Top 20 most Harmful/Costly meteorological events in the US, years 1996-2011
A NOAA database analysis for Health and Economic consequences
Mathieu C.
2 octobre 2019
Synopsis
The file “Storm Data” comes from the U.S. National Oceanic and Atmospheric
Administration’s (NOAA) and was accessible at this url on oct. 2nd 2019.
The goal of the assessment is to produce a report that points at the most
harmful and most costly meteorological events across the US for the years we
have access to (1950 - 2011).
More details can be found in the README.md file at this GitHub url
Data Processing
Data Loading
We will begin by taking a look at the data:
# Downloading file (47Mb, might take a while depending on the connection and speeds)
# note that this step is cached and will be executed much faster if repeated:
myurl <- "https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2"
if (!file.exists("Storm_Data.csv.bz2")){
download.file(myurl, destfile = "Storm_Data.csv.bz2")
}
# Reading in the dataframe, again, might take a while.
if (!exists("StormData")){
StormData <- read.csv("Storm_Data.csv.bz2")
}
str(StormData)## 'data.frame': 902297 obs. of 37 variables:
## $ STATE__ : num 1 1 1 1 1 1 1 1 1 1 ...
## $ BGN_DATE : Factor w/ 16335 levels "1/1/1966 0:00:00",..: 6523 6523 4242 11116 2224 2224 2260 383 3980 3980 ...
## $ BGN_TIME : Factor w/ 3608 levels "00:00:00 AM",..: 272 287 2705 1683 2584 3186 242 1683 3186 3186 ...
## $ TIME_ZONE : Factor w/ 22 levels "ADT","AKS","AST",..: 7 7 7 7 7 7 7 7 7 7 ...
## $ COUNTY : num 97 3 57 89 43 77 9 123 125 57 ...
## $ COUNTYNAME: Factor w/ 29601 levels "","5NM E OF MACKINAC BRIDGE TO PRESQUE ISLE LT MI",..: 13513 1873 4598 10592 4372 10094 1973 23873 24418 4598 ...
## $ STATE : Factor w/ 72 levels "AK","AL","AM",..: 2 2 2 2 2 2 2 2 2 2 ...
## $ EVTYPE : Factor w/ 985 levels " HIGH SURF ADVISORY",..: 834 834 834 834 834 834 834 834 834 834 ...
## $ BGN_RANGE : num 0 0 0 0 0 0 0 0 0 0 ...
## $ BGN_AZI : Factor w/ 35 levels ""," N"," NW",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ BGN_LOCATI: Factor w/ 54429 levels "","- 1 N Albion",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ END_DATE : Factor w/ 6663 levels "","1/1/1993 0:00:00",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ END_TIME : Factor w/ 3647 levels ""," 0900CST",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ COUNTY_END: num 0 0 0 0 0 0 0 0 0 0 ...
## $ COUNTYENDN: logi NA NA NA NA NA NA ...
## $ END_RANGE : num 0 0 0 0 0 0 0 0 0 0 ...
## $ END_AZI : Factor w/ 24 levels "","E","ENE","ESE",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ END_LOCATI: Factor w/ 34506 levels "","- .5 NNW",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ LENGTH : num 14 2 0.1 0 0 1.5 1.5 0 3.3 2.3 ...
## $ WIDTH : num 100 150 123 100 150 177 33 33 100 100 ...
## $ F : int 3 2 2 2 2 2 2 1 3 3 ...
## $ MAG : num 0 0 0 0 0 0 0 0 0 0 ...
## $ FATALITIES: num 0 0 0 0 0 0 0 0 1 0 ...
## $ INJURIES : num 15 0 2 2 2 6 1 0 14 0 ...
## $ PROPDMG : num 25 2.5 25 2.5 2.5 2.5 2.5 2.5 25 25 ...
## $ PROPDMGEXP: Factor w/ 19 levels "","-","?","+",..: 17 17 17 17 17 17 17 17 17 17 ...
## $ CROPDMG : num 0 0 0 0 0 0 0 0 0 0 ...
## $ CROPDMGEXP: Factor w/ 9 levels "","?","0","2",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ WFO : Factor w/ 542 levels ""," CI","$AC",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ STATEOFFIC: Factor w/ 250 levels "","ALABAMA, Central",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ ZONENAMES : Factor w/ 25112 levels ""," "| __truncated__,..: 1 1 1 1 1 1 1 1 1 1 ...
