1. Introduction
Natural hazards occur frequently around the world. Some natural hazards turn into disasters after they cause tangible (or physical) impact, such as human fatalities and property damages; and intangible (non-physical) impacts, such as psychological, mental, and political wounds [
1]. Research has shown that human casualties and economic losses due to natural disasters have been increasing over the last five decades [
2,
3]. This trend is likely to continue due to growing urbanization, rising populations, deepening industrialization, and worsening global environmental and climate change [
4]. For example, from 1950 to 1979, 1779 natural disasters occurred across the world, causing 4,860,449 casualties, 1,372,606 injuries, and
$78 billion in property damages. However, from 1980 to 2015 the reported number of natural disasters increased to 11,494 or about 6.5 times as many as those in the period of 1950–1979: Causing 2,599,237 fatalities, 6,354,195 injuries and
$2.71 trillion in property damages [
5].
In order to effectively prepare for, respond to, and recover from natural disasters, good hazard risks and assessments at various spatial and temporal scales are essential [
6,
7]. This research proposes and applies a new quantitative model to assess physical impacts of world natural disasters as risks at the country-level, using the international Emergency Disaster Database (EM-DAT)—compiled by the Centre for Research on the Epidemiology of Disaster (CRED) [
8]. The model takes country-specific posterior data on natural disasters for the period of 1900–2015, including disaster occurrences, classifications, and disaster impacts, and produces disaster risk estimates, rankings, and correlations to country socio-economic attributes. The proposed model is validated by Pearson and Spearman correlations, for the values and ranks of the estimated and observed total disaster impacts. The country-level risk assessment may be coarse due to using ‘country’ as the spatial unit, especially for large countries in which intra-country variability may be quite high. However, it is still worthy to reveal inter-country differences, especially global hot-spots, in risk impacts for historically reported natural disasters in the world. A comparative global risk analysis provides useful insights for country-specific or global humanitarian policies toward minimizing natural disaster impacts, and promoting sustainable development. Perhaps more importantly, as the literature review section illustrates below, while various sub-country spatial scales have been used [
1], using country as the spatial scale is relatively rare in disaster risk research [
4,
5], especially from the global country comparative perspective for multiple natural disasters over the previous 115 years.
3. Methods and Materials
The methodology consists of a set of notations in
Figure 1 and a risk assessment model, in which we consider country
k as the spatial unit, and hazard occurrences
i for the natural disaster group
m = 1. This includes sub-groups
g, with each resulting in some disaster impacts
j, including fatality, injury, affected, and damage. Note that the model collapses all time units (e.g., year) into the 1900–2015 period. Therefore, the model primarily is a spatial one for estimating cumulative expected risks for countries based on multiple impacts. Although disasters happen seemingly randomly, the model does not consider temporal variations, nor spatial variations within country over the period. The model’s main purpose is to estimate country-level expected risks in fatalities, injuries, people affected, and economic damages.
3.1. Expected Risk Model
Let
be the occurrence
i of subgroup g under the natural hazard causing disaster impact of type
j in country
k for the period of 1900–2015. Such an ex ante occurrence generated disaster impact
. Define the physical loss
as the impact reported either as human fatality, people injured, people affected, or property damage in EM-DAT. We have:
The expected risk for disaster impact
j for a country
k,
is the sum of the products of each possible natural disaster impact times its associated probability—it is a weighted average of the various possible natural disaster impacts, with the weights being their probabilities of occurrence in country
k:
A percentage of the expected risk for country
k,
, relative to the world, can be calculated as:
The dispersion of the expected risks are important to the understanding of disaster impacts. The tighter the dispersion, the more likely it is that the actual disaster impact will be close to the expected risk. Consequently, the less likely it is that the actual disaster loss will end up far below or above the expected loss. Thus, the less spread around the expected risk means the lower the disaster impact for a country.
