Natural Hazard Histories Dataset

Hurricane data was downloaded as a shapefile and was originally titled "IBTrACS.since1980.list.v04r00.lines". This shapefile, in addition to a 2018 US Census Bureau's county boundaries shapefile, was brought into ArcGIS Pro version 2.6.3. A definition query was used to limit hurricane tracks or features to just those of interest - hurricane tracks occurring in or after the year 2000. To do so, the query was written as "year >= 2000 and (TRACK_TYPE = 'main' or TRACK_TYPE = 'PROVISIONAL') and USA_STATUS LIKE 'H%' and USA_SSHS >= 1". Once limited, a copy of this shapefile was created.

To illustrate possible area of impact, each feature within this copy was buffered. To create these buffers two new fields or columns were calculated. Four 34-knot wind radii fields were provided (one for each quadrant). The max value across these four fields was found, multiplied by 1852 (converting from nautical miles to associated map unit, meters), and reported within the first field. Then, the average of these max values per each hurricane category was found using the summary statistics tool. Each category's mean max value was applied to all associated features and recorded in the second field. The buffer was based on this second field. Within this tool, a dissolve field can be specified. Features were buffered and subsequently dissolved by year, so that one feature would represent total exposure for a given year. Please note that one warning message, 000635: Skipping feature <value> because of NULL or EMPTY geometry, was generated upon the buffer tool's completion. Features with a "Shape_Length" equal to 0 were the reason for this message.

To calculate percent area of county exposed to hurricane events, this buffer output was summed to the US county shapefile via the sum within tool. Within this tool, the group by field was specified as "year", so that percent area per county per year was recorded. This tool generated a feature class containing all US counties, and a join table which held percent area per year by county. Counties with no percent area for a given year were not reported within the join table. Sum within trials and associated join tables for the contiguous US, HI, and PR (regions with percent area recorded) were exported as csv files and brought into Excel for further analysis.

"TRACK_TYPE" was specified as either "main" or "PROVISIONAL" because the remaining values, "spur" and "PROVISIONAL_spur", essentially represent alternate positions of the storm, which were noted to be short lived and not included when counting storms. For more information on "TRACK_TYPE" please refer to the IBTrACS v04 column documentation. 

The four 34-knot wind radii quadrants or tropical storm force wind fields were chosen as the buffer's starting point, instead of the 50-knot or 64-knot quadrants, because they illustrate the largest possible extent or area exposed to a hurricane. The buffer was not based on each 34-knot wind radii quadrant value, or even the max value found across these four fields for each feature, because many of these quadrant fields contained a value of zero; this value represented no data. Without an average max value per category, these features would not have been buffered. The concept of finding an average value per category to represent possible area of exposure was derived from FEMA's NRI technical document (https://www.fema.gov/sites/default/files/documents/fema_national-risk-index_technical-documentation.pdf) - see hurricane, spatial processing section for reference.

The same processing methods used for hurricane events were applied to tropical storm data. The only exception being the definition query used to limit the original shapefile mentioned previously. The query was written as "year >= 2000 and (TRACK_TYPE = 'main' or TRACK_TYPE = 'PROVISIONAL') and USA_STATUS = 'TS' and USA_SSHS = 0".

Refer to rationale provided under the previous process step.

Tornado data was downloaded as a shapefile and was originally titled "1950-2018-torn-aspath". This shapefile, in addition to a 2018 US Census Bureau's county boundaries shapefile, was brought into ArcGIS Pro version 2.6.3. A definition query was used to limit tornado tracks or features to just those of interest - tornadoes occurring in or after the year 2000 with a magnitude greater than or equal to 0. To do so, the query was written as "yr >= 2000 and mag >= 0". Once limited, a copy of this shapefile was created.

