The International Disaster Database, EMDAT from the Center for Research on the Epidemiology of Disasters (CRED, Belgium) is often used as a reference for losses on human life and property resulting from select natural and man-made disasters. This database has over 21,000 country-level records from 1900 to present. Data is available for free from EMDAT.
Major issues with EMDAT data are as follows.
Location associated with a disaster is often a province/state within a country. It would be immensely valuable to have the approximate geographical boundaries or coordinates of the disaster locations and affected spatial domain. EMDAT does not provide such information. A shapefile associated with each file would be a great place to start. The package emdatr addresses some the above issues. The goal of the package is to improve the EMDAT database by promoting its use, shedding light on its limitations, and making analysis of the data easier using R.
After installing the package, load the package along with RCurl (for data extraction from bitbucket.org), ggplot (for graphics) and plyr (for data manipulation).
library(emdatr)
library(RCurl)
library(ggplot2)
#> Warning: package 'ggplot2' was built under R version 3.1.3
library(plyr)
The single main function provided by emdatr is extract_emdat. This could be used to extract a sample of the EMDAT data (which comes with this package) or the entire data. First, load the sample data that comes with the package.
losses_2014 <- extract_emdat()
dim(losses_2014)
#> [1] 473 18
head(losses_2014)
#> Start End Country
#> 1 01/02/2014 07/02/2014 Afghanistan
#> 2 24/04/2014 02/05/2014 Afghanistan
#> 3 03/06/2014 10/06/2014 Afghanistan
#> 4 10/10/2014 10/10/2014 Angola
#> 5 10/05/2014 10/05/2014 United Arab Emirates
#> 6 25/01/2014 25/01/2014 Argentina
#> Location
#> 1 Jauzjan, Faryab, Kunduz, Kabul
#> 2 Badakhshan, Jawzjan, Faryab, Sar-e-Pul, Bagdghis, Balkh, Baghlan,Samangan, Kabul, Ghor, Logar, Takhar, Bamyan, Hirat, Parwan,Nangarhar,Khost, Kandahar, Panjsher, Hilmand, Kapisa, Nimroz, Laghman, Kunduz,Daykundi, Nuristan, Kunar provinces
#> 3 Guzargah-e-Nur district (Baghlan province)
#> 4 Near Porto Amboin
#> 5 Ruwayyah (Duba??)
#> 6 El Rodeo (Catamarca province)
#> Type SubType Killed TotAffected EstDamage
#> 1 Storm Convective storm 63 NA NA
#> 2 Flood Flash flood 431 140100 NA
#> 3 Flood Flash flood 81 10035 NA
#> 4 Transport accident Road 26 NA NA
#> 5 Transport accident Road 15 14 NA
#> 6 Landslide <NA> 24 350 NA
#> DisNo Group Year ISO_EM ISO_alpha3 ISO_cntry
#> 1 2014-0084 Meteorological 2014 AFG AFG Afghanistan
#> 2 2014-0134 Hydrological 2014 AFG AFG Afghanistan
#> 3 2014-0185 Hydrological 2014 AFG AFG Afghanistan
#> 4 2014-0394 Technological 2014 AGO AGO Angola
#> 5 2014-0150 Technological 2014 ARE ARE United Arab Emirates
#> 6 2014-0021 Geophysical 2014 ARG ARG Argentina
#> region Pop GDP
#> 1 Asia NA NA
#> 2 Asia NA NA
#> 3 Asia NA NA
#> 4 Africa NA NA
#> 5 Asia NA NA
#> 6 Americas NA NA
The default options of extract_emdat do not make any adjustments for inflation. Next, obtain the entire dataset with the inflation option enabled. This might take a few moments. The resulting dataset has all historical financial losses adjusted for inflation. If a different year of adjustment is desired, change the base_year accordingly.
losses_all <- extract_emdat(sample_only = FALSE, inflation = TRUE)
#> downloading data from bitbucket. might take a few moments...
Adjustment for inflation is currently based on the Consumer Price Index (CPI) of the United States - i.e., the adjustment factor is the ratio of CPI in the base_year and the CPI in the year of the disaster. Such an adjustment may be inappropriate since it does not directly account for economic changes in the country of occurrence. Future updates to the package could incorporate inflation adjustment using GDP, population and other socio-economic indicators.
Example graphics shown in this section are intended to duplicate some of those shown in EMDAT's ADSR report from 2013.
