The Influence of Storms and Severe Weather Events on Economy and Population Health
Vladimir Kuznetsov
December 27, 2015
Synopsis
The goal of the current project is to explore the NOAA Storm Database and answer the following questions:
- Across the United States, which types of events are most harmful with respect to population health?
- Across the United States, which types of events have the greatest economic consequences?
Analyzing data contains observations from January 03, 1951 to November 30, 2011.
The found answer to first question is that the most harmful event to public health is TORDNADO:
- Total 5633 fatalities during all period of observations
- Total 91346 injuries during all period of observations
The found answer to second question is that the most harmful event to economy is FLOOD:
- Total damage is 150319.678 millions of dollars during all period of observations
Data Processing
We start from downloading data and read it into data frame.
Load libraries and data:
# load libraries
library(ggplot2)
library(gridExtra)# get the data
fileName <- "repdata-data-StormData.csv.bz2"
if (!file.exists(fileName)){
download.file(
"https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2",
destfile=fileName)
}
# load the data
data <- read.csv(bzfile(fileName), header=TRUE, sep=",", stringsAsFactors = FALSE)
dim(data)## [1] 902297 37dates <- as.Date(data$BGN_DATE, format = "%m/%d/%Y %H:%M:%S")
print("Dates of observations:"); print(c("Date from"=min(dates), "Date to"=max(dates)))## [1] "Dates of observations:"## Date from Date to
## "1950-01-03" "2011-11-30"Our data have 902297 rows (observations) and 37 columns (variables).
Preprocessing the data
Due to the Storm Data Documentation for our project we select the next variables:
- BGN_DATE, stands for date of event
- EVTYPE, stands for event type (e.g. tornado, snow, etc.)
- FATALITIES, will use to measure damage to population health
- INJURIES, will use to measure damage to population health
- PROPDMG, stands for property damage
- CROPDMG, stands for crop damage
- PROPDMGEXP
- CROPDMGEXP
PROPDMGEXP and CROPDMGEXP contain alphabetical characters explained in documentations as:
“Alphabetical characters used to signify magnitude include “K” for thousands, “M” for millions, and “B” for billions.” Page 12
Property Damage (PROPDMG) and Crop Damage (CROPDMG) will be used to measure economical damage.
Number of Fatalities (FATALITIES) and number of Injuries (INJURIES) will be used to measure health damage.
Data transformation:
# remove unnecessary variables
data.tidy <- data[, c("BGN_DATE", "EVTYPE", "FATALITIES", "INJURIES",
"PROPDMG", "PROPDMGEXP", "CROPDMG", "CROPDMGEXP")]
# set type to upper case
# to union names like "Wintry mix" and "Wintry Mix"
data.tidy$EVTYPE = toupper(data.tidy$EVTYPE)
# remove old variable from memory
rm(data)
# function to convert damage to dollars
calculate_amount <- function (sum, exp){
exp <- toupper(exp)
if (exp == "H") ## hunderds
sum <- sum * 10^2
else if (exp == "K") # thousands
sum <- sum * 10^3
else if (exp == "M") # millions
sum <- sum * 10^6
else if (exp == "B") # billions
sum <- sum * 10^9
else if (exp %in% c("+","-","?"))
sum <- sum
else if (!is.na(as.numeric(exp))) # defined as number
sum <- sum * 10^as.numeric(exp)
sum
}
# Convert Crop and Property Damage to dollars
data.tidy$PROPDMG <- mapply(calculate_amount, data.tidy$PROPDMG, data.tidy$PROPDMGEXP)
data.tidy$CROPDMG <- mapply(calculate_amount, data.tidy$CROPDMG, data.tidy$CROPDMGEXP)
# select year from date
data.tidy$YEAR <- as.numeric(format(
as.Date(data.tidy$BGN_DATE, format = "%m/%d/%Y %H:%M:%S"), "%Y"))
# remove unused columns
data.tidy$PROPDMGEXP <- NULL
data.tidy$CROPDMGEXP <- NULL
data.tidy$BGN_DATE <- NULL
# rename columns
colnames(data.tidy) <- c("Event", "Fatalities", "Injuries", "PropertyDamage", "CropDamage", "Year")
# now data is tidy
print("Preview piece of data");print(head(data.tidy), row.names = F)## [1] "Preview piece of data"## Event Fatalities Injuries PropertyDamage CropDamage Year
## TORNADO 0 15 25000 0 1950
## TORNADO 0 0 2500 0 1950
## TORNADO 0 2 25000 0 1951
## TORNADO 0 2 2500 0 1951
## TORNADO 0 2 2500 0 1951
## TORNADO 0 6 2500 0 1951Results
1. The most harmful types of events with respect to population health
Find top Event’s Types by Fatalities and by Injuries.
