Global Risk and Vulnerability Index Trends Per Yea

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/252370255
Global Risk And Vulnerability Index Trends per Year (GRAVITY)
Article
CITATIONS READS
27 239
4 authors, including:
Hy Dao Frederic Mouton

University of Geneva University Joseph Fourier - Grenoble 1
42 PUBLICATIONS 820 CITATIONS 16 PUBLICATIONS 603 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
blueDot project (Environmental footprints of nations: national performances and global priorities based on Planetary Boundaries) http://bluedot.world/ View project
GOFORBO View project
All content following this page was uploaded by Frederic Mouton on 27 July 2014.
The user has requested enhancement of the downloaded file.

Global Risk And Vulnerability Index
Trends per Year (GRAVITY)
Phase IV: Annex to WVR and Multi Risk Integration
A technical report for:
United Nations Development Programme

Bureau of Crisis Prevention & Recovery (UNDP/BCPR)
Hy Dao & Pascal Peduzzi

United Nations Environment Programme
Global Resource Information Database - Geneva
UNEP/DEWA/GRID-Geneva, May 2003
Contributors (by alphabetic order)
Main scientific collaborators:
Dr. Hy Dao, Christian Herold, Dr. Frédéric Mouton, Pascal Peduzzi.
Additional short term collaborators:

Stéphane Kluser, Antonio Martin-Diaz, Ola Nordbeck, Damien Rochette, Dr Thao Ton-
That, Bernard Widmer.
Other acknowledgements
We would like also to thanks the following persons whose work has directly or
indirectly contributed to the success of this research: Brad Lyon and his colleagues from IRI,
Columbia University for his methodology on determining physical drought. Debarati Guha-
Sapir and Regina Below from CRED for their kind and continuous support. Haris Sanahuja
consultant at UNDP/BCPR for his support on socio-economical data sets and verification of
the CRED drought/famine information. The GEO3 team from UNEP/GRID-Geneva for
prioritising their tasks according to our needs in socio-economical data sets and our warm
thanks to the World Meteorological Organisation (WMO) and its collaborators, Nanette
Lombarda and their contacts Greg Holland, Bruce Harper for their very useful advices in
tropical cyclones.
Disclaimer
The designations employed and the presentation of material on the maps do not imply
the expression of any opinion whatsoever on the part of UNDP/BCPR, UNEP or the
Secretariat of the United Nations concerning the legal status of any country, territory, city or
area or of its authorities, or concerning the delimitation of its frontiers or boundaries.
Table of content
CONTRIBUTORS (BY ALPHABETIC ORDER) .................................................................... 2
TABLE OF CONTENT ........................................................................................................ I
LIST OF TABLES ............................................................................................................. II
LIST OF EQUATIONS ...................................................................................................... II
LIST OF FIGURES............................................................................................................ II
TABLE OF ACRONYMS .................................................................................................. III
1 WORKING DEFINITIONS AND FORMULAE ..................................................... 1
1.1. HAZARDS, VULNERABILITY AND RISK, DEFINITIONS AND CONCEPTS .................... 1
1.2. FORMULA AND METHOD FOR ESTIMATING RISK AND VULNERABILITY.................. 2
2 CHOICE OF INDICATORS...................................................................................... 3
2.1. SPATIAL AND TEMPORAL SCALES........................................................................... 3
2.2. RISK INDICATORS ................................................................................................... 3
2.3. VULNERABILITY INDICATORS ................................................................................ 4
2.4. DATA SOURCES....................................................................................................... 6
3 COMPUTATION OF PHYSICAL EXPOSURE ..................................................... 7
3.1. GENERAL DESCRIPTION .......................................................................................... 7
3.2. THE CASE OF CYCLONES ......................................................................................... 9
3.3. THE CASE OF FLOODS ........................................................................................... 10
3.4. THE CASE OF EARTHQUAKES ................................................................................ 10
3.5. THE CASE OF DROUGHTS ...................................................................................... 12
4 STATISTICAL ANALYSIS : METHODS AND RESULTS................................. 13
4.1. DEFINING A MULTIPLICATIVE MODEL .................................................................. 13
4.2. DETAILED PROCESS .............................................................................................. 13
4.3. MAPPING RISK...................................................................................................... 15
4.4. CYCLONES ............................................................................................................ 15
4.5. FLOODS................................................................................................................. 17
4.6. EARTHQUAKES ..................................................................................................... 18
4.7. DROUGHT ............................................................................................................. 20
5 MULTIPLE RISK INTEGRATION ....................................................................... 22
5.1. METHODS ............................................................................................................. 22
5.2. RESULTS ............................................................................................................... 24
6 FINAL COMMENTS AND RECOMMENDATIONS .......................................... 26
6.1. GENERAL COMMENTS........................................................................................... 26
6.2. RECOMMENDATIONS ............................................................................................ 27
7 REFERENCES .......................................................................................................... 30
i
List of tables
TABLE 1. ADVANTAGES AND INCONVENIENCES OF RESPECTIVE RISK INDICATORS....................4
TABLE 2. VULNERABILITY INDICATORS ......................................................................................5
TABLE 3. DATA SOURCES FOR HAZARDS .....................................................................................6
TABLE 4. DATA SOURCES FOR VICTIMS, POPULATION AND VULNERABILITY FACTORS ..............7
TABLE 5. WIND SPEEDS AND APPELLATIONS ...............................................................................9
TABLE 6. DEFINITION OF RADIUS FOR EARTHQUAKES BASED ON BRACKETED DURATION
(EXPRESSED IN SECONDS). ..................................................................................................11
TABLE 7. DEFINITION OF DROUGHT ...........................................................................................13
TABLE 8. EXPONENT AND P-VALUE FOR CYCLONES MULTIPLE REGRESSION ............................16
TABLE 9: EXPONENT AND P-VALUE FOR FLOOD INDICATORS ...................................................17
TABLE 10: EXPONENT AND P-VALUE FOR EARTHQUAKE MULTIPLE REGRESSION .....................19
TABLE 11: EXPONENT AND P-VALUE FOR DROUGHT MULTIPLE REGRESSION ...........................21
TABLE 12. LIST OF THE 14 COUNTRIES WITH DRI COMPUTED BUT WITHOUT OBSERVED
CASUALTIES ........................................................................................................................24
TABLE 13. LIST OF THE 37 COUNTRIES WITH RECORDED CASUALTIES IN CRED BUT NOT
MODELLED ..........................................................................................................................25
TABLE 14. LIST OF THE TWO COUNTRIES ABSENT OF BOTH CRED AND MODEL ......................25
TABLE 15. RESULT FROM THE CLASSIFICATION ........................................................................25
List of Equations
EQUATION 1: SIMPLIFIED EQUATION OF RISK ..............................................................................2
EQUATION 2: SIMPLIFICATION OF RISK EVALUATION USING PHYSICAL EXPOSURE ....................2
EQUATION 3: ESTIMATION OF THE TOTAL RISK ...........................................................................3
EQUATION 4: COMPUTATION OF PHYSICAL EXPOSURE ...............................................................7
EQUATION 5: PHYSICAL EXPOSURE CALCULATION WITHOUT FREQUENCY.................................8
EQUATION 6 : COMPUTATION OF CURRENT PHYSICAL EXPOSURE ...............................................8
EQUATION 7: FROM PROBABILITY TO ANNUAL FREQUENCY FOR CYCLONES ..............................9
EQUATION 8. ESTIMATION OF KILLED .......................................................................................13
EQUATION 9. LOGARITHM PROPERTIES .....................................................................................13
EQUATION 10. TRANSFORMATION FOR VARIABLES RANGING BETWEEN 0 AND 1 ....................14
EQUATION 11 MULTIPLE LOGARITHMIC REGRESSION MODEL FOR CYCLONES .........................16
EQUATION 12. MULTIPLE LOGARITHMIC REGRESSION MODEL FOR FLOODS .............................17
EQUATION 13. MULTIPLE LOGARITHMIC REGRESSION MODEL FOR EARTHQUAKES .................19
EQUATION 14: MULTIPLE LOGARITHMIC REGRESSION MODEL FOR DROUGHT .........................20
EQUATION 15. COMPUTATION OF THE MULTIPLE RISK BY SUMMING THE CASUALTIES AS
MODELLED FOR RISK FOR CYCLONE, FLOOD, EARTHQUAKE & DROUGHT .........................24
List of Figures
FIGURE 1: EXAMPLE OF PHYSICAL EXPOSURE FOR TROPICAL CYCLONES................................10
FIGURE 2: POPULATION, FREQUENCY AND PHYSICAL EXPOSURE FOR FLOODS ........................10
FIGURE 3: POPULATION, INTENSITY AND PHYSICAL EXPOSURE FOR EARTHQUAKES ................12
FIGURE 4: SCATTER PLOT OF THE OBSERVED NUMBER OF PEOPLE KILLED BY WINDSTORMS
(CRED FIGURES) AND THE MODEL PREDICTIONS ..............................................................16
FIGURE 5. SCATTER PLOT OF THE OBSERVED NUMBER OF PEOPLE KILLED BY FLOODS (CRED
FIGURES) AND MODEL PREDICTIONS...................................................................................18
FIGURE 6: SCATTER PLOT OF THE OBSERVED NUMBER OF PEOPLE KILLED BY EARTHQUAKES
FIGURE 7: SCATTER PLOT OF THE OBSERVED NUMBER OF PEOPLE KILLED BY DROUGHTS
FIGURE 8 PROCESS FOR MULTIPLE RISK INTEGRATION .............................................................23
FIGURE 9 SCATTER PLOT OF EXPECTED CASUALTIES FROM MULTIPLE MODEL (WINDSTORM,
DROUGHT, EARTHQUAKE, FLOOD)......................................................................................26
ii
Table of acronyms
AGREMP Percentage of labour force in agricultural sector
BCPR Bureau for Crisis Prevention and Recovery
CRED Centre for Research on Epidemiology of Disasters
FAO Food and Agriculture Organisation
GDPAGR Percentage of agriculture’s dependency for GDP
GDPCAP Gross Domestic Product per capita
GEO Global Environment Outlook
GIS Geographical Information System
GLASOD Human Induced Soil Degradation
GRAVITY Global Risk And Vulnerability Index Trend per Year
HDI Human Development Index
HPI Human Poverty Index
IFRC International Federation of the Red Cross
IRI International Research Institute for Climate Prediction
ISDR International Strategy of Disaster Reduction
PhExp Physical Exposure (if not specified, for drought)
U5MORT Under five years old mortality rate
UNDP/BCPR United Nation Development Programme, Bureau for Crisis
Prevention and Recovery
UNEP/GRID United Nation Environment Programme, Global Resource
Information Database
WATRUR Percentage of population having access to improved water supply
in rural area
WATTOT Percentage of population having access to improved water supply
WATURB Percentage of population having access to improved water supply
in urban area
iii
Technical annex
Foreword
This technical annex describes the concepts, data and methods applied to achieve the
Disaster Risk Index (DRI).
Results of the different analysis are presented in chapter IV.
A set of recommendations is also provided at the end of the annex on the appropriate
use of the DRI as well as suggested future improvements to the methodology.
1 WORKING DEFINITIONS AND FORMULAE

