Worst-damage weather events in the US

Synopsis

This analysis uses the U.S. National Oceanic and Atmospheric Administration’s (NOAA) Storm Database to answer the following questions:

1. Across the United States, which types of events are most harmful with respect to population health?
2. Across the United States, which types of events have the greatest economic consequences?

Note that the database contains events from years between 1950 and 2011.

The analysis relies on the Storm Data Preparation instruction document issued by the National Weather Service department of NOAA.

Data Processing

The storm data is read into object D, which is used throughout the analysis as the source of data. Results of data transformations are saved in this object.

download.file("https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2",
              "StormData.csv.bz2")
D <- read.csv("StormData.csv.bz2", stringsAsFactors = FALSE, nrows = 1e6)

There are two major steps of data processing in this analysis:

1. Based on the Storm Data Preparation document, the property and crop damages should be stored as pairs of value and expression in each record, where expression should indicate the magnitude of the value. For example, the value 100 with the expression M means 100 million. This data processing step sets all value to value * expression, and all expressions to U, which indicates the magnitude of the expression is 1. Furthermore a new variable is added to the dataset, which is simply the sum of property and crop damage values.
2. Based on the Storm Data Preparation document, there should be only a certain number of different weather event types in the dataset (valid event types). However, this is not true in practice: there are way more event types than expected (non-valid event types). The aim of this step is to unite as much non-valid event types as possible into as less event types (not necessarily valid ones) as possible.

These steps are detailed in the following sections.

Transformation of property and crop damages

Property and crop damages are stored in four columns in the original dataset:

• PROPDMG: the value of property damage;
• PROPDMGEXP: the expression of property damage;
• CROPDMG: the value of crop damage;
• CROPDMGEXP: the expression of crop damage.

Damage expression is an empty string or a single-character value, which specifies the multiplier of the damage value. Based on the Storm Data Preparation document, valid values are the empty string (no multiplication), K (thousands), M (millions) and B (billions).

In spite of this, the dataset contains several other expression characters:

• PROPDMGEXP: , -, ?, +, 0, 1, 2, 3, 4, 5, 6, 7, 8, B, h, H, K, m, M
• CROPDMGEXP: , ?, 0, 2, B, k, K, m, M

Property and crop damage values are made directly comparable with the help of the clean.dollars function.

Damage values with non-valid expression are ignored by this analysis and therefore these values are replaced with zero, regardless of the original damage value.

Damage values with valid expression are on the other hand multiplied according to the expression (see above).

clean.dollars <- function(dollar.vector, dollar.exp)
{
  valid.exp <- c("", "K", "M", "B")
  
  dollar.vector <- replace(dollar.vector, !(dollar.exp %in% valid.exp), 0)
  dollar.exp <- replace(dollar.exp, !(dollar.exp %in% valid.exp), "")

  dollar.exp <- replace(dollar.exp, dollar.exp == "", "U")

  value.map <- c(U = 1, K = 1e3, M = 1e6, B = 1e9)
  
  for (v in names(value.map))
  {
    dollar.vector[dollar.exp == v] <- 
      dollar.vector[dollar.exp == v] * value.map[v]
  }
  
  return(dollar.vector)
}

The transformation is made on both PROPDMG and CROPDMG variables.

D$PROPDMG <- clean.dollars(D$PROPDMG, D$PROPDMGEXP)
D$CROPDMG <- clean.dollars(D$CROPDMG, D$CROPDMGEXP)

Also, all the expressions are set to U, which denotes the damage values have been set to unit values in the dataset; this is done only for the sake of clarity, because in practice damage expressions are not used further in this analysis.

D$PROPDMGEXP <- "U"
D$CROPDMGEXP <- "U"

Finally, a new column called ALLDMG is created in the dataset, which is simply the sum of PROPDMG and CROPDMG in each row.

