A 5-Year Analysis of the Impact of Severe Weather Events throughout the United States

Synopsis

The objective of this analysis was to determine which types of storm events in the United States have been the most harmful to population health and which types have had the greatest economic consequences. The data included storm events from 2007 through 2011 taken from the NOAA Storm Event Database. The 10 most harmful and 10 most damaging storming events will be reported, and the states experiencing the largest impact from each of these event lists will be identified. The data were processed using R version 3.3.1 (2016-06-21).

The most harmful storm event was the tornado, producing in excess of 10,000 fatalities and injuries during this 5-year period. The most damaging storm event in terms of economic impact was flooding, result in over $15 billion of damage, followed closely by tornadoes.The states with the largest number of harmful events were Alabama and Missouri, while the state suffering the most economic damage was Texas.

Data Processing

The data used for this analysis were taken from the United States National Oceanic and Atmospheric Administration’s (NOAA) Storm Event Database. This database has been used since 1950 to record data about major storms and weather events in the U.S, such as date, location, magnitude, and estimates of fatalities, injuries, and property damage which are related to the event. There are some limitations to this database:

Only tornado events were reported from 1950 to 1954
Tornado, thunderstorm wind, and hail events were reported from 1955 to 1992
Additional events have been reported from 1993 through present, however, events prior to 2007 have been inconsistently recorded
Data from 2007 to 2011 have been tidied and grouped into 48 storm event categories, outlined by the NOAA and listed below.

The data were processed using R version 3.3.1 (2016-06-21) on a x86_64-w64-mingw32 system. Several additional packages were used to tidy the raw data and prepare the report.

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, union

library(lubridate)

## 
## Attaching package: 'lubridate'

## The following object is masked from 'package:base':
## 
##     date

library(stringr)
library(ggplot2)
library(maps)
library(grid)
library(gridExtra)

## 
## Attaching package: 'gridExtra'

## The following object is masked from 'package:dplyr':
## 
##     combine

library(pander)
library(RColorBrewer)

panderOptions('knitr.auto.asis', TRUE)
panderOptions('round', 1)
panderOptions('table.style', 'multiline')
panderOptions('table.split.table', Inf)
panderOptions('list.style', 'ordered')

all_state <- map_data("state")
states <- state.abb
dc <- data.frame(abb = "DC", region = "district of columbia")
states.dc <- data.frame(abb = state.abb, region = str_to_lower(state.name[match(states, state.abb)])) %>%
    bind_rows(dc)

## Warning in bind_rows_(x, .id): Unequal factor levels: coercing to character

## Warning in bind_rows_(x, .id): Unequal factor levels: coercing to character

noaa.events <- read.csv("noaa event types.csv", colClasses = "character") %>%
    mutate(event.type = str_to_title(event.type))

NOAA Storm Event Types

pandoc.list(as.list(noaa.events$event.type))

Astronomical Low Tide
Avalanche
Blizzard
Coastal Flood
Cold/Wind Chill
Debris Flow
Dense Fog
Dense Smoke
Drought
Dust Devil
Dust Storm
Excessive Heat
Extreme Cold/Wind Chill
Flash Flood
Flood
Frost/Freeze
Funnel Cloud
Freezing Fog
Hail
Heat
Heavy Rain
Heavy Snow
High Surf
High Wind
Hurricane (Typhoon)
Ice Storm
Lake-Effect Snow
Lakeshore Flood
Lightning
Marine Hail
Marine High Wind
Marine Strong Wind
Marine Thunderstorm Wind
Rip Current
Seiche
Sleet
Storm Surge/Tide
Strong Wind
Thunderstorm Wind
Tornado
Tropical Depression
Tropical Storm
Tsunami
Volcanic Ash
Waterspout
Wildfire
Winter Storm
Winter Weather

raw <- read.csv("repdata-data-StormData.csv.bz2", colClasses = "character")

Scope of the Analysis

The analysis was limited to events which occurred between 2007 and 2011. This date range was selected because the data during this period followed stricter NOAA specifications for categorizing storm events, resulting in significantly fewer inconsistencies than earlier time periods. Furthermore, historical data was limited to selected storm event types (e.g., only tornado, thunderstorm wind, and hail were reported from 1955 to 1992), which could bias the results towards these events.

Tidying the Data

Initial processing of the raw data began by assigning variables to the proper classes. The data were limited to events occurring on or after January 1, 2007. The full state names were added to the data set, which will be used to map the prevalence of events on a state-by-state basis.

data <- raw %>%
    transmute(begin.date = mdy_hms(BGN_DATE),
              end.date = mdy_hms(END_DATE),
              state = STATE,
              county = COUNTYNAME,
              event = str_to_title(str_trim(EVTYPE, side = "both")),
              tornado.strength = as.numeric(F),
              magnitude = as.numeric(MAG),
              fatalities = as.numeric(FATALITIES),
              injuries = as.numeric(INJURIES),
              property = as.numeric(PROPDMG),
              pde = ifelse(PROPDMGEXP == "", NA, str_to_lower(PROPDMGEXP)),
              crop = as.numeric(CROPDMG),
              cde = ifelse(CROPDMGEXP == "", NA, str_to_lower(CROPDMGEXP))) %>%
    filter(begin.date >= mdy("1/1/2007")) %>%
    left_join(states.dc, by=c("state" = "abb"))

To evaluate property and crop damage, the numerical damage figure was multiplied by the exponent to calculate the actual amount of damage. The following definitions were used for the exponents:

h = hundreds
k = thousands
m = millions
b = billions

For any other character in the exponent, the damage value was multiplied by 10.