The standard deviation,
, is commonly used as the measure of the tightness of the probability distribution around the expected or mean risk. The standard deviation is a probability-weighted average deviation from the estimated impact, and it gives an idea of how far above or below a disaster impact is likely to be.
Therefore, the natural disaster risk of country k is based not only on its expected impact in Equation (3), but also its deviation in Equation (5). A 95% confidence (p < 0.05) would provide the low-bound risk (Low-Rjk) with larger of the 0 or − 2 and the upper bound risk (High-Rjk) with + 2.
Another useful measure of expected risk is the coefficient of variation,
which is the standard deviation divided by the expected risk. This reflects the standard deviation per unit of expected risk. It provides another meaningful basis for comparison when the disaster risks on different spaces are not the same:
The Equations (1)–(6) can be applied to all countries based on their historical impacts caused by natural disasters. Since countries vary in social-economic-physical features, it would be meaningful to see the normalized expected risks, standard deviations, and coefficients of variations by these features, such as population, gross domestic product (GDP), and land area.
where
and
are normalized expected risks and standard deviations.
= social-economic features with
t = population, land area, GDP, population density (PD), or per capita GDP (PG). For example,
and
represent the expected fatalities normalized by population density and per capita GDP, respectively. In order to have a comprehensive assessment of a nation’s risk in the event of a natural disaster, the country’s expected fatality, injury, people affected, and damage, as well as their standard deviations, percentages, low and upper bounds, and ranks should all be considered. In general, the larger a country’s expected risk is, together with wider ranges, larger percentages, and higher ranks with expected and normalized risks, the riskier the country is expected to be.
3.2. Data Preparation
Three types of databases were used as inputs for the risk assessment model. First, the natural disaster database was obtained from EM-DAT, a global disaster database published by CRED in Brussels [
8]. EM-DAT was compiled from various sources, including UN agencies, non-governmental organizations, insurance companies, research institutes and press agencies and organized for disaster groups for natural disasters, technological disasters, and complex disasters—each of which is further split into subgroup, type, and subtype [
5]. Each natural disaster contains important disaster—human injuries, fatalities, people affected, and damages—by space (e.g., continent, region, country) and time (e.g., year). Second, the world country boundary GIS (Geographic Information System) database was drawn from Global Administrative Unit Layers (GAUL) for the world developed by the Food and Agriculture Organization (FAO) of the United Nations [
35]. This world boundary data was used together with the EM-DAT database for the model. Third, the world social-economic attributes of historical country population, area, and GDP data were extracted from the World Development Indicators database by the World Bank [
36].
Figure 2 outlines the necessary steps for database processing. The first step is to process tables of human injuries, fatalities, people affected, and damages by country and by year for natural disasters using the EM-DAT database for the period of 1900 to 2015. The second step is to calculate probability weighted risk impacts, their standard deviations, ranks, percentages, and correlations of socio-economic attributes by country. This step also involves validation using the observed natural disaster data and the results for the model using scatter plots and Pearson correlation and Spearman rank order correlation. The third step is to perform GIS functions using “Summarize”, “Join”, and “Field Calculator” on the tables produced in the first and second steps, and the World Boundary GIS layer. This step also produces spatial visualizations of natural disaster risks by country.
4. Results and Discussion
Expected risks and their relevant measures for ranking, dispersion, percentage, and correlation for the 208 countries selected in this study are summarized in
Table 2,
Table 3,
Table 4 and
Table 5. This only lists the top 30 countries, according to expected fatality, injured, affected, and damage.
4.1. Fatality
Table 2 ranks the top 30 countries according to the expected fatality (
RkF) and its normalized expected fatality (
RkF-PG,
RkF-PD). The table also lists rank (
RankkF), dispersion
(σkI,
Low-RkI,
High-RkI), coefficient of variance (
CVkF), and total observed fatality and estimated fatality for each of the 208 countries.
