Reproducible Research Project 2: NOAA Storm Data

Synopsis

Storms exact massive tolls on the United States every year. This report aims to quantify the effects to inform policy decision.
We find that flooding stands out as the most damaging, with tornados, cold weather and fires as top contributors as well.

We begin with the Storm Database from the National Oceanic and Atmospheric Administration, which has tabulated over 250,000
storms between 1950 and 2011. We break effects from storms down into three areas: fatalities, injuries and cost of damage. Some
subjectivity was required when sorting storms by type as the NOAA data included many similar storm type names. The three deadliest
storm types were: tornadoes, heat events and flooding. The three most injurious were: flooding, cold events and fires. And finally, the costliest were: drought, fires and cold events.

Data Processing

The data is available from NOAA and needs to be downloaded and read into an object called ‘storms’:

library(tidyverse)
library(R.utils)
knitr::opts_chunk$set(error=FALSE, warning=FALSE, message=FALSE, echo=TRUE, cache.lazy=FALSE, fig.path='figures/')

setwd("C:/Users/m304650/Documents/RStudio Projects/Reproducible-Research-Activity-Project")

if (!file.exists("repdata_data_StormData.csv.bz2"))
{
        download.file("https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2", "repdata_data_StormData.csv.bz2")
}

if (!file.exists("repdata_data_StormData.csv"))
{
        bunzip2("./repdata_data_StormData.csv.bz2", remove=FALSE)
}


storms <- read_csv("repdata_data_StormData.csv", show_col_types=FALSE)

Now that the data has been read in, it needs to be tidied.

First we need to clean up the property and crop damage. The values are stored in two columns. The first a number from 1-999, then a separate column of B, M or K for billion, million or thousand. We will create new columns called PROPERTYDAMAGE and CROPDAMAGE
by multiplying the corresponding column by its power of 10.

storms <- storms %>%
        mutate(PROPERTYDAMAGE = case_when(
                PROPDMGEXP == "B" ~ PROPDMG * 1000000000,  # B -> Billion
                PROPDMGEXP == "M" ~ PROPDMG * 1000000,     # M -> Million
                PROPDMGEXP == "K" ~ PROPDMG * 1000,        # K -> Thousand
                TRUE ~ PROPDMG                       # Default case in case of any other value
))

storms <- storms %>%
        mutate(CROPDAMAGE = case_when(
                CROPDMGEXP == "B" ~ CROPDMG * 1000000000,  # B -> Billion
                CROPDMGEXP == "M" ~ CROPDMG * 1000000,     # M -> Million
                CROPDMGEXP == "K" ~ CROPDMG * 1000,        # K -> Thousand
                TRUE ~ CROPDMG                            # Default case in case of any other value
))

We can now remove all unnecessary columns. We only need a handful.

storms <- storms %>%
        select(BGN_DATE, EVTYPE, FATALITIES, INJURIES, PROPERTYDAMAGE, CROPDAMAGE)

storms <- subset(storms, FATALITIES > 0 | INJURIES > 0 | PROPERTYDAMAGE > 0 | CROPDAMAGE > 0)

We still need to clean up the event types. There are 977 unique EVTYPE values in this dataset. We will group them in broader categories

storms$EVTYPE <- toupper(storms$EVTYPE)

#Heat
storms$EVTYPE <- gsub('.*HEAT.*', 'HEAT', storms$EVTYPE)
storms$EVTYPE <- gsub('.*WARM.*', 'HEAT', storms$EVTYPE)

#Cold
storms$EVTYPE <- gsub('.*FREEZ.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*COLD.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*LO.*TEMP.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*BITTER.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*ICE.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*WINT.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*FROST.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*HYPOTH.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*ICY.*', 'COLD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*GLAZE.*', 'COLD', storms$EVTYPE)


#Drought
storms$EVTYPE <- gsub('.*DRY.*', 'DROUGHT', storms$EVTYPE)
storms$EVTYPE <- gsub('.*DROUGHT.*', 'DROUGHT', storms$EVTYPE)

