Synopsis

This document is the result of a exploratory analysis over the U.S. National Oceanic and Atmospheric Administration’s (NOAA) storm database, and the purpose of it was to find which types of events (natural disaster) have been the most harmful to human healt and which of them have had the highest economic repercussions. This was done by first loading the data and using only the columns that the author think are necessary for the analysis. The cost to human health metric was calculated by giving a value to each according to the number of fatalities and injuries it has caused. The economic repercussions were calculated by adding the value of the damages to both property and crops. In the two questions, a plot and a table were made to visualize the results of the top 10 events.

Data Processing

The data is stored in the storm_data.csv file inside the data folder. This file was imported to R using the fread function and stored in the dat variable. This function is used because it is faster compared to read.CSV and a larga data set is being used. Only seven columns are going to be used for the analysis, they are all converted to lowercase in the preprocessing. These are:

It is also worth noting that all the used libraries are imported right away.

library(data.table)
library(dplyr)
## 
## Attaching package: 'dplyr'
## The following objects are masked from 'package:data.table':
## 
##     between, first, last
## The following objects are masked from 'package:stats':
## 
##     filter, lag
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union
library(ggplot2)
library(viridis)
## Loading required package: viridisLite
library(pander)
cols.keep <- c("EVTYPE", "FATALITIES", "INJURIES", "PROPDMG", "PROPDMGEXP",
               "CROPDMG", "CROPDMGEXP") 
dat <- fread("data/storm_data.csv", select=cols.keep)
names(dat) <- c("evtype", "fatalities", "injuries", "propdmg", "propdmgexp",
                "cropdmg", "cropdmgexp")

Human health cost

The human health cost was calculated by creating a new quantity called health.score, this metric depends on the number on injuries and the number of fatalities for an event. The first thing to do was to calculate the ratio of injuries to fatalities as an integer.

injury_fatal.ratio <- floor(sum(dat$injuries)/sum(dat$fatalities))

This quantity indicates how many more injuries are compared to the number of deaths in total. In the data set of our interest said ratio is: 9.

After that, health score will be calculated following the formula:

health.score = (f*r) + i

Where \(f\) stands for fatalities, \(i\) is the injuries, and \(r\) is the aforementioned ratio.

A new dataframe called health_cost is created using the dplyr library and the aggregate method from base R. First the dat original table is taken and then only the variables evtype, fatalities and injuries are extracted. Then the health.score is calculated using the formula. Finally this new dataframe is ordered according to health.score on a descending manner, and re-scaled the whole values to thousands.

health_cost <- dat %>% select(evtype, fatalities, injuries) %>%
    mutate(health.score = (fatalities*injury_fatal.ratio) + injuries)
health_cost <- aggregate(list(health_cost$fatalities, health_cost$injuries,
                              health_cost$health.score),
                         FUN=sum, na.rm=T, by=list(evtype=health_cost$evtype))
names(health_cost)[2:4] <- c("fatalities", "injuries", "health.score")
health_cost <- health_cost[order(-health_cost$health.score),]
row.names(health_cost) <- NULL

# Re scale to thousands
health_cost[,2:4] <- lapply(health_cost[,2:4], function(x) {
    x/1000.0
})

To plot the data, the first ten rows are extracted and saved into a temporary symbol. After that, the ggplot2 package is used to make the desirable graphic.

a <- head(health_cost, 10)
pander(a, type="grid")
evtype fatalities injuries health.score
TORNADO 5.633 91.35 142
EXCESSIVE HEAT 1.903 6.525 23.65
LIGHTNING 0.816 5.23 12.57
TSTM WIND 0.504 6.957 11.49
FLOOD 0.47 6.789 11.02
FLASH FLOOD 0.978 1.777 10.58
HEAT 0.937 2.1 10.53
RIP CURRENT 0.368 0.232 3.544
HIGH WIND 0.248 1.137 3.369
WINTER STORM 0.206 1.321 3.175 
plt <- ggplot(data=a, aes(x=injuries, y=fatalities, col=health.score)) +
    geom_point(size=3) + scale_color_gradientn(colours=viridis(10)) +
    geom_text(aes(label=evtype), size=3, vjust="inward", hjust="inward") +
    theme_grey(base_size = 14) + xlab("Injuries (thousands") + ylab("Fatalities (thousands)")

print(plt)
Ten most harmful events to human health

Ten most harmful events to human health

Economic consecuences

The conomic consequences are calculated by the total sum of the property damage and the crop damage. To obtain this number the damage values should be sclaed accordingly to the associated exponent. The code shown below performs this task and reassings the new values in the data table, then deletes both the exponents columns since they are no longer needed.