1.1. Hazards, vulnerability and risk, definitions and concepts
The terminology used in this study is based on UN and other experts. The definitions of
the concepts are provided in the following paragraphs:
• Risk : “The term risk refers to the expected losses from a particular hazard to a
specified element at risk in a particular future time period. Loss may be estimated in
terms of human lives, or buildings destroyed or in financial terms" [UNDRO 1979; in
Burton et al. 1993, p.34].
Specificity in this research : The term “risk” is used to describe potential human losses
(casualties) resulting from expected future hazard.
• Hazard : “The hazard can be defined as a potential threat to humans and their welfare"
[Smith, 1996]. The hazardous events varies in terms of magnitude as well as in
“frequency, duration, area extent, speed of onset, spatial dispersion, and temporal
spacing” [Burton et al. 1993, p.34].
Specificity in this research : Only frequencies and area extent are considered in the
model. The magnitude is taken into account indirectly when possible.
• Natural Hazards : “Represents the potential interaction between humans and extreme
natural events. It represents the potential or likelihood of an event (it is not the event
itself)” [Tobin & Montz 1997].
Specificity in this research : Four types of natural threats are included in the model
(floods, earthquakes, cyclones and droughts).
• Physical Exposure : “Elements at risk, an inventory of those people or artefacts which
are exposed to the hazard” [Coburn et al. 1991, p. 49].
Specificity in this research : Computation of population exposed to a given hazard type.
In this research the element at risk is the population.
• Vulnerability : “Reflects the range of potentially damaging events and their statistical
variability at a particular location" [Smith, 1996]. “The degree of loss to each element
should a hazard of a given severity occur” [Coburn et al. 1991, p. 49].
Specificity in this research : The discrepancies of casualties induced by different
vulnerabilities are used to identify socio-economical indicators reflecting such
vulnerabilities
• Disasters : “A sudden calamitous event producing great material damage, loss and
distres”. [Webster’s Dictionary. Found in : Carter 1991]
1
Specificity in this research : A disaster occurs when the high vulnerability of an
exposed population intersects a hazardous event of a relatively strong magnitude.
By UN definition [UNDRO, 1979], the risk is resulting from three components:

"Hazard occurrence probability, defined as the probability of occurrence of a specified
natural hazard at a specified severity level in a specified future time period, elements at risk,
an inventory of those people or artefacts which are exposed to the hazard and vulnerability,
the degree of loss to each element should a hazard of a given severity occur" [Coburn et al.
1991, p. 49].
1.2. Formula and method for estimating risk and vulnerability

The formula used for modelling risk combines the three components of the UNDRO
definition [UNDRO 1979]: the risk is a function of hazard occurrence probability, element at
risk (population) and vulnerability. The following hypothesis was made for modelling the risk:
the three factors explaining risk are multiplying each other. This was introduced because,
if the hazard is null, then the risk is null:
0 (hazard) x population x vulnerability = 0 (Risk)
The risk is also null if nobody lives in an area exposed to hazard (population = 0), same
situation if the population is invulnerable, (vulnerability = 0, induce a risk = 0). Following the
definition from UNDRO and the stated hypothesis the Equation 1 was derived:
Equation 1: Simplified equation of risk1
R = H ⋅ Pop ⋅ Vul
Where:
R is the risk, i.e. the expected human impacts (number of killed people).
H is the hazard, which depends on the frequency and strength of a given hazard,
Pop is the population living in a given exposed area.
Vul is the vulnerability and depends on socio-politico-economical context of this population.
From the previous discussion the physical exposure is defined as the combination of
both frequency and population exposed (see p.1) to a given magnitude for a selected type of
hazard. The hazard multiplied by the population can then be replaced by the physical
exposure:
Equation 2: Simplification of risk evaluation using physical exposure
R = PhExp ⋅ Vul
Where:
PhExp is the physical exposure i.e. the frequency and severity multiplied by exposed population
One way of estimating the risk is to look at impacts from previous hazardous events.
The physical exposure can be obtained by modelling the area extent affected by one event.
The frequency is computed by counting the number of events for the given area divided by the
number of years of observation (in order to achieve an average frequency per year). Using the
area affected, the number of exposed population can be extracted using a Geographical
Information System (GIS), the population affected multiplied by the frequency provides the
physical exposure. The identification of parameters leading to higher vulnerability can then be
1
The model uses a logarithmic regression, the equation is similar but with exponent to each of the
parameters.
2
carried out by replacing the risk in the equation by casualties reported in EM-Dat from CRED
and running a statistical analysis for highlighting links between socio-economical parameters,
physical exposure and observed casualties. The magnitude of the events is taken into account
only by placing a threshold above which the event is included, except for earthquakes where
the magnitude of the event is taken into account in the computation of the physical exposure.
The magnitude is one field of new improvements needed, although subjected to some
limitations as discussed later (see p.26).
The numbers of casualties can be aggregated at country level. The expected losses due
to natural hazards are equal to the sum of all types of risk faced by a population in a given area
as provided by the Equation 3:
Equation 3: Estimation of the total risk
Risk Tot = ∑ ( Risk Flood + Risk Earthquake + RiskVolcano + Risk Cyclone + ...Risk n ) 2
Providing the total risk for a country induces the need to estimate the probability of
occurrence and severity of each hazard, the number of persons affected by them, the
identification of population vulnerability and mitigation capacities. This is of course not
possible in absolute, however the aim is to provide indicators which will be refined years after
years in order to approach the concept of risk.
2 CHOICE OF INDICATORS
2.1. Spatial and temporal scales
The risk analysis was performed on a country by country basis, i.e. the 249 countries
defined in the GEO reports [UNEP 2002].
All the variables cover in principle the 21 year period ranging from 1980 to 2000. The
starting date was set in 1980 because the access to information (especially on victims) was not
considered as sufficiently homogenous and comparable before that year. The variables
introduced in Equation 2, p.2 are aggregated figures (sum, averages) of the available data for
that period, with the following major exceptions:
• Earthquake frequencies are calculated over a 36 year period, due to the longer return
period of this type of disaster (1964 is the starting date for the first global coverage
on earthquakes measurement).
• Cyclones frequencies are based on annual probabilities provided by CDIAC
[Birdwell & Daniel 1991]
• HDI is available for the following years : 1980, 1985, 1990, 1995, 2000
• Population by grid cell (for physical exposure calculations) : 1990, 1995
• Corruption Perception Index (CPI) : 1995 to 2000
2.2. Risk indicators

The risk can be expressed in different ways (e.g. number of killed, percentage of killed,
percentage of killed as compared to the exposed population) with their respective advantages
and inconveniences (as depicted in Table 1).
2
In the case of countries marginally affected by a hazard type, the risk was replaced by zero if the model
could not be computed for this hazard.
3
Table 1. Advantages and inconveniences of respective risk indicators
Indicators for risk Advantages Inconveniences

Number of killed Each human being has the 10.000 persons killed split
same “weight” between ten small countries
does not appeared in the
same way as 10.000 killed in
one country. Smaller
countries are disadvantaged.
Killed / Population It allows for comparisons The “weight” of human being
between countries. Less is not equal, e.g. one person
populated countries have the killed in Honduras equal 160
same weight as more killed in China.
populated countries.
Killed / Population exposed Regional risk is highlighted This may highlight local
even though the population problem that are not of
affected is a smaller portion national significance and give
of the total national wrong priority for a selected
population. country.
No scientific arguments can be used for selecting one indicator instead of another. At
the end this is a political decision to select the indicators that best suit the purpose. The DRI is
finally based on a combination of two indicators, namely the number of killed and killed per
population (see p. 22). The third indicator is used as a proxy for observed vulnerability, but
only for a selected hazard as exposed population to different hazard cannot be compared
without standardisation.
2.3. Vulnerability indicators