To illustrate possible area of exposure, each feature within this copy was buffered. To create these buffers two new fields or columns were calculated. A width field, called "wid", was provided. For each feature, this width value was taken, divided by 1.094 (converting from yards to associated map unit, meters), and reported within the first field. Then, the width value in meters was divided by 2 and reported in the second field, since the width represented the entire width of the event and not just storm center to edge. The buffer was based on this second field. Within this tool, a dissolve field can be specified. Features were buffered and subsequently dissolved by year, so that one feature would represent total exposure for a given year. Please note that two warning messages, "000635: Skipping feature <value> because of NULL or EMPTY geometry" and "000636: Skipping feature <value> because a negative or very small distance resulted in no geometry", were generated upon the buffer tool's completion. Features with a "Shape_Length" or "wid" equal to 0 were the reason for these messages.

To calculate percent area of county exposed to tornadoes, this buffer output was summed to the US county shapefile via the sum within tool. Within this tool, the group by field was specified as "yr", so that percent area per county per year was recorded. This tool generated a feature class containing all US counties, and a join table which held percent area per year by county. Counties with no percent area for a given year were not reported within the join table. Sum within trials and associated join tables for the contiguous US, HI, and PR (regions with percent area recorded) were exported as csv files and brought into Excel for further analysis.

The "wid" field was chosen as the buffer's starting point because, other than length, it was the only extent-related information given. Events with a magnitude of less than 0 (value of -9) were excluded from further processing because their magnitude was unknown.

Landslide data was downloaded as a shapefile and was originally titled "US_Landslide_poly". This shapefile, in addition to a 2018 US Census Bureau's county boundaries shapefile, was brought into ArcGIS Pro version 2.6.3. A definition query was used to limit landslide events or features to just those of interest - landslides occurring in or after the year 2000 with a confidence greater than or equal to 3. To do so, the query was written as "Confidence >= 3 and (Date like '2000%' or Date like '2001%'...". Please note that the date field was formatted as a text field and was the only temporal field provided. Many dates were accompanied by text in various formats. Also, "like" within the query means "begins with". Once limited, a copy of this shapefile was created.

No buffer was necessary since events were presented as polygons. Although the buffer tool was not used, the dissolve tool was still needed to collapse events into a single feature per year. The dissolve was based on a field called "YEAR", which was created by extracting the first 4 characters of the "Date" field.

To calculate percent area of county exposed to landslides, this dissolve output was summed to the US county shapefile via the sum within tool. Within this tool, the group by field was specified as "YEAR", so that percent area per county per year was recorded. This tool generated a feature class containing all US counties, and a join table which held percent area per year by county. Counties with no percent area for a given year were not reported within the join table. Sum within trials and associated join tables for the contiguous US and AK (regions with percent area recorded) were exported as csv files and brought into Excel for further analysis.

Events with a confidence value of less than 3 (values of 1 or 2) were omitted because of the uncertainty associated with the event's location. The source's metadata document provides a brief description of each confidence level. The following source offers more comprehensive definitions (https://kgs.uky.edu/kgsmap/helpfiles/landslide_help.shtm) - see below. 

"1 - Possible landslide occurred in the area: The lowest confidence level reflects the uncertain nature of some media reports and the lack of expert classification and characterization from old maps. 2 - Probable landslide in the area: Although the exact location and extent of the landslide is not documented, it is probable that a landslide did occur within close proximity to the specified location. This includes geologic mapping of landslide deposits that may correspond to multiple landslides as well as individual landslides mapped with low resolution topographic data."

Wildfire data was downloaded as a shapefile and was originally titled "InteragencyFirePerimeterHistory". This shapefile, in addition to a 2018 US Census Bureau's county boundaries shapefile, was brought into ArcGIS Pro version 2.6.3. A definition query was used to limit wildfire events or features to just those of interest - wildfires occurring in or after the year 2000. To do so, the query was written as "FIRE_YEAR like '2000%' or FIRE_YEAR like '2001%'...". Please note that the date field was formatted as a text field. Also, "like" within the query means "begins with". Once limited, a copy of this shapefile was created.

No buffer was necessary since events were presented as polygons. Although the buffer tool was not used, the dissolve tool was still needed to collapse events into a single feature per year. The dissolve was based on the "FIRE_YEAR" field.