# Function to replace the exponents
replace_exp <- function(x) {
    if (x=="K" | x=="k") {
        return(1000)
    }
    else if (x=="M" | x=="m") {
        return(1000000)
    }
    else {
        return(1)
    }
}
# Only data with kilo or mega exp will be processed
# The other ones could not be understood an so we will asume them as 1
dat$propdmgexp <- sapply(dat$propdmgexp, replace_exp)
dat$cropdmgexp <- sapply(dat$cropdmgexp, replace_exp)

# Multiplying the value for the correspondent exponent and dropping exp columns
dat$propdmg <- dat$propdmg * dat$propdmgexp
dat$cropdmg <- dat$cropdmg * dat$cropdmgexp
dat[,c("propdmgexp", "cropdmgexp"):=NULL]

The next step is to create a new dataframe called monetary, this object is used to store the damage values. Said values are obtained using a dplyr pipeline to extract the needed columns and calculate the new total value -which is the sum of property plus crop damage-. Then the base R aggegate function is used to obtain a total value per event. Finally all those values are re-scaled to millions.

# Calculating the dmg values
monetary <- dat %>% select(evtype, propdmg, cropdmg) %>%
    mutate(total = propdmg + cropdmg)
monetary <- aggregate(list(monetary$propdmg, monetary$cropdmg,
                           monetary$total),
                         FUN=sum, na.rm=T, by=list(evtype=monetary$evtype))
names(monetary)[2:4] <- c("propdmg", "cropdmg", "total")
monetary <- monetary[order(-monetary$total),]
row.names(monetary) <- NULL
# Re scale to millions
monetary[,2:4] <- lapply(monetary[,2:4], function(x) {
    x/1000000.0
})

To plot the data, the first ten rows are extracted and saved into a temporary symbol. After that, the ggplot2 package is used to make the desirable graphic.

b <- head(monetary, 10)
pander(b, type="grid")
evtype propdmg cropdmg total
TORNADO 51637 415 52052
FLOOD 22158 5662 27820
HAIL 13932 3026 16958
FLASH FLOOD 15141 1421 16562
DROUGHT 1046 12473 13519
HURRICANE 6168 2742 8910
TSTM WIND 4485 554 5039
HURRICANE/TYPHOON 3806 1098 4904
HIGH WIND 3970 638.6 4609
WILDFIRE 3725 295.5 4021 
plt <- ggplot(data=b, aes(x=propdmg, y=cropdmg, col=total)) +
    geom_point(size=3) + scale_color_gradientn(colours=viridis(10)) +
    geom_text(aes(label=evtype), size=3, vjust="inward", hjust="inward") +
    theme_grey(base_size = 14) + xlab("Property damage (millions)") +
    ylab("Crop Damage (millions)")


print(plt)
Economic impact of the ten most devastating events

Economic impact of the ten most devastating events

Results

It is found that, using our metrics, the tornadoes are by far the most harmful events to humna health in the United States, having a healt.score of 142.043 which is 6 times higher than the second event (Excessive heat). This could be seen in the plot, where the point representing the event is an outlier. The total number of fatalities by the event are: 5.633, and the number of injuries are: 91.346.

In the economic metric, tornadoes also first place with a total damage value of $5.205211410^{4} millions. This almost doubles the second event in floods. The property damage caused by tornadoes is: 5.163716110^{4}, and the crop damage caused is: 414.95327.

In conclusion, according to the U.S. National Oceanic and Atmospheric Administration’s the tornadoes are the most harmful events for human lives and also the one event that caused the most economic losses.