D$ALLDMG <- D$PROPDMG + D$CROPDMG

Transformation of event types

Valid event types are defined by the Storm Data Preparation instructions. These are listed in the valid.events object.

valid.events <- c('Astronomical Low Tide', 'Avalanche', 'Blizzard', 'Coastal Flood', 'Cold/Wind Chill', 'Debris Flow', 'Dense Fog', 'Dense Smoke', 'Drought', 'Dust Devil', 'Dust Storm', 'Excessive Heat', 'Extreme Cold/Wind Chill', 'Flash Flood', 'Flood', 'Frost/Freeze', 'Funnel Cloud', 'Freezing Fog', 'Hail', 'Heat', 'Heavy Rain', 'Heavy Snow', 'High Surf', 'High Wind', 'Hurricane/Typhoon', 'Ice Storm', 'Lake-Effect Snow', 'Lakeshore Flood', 'Lightning', 'Marine Hail', 'Marine High Wind', 'Marine Strong Wind', 'Marine Thunderstorm Wind', 'Rip Current', 'Seiche', 'Sleet', 'Storm Surge/Tide', 'Strong Wind', 'Thunderstorm Wind', 'Tornado', 'Tropical Depression', 'Tropical Storm', 'Tsunami', 'Volcanic Ash', 'Waterspout', 'Wildfire', 'Winter Storm', 'Winter Weather')

The dataset has a column called EVTYPE, which specifies the type of the weather event the record is about. Just like in case of the damage expressions, this also contains non-valid event types, but here situation is much worse: there are 985 different values. These include variants which are same except for the casing or typos, synonyms, mixed types, new types, free-text comments etc.

Here are 20 examples from the dataset: DRY MIRCOBURST WINDS, Dust Devil, EXCESSIVE WETNESS, HAIL/WIND, HIGH WIND, Ice Fog, Record Cold, RECORD COOL, RECORD/EXCESSIVE HEAT, RIVER FLOOD, SNOW SHOWERS, SNOW/RAIN/SLEET, Summary of March 23, THUNDERSTORM DAMAGE, THUNDERSTORM HAIL, THUNDERSTORM WINDS HEAVY RAIN, THUNDERSTORM WINDS SMALL STREA, TORRENTIAL RAIN, VOLCANIC ASH, VOLCANIC ASH.

First of all, both the list of valid events and the events in the dataset are converted to upper case. The original EVTYPE column is retained in the dataset, the upper case variants are put to a new column called EVTYPE2. Note that all the further transformations are done on this column.

valid.events <- toupper(valid.events)
D$EVTYPE2 <- toupper(D$EVTYPE)

Then some very general statistics is calculated about the wrong event types.

valid.count <- sum(table(D$EVTYPE2)[valid.events], na.rm = TRUE)
valid.percent <- round(valid.count / nrow(D) * 100, 2)

Based on this, the dataset has 902297 observations, and out of these 635439 pieces have a non-valid event type; this is approximately 70.42% of the data, which implies a clean-up is seriously needed.

Upper case event types

Making the event types upper case has a good effect in itself. The following snippet calculates the measure of enhancement, which is simply the difference between the previous number of different event types and the current number. Similar statistic will be used throughout the complete event type transformation process.

begin.count <- length(unique(D$EVTYPE))
ev.count <- list(prev = length(unique(D$EVTYPE)),
                 cur = length(unique(D$EVTYPE2)),
                 diff = length(unique(D$EVTYPE)) - length(unique(D$EVTYPE2)))

Upper casing unified 985 different event types into 898 pieces, which is a decrease by 87.

Remove zero-damage records

The dataset contains a large number of records about weather events with no damage, i.e. where fatalities, injuries, property damage and crop damage are all 0. These records can be ignored and thus removed, because they do not affect the outcome of the analysis.

