NOAA Storm Data Analysis

Synopsis

This report contains the analysis of the U.S. National Oceanic and Atmospheric Administration’s (NOAA) storm database ¹ and tries to answer following two questions:

Across United States, which type of events has greatest impact on population health.
Across United States, which type of events has greatest economic consequences.

Data Processing

The data is in the form of a comma-separated-value file compressed via the bzip2. Following fields are found out to be important fields to answer above questions:

EVTYPE : This indicates the type of storm event
FATALITIES + INJURIES : This indicates number of people died or injuried either directly or indirectly due to storm. Both of the events have similar weights in terms of thier impact to population health, hence simple sum of these parameters indicates the impact on population health due to given storm event.
PROPDMG (x 10^PROPDMGEXP) + CROPDMG (x 10^CROPDMGEXP): Impact on property and crops, which will be helpful to calculate the economic consequences of the storm event.

# select only required columns
rcols <- c('STATE', 'EVTYPE', 'FATALITIES', 'INJURIES', 'PROPDMG', 'PROPDMGEXP', 'CROPDMG',  'CROPDMGEXP')
storm.data = read.csv(bzfile('repdata-data-StormData.csv.bz2'), header=T, stringsAsFactors=F)[, rcols]

# This function converts the character-exponents into numeric format
# and keeps numeric exponent as-is.
unit2num <- function(u) {
  u <- switch(u,
         '+' = {0}, '-' = {0}, '?' = {0},
         'h' = {2}, 'H' = {2},
         'k' = {3}, 'K' = {3},
         'm' = {6}, 'M' = {6},
         'b' = {9}, 'B' = {9},
         u)
  if(u == '' | u == ' ') 0
  else as.integer(u)
}

# this function converts empty/space value to zero and any other 
# strings to numeric value
sanitize_nums <- function(x) {
  if(x == '' | x == ' ') 0
  else as.numeric(x)
}

# calculate the actual property damage by multiplying PROPDMG with 10^PROPDMGEXP
storm.data$PROPDMG <- apply(data.frame(storm.data$PROPDMG, storm.data$PROPDMGEXP), 1, function(x) { sanitize_nums(x[1])*10^unit2num(x[2]) })

# calculate the actual property damage by multiplying CROPDMG with 10^CROPDMGEXP
storm.data$CROPDMG <- apply(data.frame(storm.data$CROPDMG, storm.data$CROPDMGEXP), 1, function(x) { sanitize_nums(x[1])*10^unit2num(x[2]) })

# map abbrevation to state name
state.data <- data.frame(state.abb, state.name)

Impact on Population Health

Following code creates a map of the U.S states and the event which causes the most impact on public health with respect to injuries and death in particular states.

library(plyr)
library(maps)
library(mapproj)
library(RColorBrewer)

aggr <- aggregate(FATALITIES + INJURIES ~ EVTYPE + STATE, data = storm.data, sum)
names(aggr) <- c('EVTYPE', 'STATE', 'CASUALITIES')

# prepare colors
unique_causes <- unique(storm.data$EVTYPE)
set.seed(19619)
colors <- sample(colorRampPalette(c("red", "yellow", "green", "orange", "blue"))(n = length(unique_causes)))

# select the event which causes the most population health impact in each state
saggr <- ddply(aggr, 'STATE', function (X) {  X$EVTYPE[X$CASUALITIES == max(X$CASUALITIES)][1] })
unique_causes <- unique(saggr$V1)
saggr$color <- sapply(saggr$V1, function(x) { colors[match(x, unique_causes)] })

# Get actual state name needed by map function
saggr$STATE_NAME <- state.data$state.name[match(saggr$STATE, state.data$state.abb)]

# use state map to display the storm information
map("state", col = saggr$color, region = saggr$STATE_NAME, fill = TRUE, lty = 0, resolution = 0, projection="polyconic")
map("state", col = "black", fill = FALSE, add = TRUE, lty = 1, lwd = 1, projection="polyconic")
  
legend("bottomleft", unique_causes, fill = colors, cex=.56, bty = 'n')

Following code calculates the event that causes the most damage to public health across entire USA.

usa_aggr <- aggregate(FATALITIES + INJURIES ~ EVTYPE, data = storm.data, sum)
names(usa_aggr) <- c('EVTYPE', 'CASUALITIES')

max_casualties <- max(usa_aggr$CASUALITIES)
ph_event <- usa_aggr$EVTYPE[usa_aggr$CASUALITIES == max_casualties]

From above data, TORNADO causes most damage to public health.

Impact on Economy

Following code creates a map of the U.S states and the event which causes the most impact with respect to propery damage and damage to crops in particular state.

library(plyr)
library(maps)
library(mapproj)
library(RColorBrewer)

aggr <- aggregate(CROPDMG + PROPDMG ~ EVTYPE + STATE, data = storm.data, sum)
names(aggr) <- c('EVTYPE', 'STATE', 'ECON_IMPACT')

# prepare colors
unique_causes <- unique(storm.data$EVTYPE)
set.seed(8080)
colors <- sample(colorRampPalette(c("red", "yellow", "green", "orange", "blue"))(n = length(unique_causes)))

# select the event which causes the most economic impact in each state
saggr <- ddply(aggr, 'STATE', function (X) {  X$EVTYPE[X$ECON_IMPACT == max(X$ECON_IMPACT)][1] })
unique_causes <- unique(saggr$V1)
# map color to the event
saggr$color <- sapply(saggr$V1, function(x) { colors[match(x, unique_causes)] })

# Get actual state name needed by map function
saggr$STATE_NAME <- state.data$state.name[match(saggr$STATE, state.data$state.abb)]

# use state map to display the storm information
map("state", col = saggr$color, region = saggr$STATE_NAME, fill = TRUE, lty = 0, resolution = 0, projection="polyconic")
map("state", col = "black", fill = FALSE, add = TRUE, lty = 1, lwd = 1, projection="polyconic")
  
legend("bottomleft", unique_causes, fill = colors, cex=.56, bty = 'n')

Following code calculates the event that causes the most damage to property and crops across USA.

usa_aggr <- aggregate(CROPDMG + PROPDMG ~ EVTYPE, data = storm.data, sum)
names(usa_aggr) <- c('EVTYPE', 'ECON_IMPACT')

max_econ_impact <- max(usa_aggr$ECON_IMPACT)
econ_event <- usa_aggr$EVTYPE[usa_aggr$ECON_IMPACT == max_econ_impact]

From above data, FLOOD causes most damage to economy.

Results

From the above data:

Across United States, TORNADO causes most harm to population health, accountable for almost 96979 deaths or injuries.
Across United States, FLOOD causes most harm to property and crops, accountable for almost $150319678257 worth of damage.

¹ U.S. National Oceanic and Atmospheric Administration’s (NOAA) storm database