How severe weather events are harmful to population health
Leandro Guerra
Saturday, August 23, 2014
Synopsis
The goal of this report is show the most harmful events that can be affect the united states, in order to take some decisions related to the future investments in planning and management of damages. The report identifies the most significant weather event types with the largest impact on population health (as measured by mean number of combined fatalities and injuries) and the largest economic consequences (as measured by the mean property damage and mean crop damage sustained during the event). This analysis finds that it is wind-related events such as tropical storms and tornadoes that have the greatest impact on population health and cause the most property damage. Excessive wetness and temperature extremes are the events that cause the most severe crop damage.
About the Data
The weather events are divided into 13 groups:
-Convection (e.g. tornado, lightning, thunderstorm, hail) -Flood (e.g. flash flood, river flood) -Extreme temperatures (e.g. extreme cold, extreme hot) -Marine (e.g. tsunami, coastal storm, rip current, high waves, high seas) -Winter (e.g. avalanche, snow, blizzard, icy roads, freeze) -Tropical Cyclones (e.g. tropical storm, hurricane) -High Wind (e.g. winds, microburst) -Fire -Rain -Drought/Dust (e.g. drought, dust storm, dust) -Landslide -Fog -Others
Data Processing
#Setting WD
setwd("C:/Users/Leandro/Google Drive/Coursera/DATASCIENCE")
#Unzip and read .csv file into the variable data
unzip <- bzfile("repdata-data-StormData.csv.bz2", "r")
data <- read.csv(unzip, stringsAsFactors = FALSE)
close(unzip)Select useful data
Subsetting data into variables that are needed and adding a new variable.
x <- which(colnames(data) %in% c("BGN_DATE", "PROPDMG", "CROPDMG", "EVTYPE",
"INJURIES", "FATALITIES"))
data <- data[, x]
head(data)## BGN_DATE EVTYPE FATALITIES INJURIES PROPDMG CROPDMG
## 1 4/18/1950 0:00:00 TORNADO 0 15 25.0 0
## 2 4/18/1950 0:00:00 TORNADO 0 0 2.5 0
## 3 2/20/1951 0:00:00 TORNADO 0 2 25.0 0
## 4 6/8/1951 0:00:00 TORNADO 0 2 2.5 0
## 5 11/15/1951 0:00:00 TORNADO 0 2 2.5 0
## 6 11/15/1951 0:00:00 TORNADO 0 6 2.5 0#Formatting date and time
data$YEAR <- as.integer(format(as.Date(data$BGN_DATE, "%m/%d/%Y 0:00:00"), "%Y"))
head(data)## BGN_DATE EVTYPE FATALITIES INJURIES PROPDMG CROPDMG YEAR
## 1 4/18/1950 0:00:00 TORNADO 0 15 25.0 0 1950
## 2 4/18/1950 0:00:00 TORNADO 0 0 2.5 0 1950
## 3 2/20/1951 0:00:00 TORNADO 0 2 25.0 0 1951
## 4 6/8/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 5 11/15/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 6 11/15/1951 0:00:00 TORNADO 0 6 2.5 0 1951#To uppercase
data$EVTYPE <- toupper(data$EVTYPE)
head(data)## BGN_DATE EVTYPE FATALITIES INJURIES PROPDMG CROPDMG YEAR
## 1 4/18/1950 0:00:00 TORNADO 0 15 25.0 0 1950
## 2 4/18/1950 0:00:00 TORNADO 0 0 2.5 0 1950
## 3 2/20/1951 0:00:00 TORNADO 0 2 25.0 0 1951
## 4 6/8/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 5 11/15/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 6 11/15/1951 0:00:00 TORNADO 0 6 2.5 0 1951# creates new variable
data$ECONOMICDMG <- data$PROPDMG + data$CROPDMG
head(data)## BGN_DATE EVTYPE FATALITIES INJURIES PROPDMG CROPDMG YEAR
## 1 4/18/1950 0:00:00 TORNADO 0 15 25.0 0 1950
## 2 4/18/1950 0:00:00 TORNADO 0 0 2.5 0 1950
## 3 2/20/1951 0:00:00 TORNADO 0 2 25.0 0 1951
## 4 6/8/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 5 11/15/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 6 11/15/1951 0:00:00 TORNADO 0 6 2.5 0 1951
## ECONOMICDMG
## 1 25.0
## 2 2.5
## 3 25.0
## 4 2.5
## 5 2.5
## 6 2.5# Select only positive value data
data <- subset(data, data$FATALITIES > 0 | data$ECONOMICDMG > 0 | data$INJURIES >
0)
head(data)## BGN_DATE EVTYPE FATALITIES INJURIES PROPDMG CROPDMG YEAR
## 1 4/18/1950 0:00:00 TORNADO 0 15 25.0 0 1950
## 2 4/18/1950 0:00:00 TORNADO 0 0 2.5 0 1950
## 3 2/20/1951 0:00:00 TORNADO 0 2 25.0 0 1951
## 4 6/8/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 5 11/15/1951 0:00:00 TORNADO 0 2 2.5 0 1951
## 6 11/15/1951 0:00:00 TORNADO 0 6 2.5 0 1951
## ECONOMICDMG
## 1 25.0
## 2 2.5
## 3 25.0
## 4 2.5
## 5 2.5
## 6 2.5Data aggregation
library(plyr)## Warning: package 'plyr' was built under R version 3.1.1# data aggregated by YEAR & EVTYPE.
