledge. It is preferable for prediction errors

to give commission error, i.e. that low attention events are classified as high attention

events, rather than omission error where high attention events are classified as low

attention. Excessive occurrence of commission errors will, however, result in a ‘cry-

wolf effect’, in which case the users will loose confidence in the output. A

dichotomous categorisation lacks the ability to classify uncertain events into a middle

category, which leads higher rate of omission and commission in both categories. A

three-level ordinal output is hence preferable.

Situation reports The suitability of the OCHA situation report frequency as an indicator of the

size of the attention that the international community gave an event was investigated.

In relation to the geographical spread of the international responders, the number of

reports show distinct pattern. Figure 10.3 is a precursor to Figure 10.2, showing the

relation between the number of sitreps, human loss and international financial aid.

Figure 10.3 Situation reports, human loss and financial aid (n=53)

- 158 -

Three groups of events can be identified in Figure 10.3, low attention events in

lower left, intermediate attention events in the centre and high attention events

stretching to the upper right. These three groups are correlated to the geographical

spread of the extra-national responders. The high attention events, receiving more

than four situation reports, have a wide range of international responders, including

those from other continents. Intermediate attention events, with two to four reports,

show a concentration on regional responses with inter-continental responses being

rare. Low attention events, one or no situation report, are not well covered by the

collected data. All signs point to these events being dealt with on a domestic basis,

resulting in little information on the events in the international domain. These groups

are also related to the amount of financial aid received. The total financial aid in the

case studies does not exceed USD 200 000 for any event that generated fewer than five

sitreps. The final categorisation of the DV is listed in Table 10.3.

Table 10.3 Indicator categorisation

Role Continuous Variable

Categorised Variable

Cate-gories

Category cut points

DV Sitrep AttCat 3 Low (<2 reports), Intermediate and High (>4 reports)

Magnitude MagCat 3 Low (<5), Intermediate and High (>6) Depth Shallow 2 Depths less then 40 km are categorised

as Shallow. 50kmPop Rural 2 Rural (<45 000 persons) and Urban GNA Vulnerable 2 GNA>1,25 is categorised as Vulnerable. UGrowth HighGrowth 2 Growth rate above 4% is categorised as

HighGrowth. Exposure Exposed 2 Exposure >500 is categorised as

Exposed. LocalTime. Night 2 Time after 21:00 and before 07:00 is

categorised as Night.

WPFI Open 2 Nations scoring below 50 are categorised as open.

Source: Author

Limitations A recurring challenge lies in the limited population of case study events. As can

be seen in Table 10.4, all high attention events took place in Afghanistan and Iran.

The inclusion of IVs that filter out Afghanistan and Iran results in a model that

predicts if the event occurred in those countries rather than if the event was of a

character that is likely to receive attention. Notwithstanding this limitation, as long as

the predictions of all three categories are possible without complete separation of the

above countries, there is no need to reduce the output to a dichotomy.

- 159 -

Table 10.4 Summary of case studies per DV categories

Observed Attention Category (DV) Country Low Intermediate High Total Afghanistan 7 3 3 13 China 4 0 0 4 Iran 21 5 4 30 Kazakhstan 1 0 0 1 Kyrgyzstan 0 1 0 1 Pakistan 2 3 0 5 Tajikistan 3 0 0 3 Turkmenistan 1 0 0 1 Total 39 12 7 58

10.3.2 IV categorisation The values on the IVs are categorised based on theoretical and empirical

knowledge of the subject. Post-disaster indicators of loss and international response

are used in the classification of the values on the IVs. The two main indicators used

for the categorisation of the IV values are the total amount of financial aid and the

total human loss, i.e. the sum of the final reports of injured and killed. These two

indicators were also used in the categorisation of the DV. Care has to be taken not to

categorise the indicators in categories that suit the output, rather than categorised that

provide a natural representation of the indicator. The small population of case studies

and the limited number of involved indicators makes this task demanding. Although

the number of sitreps itself, i.e. the DV, has not been an input in the categorisation of

the values on the IVs, the total human loss and foreign financial aid have.

Preparedness and exposure Judging by the frequency of earthquakes since 1980, the case study countries are

split in two natural groups. The purpose of the categorisation is to divide the most

earthquake-prone countries from the rest on a national level. The categories are

presented in Table 10.5.

- 160 -

Table 10.5 Earthquake exposure categorisation

Country ICDO Earthquake frequency

Exposed

Iran No 1688 Yes Afghanistan No 1126 Yes Pakistan Yes 1139 Yes China Yes 851 Yes Uzbekistan No 430 No Tajikistan No 385 No Kyrgyzstan No 358 No Kazakhstan Yes 268 No Turkmenistan No 241 No

Source: Author; NEIC 2006; ICDO 2002

Urban growth When looking closer at the figures for urban growth (see Table 10.6)

Afghanistan and Pakistan distinguish themselves as experiencing exceptionally high

urban growth whereas Kazakhstan and Iran are experiencing relatively low urban

growth. The purpose of the classification is to extract countries where extreme urban

growth might be leading to increased vulnerability. A classification with Afghanistan

and Pakistan in the high category is hence in order.

Table 10.6 Urban growth categorisation

Country Urban growth rate26

Growth

Afghanistan 4.88 High Pakistan 4.17 High China 2.94 Low Tajikistan 2.81 Low Uzbekistan 2.71 Low Turkmenistan 2.46 Low Kyrgyzstan 1.81 Low Iran 1.23 Low Kazakhstan 0.82 Low

Source: HABITAT 2003; Author

Openness The degree of openness of the case study countries covers a wide spectrum with

Tajikistan being very open and Turkmenistan being one of the most secluded

countries in the world. On a global scale only Tajikistan and post-Taliban

Afghanistan are close to having press freedom. To use the 2004 index as a measure of

26 Projected percent of urban population increase from 2000 to 2005

- 161 -

openness for the whole study period skews the situation in Taliban Afghanistan;

which by any measure was not an open society.

Table 10.7 Openness categorisation

Country WPFI Open Tajikistan 27.75 Yes Afghanistan 28.25 Yes Kyrgyzstan 35.25 Yes Kazakhstan 44.17 Yes Uzbekistan 52.13 No Pakistan 61.75 No Iran 78.30 No China 92.83 No Turkmenistan 99.83 No

Source: RSF 2006; Author

Vulnerability All the case study countries are vulnerable. Although the chosen dichotomous

categorisation is labelled vulnerable, this does not mean that the non-vulnerable

countries are considered very resilient. According to the GNA, the country with the

most pressing need for external aid in the case study area is Afghanistan, followed by

Kyrgyzstan. The case study countries are evenly distributed on the GNA. Using the

same approach as in the categorisation of the population indicator, the countries least

likely to need external assistance for an event of average impact are Pakistan and Iran.

Table 10.8 Vulnerability categorisation

Country GNA Vulnerable Afghanistan 2.71 Yes Kyrgyzstan 2.38 Yes China 2.29 Yes Kazakhstan 2.13 Yes Turkmenistan 2 Yes Tajikistan 1.63 Yes Uzbekistan 1.37 Yes Iran 1.25 No Pakistan 1 No

Source: Billing and Siber 2003; Author

An additional measure of vulnerability is the 50 km population density.

Uninhabited areas or areas with low density are less likely to sustain damage

validating an international response whereas damages in urban areas with RC

structures increase the relevance of an international intervention (Walker 1991). The

circle with 50 km radius that is used has an area of about 7 850 km2. The United

- 162 -

Nations Statistics Division (UNSD) recognises that it is not a straightforward task to

create a uniform quantitative definition of an urban area:

Because of national differences in the characteristics that distinguish urban from rural areas, the distinction between the urban and the rural population is not yet amenable to a single definition that would be applicable to all countries or, for the most part, even to the countries within a region. […] a distinction by urban and rural based solely on the size of the population of localities does not always offer a satisfactory basis for classification (UNSD 2006).

Collapsing structures is what kills people in earthquakes and the definition of

urban sought in this study is vaguely defined ‘built-up areas’. Determining this based

on the population density is not always feasible as pointed out in the UNSD quote.

Being subjective and likely to change even within countries, a standard for

categorising urban areas is not practicable. The determination of the less densely

inhabited areas could, however, be feasible. A separation of the uninhabited or

agricultural areas from other areas is an asset in the model development. Figure 10.4

shows the distribution of 50 km radius population in the case studies. There are many

case studies with population less than 50 000. This equals about six people per km2. It

is hard to imagine a situation were an area with so low population density would be

effected severely enough to justify an international intervention. Somewhat

arbitrarily 50 000 people with 50 km of the epicentre is adopted as a threshold to

identify agricultural or uninhabited areas.

- 163 -

Case study ID

2100000

2000000

1900000

1800000

1700000

1600000

1500000

1400000

1300000

1200000

1100000

1000000

900000

800000

700000

600000

500000

400000

300000

200000

100000

Source: Author; Landscan; INTEREST database

Figure 10.4 Distribution of 50km radius population in the case studies

Earthquake characteristics Empirically, the exact time when an earthquake strikes is not as important as

knowledge of whether the local population was awake or asleep (Alexander 2000b).

The local time was hence classified into day and night with input from Coburn and

Spence (2002:341) (Table 10.3). As for the earthquake depth, the human loss is

virtually zero for the ten case study events with depths that exceeded 40 km. This is

slightly shallower than Bolt’s (2004) 70 km definition. Although Bolt’s (2004)

categorisation may be a better representation of the geophysical characteristics of the

event, the 40 km delimiter does fit better with the consequences of the earthquake

observed in the case study area.

Without the use of an intensity raster to represent the strength of the

earthquake, the use of the magnitude in the modelling is going to be very

approximate. A service proving real-time intensity raster for the developing world is

not currently available. Instead, the magnitude is categorised to set a lower

magnitude limit under which earthquakes will not result in a severe impact and an

upper magnitude limit above which earthquakes have the potential of causing severe

damage. This leaves three categories to be defined. Coburn and Spence (2002:20)

- 164 -

states that earthquakes with magnitude lower than 5 only result in localized effects.

This is adopted as the lower limit. All of the 25 most lethal earthquakes in the

twentieth century had magnitudes of about 6 and above (Coburn and Spence 2002:7).

This limit will suit the intention of the upper limit.

Outlier events To some extent, the outliers provide important information on the extraordinary

events that are of interest to predict. Therefore, any exclusion from the study of an

outlying event should only be made after careful consideration on whether an

outlier’s abnormal characteristic is a disturbance and not a relevant indication of

international attention. Outliers exist in the 50 km population attribute, in the

hypocentral depth and in the numbers of sitreps. An outlier event that requires

exclusion is the 2001 Gujarat earthquake. The seismic characteristics for this event

relates to the initial quake that struck the region around the city of Gujarat in India.

However, the impact data in the database relates to the impact caused in Pakistan, far

from the epicentre. There is hence no connection between the seismic characteristic

and indicators in the database. This exclusion reduces the number of case studies to

be used in the model development to 58. The final list of variables and their purpose

in the development of a model is displayed in Table 10.9. There are not instances of

empty cells in the uni-variable analysis.

- 165 -

Table 10.9 Data mining start variables

Label Code

N Marginal

Percentage AttCat 1 Low 39 67% 2 Intermediate 12 21% 3 High 7 12% MagCat 1 Strong 22 38% 2 Intermediate 23 40% 3 Weak 13 22% Shallow 0 Deep 9 15.5% 1 Shallow 49 84.5% Urban 0 Rural 26 45% 1 Urban 32 55% Night 0 Day 44 76% 1 Night 14 24% Vulnerable 0 Resilient 35 60% 1 Vulnerable 23 40% Open 0 Closed 40 69% 1 Open 18 31% HighGrowth 0 Low Growth 40 69% 1 High Growth 18 31% Exposed 0 Low Exposure 6 10% 1 High Exposure 52 90% Valid 58 100.0% Missing 0 Total 58

10.4 Data mining The data mining follows Hosmer and Lemeshow’s (2000) process outlined in

section 5.3.2. The indicators have been examined for empty cells and outliers above,

which constitutes the first step in the process recommended by Hosmer and

Lemeshow (2000).

10.4.1 Multi-variable analysis input selection A SPSS analysis of the variables and categorisations in Table 10.9 suggest that a

complete model would result in 55% empty cells. The statistical package warns of

complete separations in the data. A recurring problem in ordinal regression is final

models with quasi-complete separation in the data (Tabachnick and Fidell 2001).

Complete separation occurs when a set of IVs completely determines a category

output on the DV. The MagCat and Shallow indicators almost completely separate the

low attention events. This situation is, however, logical. Deep earthquakes with low

or intermediate magnitude are very unlikely to cause significant damage on the

- 166 -

surface. To make the model certain in this regard, the event population would have

to be increased to include deep or low magnitude earthquakes that resulted in an

intermediate or high international attention. To obtain a sufficient number of events

of that type the study would have to include events from outside central Asia. This

would increase the complexity of the study beyond the scope of this project. The

current set of variables builds on the assumption of relative cultural and socio-

economic coherence among the case study countries. An expansion of the study

would require the identification and collection of indicators for numerous events

outside the case study region. This is not feasible due to the limited time and

resources available to the project.

When analysing the descriptive statistics of the categorised variables using

cross-tabulation some variables stand out as having an unexpected pattern or not

adding to the predictive power of the model. The distribution of Night over the low

and intermediate AttCat is even with three times as many events occurring at daytime,

i.e. the fraction of events occurring at night is the same for daytime events (see Figure

10.5). However, in the high attention category there are six times as many daytime

events as there are night-time events. This is an unexpected situation. Previous

research (Alexander 2000b; Coburn and Spence 2002) have shown the time of day

impact to affect the amount of human losses. After investigation it became clear that

the anomaly was caused by an uneven distribution of strong earthquakes over night

and day (see Table 10.10). The indicator may hence still contribute to the model when

combined with the MagCat indicator and is passed on to the variable importance

analysis.

Table 10.10 Distribution of earthquakes over night and day

MagCat Total Weak Intermediate Strong

LocalTime Day 8 19 17 44 Night 5 3 6 14 Total 13 22 23 58

- 167 -

Figure 10.5 Distribution of cases over ‘Night’

The Exposed indicator has no high attention events in countries with low

exposure (see Figure 10.6). A reclassification to increase the number of events in the

low prevalence category would leave Afghanistan and Iran in the high exposure

category and thus make the indicator cause complete separation of high attention

events. A reclassification would also leave China out of the high exposure group,

which does not match with empirical knowledge. The indicator is passed on to the

variable importance analysis.

Figure 10.6 Distribution of cases over ‘Exposed’

The indicators Urban and HighGrowth both follow the pattern that would be

expected. High attention events are more than six times more common in urban

areas, with the division between rural and urban being equal in the other attention

categories. Events in HighGrowth countries receive higher international attention.

10.4.2 Variable importance analysis The SPSS parameter estimate table for the complete model served as the basis

for this phase of the model development (see Table 14.3 in the appendix). The Wald

- 168 -

score is used to remove indicators with the least influence on the model outcome. In

the complete model the least influential indicator is Open. When removed the r2 drops

to 0.655 with no negative influence on the classification table. This reduces the

number of empty cells and the statistical software no longer warns of complete

separation. An exclusion of Open would be beneficial to the model development.

Open represents a central characteristic that is being investigated as part of this

research project. Exclusion would hence deprive the model of an indicator that is of

interest to the research. Closer examination reveals that Open is correlated to the

vulnerability indicator. Open is consequently left out temporarily to be re-examined

in the model variable interaction phase.

The second least influential indicator is the intermediate category on MagCat.

To test its importance to the model, this category is collapsed into the low magnitude

category. In doing so the r2 drops from 0.706 for the initial model to 0.330 for the

model with two categories of MagCat and also making the predictive power of the

model statistically insignificant. The MagCat hence has to remain with all three

categories.

