Among extreme weather events, floods cause the most economic damage, while tornados are the most dangerous.
Synopsis
We examine National Weather Service Storm Data to discover what types of extreme weather events are responsible for the most economic damage (measured in property damage and drop damage done), and what types are the most dangerous (measured in fatalities and injuries attributed to the storm). We find that floods are clearly responsible for the most economic damage, followed by hurricanes, tornados, and storm surges. Tornados, meanwhile, are the most dangerous natural weather events, followed by heat, floods, wind, and lightning.
Processing
We first clean the data to prepare for analysis.
We first discard unwanted columns, as well as data which has nonsensical entries in either of the “exponent” columns.
importCols <- rep("NULL", 37)
importCols[c(8,23:28)] <- NA
download.file(url = 'https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2',
destfile = "weather_data.csv.bz2", method="curl")
weatherData <- read.csv("weather_data.csv.bz2", colClasses = importCols)
exps <- c('h','H','k','K','m','M','b','B')
cleanData <- weatherData[((weatherData$PROPDMG == 0 | weatherData$PROPDMGEXP %in% exps)
& (weatherData$CROPDMG == 0 | weatherData$CROPDMGEXP %in% exps)), ]
multVals <- c(100, 100, 1000, 1000, 1000000, 1000000, 1000000000, 1000000000, 0)
multUnits <- c('h','H','k','K','m','M','b','B','NOMATCH')
multLookup <- cbind(multUnits, multVals)
cleanData$PROPDMG <- as.numeric(cleanData$PROPDMG) *
multVals[match(cleanData$PROPDMGEXP, multUnits, nomatch = 9)] / 1000
cleanData$CROPDMG <- as.numeric(cleanData$CROPDMG) *
multVals[match(cleanData$CROPDMGEXP, multUnits, nomatch = 9)] / 1000
cleanData$FATALITIES <- as.numeric(cleanData$FATALITIES)
cleanData$INJURIES <- as.numeric(cleanData$INJURIES)
cleanData$EVTYPE <- toupper(as.character(cleanData$EVTYPE))
cleanData$DMG <- cleanData$PROPDMG + cleanData$CROPDMGNote that at this point, dollar values are in units of thousands of dollars.
It will be useful for us to combine many event types that mean essentially the same types of events. Note that we do not combine ALL similar event types, as we will throw many out at a later stage. We simply combine similar event types that might end up appearing at the last stage of our analysis.
blizzardType <- c('BLIZZARD')
cleanData$EVTYPE[cleanData$EVTYPE %in% blizzardType] <- 'BLIZZARD'
floodType <- c('COASTAL FLOOD', 'FLASH FLOOD', 'FLASH FLOOD/FLOOD', 'FLASH FLOODING',
'RIVER FLOOD', 'FLOOD/FLASH FLOOD', 'URBAN/SML STREAM FLD',
'FLOODING')
cleanData$EVTYPE[cleanData$EVTYPE %in% floodType] <- 'FLOOD'
freezeType <- c('DAMAGING FREEZE', 'FREEZE', 'FROST/FREEZE')
cleanData$EVTYPE[cleanData$EVTYPE %in% freezeType] <- 'FREEZE'
droughtType <- c('DROUGHT')
cleanData$EVTYPE[cleanData$EVTYPE %in% droughtType] <- 'DROUGHT'
heatType <- c('EXCESSIVE HEAT', 'HEAT', 'HEAT WAVE', 'EXTREME HEAT',
'UNSEASONABLY WARM AND DRY', 'RECORD/EXCESSIVE HEAT',
'UNSEASONABLY WARM')
cleanData$EVTYPE[cleanData$EVTYPE %in% heatType] <- 'HEAT'
coldType <- c('EXTREME COLD', 'EXTREME COLD/WIND CHILL', 'COLD/WIND CHILL',
'EXTREME WINDCHILL', 'COLD AND SNOW', 'HYPOTHERMIA/EXPOSURE',
'LOW TEMPERATURE')
cleanData$EVTYPE[cleanData$EVTYPE %in% coldType] <- 'COLD'
hailType <- c('HAIL', 'HAILSTORM')
cleanData$EVTYPE[cleanData$EVTYPE %in% hailType] <- 'HAIL'
rainType <- c('HEAVY RAIN', 'HEAVY RAIN/SEVERE WEATHER', 'FREEZING RAIN')
cleanData$EVTYPE[cleanData$EVTYPE %in% rainType] <- 'RAIN'
snowType <- c('HEAVY SNOW')
cleanData$EVTYPE[cleanData$EVTYPE %in% snowType] <- 'SNOW'
windType <- c('HIGH WIND', 'HIGH WINDS', 'STRONG WIND', 'THUNDERSTORM WIND',
'THUNDERSTORM WINDS', 'TSTM WIND', 'TSTM WIND/HAIL',
'MARINE STRONG WIND', 'MARINE THUNDERSTORM WIND',
'MARINE TSTM WIND', 'STRONG WINDS')
cleanData$EVTYPE[cleanData$EVTYPE %in% windType] <- 'WIND'
