Storms and other severe weather events can cause both public health and economic problems for communities and municipalities. Many severe events can result in fatalities, injuries, and property damage, and preventing such outcomes to the extent possible is a key concern.
This project involves exploring the U.S. National Oceanic and Atmospheric Administration's (NOAA) storm database. This database tracks characteristics of major storms and weather events in the United States, including when and where they occur, as well as estimates of any fatalities, injuries, and property damage.
The analysis on the storm event database revealed that tornadoes are the most dangerous weather event to the populations health. The second most dangerous event type is excessive heat. The economic impact of weather events was also analyzed. Flash floods and thunderstorm winds caused billions of dollars in property damages between 1950 and 2011. The largest damage to crops were caused by droughts, followed by floods and hailing.
The analysis was performed on Storm Events Database, provided by National Climatic Data Center. The data is from a comma-separated-value file available here. There is also some documentation of the data available here.
The first step is to read the data into a data frame.
storm <- read.csv(bzfile("data/repdata-data-StormData.csv.bz2"))Before the analysis, the data need some preprocessing. Event types don't have a
specific format. For instance, there are events with types Frost/Freeze,
FROST/FREEZE and FROST\\FREEZE which refer to the same type of
event. So we will translate all letters to lowercase for uniformity.
# number of unique event types
length(unique(storm$EVTYPE))## [1] 985
# translate all letters to lowercase
event_types <- tolower(storm$EVTYPE)
# replace all punct. characters with a space
event_types <- gsub("[[:blank:][:punct:]+]", " ", event_types)
length(unique(event_types))## [1] 874
# update the data frame
storm$EVTYPE <- event_typesNo further data preprocessing was performed although the event type field can be
processed further to merge event types such as tstm wind and thunderstorm wind.
After the cleaning, as expected, the number of unique event types reduce
significantly. For further analysis, the cleaned event types are used.
To find the event types that are most harmful to population health, the number of casualties are aggregated by the event type.
library(plyr)
casualties <- ddply(storm, .(EVTYPE), summarize,
fatalities = sum(FATALITIES),
injuries = sum(INJURIES))
# Find events that caused most death and injury
fatal_events <- head(casualties[order(casualties$fatalities, decreasing = T), ], 10)
injury_events <- head(casualties[order(casualties$injuries, decreasing = T), ], 10)######Top 10 events that caused largest number of deaths are
fatal_events[, c("EVTYPE", "fatalities")] EVTYPE fatalities
741 tornado 5633
116 excessive heat 1903
138 flash flood 978
240 heat 937
410 lightning 816
762 tstm wind 504
154 flood 470
515 rip current 368
314 high wind 248
19 avalanche 224
######Top 10 events that caused most number of injuries are
injury_events[, c("EVTYPE", "injuries")] EVTYPE injuries
741 tornado 91346
762 tstm wind 6957
154 flood 6789
116 excessive heat 6525
410 lightning 5230
240 heat 2100
382 ice storm 1975
138 flash flood 1777
671 thunderstorm wind 1488
209 hail 1361
To analyze the impact of weather events on the economy, available property damage and crop damage reportings/estimates were used.
In the raw data, the property damage is represented with two fields, a number
PROPDMG in dollars and the exponent PROPDMGEXP. Similarly, the crop damage
is represented using two fields, CROPDMG and CROPDMGEXP. The first step in the
analysis is to calculate the property and crop damage for each event.
exp_transform <- function(e) {
# h -> hundred, k -> thousand, m -> million, b -> billion
if (e %in% c('h', 'H'))
return(2)
else if (e %in% c('k', 'K'))
return(3)
else if (e %in% c('m', 'M'))
return(6)
else if (e %in% c('b', 'B'))
return(9)
else if (!is.na(as.numeric(e))) # if a digit
return(as.numeric(e))
else if (e %in% c('', '-', '?', '+'))
return(0)
else {
stop("Invalid exponent value.")