From the entire dataset, identify natural disasters only.
nat_data <- losses_all[losses_all$Group %in% c("Climatological", "Geophysical",
"Hydrological", "Meteorological"), ]
nat_data <- droplevels(nat_data)
# assign missing value to 0s before using cbind in aggregate
nat_data$Killed[is.na(nat_data$Killed)] <- 0
nat_data$TotAffected[is.na(nat_data$TotAffected)] <- 0
nat_data$Year <- as.factor(nat_data$Year)
Identify number killed and affected per year from 1990 through 2013.
gfx_deaths <- aggregate(cbind(Killed, TotAffected) ~ Year, data = nat_data, FUN = sum)
# total in millions
gfx_deaths$Total <- (gfx_deaths$Killed + gfx_deaths$TotAffected)/10^6
gfx_deaths <- gfx_deaths[, c("Year", "Total")]
gfx_deaths <- gfx_deaths[gfx_deaths$Year %in% seq(1990, 2013), ]
gfx_deaths <- droplevels(gfx_deaths)
Plot number killed or affected by year, similar to the barplot in EMDAT's ADSR report from 2013 (Figure 1, pg. 4 of the ADSR report).
gfx_bar <- ggplot(gfx_deaths, aes(x = Year, y = Total))
gfx_bar <- gfx_bar + geom_bar(position = "dodge", stat = "identity", fill = "blue")
gfx_bar <- gfx_bar + ylab("Reported Victims (in Millions)")
gfx_bar <- gfx_bar + ylim(0, 800)
gfx_bar <- gfx_bar + theme(axis.text.x = element_text(angle = 45, hjust = 1))
gfx_bar <- gfx_bar + geom_text(aes(label = round(Total), hjust = 0.5, vjust = 0), size = 4)
plot(gfx_bar)
Number of events per year from 1990 through 2013.
gfx_events <- as.data.frame(table(nat_data$Year), stringsAsFactors = FALSE)
colnames(gfx_events) <- c("Year", "Total_Events")
gfx_events <- gfx_events[gfx_events$Year >= 1990 & gfx_events$Year <= 2013, ]
gfx_events[gfx_events$Year == 2002, ]
#> Year Total_Events
#> 103 2002 422
Plot number of events by year, similar to the lineplot in EMDAT's ADSR report 2013 (Figure 1, pg. 4 of the ADSR report). Note that the number of events in 2002 were reported to be 428 in the ADSR 2013 report. But the same number in the 2012, 2011, 2010, 2009 and 2008 reports is 428, 421, 421, 422 and 421, respectively!
gfx_line <- ggplot(gfx_events, aes(x = Year, y = Total_Events, group = 1))
gfx_line <- gfx_line + geom_line()
gfx_line <- gfx_line + ylab("Disasters Per Year")
gfx_line <- gfx_line + ylim(0, 500)
gfx_line <- gfx_line + theme(axis.text.x = element_text(angle = 45, hjust = 1))
plot(gfx_line)
In order to replicate the graphic on top 10 countries by loss (Figure 3 and 6, pg. 15-17 of the ADSR report), a generic function is developed below which could not only be used with loss but also other variables.
Fn_Get_Top_Countries <- function(input_df, var_name, plot_title) {
var_vec <- c("Events", "EstDamage", "TotAffected", "Killed")
stopifnot(identical(colnames(input_df), colnames(nat_data)))
stopifnot(var_name %in% var_vec)
fun_name <- "sum"
if (var_name == "Events") {
fun_name <- "length"
var_name <- "Year"
}
# summary by country per natural disaster group
data_by_group <- aggregate(as.formula(paste(var_name, " ~ ISO_cntry + Group")),
data = input_df, FUN = fun_name)
colnames(data_by_group) <- c("Country", "Group", var_name)
# totals by country
data_agg <- aggregate(as.formula(paste(var_name, " ~ ISO_cntry")), data = input_df,
FUN = fun_name)
colnames(data_agg) <- c("Country", "Totals")
data_agg <- data_agg[order(data_agg$Totals, decreasing = TRUE), ]
cntrys_10 <- data_agg$Country[1:10]
# merge above two data frames
out_df <- merge(data_by_group, data_agg, by = "Country")
out_df <- out_df[order(out_df$Totals, decreasing = TRUE), ]
out_df <- out_df[out_df$Country %in% cntrys_10, ]
out_df <- droplevels(out_df)
out_df$Country <- factor(out_df$Country, levels = rev(cntrys_10))
# percentage share
out_df$Pers <- out_df[, var_name] * 100/out_df$Totals
return(out_df)
}
Use the above function to get natural disaster counts by disaster Group for 2013 for the top 10 countries.