hurricaneType <- c('HURRICANE', 'HURRICANE ERIN', 'HURRICANE OPAL',
'HURRICANE/TYPHOON')
cleanData$EVTYPE[cleanData$EVTYPE %in% hurricaneType] <- 'HURRICANE'
iceType <- c('ICE STORM')
cleanData$EVTYPE[cleanData$EVTYPE %in% iceType] <- 'ICE'
landslideType <- c('LANDSLIDE')
cleanData$EVTYPE[cleanData$EVTYPE %in% landslideType] <- 'LANDSLIDE'
lightningType <- c('LIGHTNING')
cleanData$EVTYPE[cleanData$EVTYPE %in% lightningType] <- 'LIGHTNING'
thunderstormType <- c('SEVERE THUNDERSTORM')
cleanData$EVTYPE[cleanData$EVTYPE %in% thunderstormType] <- 'THUNDERSTORM'
surgeType <- c('STORM SURGE', 'STORM SURGE/TIDE')
cleanData$EVTYPE[cleanData$EVTYPE %in% surgeType] <- 'STORM SURGE'
tornadoType <- c('TORNADO', 'TORNADOES, TSTM WIND, HAIL')
cleanData$EVTYPE[cleanData$EVTYPE %in% tornadoType] <- 'TORNADO'
tropicalType <- c('TROPICAL STORM', 'TROPICAL STORM GORDON')
cleanData$EVTYPE[cleanData$EVTYPE %in% tropicalType] <- 'TROPICAL STORM'
typhoonType <- c('TYPHOON')
cleanData$EVTYPE[cleanData$EVTYPE %in% typhoonType] <- 'TYPHOON'
wildfireType <- c('WILDFIRE', 'WILD FIRES', 'WILD/FOREST FIRE')
cleanData$EVTYPE[cleanData$EVTYPE %in% wildfireType] <- 'WILDFIRE'
winterstormType <- c('WINTER STORM', 'WINTER WEATHER', 'WINTER WEATHER/MIX',
'WINTER STORMS')
cleanData$EVTYPE[cleanData$EVTYPE %in% winterstormType] <- 'WINTER STORM'
fogType <- c('DENSE FOG', 'FOG')
cleanData$EVTYPE[cleanData$EVTYPE %in% fogType] <- 'FOG'
ripType <- c('RIP CURRENT', 'RIP CURRENTS')
cleanData$EVTYPE[cleanData$EVTYPE %in% ripType] <- 'RIP CURRENTS'
surfType <- c('HEAVY SURF/HIGH SURF', 'HIGH SURF', 'HEAVY SURF')
cleanData$EVTYPE[cleanData$EVTYPE %in% surfType] <- 'HIGH SURF'Results
Economic Costs of Weather Events
For damage, we will sum up property damage and crop damage and simply compare the total amount of damage caused by each event type.
dmgData <- aggregate(cleanData$DMG, by=list(cleanData$EVTYPE), FUN=sum, na.rm=TRUE)
names(dmgData) <- c('EVTYPE', 'DMG')
dmgData$DMG <- as.numeric(dmgData$DMG)We are interested in the events that cause the most damage. Simply finding the max might be good enough, but looking at the different event types shows that the same types of event are sometimes coded differently due to typos or simply variation in style. We will therefore try to add together events that are actually essentially the same.
Instead of examining every event type, we discard those types which have done total damage less than one one-thousandth of the maximum. These event types will not contribute enough to change the final analysis, as we shall see.
maxDmgData <- dmgData[dmgData$DMG >= max(dmgData$DMG) / 1000, ]Before we examine the results, we add the damage that we just dropped, all added together under the “DROPPED” label, to make sure that we did not remove pertinent information.
missingEventsDmg <- sum(dmgData$DMG[dmgData$DMG < max(dmgData$DMG) / 1000])
maxDmgData <- rbind(maxDmgData, c('DROPPED', missingEventsDmg))
maxDmgSortedData <- maxDmgData[order(as.numeric(maxDmgData$DMG),
decreasing = TRUE),]
maxDmgSortedData## EVTYPE DMG
## 133 FLOOD 179337997.82
## 329 HURRICANE 90109897.81
## 707 TORNADO 58904440.59
## 553 STORM SURGE 47965579
## 186 HAIL 18974211.67
## 820 WIND 17787622.56
## 71 DROUGHT 15018672
## 339 ICE 8979692.81
## 817 WILDFIRE 8793313.13
## 720 TROPICAL STORM 8383236.55
## 832 WINTER STORM 6758679.25
## 443 RAIN 3941844.44
## 22 DROPPED 2238313.77
## 150 FREEZE 1865826
## 48 COLD 1410307.4
## 630 THUNDERSTORM 1207480.55
## 517 SNOW 1082079.79
## 374 LIGHTNING 940751.37
## 216 HEAT 924549.25
## 20 BLIZZARD 771273.95
## 749 TYPHOON 601055
## 360 LANDSLIDE 344613Note that the total amount of damage done by dropped events is much smaller than the damage done by the most economically impactful events, and therefore will not affect the analysis of which events cause the most economic damage.
