Severe Weather in the USA: Human Health and Economics consequences
Sitaram Gautam
January 31, 2016
Synopsis
This is an exploratory analysis based on data set provided by National Weather services. The dataset consists of historical severe weather events, timeline, duration, fatalities, injuries and etc. This analytis focuses on population health and economics consequences by event type and find out which of the events are the most harmful
Data Processing
This analysis uses following packages. dplyr to mutate, aggregate, filter and arrange data. stringr is used to parse strings and put into array, and finally ggplot2 is used to show the charts. Please make sure these packages are installed before running the code.
library(dplyr)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionlibrary(stringr)
library(ggplot2)The original file is unzipped and placed in the current working directory. The dataset consists too many fields that the scope of this analysis cannot cover. To minimize the file size and data table clean, only the fields of interest are read.
## Read certain columns to be used in the analysis
storm_dataset <-
read.csv(file = "repdata-data-StormData.csv", header = TRUE,stringsAsFactors = FALSE)[,c(
"BGN_DATE", "END_DATE", "STATE", "EVTYPE", "FATALITIES", "INJURIES", "PROPDMG", "CROPDMG"
)]The analysis is divided in two categories: Most harmful weather events by population health and by economic consequeces. For population health anaylsys, number of fatalities and injuries are the key metrices. For economic, number of corp. damages and number of property damages are taken into account. On top of category specific metrices, information such as how many states the event has occured, how many years, how many months out of 12 months are also considered to come up with the sevirity score.
Clean up the date formart etc.Also add month and year field to the dataset with mutate function in dplyr.Distinct number of years, months and states are captured which tells the scope of certain event type by timeline as well as by geolocation.
## Keep only date part from BGN_DATE and END_DATE
storm_dataset <- mutate(
storm_dataset,
BGN_DATE = str_split_fixed(storm_dataset$BGN_DATE, " ", 2)[,1],
END_DATE = str_split_fixed(storm_dataset$END_DATE, " ", 2)[,1],
EVTYPE = tolower(EVTYPE),
YEAR = as.numeric(str_split_fixed(BGN_DATE, "/", 3)[,3]),
MONTH = as.numeric(str_split_fixed(BGN_DATE, "/", 3)[,1])
)
total_num_year <-
as.numeric(summarize(storm_dataset, n_distinct(YEAR)))
total_num_month <-
as.numeric(summarize(storm_dataset, n_distinct(MONTH)))
total_states <-
as.numeric(summarize(storm_dataset, n_distinct(STATE)))Code to come up with the severity score for populaiton health:
total_fatal <- as.numeric(sum(storm_dataset$FATALITIES))
total_injur <- as.numeric(sum(storm_dataset$INJURIES))
## Group and Summarize the data
hlth_by_event_type <- storm_dataset %>%
filter(FATALITIES > 0 | INJURIES > 0) %>%
group_by(EVTYPE) %>%
summarize(
event_count = n(),
fatal = sum(FATALITIES),
injur = sum(INJURIES),
num_year = n_distinct(YEAR),
num_month = n_distinct(MONTH),
num_states = n_distinct(STATE),
severity_score = (3 * fatal / total_fatal) * 100 + (2 * injur / total_injur) *
100 + (num_year / 62) * 100 + (num_month / 12) * 100 + (num_states / total_states) *
100
) %>%
arrange(desc(severity_score))Similarly, for the severity score related to economic consequences is calculated following code chunk.
total_prop_dmg <- as.numeric(sum(storm_dataset$PROPDMG))
total_corp_dmg <- as.numeric(sum(storm_dataset$CROPDMG))
## Group and Summarize the data
econ_by_event_type <- storm_dataset %>%
filter(CROPDMG > 0 | PROPDMG > 0) %>%
group_by(EVTYPE) %>%
summarize(
event_count = n(),
prop_damage = sum(PROPDMG),
corp_damage = sum(CROPDMG),
num_year = n_distinct(YEAR),
num_month = n_distinct(MONTH),
num_states = n_distinct(STATE),
severity_score = (prop_damage / total_prop_dmg) * 100 + (corp_damage / total_corp_dmg) *
100 + (num_year / 62) * 100 + (num_month / 12) * 100 + (num_states / total_states) *
100
) %>%
arrange(desc(severity_score))Results
First, let’s preview the final aggregated resultset with calculated severity score. The resultset is ordered by the seveeiry score in decreasing order. Higher the severity score, more harmful the event type is.