The third least influential indicator is population. Its exclusion from the model

reduces the r2 to 0.615, which is not a major reduction. However, by examining the

classification table it becomes clear that the distribution of classification errors have

shifted towards omission errors, i.e. that high attention events no longer are correctly

predicted as such. This cannot be accepted and the population indicator will hence be

restored in the model.

Based on the parameter values for the Night indicator, the events occurring

during day are expected to receive more international attention. This shows that the

uneven distribution of high magnitude events over day and night distorts the input of

this variable. The empirical knowledge published in literature (Alexander 2000b) is

that vulnerability is increased at night, which results in increased human losses.

Human losses in turn have been shown to be correlated with international attention.

Experiencing the opposite in the model is intriguing. If the situation is not caused by

the small population of events, a potential hypothesis would include an affect on the

international media coverage. Day-time events in central Asia will surface in the

morning news in Western Europe, which could spur media coverage. This will,

- 169 -

however, not be investigated as part of this thesis. In conclusion, the Night indicator

will be excluded from the study due to its unclear effect on international attention.

Individual exclusion of the Exposed, Vulnerable or HighGrowth indicators

significantly reduce the r2 but the impact on the classification table is minimal. The

lack of one, two or all three of the indicators introduces commission errors in the high

attention category and omission in the intermediate attention category to the low

attention category. Although commission in the high-level category is preferable, the

introduction of omission errors in the intermediate attention category makes the

exclusion of either indicator undesirable.

10.4.3 Main effects analysis For ordinal regression the SPSS package provides a set of link functions. The

Cauchit link function is optimized for models where one extreme needs to be

predicted (SPSS 2003; Zelterman 2006:76). This purpose fits well with this research

project because it is the minority of high attention events that are of greatest relevance

of being predicted. However, when using a cauchit link-function the standard

deviations of the resulting model are large (see Table 14.4), indicating numerical

problems in the model (Hosmer and Lemeshow 2000). When using the Logit link-

function the standard deviations are reduced to acceptable levels, but the omission of

high and intermediate attention events is increased. Hosmer and Lemeshow

(2000:141) stated that high standard deviations are a sign of problems in the model.

This does not mean that the model is useless. They write that if the model parameters

show high standard deviation, the user has to be alert for signs of complete

separation, empty cells and co-linearity. The SPSS software does not detect complete

separation in the case study data when using cauchit. The model results in 55 percent

of empty cells, which is high but not too high for a useful prediction (SPSS 2003). The

cauchit link-function is hence adopted for the model.

10.4.4 Model variable interaction Ordinal regression is sensitive to correlations among the IVs. Among the IVs a

correlation exists between Open and Vulnerable27. This means that the power of the

statistical analysis could be improved by omitting one of the two indicators.

Although not significant, the high correlation between the two indexes is interesting.

27 r2=-0.67, n=58, non-significant

- 170 -

The two indicators do not overlap in their measurements, which mean that there is a

correlation between ECHO’s needs assessment and press freedom. This is, however,

not the focus of this study and the investigation will be left for future research.

Returning to the model development, an alternative is to include the two indicators as

interaction variables. Testing does show that this results in complete separation and

lower predictive power of the model. In a choice to include one of the two indicators,

the most scientific and stable measurement should be selected for the sake of

feasibility in future usage. The WPFI, that the Open category is based on, is developed

using qualitative data in the form of questionnaires completed by reporters in the

various countries. Although the WPFI ranking process is described on the RWB

website (RSF 2006), it is subjective and not transparent. The basis for the vulnerable

indicator, GNA, is a well-documented (Billing and Siber 2003) quantitative composite

index that is in use by established development assistance organisations. The 2005

GNA is hence chosen to represent vulnerability as well as the aspects represented by

the WPFI.

The model output at this stage, the preliminary final model according to the

terminology used by Hosmer and Lemeshow (2000:99), is presented in Table 10.12.

The model is developed using the indicators listed in Table 10.11.

Table 10.11 Full model parameter estimates (Cauchit)

95% Confidence

Interval

Estimate Std. Error Wald

Lower Bound

Upper Bound

Threshold [AttCat = 1] 1.94 1.56 1.54 -1.12 5.00 [AttCat = 2] 4.44 1.76 6.36 0.99 7.90 Location [Shallow=0] -3.08 1.53 4.04 -6.09 -0.08 [Vulnerable=0] 2.12 1.13 3.52 -0.09 4.33 [MagCat=1] 5.74 1.82 9.91 2.17 9.32 [MagCat=2] 1.78 1.57 1.28 -1.30 4.87 [Urban=0] -2.05 1.04 3.87 -4.10 -0.01 [HighGrowth=0] -3.94 1.17 11.24 -6.24 -1.64 [Exposed=0] 4.44 2.04 4.72 0.43 8.44

Redundant parameters removed Source: Author; INTEREST Database

- 171 -

Table 10.12 Full model ordinal predictions (Cauchit)

MagCat Shallow Pop Vulnerable Growth Exposed AttCat (frequency) Low Intermed. High Strong Deep Rural Vulnerable Low Low Observed 1.00 0.00 0.00 Expected 0.71 0.24 0.06 Chi-square deviance 0.64 -0.56 -0.24 High High Observed 2.00 1.00 0.00 Expected 2.21 0.63 0.16 Chi-square deviance -0.28 0.53 -0.42 Shallow Rural Resilient Low High Observed 1.00 2.00 1.00 Expected 1.66 2.06 0.27 Chi-square deviance -0.67 -0.06 1.44 High High Observed 1.00 1.00 0.00 Expected 0.07 0.08 1.85 Chi-square deviance 3.54 3.26 -4.90 Urban Resilient Low High Observed 0.00 3.00 3.00 Expected 0.41 3.12 2.47 Chi-square deviance -0.67 -0.10 0.44 Vulnerable Low High Observed 3.00 0.00 0.00 Expected 2.15 0.68 0.17 Chi-square deviance 1.09 -0.94 -0.42 High High Observed 0.00 0.00 3.00 Expected 0.12 0.19 2.69 Chi-square deviance -0.36 -0.45 0.59 Intermed Deep Rural Resilient Low High Observed 1.00 0.00 0.00 Expected 0.97 0.01 0.02 Chi-square deviance 0.17 -0.09 -0.14 Vulnerable High High Observed 1.00 0.00 0.00 Expected 0.96 0.01 0.02 Chi-square deviance 0.20 -0.12 -0.16 Urban Vulnerable High High Observed 1.00 0.00 0.00 Expected 0.93 0.04 0.04 Chi-square deviance 0.28 -0.20 -0.19 Shallow Rural Resilient Low High Observed 4.00 0.00 0.00 Expected 3.83 0.07 0.10 Chi-square deviance 0.43 -0.26 -0.33 High High Observed 1.00 1.00 0.00 Expected 0.41 1.41 0.18 Chi-square deviance 1.03 -0.64 -0.44 Vulnerable Low Low Observed 2.00 1.00 0.00 Expected 2.76 0.13 0.11 Chi-square deviance -1.62 2.45 -0.34 Urban Resilient Low High Observed 7.00 0.00 0.00 Expected 6.29 0.43 0.28 Chi-square deviance 0.89 -0.68 -0.54 Vulnerable Low High Observed 1.00 0.00 0.00 Expected 0.96 0.01 0.02 Chi-square deviance 0.20 -0.12 -0.16 High High Observed 1.00 2.00 0.00 Expected 1.27 1.52 0.20 Chi-square deviance -0.32 0.55 -0.47 Weak Deep Rural Resilient Low High Observed 1.00 0.00 0.00 Expected 0.98 0.00 0.02 Chi-square deviance 0.14 -0.07 -0.13 Urban Vulnerable High High Observed 1.00 0.00 0.00 Expected 0.96 0.01 0.02 Chi-square deviance 0.20 -0.12 -0.16 Shallow Rural Resilient Low High Observed 2.00 0.00 0.00 Expected 1.94 0.02 0.04 Chi-square deviance 0.24 -0.13 -0.20 Vulnerable Low Low Observed 1.00 0.00 0.00 Expected 0.96 0.02 0.02 Chi-square deviance 0.20 -0.12 -0.16 High High Observed 1.00 0.00 0.00 Expected 0.96 0.01 0.02 Chi-square deviance 0.20 -0.12 -0.16 Urban Resilient Low High Observed 5.00 0.00 0.00 Expected 4.77 0.09 0.13 Chi-square deviance 0.49 -0.31 -0.37 High High Observed 0.00 1.00 0.00 Expected 0.17 0.73 0.10 Chi-square deviance -0.46 0.61 -0.33 Vulnerable Low Low Observed 1.00 0.00 0.00 Expected 0.91 0.05 0.04 Chi-square deviance 0.31 -0.23 -0.20

Source: Author, INTEREST Database

- 172 -

10.5 Evaluation and validation framework The pseudo-r2 was used to compare sub-models in the development of the main

model. This is, however, only good for providing a rough comparison between

models. To evaluate the final model the only tool is classification table (Table 10.12) in

the appendix. The table enables detailed analysis of classification errors. The

classification errors have been extracted from the main table to Table 10.13. These

results and the general usability of the model will be analysed further in section 11.3

in the next chapter.

Table 10.13 Classification errors

Country Event ID Observed AttCat

Predicted AttCat

Predicted Probability28

Classification difference

Pakistan 48 2 3 90% 1 Pakistan 17 1 3 90% 2 Afghanistan 3 1 2 50% 1 Iran 28 1 2 50% 1 Pakistan 5 1 2 70% 1

-- -- -- -- -- -- Afghanistan 42 2 1 75% -1 Kyrgyzstan 14 2 1 90% -1 Iran 59 3 2 50% -1 Iran 46 3 2 50% -1 Iran 18 3 2 50% -1 Iran 20 3 2 50% -1

Live event testing Since the completion of the model, two earthquakes that resulted in

international intervention have occurred in the case study region. These are the

catastrophic October 2005 Kashmir earthquake and the March 2006 earthquake in

Lorestan province in Iran. These events could not be investigated in detail due to

their occurrence relatively late in the research project. However, using Table 10.12 the

Kashmir earthquake is classified as an intermediate attention event with 51 percent

likelihood and low attention at 40 percent likelihood. This serious misclassification

shows the unsuitability of the fixed 50km radius representation of the earthquake.

Strong events like that in Pakistan have serious affects along a fault that can be in

excess of 1 000 km long. The Lorestan event was observed as intermediate with three

sitreps and 63 persons reported dead. The first earthquake was not as strong as the

28 Rounded

- 173 -

aftershocks, which caused problems in the model. However, regardless of the

classification of the earthquake as low or intermediate magnitude the model

prediction is for a low attention event with a probability between 90 and 95 percent.

Conceptual final model The conceptual of the model as now stands is provided in Figure 10.7.

Although the losses are not explicitly calculated in the model, their conceptual

location is provided in the figure. The GNA is used both for indication of needs,

which is its original intention, and as a proxy indicator of the press freedom and level

of democratisation. In section 10.4.4, the GNA was selected for both roles due to the

co-linearity between the WPFI and the GNA. This model will be used in the analysis

of the model in section 11.3 in the next chapter.

Figure 10.7 Conceptual final model

10.6 Systems Design and Implementation summary A prototype model of the international attention was developed in this chapter.

The model uses a range of categorised predictors (IVs) to determine the value of the

dependant variable (DV). The dependent variable is the categorised number of UN

situation reports. This DV was chosen because it better represents the overall

- 174 -

international attention paid to an event. All the variables were categorised to absorb

the uncertainty in them. In the data mining process, the IVs were filtered so that only

those with the greatest predicting power of the DV were included.

Preliminary testing of the model shows that it is accurate in 81 percent of the

events. This is likely to be inadequate for it to be used by any of the studied user

groups in its current form. The model will be tested and analysed further in section

11.3 in the next chapter.

- 175 -

11 EVALUATION The structure of the evaluation reflects the research objectives: User

requirements and system relevance; Quantification of the modelled context; and,

Development and testing of a prototype model.

11.1 Objective 1: User requirements and System relevance Part of the first objective of the thesis was to develop a set of user requirements

including thresholds for timeliness, accuracy and notification content. Existing

systems in this category were reviewed in section 4.2.2 and analysed in relation to the

user requirements in section 9.2. Those discussions are brought together here.

11.1.1 Relevance of international alert systems The first objective was targeted with a research question as to in which decision

and how that the international relief community should be supported. The purpose

of an alert system is not to initiate the international response but to initiate the

collection of further information from conventional sources to support or disapprove

the requirement of international relief. Without the system, decision makers would

either rely solely on alert systems activated by seismic characteristics of an event or on

on-site sources such as media and resident representatives. These are not optimal

solutions. The alert systems based on seismic data produce many false positives and

the on-site sources may involuntarily become incommunicado due to the effects of the

disaster.

For which types of events is it relevant? But for which events are alerting of the proposed kind beneficial? If accepting

Wyss’ (2004b) claims that the events with the most extreme impact are accurately

detected by loss assessment models and properly acted upon by international

community on the basis of the loss data, the events of interest for this study are the

intermediate humanitarian impact events where the demand for international relief is

not immediately apparent. The remoteness of an area or aspects such as unforeseen

vulnerability or reduced local coping capacities can obscure need. This was

exemplified in one of the case studies presented in section 7.3.2 where an unexpected

secondary disaster increased the need for international relief. An option is to send

experts to the area to evaluate the need, but that takes time. According to the

UNDAC (2000) guidelines, the intention is that a needs assessment team should reach

a disaster area within 24 hours after the relevant authorities have taken the decision to

- 176 -

send them. The information gathered by the team is, however, likely to be of little use

for more time-sensitive forms of relief, such as SAR teams or medical supplies.

Berthlin requested alerts within an hour for these types of aid and more detailed

contextual information within six hours after the event. The alert systems hence have

a window of about an hour in which to operate and, if accurate, they provide the

decision maker with additional time to improve the international response. The

research does not show how big this improvement is, but it will be moderate at best.

Time-sensitive relief dispatched internationally is bound to be less cost-efficient than

local mitigation and preparedness efforts (Walker 1991).

Normative benefit International alert systems are not only relevant for providing more time for the

entry decision. Once the entry decision has been taken, the main bottleneck is the

arrangement of logistics. The lack of high quality information and the arrangement of

logistics are not always the sources of the temporal bottlenecks in disaster relief.

Political agendas in both the responding nation(s) and the affected nation(s) can

postpone the acceptance of an event as being a disaster as described by Albala-

Bertrand (1993). Olsen et al (2003) argued that media attention could occasionally add

confusion and delay or distort the response even further by giving disproportioned

exposure to an event, although both Berthlin and Alexander (2000a:85) did not see

media or politics as powerful in influencing the short-term relief.

Koethe (2003) claims that a rule-driven analysis of data provides an objective

platform to inform decision makers. Such a platform could be used to speed up the

response process. It is the researcher’s opinion that alert systems are an example of

such a platform. Taken to its extreme, an accurate future model could be applied in

real-time to give normative suggestions to the decision maker. In a refined form,

these platforms could serve to alert the users of forgotten crises through the detection

of anomalies in the level of international attention given to events.

Summary of relevance Provided that the palliative international post-disaster relief continues to be

seen as a valid form of support to developing countries, alert systems for use by the

international community, as opposed to the affected people, can be cost-effective.

Useful tools for this task are transparent in their assumptions and have outputs that

are timely, accurate and pertinent to the task. Furthermore, cost-effectiveness requires

that a useful tool is developed and maintained with low cost.

- 177 -

11.1.2 Timeliness, Accuracy and Completeness The requirements were examined in relation to the quality of the system output,

covering the aspects of: timeliness, accuracy and completeness.

Timeliness requirement From the interviews with the implementing organisation it was clear that the

practitioners needed an alert within one hour after an event impact and that a final

entry decision had to taken within six hours. The one-hour limit for the initial alert

was under the assumption that the alert contained limited information relating to the

characteristics of the hazard, possibly coupled with socioeconomic data and maps.