To calculate percent area of county exposed to wildfire, this dissolve output was summed to the US county shapefile via the sum within tool. Within this tool, the group by field was specified as "FIRE_YEAR", so that percent area per county per year was recorded. This tool generated a feature class containing all US counties, and a join table which held percent area per year by county. Counties with no percent area for a given year were not reported within the join table. Sum within trials and associated join tables for the contiguous US, AK, HI, and PR were exported as csv files and brought into Excel for further analysis.

Drought data was pulled from US Drought Monitor's web database using R version 3.6.0. From a US Census Bureau county FIPS worksheet, a list of unique FIPS codes for each county in the United States was created as an R object. The FIPS codes were used to loop through the API pull-request URL so that drought severity statistics by percent area could be pulled between 1/1/2000 and 12/31/2020 for all US counties. Statistics type 1 is the measure of interest. The output data frames were stacked together to produce one large file to cleanse. Because the data came in a weekly format, the end-date of the week was used to determine the corresponding year. An average of all week data per county, per year became the average percent area of drought metric. We chose to only include D2 drought intensity and greater (D3 and D4). 

D0 and D1 (abnormally dry and moderate drought) intensity values were excluded because lower classifications seem to be more widespread and variable from year to year. We were most interested in areas exposed to persistent drought. Given this, it was most appropriate to start with D2 (severe drought) as our lowest intensity level. Please refer to the data source link for more information on drought level classification.

Coastal flood data was downloaded as a geodatabase, which contained one raster file per state. These raster files, in addition to a 2018 US Census Bureau's county boundaries shapefile, were brought into ArcGIS Pro version 2.6.3. Each raster was converted to a polygon feature class, and subsequently merged into one. A definition query was used to limit coastal flood features to just those of interest - possible coastal flood areas with a hazard value or "gridcode" greater than or equal to 1. To do so, the query was written as "gridcode >= 1".

To calculate percent area of county possibly exposed to coastal flooding, this merged feature class was summed to the US county shapefile via the sum within tool. Please note that for the contiguous US, this county shapefile was split up into smaller subsets before running sum within trials. Each sum within trial generated one feature class containing a number of US counties and associated percent area. Sum within trials for the contiguous US were stitched back together before exporting. Unlike previous hazards, the group by field was not necessary because only one year or data release was provided. Sum within trials for the contiguous US, HI, and PR were exported as csv files and brought into Excel for further analysis.

As mentioned previously, coastal flood data was not available for AK. Instead of omitting AK entirely, a subset of inland flood data was used to replicate one of the layers used to create the coastal flood composite layer. Inland flood data was clipped to AK, limited to only those areas of moderate (0.2% annual chance) to high risk (1% annual chance) of flooding and dissolved to a single feature before running the sum within tool.

The "gridcode" field represents the number of flood hazards that may occur at that specific pixel (for raster data) or feature's location. It was not made clear why NOAA reported on locations with a "gridcode" of 0, and how these areas would differ from those outside the bounds of these raster files. For this reason, 0 values were excluded from further analysis.

For every other hazard, the processing methods mention a need for the dissolve tool. Such was not necessary here. Event based hazards required a single feature per year. For example, inland flooding was dissolved to a single feature since only one year or data release was available. Inland flooding, like all other vector-derived hazards, was represented by millions of intricate polygon features, where it was possible and likely that many features overlapped several other features. This issue of overlapping features causes percent area values per county to balloon and over-represent the hazard. Coastal flooding was derived from raster data, so unlike vector data, overlapping features was a non-issue.

NOAA's Coastal Flood Exposure Mapper displays the source data provided to me by the NOAA affiliate mentioned within this hazard's data source section of this document. Within the mapper, this data is called the "coastal flood hazard composite layer". An information tab is available for this layer, which gives a brief summary of the five hazards used to create the composite layer. One of these hazards was a subset of the same source data used for inland flooding and could easily be replicated. We felt it was important to represent AK if possible and this was the best available data to do so.

