The basic goal of this report is to explore the National Weather Service Storm Database and answer some basic questions about severe weather events. The data analysis aims to answer the following questions.
Across the United States, which types of events (as indicated in the EVTYPE variable) are most harmful with respect to population health?
Across the United States, which types of events have the greatest economic consequences?
The data was obtained from the database of National Weather Service which contained natural events and their effects on the United States from 1950 to 2011. The data and the documentation are available from the course website of Data Science Specialization on Coursera. Only data from 1996 were considered as previous years did not record all types of events. Out of the initial 902297 observations, 201313 observations were considered which were recorded after 1996 and had positive number of fatalities, injuries, property damage, or crop damage. The analysis showed that tornadoes caused the most injuries whereas excessive heat caused the most fatalities. Furthermore, droughts caused the most amount of crop damage and floods caused the most property damage.
From the 37 variables in the original dataset, 9 were utilized to answer the two questions above. We list below the 9 variables in the format of
First, we filtered the data to only include events from 1996 to 2011 because not all types of weather events were recorded prior to 1996. The questions could be answered with simple summary statistics of the data. We calculated the total number of fatalities and injuries for each event type from 1996 to 2011 and plotted 10 observations with the highest number of fatalities and injuries. Similarly, we calculated the total amount of crop damage and property damage (scaled by the exponent variables) and observed the top 10 event types. Afterwards, out of curiosity, we repeated the calculations but for each state rather than the type of weather event, and plotted a choropleth map using ggplot2 and plotly.
For the effect on public health, there are two types of harm to be considered which includes fatalities and injuries. From 1996 to 2011 in the United States, tornadoes resulted in the highest number of injuries (20667), whereas excessive heat resulted in the highest number of fatalities (2034).
Below is a map showing the total fatalities and injuries by state. Interestingly, Texas seems to have been impacted the most, with almost ten thousand total number of fatalities and injuries (9978).
For economic consequence, floods caused the most amount of property damage from 1996 to 2011 with almost $160 billion worth of damage. With respect to crops, drought was the single greatest cause of damage, costing almost $14 billion in damage.
Whereas Texas saw the largest number of fatalities and injuries from 1996 to 2011, California took the biggest hit economically during the same period of time, with over $125 billion in property and crop damage combined.
In this report, we used the National Weather Services’ Storm database to find out which type of weather events caused the greatest damage in population health as well as the economy of U.S. From this brief analysis, we conclude that tornadoes, excessive heat, floods, droughts caused the greatest amount of damage to public health as well as the economy from 1996 to 2011. It would be greatly beneficial to focus our attention and efforts into dealing with the 4 weather events above in order to prevent or alleviate health and economic consequences from any future events. However, it should also be noted that this report only takes an initial look at the numbers without consideration of the magnitude or the location of these events. For future analysis, it would be beneficial to expand the analysis to include variables such as the magnitude, location, and the duration.
The code for the analysis is included below along with explanations.
To begin the analysis, I loaded in the .csv file and some packages.
# Download file
linkurl <- "https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2"
download.file(linkurl, "stormdata.csv")
# Load the data
storm <- read.csv("stormdata.csv", stringsAsFactors = FALSE)library(dplyr)
library(lubridate)
library(ggplot2)
library(reshape2)
library(plotly)Subsequently, I created a new data frame containing only the nine relevant variables out of 37 that were provided. I further subsetted by including only the observations with fatalities, injuries, property damage, or crop damage greater than 0. All variable names were formatted to proper format and variation in the event type (i.e “flashflood” “flood”) were removed.
# Make new data frame with the relevant variables.
after1996 <- storm %>%
select(BGN_DATE, STATE, EVTYPE, FATALITIES, INJURIES,
PROPDMG, PROPDMGEXP, CROPDMG, CROPDMGEXP) %>%
# Subset with only the rows containing data from January 1996
# This is when they started collecting all event types
filter(mdy_hms(BGN_DATE) >= "1996-01-01",
FATALITIES > 0 |
INJURIES > 0 |
CROPDMG > 0 |
PROPDMG > 0
)
# Clean up the variable names, format errors in event type 'evtype'
names(after1996) <- tolower(names(after1996))
names(after1996) <- sub("_", ".", names(after1996))
