Introduction

In this study we have analysed the NOAA Storm Database in order to determine what are the worst natural catastrophic events, both in terms of public health and in economic impact.

The U.S. National Oceanic and Atmospheric Administration’s (NOAA) storm database tracks characteristics of major storms and weather events in the United States, including when and where they occur, as well as estimates of any fatalities, injuries, and property damage.

The database currently contains data from January 1950 to January 2017, as entered by NOAA’s National Weather Service (NWS).

The database can be found on:

https://www.ncdc.noaa.gov/stormevents/ftp.jsp

RPubs version: http://rpubs.com/erickfis/noaa

GitHub version, with code included and pdf version: https://github.com/erickfis/NOAA-Storm-Database

Objective

The goal of this study is to answer the questions:

  1. Across the United States, which types of events were the most harmful with respect to population health ever recorded in a single occurrence?

  2. Which types of events caused most harm to population health along all those years?

  3. Which types of events had the greatest economic consequences in a single occurrence?

  4. Which types of events had the greatest economic consequences along all those years?

  5. Which were the places that were subject to the greatest losses, both in terms of human health and economic losses.

Methods

To answer each one of those questions, we did a very simple descriptive analysis of data.

We used R tools to filter, sort and combine data, so we could get the total sum of fatalities, injuries and economic losses.

Data Processing

Software and instructions for reproducibility

This study was conducted under the following software:

  • R version 3.4.1 (2017-06-30)
  • Platform x86_64-pc-linux-gnu
  • OS: Linux Mint 18.1 Serena

Using the same versions under the same OS should guaranty reproducibility.

This script downloads the relevant raw data from NOAA repository, under ftp://ftp.ncdc.noaa.gov/pub/data/swdi/stormevents/csvfiles/, and process it.

It is not necessary to keep the files downloaded after running this script because it writes the processed data to the file “data/harm.rda”

Then everytime it is runned, it checks for new data on the ftp server. If there is new data, it will download the files missing and raw process everything again.

Raw data processing

In order to answer our questions, the original database needed to be treated from its raw form to a more useful format.

The necessary transformations were:

  • sanitized var names
  • evaluated duration of events, however they are not useful
  • evaluated damages values according to multipliers provided
  • sanitized and grouped similar events: strong snow, heavy snow and light snow all became just “snow”
  • sanitized county names

This database has 1436408 observations. Each observation corresponds to an event occurrence.

To determine the most harmful events to human health, we checked the variables related to human health, which are “fatalities” and “injuries”.

To determine the most harmful events to economy, we checked the variables related to economic measures, from “propdmg” through “cropdmgexp”.

Also, in order to analyse various occurrences of the same event, we measured the duration of the event, its magnitude and where the event occurred (state and county name).

This is a really big database whose data has been being registered by a lot of different people since 1950. Thus, as expected, there are variations on how people registered events.

For example, the string “snow” was used to register a lot of events. They are the same type of event, but count as different:

This is why we decided to filter those events: we grouped them by its common strings.

Human health: the most harmfull events

We have determined what events did more harm to human health.

There were occurrences that caused zero fatalities but a lot of injuries. The inverse is also true, so we did a separate analysis to fatal and non-fatal events.

Fatal Occurrences

Most fatal in a single occurrence

Most fatal in a single occurrence

In order to determine what were the most fatal events in a single occurrence, we need to see how fatalities are distributed along the occurrences.

Looking at this distribution, we can infer that the vast majority of those occurrences were not fatal at all: 99.3% occurrences didn’t caused any fatalities.

On the other hand, fatal occurrences had to have at least 1 fatality.

Now, among the fatal occurrences, we are interested in the ones whose fatalities are beyond the confidence interval, ie. above 99% of the most common values.

Looking at this distribution, we can infer that 99.8% of the fatal occurrences caused up to 56.963 fattalities.

Population distribution for fatalities / occurrences

Population distribution for fatalities / occurrences

In this study, we looked on the 1% deadliest occurrences.