nat_2013 <- nat_data[nat_data$Year == 2013, ]
nat_2013 <- droplevels(nat_2013)
gfx_2013_counts <- Fn_Get_Top_Countries(nat_2013, "Events")
head(gfx_2013_counts, 10)
#> Country Group Year Totals Pers
#> 32 China Meteorological 15 42 35.714
#> 33 China Hydrological 14 42 33.333
#> 34 China Geophysical 11 42 26.190
#> 35 China Climatological 2 42 4.762
#> 155 United States Climatological 6 28 21.429
#> 156 United States Meteorological 15 28 53.571
#> 157 United States Hydrological 7 28 25.000
#> 65 Indonesia Hydrological 9 16 56.250
#> 66 Indonesia Geophysical 7 16 43.750
#> 115 Philippines Geophysical 1 14 7.143
Barplot of top 10 countries by number of natural disasters in 2013. Compare with Figure 3, pg. 15 of the ADSR report from 2013.
gfx_bar <- ggplot(gfx_2013_counts, aes(x = Country, y = Year, group = Group))
gfx_bar <- gfx_bar + geom_bar(aes(fill = Group), position = "stack", stat = "identity")
gfx_bar <- gfx_bar + ylab("Number of Events") + xlab(NULL)
gfx_bar <- gfx_bar + coord_flip()
plot(gfx_bar)
Use the above function to get natural disaster losses by disaster Group for 2013 for the top 10 countries.
gfx_2013_losses <- Fn_Get_Top_Countries(nat_2013, "EstDamage")
head(gfx_2013_losses, 10)
#> Country Group EstDamage Totals Pers
#> 15 China Hydrological 16598600 35410900 46.874
#> 16 China Meteorological 10787000 35410900 30.462
#> 17 China Geophysical 8025300 35410900 22.663
#> 23 Germany Hydrological 12900000 17700000 72.881
#> 24 Germany Meteorological 4800000 17700000 27.119
#> 63 United States Hydrological 2275000 17581400 12.940
#> 64 United States Climatological 706400 17581400 4.018
#> 65 United States Meteorological 14600000 17581400 83.042
#> 48 Philippines Meteorological 10136563 12422810 81.596
#> 49 Philippines Hydrological 2234788 12422810 17.989
Pieplot of these top 10 countries. Compare this with Figure 6, pg. 17 of the ADSR report from 2013. Note that the percentage share of the Group is not always the same between these two graphics.
gfx_pie <- ggplot(gfx_2013_losses, aes(x = "", y = Pers, fill = Group))
gfx_pie <- gfx_pie + facet_wrap(~Country)
gfx_pie <- gfx_pie + geom_bar(width = 1, stat = "identity")
gfx_pie <- gfx_pie + coord_polar(theta = "y")
gfx_pie <- gfx_pie + theme(axis.ticks = element_blank(), axis.text.y = element_blank(),
axis.text.x = element_blank())
gfx_pie <- gfx_pie + xlab("") + ylab("")
plot(gfx_pie)
Map 3 of the ADSR Report (see pg. 34) is confusing because the color scheme of the barplots and the color scheme of the continental regions in the map overlap. Below code reproduces the statistics presented in Map 3.
First, compute the regional disaster losses and the percent share of each region within each Group.
gfx_reg1 <- ddply(nat_2013[, c("EstDamage", "Group", "region")],
.(region, Group),
summarize,
tot_by_group = sum(EstDamage, na.rm = TRUE))
gfx_reg2 <- ddply(nat_2013[, c("EstDamage", "Group", "region")],
.(Group),
summarize,
tot_by_reg = sum(EstDamage, na.rm = TRUE))
gfx_reg <- merge(gfx_reg1, gfx_reg2, by = "Group", all.x = TRUE)
gfx_reg$share <- gfx_reg$tot_by_group * 100 / gfx_reg$tot_by_reg
head(gfx_reg)
#> Group region tot_by_group tot_by_reg share
#> 1 Climatological Africa 64000 3159400 2.02570
#> 2 Climatological Americas 1906400 3159400 60.34057
#> 3 Climatological Asia 0 3159400 0.00000
#> 4 Climatological Europe 0 3159400 0.00000
#> 5 Climatological Oceania 1189000 3159400 37.63373
#> 6 Geophysical Americas 4000 9082859 0.04404
Plot percent share of each region within each Group.
gfx_bar <- ggplot(gfx_reg, aes(x = Group, y = share, group = region))
gfx_bar <- gfx_bar + geom_bar(aes(fill = Group), position = "dodge", stat = "identity")
gfx_bar <- gfx_bar + facet_wrap(~region, scales = "free_y")
gfx_bar <- gfx_bar + ylab("Percent Share") + xlab(NULL)
gfx_bar <- gfx_bar + theme(axis.text.x = element_blank(), axis.ticks.x = element_blank())
plot(gfx_bar)