Inland flood data was downloaded as a geodatabase, which contained 8 feature classes and 2 tables. The "S_Fld_Haz_Ar" feature class, in addition to a 2018 US Census Bureau's county boundaries shapefile, were brought into ArcGIS Pro version 2.6.3. A definition query was used to limit inland flood features to just those of interest - all designated flood zones, except for those defined as "Area not included", "D", or "Open Water". To do so, the query was written as "FLD_ZONE <> 'AREA NOT INCLUDED' Or FLD_ZONE <> 'D' Or FLD_ZONE <> 'OPEN WATER'". Once limited, a copy of this feature class was created.

Before calculating percent area, the check geometry and repair geometry tool was executed. After which, the newly created feature class was clipped to each region and then dissolved into a single feature.

To calculate percent area of county possibly exposed to inland flooding, each region's clipped and dissolved output was summed to the region's county shapefile via the sum within tool. Please note that for the contiguous US, this county shapefile was split up into smaller subsets before running sum within trials. Each sum within trial generated one feature class containing a number of US counties and associated percent area. Sum within trials for the contiguous US were stitched back together before exporting. Unlike previous hazards, the group by field was not necessary because only one year or data release was provided. Sum within trials for the contiguous US, AK, HI, and PR were exported as csv files and brought into Excel for further analysis.

From the 8 feature classes provided, the "S_Fld_Haz_Ar" feature class was selected for further analysis because it was the only polygon data with information pertaining to flood zone type. These zone types give context to how often an area might experience some degree of inland flooding. This information was necessary for the definition query mentioned above. The flood zone types, "Area not included", "D", or "Open Water", were excluded because no likelihood of exposure to inland flooding is associated with these values. "D" is said to represent an area of undetermined flood risk. 

The check and repair geometry tools were necessary because invalid topology error messages were thrown when clipping to each region. The main issue was self-intersection of features. After executing these, the aforementioned feature class was clipped and dissolved to each region to reduce processing time when using the sum within tool. 

Earthquake data was downloaded in shapefile format. Two shapefiles, one for the central and another for the eastern US, were provided for each one-year short-term induced seismicity model. Such models were only available for 2016, 2017, and 2018. These shapefiles, in addition to a 2018 US Census Bureau's county boundaries shapefile, were brought into ArcGIS Pro version 2.6.3. Shapefiles for each year were merged. After which, a definition query was used to limit chance of damage from an earthquake to just those of interest - chance of damage greater than or equal to 1%. To do so, the query was written as "ValueRange <> '<1'". Once limited, a copy of each shapefile was created.

No buffer was necessary since features were presented as polygons. These polygons represented probability and not events, so a dissolve was not necessary either. 

To calculate percent area of county possibly exposed to earthquakes, each merged feature class was summed to the US county shapefile via the sum within tool. Each sum within trial generated one feature class containing all contiguous US counties and associated percent area. All sum within trials for the contiguous US were stitched back together before exporting. Unlike previous hazards, the group by field was not necessary because one sum within trial was run for each year. The merged sum within trial for just the contiguous US was exported as csv file and brought into Excel for further analysis.

Sum within trials for AK, HI, and PR were not conducted because no data was available for these regions. The data provided was probabilistic, and therefore could not be replicated with any degree of confidence. Long-term seismicity data was available for a number of years for these regions but was deemed not useful since no spatial data was offered.

Processing outside of ArcGIS Pro:

Exports for sum within trials and associated join tables, per each region for each hazard, were joined in Excel version 2102 via the "Join_ID" column. Once all regions were joined per each hazard, all counties with a value for percent area were stacked together. 

Each hazard matrix was added to a new workbook where zero filling could be completed. Within this, counties which were not included within associated join tables were added back in and assigned a value of 0. This was done on a per hazard basis, for each county and for each year within the hazard's spatial and temporal extent. For example, every county within each year from 2000-2018 was represented, regardless of its percent area value, for tornadoes. The year 2019 was not included for tornadoes because no data was provided for that year.