A new function called update.evcount is introduced which updates the ev.count counter list after a transformation step is completed. This function is used in all further steps of data processing.

update.evcount <- function(prev.count, cur.count)
{
  list(prev = prev.count,
       cur = cur.count,
       diff = prev.count - cur.count)
}

Zero-damage records are removed with a simple call to dplyr::filter function. Some statistics is also calculated here.

nrow.old <- nrow(D)

D <- D %>% filter(FATALITIES != 0 | INJURIES != 0 | PROPDMG != 0 | CROPDMG != 0)

nrow.diff <- nrow.old - nrow(D)
nrow.diffpc <- round(nrow.diff / nrow.old * 100, 2)

ev.count <- update.evcount(ev.count$cur, length(unique(D$EVTYPE2)))

This action removed 647908 records, i.e. approximately 71.81% of the data, and also unified 898 different event types into 443 pieces, which is a decrease by 455.

Remove junk from event type strings

This step takes the list of unique event type strings, and removes junk characters and words.

A new function called update.evtype2 is introduced which updates the EVTYPE2 column of the dataset with the new names, line by line. This function is used in several steps of the data processing.

update.evtype2 <- function(evtype2, old.values, new.values)
{
  for (i in 1:nrow(D))
  {
    index <- i
    evtype2[[index]] <- new.values[old.values == evtype2[index]]
  }

  return(evtype2)
}

The junk removal consists of the following steps:

1. non-alphabetical characters are replaced by spaces;
2. leading and trailing spaces are removed;
3. excessive spaces between alphabetical characters are removed (i.e. multiple spaces to single space);
4. space followed by a single character is removed;
5. AND and ON between two other words are removed.

evtypes.orig <- sort(unique(D$EVTYPE2))

evtypes.new <- gsub("[^A-Z]", " ", evtypes.orig)
evtypes.new <- gsub("^ +| +$", "", evtypes.new)
evtypes.new <- gsub("([A-Z]) +([A-Z])", "\\1 \\2", evtypes.new)
evtypes.new <- gsub(" [A-Z]{1}$", "", evtypes.new)
evtypes.new <- gsub(" (AND|ON) ", " ", evtypes.new)

D$EVTYPE2 <- update.evtype2(D$EVTYPE2, evtypes.orig, evtypes.new)

ev.count <- update.evcount(ev.count$cur, length(unique(D$EVTYPE2)))

Junk removal unified 443 different event types into 373 pieces, which is a decrease by 70.

Change plurals to singles

Several event types in the dataset differ only in pluralization, which are usually unnecessary duplicates of other event types.

First a list of suspected plurals is created by creating a vector of unique words of the event types, and then words ending with an S character are filtered out.

all.tokens <- strsplit(unique(D$EVTYPE2), split = " ")
all.tokens <- sort(unique(unlist(all.tokens)))
plurals <- grep("[A-Z]+S( |$)", all.tokens, value = TRUE)

This gives the following result: CONDITIONS, CURRENTS, FIRES, FLOES, FLOODS, GRASS, HAZARDOUS, JAMS, LANDSLIDES, MUDSLIDES, RAINS, ROADS, SEAS, SLIDES, SQUALLS, STORMS, SWELLS, TEMPERATURES, THUNDERSTORMS, THUNDERSTORMWINDS, TORNADOES, TREES, WAVES, WETNESS, WILDFIRES, WINDS, WINDSS, WINS.

Out of these HAZARDOUS and WETNESS are not pluralized words, and therefore they are removed from the plural list. WINDSS has two S characters at the end most probably by mistake, and as it has the same stem as WINDS, which is already on the list, WINDSS is also removed. For the remaining words, the trailing S character is simply removed to obtain the single form.

plurals <- plurals[!(plurals %in% c("HAZARDOUS", "WETNESS", "WINDSS"))]
singles <- gsub("S$", "", plurals)

Finally, the list of single word forms is used to build a regular expression, which is used to replace plurals to singles in the list of event types. The EVTYPE2 column is updated with the results.

evtypes.orig <- sort(unique(D$EVTYPE2))

regex.str <- paste0("(", paste(singles, collapse = "|"), ")", "S+( |$)")
evtypes.new <- gsub(regex.str, "\\1\\2", evtypes.orig)

D$EVTYPE2 <- update.evtype2(D$EVTYPE2, evtypes.orig, evtypes.new)

ev.count <- update.evcount(ev.count$cur, length(unique(D$EVTYPE2)))

Changing plural words to single words unified 373 different event types into 341 pieces, which is a decrease by 32.