## $ LATITUDE : num 3040 3042 3340 3458 3412 ...
## $ LONGITUDE : num 8812 8755 8742 8626 8642 ...
## $ LATITUDE_E: num 3051 0 0 0 0 ...
## $ LONGITUDE_: num 8806 0 0 0 0 ...
## $ REMARKS : Factor w/ 436781 levels "","-2 at Deer Park\n",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ REFNUM : num 1 2 3 4 5 6 7 8 9 10 ...Check for NA’s
From the above call to str() function, the database seems pretty organized.
Let’s check for missing values:
na_check <- lapply(StormData, is.na)
na_check <- lapply(na_check, sum)
sort(unlist(na_check), decreasing = T)## COUNTYENDN F LATITUDE LATITUDE_E STATE__ BGN_DATE
## 902297 843563 47 40 0 0
## BGN_TIME TIME_ZONE COUNTY COUNTYNAME STATE EVTYPE
## 0 0 0 0 0 0
## BGN_RANGE BGN_AZI BGN_LOCATI END_DATE END_TIME COUNTY_END
## 0 0 0 0 0 0
## END_RANGE END_AZI END_LOCATI LENGTH WIDTH MAG
## 0 0 0 0 0 0
## FATALITIES INJURIES PROPDMG PROPDMGEXP CROPDMG CROPDMGEXP
## 0 0 0 0 0 0
## WFO STATEOFFIC ZONENAMES LONGITUDE LONGITUDE_ REMARKS
## 0 0 0 0 0 0
## REFNUM
## 0We can see that the database is an “all or nothing” type: either a variable
has (almost) all it’s values filled in, or almost all the data is missing.
Fortunately, in the “missing” camp, we have only two variables F and
COUNTYENDN. Since these variables don’t seem to be relevant to the questions
at hands, we won’t have to take further actions in their regard.
Data modifications
This will require some packages, let’s load them:
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, unionlibrary(ggplot2)
library(reshape2)
library(pdftools)
library(stringr)
library(stringdist)Let’s first create a second database, StormData_mod, for three main reasons:
- We avoid having to rebuild the original (and heavy) dataframe with every knit.
- We can keep track of changes if needed.
- We can subset the original dataframe to work faster.
We will start by subsetting by date > 01.01.1996. Why? Because this is when
people started tracking all 48 events, not just some of them (cf this thread).
We’ll keep only the variables we will later use as well.
testdate <- as.POSIXlt(
as.character(StormData$BGN_DATE),
format = "%m/%d/%Y %H:%M:%S") > "1996-01-01"
conditions <- (StormData$CROPDMG > 0 | StormData$PROPDMG > 0 |
StormData$INJURIES > 0 | StormData$FATALITIES > 0) & testdate
StormData_mod <- subset(StormData, conditions)
StormData_mod <- StormData_mod[,c("EVTYPE","INJURIES", "FATALITIES", "CROPDMG", "CROPDMGEXP",
"PROPDMG", "PROPDMGEXP")]After some exploration, we can notice something about EVTYPE. There are
duplicates. For example: THUNDERSTORM WIND appears a whole lot of times under
slightly different names.
This is a problem. Events will need to be grouped under the official
48 names below (found in the pdf file in code, from a link provided
on Coursera):
pdf_url <- paste0("https://d396qusza40orc.cloudfront.net/",
"repdata%2Fpeer2_doc%2Fpd01016005curr.pdf")
my_pdf <-
paste0(pdf_text(pdf_url)[c(2:4)])
my_str <- unlist(strsplit(my_pdf, "(\\r)?\\n"))
my_str <- my_str[47:128]
my_str_short <-
str_match(string = my_str,
pattern = "7.([0-9]+)[.]? +(.*?) *(?:[(][A-Z][)])? *[.]{3,}")
my_str_short <- toupper(my_str_short[!is.na(my_str_short[,3]),3])
rm(my_pdf, my_str)
my_str_long <- data.frame(Value = c(1:4,
rep(5,2), 6:12,
rep(13,2), 14:16,
rep(17,2), 18:24,
rep(25,2), 26:48),
name = unlist(strsplit(my_str_short, "[/]")))
print(my_str_short)## [1] "ASTRONOMICAL LOW TIDE" "AVALANCHE"
## [3] "BLIZZARD" "COASTAL FLOOD"
## [5] "COLD/WIND CHILL" "DEBRIS FLOW"
## [7] "DENSE FOG" "DENSE SMOKE"
## [9] "DROUGHT" "DUST DEVIL"
## [11] "DUST STORM" "EXCESSIVE HEAT"
## [13] "EXTREME COLD/WIND CHILL" "FLASH FLOOD"
## [15] "FLOOD" "FREEZING FOG"
## [17] "FROST/FREEZE" "FUNNEL CLOUD"
## [19] "HAIL" "HEAT"
## [21] "HEAVY RAIN" "HEAVY SNOW"
## [23] "HIGH SURF" "HIGH WIND"
## [25] "HURRICANE/TYPHOON" "ICE STORM"
## [27] "LAKESHORE FLOOD" "LAKE-EFFECT SNOW"
## [29] "LIGHTNING" "MARINE HAIL"
## [31] "MARINE HIGH WIND" "MARINE STRONG WIND"
## [33] "MARINE THUNDERSTORM WIND" "RIP CURRENT"
## [35] "SEICHE" "SLEET"
## [37] "STORM TIDE" "STRONG WIND"
## [39] "THUNDERSTORM WIND" "TORNADO"
## [41] "TROPICAL DEPRESSION" "TROPICAL STORM"
## [43] "TSUNAMI" "VOLCANIC ASH"
## [45] "WATERSPOUT" "WILDFIRE"
## [47] "WINTER STORM" "WINTER WEATHER"After (lots) of trial and error, one of the main issues to match is the fact
that some of the terms are abbreviated, like TSTM and FLD.