Berthlin’s time limit for the entry decision was six hours. Table 9.3 showed that,

under these constraints, numerical models stand out as the most promising option for

remote assessment.

Accuracy requirement To both Berthlin and Suarez, as part of implementing organisations, the

accuracy of the supplied information was irrelevant as long as the degree of accuracy

was known. These statements were at odds with the researcher’s observation of the

development of the GDACS tool. When supplied with confidence intervals users

complained of the tool being overly complex or irrelevant when the confidence

interval was too great. To confirm this observation some users unregistered from the

alert service, when false positive alerts were issued. The inconsistencies of the user

expectations on the system are likely to relate to the type of host organisation that the

user is in.

Completeness requirement In the investigation of the completeness in Chapter 10 it became clear that the

completeness depended on the intended use of the information. An alert cannot be

expected to suffice as the only input to an entry decision. Instead the function of an

alert should be to active a process of intelligence gathering. The completeness hence

relates to the question on whether international relief is required. Users in the

implementing organisation suggested that this could be answered dichotomously.

The main drawback of a dichotomous answer is that it does not communicate the size

or likelihood of an international intervention. Still, from the user’s point of view it is

only important to know if a response is necessary based on his or her conceptions of a

response.

- 178 -

Heterogeneity of the user requirements In the systems analysis it was found that the user requirements of the funding

and co-ordinating organisations were not as well defined as those for the

implementing organisation. In their statements and through the researcher’s

observation of their use of the GDACS tool, the funding organisation did not seem to

put this kind of alert as a priority. Although not explicitly stated, their preference

seemed to be to rely on conventional information channels, like resident

representatives and partner implementing organisations. In the meetings with the co-

ordinating organisation the general impression was of them being content with

disseminating whatever information that is provided to them as long as the

information in question was not “irrelevant or politically tainted”.

The fragmentation of the user requirements becomes clearer in an analysis of the

decision processes. Once an implementing organisation is made aware of a disaster,

the decision making process starts with an ‘entry decision’. In the case of the SRSA,

the funding for the international interventions is pre-approved by the Swedish

government. For implementing organisations that do not have a funding reserve or

are independent from government, the entry decision is dependent on external

funding sources and an entry decision in a funding organisation. Before an entry

decision is taken in the SRSA, several filters in the form of relief professionals and

phenomena experts are applied to evaluate the information at hand to exclude events

that are unlikely to require assistance from the SRSA. Events that potentially could

benefit from SRSA assistance are passed on to more senior decision makers. If the

information at hand is insufficient or uncertain, the decision maker can wait for

information that is more complete or with higher accuracy. However, the longer the

waiting time, the lower the benefit of a potential relief effort. Benini et al (2005)

suitably term this equilibrium as “Speed kills vs. Victims cannot wait”. In these cases,

making an extemporaneous decision, thus shortening the response time and using

information of lower quality, increases the probability of a suboptimal decision.

Moreover, the commitment of valuable assets, including needs assessment experts, to

an event that turns out not to be a disaster could result in a reduction of the resources

available for future disasters. According to the responses received in the interviews,

situations like this existed both in the implementing and in the funding organisations.

Whether this was a problem in the co-ordinating organisation was, however, less clear

- 179 -

because they act on requests from actors like the affected country and the

implementing and funding organisations.

Summary of user requirements In summary, it was found that requirements vary significantly between user

groups. Although there is no consensus on the required degree of accuracy or

content, it is clear that an alert should be received by the users within hours following

an event. The alert will prolong the time available to collect further information and

to make an entry decision. For an alert system to be trustworthy it has be accurate,

but that is not all. Both users and literature gave particular weight to transparency of

the assumptions made by the system for it to be trustworthy.

11.1.3 The shortcomings of existing systems All alert and planning systems identified as part of this study are focused on

loss estimation. When asked whether loss data are important, potential users from

both funding and implementing organisations answered positively. It is the

researcher’s opinion that the number of collapsed structures, the number injured and

killed should not be directly translated to a requirement for an international

intervention because they are measures of loss and not of needs. Nor do those figures

indicate international response in the past. The correlation matrix in Figure 10.2

shows that loss indicators have a weak correlation to the reported need and resulting

response.

Loss-centred modelling is particularly inappropriate in low causality events

because there are no common definitions of loss. For instance, in earthquakes with

little structural collapse, most mortality results from heart attacks (Alexander

1993:466) and it is not always certain if the earthquake is the cause for the death of a

person that already was in ill health. According to Berthlin and Albala-Bertrand

(1993) international responses to these types of events, where the benefit of the

international response to those suffering is expected to be marginal, are more reliant

on intangible political factors that do not form part of the examined loss assessment

models. Furthermore, when asked what information that was useful in the initial

stages following an event, the interviewed and observed decision makers only gave

loss estimation a limited role. Knowledge of local resilience, for instance predominant

building material or response capacities, can give the international response

community qualitative and macroscopic indications of indirect needs that can be just

- 180 -

as relevant to the international response process as quantitative loss data. With

regards to international SAR responses, Walker (1991) writes that it is only useful in

the collapse of reinforced concrete (RC) structures. An alert based on loss assessment

and intended for a SAR response should hence consider how many of the expected

losses that occurred in RC structures. This is not the case in any of the studied

systems.

Alert systems overemphasised It is relevant to discuss the labelling of earthquake alert systems as being early

warning systems (EWS). The type of earthquake alert systems analysed in this

research project can by no standard be seen as ‘early’. The alerts are issued after the

event has taken place. It is the researcher’s opinion that ‘EWS’ is not a term well

suited for use in relation to earthquakes. The fastest and ‘earliest’ warning systems

for earthquakes available today give 20-30 seconds of warning (Coburn and Spence

2002:78), which at best can be seen as a ‘warning system’.

Looking at the projects included in the book “Early Warning Systems for

Natural Disaster Reduction” edited by Zschau and Küppers (2003) it is clear that the

majority of the projects fall into Kersten’s (2000) and O’Brien’s (1999) definitions of a

DSS. Some of the systems are not intended for ‘early warning’ or ‘warning’ at all and

many systems include functionality that exceeds what is normally interpreted as a

‘warning’. The word ‘warning’ implies a limitation to the issuing of an alert, whereas

many systems provide decision support on possible actions by the user and allows for

the analysis of scenarios. EWS is consequently an unsuitable term that confuses

matters. Terminology aside, as discussed above, the earthquake alert systems can be

of benefit, even pivotal, if they function as required. However, any EWS surrounded

by weak links to the preceding and subsequent phases in the disaster management

cycle will fail. In 1995 the IFRC (1995:35-36) stated that disaster managers and

organisations funding projects must accept that the development of alert tools is a

palliative form of preparedness and not a silver bullet. Nevertheless, the conference

on early warning in Bonn 2006 provided several examples where emphasis on pre-

disaster activities in exposed countries was put on alert tools.

Summary of shortcomings The current alert systems are focused on quantitative loss assessments as the

core of the decision support although there are signs that other types of information

are at least as important.

- 181 -

It should be clear that alert systems are only one link in the chain leading up to

an efficient response or a prevented disaster. There is a risk that the tangibility of

these systems makes them unjustifiably attractive investments for international

donors. Without properly funded mitigation and preparedness efforts leading up to

the alerting and without proper response plans leading out of it, the efficiency of the

international response is destined to be limited by the weakest link in the chain.

11.2 Objective 2: Quantifying the international actions The purpose of this objective is to provide a sufficient amount of accurate

quantitative data to solve the third objective. The suitability of the applied

quantifications in this data collection process is discussed here. A second task part of

the second objective was a preliminary evaluation of the patterns in the international

community actions to pave the way for a more targeted search as part of the third

objective.

11.2.1 Challenging the quantifications and categorisations To determine the preference of the international community, an indicator of

event ‘size’ or relevance has to be identified. The definition of an objective ‘size’ of

international attention is a subjective and contentious issue deemed virtually

impossible by for instance Alexander (2000a:192). As in the research by Olsen et al

(2003) and Albala-Bertrand (1993), this study found that quantitative measurements of

loss, needs and response are all inappropriate indicators of international attention

based on them either simply not correlating with a qualitative estimation of

international attention or on the basis of the relevant data not being possible to collect.

Content analysis of frequency proved to be a promising method in this context. Based

on Figure 10.2 and further analysis, the frequency of sitreps was adopted as the

indicator of international attention size. The available options were discussed in

Chapter 9.

It should be clear that the use of sitrep frequency as an international attention

indicator is a compromise. Ample criticism can be made at this method. The case

studies that took place before the UN Department of Humanitarian Affairs (DHA)

had been reformed into the Office for the Coordination of Humanitarian Affairs do

not have sitreps. Instead the DHA issued telex and fax ‘flash messages’ that are

considerably shorter than the current sitreps. Even though these messages have been

digitised and included in the research database, it is reasonable to assume that the

- 182 -

volume of digitised information for an event in 1992 is considerably less than a similar

event ten years later and that the frequency of messages sent via fax or telex is lower.

Furthermore, the data on loss and financial aid used to test the suitability of the sitrep

frequency as an attention indicator may contain systemic bias. The sitreps are an

important source of data on loss and response. Events with multiple sitreps do

consequently have more data from more sources. It is possible that the higher donor

exposure provided by the sitreps lead to increased donations.

The plurality of sources gives a balanced picture of the real loss and response.

Seeing that events with multiple sitreps are linked to greater losses and greater

financial response, it means that the uncertainty is larger in events with limited losses

and little financial response. This does not suggest that the use of the sitrep frequency

as an indicator of international attention is incorrect, but that it is unsuitable for

analysis of low attention events.

Like the quantification, the applied categorisations are subject to criticism. They

can be seen as arbitrary and in some cases, like in the categorisation of the earthquake

magnitude, their natural relation to the categorised subject can be questioned.

Admittedly, it is not true that earthquakes with depths exceeding 40 km always are

harmless, nor is it impossible for an event with fewer than five sitreps to receive more

than USD 200 000 in aid. Additionally, the earthquake magnitude scale is logarithmic

and a seismologist might argue that the fixed thresholds used for its categorisation do

not represent the phenomena in a natural way.

The categorisation is, however, an appropriate solution in light of the limitations

put on the study by the small population of events and the limited resources available

to collect and analyse data. The categorisation is required to absorb the inaccuracy in

the data, to group events according to rough characteristics and to facilitate the

statistical analysis. The categories are not used in a sequential decision-tree manner,

but converted through the ordinal regression to provide linear relationships to the

DV. That way, no single category is able to determine the output on the DV on its

own. It is true that more effort could have been invested in finding more exact ways

to categorise these attributes. With more time, events outside of the case study area

could have been included and analysed for differences in the way that characteristics

should be categorised. Additional time for such analysis was, however, not available.

- 183 -

Summary of quantification defence The frequency of sitreps is indicative of international response size. This

attribute can, however, be criticized when used to measure small events or event that

occurred decades apart. The categorisation is required to counteract the uncertainty

in the data. The categorisation was made as accurate as possible with the resources

available, but it is acknowledged that it could have been better.

11.2.2 Patterns in international actions The purpose of analysing the behavioural patterns before developing a

prognostic model was both to ascertain whether there were signs of patterns that

could be replicated and to determine if these patterns followed logic and morality.

Before the development of the prognostic model, Figure 10.2 indicated a weak

correlation between human casualties and the international financial aid as well as the

sitrep frequency. Additional analysis of factors like non-UN reporting frequency and

relief item donations also showed slight increase with an increase of human

casualties. This arguably shows that the morality of the international community is

not systematically flawed. Figure 10.2 would, however, have been more relevant if it

had been possible to include quantitative indicators of need, rather than loss. In the

analysis, the collected data on needs were found to be of inadequate quality and

coverage for such analysis.

As in the investigation of the user requirements on alert systems, coherent

groups were identified in the analysis of behaviour. Whereas the implementing

organisations could see a potential benefit in these systems, the co-ordinating

organisation seemed complacent to its benefits and the funding organisations

reluctant to change. These three groups of users fit in the DSS pyramid in Figure 3.1.

The implementing organisations are operational users taking on repetitive tasks of

SAR of largely similar, pre-specified, character from event to event. This is reflected

by Berthlin’s statement that the equipment and staff roles used for international

interventions by the SRSA have been the same for years. The co-ordinator

organisations fit a pseudo-tactical profile described in section 3.2. The unconventional

characteristic of it not having any formal power over the units that it is co-ordinating

makes it fall outside O’Brien’s (1999) pyramid. Nevertheless, in its role in the process

it has a wide scope, handling several types of disasters and responses with exclusively

external contacts, making its environment more changing and with greater similarities

- 184 -

to a tactical organisation than an operational one. With an even wider scope than the

co-ordinating organisation and with political roots, the funding organisation includes

both elements of both tactical and strategic decisions.

These differences make the three types of organisations’ decisions and actions

fundamentally dissimilar. Some commonalities do, however, exist. Through the

interview with Berthlin the most important sources of information for the

international community were determined to be:

• on-site contacts and representatives;

• international contacts with organisations of a similar type;

• the OCHA - Reliefweb and VOSOCC; and,

• the affected government.

Summary of identified patterns There were patterns in the action of the international community but the

patterns were linked to the type of actor. Preliminary analysis showed that the

actions did not systematically conflict with morality. The frequency of sitreps is

consequently a suitable proxy indicator of international attention to an event.

11.3 Objective 3: A prototype model The purpose of this prototype is to examine the feasibility of developing a non-

loss based numerical model that overcomes the complexity of communicating the

uncertainty of the output. This discussion is centred on the model development

process and the results presented in Table 10.13 (page 170) and Table 10.12 (page 171).

Each event is referred to by its location and its event ID (see Table 10.13) in the

INTEREST Database. The discussion is structured based on the type of

misclassification: under- and over-predictions. This should not be interpreted as if all

‘correct’ predictions are accepted as such. The received amount of attention is likely

to be wrong in some of those events. However, the main purpose of the discussion is

to evaluate the performance in terms of strengths and weaknesses of the model and

not the performance of the international community. This reduces the relevance of a

discussion on the accurately predicted events. For the misclassified events, the

discussion aims to discern whether (1) the observed level of international attention

was suitable, (2) whether the predicted level of international attention was accurate,

and (3) what the potential causes to an inaccurate prediction were.

- 185 -

The performance of the model is discussed in relation to its underlying

conceptual parts outlined in Table 10.1 (Classification of indicators, according to

purpose). Although the model is not attempting a sequential prediction of impact,

needs (as outlined in Figure 10.1) and response, the model takes those elements into

consideration in a non-sequential fashion. The three elements represented in the

model are hence impact, needs and political factors encouraging a response. The

discussion applies these three elements in the search for model weaknesses.

As described in the methodology chapter, the model looks at cells and estimates

the probability that a new event will belong to one of the three groups of attention

based on the patterns of the case studies. An example from Table 10.12 is that shallow

and high magnitude earthquakes in urban areas of a vulnerable countries result in

almost 90 percent (2.69/3) of the cases being predicted to receive high attention, 6

percent receiving intermediate attention and 4 percent receiving low levels of

attention. This corresponds quite well with the three cases in this cell, which were all

observed in the high attention category.

11.3.1 Under-prediction The under-predictions are the events that were predicted to fall in a lower

category than turned out to be the case. This is the more serious type of classification

error, particularly when it occurs in observed high attention events.

