Impact of U.S. Weather Events 2002-2011
David M. Leonard
April 9, 2017
Synopsis
The National Oceanic and Atmospheric Agency (NOAA) maintains records of weather events in the United States and their estimated impact in terms of human casualties and economic costs. This analysis looks at those records over the 10 year period from 2002 through the end of 2011, and answers the questions: what type of weather event caused the greatest human casualties, and what events had the highest economic costs?
The analysis shows that tornadoes were responsible for the highest casualties, in terms of both injuries, and fatalities. From an economic perspective, the most costly type of weather event was flooding, which caused significant property damage as well as some crop damage. However, the most significant crop damage was caused by droughts.
Data Processing
The NOAA weather event data that was used was provided in a compressed file format. It contains many fields that describe each weather event, but for the purposes of this analysis, only the following fields were considered:
- EVTYPE (Event Type): Describes the type of the particular event, such as Flood or Tornado
- BGN_DATE: the date on which the given event began
- FATALITIES: the number of deaths resulting from the event
- INJURIES: the number of non-fatal injuries caused by the event
- PROPDMG: the cost, in U.S. dollars, of property damage caused by the event
- PROPDMGEXP: a multiplier for the property damage cost value (for example, “K” indicates that the property damage value should be multiplied by one thousand, “M” indicates one million)
- CROPDMG: the cost, in U.S. dollars, of the crop damage caused by the event
- CROPDMGEXP: the multiplier to be used for the crop damage cost value
After downloading the file from the specified website, it was loaded into an R data frame. The following preliminary processing of the data was then performed:
- The year in which each weather event occurred was extracted from the field BGN_DATE and saved in a new field
- The fields CROPDMGEXP and PROPDMGEXP were converted from text symbolic values into numeric values that could be multiplied directly against the crop and property damage costs to determine the actual cost. This mapping was based on using the NOAA interactive site to look up entries and compare values to the appropriate exponent mapping.
- Two new fields were created that hold the actual costs from property and crop damage.
- Finally, a new table that summarizes data by event type over the period from 2002 to 2011 was created; this was used to generate charts that illustrate the findings.
library(dplyr, quietly = T)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionlibrary(ggplot2)
library(RColorBrewer)
library(knitr)# Read in the NOAA Storm Data
url <- "https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2"
bz2 <- "NOAA_Storm_Data.csv.bz2"
csv <- "NOAA_Storm_Data.csv"
download.file(url, bz2)
df <- read.csv(bz2, stringsAsFactors = FALSE) # read.csv knows how to unzip bz2 files...
df$BGN_YEAR <- substring(df$BGN_DATE,1,regexpr(" ", df$BGN_DATE)-1)
df$BGN_YEAR <- substring(df$BGN_YEAR,nchar(df$BGN_YEAR)-3,10)
df$EVTYPE <- as.factor(df$EVTYPE)
# Create a table of multipliers for the exponent column
exptbl <- data.frame(explabel = (c("k","K","m","M","b","B","0","1","2","3","4","5","6","7","8","+","?","h","H","-","")),
propmultiplier = c(1e+03,1e+03,1e+06,1e+06,1e+09,1e+09,10,10,10,10,10,10,10,10,10,1,0,100,100,0,0))
exptbl$explabel = as.character(exptbl$explabel)
# Add a column with the numeric value of the property damage multiplier
df <- df %>%
inner_join(exptbl, by = c("PROPDMGEXP" = "explabel"))
names(exptbl) <- c("explabel","cropmultiplier")
# Now do the same for the crop damage multiplier
df <- df %>%
inner_join(exptbl, by = c("CROPDMGEXP" = "explabel"))
# Create new columns that have the crop damage and property damage costs
df$cropdmgcost <- df$CROPDMG*df$cropmultiplier
df$propdmgcost <- df$PROPDMG*df$propmultiplier
# Now write out the dataframe
#write.csv(df,"dfdata.csv")
#df <- read.csv("dfdata.csv", stringAsFactors = F)
# Ready to build a summary table from which we will gather all needed statistics
evtsum <- df %>%
filter(BGN_YEAR > "2001") %>%
group_by(EVTYPE) %>%
summarize(occurrences = length(EVTYPE),
casualties = sum(FATALITIES+INJURIES),
fatalities = sum(FATALITIES),
injuries = sum(INJURIES),
casperevt = sum(FATALITIES+INJURIES)/length(EVTYPE),
fatperevt = sum(FATALITIES)/length(EVTYPE),
cost = sum(cropdmgcost) + sum(propdmgcost),
propcost = sum(propdmgcost),
cropcost = sum(cropdmgcost),
costperevt = (sum(cropdmgcost) + sum(propdmgcost))/length(EVTYPE))
totcas <- sum(evtsum$casualties)
totinj <- sum(evtsum$injuries)
totfat <- sum(evtsum$fatalities)
totfat <- sum(evtsum$fatalities)
totcost <- sum(evtsum$cost)
totpropcost <- sum(evtsum$propcost)
totcropcost <- sum(evtsum$cropcost)Results
In total, weather events resulted in 29,609 injuries, and 5,048 fatalities between 2002 and 2011. During the same time period, weather events resulted in property damage totaling $315B, and crop damage of $18.5B.