head(hlth_by_event_type)## Source: local data frame [6 x 8]
##
## EVTYPE event_count fatal injur num_year num_month num_states
## (chr) (int) (dbl) (dbl) (int) (int) (int)
## 1 tornado 7928 5633 91346 62 12 49
## 2 lightning 3305 816 5230 19 12 55
## 3 tstm wind 2930 504 6957 23 12 49
## 4 flash flood 931 978 1777 19 12 50
## 5 flood 410 470 6789 19 12 48
## 6 high wind 525 248 1137 19 12 51
## Variables not shown: severity_score (dbl)head(econ_by_event_type)## Source: local data frame [6 x 8]
##
## EVTYPE event_count prop_damage corp_damage num_year num_month
## (chr) (int) (dbl) (dbl) (int) (int)
## 1 tornado 39361 3212258.2 100018.52 62 12
## 2 hail 25969 688693.4 579596.28 19 12
## 3 flash flood 20659 1420124.6 179200.46 19 12
## 4 flood 10058 899938.5 168037.88 19 12
## 5 tstm wind 61476 1335995.6 109202.60 14 12
## 6 lightning 10360 603351.8 3580.61 19 12
## Variables not shown: num_states (int), severity_score (dbl)As seen in the ranking above, tornado is the most harmful event related to both population health and economy. For the economic consequences, hail ranks the second whereas hail is ranked number 9 in terms of population health consequences. It makes sense as hail is usually not so much life threating events but it does cause tremendous amount of economic damages.
The findings are presented in following two graphs. Only top 10 events are presented to make it look clean. I took a chunk of code from internet for multiple plots in a chart.
## Taken from http://www.cookbook-r.com/Graphs/Multiple_graphs_on_one_page_(ggplot2)/
multiplot <- function(..., plotlist=NULL, file, cols=1, layout=NULL) {
library(grid)
plots <- c(list(...), plotlist)
numPlots = length(plots)
if (is.null(layout)) {
layout <- matrix(seq(1, cols * ceiling(numPlots/cols)),
ncol = cols, nrow = ceiling(numPlots/cols))
}
if (numPlots==1) {
print(plots[[1]])
} else {
grid.newpage()
pushViewport(viewport(layout = grid.layout(nrow(layout), ncol(layout))))
for (i in 1:numPlots) {
matchidx <- as.data.frame(which(layout == i, arr.ind = TRUE))
print(plots[[i]], vp = viewport(layout.pos.row = matchidx$row,
layout.pos.col = matchidx$col))
}
}
}Viz: Impact by different weather events by Year
final_summary_for_graph <- storm_dataset %>%
filter(
EVTYPE == 'tornado' |
EVTYPE == "flash flood" |
EVTYPE == "hail" |
EVTYPE == "tstm wind" |
EVTYPE == "lightning" |
EVTYPE == "high wind" |
EVTYPE == "heavy rain" |
EVTYPE == "excessive heat" |
EVTYPE == "string wind" |
EVTYPE == "thunderstorm wind"
) %>%
group_by(YEAR, EVTYPE) %>%
summarize(
prop_damage = sum(PROPDMG),
corp_damage = sum(CROPDMG),
fatal = sum(FATALITIES),
injur = sum(INJURIES),
severity_score = (prop_damage + corp_damage + fatal + injur)
) %>%
arrange(desc(severity_score))
p1 <- ggplot(final_summary_for_graph, aes(YEAR, fatal))
p1 <- p1 + geom_area(aes(colour = EVTYPE, fill= EVTYPE), position = 'stack')
p2 <- ggplot(final_summary_for_graph, aes(YEAR, injur))
p2 <- p2 + geom_area(aes(colour = EVTYPE, fill= EVTYPE), position = 'stack')
p3 <- ggplot(final_summary_for_graph, aes(YEAR, corp_damage))
p3 <- p3 + geom_area(aes(colour = EVTYPE, fill= EVTYPE), position = 'stack')
p4 <- ggplot(final_summary_for_graph, aes(YEAR, prop_damage))
p4 <- p4 + geom_area(aes(colour = EVTYPE, fill= EVTYPE), position = 'stack')
## Plot fatalities and injuries count by year
multiplot(p1, p2, cols=1)
## Plot crop and prop damages by year
multiplot(p3, p4, cols=1)