Once the previous step was executed for each hazard, one sheet containing percent area for all counties for each year from 2000-2019 was created within a separate workbook. Each hazard was joined to this final matrix via a concatenated column containing "GEOID" and "YEAR". N/A values were replaced with -999, since N/A represented no data provided for a county within a given year.

Zero filling was deemed appropriate because it represents counties with no exposure, which are arguably just as important as those counties with some degree of exposure. The best available data sources were used for each hazard, but that isn't to say that each source doesn't have its limitations and possible misreporting of events. Given this, zero exposure in terms of percent area does not mean such counties contain zero risk or likelihood of exposure. In the context of this dataset, zero percent area simply translates to no occurrence of a defined hazard event within that county for that year.

N/A values were replaced with -999 so that counties with no data could be easily identified.

Importing final matrix back into ArcGIS Pro:

The final data matrix was then joined to a copy of the US Census Bureau's 2018 county boundary shapefile in ArcGIS Pro version 2.6.3 via the "GEOID" field. Please note that this copy contained the US Virgin Islands 3 main islands; these were removed after the join was completed. A total of 64,400 records remained (3,220 counties multiplied by 20 years). Additionally, counties with a percent area value greater than 100% were capped at 100%. Such values only exceeded 100% by less than 1%.

This join was necessary to create a spatial data set. Despite its current spatial context, this feature class can be easily exported as a table for use in other applications. 

The presence of values exceeding 100% by less than 1% were deemed an artifact of the sum within tool in ArcGIS Pro. Because of this, capping at 100% seemed appropriate.

Hurricane and Tropical Storm Data Source: International Best Track Archive for Climate Stewardship (IBTrACS) 
Data Source Link: https://www.ncdc.noaa.gov/ibtracs/index.php?name=ib-v4-access 
*Data updated weekly*

Downloaded "IBTrACS.since1980.list.v04r00.lines.zip" on 03/18/2021. This zip file contains a polyline shapefile, which houses all recorded hurricane tracks (both preliminary and final tracks) worldwide from 1980 to current. A technical document and column documentation pdf relating to this "version 4" data download are available as well. To view these, please visit the data source link listed above.

Tornado Data Source: NOAA's Storm Prediction Center (SPC)
Data Source Link: https://www.spc.noaa.gov/gis/svrgis/
*Data updated at unknown frequency. At time of download, data was last updated 10/01/2019*

Downloaded "1950-2018-torn-aspath.zip" on 09/02/2020. This zip file contains a polyline shapefile, which houses all recorded tornado paths within the US (including US territories) from 1950 to 2018. A database description document and changelog are available as well. To view these, please visit the data source link listed above. 

Landslide Data Source: USGS's Landslide Inventories across the United States
Data Source Link: https://www.sciencebase.gov/catalog/item/5c7065b4e4b0fe48cb43fbd7
*Data updated annually. At time of download, data was last updated 03/21/2019 (even though publication date is listed as 09/19/2019)*

Downloaded "US_Landslide_1.zip" on 09/30/2020. This zip file contains a polygon shapefile, which houses all recorded landslide incidents within the US (including US territories) from 1900 to 2019. A metadata document provided as an xml is available as well. To view this document, please visit the data source link listed above.

Wildfire Data Source: National Interagency Fire Center, Interagency Fire Perimeter History - All Years
Data Source Link: https://data-nifc.opendata.arcgis.com/datasets/interagency-fire-perimeter-history-all-years/explore?location=43.578757%2C63.134208%2C3.57
*Data updated at unknown frequency. At time of download, data was last updated 06/12/2020*

Downloaded "Interagency_Fire_Perimeter_History_-_All_Years.zip" on 09/08/2020. This zip file contains a polygon shapefile, which houses all recorded wildfire incidents within the US (including US territories) from 1835 to 2019. An attributes pdf document is available as well. To view this document, please visit the data source link listed above.

