FEMA Disaster Declarations

What can we learn from disaster data in the US?

Posted by Maya Sandler on November 28, 2020

Intent

Maybe it is the timing, COVID-19 pandemic and the huge California fires going around, but I was interested in specific information regarding disasters in the US in general, and specifically in California.

I hypothesize that natural disasters are not random, and that specific states are more prone to specific disasters. This is important to determine the potential risk a location has for specific type of disaster, in order to know how to prepare for one.

To test my hypothesis, I used analyzing data of all disasters reported in the USA between the years 1953-2020, collected by FEMA. The data was imported to MySQL, preprocessed, analyzed and visualized via Tableau.

I used the data to answer these questions:

Is there a change in the disasters along the years? Are there specific types that change?
What is the different between difference states in the total number of disaster incidents? Are there specific disaster types per state?
How do seasons effect the amount of specific disaster types?

In the last part of the project, I have given my conclusions about disasters and their geographic location across the USA.

Description of the Data

The database contains the following data:

femaDeclarationString
disasterNumber: A sequentially assigned number
state: State abbreviation
declarationType: Type of disaster
declarationDate: Date the disaster was declared
fyDeclared: Fiscal year that the disaster occurred
incidentType: Biological, hurricane, etc.
declarationTitle: Specific types (i.e. COVID-19, hurricane Matthew)
ihProgramDeclared: Denotes whether household assistance (IH) was declared
iaProgramDeclared: Denotes whether individual assistance (IA) was declared
paProgramDeclared: Denotes whether public assistance (PA) was declared
hmProgramDeclared: Denotes whether hazard mitigation (HM) was declared
incidentBeginDate
incidentEndDate
disasterCloseoutDate
fipsStateCode
fipsCountyCode
placeCode: FEMA code
designatedArea: County or area name
declarationRequestNumber
hash: MD5 Hash
lastRefresh
id: Unique ID assigned to the record

Import the Dataset

The data has been obtained from FEMA website and downloaded directly here.

The data was imported to MySQL using a Python code (because my computer refused to import the data for security reasons).

Preprocessing

Review of Flues and Missing Information

Through out data of disasters before 1965 (161 out of total of 60,296 incidents), as they were rarely reported in the dataset and therefore, will create false results in comparison analyses. Also, FEMA was established in 1979, so earlier data is lacking due to under reporting.
Detected missing data of incidentEndDate (8339 incidents), disasterCloseoutDate (56419 incidents), hash (68 incidents). This doesn't effect the data, but it is important to know the limitations of the database and to allow Null in these columns.

Changing Specific Characters in String that Interrupt Code

In several strings there were specific characters in the string that interrupted with the analysis code (apostrophe " ' "). To avoid these disruptions, the apostrophes were replaced by a double slash ("//") in this python code, as part of the import data code:

State Name Encoding

Sate appear in the database as a numeric code and a two-letter state abbreviation. In order to categorize and later visualize the data per state, a new chard with state names was created in MySQL:

Adding Season information

To inquire about specific seasons and the likelihood of having specific disasters in them, a seasons table was added to the database in MySQL. This was done according to year's quarter (1st = Jan-March, 2nd = April-June, 3rd = July-Sep, 4th = Oct-Dec):

Great! The data is now cleaned, supplemented with state-names, and ready to be analyzed and draw insights.

Results:

Distribution of incidents along the years

To answer the first question, we begin by finding out if there a change in the disasters quantity along the years I looked into all disaster cases along the years. The data shows a significant increase of the total amount of cases, as demonstrated from the polynomial (3 degrees; R squared = 0.33; P-value < 0.001) trendline (in black) shows in the image below.

Are there specific types that contribute to the increase?

To answer this question, I filtered the data according to disaster type. This way, we can see the quantity of each disaster type along the year and look for a significant change.

The results show that there is a significant increase in:

(1) Severe Storms, significantly increasing after the year 1990:

(2) Hurricanes, significantly increasing after the year 1996:

(3) fires, significantly increasing after the year 1996:

(4) Severe ice storms, significantly increasing after the year 2001:

(5) A highly significant increase of biological incidents (COVID-19 cases) in 2020:

Comparing States

The second enquiry I looked into, was the differences between states in the total number of disaster incidents, and weather there are specific disaster types per state. To answer this, I filtered the states as well as disaster types, and found some interesting insights:

(1) The main contributor states to Hurricanes in the US are:

Virginia - maintain an overall constant high rate of hurricanes since 1996 (14-40%).
Florida - can reach 20-100% of the overall hurricane incidents in a fiscal year, since 1998.
North Carolina - has a constant high contribution (10-38% per year) to hurricane cases in the US since 1996.

(2) The main contributor states to fires in the US are:

Texas - which had major fires that constitute 62%-79% in specific years, but most years reaching only 2-5%.
California - maintain an overall constant 15-50% of the fires along the years, without a significant increase in recent years (as I was suspecting in the beginning).
Colorado - which contributed to fires in the US mainly between the years 1998-2001 and after that decreased the fire cases significantly (from 16-65% to 0-3%).

Season Effect On Disaster Types

The third enquiry I looked into was weather the seasons has an affect on specific disaster types. To answer this, I filtered the seasons and disaster types and found several interesting insights:

(1) Storms:

Even though typhoon, tornado and hurricane are all weather phenomenon different in their oceanic origin, each type of storm showed a "favorite" season that was significantly different from the others. The below image illustrates the seasons when these storms are most likely to happen. It however doesn't take into account overall increase in the intensity of storms, that is reported in research (as noted here).

(2) Earthquakes and volcanos:

The analysis of FEMA database showed that the time in the year of earthquakes and volcanos are different, and therefore are unlikely to be related. However, it does not show a complete picture, as large regional earthquakes (greater than magnitude 6) are considered to be related to a subsequent eruption or to some type of unrest at a nearby volcano (as noted here). To have more insights on this relation we need the magnitude and scope of each earthquake, and this data is missing from FEMA database.

(3) Fire in California:

I know, I am biased because I live in California and right now there are multiple fire locations - and it is not considered to be "fire season". But I was intrigued to know more about the actual fire season and the quantity of reported incidents. According to this study, there are two "fire seasons": The first is in the 3rd quarter (summer), driven by a combination of warmer and drier weather (called the "Western fire season"), and the second is in the 4th and 1st quarters (winter+fall), driven by the Santa Ana winds and tend to spread three times faster and burn closer to urban areas than the Western fire season.

MY FEMA analysis, shown in the below image, reflects the two picks in fire seasons in summer and fall and answer to the quantity, but does not reflect the intensity and destructive properties of the fires. This analysis does not show an increase of one of the types of fire seasons, but does indicate an increase in fires in both seasons since the year 2002:

Conclusions

We took FEMA database of disaster types, locations and times and derived several insights:

We note a significant increase in disaster quantity (total and specific types: severe storms, hurricanes, fires and severe ice storms, in addition to COVID-19 pandemic) in the US.
Specific states are more prone to specific disasters, and several states show an increase in the past 20-25 years.
Specific disaster types are more likely to occur at specific times of the year.
California has two picks of seasonal fires which increased significantly after 2002 and is mostly constans.