Worst fatal occurrences, mean = 1.9 and median = 1
rank event day state county fatalities
1 hurricane 2005-08-28 louisiana orleans 638
2 tornado 2011-05-22 missouri jasper 161
3 hurricane 2005-08-28 louisiana lower.st.bernard 140
4 tornado 1953-06-08 michigan genesee 116
5 tornado 1953-05-11 texas mclennan 114
6 hurricane 2005-08-28 mississippi harrison 97
7 heat 1999-07-28 illinois cook 93
8 tornado 1953-06-09 massachusetts worcester 90
9 tornado 1955-05-25 kansas cowley 75
10 heat 1999-07-04 pennsylvania philadelphia 58
11 tornado 1966-03-03 mississippi hinds 57

Worst fatal occurrences

The single most fatal event was a hurricane, that occurred in louisiana, orleans, on 2005-08-28, killing 638 people.

However, if we compare this single awful event to the mean of fatalities caused, we see that this is very unlikely to happen.

Most fatal in all time

Most fatal in all time

Notice that are several occurrences of the same type of event along the time.

Therefore, in order to know which is the worst type of event along all the years, we summed up the fatalities caused by each one of occurrences of this events.

Notice that we are interested only in the worst of them, ie, the ones which are above the mean.

Total fatalities by event, mean = 680.93 and median = 187.5
rank event total
1 tornado 5887
2 heat 2855
3 wind 2325
4 flood 1959
5 winter 1211
6 hurricane 1128
7 lightning 834
8 rip current 806

Total fatalities by event

The most fatal event along the time is the tornado. It has killed 5887 people until now.

Least fatal events

Just for curiosity, these are the less fatal among the fatal events:

Least fatal events
rank event total
28 tropical depression 1
27 dense smoke 2
26 sleet 2
25 waterspout 2
24 cold 4
23 dust devil 4
22 slide 4
21 sneakerwave 15
20 hail 21
19 tide 22

Injuring Occurrences

Most injuring in a single occurrence

Most injuring in a single occurrence

In order to determine what were the most injuring events in a single occurrence, we need to see how injuries are distributed along the occurrences.

Looking at this distribution, we can infer that the vast majority of those occurrences were not injuring at all: 98.5% occurrences didn’t caused any injuries

On the other hand, injuring occurrences had to have at least 1 injury.

Now, among the injuring occurrences, we are interested in the ones whose harm is beyond the confidence interval, ie. above 99% of the most common values.

Looking at this distribution, we can infer that 99.8% of the injuring occurrences caused up to 500 injuries.

Population distribution for Injuries / occurrences

Population distribution for Injuries / occurrences

In this study, we looked on the 1% most injuring occurrences.

Worst injuring occurrences, mean = 7.54 and median = 2
rank event day state county injuries
1 hurricane 2008-09-12 texas harris 2400
2 tornado 1979-04-10 texas wichita 1700
3 tornado 1953-06-09 massachusetts worcester 1228
4 tornado 1974-04-03 ohio greene 1150
5 tornado 2011-05-22 missouri jasper 1150
6 flood 1998-10-17 texas comal 800
7 tornado 2011-04-27 alabama tuscaloosa 800
8 tornado 1953-06-08 michigan genesee 785
9 hurricane 2004-08-13 florida charlotte 700
10 tornado 2011-04-27 alabama jefferson 700
11 flood 1998-10-17 texas bexar 600
12 tornado 1953-05-11 texas mclennan 597
13 tornado 1965-04-11 indiana howard 560
14 heat 2007-08-04 missouri st.louis 519
15 tornado 1966-03-03 mississippi hinds 504

Worst injuring occurrences

The single most injuring event was a hurricane, that occurred in texas, harris, on 2008-09-12, injuring 2400 people.

However, if we compare this single awful event to the mean of injuries caused, we see that this is very unlikely to happen.

Most injuring in all time

Notice that are several occurrences of the same type of event along the time.

Therefore, in order to know which is the worst type of event along all the years, we summed up the injuries caused by each one of occurrences of this events.

Notice that we are interested only in the worst of them, ie, the ones which are above the mean.

Total injuries by event, mean = 5212.52 and median = 317
rank event total
1 tornado 94815
2 heat 15436
3 wind 13518
4 flood 8809
5 winter 8251

The most injuring event along the time is the tornado. It has injuried 94815 people until now.