#Fire
storms$EVTYPE <- gsub('.*FIRE.*', 'FIRE', storms$EVTYPE)


#Flood
storms$EVTYPE <- gsub('.*FLOOD.*', 'FLOOD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*TIDE.*', 'FLOOD', storms$EVTYPE)
storms$EVTYPE <- gsub('.*DAM.*', 'FLOOD', storms$EVTYPE)

#Landslide
storms$EVTYPE <- gsub('.*LANDSLIDE.*', 'LANDSLIDE', storms$EVTYPE)
storms$EVTYPE <- gsub('.*LANDSLUMP.*', 'LANDSLIDE', storms$EVTYPE)
storms$EVTYPE <- gsub('.*SLIDE.*', 'LANDSLIDE', storms$EVTYPE)

#Avalanche
storms$EVTYPE <- gsub('.*AVALANCHE.*', 'AVALANCHE', storms$EVTYPE)
storms$EVTYPE <- gsub('.*AVALANCE.*', 'AVALANCHE', storms$EVTYPE)

#Ocean
storms$EVTYPE <- gsub('.*SURF.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*MARINE.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*SEA.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*BEACH.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*COASTAL.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*WAVE.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*RIP.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*TSUNAMI.*', 'OCEAN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*SURGE.*', 'OCEAN', storms$EVTYPE)

#Snow
storms$EVTYPE <- gsub('.*SNOW.*', 'SNOW', storms$EVTYPE)
storms$EVTYPE <- gsub('.*BLIZZARD.*', 'SNOW', storms$EVTYPE)

#Tornado
storms$EVTYPE <- gsub('.*NADO.*', 'TORNADO', storms$EVTYPE)
storms$EVTYPE <- gsub('.*TORNDAO.*', 'TORNADO', storms$EVTYPE)
storms$EVTYPE <- gsub('.*LANDSPOUT.*', 'TORNADO', storms$EVTYPE)
storms$EVTYPE <- gsub('.*DUST DEVIL.*', 'TORNADO', storms$EVTYPE)

#Hail
storms$EVTYPE <- gsub('.*HAIL.*', 'HAIL', storms$EVTYPE)

#Hurricane
storms$EVTYPE <- gsub('.*HURRICANE.*', 'HURRICANE', storms$EVTYPE)
storms$EVTYPE <- gsub('.*TROPICAL.*', 'HURRICANE', storms$EVTYPE)
storms$EVTYPE <- gsub('.*TYPHOON.*', 'HURRICANE', storms$EVTYPE)

#Lightning
storms$EVTYPE <- gsub('.*LIGHTNING.*', 'LIGHTNING', storms$EVTYPE)
storms$EVTYPE <- gsub('.*THUNDER.*', 'LIGHTNING', storms$EVTYPE)


#Volcanic
storms$EVTYPE <- gsub('.*VOLCA.*', 'VOLCANIC', storms$EVTYPE)

#Wind
storms$EVTYPE <- gsub('.*WIND.*', 'WIND', storms$EVTYPE)

#Rain
storms$EVTYPE <- gsub('.*RAIN.*', 'RAIN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*BURST.*', 'RAIN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*SPOUT.*', 'RAIN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*WET.*', 'RAIN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*PRECIP.*', 'RAIN', storms$EVTYPE)
storms$EVTYPE <- gsub('.*HAIL.*', 'RAIN', storms$EVTYPE)

##Remove any one-off storm types left over
evtype_counts <- table(storms$EVTYPE)
unique_evtypes <- names(evtype_counts[evtype_counts == 1])
storms <- storms[!storms$EVTYPE %in% unique_evtypes, ]

Data Analysis

We want to look at different storm types and their fatalities, injuries and economic impact. A separate dataframe is made to
analyze these three impacts:

stormFatalities <- storms %>%
        group_by(EVTYPE) %>%
        summarize(AVGFATAL = mean(FATALITIES), TOTALFATAL = sum(FATALITIES)) %>%
        arrange(desc(TOTALFATAL))

stormInjuries <- storms %>%
        group_by(EVTYPE) %>%
        summarize(AVGINJURIES = mean(INJURIES), TOTALINJURIES = sum(INJURIES)) %>%
        arrange(desc(TOTALINJURIES))


stormCost <- storms %>%
        group_by(EVTYPE) %>%
        summarize(AVGCOST = mean(PROPERTYDAMAGE + CROPDAMAGE), TOTALCOST = sum(PROPERTYDAMAGE + CROPDAMAGE)) 

Results

Now that we have our tidy data, we want to visualize the 10 worst storms in each category. We will begin with the deadliest storms.