The socio-economical parameters were chosen to reflect the level of quality of different
constituents of a civil society such as (Table 2):
4
Table 2. Vulnerability indicators
Categories of Indicators Drought Flood Source3

vulnerability Earthqu.
Cyclones
Economic Gross Domestic Product per inhabitant at X X WB
purchasing power parity
Human Poverty Index (HPI) X UNDP
Total dept service (% of the exports of X WB
goods and services),
Inflation, food prices (annual %), X WB
Unemployment, total (% of total labour X ILO
force)
Type of %age of arable land X FAO
economical %age of urban population X UNPOP
activities %age of agriculture’s dependency for GDP X WB
%age of labour force in agricultural sector X FAO
Dependency and Forests and woodland (in %age of land X FAO
quality of the area),
environment. %age of irrigated land X FAO
Human Induced Soil Degradation X UNEP
(GLASOD)
Demography Population growth, X UNPOP
Urban growth, X GRID4
Population density, X GRID5
Age dependency ratio, X WB
Health and Average calorie supply per capita, X FAO
sanitation %age of people with access to adequate X WHO /
sanitation, UNICEF
%age of people with access to safe water X X WHO /
(total, urban, rural) UNICEF
Number of physicians (per 1000 inh.), X WB
Number Hospital Beds X WB
Life Expectancy at birth for both Sexes X UNPOP
Under five years old mortality rate X UNPOP
Politic Transparency’s CPI (index of corruption) X TI
Early warning Number of Radios (per 1000 inh.) X WB
capacity
Education Illiteracy Rate, X WB
School enrolment, X UNESCO
Secondary (% gross), X UNESCO
Labour force with primary, secondary or X WB
tertiary education
Development Human Development Index (HDI) X X UNDP
3
FAOSTAT (Food and Agriculture Organisation, FAO) / GRID: UNEP/Global
Resource Information Database / WB: World Development Indicators (World Bank) / TI:
Transparency International / UNDP: Human Development Report (UNDP) / ILO:
International Labour Office / UNPOP: UN Dep. Of Economic and Social Affairs/Population
Division. Most of the data were reprocessed by the UNEP Global Environment Outlook team.
Figures are available at the GEO Data Portal (UNEP), http://geodata.grid.unep.ch
4
calculated from UNPOP data
5
calculated from UNEP/GRID spatial modelling based on CIESIN population data.
5
The list of factors to be considered for the analysis was set on the basis of the following
criteria:
• Relevance : select vulnerability factors (outputs orientated, resulting from the
observed status of the population), not based on mitigation factors (inputs, action
taken). Example : school enrolment rather than education budget.
• Data quality and availability : data should cover the 1980-2000 period and most of
the 249 countries. Examples of rejected variables for the previous explained
reasons: % of persons affected by AIDS, level of corruption, number of hospital
beds per inhabitants.
2.4. Data sources

Table 3. Data sources for hazards
Hazard type Data source

Earthquakes Council of the National Seismic System (as of 2002), Earthquake
Catalog, http://quake.geo.berkeley.edu/cnss/
Cyclones Carbon Dioxide InformationAnalysis Center (1991), A Global
Geographic Information System Data Base of Storm Occurrences and
Other Climatic Phenomena Affecting Coastal Zones,
http://cdiac.esd.ornl.gov/
Floods U.S. Geological Survey (1997), HYDRO1k Elevation Derivative
Database, http://edcdaac.usgs.gov/gtopo30/hydro/
Droughts IRI/Columbia university, National Centers for Environmental Prediction

(physical drought) Climate Prediction Center (as of 2002), CPC Merged Analysis of
Precipitation (CMAP) monthly gridded precipitation,
http://iridl.ldeo.columbia.edu/
6
Table 4. Data sources for victims, population and vulnerability factors
Theme Data source

Victims (killed) Université Catholique de Louvain (as of 2002), EM-DAT: The
OFDA/CRED International Disaster Database, http://www.cred.be/ (for
droughts, victims of famines were also included on a case by case basis by
UNDP/BCPR)
Population CIESIN, IFPRI, WRI (2000), Gridded Population of the World (GPW),
(counts) Version 2, http://sedac.ciesin.org/plue/gpw/
UNEP,CGIAR,NCGIA (1996), Human Population and Administrative
Boundaries Database for Asia,
http://www.grid.unep.ch/data/grid/human.php
Vulnerability factors
Human UNDP (2002), Human Development Indicators, http://www.undp.org/
Development
Index (HDI)
Corruption Transparency international (2001), Global Corruption Report 2001,
Perceptions Index http://www.transparency.org/
(CPI)
Soil degradation ISRIC, UNEP (1990), Global Assessment of Human Induced Soil
(% of area Degradation (GLASOD), http://www.grid.unep.ch/data/grid/gnv18.php
affected)
Other socio- UNEP/GRID (as of 2002), GEO-3 Data portal,
economic http://geodata.grid.unep.ch/ (data compiled from World Bank, World
variables Resources Institute, FAO databases)
3 COMPUTATION OF PHYSICAL EXPOSURE

3.1. General description
In broad term, the physical exposure was estimated by multiplying the hazard frequency
by the population living in the exposed area. The frequency of hazard was derived for
different strengths of events and the physical exposure was computed as in Equation 4:
Equation 4: Computation of physical exposure
PhExp nat = ∑ Fi ⋅ Popi

Where:
PhExpnat is the physical exposure at national level
Fi is the annual frequency of a specific magnitude event in one spatial unit
Popi is the total population living in the spatial unit
For the case of earthquakes, the computation of a frequency could not be derived as the
available information consisted on a 90% probability of an earthquake to be smaller than a
given magnitude. To overcome this difficulty, the physical exposure computation was made
by adding the population affected and then divided by the number of year as shown in
Equation 5.
7
Equation 5: Physical exposure calculation without frequency
Popi
PhExp = ∑
Yn
Where:
Popi is the total population living in a particular buffer which radius from the epicentre varies
according to the magnitude.
Yn is the length of time in year for the cyclones (11)
PhExp is the total physical exposure of a country is the sum of all physical exposure of this country
Once the hazarduous event area was computed using UNEP/GRID-Geneva methods for
earthquakes, floods and cyclones or using IRI’s method for drought, then the affected
population was computed for each affected area and then this number was aggregated at
national level as needed in order to associate the victims from the last 21 years with the
physical exposure and with socio-economical variables.
Depending on type of hazards and the quality of data, different methods were applied.
Extraction of population was based on the CIESIN, IFPRI, WRI Gridded Population of the
World (GPW, Version 2) at a resolution of 2.5’ 6 (equivalent to 5 x 5 km at the equator). This
layer was further completed by Human Population and Administrative Boundaries Database
for Asia (UNEP) for Taiwan and CIESIN Global Population of the world version 2 (country
level data) for ex-Yugoslavia. These datasets reflect the estimated population distribution for
1995. Since population growth is sometimes very high in the 1980-2000 period, a correction
factor using country totals was applied in order to estimate current physical exposures for each
year as follows :
Equation 6 : computation of current physical exposure
Popi
PhExpi = ⋅ PhExp1995
Pop1995
Where:
PhExpi is the physical exposure of the current year
Popi is the population of the country at the current year
Pop1995 is the population of the country in 1995
PhExp1995 is the physical exposure computed with population as in 1995
Due to the resolution of the data set, the population could not be extracted for some
small islands. This has lead to the non-consideration of the small islands (even for large
archipelagos). Refined study should be carried out in a further research (see recommendations
in conclusion). Apart from these limitations, the extraction of the population living in exposed
area is a simple task performed with a GIS.
The main difficulty consists in the evaluation of hazard area extent, frequency and
intensity. At a global scale, data are not complete and generalisation is the rule. Help of
specialists was asked in order to review the necessary simplifications. Out of the four hazards
studied, only the case of floods requested the complete design of a global dataset built by
6
GPW2 was preferred to the ONRL Landscan population dataset despite its 5 times lower spatial resolution
(2.5’ against 30”) because the original information on administrative boundaries and population counts is almost
two times more precise (127,093 administrative units against 69,350 units). Furthermore the Landscan dataset is the
result of a complex model which is not explained thoroughly and which is based, among other variables, on
environmental data (land-cover), making it difficult to use for further comparison with environmental factors
(circularity).
8
linking CRED information with USGS watersheds. Drought maps were provided by the
International Research Institute for Climate Prediction (IRI). For the other hazards,
independent global datasets had already been updated, compiled or modelled by UNEP/GRID-
Geneva and were used to extract the population. The Mollweide equal-area projection was
used when calculations of areas were needed.
Although the quality can always be improved, the greatest care was taken and the level
of accuracy achieved is believed to be relevant and appropriate for a global scale study.
3.2. The case of cyclones

The absence of complete coverage (i.e. India, Bangladesh and Pakistan data) prevents
the use of the UNEP/GRID-Geneva PREVIEW Global Cyclones Asymmetric Windspeed
Profile dateset. The data used to map cyclone hazard areas are produced by the Carbon
Dioxide Information Analysis Centre [Birdwell & Daniel 1991]. These delineate annual
probabilities of occurrence of tropical cyclones. The spatial unit is a 5 x 5 decimal degrees
cell. Probabilities are based on tropical cyclones activity of a specific record period, except for
several estimated values attributed to areas that may present occasional activity but where no
tropical cyclones were observed during the record period.
Table 5. Wind speeds and appellations
Wind speeds Name of the phenomenon
≥ 17 m/ s Tropical storms
≥ 33 m/ s Hurricanes, typhoons, tropical cyclones, severe
cyclonic storms (depending on location7)
≥ 65 m/ s Super typhoons
Saffir-Simpson tropical cyclones classification is based on the "maximum sustained

surface wind". With winds of less than 17 m/s, they are called "tropical depressions". If the
wind reaches speeds of at least 17 m/s, they are called "tropical storms". If the wind speed is
equal to or greater than 33 m/s, they get one of the following names, depending on their
location7: "hurricanes", "typhoon", "severe tropical cyclone", "severe cyclonic storm" or
"tropical cyclone". At last, if the wind reaches speeds of 65 m/s or more, they are called "super
typhoons" [Landsea 2000].
The CDIAC is providing probability of occurrence for these three types of events. The
average frequency (per year) was computed using Equation 7:
Equation 7: From probability to annual frequency for cyclones
E ( x) = λ = − ln(1 − P( x ≥ 1))
Where:
E(x) is the statistical expectation, i.e. the average number of events per year = λ
P(x) is the probability of occurrence
Hurricanes: North Atlantic Ocean, Northeast Pacific Ocean east of the dateline, or the South Pacific
Ocean east of 160E);
Typhoon : Northwest Pacific Ocean west of the dateline,
Severe tropical cyclone: Southwest Pacific Ocean west of 160E and Southeast Indian Ocean east of 90E,
Severe cyclonic storm: North Indian Ocean,
Tropical cyclone: Southwest Indian Ocean
Source: NOAA/AOML, FAQ: Hurricanes, Typhoons, and Tropical Cyclones,
http://www.aoml.noaa.gov/hrd/tcfaq/tcfaqA.html#A1
9
Figure 1: Example of physical Exposure for Tropical Cyclones
To obtain the physical exposure, a frequency per year is derived for each cell. Cells are
divided to follow country borders, then population is extracted and multiplied by the
frequency in order to obtain the average yearly physical exposure for each cell. This physical
exposure is then summed by country for the three types of cyclones.
Physical exposure to tropical cyclones of each magnitude was calculated for each
country using Equation 4 (p.7).
3.3. The case of floods