Total Injuries by event

Least injuring events

Just for curiosity, lets show now what are the less injuring among the injuring events:

Least injuring events
rank event total
31 funnel 3
30 tropical depression 3
29 waterspout 3
28 other 4
27 drought 8
26 sleet 10
25 sneakerwave 12
24 slide 13
23 cold 15
22 dense smoke 17

Economy: the the most harmfull events

We have determined what events did more harm to economy, both in terms of property and crops damage.

There were events that causes zero property damage but a lot of crop damage. The inverse is also true, so we did a separate analysis to property VS crop damaging events.

Property losses

Most Property Damaging event in a single occurrence

In order to determine what were the most property damaging events in a single occurrence, we need to see how damages are distributed along the occurrences.

Looking at this distribution, we can infer that 99.8% of the occurrences caused less than $38,000,000 in losses.

On the other hand, damaging occurrences had to have damages above zero.

Now, among the damaging occurrences, we are interested in the ones whose damages are above 99.8% of the most common values.

Looking at this distribution, we can infer that 99.8% of the damaging occurrences caused up to $125,282,000 in losses.

Population distribution for losses / occurrences

Population distribution for losses / occurrences

In this study, we looked on the 1% most harmful occurrences.

Worst property damaging occurrences, mean = $1,129,938 and median = $10,000
rank event day state county value
1 tide 2005-08-29 louisiana orleans $17,900,000,000
2 hurricane 2005-10-24 florida .palm.beach $10,000,000,000
3 flood 2012-10-29 new.jersey eastern.ocean $7,500,000,000
4 tide 2005-08-29 mississippi harrison $5,630,000,000
5 storm 2001-06-05 texas harris $5,030,000,000
6 flood 2012-10-28 new.jersey eastern.monmouth $5,000,000,000
7 flood 2012-10-28 new.jersey western.monmouth $5,000,000,000
8 hurricane 2004-09-13 florida .escambia $4,000,000,000
9 tide 2008-09-12 texas galveston $4,000,000,000
10 hurricane 2005-08-28 louisiana orleans $3,560,000,000
11 tide 2005-08-29 mississippi hancock $3,380,000,000
12 tide 2005-08-29 louisiana st.tammany $3,030,000,000
13 tide 2005-08-29 louisiana lower.plaquemines $3,030,000,000
14 tide 2005-08-29 louisiana lower.st.bernard $3,020,000,000
15 tide 2005-08-29 louisiana upper.st.bernard $3,020,000,000
16 flood 1997-04-18 north.dakota grand.forks $3,000,000,000
17 hurricane 1999-09-15 north.carolina alamance $3,000,000,000
18 hurricane 2004-08-13 florida charlotte $3,000,000,000
19 tide 2008-09-12 texas harris $3,000,000,000
20 hurricane 2005-08-28 mississippi harrison $2,940,000,000

Worst property damaging occurrences

The single most economic damaging event to properties was a tide, that occurred in louisiana, orleans, on 2005-08-29, causing U$ $17,900,000,000 in losses.

Most Property Damaging event in all time

Notice that are several occurrences of the same type of event along the time.

Therefore, in order to know which is the worst type of event along all the years, we summed up the losses caused by each one of occurrences of this events.

Notice that we are interested only in the worst of them, ie, the ones which are above the mean.

Total property losses by event, mean = $11,433,157,544 and median = $187,530,000
rank event total
1 hurricane $87,005,170,310
2 flood $83,736,516,680
3 tornado $63,741,926,192
4 tide $54,155,102,600
5 wind $25,476,530,618
6 hail $25,402,474,594
7 storm $16,754,690,360

Total Property Damages by event

The most property damaging event along the time is the hurricane. It has caused $87,005,170,310 in losses.

Least property damaging events

Just for curiosity, these are the less damaging events:

Least property damaging events
rank event total
33 other $1,000
32 sneakerwave $100,000
31 funnel $123,100
30 dense smoke $130,000
29 rip current $163,000
28 volcanic ash $500,000
27 dust devil $1,177,430
26 seiche $1,402,000
25 sleet $3,084,000
24 avalanche $4,060,050
e

Crop losses

Most Crop Damaging event in a single occurrence

In order to determine what were the most crop damaging events in a single occurrence, we need to see how damages are distributed along the occurrences.