High attention events predicted as intermediate The four high attention cases that were misclassified as intermediate attention

events all occurred in Iran: Bam 2003 (id:59), Quazvin 2002 (id:46), Ardebil 1997 (id:18)

and Qayen 1997 (id:20). These disasters were all the result of shallow, high

magnitude earthquakes in urban areas. The events in Ardebil and Quazvin received

five and six sitreps respectively and thus just barely made it to the high attention

category in the categorisation. However, in reality, all four events are rightly high

attention events, as they resulted in both a large number of human casualties as well

as in substantial foreign financial aid. The Bam earthquake received 14 sitreps, the

highest number in the study, but was nevertheless predicted to fall in the intermediate

attention category. Erroneously, the initially reported epicentre was outside the city

and lead to it being classified as a rural event. The high seismic vulnerability of the

local construction material was also unrepresented in the model. The loss estimation

facet of the model was hence inadequate. Potential contributing factors to the

- 186 -

underestimations for all the events in Iran include systemic increased international

attention to events in Iran, or insufficient temporal or spatial resolution of the

indicators, or missing indicators. The model element at fault is hence a combination

of any of the involved factors of loss, needs or response.

Intermediate attention events predicted as low Two observed intermediate attention events were predicted to receive low

attention. These were the 2002 Dakhli event (id:42) in Afghanistan and the 1997 Ak-

Tala event (id:14) in Kyrgyzstan. The Ak-Tala event affected around 1 200 people in a

couple of villages and made 30 people homeless. The widespread damage to

infrastructure and housing delivered a huge blow to the poor country. However, no

serious injuries or deaths were reported. Two sitreps were released with focus on

recovery of infrastructure and housing. An appeal for international support was

made by the host government a week after the earthquake. The media lime-light in

the developed world was at the time occupied by severe snow storms in the US and

torrential rain and landslides in southern Europe. The attention that the event

received is barely justified, but supported by the high vulnerability of the country and

the donor countries’ willingness to provide support to the nation that had been

relatively spared from sudden-onset events. The fault is hence in the political element

of the model. The Dakhli event, on the other hand, is a definite case of an

intermediate attention event with deaths in the hundreds in both Afghanistan and

Tajikistan. Due to the exceptionally complex nature of the event, as described in

section 7.3.2, the resulting international relief mission was sizeable and multifaceted.

With four sitreps, this event is bordering to a high attention event. The major impacts

were caused by the secondary disasters, without which it is likely that the

international community would not have paid any attention. The international relief

did not include traditional forms of aid, like SAR and medical assets, but was centred

on airborne transport of relief in-country and on equipment and expertise to deal with

the natural dam and landmine threat.

11.3.2 Over-prediction The overestimated events were predicted by the model to receive more attention

than they did. This type of misclassification does not have as dire consequences as

under-predictions. Still, it would be preferable if the model dealt with these cases

correctly. Three of the five events in this category took place in Pakistan: the 1997

- 187 -

Harnai event (id:17), the 2002 second Diamer event (id:48) and the 1994 Hindukush

event (id:5). The other two occurred in Iran and Afghanistan.

Incorrect predictions of high attention level The Harnai event was predicted to receive high attention, but received none.

The high magnitude earthquake (7.3Mw) combined with a relatively densely

populated surrounding spelled disaster. The hypocentral depth of the earthquake

was, however, not determined exactly but expected to be shallow. It is not clear from

the data in the NEIC database if in retrospect it was determined to have been shallow.

Although initial reports claimed that the event resulted in deaths in the hundreds and

even thousands, the final figure was 40. The event did rightly not receive any

international attention. It is likely that the hypocentre was deep and therefore made

the effects of the earthquake less harmful. It is also possible that local geology or

architecture not part of the model lowered the vulnerability of the affected area.

Nevertheless, the fault here is likely in the impact element of the model.

The second Diamer event was predicted to receive high attention, but was only

observed to receive intermediate attention. The event was caused by a strong

earthquake that hit the area just months after an almost equally strong earthquake

had hit. The vulnerability of the already exposed local residents caused by this

double strike is not adequately modelled in the prediction. Neither is the situation

that the area already had received substantial aid following the first event. On its

own, the event would have called for greater attention; however, in combination with

the preceding event, this was not the case. This highlights the need to include

indicators of the context in which the earthquake strikes to strengthen the needs

element of the model.

Low attention events predicted as intermediate Three events were observed to receive low attention but predicted to receive

intermediate level of attention. These were the 1994 Mazar-I-Sharif event (id:3) in

Afghanistan, the 1994 Hindukush event (id:5) in Pakistan and the 1999 Bandar-e-

Abbas event (id:28) in Iran. A potential cause of the misclassification in the

Hindukush event is the categorisation of the indicators. The event had a population

just exceeding 45 000, the threshold dividing the urban and rural categories and

hypocentral depth just shallower than the 40 km threshold for the shallow category.

The event could hence just as well been classified as deep focus and low population

density and would in that case have been correctly classified as a low attention event.

- 188 -

This puts the fault in the modelling methodology and to the insufficient size of the

population of analysed events.

The Bandar-e-Abbas event was caused by a strong and shallow earthquake with

local population just exceeding the threshold to be categorised as urban. There were

no reported casualties of the event and only one sitrep issued and that with limited

information. This misclassification can be the result of an inexact impact element or

an example of an event in which the issuing of sitrep has been non-standard.

The Mazar-I-Sharif event resulted in a final death toll of 160 people and tens of

thousands of damaged buildings. Although needs (clothing, tents, water, cooking

material etc.) for international relief were outlined in the sitrep issued for the event,

there are no reports of aid actually having been dispatched. An intermediate level of

international attention would have been appropriate for this event and it is not clear

why it did not materialise. A potential cause is the concurrent landfall of a serious

cyclone in Bangladesh absorbing the attention from the international media.

11.3.3 Weaknesses Eleven events out of 58 were misclassified, equalling 19 percent. The most

serious errors being the high attention events in Iran being classified as intermediate

attention events. However, considering the context of the misclassifications the model

performs with potential. It is clearly not accurate enough to be used as is by users. It

does, however, provide fertile ground for the development of future models, as will

be discussed in section 12.2.

The misclassifications put emphasis on the requirement for a more

comprehensive impact-component in the model and on a greater population of events

to allow for the analysis of continuous variables, rather than their categorised

versions. The indicators in the final model have their centre of gravity on the impact

estimation in Table 10.1. In Figure 10.1 it was made clear that the intention of the

research project was to bypass those two stages in the estimation process to proceed

directly to a prognosis of the resulting international attention. The final list of suitable

IVs (see Table 14.4 on page 213) are, however, all already used in existing impact

estimation models. This puts into question whether the model is predicting the event

impact rather than the resulting international attention. This does not cast doubt on

whether the research is novel or relevant. The innovative characteristics of the project

are not in the selection of predictor variables but in the development of a probabilistic

- 189 -

model and in the use of the sitrep frequency as a dependent variable. The lack of

indicators targeted at media influence, political relations, or international presence is,

however, a weakness. These weaknesses are discussed further in the next chapter.

- 190 -

12 CONCLUSION This project aimed to improve the international relief to sudden-onset disasters

by identifying novel ways of supporting the decision process surrounding it. It did so

by focusing on the alerting of decision makers in the international donor, implementer

and co-ordinator organisations. The final model was accurate in 81 percent of the

events. The errors that were made were, however, serious. The model was also tested

on two events that occurred in the late stages of the research project, but with

disappointing results. Due to the high level of inaccuracy, the model can not be used

by the international relief community. It does, however, provide a concept and

methods that can be used to improve existing alert tools. What this research has

shown is that although the prediction of the international attention is difficult, it is

feasible.

12.1 Aim and objectives The first objective was to establish a set of user requirements on an alert tool and

to determine the relevance of such a tool to the users. An investigation into the

relevance showed that they have the potential of giving decision makers more time to

collect information from conventional sources. Current tools for this kind of alerting

are based on estimated human losses or, in the case of earthquakes, on the seismic

magnitude. Instead of estimating losses, this study attempted to estimate the

resulting international response. The implementing organisation required an alert

within one hour following a potential disaster in order to take an entry decision

within six hours. Their temporal bottle-neck was the preparation of logistic,

particularly air transport. The requirement on the content of the alert depended on its

purpose. For an initial alert, all that was required by the implementing organisation

was a notification of whether international relief would be required. For the

stakeholders, the accuracy of the alert was secondary to its speed, but they also

pointed out that the level of accuracy should be known.

The second objective was to collect, quantify and store information produced by

stakeholders in the context of international relief missions. A relational MySQL

database, called the INTEREST database, was developed for this purpose. To enable

the quantification and storage of uncertain data extracted from reports in the

international relief context a set of taxonomies of data on loss, needs and relief were

developed. These taxonomies allowed for approximate data to be stored at a higher

- 191 -

level of generalisation. After an exploratory analysis of the database, targeting

alternatives for proxy indicators of international attention, the frequency of OCHA

Situation Reports (sitreps) was chosen. The data that were collected and stored in the

INTEREST database as part of this objective proved to vastly exceed what was

required to achieve the third objective.

The third objective was to develop and evaluate a prototype DSS. A range of

predictor variables (IVs) were evaluated for their predictive power on the frequency

of sitreps (the DV). Ordinal logistic regression was applied to achieve a three-level

ordinal categorical output. This ordinal alert level indicates the expected amount of

information that will be generated on the event in the international community; which

was defined as the international attention. As with alerts based on loss estimations,

there is a need to communicate the level of certainty in the output to the users of the

alert. To avoid making the output too complex, which is a current problem for users

of loss-based alert tools, the output alert level is coupled with a probability. The

calculation of the probability is facilitated by the use of logistic regression. The three

levels of ordinal alerts are:

1. No international attention: Little or no information on the event is

expected to be generated in the international community. Local and

national stakeholders will respond the event.

2. Intermediate international attention: Some information is expected to be

generated on the even in the international community. Regional and

some international stakeholders will respond to the event.

3. High international attention: A significant amount of information is

expected to be generated on the event by the international community.

Many international stakeholders will respond to the event.

The result of this thesis has the potential to fulfil its aim to improve the decision

making in relation to international relief missions to sudden-onset disasters. The

current model will not be of direct help to the decision process due to its low

accuracy.

12.1.1 Lessons learnt The project has provided a potentially important input to the domain of

international disaster alerting. In addition to the concept and methods, it also

- 192 -

includes a database on the information flow surrounding the case studies. This

database contains information on the workflow surrounding all the 59 central Asian

case studies. It can be used for a range of research relating to international post-

earthquake relief processes.

The main two obstacles in the development of the model were the uncertainty in

the collected data and the relatively small sample of events. The predictor variables

were categorised to better cope with the uncertainty and the number of predictor

variables in the model development was reduced to avoid the small sample to

negatively affecting the statistical analysis. This limitation could be avoided in future

research by increasing the population of studied events to include earthquakes in

developing countries world-wide.

Due to the restriction of the number of predictors that could be included in the

final model the predictors that were not related to the loss assessment model could

not be included. If future models take benefit of recent advances in the domain of

remote loss assessment and use loss estimations as predictors, instead of attempting to

emulate the losses indirectly, additional predictors with political emphasis could be

included. Furthermore, if estimated losses were used as a predictor the model could

become hazard-independent. This could allow for the inclusion of other types of

sudden-onset hazards in the modelling.

Further studies of the user requirements of the various users of alerts tools

could prove beneficial. A focus study of one of the three groups identified as part of

this study, implementing (operational decisions) organisations, co-ordinating (tactical

decisions) organisations and funding (strategic decisions) organisation, is likely to

reveal additional requirements.

12.2 Future research There are two directions that the research can take based on the results of this

thesis: (1) The expansion of the developed model and concept to create a more

accurate or more geographically applicable model, or (2) the use of the developed

database for other purposes in relation to international relief. These two directions

are discussed separately below.

- 193 -

12.2.1 Potential model improvements The model development was hampered by low data quality, worsened by a

limited population of case study events. In turn, these two factors lead to limitations

in the choice of indicators and in the choice of analytical methods. By expanding the

analysis to include all earthquake-prone countries the greater population of events

could allow for better analysis. A geographically expanded search for patterns in the

deviations of the international attention would reveal if the country-dependent

misclassification detected as part of this study are real or simply by-products of the

limited sample.

Identify and include missing indicators The use of ordinal regression put restrictions on the size of the set of indicators

that could be included in the modelling. The decision only to include the most

influential indicators leads to an emphasis in the final model on the indicators of the

impact element. Consequently, the indicators of the desired non-impact aspects

presented in section 10.2 could not be included to the desired extent. Having the non-

impact elements in the centre of interest for the research project is a setback. As a

result, the final model is completely hazard dependent and is missing several factors

that affect the international response. For instance, the analysis of the model output

showed that knowledge of the composite GNA index or the level of press freedom

does not support the determination of the likelihood of international relief being

requested by the affected nation and thus spurring the international response and the

number of sitreps. Furthermore, Walker (1991) stated that international SAR relief is

only required in the response to collapses of RC structures. The current model does

not recognise this fact and treat all types of structures equally. It is safe to assume that

there are additional such indicators that have to be included in future models to

achieve increased accuracy in the prediction of the international attention. It is not

clear which these indicators are, but they will relate to the non-impact characteristics

outlined in section 10.2. There is also scope for improvement of the already included

indicators of macroscopic vulnerability. The current model suffers from insufficient

spatial or temporal resolution of the applied indicators of for instance population

density (valid 2004 only), GNA (valid for 2005 only) and WPFI (valid for 2005 only).

This study did not take into account the individual bi-lateral relations between

donor nations and the affected country. A more detailed study where indicators of

the health of bi-lateral relations, such as trade, is included could improve the model’s

- 194 -

ability to predict the actions of major donors and to better attribute the causes of

irregular donations.

The attempts at including media coverage and impact in the model were

unsuccessful due to the incomplete coverage of the collected media reports and due to

the insufficient resources available for quantifying the content of individual reports.

For instance, the spatial analysis of the issuing of media reports was made impossible

due to insufficient meta-data in reports issued by the media. In a practitioner

conference Berthlin mentioned that information ‘black-holes’, i.e. areas from which no

reports emanate, can be used as crude indication of the geographical spread of an

event. This idea was of interest to the research project. However, without a geo-

referenced point of origin of the media reports, such analysis could not be made.

Although the media often include a reference to the field office that produced a

report, this is provided with a very low resolution. For instance, for the international

media organisations included in the study it was common to only have one office

covering the whole region. A solution would have been to analyse national media.

However, access to national media records for the studied period requires on-site

visits and the requirement of translation would have been overwhelming. Although

significant amounts of data on media activity were collected for the case studies, the

analysis was limited and future research should expand in this field.

Reduce hazard dependence Low data quality was also prominent in the representation of the sudden-onset

hazard in the model. For instance, the earthquake magnitude first reported by the

seismological institutions is often inaccurate for very strong earthquakes, which could

distort the loss assessment component of the international attention prototype.

Similarly, the initial reports of hypocentral depth are approximations made by

seismologists based on experience (Sambridge et al 2003) and when a definite depth

cannot be determined for an earthquake that is suspected to be shallow it is reported

by the NEIC with the default value of 33km (Menke and Levin 2005). The depth of

the earthquake is just as important as its magnitude in the impact assessment (Wyss

2004b). The combination of an approximate depth and uncertain magnitude therefore

introduces additional vagueness in the model. In addition, there are several aspects

of the earthquake hazard that are currently not represented in the international

attention model. For instance, very strong earthquakes affect tall and big buildings on

greater distance due to the frequency of the shaking (Bolt 2004:16-17). With no

- 195 -

identified data source with global coverage on building height, this could not be

added to the international attention model. The spatial distribution of the impact is

not accurately represented by the 50 km radius circle currently applied in the model

(Hewitt 1997:220), lacking any consideration to factors like hypocentral depth,

magnitude, fault shape and local geology. As discussed in the methodology chapter,

this adds up to a situation where the appropriateness of the depiction of the

earthquake is questionable.