For the floods the method is slightly different. No global database on floods could be
found except the one from Dartmouth Flood Observatory, but which was not covering the
period of interest. Due to the lack of information on the duration and severity of floods, only
one class of intensity could be made. Using the information in the column “comment” (e.g.
name of town, river, valley,…) in the CRED database, a georeference of the floods was
produced and a link between the watersheds and the events was made. Watersheds affected
were mapped for the period 1980-2000. A frequency was derived for each watershed by
dividing the total number of events by 21 years. The watersheds were then splitted to follow
country borders, then population was extracted and multiplied by the frequency. The average
yearly physical exposure was then summed at a country level using Equation 4 (p.7).
Figure 2: population, Frequency and Physical exposure for floods
Population Frequency for each watershed National physical exposure per
year
3.4. The case of earthquakes

A choice was made to produce seismic hazard zones using the seismic catalogue of the
CNSS (Council of the National Seismic System). The earthquakes records of the last 21 years
(1980-2000) were grouped in five magnitude classes and a buffer with a radius length from the
10
epicentres varying according to these classes. Choices of specific radius were made
considering the table Table 6.
Table 6. Definition of radius for earthquakes based on bracketed duration
(expressed in seconds).
Distance Magnitude
(km) 5.5 6.0 6.5 7.0 7.5 8.0 8.5
10 8 12 19 26 31 34 35
25 4 9 15 24 28 30 32
50 2 3 10 22 26 28 29
75 1 1 5 10 14 16 17
100 0 0 1 4 5 6 7
125 0 0 1 2 2 3 3
150 0 0 0 1 2 2 3
175 0 0 0 0 1 2 2
200 0 0 0 0 0 1 2
table source [Bolt et al. 1975] Acceleration > 0.05 g = ~ 0,49 m/s2, frequency > 2 Hz
These above numbers delineate the estimations of ground motions duration for specific
acceleration and frequency ranges, according to magnitude and distance from epicentre [Bolt
et al. 1975]. Bracketed duration is “the elapsed time (for a particular acceleration and
frequency range) between the first and last acceleration excursions on the record greater than
a given amplitude level (for example, 0.05 g)” [Bolt et al. 1975].
According to these figures, a specific buffer distance is defined for each class of
magnitude to limit area affected by ground motions: 75 km for Magnitude ≤ 6.2, 125 km for
M = 6.3 – 6.7, 150 km for M = 6.8 – 7.2, 175 km for M = 7.3 – 7.7, 200 km for M ≥ 7.8. This
is a general approach that does not take into account any regional effects, for instance soil
conditions or geotectonic characteristics.
Physical exposure to earthquakes was calculated for each country and each magnitude
class using Equation 5 (p. 8), for details on how this could be improved, see recommendations
(p. 27).
11
Figure 3: Population, intensity and physical exposure for earthquakes
Population Intensity National physical exposure per year
3.5. The case of droughts

Identification of drought
Explanations from the author : Brad Lyon, International Research Institute for
Climate Prediction (IRI)
The data used in the analysis consists of gridded monthly precipitation for the globe for
the period 1979-2001. This dataset is based on a blend of surface station observations and
precipitation estimates based on satellite observations. Further details of the data are described
at the end of this section.
The first step in assessing the exposure to meteorological drought was to compute, for
each calendar month, the median precipitation for all grid points between the latitudes of 60S
and 70N over the base period 1979-2001 (the 23-yr. period for which the data was available).
Next, for each gridpoint, the percent of the long-term median precipitation was computed for
every month over the period Jan 1980 to Dec 2000. For a given month, gridpoints with a
long-term median precipitation of less than 0.25 mm/day were excluded from the analysis.
Such low median precipitation amounts can occur either during the "dry season" at a given
location or in desert regions and in both cases our definition of drought does not apply.
Finally, a drought "event" was defined as having occurred when the percent of median
precipitation was at or below a given threshold for at least 3 consecutive months. The different
thresholds considered were 50%, 75% and 90% of the long-term median precipitation with the
lowest percentage indicative of the most severe drought according to this method. The total
number of events over the period 1980-2000 were thus determined for each gridpoint and the
results plotted on global maps.
Computation of physical exposure

Using the IRI/Columbia University data set, the physical exposure was estimated by
multiplying the frequency of hazard by the population living in the exposed area. The events
were identified using different measurements based on severity and duration as described in
Table 7. For each of the 6 following definitions, the frequency was then obtained by dividing
the number of events by 21 years, thus providing an average frequency of events per year.
12
Table 7. Definition of drought
Duration Severity
3 months 90% of median precipitation 1979-2001 (-10%)
The physical exposure was computed as in Equation 4 (p.7) for each of the drought
definitions. The statistical analysis selected the best fit which was achieved with droughts of
three month duration and 50% decrease in precipitation.
4 STATISTICAL ANALYSIS : METHODS AND RESULTS

4.1. Defining a multiplicative model
The statistical analysis is based on two major hypotheses. Firstly, that the risk can be
approached by the number of victims of past hazardous events. Secondly, that the equation of
risk follows a multiplicative model as in Equation 8:
Equation 8. Estimation of killed
α
K = C ⋅ ( PhExp)α ⋅ V1α1 ⋅ V2α 2 ... ⋅ V p p
Where:
K is the number of persons killed by a certain type of hazard.
C is the multiplicative constant.
PhExp is the physical Exposure: population living in exposed areas multiplied by the frequency of
occurrence of the hazard.
Vi are the socio-economical parameters.
αi are the exponent of Vi , which can be negative (for ratio)
Using the logarithmic properties, the equation could be written as follows:

Equation 9. Logarithm properties
ln( K ) = ln(C ) + α ln( PhExp) + α 1 ln(V1 ) + α 2 ln(V2 ) + ... + α p ln(V p )
This equation provides a linear relation between logarithmic sets of values. Significant
socio-economical parameters Vi (with transformations when appropriate) and exponents αi
could be determined using linear regressions.
4.2. Detailed process

Data on victims
The number of killed was derived from the CRED database, and computed as the
average number of killed per year over the 1980-2000 period.
13
Filtering the data
The statistical models for each disaster type were based on subsets of countries, from
which were excluded:
• Countries with no physical exposure or no victims reported (zero or null values)
• Countries with dubious data on physical exposure (e.g. the case of Kazakhstan
for floods) or socio-economic factors (100% access to water in North Korea)
• Countries with low physical exposure (smaller than 2 percent of the total
population) because socio-economical variables are collected at national scale.
Attempts delineate that the exposed population needs to be of some significance
at national level to reflect a relationship in the model.
• Countries without all the selected socio-economic variables.
• Eccentric values, when exceptional events or other factors would clearly show
abnormal level of victims like hurricane Mitch in Nicaragua and Honduras or
droughts in Sudan and Mozambique (i.e. probably more related to political
situation than from physical drought).
Transformation of variables
The average of socio-economical parameters was computed for the 21 year period. For
some of the indicators the logarithm was computed directly, for other parameters expressed in
percentage, a transformation was applied in order that all variables would range between -∞
and +∞. This appeared to be relevant as some of the transformed variables were proved to be
significant in the final result. For others no logarithmic transformation was needed, for
instance the population growth already behaves in a cumulative way.
Equation 10. Transformation for variables ranging between 0 and 1
Vi
Vi ' =
(1 − Vi )
Where:
V’i is the transformed variable (ranging from -∞ to +∞)
Vi is the socio-economical variable (ranging from 0 to 1)
1000
transformed value (V')
100
10
0.1
0.01
0.001
0 0.2 0.4 0.6 0.8 1
original value (V)
Choice between variables

One important condition, when computing regressions, is that the variables included in a
model should be independent, i.e. the correlation between two sets of variables is low. This is
14
clearly not the case of HDI and GDPcap purchasing power parity (further referred as
GDPcap), which are highly correlated. GDPcap was more used than HDI uniquely because
HDI was not available for several countries. In order to keep the sample as complete as
possible a choice of available variables had to be made. This choice has been performed by the
use of both matrix-plot and correlation-matrix (using low correlation, hence low p-value, as
selection criteria).
The stepwise approach