#ddply -> For each subset of a data frame, apply function then combine results into a data frame.
eventYear <- ddply(data[, -1], .(YEAR, EVTYPE),
.fun = function(x) {
return(
c(sum(x$FATALITIES), sum(x$ECONOMICDMG), sum(x$INJURIES))
)
}
)
names(eventYear) <- c("YEAR", "EVTYPE", "FATALITIES", "ECONOMICDMG", "INJURIES")
head(eventYear)## YEAR EVTYPE FATALITIES ECONOMICDMG INJURIES
## 1 1950 TORNADO 70 16999 659
## 2 1951 TORNADO 34 10561 524
## 3 1952 TORNADO 230 16680 1915
## 4 1953 TORNADO 519 19182 5131
## 5 1954 TORNADO 36 23368 715
## 6 1955 TORNADO 129 27716 926Grouping the events We grouped the events by its related categories
#Function that calculates the events by categories (13 categories described in the synopsis)
#grepl -> search for matches to argument pattern within each element of a character vector
eventCategory <- function(x) {
ev <- x$EVTYPE[1]
if (grepl("LIG(H|N)T(N|)ING|TORNADO|T(H|)U(N|)(DER|ER|DEER|DERE)(STORM|STROM|TORM)|TSTM|HAIL",
ev)) {
category <- "Convection"
} else if (grepl("WINT(ER|RY)|ICE|AVALANC(H|)E|SNOW|BLIZZARD|FREEZ|ICY|FROST",
ev)) {
category <- "Winter"
} else if (grepl("COLD|HEAT|HOT|TEMPERATURE|COOL|WARM", ev)) {
category <- "Extreme Temp"
} else if (grepl("FLOOD| FLD$", ev)) {
category <- "Flood"
} else if (grepl("COASTAL|TSUNAMI|RIP CURRENT|MARINE|WATERSPOUT|SURF|SLEET|SEAS|(HIGH|RISING|HEAVY) (WAVES|SWELLS|WATER)",
ev)) {
category <- "Marine"
} else if (grepl("TROPICAL|HURRICANE|STORM SURGE|TYPHOON", ev)) {
category <- "Tropical Cyclones"
} else if (grepl("WIND|MICROBURST", ev)) {
category <- "High Wind"
} else if (grepl("FIRE", ev)) {
category <- "Fire"
} else if (grepl("RAIN|PRECIP", ev)) {
category <- "Rain"
} else if (grepl("DROUGHT|DUST", ev)) {
category <- "Drought/Dust"
} else if (grepl("LANDSLIDE|MUD.*SLIDE", ev)) {
category <- "Landslide"
} else if (grepl("FOG|VOG", ev)) {
category <- "Fog"
} else {
category <- "Others"
}
x$EVGROUP <- rep(category, dim(x)[1])
return(x)
}
eventYear <- ddply(eventYear, .(EVTYPE), .fun = eventCategory)
head(eventYear)## YEAR EVTYPE FATALITIES ECONOMICDMG INJURIES EVGROUP
## 1 2001 HIGH SURF ADVISORY 0 200 0 Marine
## 2 2000 FLASH FLOOD 0 50 0 Flood
## 3 1999 TSTM WIND 0 100 0 Convection
## 4 2000 TSTM WIND 0 8 0 Convection
## 5 1998 TSTM WIND (G45) 0 8 0 Convection
## 6 1994 ? 0 5 0 Others#We organize the data to show FATALITIES, ECONOMICDMG and INJURIES
#by YEAR and EVGROUP
groupYear <- ddply(eventYear, .(YEAR, EVGROUP), .fun = function(x) {
return(c(sum(x$FATALITIES), sum(x$ECONOMICDMG), sum(x$INJURIES)))
})
names(groupYear) <- c("YEAR", "EVGROUP", "FATALITIES", "ECONOMICDMG", "INJURIES")
head(groupYear)## YEAR EVGROUP FATALITIES ECONOMICDMG INJURIES
## 1 1950 Convection 70 16999 659
## 2 1951 Convection 34 10561 524
## 3 1952 Convection 230 16680 1915
## 4 1953 Convection 519 19182 5131
## 5 1954 Convection 36 23368 715
## 6 1955 Convection 129 27716 926# calculate average annual damage by group
eventFirstYear <- ddply(groupYear, .(EVGROUP), .fun = function(x) {
return(c(min(x$YEAR)))
})