Note: This step cannot handle special plural forms (like minima and maxima). However, this should not be an issue as no such event type was observed in the dataset.

Replacement of specific words

The result of the previous steps was checked manually for further possible enhancement of the set of event types. The result of the analysis is shown in the table below. This table is stored in the replacement.table data frame. For the sake of clarity, the R script of this is not echoed here, only the content of the replacement table is shown below.

	From	To	Type
1		UNNAMED	synonym
2	AVALANCE	AVALANCHE	correction
3	BLOWING SNOW	BLIZZARD	synonym
4	BRUSH FIRE	WILDFIRE	synonym
5	COASTAL SURGE	STORM SURGE TIDE	synonym
6	COASTALSTORM	COASTAL STORM	correction
7	DRY MICROBURST WIND	DRY MICROBURST	synonym
8	EROSION CSTL FLOOD	COASTAL FLOOD EROSION	correction
9	EXTREME COLD	EXTREME COLD WIND CHILL	synonym
10	EXTREME WINDCHILL	EXTREME COLD WIND CHILL	synonym
11	FLASH FLOOD FLOOD	FLASH FLOOD	correction
12	FLASH FLOOD FROM ICE JAM	ICE JAM FLOOD	correction
13	FLASH FLOOD THUNDERSTORM WI	FLASH FLOOD THUNDERSTORM WIND	correction
14	FLOOD FLASH	FLASH FLOOD	correction
15	FLOOD FLASH FLOOD	FLASH FLOOD	correction
16	FLOOD FLASHFLOOD	FLASH FLOOD	correction
17	FLOOD RIVER FLOOD	FLOOD	correction
18	FLOODING	FLOOD	general
19	FOREST FIRE	WILDFIRE	synonym
20	FREEZE	FROST FREEZE	synonym
21	FREEZING RAIN	SLEET	synonym
22	FROST	FROST FREEZE	synonym
23	GRAS FIRE	WILDFIRE	synonym
24	GROUND BLIZZARD	BLIZZARD	correction
25	GUSTY WIND HVY RAIN	GUSTY WIND HEAVY RAIN	correction
26	HAIL DAMAGE	HAIL	correction
27	HAILSTORM	HAIL	synonym
28	HARD FREEZE	FROST FREEZE	synonym
29	HAZARDOUS SURF	HIGH SURF	synonym
30	HEAVY SNOW SQUALL	BLIZZARD	synonym
31	HEAVY SURF HIGH SURF	HIGH SURF	correction
32	HIGH SQUALL	HIGH WIND	synonym
33	HIGH SWELL	HIGH SURF	synonym
34	HIGH WAVE	HIGH SURF	synonym
35	HIGH WIND DAMAGE	HIGH WIND	correction
36	HURRICANE EDOUARD	HURRICANE	correction
37	HURRICANE EMILY	HURRICANE	correction
38	HURRICANE ERIN	HURRICANE	correction
39	HURRICANE FELIX	HURRICANE	correction
40	HURRICANE GORDON	HURRICANE	correction
41	HURRICANE HURRICANE	HURRICANE	correction
42	HURRICANE OPAL	HURRICANE	correction
43	HURRICANE OPAL HIGH WIND	HURRICANE HIGH WIND	correction
44	HVY RAIN	HEAVY RAIN	correction
45	ICE JAM FLOOD	ICE JAM FLOOD	correction
46	ICE JAM FLOOD MINOR	ICE JAM FLOOD	correction
47	ICE ON ROAD	ICY ROAD	correction
48	ICE ROAD	ICY ROAD	correction
49	LAKE FLOOD	LAKESHORE FLOOD	correction
50	LATE SEASON SNOW	SNOW	synonym
51	LIGHT FREEZING RAIN	SLEET	synonym
52	LIGHT SNOWFALL	LIGHT SNOW	synonym
53	LIGHTING	LIGHTNING	correction
54	LIGHTNING FIRE	LIGHTNING	correction
55	LIGHTNING INJURY	LIGHTNING	correction
56	LIGHTNING THUNDERSTORM WIN	LIGHTNING THUNDERSTORM WIND	correction
57	LIGHTNING WAUSEON	LIGHTNING	correction
58	LIGNTNING	LIGHTNING	correction
59	LOW TEMPERATURE	COLD WIND CHILL	synonym
60	MAJOR FLOOD	FLOOD	correction
61	MARINE MISHAP	MARINE ACCIDENT	synonym
62	MICROBURST WIND	MICROBURST	synonym
63	MINOR FLOOD	FLOOD	correction
64	MIXED PERCIP	MIXED PERCIPITATION	correction