Earthquakes were chosen as archetypes of sudden-onset disaster, but the

researcher is not a seismologist or an earthquake engineer. Scientists in those two

domains have long been investigating numerical models for impact estimation for

earthquakes. The purpose of this project is not to model the earthquake, but to model

the resulting international response. The international attention model is, however,

indirectly dependent on the losses caused by the hazard. Although this step is not

explicitly calculated in the model, the impact assessment element of the international

attention prototype is simplistic. If other impact estimation models developed by

phenomena experts can provide real-time estimations of the impact, this would

remove the need for the international attention model to attempt to do so. Output

data from impact estimation models, like numbers of killed and injured, could be

used as input to the international attention model. It would make sense to build upon

existing earthquake loss estimation models that operates in real time for earthquakes

anywhere in the world. Such tools did not exist when this research project started,

but with the continuous development of tools like PAGER and QUAKELOSS, this

functionality is only around the corner. If the international attention model could

benefit from an impact estimation output, it would mean that impact estimations of

other hazards could be used as a model input and thus make the international

attention model hazard independent.

In summary, the distancing of the project from the earthquake hazard would kill

two birds with one stone: it would open the door to multi-hazard applicability and it

could increase the accuracy of the attention prognosis through a focus on the

strengths of the model.

Improve user requirements The researcher did his outmost to live up to the advice of Tsui in the creation of

early warning systems to “Define user needs and utilise data sets and formats that

- 196 -

directly support decision-making” (2002:14). The research process revealed separate

user groups with conflicting expectations on an alert system. The differences between

the user groups were not fully examined due to time limitations. These differences

should be investigated further to achieve a better understanding of the requirements

of each user group: Strategic, Tactical and Operational users. The corresponding three

groups identified as part of this study, donors, co-ordinators and implementers are

likely to be an incomplete set of all the types of users. The identification of all users

groups in the context of international relief to sudden-onset disasters and the

determination of the possible ways to support work of the various groups with DSS,

in addition to alert systems, would open new paths for the research and would

facilitate the improvement of the prototype developed in this project.

12.2.2 Database use for other applications A significant amount of data has been collected as part of this research project,

but only a small fraction was used in the development of the prototype. The data and

patterns found in the INTEREST database provide fertile ground for future research

projects in areas not necessarily related to DSS or earthquakes. Some ideas of

prospective research subjects are presented here.

Evaluating international aid impact If future prognostic models are to adopt a normative stance and attempt to

detect forgotten crises, the suitability of international aid in the case studies will have

to be examined. Sundnes and Birnbaum (2003) presented a conceptual model of one

of the challenges faced in that task (see Figure 12.1). They call their model the Best

Outcome Without Assistance (BOWA).

- 197 -

Source: Sundnes and Birnbaum (2003)

Figure 12.1 The BOWA model

Figure 12.1 show how the best outcome with regards to restoring the

functionality in an affected society can be affected positively or negatively by external

interventions and look like a failure even though it was a success or vice versa. Using

data collected in this project, complemented with additional qualitative data on the

aid impact collected on the ground, this analysis is feasible. By combining data on

external relief data with proxy indicators of societal functionality (e.g. number of

homeless, number of injured without treatment); the suitability of individual

interventions could be assessed. The best option would of course be to combine such

a quantitative analysis with interviews with the affected community to evaluate their

satisfaction with the international relief. Similarly, the differences between the

reported needs and the resulting international response stored in the database could

help to explain the selectivity in international relief.

Improved taxonomies of domain data The current taxonomies for loss, needs and relief, where developed on a basis of

trial and error. A review of their suitability for use in other regions could provide an

input on how to improve the detail of the classification while maintaining its wide

geographical applicability. This would allow for more detailed analysis and

comparison of international relief to sudden-onset disasters globally.

Time-series analysis The data collection for this project included not only the final established figures

for the case studies, but also all the changes leading from the initial reports to those

- 198 -

final figures. This data collection was complicated and time-consuming. However,

apart from the explorative analysis (see appendix A-3), the time-series data remained

untouched in the analysis. It is the researcher’s belief and hope that the time-series

data that were superfluous to this study will come to use in other research projects

interested in the information flow and data quality surrounding international relief

interventions.

- 199 -

13 REFERENCES ACC, 1995, “Statement on the Role and Functioning of the Resident Coordinator System”,

United Nations Administrative Committee on Coordination (ACC), ACC/1995/1

ADPC, 2000, “Post-Disaster Damage Assessment and Need Analysis”, Asian Disaster Preparedness Centre, Pathumthani, Thailand

AIMS, 2004, “Who’s doing what where report”, Kabul: Afghanistan Information Management Service

Albala-Bertrand, J.M, 1993, Political Economy of Large Natural Disasters, Oxford: Clarendon Press

Aleskerov, F., A.I. Say, A. Toker, H.L. Akin and G. Altay, 2005, “A cluster-based decision support system for estimating earthquake damage and casualties”, Disasters, 29(3):255-276, Oxford: Blackwell Publishing

Alexander, D.E, 1984, “Death and Injuries in earthquakes”, Disasters, 9 (1), 57-60, Oxford: Blackwell Publishing

Alexander, D.E, 1993, Natural Disasters, London: Routledge

Alexander, D.E, 1995, “Armenian earthquake leads to calls for building reform”, in F.N. Magill (ed.) Great Events from History II: Ecology and the Environment, Vol. 5, 1985-1994, Pasadena, California: Salem Press, 1926-1931.

Alexander, D.E, 2000a, Confronting Catastrophe, Harpenden: Terra Publishing.

Alexander, D.E, 2000b, “On the Spatial Pattern of Casualties in Earthquakes - Forecasting mortality for various kinds of earthquakes”, Annals of Epidemiology, Volume 10, Number 1, January 2000, pp. 1-4(4), Elsevier Science

Alexander, D.E, 2002, Principles of Emergency Planning and Management, Harpenden: Terra Publishing

Al-Khudhairy, D.H.A., I. Caravaggi and S. Giada, 2003, “Application of change detection analysis and object-oriented classification techniques for structural damage assessment – Study sites: Jenin and FYROM”, Ispra: European Commission Joint Research Centre, Technical Note No. I.0.03.06

Al-Khudhairy, D.H.A., I. Caravaggi, S. Giada, I.M. Shepherd and D. Eriksson, 2002a, “Use of commercial very high-resolution data for timely humanitarian relief and aid development support”, Ispra: European Commission Joint Research Centre, Technical Note No. I.02.50

Al-Khudhairy, D.H.A., I. Caravaggi, S. Giada and I.M. Shepherd, 2002b, “Application of remote sensing, object-oriented classification techniques and GIS for post-conflict structural damage assessment”, no. I.02.94, Ispra: European Commission Joint Research Centre

Al-Khudhairy, D.H.A. and S. Giada, 2002, “A road map to acquiring time-critical EO data to respond to humanitarian crises situations”, Ispra: European Commission Joint Research Centre, Technical Note No. I.03.03

Altan, O., 2005, “Use of Photogrammetry, Remote Sensing and Spatial Information Technologies in Disaster Management, especially Earthquakes”, in Geo-information for Disaster Management, 311-322, Van Oosterom, P., S., Zlatanova and E.M. Fendel (eds.), Springer

Arney, G., 1990, Afghanistan, London: Mandarin.

Andersen, E.S and P. Gottschalk, 2001, Information Technology Management, Oslo: Scandinavian University Press / Universitetsforlaget

- 200 -

Atwood, L.E. and A.M. Major, 1998, “Exploring the "cry wolf" hypothesis”, International Journal of Mass Emergencies and Disasters, 16(3): 279-302.

Badal, J., M. Vazqyez-Prada and A. Gonzalez, 2004, “Preliminary quantitative assessment of earthquake casualties and damages”, University of Zaragoza, Proceedings of the European Seismological Commission XXIX General Assembly

Bankoff, G. (ed.), G. Frerks, D. Hilhorst, 2003, Mapping vulnerability: Disasters, Development, and People, London: Earthscan publications

Benini, A.A., 1997, “Uncertainty and Information Flows in Humanitarian Agencies”, Disasters, 21(4):335-353, Oxford: Blackwell Publishing

Benini, A.A., 1998, “Network without a center? A case study of an organizational network responding to an earthquake”, Natural Hazards Research Working Paper #100, University of Colorado

Benini, A.A., W. Barron, C. Conley, J. Donahue, L. Gaym, S. Messick and I. Shaikh, 2005, “Rapid humanitarian assessments – How rational? A Value-of-Information Study of Two Assessments in Iraq [Navigating Post-Conflict Environments series]", Washington D.C.: Vietnam Veterans of America Foundation, http://www.vvaf.org/references/reports/rapid-assessments.pdf, accessed 19 October 2005

Benthall, J., 1993, Disasters, relief and the media, New York: I.B. Tauris & Co Ltd.

Beroggi, G.E.G and W.A. Wallace, 1995, “Decision Support for Real-Time Management of Hazardous Operations”, Computer Supported Risk Management, 195-210, Beroggi, G.E.G and W.A. Wallace, Kluwer Academic Publishers

Berz, G. and A. Siebert, 2004, World of natural disasters, CD-ROM, Munich: Munich Re Group

Best, C, E. van der Goot, K. Blackler, T. Garcia, D. Horby, R. Steinberger and B. Pouliquen, 2005, “Mapping world events”, in Geo-information for Disaster Management, 683-696, Van Oosterom, P., S., Zlatanova and E.M. Fendel (eds.), Springer

Billing, P. and V. Siber, 2003, “ECHO Strategy 2004: Global humanitarian Needs Assessment (GNA); Methodological notes”, Brussels: European Commission, ECHO 4/PB D(2003)

Billing, P., 2004, “Policy guidelines regarding ECHO entry and exit strategies”, Brussels: European Commission, ECHO 4, this information was also given to me verbally from the author in meetings in 2004

Bhaduri, B., E. Bright, P. Coleman and J. Dobson, 2002, “LandScan: Locating People is What Matters”, Geoinformatics Vol. 5, No. 2, pp. 34-37

Bloom, D.E., P.H. Craig and P.N. Malaney, 2001, “The Quality of Life in Rural Asia – A study of rural Asia - volume 4”, Asian Development Bank, Oxford: Oxford University Press, p47-64, p183-190

Blaikie, N., 2000, Designing Social Research, Oxford: Blackwell Publishing.

Bolt, B.A., 2004, Earthquakes, 5th ed., New York: W.H. Freeman and Company

Bonilla, M.G, R.K. Mark and J.J. Lienkaemper, 1984, “Statistical Relations Among Earthquake Magnitude, Surface Rupture Length, and Surface Fault Displacement”, USGS, Report 84-256

Broad and Sanger, 2006, “New worry rises after Iran claims nuclear steps”, International Herald Tribune/New York Times, Monday, April 17, 2006

Brossel, V, 2002, “Afghanistan: The situation of press freedom one year after the fall of the Taliban”, November 2002, Paris: Reporters Without Borders

- 201 -

Brzev S., M.R. Greene, S.M. Alcocer, C. Arnold, S. Cherry, C. Comartin, I.Davis, F.Naeim and R.Sinha, 2001, “Development of a web-based encyclopedia of housing construction types in seismically prone areas of the world”, Proceedings, 12th European Conference on Earthquake Engineering, Ref 821, Elsevier Science

Cahill, K.M., (ed.), Emergency Relief Operations, New York: The International Centre for Health and Cooperation

Campbell, J.B., 2002, Introduction to Remote Sensing, 3rd ed., Taylor & Francis.

CASRI, 2006, Central Asian Seismic Risk Initiative, website, http://www.casri.org/, accessed April 2006

Chung, H.C, B. Adams, S. Huyck, S. Ghosh and R. Eguchi, 2003, “Remote Sensing for Building Inventory Update and Improve Loss Estimation in HAZUS99”, MCEER, Proceedings, Workshop on Application of Remote Sensing Technologies for Disaster Response, Beckman Center of the National Academies, Irvine, CA, September 12, 2003

CIA, 2006, “2005 World Fact Book”, Langley: Central Intelligence Agency, website, http://www.cia.gov/cia/publications/factbook/, Accessed April 2006

Coburn, A. and R. Spence, 2002, Earthquake Protection, 2nd edition, Chichester: John Wiley & Sons Ltd

Comfort, L., 2000,”Information technology and the efficiency in disaster response: The Marmara, Turkey earthquake, 17 August 1999”, Quick response report 130, Pittsburgh: University of Pittsburgh, NSF Grant: CMS-9632458

Comfort, L., M. Dunn, D. Johnson, A. Zagoreckim and R, Skertich, 2004, “Coordination in complex systems: Increasing the efficiency in disaster mitigation and response”, International Journal of Emergency Management, (IJEM), Vol. 2, No. 1/2

Currion, P., 2003, “Surviving ‘Droughts’ and ‘Floods’: Stretching the metaphor for Humanitarian Information Management”, Symposium: Toward an International System Model in Emergency Management, http://www.riskinstitute.org , Fairfax: Public Entity Risk Institute (PERI)

Dalton, M., K. von Hippel, R. Kent and R. Maurer, 2003, “Study four: Changes in humanitarian financing: Implication for the United Nations”, New York: UNOCHA

Darcy, J. and C-A. Hofmann, 2003, “According to Need? Needs assessment in the humanitarian sector”, HPG Report 15, London: Overseas Development Institute – Humanitarian Policy Group

Darcy, J. and C-A. Hofmann, 2004, “Measuring the impact of humanitarian aid – A review of current practice”, HPG Report 17, London: Overseas Development Institute – Humanitarian Policy Group

De Groeve, T. and D. Ehrlich, 2002, “DMA Earthquake Alert Tool – A decision support tool for humanitarian aid”, Ispra: European Commission Joint Research Centre, Technical Note No. I.02.75

De Groeve, T. and D. Eriksson, 2005, “An evaluation of the performance of the JRC earthquake alert tool”, Ispra: European Commission Joint Research Centre

De Ville de Goyet, C., 1993,”Post disaster relief: The supply chain management challenge”, Disasters 20(3), Blackwell

Dilley, M., 2004, “Natural Disaster Hotspots: A Global Risk Analysis”, Powerpoint presentation, International Research Institute for Climate Prediction

Durch , W.J., 2004, “Strengthening UN Secretariat Capacity for Civilian Post-Conflict Response”, Proceedings: Strengthening the UN's Capacity on Civilian Crisis Management, Copenhagen, Denmark, 8-9 June 2004

- 202 -

Earle, P.S., D. Wald, L. Lastowka and M. Donnelly, 2003, “Rapid assessment of shaking impact following global earthquakes”, Eos Transactions, AGU, 84(46), Fall Meeting Suppl.

Earle, P.S., J. David, J. Wald and L. Lastowka, 2005, “PAGER—Rapid Assessment and Notification of an Earthquake’s Impact”, USGS Factsheet 2005-3026, Boulder: USGS

Ebersole, J.M., 1995, “Mohonk criteria for humanitarian assistance in complex emergencies”, Disaster Prevention and Management, Vol 4, No 3, pp 14-24, MCB University Press

ECHO, 2002, “Evaluation of the first DIPECHO action plan for South Asia 2002”, Harold Lockwood (Team Leader), Anthony Conlay, Brussels: ECHO

EERI, 2003, “The Boumerdes, Algeria, Earthquake of May 21, 2003 – EERI Learning from Earthquakes Reconnaissance Report”, no 2003-04, Earthquake Engineering Research Institute

Eguchi, R.T., C.K. Huyck, B.J. Adams, B. Mansouri, B. Houshmand and M. Shinozuka, 2003, “Resilient Disaster Response: Using Remote Sensing Technologies for Post-earthquake Damage Detection”, Buffalo: Multidisciplinary Center for Earthquake Engineering Research (MCEER)

EMM, 2006, website, Electronic Media Monitor, http://press.jrc.it, accessed April 2006

Eriksson, D. and D. Stanojlovic, 2000, “EOD IS – Ett system införande med omfattande krav på organisationsanpassning”, BSc thesis, Linköping University Department of Computer and Information Science, ISRN LIU-IDA-C—99/34--SE

Eriksson, D., 2005, “An Evaluation of Data Sources for Entry Decision Support in Rapid-onset Disaster”, Ispra: European Commission Joint Research Centre, EUR21550EN

Faulkner, X., 2000, Usability Engineering, Palgrave.