names(eventFirstYear) <- c("Weather.Event", "First.Year")
head(eventFirstYear)## Weather.Event First.Year
## 1 Convection 1950
## 2 Drought/Dust 1993
## 3 Extreme Temp 1993
## 4 Fire 1993
## 5 Flood 1993
## 6 Fog 1993As we can notice analysing the variable eventFirstYear, the weather event “Convection” has its occurency starting at the 50’s but the others events starts at 1993. In this section we subset the groupYear to analysis all the events starting from 1993
## start data analysis at 1993
groupYear <- subset(groupYear, YEAR >= 1993)
# calculate average annual damage by group
byGroup <- ddply(groupYear, .(EVGROUP), .fun = function(x) {
return(c(mean(x$FATALITIES), mean(x$ECONOMICDMG), mean(x$INJURIES)))
})
names(byGroup) <- c("EVGROUP", "AVG.FATALITIES", "AVG.ECONOMICDMG", "AVG.INJURIES")
head(byGroup)## EVGROUP AVG.FATALITIES AVG.ECONOMICDMG AVG.INJURIES
## 1 Convection 154.895 328814.6 1883.68
## 2 Drought/Dust 1.263 2388.8 25.63
## 3 Extreme Temp 190.579 1461.9 503.32
## 4 Fire 4.737 7093.9 84.63
## 5 Flood 81.737 148846.1 456.89
## 6 Fog 4.211 898.7 56.63Results
Results section 1 - Health Harmful Events
This histograms Show fatalities and injuries for weather events.
# Graph libraries
library(ggplot2)
library(scales)
# average annual populational damage by group of event
byGroup$EVGROUP <- with(byGroup, reorder(EVGROUP, -AVG.FATALITIES))
g <- ggplot(byGroup, aes(x = EVGROUP))
g + geom_histogram(aes(weight = AVG.FATALITIES, fill = AVG.FATALITIES), binwidth = 5,
color = "black") + ggtitle("Average Fatalities") + ylab("# Fatalities") +
xlab("Weather Event") + theme(axis.text.x = element_text(angle = 45, hjust = 1))
# average annual populational damage by group of event
byGroup$EVGROUP <- with(byGroup, reorder(EVGROUP, -AVG.INJURIES))
g <- ggplot(byGroup, aes(x = EVGROUP))
g + geom_histogram(aes(weight = AVG.INJURIES, fill = AVG.INJURIES), binwidth = 1,
color = "black") + ggtitle("Average Injuries") + ylab("# Injuries") + xlab("Weather Event") +
theme(axis.text.x = element_text(angle = 45, hjust = 1))
Results section 2 - Economic Harm
Histogram of weather event harm to the economy.
# average annual economical damage by group of event
byGroup$EVGROUP <- with(byGroup, reorder(EVGROUP, -AVG.ECONOMICDMG))
g <- ggplot(byGroup, aes(x = EVGROUP))
g + geom_histogram(aes(weight = AVG.ECONOMICDMG, fill = AVG.ECONOMICDMG), binwidth = 1,
color = "black") + ggtitle("Average Economic Damage") + ylab("Economic damage") +
xlab("Weather Event") + theme(axis.text.x = element_text(angle = 45, hjust = 1))
Conclusion
According to the analysis, we can notice in the results that the most harmful events for population are “Extreme temperatures” and “Convection” when we look at “Average Fatalities”. When we talk about “Average Injuries”, we have the same events, but in a different order - “Convection” and “Extreme Temperatures”.
Now, when we look at economic damage,the extremely harmful events for economy are “Convection” and “Flood”. It is quite logical think about it, but here we can prove with data.