Firstly, the top 30 countries in expected fatality spread over the global continents. With China, India, Bangladesh top all in Asia; Russia and Italy in Europe; Uganda, Niger, and Ethiopia sit high in Africa; and Guatemala, Peru, Chile and the United States hold the top spots in Americas. While Asia has the most countries ranked in the top 30 and had the most deaths, Oceania has no country listed. Secondly, China is particularly deadly with almost 50% of world fatalities, followed by India and Bangladesh at almost 17% and 9%, respectively. Thirdly, the dispersion measures are quite wide and large for top countries, but their spreads per unit expected fatality are quite different (e.g., smaller for China with CV = 0.86 and Uganda with CV = 0.72, but larger for India with CV = 1.76, Russia with CV = 1.56, Ethiopia with CV = 2.08, and Haiti with CV = 2.49). Finally, the ranks of expected deaths and ranks of normalized expected fatalities correspond, as do the observed and estimated total fatalities.
4.2. Injured
Similar to the results of expected fatalities discussed just above,
Table 3 ranks the top 30 countries according to their expected injuries (
RkI) and their normalized expected injuries (
RkI-PG,
RkI-PD). The table also lists rank (
RankkI), dispersion
(σkI,
Low-RkI,
High-RkI), coefficient of variance (
CVkI), and total observed and estimated injuries.
Firstly, top 30 countries in expected injuries are found in all continents, with China, Bangladesh, and Indonesia leading in Asia; Russia, Macedonia, and Italy in Europe; Sudan and Algeria in Africa; Peru, Guatemala, Haiti, El Salvador; and the United States in Americas. Again, Asia has the most countries ranked in the top 30 and the most deaths, while Oceania has no country listed. Secondly, China and Bangladesh are particularly vulnerable to injuries, with each having over 25% of the world total, followed by Peru, Indonesia, and Iran with 13.13%, 9.38%, and 4.48%, respectively. Thirdly, the dispersion measures are quite wide and large for top 30 countries, but their spreads per unit expected injuries are quite different (e.g., small for China with CV = 0.77 and Bangladesh with CV = 0.63, but large for Peru with CV = 2.10, Haiti with CV = 3.35, and India with CV = 2.53). Finally, the ranks of expected injuries and ranks of normalized expected injuries are quite similar, as are the total observed and estimated injuries.
4.3. Affected
Table 4 shows the top 30 countries in which people were affected by natural disasters. There are several numbers worth noting in this table. Firstly, Asian countries are expected to produce the highest expected number of people affected in the event of a disaster, For example, India had the number of people affected with more than 429 million (or 32.11%), followed by China with over 286 million (or 21.43%). These countries, alongside Bangladesh (9.74%), Philippines (6.69%), and Pakistan (3.24%) make up the top 5. The top 30 rankings also included: Brazil, the United States, Colombia, Argentina, Cuba, Mexico, Peru, and Chile in Americas; and Ethiopia, Kenya, Mozambique, Madagascar, Sudan, Niger, and Nigeria in Africa. Secondly, Australia is the only country from Oceania; as is Russia from Europe, in the top 30. Thirdly, again, Asia is the top continent with 11 countries, including the top 7, in the top 30. Fourthly, the dispersion measures are relatively significant, but the standard deviation per unit of expected people affected are relatively small, especially the top 10 with
CV = 0.53 for the United States and
CV = 0.86 for China, except for Ethiopia with
CV = 1.98. Finally, the ranks for expected people affected normalized by population density and per capita GDP are very similar, so are the observed and estimated total people affected.
These results show that casualties, injuries, and people affected by natural disasters are highly correlated to population size, distribution, and density in general [
17], and perhaps more to vulnerable population groups in particular [
37].