On the other hand, damaging occurrences had to have damages above zero.

Now, among the damaging occurrences, we are interested in the ones whose damages are above 99% of the most common values.

Looking at this distribution, we can infer that 99.8% of the damaging occurrences caused up to $196,735,000 in losses.

Population distribution for losses / occurrences

Population distribution for losses / occurrences

In this study, we looked on the 1% most harmful occurrences.

Worst crops damaging occurrences, mean = $1,805,596 and median = $20,000
rank event day state county value
1 drought 2014-12-01 california northernsanjoaquin $1,500,000,000
2 drought 2011-06-01 texas lubbock $1,050,000,000
3 drought 2006-01-01 texas montague $1,000,000,000
4 drought 2007-06-01 mississippi warren $700,000,000
5 cold 2007-01-11 california sesj $568,600,000
6 drought 2000-11-01 texas parmer $515,000,000
7 drought 1998-07-06 oklahoma choctaw $500,000,000
8 drought 1999-07-01 pennsylvania potter $500,000,000
9 hurricane 1999-09-15 north.carolina alamance $500,000,000
10 flood 2000-10-03 florida .dade $500,000,000
11 flood 2007-07-01 missouri henry $500,000,000
12 wind 1998-12-20 california southernsanjoaquin $490,500,000
13 drought 1998-12-01 texas yoakum $450,000,000
14 hurricane 2005-08-25 florida .dade $423,000,000
15 drought 2001-12-01 texas parmer $420,000,000
16 drought 2007-09-01 georgia baldwin $344,000,000
17 drought 2006-02-01 texas fannin $300,000,000
18 cold 2010-01-10 florida inlandcollier $300,000,000
19 cold 2010-01-10 florida inland.miami.dade $286,000,000
20 drought 1998-12-01 texas andrews $250,000,000

Worst crops damaging occurrences

The single most economic damaging event to crops was a drought, that occurred in california, northernsanjoaquin, on 2014-12-01, causing U$ $1,500,000,000 in losses.

Most Crop Damaging event in all time

Notice that are several occurrences of the same type of event along the time.

Therefore, in order to know which is the worst type of event along all the years, we summed up the losses caused by each one of occurrences of this events.

Notice that we are interested only in the worst of them, ie, the ones which are above the mean.

Total crops losses by event, mean = $2,857,717,345 and median = $451,354,485
rank event total
1 drought $27,454,862,620
2 flood $7,750,266,270
3 hurricane $5,341,874,800
4 cold $5,123,993,200
5 wind $4,429,685,250
6 hail $3,657,850,413

Total Crop Damages by event

The most crop damaging event along the time is the drought. It has caused $27,454,862,620 in losses.

Least crops damaging events

Just for curiosity, lets show now what are the less damaging among the events:

Least crops damaging events
rank event total
20 slide $17,000
19 tsunami $20,000
18 tide $955,000
17 surf $1,000,000
16 blizzard $7,060,000
15 lightning $7,422,670
14 debris flow $20,006,500
13 winter $46,924,000
12 snow $91,145,900
11 fire $448,468,860

Most aflicted locations

We have determined what locations had the worst outcome from those events, both in terms of human health and economic losses.

Unfortunately, these has been the worst counties for living in:

Worst fatality count

Total fatalities by county
rank state county fatalities injuries prop.dmg crop.dmg
1 louisiana orleans 649 132 $21,614,049,550 $0
2 illinois cook 565 912 $670,237,350 $0
3 pennsylvania philadelphia 387 455 $52,680,980 $0
4 nevada lasvegas 264 601 $14,762,000 $0
5 texas harris 216 2827 $10,891,093,870 $7,450,000
6 missouri jasper 178 1273 $2,864,881,330 $46,475,500
7 texas dallas 149 1757 $1,946,641,730 $1,405,000
8 louisiana lower.st.bernard 140 0 $4,845,022,000 $0
9 texas mclennan 127 657 $65,192,600 $1,710,500
10 michigan genesee 123 962 $132,151,750 $6,300,000

The county with the biggest fatality count is orleans, in louisiana, with 649 people killed.