65	MUD SLIDE	LANDSLIDE	synonym
66	MUD SLIDEURBAN FLOOD	MUD SLIDE URBAN FLOOD	correction
67	MUDSLIDE	LANDSLIDE	synonym
68	RECORD COLD	EXTREME COLD WIND CHILL	synonym
69	RECORD EXCESSIVE HEAT	EXCESSIVE HEAT	correction
70	RECORD HEAT	EXCESSIVE HEAT	correction
71	RIP CURRENTHEAVY SURF	RIP CURRENT HIGH SURF	correction
72	RIVER FLOOD	FLOOD	correction
73	RIVER STREAM FLOOD	FLOOD	correction
74	ROCK SLIDE	LANDSLIDE	synonym
75	ROUGH SURF	HIGH SURF	synonym
76	RURAL FLOOD	FLOOD	correction
77	SEVERE THUNDERSTORM WIND	THUNDERSTORM WIND	correction
78	SMALL HAIL	HAIL	correction
79	SNOW BLOWING SNOW	BLIZZARD	synonym
80	SNOW FREEZING RAIN	SNOW SLEET	synonym
81	SNOW HEAVY SNOW	HEAVY SNOW	synonym
82	SNOW SLEET FREEZING RAIN	SNOW SLEET	correction
83	SNOW SQUALL	BLIZZARD	synonym
84	SNOWMELT FLOOD	FLOOD	synonym
85	STORM FORCE WIND	THUNDERSTORM WIND	synonym
86	STORM SURGE	HIGH SURF	synonym
87	THUDERSTORM WIND	THUNDERSTORM WIND	correction
88	THUNDEERSTORM WIND	THUNDERSTORM WIND	correction
89	THUNDERESTORM WIND	THUNDERSTORM WIND	correction
90	THUNDERSTORM DAMAGE TO	THUNDERSTORM	correction
91	THUNDERSTORM HAIL	HAIL	correction
92	THUNDERSTORM WIN	THUNDERSTORM WIND	correction
93	THUNDERSTORM WIND AND	THUNDERSTORM WIND	correction
94	THUNDERSTORM WIND AWNING	THUNDERSTORM WIND	correction
95	THUNDERSTORM WIND DAMAGE	THUNDERSTORM WIND	correction
96	THUNDERSTORM WIND FUNNEL CLOU	THUNDERSTORM WIND FUNNEL CLOUD	correction
97	THUNDERSTORM WIND HAIL	THUNDERSTORM WIND HAIL	correction
98	THUNDERSTORM WIND MPH	THUNDERSTORM WIND	correction
99	THUNDERSTORM WIND TREE	THUNDERSTORM WIND	correction
100	THUNDERSTORM WINDSHAIL	THUNDERSTORM WIND HAIL	correction
101	THUNDERSTORMW	THUNDERSTORM WIND	correction
102	THUNDERSTORMWIND	THUNDERSTORM WIND	correction
103	THUNDERSTROM WIND	THUNDERSTORM WIND	correction
104	THUNDERTORM WIND	THUNDERSTORM WIND	correction
105	THUNERSTORM WIND	THUNDERSTORM WIND	correction
106	TIDAL FLOOD	FLOOD	synonym
107	TORNADOE	TORNADO	correction
108	TORNADOE THUNDERSTORM WIND HAIL	TORNADO THUNDERSTORM WIND HAIL	correction
109	TORNDAO	TORNADO	correction
110	TROPICAL STORM	TROPICAL DEPRESSION	synonym
111	TROPICAL STORM ALBERTO	TROPICAL DEPRESSION	correction
112	TROPICAL STORM DEAN	TROPICAL DEPRESSION	correction
113	TROPICAL STORM GORDON	TROPICAL DEPRESSION	correction
114	TROPICAL STORM JERRY	TROPICAL DEPRESSION	correction
115	TSTM	THUNDERSTORM	general
116	TUNDERSTORM WIND	THUNDERSTORM WIND	correction
117	TYPHOON	HURRICANE	general
118	UNSEASONABLE COLD	UNSEASONABLY COLD	correction
119	UNSEASONABLY WARM DRY	UNSEASONABLY WARM	correction
120	URBAN FLOOD	FLASH FLOOD	synonym
121	URBAN SMALL	FLASH FLOOD	correction
122	URBAN SMALL STREAM	FLASH FLOOD	correction
123	URBAN SMALL STREAM FLOOD	FLASH FLOOD	correction
124	URBAN SMALL STREAM FLOODIN	FLASH FLOOD	correction
125	URBAN SML STREAM FLD	FLASH FLOOD	correction
126	WARM WEATHER	HEAT	synonym
127	WILD FIRE	WILDFIRE	correction
128	WILD FOREST FIRE	WILDFIRE	correction
129	WIND DAMAGE	WIND	correction
130	WIND STORM	THUNDERSTORM WIND	correction
131	WINTER WEATHER MIX	WINTER WEATHER	correction
132	WINTRY MIX	WINTER WEATHER	correction