We will try to fix them before the automated attempt at matchin them.
While we’re at it, there’s a few strings that start with NON something
as well as all the micellaneous names people came up with.
NOTE: This step has been done in a very (very) empiric way, if one section
of the work needs to be improved in the future, it’s this one.
patterns <- c("FLD", "NON-? ?\\w*", "TSTM", "(LIGHT)? *FREEZ.*RAIN",
"URBAN/SML STREAM", "Ice jam flood [(]minor",
"((HEAVY|HWY|HVY)? *PRECIP(ITATION)?|RAIN)", ".*HYPOTHERMIA.*",
"TORRENTIAL RAINFALL",
"HYPERTHERMIA", "AND LIGHTNING$", ".*GUSTY WIND/",
"(HWY|HVY)", "(EXTREME|TO[R]*ENTIAL|DAMAGE|UNSEASON[A-Z]*)",
"UNSEASONAB.* WARM", "UNSEASONAB.*COLD",
"(C(OA)?ST(A)?L *FLOOD(ING)?|BEACH EROSION)",
"AGRICULTURAL", "TEMPERATURE", "(HEAVY SURF|.*)?/HIGH SURF",
"(LIGHT|FALLING|LATE SEASON)? *SNOW(FALL|/ICE)? *(SQUALL)?",
".*MICROBURST.*", "(ADVISORY|HAZARDOUS|AND [A-Z]*$)",
".*HEAVY SNOW.*", "BLOWING DUST", "COASTAL FLOOD.*",
"FOG")
replacements <- c("FLOOD", "", "THUNDERSTORM", "FROST/FREEZE",
"", "FLOOD",
"HEAVY RAIN", "COLD",
"RAIN",
"HEAT", "", "",
"HEAVY", "",
"HEAT", "COLD",
"COASTAL FLOOD",
"", "", "HIGH SURF",
"HEAVY SNOW",
"THUNDERSTORM WIND", "",
"HEAVY SNOW", "HIGH WIND", "COASTAL FLOOD",
"DENSE FOG")
StormData_mod$EVTYPE_corr <- as.character(StormData_mod$EVTYPE)
for (i in 1:length(patterns)){
StormData_mod$EVTYPE_corr <-
gsub(patterns[i], replacements[i],
StormData_mod$EVTYPE_corr, ignore.case = T, perl = T)
}
StormData_mod$EVTYPE_corr <- factor(StormData_mod$EVTYPE_corr,
levels = unique(StormData_mod$EVTYPE_corr))We’ll then attempt to match the names obtained in StormData_mod$EVTYPE_corr:
## The matching attempt:
matches <-
amatch(toupper(as.character(StormData_mod$EVTYPE_corr))
, my_str_long$name,
method = "lcs",
maxDist = 10)
