Severe Weather Event Impact Analysis

Using the U.S. National Oceanic and Atmospheric Administration’s (NOAA) storm database, we identify the weather events with the highest adverse impacts on population health and economic wellbeing. Tornadoes have the worst impact on population health: 91346 injuries and 5633 fatalities. Hurricanes and typhoons have the worst economic consequences, at $72T. From a policy perspective, it is highly important to focus on mitigating the effects of these weather event types.

In the following, I load in the data, process it, and investigate the events with the highest impacts.

Data Processing

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

storms <- read.csv("StormData.csv.bz2")
head(storms)
##   STATE__           BGN_DATE BGN_TIME TIME_ZONE COUNTY COUNTYNAME STATE  EVTYPE
## 1       1  4/18/1950 0:00:00     0130       CST     97     MOBILE    AL TORNADO
## 2       1  4/18/1950 0:00:00     0145       CST      3    BALDWIN    AL TORNADO
## 3       1  2/20/1951 0:00:00     1600       CST     57    FAYETTE    AL TORNADO
## 4       1   6/8/1951 0:00:00     0900       CST     89    MADISON    AL TORNADO
## 5       1 11/15/1951 0:00:00     1500       CST     43    CULLMAN    AL TORNADO
## 6       1 11/15/1951 0:00:00     2000       CST     77 LAUDERDALE    AL TORNADO
##   BGN_RANGE BGN_AZI BGN_LOCATI END_DATE END_TIME COUNTY_END COUNTYENDN
## 1         0                                               0         NA
## 2         0                                               0         NA
## 3         0                                               0         NA
## 4         0                                               0         NA
## 5         0                                               0         NA
## 6         0                                               0         NA
##   END_RANGE END_AZI END_LOCATI LENGTH WIDTH F MAG FATALITIES INJURIES PROPDMG
## 1         0                      14.0   100 3   0          0       15    25.0
## 2         0                       2.0   150 2   0          0        0     2.5
## 3         0                       0.1   123 2   0          0        2    25.0
## 4         0                       0.0   100 2   0          0        2     2.5
## 5         0                       0.0   150 2   0          0        2     2.5
## 6         0                       1.5   177 2   0          0        6     2.5
##   PROPDMGEXP CROPDMG CROPDMGEXP WFO STATEOFFIC ZONENAMES LATITUDE LONGITUDE
## 1          K       0                                         3040      8812
## 2          K       0                                         3042      8755
## 3          K       0                                         3340      8742
## 4          K       0                                         3458      8626
## 5          K       0                                         3412      8642
## 6          K       0                                         3450      8748
##   LATITUDE_E LONGITUDE_ REMARKS REFNUM
## 1       3051       8806              1
## 2          0          0              2
## 3          0          0              3
## 4          0          0              4
## 5          0          0              5
## 6          0          0              6

The variables of interest are the following:

  • EVTYPE : type of event

  • FATALITIES and INJURIES : count of damage to individuals

  • PROPDMG & PROPDMGEXP : property damage, given as a combination of number and magnitude designation (K for thousands, M for millions, B for billions)

  • CROPDMG & CROPDMGEXP : damage to crops plus magnitude as above

Analysis

Population Health Impacts

The database records fatalities and injuries per event. It is intuitive to count fatalities as worse than injuries, but hard to quantify just how much worse: is a fatality worth 100 injuries? 1000 injuries? And are we talking about bumped knees injuries or shattered limbs and severed spinal cords? That said, if fatalities strongly correlate with injuries, we may not need to answer that question at all.

par(mfrow = c(2,2))
plot(storms$FATALITIES, storms$INJURIES)
abline(lm(storms$INJURIES ~ storms$FATALITIES))

plot(storms$FATALITIES, storms$INJURIES, xlim = c(0,200))

plot(storms$FATALITIES, storms$INJURIES, xlim = c(0,50))

plot(log(storms$FATALITIES), log(storms$INJURIES))

We clearly see a positive correlation, but it is an ambiguous one. It is significantly skewed by the outlier of a 1995 heatwave in Northern Illinois with 583 fatalities and 0 injuries reported. Even aside from that one, as we limit the x-axis, we see a very crowded region near the base of the chart: most of the damaging events have 0-15 fatalities, but for each of those, there’s a high variation between 0-200 injuries. Using a log scale to uncrowd the plot makes the pattern of high injury variation even clearer. Though this is an entirely arbitrary decision, I will make the calculation based on a 1:100 fatality-injury conversion.