Fischer, H. W. III, 1998, “Response to disaster – Fact versus Fiction & its perpetuation”, Lanham: University Press of America, 2nd edition

Frolova, N., 2006, “Multi-Hazard Risk Assessment at different Levels with Extremum System Application”, presentation by Nina Frolova at the 2006 EWCIII in Bonn, Germany

Fujita, T.T, 1973, “Tornadoes around the world”, Weatherwise, 26, pp. 56-62

GDACS, 2006a, website, http://www.gdacs.org/

GDACS, 2006b, Global Disaster Alert and Coordination System, email alert broadcast, 27th of April 2006, http://www.gdacs.org/

Geohazards, 1996, “Urban Earthquake Risk Management Strategies for Former Soviet Central Asian Republic”, proceedings from workshop on strategies for urban earthquake risk management in Central Asia, http://www.geohaz.org/

Geohazards, 2006, Lessons for Central Asia, Geohazards International, hand-out, no publication date, http://www.geohaz.org/

Ghafory-Ashtiany, M., 1999, “Rescue operation and reconstruction of recent earthquakes in Iran”, Disaster Prevention and Management, Volume 8, Number 1, pp. 5–20

Glantz, M.H. (ed.), 2004, “Early Warning Systems: Do’s and Don’ts”, Report of Workshop on Usable Science, Shanghai, China 20-23 October 2003, http://www.ccb.ucar.edu/warning/ , accessed June 2005

GLIDE, 2006, website, http://www.glidenumber.net, accessed January 2006

Guarino, N., 1998, “Formal Ontology and Information Systems” in Formal Ontology for Information Systems, Proceedings of FOIS ’98, Trento, Italy, 6-8 June 1998, Amsterdam: IOS Press

- 203 -

Gutiérrez, E., F. Taucer, T. De Groeve, D. Al-Khudhairy and J. M. Zaldivar, 2005, “Analysis of Worldwide Earthquake Mortality using Multivariate Demographic and Seismic Data”, American Journal of Epidemiology, 161(12):1151-1158

Hamburg, M., 1991, Statistical Analysis for Decision Making, Fifth edition, Harcourt Brace Jovanovich

Hewitt, K., 1997, Regions of Risk – A geographical introduction to disasters, Essex: Addison Wesley Longman Ltd

HABITAT, 2003, Human Settlements Statistical Database version 4 (HSDB4-99), UN Human Settlements Programme, http://www.unhabitat.org/programmes/guo/guo_hsdb4.asp, annual updating, accessed January 2006

HEWS, 2006, website, http://www.hewsweb.org/, accessed February 2006

Holm, H-H, 2002, “Failing Failed States: Who Forgets the Forgotten?”, Security Dialogue, 33: 457-471, Sage

Hosmer, D.W. and S. Lemeshow, 2000, Applied Logistic Regression, 2nd edition, Toronto: John Wiley & Sons

IASC, 2006, website, http://www.humanitarianinfo.org/iasc/, accessed February 2006

ICDO, 2002, “Emergency agency of the republic of Kazakhstan”, International Civil Defence Organisation, website, http://www.icdo.org/National%20structures/, accessed January 2005

IFRC, 1993, “World Disasters Report 1993”, Geneva: IFRC

IFRC, 1995, “World Disasters Report 1995”, Geneva: IFRC

IFRC, 1999, “Vulnerability and Capacity Assessment. An International Field Guide”, Geneva, http://www.proventionconsortium.org/files/vca.pdf, accessed January 2005

IFRC, 2003, “Partnerships in profile 2002-2003”, Geneva: IFRC

IRIN, 2003, “KYRGYZSTAN: Flood and emergency preparedness in the south”, Integrated Regional Information Networks Asia, 30 June 2003, Osh: OCHA, http://www.irinnews.org/

ISDR, 2006, International Strategy for Disaster Reduction (ISDR), website, http://www.unisdr.org/eng/library/, accessed April 2006

Jeffreys, A., 2002, “Giving voice to silent emergencies”, in Forgotten Humanitarian Crises, Conference on the Role of the Media, Decision-makers and the Humanitarian Agencies, Copenhagen: Danish Refuge Council

Johnson, G.O., 1995, “GIS Applications in Emergency Management”, Computer Supported Risk Management, 133-142, Beroggi, G.E.G and W.A. Wallace (eds.), Kluwer Academic Publishers

Kaji, H., Y. Ogawa, T.M. Rufin Jr, E. Kato, H. Tsujibata, M. Ueno, Y. Tokuhisa and S. Hirabayashi, 1998, “Research report on Qayen Earthquake in Iran, May 1997”, Research report series #34, Nagoya: United Nations Centre for Regional Development

Kandeh, J.M.K., A.W. Ahadi and L. Kumar, 2005, “Using Remote Sensing Data for Earthquake Damage Assessment in Afghanistan: The Role of the International Charter”, in Geo-information for Disaster Management, 829-840, Van Oosterom, P., S., Zlatanova and E.M. Fendel (eds.), Springer

Kario, K., K. Shimanda and F. Takaku, 2005, “Management of cardiovascular risk in disaster”, Journal of the Japan Medical Association (JMAJ), 48(7):363-376

- 204 -

Keen, D. and J. Ryle, 1996, “Editorial: The Fate of Information in the Disaster Zone”, Disasters 20(3):169-172, Blackwell

Keller and Pinter, 2002, Active Tectonics – Earthquakes, Uplift, and Landscapes, 2nd ed., New Jersey: Prentice Hall

Kent, R.C., 1987, The anatomy of disaster relief, London: Pinter Publishers.

Kersten, G.E, 2000, “Decision Making and Decision Support”, In: Kersten, G.E., M. Zbigniew and A. Gar-on Yeh (eds.), Decision Support Systems for Sustainable Development: A Resource Book of Methods and Applications, Kluwer, 29-51

Khan, F.K., 1991, A Geography of Pakistan – Environment, People, and Economy, Karachi: Oxford University Press

King, D.J., 2005, “Humanitarian Knowledge Management”, Proceedings of the 2nd international ISCRAM conference, B. Carle and B. Van de Walle (eds.), Brussels, Belgium, April 2005

Kircher, C.A, 2003, “Near-real-time loss estimation using HAZUS and ShakeMap data”, SMIP03 Seminar on Utilization of Strong-Motion Data, Oakland, California, May 22, 2003: Proceedings, p59-65

Koethe, M.R., 2003, “Emergency Response Management: What Can Business Process Management Provide?” Symposium proceedings, Toward an International System Model in Emergency Management, www.riskinstitute.org, Fairfax: Public Entity Risk Institute (PERI)

Kreger, N, 2002, “Humanitarian vs. Military Mine Action”, Journal of Mine Action, Issue 8.1, June 2004

Lagorio, H.L, 1990, Earthquakes – An architect’s guide to non-structural seismic hazards, Berkeley: John Wiley & Sons Inc.

Lamontagne, M, 2005, “Making Earthquake Notifications More Useful to Emergency Managers”, Seismological Research Letters, 76(3):387-400

Laurini, R. and D. Thompson, 1999, Fundamentals of Spatial Information Systems, Academic Press

Le, C.T., 1984, Applied categorical data analysis, New York: Wiley

Leader, N., 2000, The Politics of Principle: the principles of humanitarian action in practice, Humanitarian Policy Group Report 2, London: ODI

Leader, N., 2001, “Donor governments and capacity-building in Afghanistan”, Humanitarian Exchange, Humanitarian Practice Network, London: ODI

Lomnitz, C., 1974, Developments in Geotectonics – Global Tectonics and Earthquake Risk, Amsterdam: Elsevier

Mahadevan, A., K. Ponnudurai, G.E. Kersten and R. Thomas, 2000, “Knowledge Discovery in Databases and Decision Support”, In: Kersten, G.E. , M. Zbigniew and A. Gar-on Yeh eds. Decision Support Systems for Sustainable Development: A Resource Book of Methods and Applications, Kluwer, 343-365

McConnan, I. (ed.), 2000, The Sphere Project – Humanitarian Charter and Minimum Standards in Disaster Response, Geneva: Sphere Project

Mehrotra, S., C. Butts, D. Kalashnikov, N. Venkatasubramanian, R. Rao, G. Chockalingam, R. Eguchi, B. Adams and C. Huyck, 2003, “Project Rescue: Challenges in Responding to the Unexpected”, Project RESCUE White paper,

Menke, W. and V. Levin, 2005, “A Strategy to Rapidly Determine the Magnitude of Great Earthquakes”, Eos, Vol. 86, No. 19, 10 May 2005

- 205 -

Mitchell, J.K., N. Devine and K. Jagger, 1989, “A contextual model of natural hazard”, Geographical Review, 79(4): 391-409

Mohammadi, A. and A. Elitesh (eds.), 2000, Iran and Eurasia,

Mohavedi, H., 2005, “Search, Rescue, and Care of the Injured Following the 2003 Bam, Iran, earthquake”, Earthquake Spectra, December 2005, Volume 21, Issue 1, pp. 475-485

NEIC, 2004, “Magnitude 9.0 OFF THE WEST COAST OF NORTHERN SUMATRA Sunday, December 26, 2004 at 00:58:53 UTC, Preliminary Earthquake Report”, http://neic.usgs.gov/neis/bulletin/, accessed April 2006

NEIC, 2006, website, National Earthquake Information Centre, http://neic.usgs.gov, accessed May 2006

Neuman, W.L, 2000, Social research methods, 4th edition, Needham Heights: Allyn & Bacon

Nicholds, N. and J. Borton, 1994, “The changing role of NGOs in the provision of relief and rehabilitation assistance: Case study 1 – Afghanistan/Pakistan”, ODI Working paper #74, London: Overseas Development Institute, pp. 19-25, pp. 65-97

Norman, D.A, 1998, The Design of Everyday Things, MIT Press.

O’Brien, J., 1999, Management Information Systems – Managing Information Technology in the Internetworked Enterprise, Boston: Irwin McGraw-Hill

OCHA, 1997, Bojnoord, Iran, Situation Reports, http://www.reliefweb.int

OCHA, 2002, Dahkli, Afghanistan/Tajikistan Situation Reports, http://www.reliefweb.int

OCHA, 2006, website, http://ochaonline.un.org/, accessed April 2006

OFDA, 1987, A review of OFDA involvement in search and rescue: dispatch procedures and recommendations, Washington D.C.: Office of U.S. Foreign Disaster Assistance (OFDA)

OJEC, 2001, Council Regulation (EC) No 381/2001 of February 26 2001 creating a rapid-reaction mechanism, Official Journal of the European Communities (OJEC), Vol. 44, 2nd February 2001, ISSN 0378-6978, 2001/L/57/5

Okazaki, K. (ed.), 2001, “RADIUS – Risk Assessment tools for DIagnosis of Urban areas against Seismic disasters”, Geneva: ISDR

Olsen, G.R., N. Carstensen and K. Hoyen, 2003, “Humanitarian Crises: What determines the level of emergency assistance? Media Coverage, Donor Interests and the Aid Business”, Disasters, 27(2):109-126, Oxford: Blackwell Publishing

Oosterom (van), P., S., Zlatanova and E.M. Fendel (eds.), 2005, Geo-information for Disaster Management, Springer

ORNL, 2006, Oak Ridge National Laboratory (ORNL), website, http://www.ornl.gov/sci/landscan/, accessed February 2006

Papathoma, M., 2003, “Tsunami vulnerability assessment with special reference to Greece”, PhD thesis, Coventry: Coventry University

Peduzzi, P., H. Dao and C. Herold, 2002, “Global Risk And Vulnerability Index Trends per Year (GRAVITY)”, Geneva: UNEP

POST, 2005, “Postnote: Early warnings for natural disasters”, Parliamentary Office of Science and Technology (POST), Number 239, May 2005

Recknagel, C, “Iran: Election Of Ahmadinejad Unlikely To Affect Nuclear Negotiations”, Radio Free Europe, 26 June 2005

RELEX, 2006, Rapid Reaction Mechanism, website, http://ec.europa.eu/comm/external_relations/cpcm/rrm, accessed April 2006

- 206 -

Reliefweb, 2006, website, http://www.reliefweb.int/, accessed April 2006

Renard, M-F., (ed.), 2002, China and its regions – economic growth and reform in Chinas provinces, Cheltenham: Edward Elgar Publishing, p57-87, p147-194

RSF, 2006, Reporters Sans Frontiers, website, http://www.rsf.org, accessed April 2006

Rumer, B. (ed.), 2000, Central Asia and the new global economy, New York: M.E. Sharpe, p3-86, p174-276

Porfiriev, B.N., 1998, “Issues in the definition and delineation of disasters and disaster areas”, in Quarantelli, E.L. (ed.), 1998, What is disaster? Perspectives on the question, pp. 56-75, London: Routledge

Quarantelli, E.L. (ed.), 1998, What is disaster? Perspectives on the question, London: Routledge

SAIC, 2001, “CATS Version 4 Build 60 User’s Manual”, 28 February 2001, San Diego: SAIC

Sambridge, M., T. Nicholson and O. Gudmundsson, 2003, “Earthquake location via the direct use of arrival time catalogues”, EGS - AGU - EUG Joint Assembly, Abstracts from the meeting held in Nice, France, 6 - 11 April 2003, abstract #2429

Schneider, P. and C. Drury, 1999, “HAZUS99 SR2 Technical Manual”, Federal Emergency Management Agency (FEMA), Washington, D. C., http://www.fema.gov/hazus/

Schneiderbauer, S. and D. Ehrlich, 2004, “Risk, hazard, and people’s vulnerability to natural hazards”, Ispra: European Commission Joint Research Centre, EUR 21410EN

Seibold, E, 2003, “Natural Disasters and Early Warning”, in Zschau and Küppers (eds.), Early Warning Systems for Natural Disaster Reduction, 3-14, Springer

Sen, A., 1999, Development as Freedom, Oxford: Oxford University Press

Shakhramanian, M.A. (ed.), V.I. Larionov, G.M. Nigmetov and S.P. Sutschev, 2000, “Assessment of the seismic risk and forecasting consequences of earthquakes while solving problems on population rescue”, Moscow: Russian Civil Defence Institute

Shaw, D.J.B (ed.), 1995, The Post Soviet Republics: A Systematic Geography, New York: Longman Group

Showalter, P.S., 2001, “Remote sensing’s use in disaster research: a review”, Disaster Prevention and Management, 10(1):21-29, Emerald

Smart, P., 2005, “Knowledge-intensive fusion for situational awareness”, University of Southampton, Electronics and Computer Science, DTC/WP150/Scenario, 4th October, 2005

Smillie, I. and L. Minear, 2003, “The quality of money – Donor behaviour in humanitarian financing”, Somerville: Tufts University

SPSS, 2003, Online software documentation, Statistical Package for the Social Sciences v.14, Chicago: SPSS Inc.

Stallings, R.A. (ed.), 2002, Methods of Disaster Research, United States: Xlibris.

Sundnes, K.O and M.L. Birnbaum, 1999, “Health disaster management guidelines for evaluation and research in the Utstein style”, Prehospital and Disaster Medicine, Apr-Jun,14(2):43-52, Jems Pub. Co.