We now plot the damage done by event types that do the most damage. Flooding clearly causes the most damage, with hurricanes, tornados, and storm surges following. Beyond those four types of events, relative economic impacts of weather types becomes less clear from our analysis.
par(mar=c(10,8,4,2))
barplot(as.numeric(maxDmgSortedData$DMG[1:15])/1000000, las = 2, ps = 7,
names.arg = maxDmgSortedData$EVTYPE[1:15], ylim = c(0,200),
main = "Economic Damage done by Weather Events",
ylab = "Economic Damage (in billions of dollars)")
Human Costs of Weather Events
We cannot justify simply summing deaths and injuries caused by extremem weather events, so we will analyze these statistics separately.
fatalityData <- aggregate(cleanData$FATALITIES, by=list(cleanData$EVTYPE), FUN=sum,
na.rm=TRUE)
injuryData <- aggregate(cleanData$INJURIES, by=list(cleanData$EVTYPE), FUN=sum,
na.rm=TRUE)
names(fatalityData) <- c('EVTYPE', 'FATALITIES')
names(injuryData) <- c('EVTYPE', 'INJURIES')
fatalityData$FATALITIES <- as.numeric(fatalityData$FATALITIES)
injuryData$INJURIES <- as.numeric(injuryData$INJURIES)As with economic information, we are interested in which type of event is most dangerous to our health. We can therefore discard events as in the analysis of the economic question.
maxInjData <- injuryData[injuryData$INJURIES >= max(injuryData$INJURIES) / 1000, ]
maxFatData <- fatalityData[fatalityData$FATALITIES >=
max(fatalityData$FATALITIES) / 1000, ]missingEventsInj <- sum(injuryData$INJURIES[injuryData$INJURIES <
max(injuryData$INJURIES) / 1000])
missingEventsFat <- sum(fatalityData$FATALITIES[fatalityData$FATALITIES <
max(fatalityData$FATALITIES) / 1000])
maxInjData <- rbind(maxInjData, c('DROPPED', missingEventsInj))
maxFatData <- rbind(maxFatData, c('DROPPED', missingEventsFat))
maxInjSortedData <- maxInjData[order(as.numeric(maxInjData$INJURIES),
decreasing = TRUE),]
maxFatSortedData <- maxFatData[order(as.numeric(maxFatData$FATALITIES),
decreasing = TRUE),]
maxInjSortedData## EVTYPE INJURIES
## 707 TORNADO 91322
## 820 WIND 11327
## 216 HEAT 9176
## 133 FLOOD 8674
## 374 LIGHTNING 5228
## 339 ICE 2112
## 832 WINTER STORM 1808
## 817 WILDFIRE 1606
## 186 HAIL 1360
## 329 HURRICANE 1323
## 147 FOG 1076
## 517 SNOW 1052
## 23 DROPPED 924
## 20 BLIZZARD 805
## 481 RIP CURRENTS 529
## 93 DUST STORM 440
## 720 TROPICAL STORM 383
## 48 COLD 320
## 443 RAIN 274
## 277 HIGH SURF 244
## 169 GLAZE 216
## 11 AVALANCHE 170
## 747 TSUNAMI 129maxFatSortedData## EVTYPE FATALITIES
## 707 TORNADO 5655
## 216 HEAT 3165
## 133 FLOOD 1537
## 820 WIND 1154
## 374 LIGHTNING 816
## 481 RIP CURRENTS 572
## 48 COLD 465
## 832 WINTER STORM 277
## 11 AVALANCHE 224
## 27 DROPPED 166
## 277 HIGH SURF 154
## 329 HURRICANE 132
## 517 SNOW 130
## 443 RAIN 105
## 20 BLIZZARD 101
## 339 ICE 95
## 817 WILDFIRE 90
## 147 FOG 80
## 720 TROPICAL STORM 66
## 360 LANDSLIDE 38
## 747 TSUNAMI 33
## 553 STORM SURGE 24
## 93 DUST STORM 22
## 186 HAIL 15
## 489 ROUGH SEAS 8
## 169 GLAZE 7
## 402 MARINE MISHAP 7par(mar=c(10,8,4,2))
barplot(as.numeric(maxInjSortedData$INJURIES[1:15]), las = 2, ps = 7,
names.arg = maxInjSortedData$EVTYPE[1:15],
main = "Injuries by Weather Events", ylim = c(0,100000),
ylab = "Injuries")
barplot(as.numeric(maxFatSortedData$FATALITIES[1:15]), las = 2, ps = 7,
names.arg = maxFatSortedData$EVTYPE[1:15],
main = "Fatalities by Weather Events", ylim = c(0,6000),
ylab = "Fatalities")
We see from the above graphs that tornados are by far the most dangerous natural weather events, followed by heat, floods, wind, and lightning, in approximately that order (that order does not quite match the injury statistics, those four events cause similar numbers of injuries, and as fatalities are more serious, the fatality statistics ought to be given more weight in the analysis). After these five types of events, analysis becomes difficult with our methods.