Storms by Fatalities

knitr::kable(stormFatalities[0:10,], caption = "10 Deadliest Storm Types")

10 Deadliest Storm Types

EVTYPE	AVGFATAL	TOTALFATAL
TORNADO	0.1413171	5663
HEAT	3.2133468	3178
FLOOD	0.0471918	1536
LIGHTNING	0.0147193	1018
WIND	0.0129691	948
COLD	0.2417521	872
OCEAN	0.6574288	854
SNOW	0.1177862	249
AVALANCHE	0.8333333	225
HURRICANE	0.2917271	201

ggplot(stormFatalities[0:10,], aes(x = reorder(EVTYPE, TOTALFATAL), y = TOTALFATAL, fill = EVTYPE)) +
        geom_bar(stat = "identity") +
        coord_flip() +  
        labs(title = "10 Deadliest Storm Types",
                x = "Storm",
                y = "Total Fatalities") +
        theme_minimal() +
        theme(axis.text.x = element_text(angle = 45, hjust = 1))

As you can see, tordanos kill the most Americans, followed by heat and flooding events. Four more: lightning, wind, cold and oceanic events make up most of the rest of fatalities.

Storms by Injuries

knitr::kable(stormInjuries[0:10,], caption = "10 Most Injurious Storms")

10 Most Injurious Storms

EVTYPE	AVGINJURIES	TOTALINJURIES
TORNADO	2.2820852	91450
HEAT	9.3458038	9243
WIND	0.1204837	8807
FLOOD	0.2646553	8614
LIGHTNING	0.1110886	7683
COLD	1.3146659	4742
SNOW	0.9262062	1958
RAIN	0.0645971	1802
HURRICANE	2.4905660	1716
FIRE	1.2772041	1608

ggplot(stormInjuries[0:10,], aes(x = reorder(EVTYPE, TOTALINJURIES), y = TOTALINJURIES, fill = EVTYPE)) +
        geom_bar(stat = "identity") +
        coord_flip() +  
        labs(title = "10 Most Injurious Storms",
                x = "Storm",
                y = "Total Injuries") +
        theme_minimal() +
        theme(axis.text.x = element_text(angle = 45, hjust = 1))

Similar results for injuries as seen for flooding, with tornadoes causing the most damage

Costliest Storms

knitr::kable(stormCost[0:10,], caption = "10 Costliest Storms")

10 Costliest Storms

EVTYPE	AVGCOST	TOTALCOST
AVALANCHE	32302.96	8721800
COLD	5413232.37	19525529161
DENSE FOG	130729.73	9674000
DROUGHT	43551940.87	15025419600
DUST STORM	83970.87	8649000
FIRE	7073002.49	8904910130
FLOOD	5670187.33	184553257197
FOG	122948.60	13155500
FUNNEL CLOUD	14969.23	194600
HEAT	935091.03	924805030

costliestStorms <- stormCost[0:10, ]
        ggplot(stormCost[0:10, ], aes(x = reorder(EVTYPE, TOTALCOST), y = TOTALCOST, fill = EVTYPE)) +
        geom_bar(stat = "identity") +
        coord_flip() +  
        labs(title = "10 Costliest Storms",
                x = "Storm",
                y = "Total Cost ($)") +
        theme_minimal() +
        theme(axis.text.x = element_text(angle = 45, hjust = 1))

We see that flooding is by far the costliest storm type. Almost to an extenct that others do not matter.