Suslov, S.P., 1961, Physical Geography of Asiatic Russia, San Francisco: W.H. Freeman, p521-575

Swiatek, J.A. and D.C. Kaul, 1999, “Crisis Prediction Disaster Management”, SAIC Science and Technology Trends II, http://cats.saic.com/src/technical.html

Swift, J., 1989, “Why are poor people vulnerable to famine?”, Institute of Development Studies Bulletin, Volume 20, Number 2, April 1989, Sussex: Institute of Development Studies

- 207 -

Tabachnick, B.G. and L.S. Fidell, 2001, Using Multivariate Statistics, Needham Heights: Allyn and Bacon

Tailhades, M and M.J. Toole, 1991, “Disasters: what are the needs? How can they be assessed?”, Tropical Doctor, 21(s1), 18-23

Tashakkori, A. and C. Teddlie, 1998, Mixed methodology – Combining Quantitative and Qualitative Approaches, Applied Social Science Research Methods Series, Vol.46, London: Sage Publications

Tatar, M., Yaminifar, F., Hesami-Azar, K.H. and Hosseinzadeh, N., 1997, “Bojnoord Earthquake of 1997”, IIEES Report No. 76-97-21, IIEES: Tehran.

Taylor, P.D., 2003, “ECHO in Tajikistan and Central Asia”, Brussels: ECHO

Thomas, A., J. Chataway and M. Wuyts (eds.), 1998, Finding out Fast – Investigative skills for Policy and Development, London: Sage Publications

Thywissen, K., 2006, Components of Risk – A comparative glossary, Bonn: United Nations University

Tobin, G.A. and B.E. Montz, 1997, Natural Hazards: Explanation and Integration, New York: The Guilford Press

Tralli, D.M, 2000, Assessment of Advanced Technologies for Loss Estimation, Buffalo: Multidisciplinary Center for Earthquake Engineering Research (MCEER), MCEER-00-SP02

Tsui, E. (ed.), 2002, “Symposium on best practices in humanitarian information exchange”, proceedings, 5-8 February, 2002, Palais de Nations, Geneva: Reliefweb

Tsui, E., 2003, “Initial response to Complex Emergencies and Natural Disasters”, in K.M. Cahill (ed.), Emergency Relief Operations, New York: The International Centre for Health and Cooperation

Twigg, J, 2003, “The Human Factor in Early Warnings: Risk Perception and Appropriate Communications”, 19-26, in Zschau and Küppers (eds.), Early Warning Systems for Natural Disaster Reduction, Springer

Twigg, J, 2004, Disaster risk reduction – Mitigation and preparedness in development and emergency programming, Overseas Development Institute, Humanitarian Practice Network, Good Practice Review, Number 9, March 2004

UN, 2003, The UN System in the Kyrgyz Republic Common Country Assessment 2003, New York: United Nations, http://www.undp.kg/english/publications.phtml

UNAMA, 2003, “Disaster Management Framework for Afghanistan”, Kabul: United Nations Assistance Mission in Afghanistan

UNDAC, 2000, UNDAC Field Handbook, 3rd edition, UN Disaster Assessment and Coordination team, Geneva: OCHA

UNDP, 2004,”Human Development Report 2004 – Cultural Liberty in Today’s Diverse World”, New York: United Nations Development Programme, HDRO

UNDRO, 1989, “Report on International relief assistance for the earthquake of 7 December 1988 in the Soviet Socialist Republic of Armenia”, Geneva: United Nations Disaster Relief Organization

UNESCO, 2003, “Disaster reduction in Iran”, website, http://www.unesco.org/disaster, accessed September 2003

UNSD, 2006, United Nations Statistics Division, “Population density and urbanization”, website, http://unstats.un.org/unsd/demographic/sconcerns/densurb/, accessed April 2006

- 208 -

USAID, 1994, “Field operations guide for Disaster Assessment and Response”, Office for Foreign Disaster Aid (OFDA), USAID

USAID, 2005, “Youth and Conflict – A toolkit for intervention”, USAID Office of Conflict Management and Mitigation, Washington: USAID

Vatsa, K.S and F. Krimgold, 2000,”Financing disaster mitigation for the poor”, in Managing Disaster Risk in Emerging Economies, 129-136, Kreimer and Arnold (eds.), Washington: World Bank

Vereign, H., 1998, “Data Quality Measurement and Assessment”, NCGIA Core Curriculum in GISciene, http://www.ncgia.ucsb.edu/giscc/units/u100/u100.html, posted March 23, 2003, accessed June 2004

VOSOCC, 2006, Virtual On-Site Operations and Coordination Centre, http://ocha.unog.ch/virtualosocc/, accessed April 2006

Walker, P., 1991, “Field Studies Paper No.1 – International Search and Rescue Teams a league discussion paper”, April 1991, Geneva: League of the Red Cross and Red Crescent Societies

Wallace, W.A and F. De Balogh, 1985, “Decision Support Systems for Disaster Management”, Public Administration Review, Vol. 45, Special Issue: Emergency Management: A Challenge for Public Administration, January, 1985, 134-146

Walter, J. (ed.), 2004, World Disasters Report – Focus on community resilience, Geneva: International Federation for the Red Cross and Red Crescent Societies

Weatherwise, 1974, “The hurricane disaster-potential scale”, Weatherwise, 27, pp. 169

Westervelt, J.D. and M. Shapiro, 2000, Combining Scientific Models into Management Models, Proceedings 4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4): Problems, Prospects and Research Needs. Banff, Alberta, Canada, September 2 - 8, 2000.

Wijkman, A. and L. Timberlake, 1984, Natural Disasters: Acts of God or acts of Man?, Earthscan, Washington D.D.

Wisner, B., P. Blaikie, T. Cannon and I. Davis, 2004, At Risk – Natural Hazards, People’s vulnerability and disasters, London: Routledge

Wisniewski, M., 1997, Quantitative methods for decision makers, 2nd edition, Pearson Higher Education

Whitman, R., V. Anagnos, T. Kircher, C. Lagorio, H. J. Scott Lawson and R. Schneider, P., 1997, “Development of a National Earthquake Loss Estimation Methodology”, Earthquake Spectra, Vol. 13(4), 643-662, Earthquake Engineering Research Institute (EERI)

Woodward, R.L, H.M. Benz, K.M. Shedlock and W.M. Brown III, 1997, “Taking the earth’s pulse”, U.S. Geological Survey Fact Sheet 103-97

Wyss, M., 2004a, “Verifying the validity of human loss estimates due to earthquakes, using QUAKELOSS”, Geophysical Research Abstracts, Vol. 6, 05841

Wyss, M., 2004b, “Earthquake Loss Estimates in Real Time Begin to Assist Rescue Teams Worldwide”, EOS, Vol. 85, No. 52, 565-572, 28 December 2004, American Geophysical Union

Wyss, M., 2005, “Human losses expected in Himalayan earthquakes”, Natural Hazards, 34(3), pp.305-314.

Wyss, M., 2006, “WAPMERR: World Agency Of Planetary Monitoring And Earthquake Risk Reduction”, website, http://www.wapmerr.org/, accessed March 2006

- 209 -

Xinhua, 2003, “Earthquake Provision and Disaster Alleviation in China”, Statement of Li Qianghua of the China Seismology Bureau, Xinhua news agency, September 5, 2003

Yuan, Z., 2003, “Development of a GIS interface for seismic hazard assessment”, thesis, Enschede: ITC

Zelterman, D., 2006, Models for Discrete Data, Revised edition, Oxford: Oxford University Press

Zerger, A. and D.I. Smith, 2003, “Impediments to using GIS for real-time disaster decision support”, Computers, Environment and Urban Systems, 27(2003), pp.123-141, Pergamon

Zimmerman, H., 2002, “Emergency services across the borders: Communications for Decision-making in Disaster Management”, proceedings 10th CEPT Conference, Vienna: European Conference of Postal and Telecommunications Administrations

Zschau, J. and A.N. Küppers (eds.), 2003, Early Warning Systems for Natural Disaster Reduction, Springer

- 210 -

INDEX Absolute number.................................................. 84 Accumulative number.......................................... 84 Accuracy ............................................................... 91 Afghanistan 58, 64, 76, 103, 104, 106, 107, 108,

111, 115, 116, 117, 118, 127, 152, 158, 159, 160, 161, 167, 172, 186, 187, 199, 200, 203, 204, 205, 207, 211, 212, 216

Black-hole...................................92, 125, 144, 194 CATS...................................................................... 38 China58, 78, 103, 104, 107, 109, 118, 159, 160,

161, 167, 202, 206, 209, 211, 212 Classification error ............................................... 91 Co-linearity..............................................................xii Commission error.......................31, 142, 157, 169 Completeness ..............................91, 92, 135, 137 Contingency cleaning........................................... 79 Cost-benefit .............................................................8 Data mining................................... xii, 87, 165, 228 Data quality .... 21, 35, 36, 90, 91, 135, 193, 194,

198 Department of Humanitarian Affairs............. x, 181 Disaster management cycle ...................................7 DMA Earthquake Alert Tool ................................. 39 Early warning........................................................ 29

Cascade.....................................................30, 47 Five W’s............................................................ 29 Situational awareness .................................... 29

Earthquake engineering .................................... 102 Earthquake prediction ....................................... 101 Empty cells .....88, 151, 152, 153, 156, 157, 164,

165, 168 Entry decision xii, 3, 52, 119, 124, 130, 131, 133,

176, 178 EUSC ........................................................... x, 56, 61 FEMA.................................................. x, 37, 38, 206 Fire................................................26, 38, 101, 102 GDACS . i, x, 39, 40, 41, 47, 48, 55, 61, 113, 115,

116, 117, 129, 135, 140, 142, 147, 151, 177, 202

Global Needs Assessmentx, 64, 69, 77, 130, 153, 156, 158, 161, 170, 193, 200

Gujarat earthquake.............................58, 134, 164 HAZUS.......................................................x, 37, 204 HEWSWeb............................................................. 38 Image pair ................................................... 31, 140 Intensity raster .......................................................xii Iran.....1, 16, 43, 58, 67, 90, 103, 104, 105, 107,

108, 109, 113, 114, 115, 118, 152, 158, 159, 160, 161, 167, 172, 185, 187, 188, 200, 202, 203, 205, 207, 211, 212, 216

Kazakhstan ....58, 103, 104, 107, 108, 110, 118, 159, 160, 161, 203, 211

KDD process .50, 51, 53, 87, 119, 148, 150, 214 Kyrgyzstan ......58, 103, 104, 107, 108, 110, 117,

118, 159, 160, 161, 172, 186, 212 Landscan...............................................64, 76, 156 Landslide..................41, 101, 111, 113, 115, 186 Link-function ............................................ xii, 86, 88 Liquefaction ....................................................... 101 Logistic regression ...................... xii, 86, 87, 88, 89 Multiple-regression .............................................. 86 Non-quantified attribute ...................................... 84 OLAP ............................................... x, xii, 23, 38, 39 Omission.......................................................91, 141 Ontology............................... 49, 52, 61, 62, 68, 73 Openness .................................................. 160, 161 Ordinal regression.... xii, 2, 86, 87, 156, 165, 169,

182, 193 PAGER........ x, 41, 44, 45, 75, 135, 140, 195, 202 Pakistan... 43, 58, 103, 104, 107, 108, 111, 117,

118, 159, 160, 161, 164, 172, 186, 187, 204, 205, 211, 212

Pressure And Release model .......................... 9, 10 Prevention measures..............................................8 QUAKELOSS ...... 43, 45, 135, 142, 143, 195, 208 RADIUS ................................................................. 39 Real-time .. 23, 28, 33, 38, 39, 42, 43, 46, 47, 52,

75, 101, 126, 129, 135, 142, 152, 154, 156, 176, 195, 204, 209

Relational database..............................................xiii Resolution ............................................................ 91 Risk..........................................................................9 Sudan field visit .............................................56, 61 Swedish Rescue Services Agency ..... x, xi, 55, 119,

120, 121, 122, 123, 124, 126, 178, 183 Tajikistan58, 103, 104, 106, 107, 108, 111, 115,

116, 117, 118, 159, 160, 161, 186, 205, 207, 211

Taliban........................................64, 108, 160, 200 Tele-assessment .................................................. 28 Tsunami.................................................38, 99, 101 Turkmenistan .58, 103, 104, 107, 112, 113, 159,

160, 161, 211 Urban growth....................... 12, 78, 155, 156, 160 Uzbekistan......58, 103, 104, 107, 108, 112, 113,

118, 160, 161 Vulnerability

ex ante ..........................................11, 20, 68, 69 ex post ................................................ 11, 68, 69

Window of opportunity............................................8 World Press Freedom Index .. xi, 64, 78, 154, 156,

158, 161, 170

- 211 -

14 APPENDICES

A-1 Case study descriptives Table 14.1 Case studies and the amount of linked data (two pages)

Year Country Name/Location Sources Reports Attributes 2005 Iran Saravan 3 3 N/A

2005 Iran Zarand 10+ 152 N/A

2004 Iran Mazanderan 5+ 23 N/A

2004 Afghanistan Hindu Kush 3 3 N/A

2004 Pakistan Mansehra 5+ 18 N/A

2003 Iran Bam 18 ?? 307

2003 Iran Masjed 1 1 3

2003 Iran Jahrom 1 2 4

2003 Iran Torbat-e Jam 1 1 2

2003 Kazakhstan Lugovoy 4 3 25

2003 Afghanistan Yakabagh 3 4 6

2003 China Jiashi 2 3 7

2003 Iran Nourabad 2 2 5

2002 Iran Sanandaj 1 2 2

2002 Pakistan 2nd Diamer 5 5 45

2002 Pakistan Diamer 7 6 28

2002 Iran Soleyman (Masjedsoleyman)

2002 Iran Quazvin 11 18 61

2002 Iran Kermanchah 2 2 5

2002 Afghanistan Dawabi 5 6 19

2002 Afghanistan Nahrin 15 21 92

2002 Tajikistan Haut-Badakhchan 2 2 4

2002 Afghanistan Dakhli 7 11 45

2002 Iran Bousheher 2 2 5

2002 Tajikistan Ragoun 7 6 26

2001 Iran Birjand 2 2 3

2001 Afghanistan Gumbahar 2 2 5

2001 Afghanistan Faizabad 2 2 3

2001 Pakistan Badin (Gujarat) 6 8 32

2000 Turkmenistan Balkan Oblast 3 4 4

2000 Tajikistan Khasanov 3 3 11

2000 Iran Mohammadieh 2 2 3

2000 Iran Kachmar 2 3 6

2000 Afghanistan Hindu Kush 2 3 3

1999 Iran Ali-Abad 2 2 3

1999 Iran Shiraz 4 5 13

1999 Iran Bandar-E-Abbas 3 3 9

1999 Afghanistan Baraki Barak (Shaikabad)

6 6 40

1998 Iran Khonj 1 2 3

1998 Iran Lerik 3 4 6

1998 Afghanistan Rustaq (2nd) 19 36 189

1998 Iran Birjand 5 6 9

1998 Iran Golfbaft 4 6 8

1998 Afghanistan Rustaq 31 54 357

- 212 -

Year Country Name/Location Sources Reports Attributes 1997 Iran Qayen (Birjand/Ardekul) 17 45 344

1997 China Kashi (Jiashi) 7 11 22

1997 Iran Ardebil 20 24 129

1997 Pakistan Harnai 3 6 6

1997 Iran Bojnoord (Khorasan) 11 14 75

1997 China Jiashi 8 8 23

1997 Kyrgyzstan Ak-Tala 4 4 14

1996 China Artux 2 4 6

1996 China Lijang 13 124

1996 Afghanistan Maimana 4 3 16

1994 Pakistan Hindukush 2 3 6

1994 Iran Shiraz 3 3 9

1994 Afghanistan Mazar-I-Sharif 3 5 18

1994 Iran Firozabad 7 5 20

1994 Iran Sefid Abeh 9 7 16

1993 Iran Gachsaran 2 2 7

Table 14.2 Example USGS Long earthquake notification message

Region: SOUTHERN QUEBEC, CANADA Geographic coordinates: 45.026N, 73.881W Magnitude: 3.7 Ml Depth: 12 km Universal Time (UTC): 9 Jan 2006 15:35:40 Time near the Epicenter: 9 Jan 2006 10:35:40 Local time in your area: 9 Jan 2006 08:35:40 Location with respect to nearby cities: 19 km (12 miles) NE (54 degrees) of Chateaugay, NY 23 km (15 miles) NW (310 degrees) of Altona, NY 24 km (15 miles) WNW (287 degrees) of Mooers, NY 60 km (37 miles) SSW (192 degrees) of Laval, Québec, Canada 60 km (37 miles) SSW (204 degrees) of Montréal, Québec, Canada

ADDITIONAL EARTHQUAKE PARAMETERS event ID : LD 1017309 version : 1 number of phases : 24 rms misfit : 0.24 seconds horizontal location error : 0.4 km vertical location error : 1.1 km maximum azimuthal gap : 75 degrees distance to nearest station : 31.3 km Flinn-Engdahl Region Number = 447 This is a computer-generated message and has not yet been reviewed by a seismologist. For subsequent updates, maps, and technical information, see: http://earthquake.usgs.gov/recenteqsUS/Quakes/ld1017309.htm or http://earthquake.usgs.gov/

Source: NEIC 2006

- 213 -

A-2 Model development

Table 14.3 Starting model parameters (Cauchit)

Estimate Std. Error Wald 95% Confidence Interval Lower Bound Upper Bound Threshold [AttCat = 1] 4.50 4.75 0.90 -4.81 13.81 [AttCat = 2] 10.03 6.67 2.26 -3.05 23.10 Location [Shallow=0] -7.76 4.14 3.51 -15.87 0.36 [Shallow=1] 0 . . . . [Vulnerable=0] 615.86 3.04 40936.51 609.89 621.82 [Vulnerable=1] 0 . . . . [MagCat=1] 9.59 4.85 3.92 0.09 19.09 [MagCat=2] 2.16 2.70 0.64 -3.14 7.45 [MagCat=3] 0 . . . . [Population=0] -1.33 1.58 0.71 -4.43 1.77 [Population=1] 0 . . . . [HighGrowth=0] -5.04 3.08 2.69 -11.07 0.98 [HighGrowth=1] 0 . . . . [Exposed=0] 4.50 3.77 1.42 -2.90 11.90 [Exposed=1] 0 . . . . [Open=0] -614.30 0 . -614.30 -614.30 [Open=1] 0 . . . . [Night=0] 3.70 2.92 1.61 -2.01 9.42 [Night=1] 0 . . . .