Worst injuries count

Total injuries by county
rank state county fatalities injuries prop.dmg crop.dmg
1 missouri st.louis 65 3144 $1,461,882,880 $10,500
2 texas harris 216 2827 $10,891,093,870 $7,450,000
3 missouri st.louis. 118 2701 $79,552,000 $5,000
4 texas wichita 55 1853 $310,822,880 $0
5 texas dallas 149 1757 $1,946,641,730 $1,405,000
6 alabama jefferson 117 1700 $2,037,082,100 $3,355,000
7 massachusetts worcester 96 1292 $286,072,530 $0
8 ohio greene 40 1278 $289,980,757 $540,000
9 missouri jasper 178 1273 $2,864,881,330 $46,475,500
10 oklahoma oklahoma 79 1253 $1,356,150,290 $8,330,000

The county with the biggest injuries count is st.louis, in missouri, with 3144 people injuried.

Worst property losses

Total property losses by county
rank state county fatalities injuries prop.dmg crop.dmg
1 louisiana orleans 649 132 $21,614,049,550 $0
2 texas harris 216 2827 $10,891,093,870 $7,450,000
3 florida .palm.beach 5 7 $10,828,630,000 $75,000,000
4 mississippi harrison 110 90 $8,870,659,460 $0
5 new.jersey eastern.ocean 16 112 $8,116,441,690 $10
6 new.jersey eastern.monmouth 13 397 $6,527,278,550 $0
7 louisiana st.tammany 8 89 $5,677,642,950 $0
8 florida .escambia 14 0 $5,632,695,000 $25,300,000
9 texas galveston 44 259 $5,359,033,770 $109,602,000
10 new.jersey western.monmouth 6 84 $5,267,488,450 $0

The county with the biggest property losses is orleans, in louisiana, with $21,614,049,550 in losses.

Worst crops losses

Total crops losses by county
rank state county fatalities injuries prop.dmg crop.dmg
1 texas lubbock 37 679 $2,007,426,360 $2,439,945,000
2 texas montague 5 42 $118,971,700 $1,963,106,500
3 california northernsanjoaquin 14 25 $6,058,500 $1,520,000,000
4 texas parmer 1 23 $44,657,090 $1,181,360,000
5 florida .dade 10 1 $693,020,000 $1,168,000,000
6 california sesj 28 64 $6,167,300 $992,223,000
7 mississippi warren 43 341 $314,976,880 $728,657,000
8 california ecentralsj 43 123 $7,706,800 $578,212,000
9 california southernsanjoaquin 1 22 $18,657,000 $517,800,000
10 north.carolina alamance 2 8 $3,005,157,200 $503,166,000

The county with the biggest croperty losses is lubbock, in texas, with $2,439,945,000 in losses.

Results

Population Health

Population Health: fatalities and injuries

Population Health: fatalities and injuries

The single most fatal event was a hurricane, that occurred in louisiana, orleans, on 2005-08-28, killing 638 people.

The most fatal event along the time is the tornado. It has killed 5887 people until now.

The single most injuring event was a hurricane, that occurred in texas, harris, on 2008-09-12, injuring 2400 people.

The most injuring event along the time is the tornado. It has injuried 94815 people until now.

Economic Damages

Economic Damages: property and crops

Economic Damages: property and crops

The single most economic damaging event to properties was a tide, that occurred in louisiana, orleans, on 2005-08-29, causing U$ $17,900,000,000 in losses.

The most property damaging event along the time is the hurricane. It has caused $87,005,170,310 in losses.

The single most economic damaging event to crops was a drought, that occurred in california, northernsanjoaquin, on 2014-12-01, causing U$ $1,500,000,000 in losses.

The most crop damaging event along the time is the drought. It has caused $27,454,862,620 in losses.

Most aflicted locations

The county with the biggest fatality count is orleans, in louisiana, with 649 people killed.

The county with the biggest injuries count is st.louis, in missouri, with 3144 people injuried.

The county with the biggest property losses is orleans, in louisiana, with $21,614,049,550 in losses.

The county with the biggest croperty losses is lubbock, in texas, with $2,439,945,000 in losses.

Distribution of data

Population distributionPopulation distribution