harmfulEvents <- aggregate(cbind(Fatalities, Injuries) ~ Event, data.tidy, FUN=sum)
topFatalities <- harmfulEvents[head(order(harmfulEvents$Fatalities, decreasing = TRUE), 10), 1:2]
topInjuries <- harmfulEvents[head(order(harmfulEvents$Injuries, decreasing = TRUE), 10), c(1, 3)]
# Show top 10 by Fatalities
print("Top 10 by Fatalities");print(topFatalities, row.names = F)## [1] "Top 10 by Fatalities"## Event Fatalities
## TORNADO 5633
## EXCESSIVE HEAT 1903
## FLASH FLOOD 978
## HEAT 937
## LIGHTNING 816
## TSTM WIND 504
## FLOOD 470
## RIP CURRENT 368
## HIGH WIND 248
## AVALANCHE 224# Show top 10 by Injuries
print("Top 10 by Injuries");print(topInjuries, row.names = F)## [1] "Top 10 by Injuries"## Event Injuries
## TORNADO 91346
## TSTM WIND 6957
## FLOOD 6789
## EXCESSIVE HEAT 6525
## LIGHTNING 5230
## HEAT 2100
## ICE STORM 1975
## FLASH FLOOD 1777
## THUNDERSTORM WIND 1488
## HAIL 1361Now lets’s show top events in plots:
plot1 <-
ggplot(topFatalities,
aes(x=reorder(Event, Fatalities, function(x) x*-1), y=Fatalities)) +
geom_bar(stat = "identity", position = "stack", aes(fill = Fatalities)) +
theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
labs(title="Top 10 Events by Fatalities in 1950-2011") +
labs(x="", y="Total Fatalities") +
scale_fill_continuous(low="darkorange", high="darkred",
limits=c(0, topFatalities$Fatalities[1]))
plot2 <-
ggplot(topInjuries,
aes(x=reorder(Event, Injuries, function(x) x*-1), y=Injuries)) +
geom_bar(stat = "identity", position = "stack", aes(fill = Injuries)) +
theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
labs(title="Top 10 Events by Injuries in 1950-2011") +
labs(x="", y="Total Injuries") +
scale_fill_continuous(low="darkorange", high="darkred",
limits=c(0, topInjuries$Injuries[1]))
grid.arrange(plot1, plot2, nrow = 2)
RESULTS:
- The most harmful event by fatalities is TORNADO with total 5633 fatalities.
- The most harmful event by injuries is TORNADO with total 91346 injuries
2. The most harmful types of events with the greatest economic consequences
Add Property Damage to Crop Damage and find top 10 events with the greates economic consequences.
damage <- aggregate(PropertyDamage + CropDamage ~ Event, data.tidy, FUN=sum)
colnames(damage)[2] <- "TotalDamage"
topDamage <- damage[head(order(damage$TotalDamage, decreasing = TRUE), 10), ]Top 10 events with economic damage in millions of dollars.
topDamage$TotalDamage <- topDamage$TotalDamage/1e6
# Show top 10 events
print("Top 10 by financial damage");print(topDamage, row.names = F)## [1] "Top 10 by financial damage"## Event TotalDamage
## FLOOD 150319.678
## HURRICANE/TYPHOON 71913.713
## TORNADO 57362.334
## STORM SURGE 43323.541
## HAIL 18761.222
## FLASH FLOOD 18243.991
## DROUGHT 15018.672
## HURRICANE 14610.229
## RIVER FLOOD 10148.405
## ICE STORM 8967.041Now let’s show the plot.
update_geom_defaults("bar", list(fill = "red"))
plot1 <- ggplot(topDamage,
aes(x=reorder(Event, TotalDamage, function(x) x*-1),
y=TotalDamage))
plot1 <- plot1 + geom_bar(stat = "identity", position = "stack", aes(fill=TotalDamage))
plot1 <- plot1 + theme(axis.text.x = element_text(angle = 45, hjust = 1))
plot1 <- plot1 + labs(title="Top 10 Events by Economic Damage in 1950-2011")
plot1 <- plot1 + labs(x="", y="Total Damage in million dollars")
plot1 <- plot1 + scale_fill_continuous(low="darkorange", high="darkred",
limits=c(0, topDamage$TotalDamage[1]), name="Total Damage")
plot1
RESULTS:
- The most harmful event by economic damage is FLOOD with total damage amount 150319.678 millions of dollars.