Analysis of Health Hazards and Economic Impacts of Extreme Weather

Synopsis

This report summarizes an analysis of the NOAA extreme weather records. The data was subsetted to include only the years 1996-2011 for the 50 United States and the District of Columbia.

The analysis presented here shows that clearly, tornados, floods, and exterme heat present the greatest health related hazards from extreme weather. Extreme heat contributes the most to fatalities, while tornadoes produce the most injurues. Flooding, hurricanes, and tornadoes contribute most to economic damage with flooding being a much larger contributor than any other weather event.

As a side note: For improved future reporting, it is recomended that event types be explicitly enumerated and thier criteria, firmly established.

Data Processing

Obtain the NOAA storm data from the course website.

fname="./stormdata.bz2"
if(!file.exists(fname)) {
    url <- "http://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2"
    download.file(url, fname, mode="wb")
}
stormdata=read.csv(fname)

We don’t need all of the data to answer the questions. Because the older data is less consistent in the way it was collected, and is likely, less accurate. Based on converstaions in this discussion thread (https://class.coursera.org/repdata-006/forum/thread?thread_id=125) we make the decision to use data starting from 1996.

stormdata$DATE=as.Date(factor(stormdata$BGN_DATE), format("%m/%d/%Y"))
stormdata$YEAR=as.numeric(format(stormdata$DATE, "%Y"))
stormdata=subset(stormdata,YEAR>=1996)

The data obtained contains values for U.S. States and territories, but also data from sources that are unclear. For the purposes of this analysis, we will limit the scope to only the 50 states and the District of Columbia.

statelist=c("AK","AL","AR","AZ","CA","CO","CT","DC","DE","FL","GA","HI","IA","ID","IL","IN","KS","KY","LA","MA","MD","ME","MI","MN","MO","MS","MT","NC","ND","NE","NH","NJ","NM","NV","NY","OH","OK","OR","PA","RI","SC","SD","TN","TX","UT","VA","VT","WA","WI","WV","WY")
statedata=subset(stormdata, STATE %in% statelist)
statedata=droplevels(statedata)

The naming conventions are inconsistent. We try to make a reasonable attempt to group similary named events. We could spend more time here cleaning the data set, but this should be suffucient to identify the major contributors being assessed.

statedata$EVTYPE=with(statedata, gsub(".*UNSEASONABL.*|WARM WEATHER|UNUSUALLY WARM|VERY WARM|PROLONG WARMTH|MONTHLY PRECIPITATION|UNUSUAL WARMTH","UNSEASONABLE WEATHER", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*WINTER.*|.*SNOW.*|.*BLIZZARD.*|.*ICE.*|.*FROST.*|.*FREEZE.*|.*COLD.*|.*WINTRY.*|.*FREEZING.*|.*ICY.*","WINTER WEATHER", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*SLEET.*","SLEET", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*HEAT.*|EXTREME HEAT|RECORD WARMTH|HIGH TEMPERATURE RECORD|RECORD HIGH|HOT SPELL|.*RECORD WARM.*", "EXTREME HEAT", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*THUNDERSTORM.*","THUNDERSTORM", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*MICOBURST.*|.*DOWNBURST.*","MICROBURST", EVTYPE, ignore.case=T))
match=agrep("MICROBURST", statedata$EVTYPE, max=2, ignore.case=TRUE)
statedata$EVTYPE[match]="MICROBURST"

statedata$EVTYPE=with(statedata, gsub("GRASS FIRES?|.*WILD ?FIRE|.*BRUSH ?FIRE|.*FOREST ?FIRE","WILD/FOREST FIRE", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub("ABNORMALLY WET|EXTREMELY WET|EXCESSIVE WETNESS|WET MONTH|WET YEAR","WET CONDITIONS", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub("LAND ?SLIDES?|LANDSLUMP|MUD ?SLIDE|ROCK ?SLIDE|.*LANDSLIDE.*","LANDSLIDE", EVTYPE, ignore.case=T))
match=agrep("LANDSLIDE", statedata$EVTYPE, max=2, ignore.case=TRUE)
statedata$EVTYPE[match]="LANDSLIDE"