Table 14.4 Full model parameter estimates (Cauchit)

Std. Error 95% Confidence Interval Lower Bound Upper Bound Threshold [AttCat = 1] Low 4.39 -4.85 12.34 [AttCat = 2] Intermediate 6.23 -3.62 20.79 Location [MagCat=1] High 6.77 -1.70 24.82 [MagCat=2] Intermediate 4.43 -4.70 12.67 [Shallow=0] 2.96 -10.26 1.35 [Population=0] 2.99 -10.15 1.57 [Vulnerable=0] 3.16 -0.82 11.57 [HighGrowth=0] 4.09 -16.65 -0.61 [Exposed=0] 4.96 -0.94 18.51

Redundant parameters removed Source: Author; INTEREST Database

- 214 -

A-3 Exploratory analysis This chapter presents the results of the preparatory iterative statistical analysis

conducted as part of O’Brien’s (2002) Systems Design. The results presented here

guided the subsequent development of the prognostic model by providing an

overview of the character of the case studies. As such it can be seen as having been

part of the problem definition and data selection phases of the KDD process, taking

place before and in parallel with the Systems Analysis and Implementation.

Numerous probes of the data did not provide any useful result in relation to the final

research subject. The analysis presented here is hence a summary.

Descriptive analysis A more targeted categorisation of the case study countries was adopted in the

descriptive analysis. In addition to the aforementioned categories, all countries with a

GDP/capita (see Table 7.1) exceeding USD 4 000 will be defined as Rich. Just as with

the categorisation of vulnerability, this does only mean that they are richer among the

case study countries. In the descriptive analysis, to circumvent the problems

generated by the use of non-standard units by the reporting agencies, the data derived

through content analysis is used for frequency analysis only. In other words, the

amount that is requested of a particular relief item is not taken into account. The

number of times that the request was made forms the basis for the analysis. Figure

14.1 shows the distribution of the types of relief item requests made from the affected

areas by either the UN or the affected government. Shelter and food are the most

common requests with financial aid representing 10% of the requests. As can be seen

in Figure 14.2 there are some differences in the patterns of requests made by rich and

poor countries. Requests for shelter are more common in poor countries and requests

for specialised equipment, e.g. excavators and SAR equipment, are more common in

rich countries.

- 215 -

Shelter42%

Food19%

Financial10%

Health9%

Logistics8%

Equipment6%

Other5%

Fuel1%

Poor Rich

OtherFuelGenericEquipmentLogisticsHealthFinancialFoodShelter

Figure 14.1 Relief requests Figure 14.2 Relief request distribution by wealth

International-Government

United Nations INGO Mixed National-Government

National-NGO Commercial

TajikistanPakistan, Islamic Republic ofKazakhstan, Republic ofIran, Islamic Republic ofChina, People's Republic ofAfghanistan

Sum of Donations

Origin

Country

Figure 14.3 Donation destination per origin category

- 216 -

Similar to the needs reporting, to avoid issues with the units used in response

reporting, the amounts are not considered. A donation is hence defined as a

notification from an organisation that they have dispatched a certain type of relief.

The frequency is analysed on an attribute level. In other words, if an organisation

reports to have dispatched money and food it will be counted as two donations. With

this definition of donation the distribution of donations among the donor

organisations can be seen in Figure 14.3. Looking at the donation recipients in Figure

14.4 it is clear that Afghanistan and Iran received the bulk of the donations in the

study and that international governments and the United Nations are the most active

donors. The INGOs in the case studies had the broadest geographical spread in their

donations. In Figure 14.5 the focus is shifted to the content of the donations. The

international commercial organisation donations are a very few and cannot represent

the distribution of all commercial organisations. Two trends are visible in the graph.

International organisations tend to rely on financial donations whereas national

organisations put their emphasis on tangible relief such as shelter, food and logistics.

The difference is clearest when comparing international governments with national

NGOs. The above graphs are produced using the top level relief item types, i.e.

shelter, financial, food, etc. To give an example of the resolution of the database

Figure 14.7 shows the Tier 2 (see Table 5.5) of the Shelter donations. When comparing

Figure 14.6 with Figure 14.1, above, it is clear that the requested relief, in terms of the

number of donations, is not equal to the dispatched relief. This can obviously be due

to the individual dispatches of relief being more voluminous, but it is an indication of

a potential discrepancy.

- 217 -

Afghanistan Iran, IslamicRepublic of

China, People'sRepublic of

Pakistan, IslamicRepublic of

Kazakhstan,Republic of

Tajikistan

CommercialInternational-GovernmentInternational-NGOMixedNational-GovernmentNational-NGOUnited Nations

Sum of Donations

Country

Origin

Figure 14.4 Donation origin per recipient

Government United Nations NGO Commercial Government NGO Mixed

International National Mixed

FuelEquipmentGenericHuman ResourcesHealthLogisticsFoodShelterFinancial

Figure 14.5 Donation type distribution per origin category

- 218 -

Financial34%

Shelter25%

Food10%

Logistics10%

Health8%

Human Resources5%

Generic5%

Fuel1%

Equipment2%

Figure 14.6 Donations

Blankets

Clothing

Shelter

Plastic Sheeting

Tarpaulins

Heaters

Ground Sheet

Rubbhall

Floor cover

Generators

Figure 14.7 Tier 2 shelter donations

- 219 -

Time-series analysis Figure 5.7 shows how the difference between the available minimum and

maximum values of a generic indicator changes over time. The only general rule for

the difference between the maximum and minimum is that it eventually reaches zero

when a definite value is agreed upon. The average time for this is in the cases studies

were, depending on the severity of the event, about a week for loss indicators. Data

on dispatched relief material can not be analysed in the same manner because there

was no source claiming to have an absolute truth at any point. Reports from

individual organisations would say for instance that “we have dispatched 100 tents”.

Data on what was actually dispatched and what was actually received does not exist.

Relief material often kept being dispatched many weeks after the disaster onset and,

at that time, tended to focus on recovery. There were only a few instance were final

figures of donated relief material were provided. In those cases, the data were

provided in post disaster academic reports (e.g. Kaji 1998) or through the VOSOCC.

The accuracy of relief and needs data over time was hence impossible to discern. Loss

data could, however, be extracted from the sitreps, media, national government

reports and INGO/NGOs.

By querying the case study database for all attributes on human injuries using

“persons” as the unit and filter this for absolute figures, it was possible get a closer

look on the accuracy of the data. The commonly reported characteristics include the

number of deaths, the number of injured and the number of structures damaged or

destroyed. Only in rare cases are relative measurements, such as mortality, used. The

last report released on injuries is taken as the real and final figure. The source of the

final report is either the UN or the CRED EM-DAT. Some uncertainties still exist in

the interpretation of what constitutes an injury and possibly what constitutes a

person29, but with the available data, this is as close to objectivity as is achievable.

Figure 14.8 shows the resulting graph of reporting accuracy. The first twenty-four

hours following an event the reported number of injured persons is on the average of

from the final figure by a factor of 16. This number is driven up by outliers in the case

study data. The extremes, which are off the final figure with close to a factor of one

29 This refers to the use of “person” strictly for a civilian person. It is likely that domestic emergency workers and volunteers make up a notable fraction of the human losses in low casualty events.

- 220 -

hundred, are all from Chinese events. Overall, the average of the reports preceding

the final correct report is off target by factor of four.

100.00

120.00

24 96 168 744

Max DeviationAverage Deviation

Figure 14.8 Injury reporting accuracy

Loss Need Response

GovernmentUnited NationsNGO

Average of HourDelay

InfoType

SourceType

Figure 14.9 Average time until first report release

- 221 -

No interestLittle interest

Intermediate interestSubstantial interest

Long interest5

Figure 14.10 Media perseverance per Events

LossTotal

02000400060008000

0 2000 4000 6000

0 200040006000800010000

InterAid

MediaPersever

0 100200300400500600

0 10 20 30 40

MediaFreq

0100200300400500600

NonMediaAttribFreq

0 100 200 300

Source: Author; INTERST database

Figure 14.11 Correlation matrix of media exposure30

30 Legend: Total Human Loss (LossTotal), International Financial Aid (InterAid), Media Perseverance (MediaPersever), Media Reporting Frequency (MediaFreq), and Non-Media Attribute Frequency (NonMediaAttribFreq)

- 222 -

Reporting speed Looking at the reporting speed of the affected government, the United Nations

and NGOs a pattern emerges. Figure 14.9 shows the average delay until the issuing of

the first report according to the source type and report content type. The affected

government is, as would be expected, the first to report loss and needs. The

dispatched relief report is provided equally fast by the UN and the affected

government. The averages are distorted by several outlier events for which the first

reports were issued exceptionally late. The corresponding figure for media reports,

which are not categorised according to their content, is 12.5 hours.

Media exposure Olsen et al (2003) show that media, alongside the agendas of NGOs and geo-

political actors, play an important role in the decision process of international donors.

Media influence can hence not be ignored when attempting to predict the actions of

the international community. The media data were, however, the data with the most

inconsistent coverage for the case studies. This is due to the media reports being

perishable, particularly before the emergence of the Internet and the data collection

thus not being complete. Media exposure is in this study divided into measurements

of perseverance and frequency. Other studies like Best et al (2005) have used finer or

more targeted division using linguistic measurements such as the length of the

reported text or the frequency of certain words in the text. The media data collected

on the case studies would allow for a similar division, but it would require additional

work in digitising the more than 10 000 reports. It is important to note that only

textual media has been entered in the database. Radio and television is hence not

included. The case studies include 24 events that resulted in an international

response. 16 of those events have associated media reports. The Media frequency is a

measure of the number of reports issued in relation to a particular event. The Media

perseverance is defined as the time from the first to the last report issued by media on a

particular event. Figure 14.10 is developed using the perseverance level categories in

Table 14.5. In the cases for which less than two reports were issued, the time

difference is set to zero.

- 223 -

Table 14.5 Media perseverance categories

Time Perseverance level 0 None

=< 24 hours Little > 24 hours Intermediate

>7 days Substantial >31 days Long

Source: Author

From Figure 14.10 it is possible to draw the conclusion that the media has a

fairly set mind when determining for how long they provide coverage for an

earthquake. If it is a story, it receives coverage for up to a week and for some events

longer. If it is not a story, it will be mentioned only once or not at all. It is rare that an

event only features in media a couple of times during the first day after the disaster

impact.

Does media reaction time affect overall response time? Some studies (Benthall 1993:36; Coburn and Spence 2002:96-97) have mentioned

the role of the media as a source of early warning. Those studies have the relief

activities in the affected area in focus and the early warning is intended for the local

population and emergency organisations. On an international level, media could play

a role in alerting the international relief community. The current temporal resolution

of the case study data makes it difficult to analyse the relation between media reaction

and international response. Although each media report is provided with a release

time, it is often inaccurate. As can be seen in Figure 14.12 there is one case where the

response report supposedly was released before the event it claims to respond to took

place. That event started in the late evening and the first response was reported the

same day but without an exact time of response. When no time is reported the default

time used for the calculation is 00:01 of the reported day; which results in a negative

response time. The maximum potential error caused by the same type of fault in

other events is twelve hours.

- 224 -

0 10 20 30 40 50 60

Media delay (h)

Figure 14.12 Media reporting delay and response delay

Figure 14.12 shows no correlation between the delay in media reporting and the

delay in the international response reporting. An interesting pattern is that media is

faster to report events that the international community is slow to respond to and vice

versa.

What attracts media exposure? From Figure 14.11 it is clear that the events that receive greater media exposure

are those with a high human loss, large amounts of international financial aid and

high non-media reporting frequency. It is the media reporting frequency that shows

the strongest correlation with the media perseverance. Although not included in the

correlation matrix, the frequency of sitreps is highly correlated to non-media

reporting frequency. The media reporting frequency will therefore also be correlated

to the sitrep frequency. These relationships will be investigated further in Chapter 10.

- 225 -

A-4 INTEREST Database The screen-shots in this section are taken from the INTEREST software package

developed by the researcher as part of the project. The images show examples of the

built-in functionality for collecting, storing and analysing data collected on the

international response to a disaster.

Figure 14.13 Earthquake (seismic) report view

Figure 14.14 Main menu

- 226 -

Figure 14.15 Administration menu

Figure 14.16 Event population distribution view

- 227 -

Figure 14.17 Database event view

- 228 -

Figure 14.18 Data mining view

- 229 -

Figure 14.19 Database Entity-Relationship diagram

Modelling projections of international response to sudden-onset disasters - Development of a...

Documents

Transcript of Modelling projections of international response to sudden-onset disasters - Development of a...

University of Groningen International fund flows: surges, sudden ...

Vulnerability, disasters and culture

Forecasting characteristic earthquakes in a minimalist model

Earthquakes - Clay Community Schools

Climate change and natural disasters - EconStor

Facing Hazards and Disasters - Santa Rosa County

PIPELINES DURING EARTHQUAKES

SPATIAL ANALYSIS OF EARTHQUAKES IN IRAQ USING ...

Glob Warm Earthquakes landslides

Earthquakes and economic growth - EconStor

Can Youth Quotas Help Avoid Future Disasters?

Climate change and disasters

WEAVING HOPES AFTER DISASTERS - Caritas international

Causes of Earthquakes and their effects

Readiness in Times of Emergencies & Disasters (RTED)

Measuring Earthquakes from Optical Satellite Images

Disasters Implied by Equity Index Options

CATalytic insurance : the case of natural disasters

Population Movement and Sudden Cardiac Arrest Location

Chapter 1 Sudden Lowering of Stage Lights