storms$health_impact <- 100* storms$FATALITIES + storms$INJURIES
library(dplyr)
## 
## Attaching package: 'dplyr'
## The following object is masked _by_ '.GlobalEnv':
## 
##     storms
## The following objects are masked from 'package:stats':
## 
##     filter, lag
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union
population_impact <- storms %>% 
        group_by(EVTYPE) %>%
        summarize(total_impact = sum(health_impact), 
                  avg_impact = mean(health_impact))

population_impact %>% arrange(desc(avg_impact)) %>% head()
## # A tibble: 6 x 3
##   EVTYPE                     total_impact avg_impact
##   <chr>                             <dbl>      <dbl>
## 1 TORNADOES, TSTM WIND, HAIL         2500      2500 
## 2 COLD AND SNOW                      1400      1400 
## 3 TROPICAL STORM GORDON               843       843 
## 4 RECORD/EXCESSIVE HEAT              1700       567.
## 5 EXTREME HEAT                       9755       443.
## 6 HEAT WAVE DROUGHT                   415       415
population_impact %>% arrange(desc(total_impact)) %>% head()
## # A tibble: 6 x 3
##   EVTYPE         total_impact avg_impact
##   <chr>                 <dbl>      <dbl>
## 1 TORNADO              654646     10.8  
## 2 EXCESSIVE HEAT       196825    117.   
## 3 FLASH FLOOD           99577      1.83 
## 4 HEAT                  95800    125.   
## 5 LIGHTNING             86830      5.51 
## 6 TSTM WIND             57357      0.261

Results

Ostensibly, “TORNADOES, TSTM WIND, HAIL” has the largest average impact, at 70, but a closer look reveals that we’re dealing with a recording fluke. A single event with an impact of 70 was labeled with a unique title, yielding an erroneous average of 2500. Most of the top average entries are such single-entry events.

Total impact counts are far less sensitive to such input errors, and tornadoes clearly take the lead with a total impact of 654,646. Disaggregating that total figures yields the following figures:

tornado_focus <- storms %>% 
        filter(EVTYPE == "TORNADO") %>%
        select(EVTYPE, FATALITIES, INJURIES) %>%
        summarise(num = n(),
                  tot_fatalities = sum(FATALITIES),
                  avg_fatalities = mean(FATALITIES),
                  tot_injuries = sum(INJURIES),
                  avg_injuries = mean(INJURIES))
tornado_focus
##     num tot_fatalities avg_fatalities tot_injuries avg_injuries
## 1 60652           5633      0.0928741        91346     1.506067

Economic Impacts

The dataset reports property and crop damage, both relevant to the question. The only caveat is that each is reported as a combination of number and order of magnitude, but that is easily corrected.

storms$economic_impact <- case_when(
        storms$PROPDMGEXP == "" ~ 0,
        storms$PROPDMGEXP == "K" ~ storms$PROPDMG * 1000,
        storms$PROPDMGEXP == "M" ~ storms$PROPDMG * 1000000,
        storms$PROPDMGEXP == "B" ~ storms$PROPDMG * 1000000000
) + case_when(
        storms$CROPDMGEXP == "" ~ 0,
        storms$CROPDMGEXP == "K" ~ storms$CROPDMG * 1000,
        storms$CROPDMGEXP == "M" ~ storms$CROPDMG * 1000000,
        storms$CROPDMGEXP == "B" ~ storms$CROPDMG * 1000000000)

economic_impact <- storms %>% 
        group_by(EVTYPE) %>%
        summarize(total_impact = sum(economic_impact), 
                  avg_impact = mean(economic_impact))

Results

economic_impact %>% arrange(desc(avg_impact)) %>% head()
## # A tibble: 6 x 3
##   EVTYPE                     total_impact  avg_impact
##   <chr>                             <dbl>       <dbl>
## 1 TORNADOES, TSTM WIND, HAIL   1602500000 1602500000 
## 2 HEAVY RAIN/SEVERE WEATHER    2500000000 1250000000 
## 3 HURRICANE/TYPHOON           71913712800  817201282.
## 4 STORM SURGE                 43323541000  165990579.
## 5 WILD FIRES                    624100000  156025000 
## 6 EXCESSIVE WETNESS             142000000  142000000
economic_impact %>% arrange(desc(total_impact)) %>% head()
## # A tibble: 6 x 3
##   EVTYPE            total_impact avg_impact
##   <chr>                    <dbl>      <dbl>
## 1 HURRICANE/TYPHOON  71913712800 817201282.
## 2 STORM SURGE        43323541000 165990579.
## 3 HURRICANE          14610229010  83966833.
## 4 RIVER FLOOD        10148404500  58661298.
## 5 TROPICAL STORM      8382236550  12148169.
## 6 WILDFIRE            5060586800   1832882.

Average damage values, again, show noise from single-occurrence events. Total impact figures are more robust, and show hurricanes & typhoons clearly in the lead with $71,913,712,800--that’s nearly 72 trillion dollars. Hurricane as a separate category comes in third at $14.7T, but since I am not manually totaling other duplicate categories, I will let this one be, too.