statedata$EVTYPE=with(statedata, gsub("DRYEST MONTH|.*DROUGHT.*|ABNORMALLY DROUGHT|VERY DRY.*|DRY|DRY SPELL|DRY CONDITIONS","DROUGHT", EVTYPE, ignore.case=T))
match=agrep("DROUGHT", statedata$EVTYPE, max=2, ignore.case=TRUE)
statedata$EVTYPE[match]="DROUGHT"

statedata$EVTYPE=with(statedata, gsub(".*HAIL.*","HAIL", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*LIGHTNING.*","LIGHTNING", EVTYPE, ignore.case=T))
match=agrep("LIGHTNING", statedata$EVTYPE, max=2, ignore.case=TRUE)
statedata$EVTYPE[match]="LIGHTNING"

statedata$EVTYPE=with(statedata, gsub(".*TROPICAL STORM.*","TROPICAL STORM", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*TORNADO.*|.*FUNNEL.*|.*WATERSPOUT.*|LANDSPOUT","TORNADO", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*HURRICANE.*|.*FLOYD","HURRICANE", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*FLOOD.*|FLOODING/TIDE|.*RAIN.*|STORM SURGE|.*STREAM FLD|.*SMALL STREAM.*|.*HIGH WATER.*|RAPIDLY RISING WATER", "FLOODING", EVTYPE, ignore.case=T))
match=agrep("FLOOD", statedata$EVTYPE, max=2, ignore.case=TRUE)
statedata$EVTYPE[match]="FLOODING"

statedata$EVTYPE=with(statedata, gsub(".*SURF.*|.*RED FLAG.*","HIGH/HAZARDOUS SURF", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*WIND.*|.*WND.*|.*GUSTNADO.*","HIGH WIND", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*MIXED PRECIP.*","MIXED PRECIPITATION", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*BLOW-?OUT TIDE.*","BLOW-OUT TIDE", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*HYPOTHERMIA/EXPOSURE.*","HYPOTHERMIA/EXPOSURE", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*DUST.*","DUST STORM", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*VOLCAN.*","VOLCANIC ACTIVITY", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*DAM.*","DAM BREACH", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*WALL CLOUD.*","WALL CLOUD", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*FOG.*","FOG", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub("*.RIP CURRENT.*","RIP CURRENT", EVTYPE, ignore.case=T))

statedata$EVTYPE=with(statedata, gsub(".*SUMMARY.*","OTHER", EVTYPE, ignore.case=T))

# Turn the event types back into factor variables.
statedata$EVTYPE=factor(statedata$EVTYPE)

unique(statedata$EVTYPE)

##  [1] WINTER WEATHER         TORNADO                HIGH WIND             
##  [4] HAIL                   FLOODING               EXTREME HEAT          
##  [7] LIGHTNING              RIP CURRENTS           Other                 
## [10] DROUGHT                Temperature record     WILD/FOREST FIRE      
## [13] DUST STORM             MICROBURST             FOG                   
## [16] THUNDERSTORM           HIGH/HAZARDOUS SURF    Marine Accident       
## [19] AVALANCHE              TROPICAL STORM         COASTAL STORM         
## [22] WET CONDITIONS         HURRICANE              Beach Erosion         
## [25] UNSEASONABLE WEATHER   LANDSLIDE              Coastal Storm         
## [28] Glaze                  Heavy Precipitation    Record temperature    
## [31] MIXED PRECIPITATION    OTHER                  Metro Storm, May 26   
## [34] No Severe Weather      Record Temperatures    NONE                  
## [37] DAM BREACH             BLOW-OUT TIDE          RIP CURRENT           
## [40] RECORD TEMPERATURES    HYPOTHERMIA/EXPOSURE   COASTALSTORM          
## [43] BEACH EROSION          ABNORMAL WARMTH        GLAZE                 
## [46] COASTAL EROSION        SEICHE                 TSTM                  
## [49] HYPERTHERMIA/EXPOSURE  RECORD COOL            HOT WEATHER           
## [52] RECORD TEMPERATURE     VOLCANIC ACTIVITY      COOL SPELL            
## [55] HEAVY SEAS             VOG                    MONTHLY TEMPERATURE   
## [58] DRIEST MONTH           RECORD PRECIPITATION   SLEET                 
## [61] HIGH SEAS              SMOKE                  ROGUE WAVE            
## [64] NORTHERN LIGHTS        ASTRONOMICAL HIGH TIDE DROWNING              
## [67] TROPICAL DEPRESSION    TSUNAMI                DENSE SMOKE           
## [70] ASTRONOMICAL LOW TIDE 
## 70 Levels: ABNORMAL WARMTH ... WINTER WEATHER