4.4. Damage
Table 5 provides the top 30 countries which suffered severe physical damage by natural disasters. The damages were measured by monetary value, in
$1000 U.S. dollars. It should be noted that the United States, which was ranked 21st in expected fatalities; 15th in expected injuries, 9th in people affected, received the highest ranking with 37.63% of the world expected damages, followed by Mexico, Chile, Brazil, Canada, Cuba and Argentina in Americas. In addition, some new faces from developed countries—for example: Italy, Germany, France, United Kingdom in Europe; and Australia and New Zealand in Oceania—also appeared in the top 30 countries. Many Asian countries, such as China, Japan, China (12.14%), Japan (11.30%), Thailand, India, Koreas, were in the top 30. The ranges are higher for the top countries, while the dispersions of per unit expected damage are fairly small (e.g., the United States with
CV = 0.45), except New Zealand with
CV = 1.86 and Japan with
CV = 1.16. Finally, the ranks for the expected damages and the normalized expected damages quite match in values, so do the magnitudes for the total observed and estimated damages.
Properties are used by people for social, cultural, and economic activities. Therefore, property damages are presumably highly correlated with population size, distribution, and density. While many countries with high fatalities and injuries are also high in property damages,
Table 5 shows this is not always the case. For example, the United States ranks quite differently in the categories of fatality, injury, and in property damage, as do some other developed nations. This observation is partially related to property damage valuation parities, or metrological differences [
38].
4.5. Model Performance
The similar ranks between the expected risks and the normalized expected risks and the similar and values between the total observed and estimated risks indicate that the model performs reasonably well. We used two methods to evaluate the model’s performance: the Spearman rank correlation, and the Pearson correlation. Both are shown in scatter plots, with correlation and
R and
R2 values, in
Figure 3. The Spearman rank correlations were calculated for
Rankkj vs.
Rankkj-PG,
Rankkj-PD. Whereas, the Pearson correlations were computed for estimated and observed totals for fatality, injured, affected, and damage. It is believed that the higher the
R and
R2 values, the better the correlations, and the better the model’s performance.
Shown in
Figure 3, the Spearman rank correlations for the expected risks and normalized expected risks are very high, ranging from the low 0.904 for fatality (a)
RankkF vs.
RankkF-PD to the high 0.972 for affected (c)
RankkA vs.
RankkA-PG. The Pearson correlations are also very high, ranging from the low 0.863 for affected (g) to the high 0.966 for damage (h). It is interesting to note that from
Figure 3 (a–d) the lower or higher ranked countries in particular have much higher correlations. This is especially true, due to the small but similar expected and normalized expected risks for countries ranked after 150 for injured (b), and after 175 for damaged (d).
The scatter plots
Figure 3 (e–h) show all observed and estimated data points, including top ranked countries with extremely large total risks. These top hotspots may be statistically regarded as outliers or influential points, but they are not removed due to their apparent importance for global risk analysis, especially for ranking.
4.6. Natural Disaster Hot-Spots
Table 6 summarizes the percentage shares of the expected and normalized expected fatality, injured, affected, and damage by the top 10, 20, and 30 countries, and then the remaining 178 countries. Their percentage shares of total estimated and observed disaster impacts are also shown in
Table 6. The top 10, 20, and 30 countries respectively accounted for more than 90%, 95%, and 97% for expected fatality (
RkF); 88%, 94%, and 97% for expected injury (
RkI); 81%, 88%, and 91% for expected people affected; and 74%, 85%, and 91% for expected damage. Meaning that if historical trends and patterns continue, more than 91% of losses from future natural disasters are expected to happen in those 30 nations, especially those in the top 10. The remaining 178 countries only shared 2.52%, 2.92%, 8.13%, and 8.82% of the expected risks, implying quite a few countries, especially smaller ones, did not experience many natural disasters.
The similar patterns can be found in percentage shares of expected risks normalized by: Population density and per capita GDP, total observed and estimated risks, and the risk dispersion (
σkj). Referencing to the specific top 30 countries in
Table 3,
Table 4,
Table 5 and
Table 6, we can say that the natural disaster hotspots are mostly in Asia (e.g., China, India, Bangladesh), a few in North America (e.g., the United States), some in Europe (e.g., Russia, Italy, Germany), and Africa (e.g., Ethiopia, Sudan, Niger, Algeria). While Oceania and small countries are relatively safe places for natural disasters.