This was determined by cross-referencing damage values with the online NOAA Storm Event Database, which indicated that these miscellaneous characters were erroneously placed in the damage exponent column when in fact they should have been the last digit of the damage value. Since these values were very small, the damage value was rounded to the approximate value through multiplying by 10.

data <- mutate(data, pde = str_replace(pde, "([^bhkm])", 10),
              pde = str_replace(pde, "h", 100),
              pde = str_replace(pde, "k", 1000),
              pde = str_replace(pde, "m", 1000000),
              pde = str_replace(pde, "b", 1000000000),
              pde = as.numeric(pde),
              property.damage = ifelse(!is.na(pde), property * pde, property),
              cde = str_replace(cde, "([^bhkm])", 10),
              cde = str_replace(cde, "h", 100),
              cde = str_replace(cde, "k", 1000),
              cde = str_replace(cde, "m", 1000000),
              cde = str_replace(cde, "b", 1000000000),
              cde = as.numeric(cde),
              crop.damage = ifelse(!is.na(cde), crop * cde, crop))

Summarizing the Data

To summarize the data, the total number of fatalities, injuries, property damage, and crop damage for each storm event were calculated. The total harm of the event was then determined by adding the number of fatalities and injuries. The total damage produced by the event was determine by adding the amount of property damage with the amount of crop damage.

events.total <- group_by(data, event) %>%
    summarize(count = n(),
              fatalities = sum(fatalities),
              injuries = sum(injuries),
              property.damage = sum(property.damage),
              crop.damage = sum(crop.damage)) %>%
    mutate(harm = fatalities + injuries,
           damage = property.damage + crop.damage)

Results

Events included in this analysis occurred between 2007 and 2011. In this analysis, harmful to the population health was defined as causing a fatality or injury, while economic consequence was defined as causing damage to property or crops.

The Most Harmful and Most Damaging Storm Event Types

The total harm and damage were calculated for each event type, and the top 10 events in each group were selected. The total amount of damge will be reported as $Billions. The 10 most harmful event types and 10 most damaging event types are reported in figure 1 below.

A full list of results for all 48 storm event types can be found in table 1 in the appendix at the end of this document.

top.harm <- select(events.total, event, harm) %>%
    arrange(desc(harm)) %>%
    top_n(10, harm)

top.damage <- select(events.total, event, damage) %>%
    mutate(damage = damage / 1000000000) %>%
    arrange(desc(damage)) %>%
    top_n(10, damage)

cols <- brewer.pal(5, "Blues")
graph1 <- ggplot(top.harm, aes(x=event, y=harm)) +
    geom_bar(stat="identity", fill=cols[3], color="black") + 
    ggtitle("A. Most Harmful Storm Events") +
    xlab("Storm Event") +
    ylab("Number of Harmful Events") +
    scale_x_discrete(limits=top.harm$event) +
    theme_bw() +
    theme(axis.text.x = element_text(angle=30, hjust=1, vjust=1))

cols <- brewer.pal(5, "Greens")
graph2 <- ggplot(top.damage, aes(x=event, y=damage)) +
    geom_bar(stat="identity", fill=cols[3], color="black") + 
    ggtitle("B. Most Damaging Storm Events") +
    xlab("Storm Event") +
    ylab("Damage Totals in Billions ($)") +
    scale_x_discrete(limits=top.damage$event) +
    theme_bw() +
    theme(axis.text.x = element_text(angle=30, hjust=1, vjust=1))

grid.arrange(graph1, graph2, ncol = 2, top="Figure 1. Top 10 Most Harmful and Most Damaging Events from 2007 to 2011")

States with the Highest Prevelance of Harmful and Damaging Storm Events

The 10 most harmful and 10 most damaging storm event types were then totaled on a state-by-state basis to determine the prevalence of each event type within the 50 U.S. states and District of Columbia. The incidence of these event types, ranging from least to most, can be found in figure 2 below.