}
}prop_dmg_exp <- sapply(storm$PROPDMGEXP, FUN=exp_transform)
storm$prop_dmg <- storm$PROPDMG * (10 ** prop_dmg_exp)
crop_dmg_exp <- sapply(storm$CROPDMGEXP, FUN=exp_transform)
storm$crop_dmg <- storm$CROPDMG * (10 ** crop_dmg_exp)# Compute the economic loss by event type
library(plyr)
econ_loss <- ddply(storm, .(EVTYPE), summarize,
prop_dmg = sum(prop_dmg),
crop_dmg = sum(crop_dmg))
# filter out events that caused no economic loss
econ_loss <- econ_loss[(econ_loss$prop_dmg > 0 | econ_loss$crop_dmg > 0), ]
prop_dmg_events <- head(econ_loss[order(econ_loss$prop_dmg, decreasing = T), ], 10)
crop_dmg_events <- head(econ_loss[order(econ_loss$crop_dmg, decreasing = T), ], 10)Top 10 events that caused most property damage (in dollars) are as follows
prop_dmg_events[, c("EVTYPE", "prop_dmg")] EVTYPE prop_dmg
138 flash flood 6.820237e+13
697 thunderstorm winds 2.086532e+13
741 tornado 1.078951e+12
209 hail 3.157558e+11
410 lightning 1.729433e+11
154 flood 1.446577e+11
366 hurricane typhoon 6.930584e+10
166 flooding 5.920826e+10
585 storm surge 4.332354e+10
270 heavy snow 1.793259e+10
Similarly, the events that caused biggest crop damage are
crop_dmg_events[, c("EVTYPE", "crop_dmg")] EVTYPE crop_dmg
84 drought 13972566000
154 flood 5661968450
519 river flood 5029459000
382 ice storm 5022113500
209 hail 3025974480
357 hurricane 2741910000
366 hurricane typhoon 2607872800
138 flash flood 1421317100
125 extreme cold 1312973000
185 frost freeze 1094186000
The following plot shows top dangerous weather event types.
library(ggplot2)
library(gridExtra)
# Set the levels in order
p1 <- ggplot(data=fatal_events,
aes(x=reorder(EVTYPE, fatalities), y=fatalities, fill=fatalities)) +
geom_bar(stat="identity") +
coord_flip() +
ylab("Total number of fatalities") +
xlab("Event type") +
theme(legend.position="none")
p2 <- ggplot(data=injury_events,
aes(x=reorder(EVTYPE, injuries), y=injuries, fill=injuries)) +
geom_bar(stat="identity") +
coord_flip() +
ylab("Total number of injuries") +
xlab("Event type") +
theme(legend.position="none")
grid.arrange(p1, p2, main="Top deadly weather events in the US (1950-2011)")
Figure 1: Top dangerous weather event types
Tornadoes cause most number of deaths and injuries among all event types. There are more than 5,000 deaths and more than 10,000 injuries in the last 60 years in US, due to tornadoes. The other event types that are most dangerous with respect to population health are excessive heat and flash floods.
The following plot shows the most severe weather event types with respect to economic cost that they have costed since 1950s.
library(ggplot2)
library(gridExtra)
# Set the levels in order
p1 <- ggplot(data=prop_dmg_events,
aes(x=reorder(EVTYPE, prop_dmg), y=log10(prop_dmg), fill=prop_dmg )) +
geom_bar(stat="identity") +
coord_flip() +
xlab("Event type") +
ylab("Property damage in dollars (log-scale)") +
theme(legend.position="none")
p2 <- ggplot(data=crop_dmg_events,
aes(x=reorder(EVTYPE, crop_dmg), y=crop_dmg, fill=crop_dmg)) +
geom_bar(stat="identity") +
coord_flip() +
xlab("Event type") +
ylab("Crop damage in dollars") +
theme(legend.position="none")
grid.arrange(p1, p2, main="Weather costs to the US economy (1950-2011)")
figure 2: The most severe weather event types with respect to
economic cost that they have costed since 1950s
Property damages are given in logarithmic scale due to large range of values. The data shows that flash floods and thunderstorm winds cost the largest property damages among weather-related natural diseasters.
The most severe weather event in terms of crop damage are droughts. In the last half century, droughts have caused more than 10 billion dollars damage. Other severe crop-damage-causing event types are floods and hailing.