Figure 4 illustrates the spatial distribution of expected disaster risks at the country-level. The countries, which have a higher number of expected fatalities, injuries, people affected and damages, are represented in darker brown colors.
5. Conclusions and Remarks
A natural hazard (such as floods, earthquakes, and hurricanes) is an event that happens in ‘mother nature’. However, a natural disaster causing deaths, injuries, and property losses occurs through interactions between natural and the man-made environments. No country can be immune from natural hazards. However, some countries are suffered more from natural disasters than others. One of the major findings of this research is that natural disasters occurred during the period of 1900 to 2015 severely impacted human lives and economies of countries—especially the top 10, 20, and 30 countries ranked by expected risks. These top hot spots are also large, populated, developed, or rapidly developing ones. For example, the United States received the highest ranking in terms of natural disaster occurrence. This was followed by China, India, the Philippines, Indonesia, Bangladesh, and Japan—all of which are located in Asia. The nations suffering the most from fatalities were China, India, Russia, Bangladesh, and Ethiopia; while the highest number of injuries happened in Peru, China, Bangladesh, Haiti, Indonesia, Japan, and India. Also, the United States was ranked first in terms of damage, which was followed by Japan, China, Italy and Germany. All of these nations with high damages also had high gross domestic products. These results indicate that natural disasters happened in nations across every continent (e.g., Asia, Europe, Africa, and America), even though some countries experienced more losses than others. These findings support the research by Dilley [
25], and Giuliani and Peduzzi [
23]. Perhaps more important is the high correlations between the expected risk and the socio-economically normalized expected risks: Indicating not only the relevance of the population density and per capita GDP for risk analyses, but the success of this model’s performance.
The model calculates expected human fatalities, injures, people affected, and economic damages along with their relevant percentages, ranges, and ranks, for 208 countries in the world. Scatter plots and Spearman rank correlations between expected risks and normalized expected risks indicate that the model perform well. In addition, the scatter plots and Pearson correlations showed that the total estimated human casualties and economic losses aligned with the corresponding observed disaster totals. For example, China, ranked first in terms of the ‘real’ number of fatalities from natural disasters, and also held the first position in ‘expected’ risk of fatalities. Similarly, India and Bangladesh, which were the second and fourth nations in terms of the ‘real’ fatalities, and took the second and third rankings, respectively, in the total estimated fatalities. Furthermore, the United States, Japan, and China, which had the first, second, and third rankings in the total observed economic damages, respectively, were ranked first, third, and second in the total estimated economic damages. The results and rankings from this model are synthetic in nature and similar to Munich [
39] on natural hazard index, and Dilley [
25] on natural disaster hotspot analysis, hence, cross-model comparative studies may be warranted.
Finally, the expected risk model, based upon the widely used EM-DAT historical natural disaster data, can be used as a new alternative approach to conduct country-level risk assessments—or risk analyses of fatality, injured, affected, and damage, especially for counties’ governments to make sound disaster preparation and mitigation decisions, policies, or plans regarding natural disasters. Local governments at the state, provincial, or municipal levels in a country can also use the model, if disaster datasets for jurisdiction-specific natural disaster information is available. The model can also be tested for specific natural disasters, especially floods, earthquakes, and hurricanes, so that more specific prevention, evaluation, mitigation, and recovery policies or plans can be made and implemented to minimize risks or losses from natural disasters. However, the reliability and validity of EM-DATA are critical for all these efforts. Therefore, cross-database comparative studies, especially for databases with similar spatial and temporal coverages as used in Giuliani and Peduzzi [
20], and Gregorowski et al. [
40], are also imperative.