Analysis of Severe Weather Events Impacts
jnabonne
7/17/2018
About
This document computes the answer to the assignment #2 of Reproducible Research course.
Dataset and corresponding Documentation
The dataset for this exercice available here.
There is also two documents to better understand its content:
Synopsis of the Analysis
The main objective of this analysis is to understand the possible impacts (economic and public health) of server weather events.
Two main questions were asked :
- which types of events are most harmful to population health?
- which types of events have the greatest economic consequences?
This analysis starts by downloading a dataset from the U.S. National Oceanic and Atmospheric Administration’s (NOAA) storm database containing characteristics of major storms and weather events in the US (time, location, impacts estimation).
The data are then cleaned to only conserve interesting measures and in adequate format for manipulation and display.
Some new tables presenting aggregated results are the created giving for each type of events their number of occurences, victims and damages amount.
The Analysis ends by displaying some graphics showing evidence that tornados and flood are respectively the worst events in term of human victims and economical damages.
Data Processing
First we load the libraries used for this analysis
library(ggplot2)
library(dplyr)Then we download the dataset from the internet.
No check on the already existing presence of the file is done, this to ensure that we always work on the latest available version of the dataset.
# DOWNLOADING DATASET (EACH TIME TO GET ITS LATEST VERSION)
file_url <- "https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2"
file_path <- "StormData.csv.bz2"
download.file(file_url, destfile=file_path, method="curl")The dataset is then loaded
# Reading (and unzipping on-the-fly) the DATASET
sdata <- read.csv("StormData.csv.bz2")
# sdata <- read.csv("StormData.csv.bz2", nrows=30000) # only to speed tests ;)Then some processing are done to the dataset to ease the rest of the analysis.
First we change the exponent factors to be applied to property and crop damages by correct numeric values. Following some reading on the internet (especially this post) the following matching as been performed
| symbol | exponent value |
|---|---|
| numeric | 10 |
| h or H | 100 |
| t or T | 1000 |
| k or K | 1000 |
| m or M | 1.000.000 |
| b or B | 1.000.000.000 |
| + | 1 |
| all others | 0 |
These exponent values are then used to redress property and crop damages.
# Mathcing exponantial factors with correct numeric values
sdata <- mutate(sdata, CROPDMGEXP=ifelse(CROPDMGEXP %in% c('h','H'), 100,
ifelse(CROPDMGEXP %in% c('t','T','k','K'), 1000,
ifelse(CROPDMGEXP %in% c('m','M'), 1000000,
ifelse(CROPDMGEXP %in% c('b','B'), 1000000000,
ifelse(CROPDMGEXP %in% 0:9, 10,
ifelse(CROPDMGEXP == '+', 1, 0)))))))
sdata <- mutate(sdata, PROPDMGEXP=ifelse(PROPDMGEXP %in% c('h','H'), 100,
ifelse(PROPDMGEXP %in% c('t','T','k','K'), 1000,
ifelse(PROPDMGEXP %in% c('m','M'), 1000000,
ifelse(PROPDMGEXP %in% c('b','B'), 1000000000,
ifelse(PROPDMGEXP %in% 0:9, 10,
ifelse(PROPDMGEXP == '+', 1, 0)))))))
# Computing final damages amount from corresponding exponential factors
sdata <- mutate(sdata,PROPDMG=PROPDMG*PROPDMGEXP,CROPDMG=CROPDMG*CROPDMGEXP)The dataset is the cleaned by removing unusefull columns.
The dates, even if not very used in this analyis (kept to allow playing with the data set in eda mode) are casted correctly to be more easily manageable.
# Keeping only interresting data
sdata <- subset(sdata, select=c("EVTYPE", "BGN_DATE", "FATALITIES",
"INJURIES", "PROPDMG", "CROPDMG"))
# Casting dates correctly (date is here optional and was kept for tests)
sdata$BGN_DATE <- as.Date(sdata$BGN_DATE, "%m/%d/%Y %H:%M:%S")
#sd1 <- sdata %>% group_by(year(BGN_DATE), EVTYPE) # requires lubridate libFinally, the dataset is grouped by event type (using the dplyr library) and summarized creating totals for the interesting measures:
- nb of occurences
- total victims, with subtotal per injuries and fatalities
- total damages, with subtotal for property and crop damages
To get plots that can easily be read by humans, we only kept the top10 events (in term of occurences, victims and damages) otherwise we would get graphics with hundreds of variables
# Grouping data by events and summarizing
sd2 <- sdata %>% group_by(EVTYPE)
sd3 <- summarize(sd2, total_events=length(EVTYPE),
total_injuries=sum(INJURIES),
total_fatalities=sum(FATALITIES),
total_victims=sum(INJURIES)+sum(FATALITIES),
total_property_damages=sum(PROPDMG),
total_crop_damages=sum(CROPDMG),
total_damages=sum(PROPDMG)+sum(CROPDMG))
# Building top10 to get cleaner plots
top10_events <- arrange(sd3, desc(sd3$total_events))[1:10,]
top10_victims <- arrange(sd3, desc(sd3$total_victims))[1:10,]
top10_damages <- arrange(sd3, desc(sd3$total_damages))[1:10,]Results
From a pure events based approach we can see on the below graphics the type of event that occurs most is HAIL.