For each type of hazard, numerous stepwise (back and forth steps) linear regressions
were performed in order to highlight significant variables. The validation of regression was
carried out using R2, variance analysis and detailed residual analysis.
Once the model was derived, the link between estimated killed and number of killed
observed was provided by both graphical plots and computation of Pearson correlation
coefficients in order to ease the visualisation of the efficiency by the readers.
If intuitively one can understand that physical exposure is positively related with the
number of victims and that GDPcap is inversely related with the number of victims (the lowest
the GDP the highest the victims), this is less obvious for other variables such as percentage of
arable land for example. This method allows the estimation of the αi coefficients. Their signs
provide information on if the variables are in a numerator or denominator position.
This model allows the identification of parameters leading to higher/lower risk, but
should not be used as a predictive model, because small differences in logarithm scale induce
large ones in the number of killed.
The results following this method are surprisingly high and relevant, especially
considering the independence of the data sources (no auto-correlation suspected), the non
consideration of the magnitude of hazard and the coarse resolution of the data at global scale.
4.3. Mapping Risk

A subjective – political – choice belonging to UNDP had to be made between the
different risk indicators (i.e. killed, killed per million inhabitant,… , see p. 3). The UNDP aims
to provide categories of countries taking into account both risk and disaster reduction
measures.
Number of categories and method of classification

The number of classes was chosen taking into account by UNDP requirements, but also
taking into consideration the levels of error and uncertainty of the data. So far, the precision
and quality of the data, as well as the sensitivity of the model do not allow the ranking of
countries. However, for the risk component, results indicate a possibility to provide five
classes of countries. This number of categories minimise the error of misclassification and is
simple enough to be incentive to persons not too familiar with statistics.
The number of classes and method for classifying the maps were chosen according to
several criteria such as the optimum number of classes for visual representation or the number
and level of errors between modelled and observed classes. According to these tests, with the
aim to minimise internal class distances and maximise distances between classes, the number
of five classes based on a Cluster analysis for grouping countries using both killed and killed
per millions inhabitant was chosen.
4.4. Cyclones
Statistical model
The multiple regression was based on 32 countries and the best fit regression line
follows the following model:
15
Equation 11 Multiple logarithmic regression model for cyclones
____ ____
ln( K ) = 0.63 ln( PhExp) + 0.66 ln( Pal ) − 2.03 ln( HDI ) − 15.86
Where:
K is the number of killed from cyclones
PhExp is the physical exposure to cyclones
____
Pal is the transformed value of percentage of arable land
___
HDI is the transformed value of the Human Development Index
Table 8. Exponent and p-value for cyclones multiple regression
32 countries B p-value8
Intercept -15.86 0.00000
ln(PhExp) 0.63 0.00000
___
ln( Pal ) 0.66 0.00013
___
ln( HDI ) -2.03 0.00095
R= 0.93, R²= 0.86, adjusted R²= 0.85
The plot delineates a nice linear distribution of the data as seen in Figure 4:
Figure 4: Scatter plot of the observed number of people killed by windstorms
(CRED figures) and the model predictions
8
In broad terms, a p-value smaller than 0.05, shows the significance of the selected indicator, however this
should not be used blindly.
16
The parameters highlighted show that besides the physical exposure, HDI and the percentage of arable land are
selected indicators for vulnerability to cyclone hazards.
The percentage of arable land is probably an indirect way of measuring the dependency of a population from the
agricultural activity. According to the analysis, a stronger dependence to agriculture is inducing a higher
vulnerability. Although this was already mentioned by experts, it is now confirmed by statistical evidences. After a
cyclone, economies relying on third sector are less affected than economy relying on agriculture, fields being
devastated. The GDPcap is strongly correlated with the HDI or negatively with the percentage of urban growth. In
most of the cases the variable GDPcap could be replaced by HDI as explained previously (see p.13). However,
these results depict with confidence that poor countries and less developed in terms of HDI are more vulnerable to
cyclones.
With a considerable part of variance explained by the regression (R2 = 0.863) and a high degree of confidence in
the selected variables (very small p-value) over a sample of 32 countries, the model achieved is solid.
In the model, the consequences of Mitch could easily be depicted. Indeed, Honduras and Nicaragua were far off the
regression line (significantly underestimated). This is explained by the incredible difference of intensity of Mitch
and other hurricanes. Mitch is a type of hazards on its own, the difference of intensity made this event impossible to
compare with the other hurricanes. This is explaining the rejection of these two countries from the model.
4.5. Floods
Statistical model
Equation 12. Multiple logarithmic regression model for floods
ln( K ) = 0.78 ln( PhExp) − 0.45 ln(GDPcap ) − 0.15 ln( D) − 5.22

Where:
K is the number of killed from floods
PhExp is the physical exposure to floods
GDPcap is the normalised Gross Domestic Product per capita (purchasing power parity)
D is the local population density (i.e. the population affected divided by the area affected)
Table 9: Exponent and p-value for flood indicators

90 countries B p-value (8)
Intercept -5.22 0.00000
ln(PhExp) 0.78 0.00000
ln(GDPcap) -0.45 0.00002
ln(Density) -0.15 0.00321
R= 0.84, R²= 0.70, adjusted R²= 0.69
17
Figure 5. Scatter plot of the observed number of people killed by floods (CRED
figures) and model predictions
Due to space constraints, only a selection of countries was included in the above scatter
plot, a comprehensive list of countries affected by floods is provided below:
Albania, Algeria, Angola, Argentina, Australia, Austria, Azerbaijan, Bangladesh, Benin, Bhutan, Bolivia,
Botswana, Brazil, Burkina Faso, Burundi, Cambodia, Cameroon, Canada, Chad, Chile, China, Colombia, Costa
Rica, Cote d'Ivoire, Czech Republic, Dominican Republic, Ecuador, Egypt, El Salvador, Ethiopia, Fiji, France,
Gambia, Georgia, Germany, Ghana, Greece, Guatemala, Haiti, Honduras, India, Indonesia, Iran (Islamic Republic
of), Israel, Italy, Jamaica, Japan, Jordan, Kenya, Lao People's Democratic Republic, Malawi, Malaysia, Mali,
Mexico, Moldova, Republic of Morocco, Mozambique, Nepal, Nicaragua, Niger, Nigeria, Pakistan, Panama, Papua
New Guinea, Paraguay, Peru, Philippines, Poland, Portugal, Republic of Korea, Romania, Russian Federation,
Rwanda, Saudi Arabia, Sierra Leone, Slovakia, South Africa, Spain, Sri Lanka, Thailand, Tunisia, Turkey, Uganda,
Ukraine, United Kingdom of Great Britain and Northern Ireland, United Republic of Tanzania, United States of
America, Viet Nam, Yemen, Zimbabwe,
The variables selected by the statistical analysis are physical exposure, GDPcap and local density of population.
GDPcap being highly correlated with HDI, this later could have been chosen as well. The GDPcap was chosen due
to slightly better correlation between the model and the observed killed, as well as because of lower p-value.
Regression analysis supposes the introduction of non-correlated parameters, thus preventing the use of all these
variables.
Without surprise, the regression proves that highly exposed and poorer populations are more subject to suffer
casualties from floods. More surprisingly, it shows that countries with low population density are more vulnerable
than countries with high population density. The part of explained variance (R2 = 0.70) associated with significant
p-value (between 10-23 and 2·10-3) on 90 countries is confirming a solid confidence in the selection of the variables
(see Table 9).
4.6. Earthquakes
Statistical model
18
Equation 13. Multiple logarithmic regression model for earthquakes
ln( K ) = 1.26 ln( PhExp) + 12,27 ⋅ U g − 16.22

Where:
K is the number of killed from earthquakes
PhExp is the physical exposure to earthquakes
Ug is the rate of urban growth (rates do not request transformation as it is already a cumulative value)
Table 10: Exponent and p-value for earthquake multiple regression
48 countries B p- value (8)
Intercept -16.22 0.000000
PhExp 1.26 0.000000
Ug 12.27 0.047686
R= 0.75, R²= 0.56, adjusted R²= 0.54
19
Figure 6: Scatter plot of the observed number of people killed by earthquakes
(CRED figures) and the model predictions
The variables retained by the regression include the physical exposure and the rate of urban growth. The part of
explained variance is smaller than for flood or cyclones (R2=0.544), however considering the small length of time
taken into account (21 years as compared to earthquakes long return period), the analysis delineates a reasonably
good relation. The physical exposure is of similar relevance than for previous cases, relevant p-value. The urban
growth was expected to be selected as indicators. A high rate of population moving into a city is usually synonym
of low quality urban planning and building standard. The urban growth is also highly negatively correlated with
GDP and HDI. Thus, similar correlation (but slightly inferior) could have been derived using HDI or GDP.
4.7. Drought
Statistical model
The regression analysis was performed using the 6 different exposure datasets derived
from IRI drought maps (see p. 12). In general, the models based on 3 month thresholds give
better results. The dataset based on a drought threshold set at 3 months at 50% below the
median precipitation 1979-2001 was finally selected as the exposure data.
follows the following Equation 14:
Equation 14: Multiple logarithmic regression model for drought
ln( K ) = 1.26 ln( PhExp3 _ 50) − 7.58 ln(WATTOT ) + 14.4