The Type column indicates the nature of replacement:

• general: it is a replacement of a deviation frequent among event types;
• synonym: the two terms have the same meaning;
• correction: this indicates a correction of a typo, duplication, etc.

Replacements are made on the result of the previous step.

First the general replacements are made. In this case replacement is made on sub-strings of event type strings.

evtypes.orig <- sort(unique(D$EVTYPE2))
evtypes.new <- evtypes.orig

R <- replacement.table %>% filter(Type == "general")

for (i in 1:nrow(R))
{
  evtypes.new <- gsub(R$From[i], R$To[i], evtypes.new)
}

After this, the synonym and correction replacements are made. Here the replacement is made only if From exactly matches the event type string.

R <- replacement.table %>% filter(Type %in% c("synonym", "correction"))

for (i in 1:nrow(R))
{
  evtypes.new <- gsub(paste0("^", R$From[i], "$"), R$To[i], evtypes.new)
}

Finally the EVTYPE2 column in the dataset, and the count statistics are updated.

D$EVTYPE2 <- update.evtype2(D$EVTYPE2, evtypes.orig, evtypes.new)

ev.count <- update.evcount(ev.count$cur, length(unique(D$EVTYPE2)))

These replacements unified 341 different event types into 212 pieces, which is a decrease by 129.

Summary of Data Processing

Up to this point, the original 985 pieces of event type were unified into 212 pieces.

Further unification would require deeper understanding of the topic, and also a deeper analysis of the remarks to understand the non-valid event types. These are not done in this analysis.

Results

This analysis deals with the following questions:

1. Across the United States, which types of events are most harmful with respect to population health?
2. Across the United States, which types of events have the greatest economic consequences?

The answers to these questions require different variables from the dataset, but they can be constructed in the same way.

top.n <- 10

selector <- c("FATALITIES", "INJURIES", "PROPDMG", "CROPDMG", "ALLDMG")

ylabels <- c("Fatalities (people)", "Injuries (people)",
             "Property damage ($ billion)", "Crop damage ($ billion)",
             "Total damage ($ billion)")
names(ylabels) <- selector

The plots will show the 10 most harmful weather events for damage types FATALITIES, INJURIES, PROPDMG, CROPDMG, ALLDMG. All of these are totals across the whole United States, per weather event.