## Building a data frame of the gradual results, was useful for testing purposes
## and might be useful if improvement is needed.
df_test_amatch <- data.frame(EVTYPE = StormData_mod$EVTYPE,
corr = StormData_mod$EVTYPE_corr,
RESULT = my_str_short[my_str_long$Value[matches]])
StormData_mod$EVTYPE_corr <- factor(df_test_amatch$RESULT,
levels = unique(df_test_amatch$RESULT))After all this wrangling we end up with a new and “clean” variable:
summary(StormData_mod$EVTYPE_corr)## WINTER STORM TORNADO THUNDERSTORM WIND
## 1463 12365 105449
## HIGH WIND FLASH FLOOD FROST/FREEZE
## 5529 19014 173
## COLD/WIND CHILL LIGHTNING HAIL
## 442 11151 22697
## FLOOD EXCESSIVE HEAT RIP CURRENT
## 10302 685 603
## HEAT HEAVY SNOW WILDFIRE
## 203 1427 1228
## ICE STORM BLIZZARD STORM TIDE
## 632 228 216
## DUST STORM STRONG WIND DUST DEVIL
## 101 3414 84
## DENSE FOG HIGH SURF HEAVY RAIN
## 159 193 1110
## MARINE HAIL AVALANCHE COASTAL FLOOD
## 3 264 200
## WATERSPOUT HURRICANE/TYPHOON TROPICAL STORM
## 29 208 410
## DROUGHT SLEET EXTREME COLD/WIND CHILL
## 263 215 1
## FREEZING FOG WINTER WEATHER FUNNEL CLOUD
## 13 547 9
## SEICHE VOLCANIC ASH MARINE THUNDERSTORM WIND
## 9 2 142
## ASTRONOMICAL LOW TIDE TROPICAL DEPRESSION MARINE HIGH WIND
## 10 35 19
## TSUNAMI LAKESHORE FLOOD MARINE STRONG WIND
## 14 5 46
## DENSE SMOKE
## 1Data Analysis
Results
Now we have all the tools to create a new dataframe, with data grouped by EVTYPE
and we’ll take a look at the highest means of fatalities and injuries combined.
health_by_event <- StormData_mod %>%
group_by(EVTYPE_corr) %>%
summarize(mean.fatalities = mean(FATALITIES, na.rm = T),
mean.injuries = mean(INJURIES, na.rm = T),
total.means = mean.fatalities + mean.injuries,
occurences = n()) %>%
arrange(desc(total.means))
head(health_by_event, 10)## # A tibble: 10 x 5
## EVTYPE_corr mean.fatalities mean.injuries total.means occurences
## <fct> <dbl> <dbl> <dbl> <int>
## 1 EXCESSIVE HEAT 2.62 9.33 12.0 685
## 2 TSUNAMI 2.36 9.21 11.6 14
## 3 HEAT 1.18 6.38 7.57 203
## 4 HURRICANE/TYPHOON 0.601 6.38 6.99 208
## 5 DENSE FOG 0.434 5.38 5.81 159
## 6 DUST STORM 0.149 3.74 3.89 101
## 7 BLIZZARD 0.307 1.69 2.00 228
## 8 HIGH SURF 0.741 1.07 1.81 193
## 9 TORNADO 0.122 1.67 1.79 12365
## 10 RIP CURRENT 0.899 0.834 1.73 603In our question we want to know how harmful an event is to population.
My first thought was to sort the events by their total means (injuries
and fatalities) combined as in the top 10 above.
However, we first have to make sure that these events occur regularly to avoid
putting too much weight into a deadly, harmful but also very unlikely event
Let’s see what the top most frequent events a.k.a. occurences looks like:
head(health_by_event[order(health_by_event$occurences, decreasing = T),], 10)## # A tibble: 10 x 5
## EVTYPE_corr mean.fatalities mean.injuries total.means occurences
## <fct> <dbl> <dbl> <dbl> <int>
## 1 THUNDERSTORM WIND 0.00362 0.0489 0.0525 105449
## 2 HAIL 0.000352 0.0337 0.0340 22697
## 3 FLASH FLOOD 0.0466 0.0880 0.135 19014
## 4 TORNADO 0.122 1.67 1.79 12365
## 5 LIGHTNING 0.0583 0.371 0.430 11151
## 6 FLOOD 0.0431 0.664 0.707 10302
## 7 HIGH WIND 0.0468 0.215 0.262 5529
## 8 STRONG WIND 0.0322 0.0876 0.120 3414
## 9 WINTER STORM 0.131 0.936 1.07 1463
## 10 HEAVY SNOW 0.0939 0.529 0.623 1427This is interesting. We can see that some newcomers like HAIL or
THUNDERSTORM WIND occur lots of times and they are harmful, not top
10 harmful but still.
Let’s try creating an index variable. It will be the product of occurences
by total.means.
Why?