Where:
K is the number of killed from droughts
PhExp3_50 is the number of people exposed per year to droughts ; a drought is defined as a period of
at least three months less or equal to 50% of the average precipitation level (IRI,
CIESIN/IFPRI/WRI)
WATTOT is the percentage of population with access to improved water supply (WHO/UNICEF)
20
Table 11: Exponent and p-value for drought multiple regression
Predictor Coef SE Coef T p-value9
Constant 14,390 3,411 4,22 0.001
Phexp3_5 1,2622 0,2268 5,57 0,000
(ln)
WATTOT -7,578 1,077 -7,03 0,000
S = 1,345 R-Sq = 0.812 R-Sq(adj) = 0.78
Figure 7: Scatter plot of the observed number of people killed by droughts (CRED
figures) and the model predictions
Rejected countries : Swaziland and Somalia (WATTOT value inexistent), North Korea
(reported WATTOT of 100% is highly doubtful), Sudan and Mozambique (eccentric values,
suggesting other explanation for casualties)
9
In broad terms, a p-value smaller than 0.05, shows the significance of the selected indicator, however this
should not be used blindly.
21
The small p-values observed are suggesting a relevant selection of the indicators among the list of available
datasets. It is to be noted that the high coefficient for WATTOT (-7.578) denotes a strong sensibility to the quality
of the data. This implies that even a change of 1% in the percentage of total access to water will induce significant
change in the results, especially for small values (where small changes have bigger influence in proportion).
If the model allows the selection of socio-economic parameters indicating the vulnerability of the population, this
model cannot be used for predictive purpose. Some inconsistencies were depicted in the data that require
verification.
The two indicators selected through the statistical analysis are not surprising. Physical exposure summarise the
frequency of hazard and the element at risk (here the population) while the percentage of population with access to
improved water supply is an obvious indicator of vulnerability to drought. This could obviously be derived through
common sense (population with good water supply less suffer from drought is not surprising). However, the fact
that it was selected and that a strong correlation could be established (R2 = 0.81) between independent datasets such
as level of precipitation, population, casualties from drought and access to water, assess the solidity of the method
as well as the reliability of these datasets for such scale.
Given the level of precision of the data and the compulsory simplification of the drought model, such match is
much higher than originally expected.
The Figure 7 shows the distribution (on a logarithmic scale) of expected casualties from drought and as predicted
from the model. A clear regression can be drawn. It is true that if Ethiopia is removed the correlation will fall to a
mere (R2 = 0.6), however the off set and the slope of the regression line do not change significantly thus assessing
the robustness of the model.
As far as 1.26 is close to 1, the number of killed people grows proportionally to physical exposure. Also the
number of killed people is decreasing as the percentage of population when improved water supply is growing.
This latter variable should be seen as an indicator of the level of development of the country as it was correlated to
other development variables, such as under five mortality rate (U5MORT, Pearson correlation r = -0.64) and Human
Development Index (HDI, r = 0.65).
There were some concerns about some features reported in CRED. Some countries with large physical exposure did
not reported any killed (United States of America, Viet Nam, Nigeria, Mexico, Bangladesh, Iran, Iraq, Colombia,
Thailand, Sri Lanka, Jordan, Ecuador, etc.). This could be for different reason: either the vulnerability is null (or
extremely low) e.g. USA, Australia or the number of reported killed from food insecurity is placed under conflict
(e.g. Iraq, Angola,…) for other countries further inquiries might be necessary.
5 MULTIPLE RISK INTEGRATION

5.1. Methods
How to compare countries and disasters
One of the main difficulties of this research might consist in the comparison between
disaster types. Indeed, how to compare an earthquake with a drought? Rapid on-set, not
predictable, with sudden and significant effect for the earthquake, slow on-set, fairly
predictable, with fuzzy boundaries difficult to determine in both spatial and temporal way for
a drought inducing long term effect. These considerations prevent the use of the proxy for
multiple hazard vulnerability as the physical exposure cannot be compared.
The idea was then to compare what is comparable: the casualties. A model for multiple
risks integration was made by adding expected casualties. In order to reduce the number of
countries with no data, preventing them to be modelled, the value “no data” for countries
without significant exposure was replaced by zero risk of casualties. Thus reducing the
number of country removed from the multiple risk model, because some data are missing for
modelling a type of hazard that do not affect these countries. By performing these boolean
conditions it was possible to incorporate 210 countries (out of 249) into the DRI.
A country was considered as not or marginally affected if the two following conditions
where gathered: a physical exposure smaller than 2% of the national population AND an
affected population smaller than 1000 per year.
The remaining 39 countries are cases with exposure and for a large majority with
recorded casualties (for 37 of them). This list of countries with casualties is therefore
important to complement the model. It identifies countries where improvement in data
22
collection is needed. Seven different cases were identified, differentiating countries marginally
affected by a specific hazard, countries affected but without data, countries with situation that
cannot be explained by the model (e.g. Mozambique, Sudan for drought, where conflict is
playing a more significant role than physical drought on food security).
Once the different cases were identified, it was possible to run a boolean process in
order to allocate the relevant values depending on the cases. The Figure 8 illustrates the
different steps for incorporating the values into a multiple risk index. Once the values for the
countries were computed, three different products were available:
- A table of values for the countries that include the data for relevant hazards or
countries without data but marginally affected (210 countries)
- A list of countries with missing data (countries with reported casualties but without
appropriate data).
- A list of countries where the model could not be applied (indicators do not capture
the situation in these countries, case of countries not explained by the model,
rejected during the analysis because the indicators are not relevant to the situation).
Figure 8 Process for multiple risk integration
The total risk computation

The multiple risk, in the case where the condition allows its computation, is computed
using the succession of formulae as described in Equation 15:
23
Equation 15. Computation of the multiple Risk by summing the casualties as
modelled for risk for cyclone, flood, earthquake & drought
____ 0.66 ___ −2.03

0.63
K cyclone ( PhExpCyclones ⋅ Pal ⋅ HDI ⋅ e −15.86 ) +
− 0.45
K floods ( PhExp 0floods
.78
⋅ GDPcap ⋅ D −0.15 ⋅ e −5.22 ) +
K earthquakes ( PhExp1earthquake
.26
s ⋅U g
12.27
⋅ e −16.27 ) +
− 7.58
K drought ( PhExp3 _ 501.26 ⋅ WATTOT ⋅ e14.4 )
Where:
e is the Euler constant (=2.718…)
PhExp is the physical exposure of selected hazard
HDI is the Human Development Index
GDPcap is the Gross Domestic Product per capita at purchasing power parity
D is the local density (density of population in the flooded area)
Ug is the Urban growth (computed over 3 years period)
Wattot is the access to safe drinking water.
Between each addition, the whole process described in Figure 8 needs to be run in order
to identify the cases where the value is replaced by zero, calculated from the selected hazard
model or placed in the “not-relevant” or “no data” lists see Table 13 and Table 14.
To classify the countries into the five categories, a cluster analysis minimising the intra-
class distance and maximising the inter-classes (K-means clustering method) was performed.
The choice to define 5 classes is discussed in page 15.
In order to take both risk indicators (killed and killed per inhabitant) into account a
Principal Component Analysis (PCA) was performed to combine the two informations into
one. Then a distinction was made between countries smaller than 30’000 km2 and with
population density higher than 100 inhabitants per km2 following UNDP/BCPR requirements.
5.2. Results
List of countries with observed casualties and risk modelled
Modelled countries without reported casualties
The Disaster Risk Index (DRI) was computed for 210 countries, this includes 14
countries where no reported casualties were reported in the last two decades (from CRED) as
seen in list in Table 12.
Table 12. List of the 14 countries with DRI computed but without observed
casualties
Barbados, Croatia, Eritrea, Gabon, Guyana, Iceland, Luxembourg, Namibia, Slovenia,
Sweden, Syrian Arab Republic, The former Yugoslav Republic of Macedonia, Turkmenistan,
Zambia.
No data, abnormal values and specific cases

Following the Principal Component Analysis transformation, inferior and superior
thresholds could be identified. This was performed on both observed and modelled casualties.
In the cases of 14 countries the model was produced even though no recorded casualties were
recorded by CRED in the last two decades. On the other hand 37 countries where casualties
were recorded, could not be modelled either because of lack of data or because they did not fit
with the model (specific socio-economical contexts).
24
Table 13. List of the 37 countries with recorded casualties in CRED but not
modelled
Afghanistan, Azerbaijan, Cuba, Democratic People's Republic of Korea, Democratic Republic
of the Congo, Djibouti, Dominica, France, Greece, Liberia, Malaysia, Montserrat, Myanmar,
New Caledonia, Portugal, Solomon Islands, Somalia, Spain, Sudan, Swaziland, Taiwan,
Tajikistan, Vanuatu, Yugoslavia, Antigua and Barbuda, Armenia, Guadeloupe, Guam, Israel,
Martinique, Micronesia (Federated States of), Netherlands Antilles, Puerto Rico, Reunion,
Saint Kitts and Nevis, Saint Lucia, United States Virgin Islands.
Table 14. List of the two countries absent of both CRED and Model
Anguilla and Bosnia-Herzegovina
The Table 15 indicates when the comparisons were possible (196) out of 249.
Table 15. Result from the classification
Modeled risk
The Figure 9 depicts the distribution of killed and killed per population for the model
with categories in different colours. One can notice the diagonal cut between the categories,
meaning that both killed and killed by population have their respective role. A small country
with lower number of casualties but high in proportion of its country being in the same
categories as large countries with higher number of casualties but with lower percentage.
25
Figure 9 Scatter plot of expected casualties from multiple model (windstorm,
drought, earthquake, flood)
6 FINAL COMMENTS AND RECOMMENDATIONS

