Effects of Severe Weather Events on Population Health, Property, and Crops
Katherine Vance
May 2, 2020
Synopsis
In this report, we attempt to determine which types of severe weather events have the most significant impact on population health and the economy. To do this, we use the U.S. National Oceanic and Atmospheric Administration’s (NOAA) storm database. In order to assess the effect of severe weather events on population health, we use the NOAA storm data to determine which types of weather events have caused the greatest number of injuries and fatalities. In order to determine which event types have the greatest economic consequences, we use the NOAA storm data to determine which types of events have caused the most crop and property damage. We find that tornados have caused the largest number of injuries and fatalities among types of severe weather events, and that hurricanes and typhoons are most likely to cause significant amounts of crop and property damage.
Data Processing
We start by reading in the data directly from the zipped .csv.bz2 file. Ensure that the Storm Data file is in the working directory.
StormData <- read.csv("repdata_data_StormData.csv.bz2", na.strings = "")We will be investigating two questions: one about the effects of different types of severe weather events on population health, and a second about the economic consequences of different types of severe weather events. We’ll subset the data to create a separate dataframe for each question, containing only the columns relevant to that question.
PopStormData <- subset(StormData, select = c(EVTYPE, FATALITIES, INJURIES))Next, we check to see if our subsetted PopStormData dataframe has any missing values.
sum(is.na(PopStormData$EVTYPE))## [1] 0sum(is.na(PopStormData$FATALITIES))## [1] 0sum(is.na(PopStormData$INJURIES))## [1] 0summary(PopStormData$EVTYPE)## HAIL TSTM WIND THUNDERSTORM WIND
## 288661 219940 82563
## TORNADO FLASH FLOOD FLOOD
## 60652 54277 25326
## THUNDERSTORM WINDS HIGH WIND LIGHTNING
## 20843 20212 15754
## HEAVY SNOW HEAVY RAIN WINTER STORM
## 15708 11723 11433
## WINTER WEATHER FUNNEL CLOUD MARINE TSTM WIND
## 7026 6839 6175
## MARINE THUNDERSTORM WIND WATERSPOUT STRONG WIND
## 5812 3796 3566
## URBAN/SML STREAM FLD WILDFIRE BLIZZARD
## 3392 2761 2719
## DROUGHT ICE STORM EXCESSIVE HEAT
## 2488 2006 1678
## HIGH WINDS WILD/FOREST FIRE FROST/FREEZE
## 1533 1457 1342
## DENSE FOG WINTER WEATHER/MIX TSTM WIND/HAIL
## 1293 1104 1028
## EXTREME COLD/WIND CHILL HEAT HIGH SURF
## 1002 767 725
## TROPICAL STORM FLASH FLOODING EXTREME COLD
## 690 682 655
## COASTAL FLOOD LAKE-EFFECT SNOW FLOOD/FLASH FLOOD
## 650 636 624
## LANDSLIDE SNOW COLD/WIND CHILL
## 600 587 539
## FOG RIP CURRENT MARINE HAIL
## 538 470 442
## DUST STORM AVALANCHE WIND
## 427 386 340
## RIP CURRENTS STORM SURGE FREEZING RAIN
## 304 261 250
## URBAN FLOOD HEAVY SURF/HIGH SURF EXTREME WINDCHILL
## 249 228 204
## STRONG WINDS DRY MICROBURST ASTRONOMICAL LOW TIDE
## 196 186 174
## HURRICANE RIVER FLOOD LIGHT SNOW
## 174 173 154
## STORM SURGE/TIDE RECORD WARMTH COASTAL FLOODING
## 148 146 143
## DUST DEVIL MARINE HIGH WIND UNSEASONABLY WARM
## 141 135 126
## FLOODING ASTRONOMICAL HIGH TIDE MODERATE SNOWFALL
## 120 103 101
## URBAN FLOODING WINTRY MIX HURRICANE/TYPHOON
## 98 90 88
## FUNNEL CLOUDS HEAVY SURF RECORD HEAT
## 87 84 81
## FREEZE HEAT WAVE COLD
## 74 74 72
## RECORD COLD ICE THUNDERSTORM WINDS HAIL
## 64 61 61
## TROPICAL DEPRESSION SLEET UNSEASONABLY DRY
## 60 59 56
## FROST GUSTY WINDS THUNDERSTORM WINDSS
## 53 53 51
## MARINE STRONG WIND OTHER SMALL HAIL
## 48 48 47
## FUNNEL FREEZING FOG THUNDERSTORM
## 46 45 45
## Temperature record TSTM WIND (G45) Coastal Flooding
## 43 39 38
## WATERSPOUTS MONTHLY PRECIPITATION WINDS
## 37 36 36
## (Other)
## 2940summary(PopStormData$FATALITIES)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.0000 0.0000 0.0000 0.0168 0.0000 583.0000summary(PopStormData$INJURIES)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.0000 0.0000 0.0000 0.1557 0.0000 1700.0000It appears that the data set has no missing values in the columns we’ll be using to answer the question of which event types are most harmful to population health.