Events Resulting In Greatest Numbers of Casualties
The chart below summarizes the weather events that resulted in the greatest human casualties over the 10-year period from 2002 to 2011.
# Build table of top causes of casualties
maxcas <- evtsum %>%
select(EVTYPE, injuries, fatalities, casualties) %>%
arrange(desc(casualties))
maxcas$pct <- maxcas$casualties/totcas
maxcas$cumpct <- maxcas$pct
# Get the cumulative percent
for (i in 2:length(maxcas$EVTYPE)) {
maxcas$cumpct[i] <- maxcas$cumpct[i-1]+maxcas$pct[i]}
# Limit the table to entries representing at least 90% of all casualties or at least 10 events
maxcas <- maxcas %>%
filter(cumpct <= 0.90 | row_number(cumpct) < 11)
topevt <- maxcas$EVTYPE[1]
topinj <- maxcas$injuries[1]
topfat <- maxcas$fatalities[1]
par(oma = c(6,1,1,1))
mcbp <- barplot(t(maxcas[,2:3]),
col = brewer.pal(3, "Paired"),
main = "Weather Events Accounting for Top 90% of Casualties, 2002-2011",
ylab = "Number of Casualties",
legend = colnames(maxcas[,2:3]),
xaxt = "n",
yaxt = "n")
axis(1,at = mcbp, labels = maxcas$EVTYPE, las = 3, cex.axis = .9)
axis(2,at = axTicks(2), labels = formatC(axTicks(2), format = "d", big.mark = ","))
As indicated in the chart, tornadoes were the largest cause of casualties, resulting in 13,588 injuries (45% of total) and 1,112 deaths (22%) over a 10 year span.
The table below gives statistics for the top events.
maxcasd <- maxcas
maxcasd$injuries <- formatC(maxcas$injuries, format = "d", big.mark = ",")
maxcasd$fatalities <- formatC(maxcas$fatalities, format = "d", big.mark = ",")
maxcasd$casualties <- formatC(maxcas$casualties, format = "d", big.mark = ",")
maxcasd$pct = paste(formatC(maxcas$pct*100, format = "f", digits = 1),"%", sep = "")
maxcasd$cumpct = paste(formatC(maxcas$cumpct*100, format = "f", digits = 1),"%", sep = "")
kable(maxcasd,
table.attr='class="myTable"',
col.names = c("Event","Injuries","Fatalities","Casualties", "% Casualties", "Cum % Casualties"),
align = c("l","r","r","r","c","c"),
format = "html",
row.names = "F",
caption = "Events with highest casualties (injuries + fatalities), in sorted order")| Event | Injuries | Fatalities | Casualties | % Casualties | Cum % Casualties |
|---|---|---|---|---|---|
| TORNADO | 13,588 | 1,112 | 14,700 | 42.4% | 42.4% |
| EXCESSIVE HEAT | 2,797 | 691 | 3,488 | 10.1% | 52.5% |
| LIGHTNING | 2,250 | 370 | 2,620 | 7.6% | 60.0% |
| THUNDERSTORM WIND | 1,400 | 130 | 1,530 | 4.4% | 64.5% |
| HEAT | 1,222 | 229 | 1,451 | 4.2% | 68.6% |
| HURRICANE/TYPHOON | 1,275 | 64 | 1,339 | 3.9% | 72.5% |
| TSTM WIND | 1,146 | 77 | 1,223 | 3.5% | 76.0% |
| FLASH FLOOD | 517 | 539 | 1,056 | 3.0% | 79.1% |
| WILDFIRE | 911 | 75 | 986 | 2.8% | 81.9% |
| HIGH WIND | 486 | 97 | 583 | 1.7% | 83.6% |
| FLOOD | 301 | 247 | 548 | 1.6% | 85.2% |
| RIP CURRENT | 208 | 340 | 548 | 1.6% | 86.8% |
| HAIL | 456 | 3 | 459 | 1.3% | 88.1% |
| WINTER WEATHER | 343 | 33 | 376 | 1.1% | 89.2% |
Events Resulting In Highest Damage Costs
The chart below shows the weather events that had the greatest economic impact over the 10-year period.