First, a grouping by EVTYPE2 is created for each of the damage types. In case of PROPDMG, CROPDMG and ALLDMG, values are divided by one billion.

tops <- list()

for (s in selector)
{
  D.grouped <- D %>%
    select(EVTYPE2, DATA = s) %>%
    group_by(EVTYPE2) %>%
    summarise(DATA = sum(DATA)) %>%
    arrange(desc(DATA)) %>%
    mutate(EVTYPE2 = factor(EVTYPE2, EVTYPE2))
  
  if (s %in% c("PROPDMG", "CROPDMG", "ALLDMG"))
    D.grouped <- D.grouped %>% mutate(DATA = DATA / 1e9)
  
  tops[[s]] <- D.grouped[1:top.n,]
}

For each of the top 10 tables, a column plot is created (but not yet shown), which will be used on the final figures.

grobs <- list()

for (s in selector)
{
  col.plot <- ggplot(tops[[s]], aes(EVTYPE2, DATA)) +
              geom_col(fill = I("steelblue")) +
              theme_light() +
              theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
              xlab("Weather event") +
              ylab(ylabels[s])
  
  grobs[[s]] <- ggplotGrob(col.plot)
}

Question 1: Across the United States, which types of events are most harmful with respect to population health?

The following table shows the top 10 most harmful weather events with respect to population health in the US. Values are given in people.

	Weather event	Fatalities	Weather event	Injuries
1	TORNADO	5633	TORNADO	91364
2	EXCESSIVE HEAT	1922	THUNDERSTORM WIND	9400
3	FLASH FLOOD	1064	FLOOD	6795
4	HEAT	937	EXCESSIVE HEAT	6575
5	LIGHTNING	817	LIGHTNING	5231
6	THUNDERSTORM WIND	706	HEAT	2102
7	RIP CURRENT	572	ICE STORM	1975
8	FLOOD	482	FLASH FLOOD	1879
9	EXTREME COLD WIND CHILL	305	WILDFIRE	1608
10	HIGH WIND	283	HIGH WIND	1440

The same result is summarised on the figure below.

Figure 1. Most harmful weather events across the US (population health)

Question 2: Across the United States, which types of events have the greatest economic consequences?

The following table shows the top 10 most harmful weather events with respect to the economy in the US. Values are given in billion dollars.

	Weather event	Property damage	Weather event	Crop damage	Weather event	Total damage
1	FLOOD	150.118974	DROUGHT	13.9725660	FLOOD	160.847331
2	HURRICANE	85.236335	FLOOD	10.7283579	HURRICANE	90.732453
3	TORNADO	56.930432	HURRICANE	5.4961178	TORNADO	57.345394
4	HIGH SURF	43.423376	ICE STORM	5.0221135	HIGH SURF	43.423381
5	FLASH FLOOD	16.989379	HAIL	3.0464705	HAIL	19.015614
6	HAIL	15.969144	FROST FREEZE	1.6300110	FLASH FLOOD	18.531344
7	THUNDERSTORM WIND	9.714430	FLASH FLOOD	1.5419653	DROUGHT	15.018672
8	WILDFIRE	8.496628	EXTREME COLD WIND CHILL	1.3300230	THUNDERSTORM WIND	10.903024
9	TROPICAL DEPRESSION	7.716128	THUNDERSTORM WIND	1.1885947	ICE STORM	8.967041
10	WINTER STORM	6.688997	HEAVY RAIN	0.7939028	WILDFIRE	8.899910

The same result is summarised on the figure below.

Figure 2. Most harmful weather events across the US (impact on economy)

Summary of Results

The analysis shows that tornados are the most harmful weather events with respect to population health - both from deaths and injuries point of view. It is also clearly visible that the impact of tornados on population health is much higher than any other weather event.

Regarding economic consequences, flood is the leading weather event for property damage and total damage. Drought is the leading weather event for crop damage, but flood is also there as second, with a relatively close value to drought.