state.harm <- filter(data, event %in% top.harm$event) %>%
    group_by(region) %>%
    summarize(count = n(),
          fatalities = sum(fatalities),
          injuries = sum(injuries)) %>%
    mutate(harm = fatalities + injuries) %>%
    arrange(harm) %>%
    inner_join(all_state, by="region")

state.damage <- filter(data, event %in% top.damage$event) %>%
    group_by(region) %>%
    summarize(count = n(),
              property.damage = sum(property.damage),
              crop.damage = sum(crop.damage)) %>%
    mutate(damage = (property.damage + crop.damage)/1000000000) %>%
    arrange(damage) %>%
    inner_join(all_state, by="region")

cols <- brewer.pal(5, "Blues")
graph1 <- ggplot() +
    geom_polygon(data=state.harm, aes(x=long, y=lat, group=group, fill=harm), colour="white") +
    scale_fill_continuous(low=cols[1], high=cols[5], guide="colorbar") +
    labs(title="A. Most Harmful Events", x="", y="", fill="Number of Harmful Events") +
    scale_x_continuous(breaks = NULL) +
    scale_y_continuous(breaks = NULL) +
    theme_bw() +
    theme(legend.position="bottom")

cols <- brewer.pal(5, "Greens")
graph2 <- ggplot() +
    geom_polygon(data=state.damage, aes(x=long, y=lat, group=group, fill=damage), colour="white") +
    scale_fill_continuous(low=cols[1], high=cols[5], guide="colorbar") +
    labs(title="B. Most Damaging Events", x="", y="", fill="Cost of Damaging Events\nin $Billion") +
    scale_x_continuous(breaks = NULL) +
    scale_y_continuous(breaks = NULL) +
    theme_bw() +
    theme(legend.position="bottom")

grid.arrange(graph1, graph2, ncol=2, top="Figure 2. Top 10 Most Harmful and Most Damaging Events by State")

References

Appendix

all.harm <- select(events.total, event, harm) %>%
    arrange(desc(harm)) 

all.damage <- select(events.total, event, damage) %>%
    mutate(damage = damage / 1000000000) %>%
    arrange(desc(damage)) 

all.harm <- bind_cols(all.harm, all.damage)

colnames(all.harm) <- c("Harmful Events","Harm Produced","Damaging Events","Damage Produced")
set.caption("Table 1. List of Harm and Damage Produced by Storm Events")
pander(all.harm)

Table 1. List of Harm and Damage Produced by Storm Events
Harmful Events	Harm Produced	Damaging Events	Damage Produced
Tornado	10471	Flood	16.85541580
Thunderstorm Wind	1521	Tornado	14.73228374
Lightning	1082	Hail	6.96779060
Excessive Heat	999	Flash Flood	5.75261413
Heat	884	Storm Surge/Tide	4.64149300
Flash Flood	609	Thunderstorm Wind	3.77156069
Wildfire	455	Hurricane	2.64811000
Winter Weather	354	Wildfire	2.22150747
Rip Current	334	High Wind	1.29262904
Flood	332	Frost/Freeze	0.94128100
Strong Wind	184	Winter Storm	0.90929000
Hail	182	Ice Storm	0.78004830
High Surf	177	Tropical Storm	0.44350735
High Wind	163	Drought	0.42650500
Tsunami	162	Lightning	0.30359543
Avalanche	138	Coastal Flood	0.16454656
Cold/Wind Chill	105	Heavy Rain	0.14039803
Winter Storm	55	Landslide	0.13581050
Extreme Cold/Wind Chill	53	Tsunami	0.13487900
Heavy Rain	52	High Surf	0.08296750
Heavy Snow	44	Heavy Snow	0.06781910
Marine Thunderstorm Wind	35	Strong Wind	0.06150406
Marine Strong Wind	34	Winter Weather	0.03417850
Dust Devil	26	Blizzard	0.02874100
Tropical Storm	21	Lake-Effect Snow	0.01670700
Ice Storm	17	Lakeshore Flood	0.00753000
Landslide	16	Extreme Cold/Wind Chill	0.00703800
Storm Surge/Tide	16	Waterspout	0.00520120
Hurricane	15	Dust Storm	0.00317800
Dust Storm	11	Dense Fog	0.00276200
Blizzard	8	Cold/Wind Chill	0.00259000
Coastal Flood	4	Avalanche	0.00238580
Dense Fog	3	Freezing Fog	0.00218200
Marine High Wind	2	Heat	0.00161000
Astronomical Low Tide	0	Tropical Depression	0.00130200
Dense Smoke	0	Excessive Heat	0.00123320
Drought	0	Marine High Wind	0.00114001
Freezing Fog	0	Marine Thunderstorm Wind	0.00048540
Frost/Freeze	0	Marine Strong Wind	0.00040233
Funnel Cloud	0	Dust Devil	0.00038113
Lake-Effect Snow	0	Astronomical Low Tide	0.00032000
Lakeshore Flood	0	Dense Smoke	0.00010000
Marine Hail	0	Seiche	0.00009000
Seiche	0	Funnel Cloud	0.00006510
Sleet	0	Marine Hail	0.00000400
Tropical Depression	0	Rip Current	0.00000100
Volcanic Ashfall	0	Sleet	0.00000000
Waterspout	0	Volcanic Ashfall	0.00000000