Next, we subset StormData to create a dataframe containing the columns relevant to answering the question of which severe weather event types have the greatest economic consequences.
EconStormData <- subset(StormData, select = c(EVTYPE, PROPDMG, PROPDMGEXP, CROPDMG, CROPDMGEXP))Next, we check to see if our subsetted EconStormData dataframe has any missing values. The property and crop damage amounts are reported in scientific notation, where PROPDMGEXP and CROPDMGEXP give the order of magnitude of the property damage and crop damage amounts, respectively.
sum(is.na(EconStormData$EVTYPE))## [1] 0sum(is.na(EconStormData$PROPDMG))## [1] 0sum(is.na(EconStormData$PROPDMGEXP))## [1] 465934sum(is.na(EconStormData$CROPDMG))## [1] 0sum(is.na(EconStormData$CROPDMGEXP))## [1] 618413summary(EconStormData$PROPDMG)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.00 0.00 0.00 12.06 0.50 5000.00summary(EconStormData$CROPDMG)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.000 0.000 0.000 1.527 0.000 990.000summary(EconStormData$PROPDMGEXP)## - ? + 0 1 2 3 4 5 6
## 1 8 5 216 25 13 4 4 28 4
## 7 8 B h H K m M NA's
## 5 1 40 1 6 424665 7 11330 465934summary(EconStormData$CROPDMGEXP)## ? 0 2 B k K m M NA's
## 7 19 1 9 21 281832 1 1994 618413The only columns with missing values are PROPDMGEXP and CROPDMGEXP, and there are a lot of missing values: 51.6386511% and 68.5376323%, respectively. We assume that entries whose PROPDMGEXP or CROPDMGEXP are missing already have their PROPDMG or CROPDMG entry in decimal notation. We will remove entries with PROPDMGEXP or CROPDMGEXP of “-”, “?”, and “+”. We will convert the letter entries (“h” or “H” for “hundred”, “k” or “K” for “thousand”, “m” or “M” for “million”, and “B” for “billion”) to numerical exponents, so that all entries in the column have the same formatting. Then we will create new columns containing the dollar amounts of property and crop damage amounts, respectively.
EconStormData <- EconStormData %>%
mutate(
cleanPROPDMGEXP = ifelse(is.na(PROPDMGEXP), 0,
ifelse(PROPDMGEXP %in% c("h", "H"), 2,
ifelse(PROPDMGEXP == "K", 3,
ifelse(PROPDMGEXP %in% c("m", "M"), 6,
ifelse(PROPDMGEXP == "B", 9,
ifelse(PROPDMGEXP %in% c("-", "?", "+"), "remove",
as.character(PROPDMGEXP)
)
)
)
)
)
)
) %>%
filter(cleanPROPDMGEXP != "remove") %>%
mutate(cleanPROPDMGEXP = as.numeric(cleanPROPDMGEXP)) %>%
mutate(cleanPROPDMG = PROPDMG * 10^cleanPROPDMGEXP)EconStormData <- EconStormData %>%
mutate(
cleanCROPDMGEXP = ifelse(is.na(CROPDMGEXP), "0",
ifelse(CROPDMGEXP %in% c("h", "H"), 2,
ifelse(CROPDMGEXP %in% c("k", "K"), 3,
ifelse(CROPDMGEXP %in% c("m", "M"), 6,
ifelse(CROPDMGEXP == "B", 9,
ifelse(CROPDMGEXP %in% c("-", "?", "+"), "remove",
as.character(CROPDMGEXP)
)
)
)
)
)
)
) %>%
filter(cleanCROPDMGEXP != "remove") %>%
mutate(cleanCROPDMGEXP = as.numeric(cleanCROPDMGEXP)) %>%
mutate(cleanCROPDMG = CROPDMG * 10^cleanCROPDMGEXP)Results
Population Health
In order to assess the effect of severe weather events on population health, we’ll create a new variable that combines the two ways the dataset measures the impact of weather events on the population: number of injuries and number of fatalities. Our new variable will be their sum. We then use this new variable to identify the ten severe weather event types with the largest total sum of injuries and fatalities over all events in the dataset.