after1996$evtype <- tolower(after1996$evtype)
after1996$evtype <- gsub(" ", "", after1996$evtype)
after1996$evtype <- gsub("-", "", after1996$evtype)
after1996$evtype <- sub("^hurricane.*|typhoon", "hurricane", after1996$evtype)
after1996$evtype <- sub(".*flood$", "flood", after1996$evtype)
after1996$evtype <- sub("tstmwind.*|thunderstorm.*", "thunderstormwind", after1996$evtype)
after1996$evtype <- sub("^winterweather.*|wintrymix", "winterweather", after1996$evtype)
after1996$evtype <- sub("^wild.*", "wildfire", after1996$evtype)
after1996$evtype <- sub("^freez.*|^frost.*|.*frost$", "frost/freeze", after1996$evtype)
after1996$evtype <- sub("^cold.*", "cold/wind chill", after1996$evtype)
after1996$evtype <- sub(".*cold$|^extreme[.]cold.*|excessivesnow|extremewindchill", "extreme cold/wind chill", after1996$evtype)
after1996$evtype <- sub("^heat.*", "excessiveheat", after1996$evtype)
after1996$evtype <- sub("^ripcurrent.*", "ripcurrent", after1996$evtype)
after1996$evtype <- sub("^stormsurge.*", "stormsurge/tide", after1996$evtype)# Plot for the top 10 causes of fatalities and injuries
population.health <- after1996 %>%
group_by(evtype) %>%
summarise(
fatalities = sum(fatalities),
injuries = sum(injuries)
) %>%
arrange(desc(fatalities), desc(injuries))
#filter out for the events with the most fatalities and injuries
top10 <- head(population.health, n = 10)
# Create long version for plotting
top10.long <- melt(top10, id.vars = "evtype")
# Plot
g <- ggplot(data = top10.long,
aes(x = evtype, y = value, fill = variable)) +
geom_bar(stat = 'identity', position = 'dodge') +
labs(title = "Total Fatalities and Injuries from 1996-2011",
x = "Event Type", fill = "Type") + coord_flip()
ggplotly(g, width = 900, height = 480)# Choropleth map for Fatalities and Injuries
# Create data frame
by_state <- after1996 %>%
group_by(state) %>%
summarise(
fatalities = sum(fatalities),
injuries = sum(injuries)
) %>%
filter(state %in% state.abb) %>%
mutate(total = fatalities + injuries)
# Create hover text
by_state$hover <- with(by_state, paste("Fatalities:", fatalities,
'<br>', "Injuries:", injuries,
'<br>', "Total:", total))
# Make state borders red
borders <- list(color = toRGB("red"))
# Set up some mapping options
map_options <- list(
scope = 'usa',
projection = list(type = 'albers usa'),
showlakes = TRUE,
lakecolor = toRGB('white')
)
plot_ly(width = 900, height = 480,
z = ~by_state$total, text = ~by_state$hover, locations = ~by_state$state,
type = 'choropleth', locationmode = 'USA-states',
color = by_state$total, colors = 'Greens', marker = list(line = borders)) %>%
colorbar(title = "Total") %>%
layout(title = 'Total Number of Fatalities and Injuries from 1996 to 2011', geo = map_options)# Plot for the top 10 causes of property and crop damage
# Create a key for exponents, replace blank strings with O
exponent.key <- c(O = 1, K = 1000, M = 1000000, B = 1000000000)
after1996$cropdmgexp[after1996$cropdmgexp == ""] <- "O"
after1996$propdmgexp[after1996$propdmgexp == ""] <- "O"
# Create new variables that sum the total property and crop damage
econ.dmg <- after1996 %>%
group_by(evtype) %>%
summarise(
property = sum(propdmg * exponent.key[propdmgexp]),
crop = sum(cropdmg * exponent.key[cropdmgexp])
)
top10prop <- head(arrange(econ.dmg, desc(property)), n = 10)[,1:2]
top10crop <- head(arrange(econ.dmg, desc(crop)), n = 10)[,c(1,3)]
# Plot property damage
gprop <- ggplot(data = top10prop,
aes(x = evtype, y = property)) +
geom_bar(stat = 'identity', position = 'dodge', fill = "darkred") +
labs(title = "Total Property Damage from 1996-2011",
x = "Event Type", y = "Total Damage in U.S Dollar") + coord_flip()
ggplotly(gprop, width = 900, height = 480)
# Plot crop damage
gcrop <- ggplot(data = top10crop,
aes(x = evtype, y = crop)) +
geom_bar(stat = 'identity', position = 'dodge', fill = "darkblue") +
labs(title = "Total Crop Damage from 1996-2011",
x = "Event Type", y = "Total Damage in U.S Dollar") + coord_flip()
ggplotly(gcrop, width = 900, height = 480)# Choropleth map for property and crop damage
# Create a key for exponents, replace blank strings with O
exponent.key <- c(O = 1, K = 1000, M = 1000000, B = 1000000000)
after1996$cropdmgexp[after1996$cropdmgexp == ""] <- "O"
after1996$propdmgexp[after1996$propdmgexp == ""] <- "O"
# Create new variables that sum the total property and crop damage
by_state2 <- after1996 %>%
group_by(state) %>%
summarise(
property = sum(propdmg * exponent.key[propdmgexp]),
crop = sum(cropdmg * exponent.key[cropdmgexp])
) %>%
filter(state %in% state.abb) %>%
mutate(total = property + crop)
# Create hover text
by_state2$hover <- with(by_state2, paste("Property:", property,
'<br>', "Crop:", crop,
'<br>', "Total:", total))
# Make state borders red
borders <- list(color = toRGB("red"))
# Set up some mapping options
map_options <- list(
scope = 'usa',
projection = list(type = 'albers usa'),
showlakes = TRUE,
lakecolor = toRGB('white')
)
plot_ly(width = 900, height = 480,
z = ~by_state2$total, text = ~by_state2$hover, locations = ~by_state2$state,
type = 'choropleth', locationmode = 'USA-states',
color = by_state2$total, colors = 'Greens', marker = list(line = borders)) %>%
colorbar(title = "Total") %>%
layout(title = 'Total Property and Crop Damage in USD from 1996 to 2011', geo = map_options)