To assess the most harmful severe weather events we will look at events that have a number of high numbers of fatalities and high number of injuries.

fatalities=with(statedata, aggregate(FATALITIES, by=list(EVTYPE), FUN=sum))
# Drop rows with zero values.
fatalities=subset(fatalities, x > 0)
fatalities=droplevels(fatalities)
#fatalities$Group.1=factor(fatalities$Group.1)

hi.fatal=subset(fatalities, x > 500)
hi.fatal=droplevels(hi.fatal)
hi.fatal

##           Group.1    x
## 19   EXTREME HEAT 2036
## 20       FLOODING 1367
## 28      HIGH WIND  628
## 35      LIGHTNING  642
## 60        TORNADO 1513
## 70 WINTER WEATHER  909

injuries=with(statedata, aggregate(INJURIES, by=list(EVTYPE), FUN=sum))
# Drop rows with zero values.
injuries=subset(injuries, x > 0)
injuries=droplevels(injuries)

hi.injury=subset(injuries, x > 2500)
hi.injury=droplevels(hi.injury)
hi.injury

##           Group.1     x
## 19   EXTREME HEAT  7683
## 20       FLOODING  8767
## 28      HIGH WIND  5117
## 35      LIGHTNING  4135
## 60        TORNADO 20669
## 70 WINTER WEATHER  3500

To ocmpute cost, we combine the reported property and crop damage. The *EXP columns are stated to contain units of thousands, millions, and billions [K,B,M of dollars]. We choose to impute invalid values by entering a multiple of one. If this is not the correct multiple, then the economic impact will be under-reported for that event.

statedata$PROPDMGEXP=sub("K","1000", statedata$PROPDMGEXP, ignore.case=T)
statedata$PROPDMGEXP=sub("M","1000000", statedata$PROPDMGEXP, ignore.case=T)
statedata$PROPDMGEXP=sub("B","1000000000", statedata$PROPDMGEXP, ignore.case=T)
statedata$PROPDMGEXP=sub("[^KMB]","1",statedata$PROPDMGEXP, ignore.case=T)
statedata$PROPDMGEXP=as.numeric(statedata$PROPDMGEXP)
statedata$PROPDMGEXP[is.na(statedata$PROPDMGEXP)]=1
#table(statedata$PROPDMGEXP)

statedata$CROPDMGEXP=sub("K","1000", statedata$CROPDMGEXP, ignore.case=T)
statedata$CROPDMGEXP=sub("M","1000000", statedata$CROPDMGEXP, ignore.case=T)
statedata$CROPDMGEXP=sub("B","1000000000", statedata$CROPDMGEXP, ignore.case=T)
statedata$CROPDMGEXP=sub("[^KMB]","1",statedata$CROPDMGEXP, ignore.case=T)
statedata$CROPDMGEXP=as.numeric(statedata$CROPDMGEXP)
statedata$CROPDMGEXP[is.na(statedata$CROPDMGEXP)]=1
#table(statedata$CROPDMGEXP)

statedata$COST = with(statedata, PROPDMG*PROPDMGEXP + CROPDMG*CROPDMGEXP)

# Sum the cost per event type.
cost=with(statedata, aggregate(COST, by=list(EVTYPE), FUN=sum))
hi.cost=subset(cost, x > 1e+10)
hi.cost=droplevels(hi.cost)

Results

To assess the most harmful severe weather events we will look at events that have a number of high numbers of fatalities and high number of injuries.

par(mar=c(10.1,6.1,4.1,2.1))
plot(hi.fatal, las=2, main="Number of fatalities per severe weather type", xlab="", ylab="")
mtext("Number of Fatalities", side=2, line=5)

plot of chunk plot harmfulness

plot(hi.injury, las=2, main="Number of injuries per severe weather type", xlab="", ylab="")
mtext("Number of Injuries", side=2, line=5)

plot of chunk plot harmfulness

To assess the economic impact, we report the evnts with have the highest costs (reported on a logarithmic scale).

par(mar=c(10.1,6.1,4.1,2.1))
plot(hi.cost, las=2, main="Economic import by severe weather type", xlab="", ylab="")
mtext("Event cost (log scale)", side=2, line=5)

plot of chunk plot economic impact