6.1. General comments
The purposes of the DRI was to identify whether global data sets could be used for
identifying population living in exposed areas and demonstrate the link between socio-
economical parameters and vulnerability. The level of correlation achieved delineates that
both physical exposure and variables tested are significant and could be used for categories of
risk identification. The correlation found is even much higher than initially thought. This is
particularly true for climatic events. Smaller correlation was achieved with earthquakes, this is
believed to be due to the long returning period, which is less compatible with the 21 year
period observed.
Except for drought, in the three other hazards the role of physical exposure appeared to
be the most significant, however socio-economical parameters such as GDPcap, HDI, urban
growth, percentage of arable land and local population density, were also selected depending
on the type of hazards. The sign of the exponents was always following what the common
sense and specialists would have recommended with the notable exception of local density for
flood. In the case of drought, the socio-economical context is playing the stronger role and is
highly sensitive to the data quality (in this case access to safe water).
All in all, the method used in this statistical analysis proved to be appropriate and allows
the identification of the parameters leading to a higher risk and vulnerability. Such a model
should not be used as a predictive model. Firstly because the level of data precision can not
reach such precision. Secondly, because a significant discrepancy of losses between two
26
events in the same country can be found. This shows the variability due to micro spatio-
temporal context. The risk maps provided in this research are not to be confused with danger
maps. At a local scale predictive model can and should be made allowing better urban
planning and improved evaluation of risk. Maps at global scale are only produced in the aim
of identifying the countries with the highest needs corresponding to the request from UNDP.
Extraordinary events – also called century disasters – do not follow the normal trend.
Hurricane Mitch (Central America, 1998), or the flood causing the landslide in Caracas
(Venezuela, 1999), earthquake in Armenia (1988), were clearly off the regression line. This is
due to the abnormal intensity of such events which do not correspond with the average
intensity. These events are (hopefully) too rare to be approached by a two decade period.
Incorporating the intensity can only be done on an event per event approach. When entering
an average intensity, the numerous low intensity events are biasing the average and finally the
intensity was rejected as explicative variable.
6.2. Recommendations
Use of the model
The very high sensitivity to the data of the drought model confirms that the equation
should not be used for predictive applications.
Socio-economical variables
Results delineate that global data sets can still be improved both in terms of precision
and completeness, however they are already allowing the comparison of countries. Other
indicators such as corruption index (transparency) or political indicators would be interesting
to test in the model, when all the countries will be available. Efforts on compilation are still
needed. Tremendous amount of work was involved (by UNEP/GRID-Geneva GEO team) to
verify and complete the data.
Floods
The geophysical data can also be improved. The watersheds for flood physical exposure
is based on a 1 km cell resolution for elevation. A new global data set on elevation from radar
measures taken from the NASA shuttle is expected in 2004. It consists in a 30m resolution
grid for the US and 90m resolution for the global coverage. This would allow the refining of
the estimated area flooded. This would be especially welcome for the central Asian countries
where the watersheds taken were of very poor resolution. Collaboration with Dartmouth Flood
Observatory would be an asset.
Earthquakes
If information on soil (quaternary rocks) and faults orientations can be implemented, it
would then be possible to compute intensity using modified Mercali scale and thus with much
higher precision on the area affected. Alternatively a method for deriving frequency based on
the Global Seismic Hazard Map from the GSHAP [Giardini 1999] could be also used.
Cyclones
Once the data from North Indian Ocean are available, a vector approach should be
applied using the PreView Global Cyclone Asymetric Windspeed Profile model developed by
UNEP/GRID-Geneva. This method computes the areas affected based on central pressure and
sustainable winds.
Drought
It might be interesting to test other precipitation data sets with higher spatial resolution,
although the resolution did not seem to be causing so much problem. The use of geo-climatic
zones might be useful in order to take into account the usual climate of a specific area. Indeed
27
a drop of 50% precipitation might not have the same consequence on a humid climate as
compared to a semi-arid area. The use of the Global Humidity Index (from UNEP/GRID
UEA/CRU) might help in differentiate these zones. Measuring food insecurity (using e.g.
information on conflict and political status) would be also a significant improvement as
compared to physical drought.
The case of small islands and archipelagos

Small islands and archipelagos are causing problems. In some cases they were too small
to be considered by the GIS automated algorithms. This was typically the case for the
population. The raster information layer for the population can not be used to extract the
population of small islands. For single island countries, the problem might be overcome by
using the population of the country, but for the others this was not possible. Indeed, when
superimposing cyclone tracks on top of archipelago, the population is needed for each island.
A manual correction is needed, but could not be performed due to the time frame of the study.
The compilation of socio-economical parameters was also not complete for the islands. This
could probably be improved by collaborating with SOPAC (Fiji).
For all these reasons, the case of small islands and archipelagos would need a separate
study and intuitively, the vulnerability for isolated countries might be different than other
connected countries.
The issue of indicators

To what extend the casualties are proportional to the significance of total losses
(including losses of livelihood)? In the case of earthquakes where no early warning exists, this
might be a good proxy, although it will highly depend if the earthquake epicentre is located in
rural or urban areas. For flood, however, the casualties are usually much smaller in relation to
losses of houses, infrastructures and crops. The ideal would be to have access to records of
livelihood losses in order to calibrate the severity of a hazard type as compared to another
(while considering the magnitude). Without such data no scientific method can be
implemented. Several options can be explored:
1. Organising a workshop with experts and asking them for a comparative severity of hazard.
2. Ask relief and aid organisations what is their average budget for recovery and mitigation
for each type of hazard.
This is a significant issue that could not be taken into account during this first research, but
needs to be incorporated in the future to improve the index quality. This would also allow to
normalise physical exposure. So far it is impossible to state how bad it is to be exposed to an
earthquake as compared to a flood, except eventually in terms of surviving chances.
Extending to other hazards