Because this way, if an event is occuring a very low amount of time but has
a very harmful outcome, it will be on par with a less harmful event that
happens much more frequently. We will log(10) this index to avoid working
with huge numbers.
Let’s do this and have a look:
health_by_event <- transform(health_by_event,
index = log((occurences * total.means), 10))
head(health_by_event[order(desc(health_by_event$index)),], 10)## EVTYPE_corr mean.fatalities mean.injuries total.means occurences
## 9 TORNADO 0.122199757 1.67141124 1.79361100 12365
## 1 EXCESSIVE HEAT 2.623357664 9.32992701 11.95328467 685
## 22 FLOOD 0.043098427 0.66375461 0.70685304 10302
## 39 THUNDERSTORM WIND 0.003622604 0.04887671 0.05249931 105449
## 26 LIGHTNING 0.058290736 0.37126715 0.42955789 11151
## 32 FLASH FLOOD 0.046649837 0.08804039 0.13469023 19014
## 16 WINTER STORM 0.131237184 0.93574846 1.06698565 1463
## 13 WILDFIRE 0.070846906 1.18566775 1.25651466 1228
## 3 HEAT 1.182266010 6.38423645 7.56650246 203
## 4 HURRICANE/TYPHOON 0.600961538 6.38461538 6.98557692 208
## index
## 9 4.345922
## 1 3.913178
## 22 3.862251
## 39 3.743196
## 26 3.680336
## 32 3.408410
## 16 3.193403
## 13 3.188366
## 3 3.186391
## 4 3.162266Maybe a figure would be prettier:
plotpop <- head(health_by_event[order(desc(health_by_event$index)),], 20)
plotpop$EVTYPE_corr <- factor(plotpop$EVTYPE_corr,
levels = as.character(plotpop$EVTYPE_corr))
warncolors <- colorRampPalette(c("red", "gold3"))
g1 <- ggplot(plotpop[c(1:6)],
aes(x=EVTYPE_corr, y = index, fill = EVTYPE_corr))
g1 + geom_bar(stat = "identity", position = "dodge") +
ylab("Health issues index")+
xlab(element_blank())+
scale_fill_manual(values = warncolors(20))+
theme_bw()+
theme(legend.position = "none")+
theme(axis.text.x = element_text(angle = 45, hjust = 0.95)) +
labs(title = paste0("Top 20 events in decreasing order of Health issues",
" index (HII).\n HII = mean of injuries + mean ",
" of fatalities * total occurences (log10)\nData for ",
"years 1996 to 2011"))
Let’s do the same with cost, using two new variables, PROPDMG and CROPDMG,
holding numbers for property damage and crop damage, respectively, they
first have to be ajusted with their exponent EXP counteparts:
v_pattern <- c("[0-9]", "^$", "[+]", "[?-]", "[hH]", "[kK]", "[mM]", "[Bb]")
v_repl <- c("10", "0", "1", "0", "100", "1000", "1000000", "1000000000")
## Modifying CROPDMGEXP
StormData_mod$CROPDMGEXP <- as.character(StormData_mod$CROPDMGEXP)
StormData_mod$PROPDMGEXP <- as.character(StormData_mod$PROPDMGEXP)
for (i in 1:8){
StormData_mod$CROPDMGEXP <-
gsub(v_pattern[i], v_repl[i], StormData_mod$CROPDMGEXP)
StormData_mod$PROPDMGEXP <-
gsub(v_pattern[i], v_repl[i], StormData_mod$PROPDMGEXP)
}
StormData_mod$CROPDMGEXP <- as.numeric(StormData_mod$CROPDMGEXP)
StormData_mod$PROPDMGEXP <- as.numeric(StormData_mod$PROPDMGEXP)cost_by_event <- StormData_mod %>%
group_by(EVTYPE_corr) %>%
summarise(mean.prop = mean(PROPDMG * PROPDMGEXP, na.rm = T),
mean.crop = mean(CROPDMG * CROPDMGEXP, na.rm = T),
mean.combi = mean.crop + mean.prop,
occurences = n(),
index = log(mean.combi*occurences, 10)) %>%
arrange(desc(index))
head(cost_by_event[order(desc(cost_by_event$mean.combi)),])## # A tibble: 6 x 6
## EVTYPE_corr mean.prop mean.crop mean.combi occurences index
## <fct> <dbl> <dbl> <dbl> <int> <dbl>
## 1 HURRICANE/TYPHOON 392879274. 25721672. 418600946. 208 10.9
## 2 STORM TIDE 221457056. 3958. 221461014. 216 10.7
## 3 DROUGHT 3977578. 50827247. 54804825. 263 10.2
## 4 TROPICAL STORM 18640184. 1652954. 20293138. 410 9.92
## 5 FLOOD 13990447. 486620. 14477067. 10302 11.2
## 6 TSUNAMI 10290143. 1429. 10291571. 14 8.16If we arrange the data by mean.combi (as seen above), we can see that the
most costly events have a very low occurence. Since the goal of this analysis
is to produce a report supposedly for preparation towards events, the formerly
used method of creating an index to weight events by occurences seems about
right.