Because the number of event types is unmanageably large and has numerous obvious coding errors, and because I don’t know enough about the dataset to fix those coding errors in an intelligent way, I will not attempt to combine apparent duplicate event types. Instead, we’ll use the event types as coded in the dataset, ignoring any errors. We’ll pick out the 10 event types with the largest sum of fatalities and injuries and focus the analysis on those event types.
PopStormData <- PopStormData %>%
mutate(PopHealth = FATALITIES + INJURIES)
EventsByPopImpact <- PopStormData %>%
group_by(EVTYPE) %>%
summarize(PopImpact = sum(PopHealth))
EventsByPopImpact <- EventsByPopImpact[order(-EventsByPopImpact$PopImpact), ]
Top10PopImpact <- EventsByPopImpact[1:10, ]In order to show the impact of the ten event types with the largest numbers of fatalities and injuries, we’ll make a bar graph showing the number of fatalities and injuries caused by all of the events of those types.
p1 <- ggplot(data=PopStormData[PopStormData$EVTYPE %in% Top10PopImpact$EVTYPE,], aes(x=EVTYPE)) + geom_col(aes(y=-FATALITIES, fill="Fatalities")) + geom_col(aes(y=INJURIES, fill="Injuries")) + coord_flip() + labs(x = "Event Type", y = "Number of Fatalities and Injuries", fill = "Legend", title = "Population Impact of Weather Events") + scale_fill_manual(values = c("Fatalities" = "red", "Injuries" = "blue"))
p1
From the bar graph, we can see that tornadoes have caused by far the largest number of fatalities and the largest number of injuries of any type of severe weather event.
Economic Effects
In order to assess the economic effects of severe weather events, we’ll look at the dollar figures of property damage and crop damage caused by the various types of events.
As for the impact on population, because the number of event types is unmanageably large and has numerous obvious coding errors, and because I don’t know enough about the dataset to fix those coding errors in an intelligent way, I will not attempt to combine apparent duplicate event types. Instead, we’ll use the event types as coded in the dataset, ignoring any errors. We’ll pick out the 10 event types with the largest total property damage and crop damage, and focus the analysis on those event types.
EventsByPropDmg <- EconStormData %>%
group_by(EVTYPE) %>%
summarize(TotalPropDmg = sum(cleanPROPDMG))
EventsByPropDmg <- EventsByPropDmg[order(-EventsByPropDmg$TotalPropDmg), ]
Top10PropDmg <- EventsByPropDmg[1:10, ]
EventsByCropDmg <- EconStormData %>%
group_by(EVTYPE) %>%
summarize(TotalCropDmg = sum(cleanCROPDMG))
EventsByCropDmg <- EventsByCropDmg[order(-EventsByCropDmg$TotalCropDmg), ]
Top10CropDmg <- EventsByCropDmg[1:10, ]We’ll reshape the data to allow us to plot property and crop damage on the same plots.
longEconStormData <- EconStormData %>%
filter(EconStormData$EVTYPE %in% Top10PropDmg$EVTYPE | EconStormData$EVTYPE %in% Top10CropDmg$EVTYPE) %>%
gather(DmgType, Dollars, cleanCROPDMG, cleanPROPDMG)In order to show the impact of the event types that carry the highest economic costs, we will make boxplots of the property and crop damage caused by events of the types in the top ten for total amounts of either property or crop damage.
p2 <- ggplot(data=longEconStormData, aes(EVTYPE, log10(Dollars + 1))) +
geom_boxplot(aes(color = DmgType)) +
theme(axis.text.x = element_text(angle=90, hjust=1)) +
scale_y_continuous(breaks = c(0, 3, 6, 9), labels = c("0", "1,000", "1,000,000", "1,000,000,000")) +
labs(x = "Event Type", y = "Economic cost, in dollars (log scale)", title = "Economic Impact of Weather Events") +
scale_color_discrete(name="Legend", labels=c("Crop Damage", "Property Damage"))
p2
From the box plots, we can see that hurricanes and typhoons have caused the most economic damage in terms of dollar costs of property and crop damage. While it appears that most events don’t cause any crop damage (since it appears that all event types have median crop damage of $0), hurricanes and hurricane/typhoons are the only event type whose third quartile is significantly more than $0, and those are also the event types with the highest median property damage amounts.