Volcanoes
An attempt was made to model volcanic eruption. It revealed the impossibility of
modelling physical exposure and vulnerability to volcanic eruption at global scale. If danger
maps can be derived for a single volcano at local scale, the variability of the volcanic
manifestations is far too complex to be generalised, it ranges from lahars (linked with
precipitation level, seismicity, topography, soils characteristics,…), tephra falls (depending
from wind direction and strenght), phreatomagmatic eruption,… . However, numerous experts
are working on surveying these activities and each volcano is well described.
1. Cautions must be taken on the use of small areas (smaller than 100 pixels) for the
calculation of population exposures, due to the precision of population data.
2. The risk results in the inversion of the frequency role: the less frequent the more
dangerous. Indeed the larger impacts are resulting from volcanoes with low
historical activity. The habitants of these areas are living in a false confidence.
28
Indeed, the volcano Ruiz (Colombia, 1985) which melting of its summit icecap
provoked the South America's deadliest eruption (22’800 killed) had a large
eruption in 1595 and smaller in 1828, 1829 and then has only been reported
smoking in 1831, 1833 [Herd, 1982]. Same situation with Pinatubo (Philippines,
1991) “Prior to 1991 Pinatubo volcano was a relatively unknown, heavily forested
lava dome complex with no records of historical eruptions.” [Global Volcanism
Program, Smithsonian Institution, 2001]. And similar with the lake Monoun
(Cameroun, 1986) “No previous eruptions are known from Lake Monoun” [Le
Marechal, 1975a]. These disasters demonstrated that low frequency does not mean
low risk, this is completely different as for the other hazards.
Data requested are probably existing. Finer resolution for elevation is a must, for
showing shape and relief of volcanoes, computing slopes and lahars danger. Remote sensing
analysis for local assessment of danger and population distribution would also be requested.
Numerous maps of volcanoes can be found at http://www.nmnh.si.edu/gvp/volcano/index.htm.
Tsunamis and Landslides
Some countries are not well represented by the model, because they are affected by
hazards which were not of global significance. This is the case of Papua New Guinea and
Ecuador, which are affected by tsunamis (respectively 67.8 and 14.3% of national casualties);
landslides are also causing significant impact in Indonesia (13,88%), Peru (33%) and Ecuador
(10.2%). As a result, the global risk is under evaluated for these countries.
Epidemics
This is more a health angle and should probably be taken care of by the World Health
Organisation (WHO). However, the appropriate sanitation, access to safe water, number of
physicians per inhabitants and other health infrastructure are also significant parameters of
development. Data on epidemics are now starting to be available. Epidemics is representing a
significant amount of casualties and AIDS is definitely impacting developing societies
especially (but not only) in Africa.
Conflicts
The case of conflicts although much more politically difficult to approach is probably
also highly correlated to human vulnerability. Results from a statistical analysis would be
extremely interesting.
Last word
These results delineate the relation between level of development and low casualties
from these four types of hazards. Stating that there is a relation can be understood both ways:
lower development may lead to higher casualties, but high hazard occurrence may also lead to
lower economical development as it destroys infrastructures and crops as well as it scares the
investors away. If higher impacts from natural hazards in developing countries were depicted,
the message should not be perceived as “developed countries should be taken as models”.
Other figures such as death from suicides, drug abuses or excess of fat food, are also leading
to numerous casualties and are highly and positively correlated with HDI!
This research underlines the usefulness of continuing the improvement of data
collection for a better identification of populations at risk. This is, however, not a final result
as such. Final results will be achieved when proper risk reduction measures will be
implemented leading to an observed decrease of casualties.
29
7 REFERENCES
Articles and books
Anand, S. and Sen, A., 2000, The Income Component of the Human Development Index, Journal of Human
Development, Vol. 1, No. 1
Anderson, M. and Woodrow, P., 1989, Rising from the Ashes: Development Strategies in Times of Disaster.
(Westview Press, Boulder).
Blaikie, P. et al., 1996, At Risk: Natural Hazards, Peoples Vulnerability and Disasters (Routledge).
Blong, R.J., 1984, Volcanic Hazards, A Sourcebook on the Effects of Eruptions (Academic Press Australia).
Bolt, B.A., Horn, W.L., Macdonald, G.A., Scott, R.F., 1975, Geological Hazards (Springer-Verlag Berlin-
Heidelberg-New York).
Burton, I., Kates, R.W. and White, G.F. 1993: The Environment as Hazard, Second Edition. New
York/London: Guilford Press, 290 pp. [Pp. 31-47]
Carter, N., 1991, Disaster Management, a disaster Manager’s Handbook (Asian Development Bank, Manilla).
Coburn, A.W., Spence, R.J.S. and Pomonis, A. 1991: Vulnerability and Risk Assessment. UNDP Disaster
Management Training Program 57 pp.
Demuth, S., Stahl, K., 2001, Assessment of the Regional Impact of Droughts in Europe (ARIDE), Final Report,
Institute of Hydrology, University of Freiburg, Freiburg, Germany.
Giardini, D., 1999, Annali di Geofisica, the global seismic hazard assessment program (GSHAP) 1992-1999,
Instituto Nazionale di Geofisica, Volume 42, N. 6, December 1999, Roma, Italy.
Herd, D.G., 1982, Glacial and volcanic geology of the Ruiz-Tolima Volcanic Complex, Cordillera Central,
Colombia: Publicaciones Geológicas Especiales del INGEOMINAS, no. 8, 48 p.
Holland, G. J., 1980, An analytic model of the wind and pressure profiles in hurricanes. Monthly Weather
Review, 108, 1212 1218.
Le Marechal, A., 1975a, Carte geologique de l'ouest du Cameroun et de l'Adamaoua, 1:1,000,000; ORSTOM.
Newhall, C. G., Self, S., 1982, The volcanic exposivity index (VEI): an estimate of explosive magniude for
historical volcanism. Jour. Geophys. Res. (Oceans & Atmospheres), 87:1231-8.
Peduzzi, P., Dao, H., Herold, C., Frédéric Mouton (2003), Global Risk And Vulnerability Index Trends per
Year (GRAVITY), Phase IIIa: Drought analysis, scientific report UNDP/BCPR, Geneva, Switzerland.
Peduzzi, P., Dao, H., Herold, C., (2002), Global Risk And Vulnerability Index Trends per Year (GRAVITY),
Phase II: Development, analysis and results, scientific report UNDP/BCPR, Geneva, Switzerland.
Peduzzi, P., Dao, H., Herold, C., and Rochette, D., (2001), Feasibility Study Report On Global Risk And
Vulnerability Index Trends per Year (GRAVITY), scientific report UNDP/BCPR, Geneva, Switzerland.
Schloemer, R. W., 1954, Analysis and synthesis of hurican wind patterns over Lake Okehoee, Fl. Hydromet
Rep. 31, 49 pp. [Govt. Printing Office, No. C30.70:31].
Simkin, T. and Siebert, L.1994, Volcanoes of the World, (Geoscience Press, Washington DC).
Smith, K., 1996, Environmental Hazards, Assessing risk and reducing disaster (Routldlege: London and New
York).
Tobin, G.A., Montz, B.E., 1997, Natural Hazards, Explanation and Integration (The Guildford Press, New
York, London).
UNDRO (United Nations Disater Relief Coordinator), 1979: Natural Disasters and Vulnerability Analysis in
Report of Expert Group Meeting (9-12 July 1979). Geneva: UNDRO. 49 pp.
UNEP, 2002, GEO: Global Environment Outlook 3. Past, present and future perspectives, UNEP.
30
References on Internet
Anderson, E., Brakenridge, G.R., 2001, NASA-supported Dartmouth Flood Observatory.
http://www.dartmouth.edu/artsci/geog/floods/index.html.
Babin, S. and Sterner, R., 2001, Atlantic Hurricane Track Maps & Images,
http://fermi.jhuapl.edu/hurr/index.html
Birdwell K.R., Daniels, R.C., 1991, A Global Geographic Information System Data Base of Storm Occurrences
and Other Climatic Phenomena Affecting Coastal Zones (1991)
http://cdiac.esd.ornl.gov/ndps/ndp035.html
CIESIN, IFPRI, WRI, 2000, Gridded Population of the World (GPW), Version 2,
http://sedac.ciesin.org/plue/gpw/
Council of the National Seismic System (as of 2002), Earthquake Catalog, http://quake.geo.berkeley.edu/cnss/
Deichmann, Uwe, 1996, GNV197 - Human Population and Administrative Boundaries Database for
Asia, UNEP/GRID-Geneva, http://www.grid.unep.ch/data/grid/gnv197.php
Donlin C. and Fitzgibbon, T., 2001, Geopubs - Online Geologic Publications of the Western United States,
USGS: http://geopubs.wr.usgs.gov/docs/wrgis/fact
Giardini, D., Grünthal, G., Shedlock K., Zhang, P., 2000, Global Seismic Hazard Assessment Program,
http://www.seismo.ethz.ch/GSHAP/
Global Volcanism Program, National Museum of Natural History, E-421, Smithsonian Institution, Washington
DC 20560-0119, http://www.nmnh.si.edu/gvp/index.htm
Global Volcanism Program, Volcanic Activity Reports, Pinatubo, Index and All Reports, National Museum of
Natural History, Smithsonian Institution, Washington:
http://www.nmnh.si.edu/gvp/volcano/region07/luzon/pinatubo/var.htm#1605
International Decade for Natural Disaster Reduction, http://www.unisdr.org/unisdr/indexidndr.html
IRI/Columbia university, National Centers for Environmental Prediction Climate Prediction Center (as of
2002), CPC Merged Analysis of Precipitation (CMAP) monthly gridded precipitation,
http://iridl.ldeo.columbia.edu/
ISRIC, UNEP, 1990, Global Assessment of Human Induced Soil Degradation (GLASOD),
http://www.grid.unep.ch/data/grid/gnv18.php
Landsea Christopher W., 2000,NOAA/AOML, FAQ: Hurricanes, Typhoons, and Tropical Cyclones.
http://www.aoml.noaa.gov/hrd/tcfaq/tcfaqA.html#A1
OFDA/CRED, 2001, EM-DAT: The OFDA/CRED International Disaster Database, http://www.cred.be/emdat
Peduzzi, P., 2000, Insight of Common Key Indicators for Global Vulnerability Mapping, presentation for the
Expert Meeting on Vulnerability and Risk Analysis and Indexing, Geneva 11-12 September 2000,
UNEP/DEWA/GRID-Geneva:
http://www.grid.unep.ch/activities/earlywarning/preview/appl/reports/reports.htm
Peduzzi, P., 2001, Project of Risk Evaluation, Vulnerability Indexing and Early Warning (PREVIEW),
UNEP/DEWA/GRID-Geneva:
http://www.grid.unep.ch/activities/earlywarning/preview/index.htm
Topinka, L., 2001, Cascades Volcano Observatory, USGS, Vancouver, Washington, USA:
http://vulcan.wr.usgs.gov/Volcanoes/
Transparency international, 2001, Global Corruption Report 2001,
http://www.transparency.org/
U.S. Geological Survey, 1997, HYDRO1k Elevation Derivative Database,
http://edcdaac.usgs.gov/gtopo30/hydro/
UNDP, 2002, Human Development Indicators, http://www.undp.org/
UNEP,CGIAR,NCGIA, 1996, Human Population and Administrative Boundaries Database for Asia,
http://www.grid.unep.ch/data/grid/human.php
UNEP/GRID (as of 2002), GEO-3 Data portal,
Http://geodata.grid.unep.ch/
VHP., 2000, Volcano Hazards Program, Strategy for reducing volcanic risk, U.S. Department of the Interior,
U.S. Geological Survey, Menlo Park, California, USA: http://volcanoes.usgs.gov/About/What/Assess/
William P. Leeman W.P., 1999, Volcanism & Volcanic Hazards Summary of basic terms and concepts, Rice
University: http://www.ruf.rice.edu/~leeman/volcanic_hazards.html
31
View publication stats

Global Risk and Vulnerability Index Trends Per Yea

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Global Risk and Vulnerability Index Trends Per Yea

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Global Risk and Vulnerability Index Trends Per Yea

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Global Risk And Vulnerability Index Trends per Year (GRAVITY)

Hy Dao Frederic Mouton

SEE PROFILE SEE PROFILE

GOFORBO View project

The user has requested enhancement of the downloaded file.

A technical report for:

United Nations Development Programme

Hy Dao & Pascal Peduzzi

Additional short term collaborators:

1 WORKING DEFINITIONS AND FORMULAE

By UN definition [UNDRO, 1979], the risk is resulting from three components:

1.2. Formula and method for estimating risk and vulnerability

2.2. Risk indicators

Indicators for risk Advantages Inconveniences

2.3. Vulnerability indicators

Categories of Indicators Drought Flood Source3

2.4. Data sources

Hazard type Data source

Droughts IRI/Columbia university, National Centers for Environmental Prediction

Theme Data source

3 COMPUTATION OF PHYSICAL EXPOSURE

PhExp nat = ∑ Fi ⋅ Popi

3.2. The case of cyclones

Saffir-Simpson tropical cyclones classification is based on the "maximum sustained

3.3. The case of floods

3.4. The case of earthquakes

3.5. The case of droughts

Computation of physical exposure

4 STATISTICAL ANALYSIS : METHODS AND RESULTS

Using the logarithmic properties, the equation could be written as follows:

ln( K ) = ln(C ) + α ln( PhExp) + α 1 ln(V1 ) + α 2 ln(V2 ) + ... + α p ln(V p )

4.2. Detailed process

original value (V)

Choice between variables

The stepwise approach

4.3. Mapping Risk

Number of categories and method of classification

ln( K ) = 0.78 ln( PhExp) − 0.45 ln(GDPcap ) − 0.15 ln( D) − 5.22

Table 9: Exponent and p-value for flood indicators

ln( K ) = 1.26 ln( PhExp) + 12,27 ⋅ U g − 16.22

ln( K ) = 1.26 ln( PhExp3 _ 50) − 7.58 ln(WATTOT ) + 14.4

5 MULTIPLE RISK INTEGRATION

Figure 8 Process for multiple risk integration

The total risk computation

____ 0.66 ___ −2.03

No data, abnormal values and specific cases

6 FINAL COMMENTS AND RECOMMENDATIONS

The case of small islands and archipelagos

The issue of indicators

Extending to other hazards

View publication stats

You might also like

__ 0.66 _ −2.03