Storm impact on the US population health and economy

Synopsis

In the current research the main goal was to apply reproducible research knowledge into the storms dataset. There were 2 main questions: which storm events affect the most on the public health and witch storm events affect the US economy the most. Results showed, that there are 15 which events, that cover more that 80% of all the impact on the public health and the US economy

Introduction

Before the data processing, first, specify the global options to show all the code and results

library(knitr)
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

opts_chunk$set(echo = TRUE, results = TRUE)

Data processing

The data for this assignment come in the form of a comma-separated-value file compressed via the bzip2 algorithm to reduce its size

fileUrl <- "https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2"
fileDest <- ("storm_data.csv.bz2")
if(!file.exists(fileDest)){
  download.file(fileUrl, fileDest)
}
storm <- read.csv("storm_data.csv.bz2")

Data manipulation

Most harmful storm events on the population health

Before analyzing, count the unique values of event types (EVTYPEs)

length(unique(storm$EVTYPE))

## [1] 985

Almost a thousand, which is many. Let’s count total injuries and fatal cases per each type and sort the result by descending order or fatal cases.

by_type <- storm %>%
  group_by(EVTYPE) %>%
  summarise(sum(INJURIES), sum(FATALITIES)) %>%
  rename(fatal = 'sum(FATALITIES)', injury = 'sum(INJURIES)') %>%
  filter(fatal != 0, injury !=0) %>%
  arrange(desc(fatal, injury))

## `summarise()` ungrouping output (override with `.groups` argument)

head(by_type)

To better understand the distribution of fatal cases, we can plot them.

plot(by_type$fatal, pch = 19, ylab = "fatal cases", xlab= "Event index", 
     main = "Distribution of fatal cases by event types")

From the plot, we see there are several event types, that has the most of fatal cases. To get the list of the events, that have the most impact, I will be using 80% rule, keep those types, that in total produce 80% of all the fatal cases.

Also, because only several events cover most of the distribution, there is no need to wrangle with other names of events.

topfatal <- by_type %>% 
  mutate(cumsum.prop = cumsum(fatal)/sum(fatal)) %>%
  filter(cumsum.prop <= 0.8)
topfatal

Thus, only 9 event types fit into the criteria and become most harmful on the population health, namely: tornado, excessive heat, flash flood, heat, lightning, tstm wind, flood, rip current, high wind.

Most harmful storm events on the US economy

In this section we will be calculating the impact on the economy by looking at the property and crop damage.

First we need to prepare the data to be proceeded.

economydmg <- select(storm, EVTYPE, PROPDMG, PROPDMGEXP, CROPDMG, CROPDMGEXP) %>%
  filter(PROPDMG != 0 | CROPDMG != 0)
head(economydmg)

unique(economydmg$PROPDMGEXP, economydmg)

##  [1] "K" "M" "B" "m" ""  "+" "0" "5" "6" "4" "h" "2" "7" "3" "H" "-"

Because the data has an exponent value, we need to create 2 new features, that multiply initial number into the exponent, where B or b = Billion, M or m = Million, K or k = Thousand, H or h = Hundred

Calculating property damage

propdmgcomb <- c()
for (i in 1:nrow(economydmg)){
  if(economydmg$PROPDMGEXP[i] == "K"){
    propdmgcomb[i] <- economydmg$PROPDMG[i] * 1000
  } else if(economydmg$PROPDMGEXP[i] == "m" | economydmg$PROPDMGEXP[i] == "M"){
    propdmgcomb[i] <- economydmg$PROPDMG[i] * 1000000
  } else if(economydmg$PROPDMG[i] == "B"){
    propdmgcomb[i] <- economydmg$PROPDMG[i] * 1000000000
  } else if(economydmg$PROPDMG[i] == "h" | economydmg$PROPDMG[i] == "H"){
    propdmgcomb[i] <- economydmg$PROPDMG[i] * 100
  } else {
    propdmgcomb[i] <- economydmg$PROPDMG[i]
  }
}
summary(propdmgcomb)

##      Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
##         0      2500     10000    618161     40000 929000000

Calculating crop damage

cropdmgcomb <- c()
for (i in 1:nrow(economydmg)){
  if(economydmg$CROPDMGEXP[i] == "K"){
    cropdmgcomb[i] <- economydmg$CROPDMG[i] * 1000
  } else if(economydmg$CROPDMGEXP[i] == "m" | economydmg$CROPDMGEXP[i] == "M"){
    cropdmgcomb[i] <- economydmg$CROPDMG[i] * 1000000
  } else if(economydmg$CROPDMGEXP[i] == "B"){
    cropdmgcomb[i] <- economydmg$CROPDMG[i] * 1000000000
  } else if(economydmg$CROPDMGEXP[i] == "h" | economydmg$CROPDMGEXP[i] == "H"){
    cropdmgcomb[i] <- economydmg$CROPDMG[i] * 100
  } else {
    cropdmgcomb[i] <- economydmg$CROPDMG[i]
  }
}
summary(cropdmgcomb)

##      Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
## 0.000e+00 0.000e+00 0.000e+00 2.004e+05 0.000e+00 5.000e+09

Now combine resulted vectors with an economydmg dataset

economydmg <- cbind(economydmg, propdmgcomb, cropdmgcomb)
names(economydmg)

## [1] "EVTYPE"      "PROPDMG"     "PROPDMGEXP"  "CROPDMG"     "CROPDMGEXP" 
## [6] "propdmgcomb" "cropdmgcomb"

Finally, calculate which storm type affect more on the economy by property damage and crop damage, and create a new feature, that combines them together.

totdamage <- economydmg %>%
  group_by(EVTYPE) %>%
  summarise(sum(propdmgcomb), sum(cropdmgcomb)) %>%
  rename(propdmgtot = 'sum(propdmgcomb)', cropdmgtot = 'sum(cropdmgcomb)') %>%
  mutate(totaldmg = propdmgtot + cropdmgtot) %>%
  arrange(-totaldmg)

## `summarise()` ungrouping output (override with `.groups` argument)

head(totdamage)

Now use the Pareto 80% rule to get a list of the most influential storm types

pareto_economy <- totdamage %>%
  mutate(dmg_cumsumprop = cumsum(totaldmg)/sum(totaldmg)) %>%
  filter(dmg_cumsumprop <= 0.8)

And make a plot with the final list

par(mar=c(8,5.5,3,2))
barplot(pareto_economy$totaldmg, names.arg = tolower(pareto_economy$EVTYPE)
        ,las = 2, main = "Storm types that affect the economy the most")
title(ylab="Damage", mgp=c(4,1,0))

Results

In the research we have found, that 9 storm types affect nearly 80% of public health. And similarly, 15 types affect 80 economy damage. In the table you can see final results

combine_result <- data.frame(rank = seq(1:9), Public.Health = topfatal$EVTYPE, 
           Economy.Damage = pareto_economy$EVTYPE)
inter <- intersect(combine_result$Public.Health, combine_result$Economy.Damage)
length(unique(c(topfatal$EVTYPE, pareto_economy$EVTYPE)))

## [1] 15

combine_result

There are 15 events, that cover more than 80% of all the public health and US economy together. But only 3 of them appear in the top 9 lists, that are: TORNADO, FLASH FLOOD, FLOOD