head(cost_by_event, 20)## # A tibble: 20 x 6
## EVTYPE_corr mean.prop mean.crop mean.combi occurences index
## <fct> <dbl> <dbl> <dbl> <int> <dbl>
## 1 FLOOD 13990447. 486620. 14477067. 10302 11.2
## 2 HURRICANE/TYPHOON 392879274. 25721672. 418600946. 208 10.9
## 3 STORM TIDE 221457056. 3958. 221461014. 216 10.7
## 4 TORNADO 1990854. 22922. 2013775. 12365 10.4
## 5 HAIL 643047. 110007. 753054. 22697 10.2
## 6 FLASH FLOOD 800582. 70206. 870788. 19014 10.2
## 7 DROUGHT 3977578. 50827247. 54804825. 263 10.2
## 8 THUNDERSTORM WIND 75063. 9644. 84707. 105449 9.95
## 9 TROPICAL STORM 18640184. 1652954. 20293138. 410 9.92
## 10 WILDFIRE 6319584. 327569. 6647154. 1228 9.91
## 11 HIGH WIND 949976. 114681. 1064657. 5529 9.77
## 12 ICE STORM 5763052. 24778. 5787830. 632 9.56
## 13 WINTER STORM 1047673. 8164. 1055837. 1463 9.19
## 14 COLD/WIND CHILL 76427. 3069605. 3146032. 442 9.14
## 15 FROST/FREEZE 114289. 7911913. 8026202. 173 9.14
## 16 HEAVY RAIN 529143. 665243. 1194386. 1110 9.12
## 17 HEAVY SNOW 477920. 49840. 527760. 1427 8.88
## 18 LIGHTNING 66638. 619. 67256. 11151 8.88
## 19 BLIZZARD 2305522. 30965. 2336487. 228 8.73
## 20 EXCESSIVE HEAT 11275. 718835. 730111. 685 8.70plotpop2 <- head(cost_by_event[order(desc(cost_by_event$index)),], 20)
plotpop2$EVTYPE_corr <- factor(plotpop2$EVTYPE_corr,
levels = as.character(plotpop2$EVTYPE_corr))
prettycolors2 <- colorRampPalette(c("darkgreen", "springgreen"))
g2 <- ggplot(plotpop2[c(1:6)],
aes(x=EVTYPE_corr, y = index, fill = EVTYPE_corr))
g2 + geom_bar(stat = "identity", position = "dodge") +
ylab("Cost Index")+
xlab(element_blank())+
theme_bw()+
theme(legend.position = "none")+
scale_fill_manual(values = prettycolors2(20))+
theme(axis.text.x = element_text(angle = 45, hjust = 0.95)) +
labs(title = paste0("Top 20 events in decreasing order of Cost",
" Index (CI).\n CI = mean of propriety damage +",
" mean of crop damage \n",
" times total occurences (log10)\nData for ",
"years 1996 to 2011"))
I will rest my case here. Further steps should be taken to be able to consider
this work for anything else than a simple assessment in the JH Datascience
Course on Coursera:
- In depth cleaning of the
EVTYPE_corrvariable to ensure the best merge of duplicates.
Probing aroud with something like :
head(df_test_amatch[df_test_amatch$RESULT == "RIP CURRENT",])## EVTYPE corr RESULT
## 406 RIP CURRENTS RIP CURRENTS RIP CURRENT
## 480 RIP CURRENTS RIP CURRENTS RIP CURRENT
## 816 RIP CURRENTS RIP CURRENTS RIP CURRENT
## 1166 RIP CURRENTS RIP CURRENTS RIP CURRENT
## 1187 RIP CURRENTS RIP CURRENTS RIP CURRENT
## 1250 RIP CURRENTS RIP CURRENTS RIP CURRENT… For all 48 events Would be great, manually correcting the last typos.
- Regional repartition to further affinate the specific risk potential.
- Ajusting the index, especially for health, to reflect the true impact on
a community