# Build table of costs
maxcost <- evtsum %>%
select(EVTYPE,
Property_Damage = propcost,
Crop_Damage = cropcost,
cost,
Cost_Per_Event = cost/occurrences) %>%
arrange(desc(cost))
maxcost$pct <- maxcost$cost/totcost
maxcost$cumpct <- maxcost$pct
# Get the cumulative percent
for (i in 2:length(maxcost$EVTYPE)) {
maxcost$cumpct[i] <- maxcost$cumpct[i-1]+maxcost$pct[i]}
# Limit table to at least 10 entries that represents at least 90% of all costs
maxcost <- maxcost %>%
filter(cumpct <= 0.90 | row_number(cumpct) < 11)
costchartcumpct <- max(maxcost$cumpct)
topevt <- maxcost$EVTYPE[1]
topcost <- maxcost$cost[1]
topprop <- maxcost$Property_Damage[1]
topproppct <- topprop/topcost
topcropcost <- max(maxcost$Crop_Damage)
topcostperocc <- max(maxcost$Cost_Per_Event)
topcropevt <- maxcost[maxcost$Crop_Damage == topcropcost, "EVTYPE"][1]
topcostperoccevt <- maxcost[maxcost$Cost_Per_Event == topcostperocc, "EVTYPE"][1]
par(oma = c(6,1,1,1))
mcbp <- barplot(t(maxcost[,2:3]),
col = brewer.pal(3, "Paired"),
main = paste("Weather Events Accounting for Top ",
formatC(round(costchartcumpct*100), format = "d"),
"% of Damage Costs, 2002-2011",
sep=""),
ylab = "Cost of Damage ($ Billions)",
legend = colnames(maxcost[,2:3]),
xaxt = "n",
yaxt = "n")
axis(1,at = mcbp, labels = maxcost$EVTYPE, las = 3, cex.axis = .9)
axis(2,at = axTicks(2), labels = formatC(axTicks(2)/1000000000, format = "f", big.mark = ",", digits = 0))
The top 10 weather events accounted for 94% of all costs. Floods represented 41% of all weather event-related damages during the period, with total costs of $137.0B, of which 97% was property damage. The largest source of crop damage costs was the event labeled “DROUGHT”, representing 29% of all such costs, for a total value of $5.4B.
The table below summarizes the cost statistics for the top causes of damage.
maxcostd <- maxcost
maxcostd$Property_Damage <- paste("$",formatC(maxcostd$Property_Damage/1000000000,
format = "f",
digits = 1,
big.mark = ","),
sep = "")
maxcostd$Crop_Damage <- paste("$",formatC(maxcostd$Crop_Damage/1000000000,
format = "f",
digits = 1,
big.mark = ","),
"B",
sep = "")
maxcostd$cost <- paste("$",formatC(maxcostd$cost/1000000000,
format = "f",
digits = 1,
big.mark = ","),
"B",
sep = "")
maxcostd$Cost_Per_Event <- paste("$",formatC(maxcostd$Cost_Per_Event,
format = "f",
digits = 0,
big.mark = ","),
sep = "")
maxcostd$pct = paste(formatC(maxcostd$pct*100, format = "f", digits = 1),"%", sep = "")
maxcostd$cumpct = paste(formatC(maxcostd$cumpct*100, format = "f", digits = 1),"%", sep = "")
kable(maxcostd,
table.attr='class="myTable"',
col.names = c("Event",
"Property Damage",
"Crop Damage",
"Total Damage",
"Cost per Occurrence",
"% Damage",
"Cum % Damage"),
align = c("l","r","r","r","r","c","c"),
format = "html",
row.names = "F",
caption = "Events with highest economic cost (property damage + crop damage), in sorted order")| Event | Property Damage | Crop Damage | Total Damage | Cost per Occurrence | % Damage | Cum % Damage |
|---|---|---|---|---|---|---|
| FLOOD | $133.4 | $3.6B | $137.0B | $247 | 41.1% | 41.1% |
| HURRICANE/TYPHOON | $69.3 | $2.6B | $71.9B | $64 | 21.6% | 62.6% |
| STORM SURGE | $43.2 | $0.0B | $43.2B | $0 | 12.9% | 75.6% |
| TORNADO | $18.4 | $0.2B | $18.6B | $1,112 | 5.6% | 81.1% |
| FLASH FLOOD | $10.7 | $0.8B | $11.5B | $539 | 3.5% | 84.6% |
| HAIL | $9.2 | $1.4B | $10.6B | $3 | 3.2% | 87.8% |
| DROUGHT | $0.8 | $5.4B | $6.3B | $0 | 1.9% | 89.6% |
| HIGH WIND | $4.8 | $0.5B | $5.3B | $97 | 1.6% | 91.2% |
| WILDFIRE | $4.8 | $0.3B | $5.1B | $75 | 1.5% | 92.8% |
| STORM SURGE/TIDE | $4.6 | $0.0B | $4.6B | $11 | 1.4% | 94.1% |
The event with the highest cost per occurrence was TORNADO, with a per-occurrence cost of $1,112. Although floods are the most costly in total, they have a lower cost per occurrence than either tornadoes or flash floods, which strike quickly and with little warning.
Conclusions
Tornadoes are by far the most dangerous type of weather event, in terms of both injuries and fatalities. This can be attributed to the fact that they strike quickly, with little warning, and have immense energy, thus making escape more challenging.
In terms of economic damage, on the other hand, floods were more costly than any other single event type; the cost per occurrence was lower than events like tornadoes, but they occurr with greater frequency. Perhaps not surprisingly, the greatest crop damage was due to droughts.
==========================================================================================