The graphic below shows the most frequent climatic events in the US between 1950 and 2011:
# ploting nb occurrences / events type
g <- ggplot(data=top10_events,
aes(x=reorder(EVTYPE, -total_events), y=total_events, fill=EVTYPE))
# printing plot with adequate legends and scaling
g + geom_bar(stat="identity") + theme_minimal() +
ggtitle("NUMBER OF OCCURENCES PER EVENTS TYPE") + labs(fill="Events") +
scale_y_continuous(name="number of occurences", labels = scales::comma) +
theme(axis.title.x=element_blank(), axis.text.x=element_blank(),
axis.ticks.x=element_blank())
This does not represent a list of the most dangerous climatic events. We should take some criteria to investigate more about these catastrophic events…
Victims based approach
But if we base our analysis on the number of victims the result is quite different; the worst event in terms of human victims is TORNADO.
Note that have been summed here injuries and fatalities (even if deaths could arguably count more) as it is difficult to create a scale between these two.
The table below displays a top10 of victims per event type:
subset(arrange(sd3, desc(sd3$total_victims))[1:10,], select=c("EVTYPE",
"total_events", "total_injuries", "total_fatalities", "total_victims"))## # A tibble: 10 x 5
## EVTYPE total_events total_injuries total_fatalities total_victims
## <fct> <int> <dbl> <dbl> <dbl>
## 1 TORNADO 60652 91346 5633 96979
## 2 EXCESSIVE H… 1678 6525 1903 8428
## 3 TSTM WIND 219940 6957 504 7461
## 4 FLOOD 25326 6789 470 7259
## 5 LIGHTNING 15754 5230 816 6046
## 6 HEAT 767 2100 937 3037
## 7 FLASH FLOOD 54277 1777 978 2755
## 8 ICE STORM 2006 1975 89 2064
## 9 THUNDERSTOR… 82563 1488 133 1621
## 10 WINTER STORM 11433 1321 206 1527The graphic below shows a top10 of the worst events in terms of victims:
# ploting nb victims / events type
g <- ggplot(data=top10_victims,
aes(x=reorder(EVTYPE, -total_victims), y=total_victims, fill=EVTYPE))
# printing plot with adequate legends and scaling
g + geom_bar(stat="identity") + theme_minimal() +
ggtitle("TOTAL NUMBER OF VICTIMS PER EVENTS TYPE") + labs(fill="Events") +
scale_y_continuous(name="number of injuries + fatalities",
labels = scales::comma) +
theme(axis.title.x=element_blank(), axis.text.x=element_blank(),
axis.ticks.x=element_blank())
Economic based approach
Now from an economic point of view, the result is again different with FLOOD arriving first in term of damages amount.
As for victims, both property and crop damages have been added together to get these results.
The table below displays a top10 of economic damages per event type:
subset(arrange(sd3, desc(sd3$total_damages))[1:10,], select=c("EVTYPE",
"total_events", "total_property_damages",
"total_crop_damages", "total_damages"))## # A tibble: 10 x 5
## EVTYPE total_events total_property_d… total_crop_dama… total_damages
## <fct> <int> <dbl> <dbl> <dbl>
## 1 FLOOD 25326 144657709800 5661968450 150319678250
## 2 HURRICAN… 88 69305840000 2607872800 71913712800
## 3 TORNADO 60652 56937162897 414954710 57352117607
## 4 STORM SU… 261 43323536000 5000 43323541000
## 5 HAIL 288661 15732269877 3025954650 18758224527
## 6 FLASH FL… 54277 16140815011 1421317100 17562132111
## 7 DROUGHT 2488 1046106000 13972566000 15018672000
## 8 HURRICANE 174 11868319010 2741910000 14610229010
## 9 RIVER FL… 173 5118945500 5029459000 10148404500
## 10 ICE STORM 2006 3944928310 5022113500 8967041810The graphic below shows the 10 worst events in terms of economic damages.
# ploting damages / events type
g <- ggplot(data=top10_damages,
aes(x=reorder(EVTYPE, -total_damages), y=total_damages, fill=EVTYPE))
# printing plot with adequate legends and scaling
g + geom_bar(stat="identity") + theme_minimal() +
ggtitle("TOTAL AMOUNT OF DAMAGES PER EVENTS TYPE") + labs(fill="Events") +
scale_y_continuous(name="amount of damages (property + crop) in USD",
labels = scales::comma) +
theme(axis.title.x=element_blank(), axis.text.x=element_blank(),
axis.ticks.x=element_blank())
This is the end, thanks for reading until it!