Drought Data Source: U.S. Drought Monitor
Data Source Link: https://droughtmonitor.unl.edu/DmData/DataDownload.aspx
*Data updated weekly*

Drought severity statistics for percent area per county for D2 drought severity and greater from 1/1/2000-12/31/2020 was pulled via API and brought into R for further processing. A metadata tab is available at the data source listed above.

Coastal Flooding Data Source: Coastal Flood Exposure Mapper, Coastal Flood Hazard Composite Layer
Data Source Link: https://coast.noaa.gov/floodexposure/#-10575352,4439107,5z
*Data updated at unknown frequency*

Downloaded "OCM_CoastalFloodHazardComposite_FY19" from a shared google drive with NOAA affiliate on 03/23/2021. This zip file contains a geodatabase. Within this, raster files for coastal flooding are provided by state. Please note data for AK is not included, as the contact stated that no data currently exists. Also, this data is not event based - it only illustrates the most up to date representation of possible exposure to coastal flooding. This contact mentioned that although a variety of data sources were used to create this composite layer, each with a different production date, that the composite layer was created at the end of the 2019 calendar year. No metadata is available for this source.

Inland Flooding Data Source: National Flood Hazard Layer (NFHL), Seamless Nationwide NFHL GIS data
Data Source Link: https://catalog.data.gov/dataset/national-flood-hazard-layer-nfhl
*Data updated monthly*

Downloaded "NFHL_Key_Layers.gdb" from the data source link on 02/09/2021. This zip file contains a geodatabase. Within this, various polygon shapefiles, which represent different aspects of inland flooding, are provided. Please note, this data is not event based - it only illustrates the most up to date representation of possible exposure to inland flooding. Two metadata documents and a database technical reference are available as well. To view these, please visit the data source link listed above. Within these metadata documents, no "last update" date pertaining to the data itself is specified. A process step within these documents mentions that the resource is updated on a monthly basis. Since these documents were last updated 11/2020, we felt it was most appropriate to state that the data was representative of 2019. We did not feel confident that inland flood data or FIRM panels for 2020 were included within this set of data.

Earthquakes Data Source: Short-term Induced Seismicity Models
Data Source Link: https://www.usgs.gov/natural-hazards/earthquake-hazards/science/short-term-induced-seismicity-models?qt-science_center_objects=0#qt-science_center_objects
*Model updated at unknown frequency. At time of download, most recent one-year seismic hazard forecast was released in 2018*

Downloaded two zip files, one for each half of the contiguous US, for each year (2016, 2017, and 2018) from the data source link on 12/31/2020. Each zip file contains a singular shapefile, which illustrates percent chance of damage from an earthquake. Please note, this data is not event based - each set of shapefiles represent a one-year probabilistic seismic hazard forecast for the central and eastern US from induced and natural earthquakes. 

2018 downloads: Please note that data illustrating minor-damage, instead of moderate damage, was downloaded because the source referenced this data when stating "For consistency, the updated 2018 forecast is developed using the same probabilistic seismicity-based methodology as applied in the two previous forecasts."
1. "Chance of potentially minor-damage ground shaking in 2018 based on the average of horizontal spectral response acceleration for 1.0-second period and peak ground acceleration for the Central and Eastern United States
2. "Chance of potentially minor-damage ground shaking in 2018 based on the average of horizontal spectral response acceleration for 1.0-second period and peak ground acceleration for the Western United States

2017 downloads:
1. "Chance of damage from an earthquake in 2017 based on the average of horizontal spectral response acceleration for 1.0-second period and peak ground acceleration for the Central and Eastern United States"
2. "Chance of damage from an earthquake in 2017 based on the average of horizontal spectral response acceleration for 1.0-second period and peak ground acceleration for the Western United States"

2016 downloads: 
1. "Chance of damage from an earthquake in 2016 based on the average of horizontal spectral response acceleration for 1.0-second period and peak ground acceleration for the Central and Eastern United States"
2. "Chance of damage from an earthquake in 2016 based on the average of horizontal spectral response acceleration for 1.0-second period and peak ground acceleration for the Western United States"

Natural Hazard Histories Dataset

Point of Contact