Storm Blackout and Restoration

All datasets · API reference · Dataset page on heimdallresearch.com

How long each county stayed dark after FEMA-declared hurricanes, tropical storms and winter storms in the United States, Puerto Rico and the U.S. Virgin Islands since 2012, and how quickly its night light returned, measured from NASA Black Marble daily satellite light readings.

Status: In preparation. Version 1.0 is estimated for October 2026, with storms from 2012 to 2017 first.

Themes	weather, energy
Coverage	FEMA-declared hurricanes, tropical storms and major winter storms; contiguous U.S., Puerto Rico, U.S. Virgin Islands
Period	2012 to present
Resolution	County by day
Free updates	Each new storm, four to six weeks after landfall
Paid updates	Within days of landfall, for a named storm or every storm in your region
License	CC-BY-4.0
Free	Every declared storm at county level, as files and through the API and MCP server with a free key.
Paid	A report for a named storm and area within days of landfall, or a regional subscription. See plans and prices.
Intended use	Comparing how quickly places recovered after major storms, studying which communities waited longest, and giving resilience, insurance and emergency-management research a consistent record across storms and years.
Not intended for	Counting customers without power, judging a single address or neighborhood, real-time response, or insurance claims on a single property.

Tables

Table	One row per	Fields
storms	event_id	6
county_summary	event_id, geoid	15
county_daily	event_id, geoid, date	8

Access

API: GET /v1/storms, GET /v1/storms/{event_id}, GET /v1/storms/{event_id}/counties/{fips}, GET /v1/storms/{event_id}/files. See the API reference.

MCP tools: get_storm, get_download_link. See Use it in your AI assistant.

Files: CSV and Parquet on Zenodo and Hugging Face, with datapackage.json, a Croissant file and SHA256SUMS, from the first release.

How values are produced

Each county in a storm’s declared area is compared night by night with its own normal light. The baseline is the median of clear, snow-free nights before the storm, leaving out nights affected by an earlier storm; when too few remain, the same weeks one year earlier are used and the row says so. After landfall each clear night is compared with the baseline, and recovery is counted from when light stays at or above half and nine-tenths of normal for three observed nights. Observation continues for up to 180 days.

Pixels flagged for cloud, poor quality or snow are excluded, and the number of usable nights travels with every value. Results are never reported below county level.

Table storms

File storms.csv. One row per storm event.

One row per event_id.

Field	Type	Unit	Description
event_id	string		NHC storm identifier or FEMA declaration number.
name	string		Storm name or declaration title.
kind	string		tropical, winter_storm, severe_ice_storm or snowstorm.
landfall	date		Landfall or incident start date.
states	string		Declared states, space separated. May be empty.
source	string		HURDAT2, NHC-ATCF-operational or OpenFEMA.

Table county-summary

File county_summary.csv. One row per storm and county.

One row per event_id, geoid.

Field	Type	Unit	Description
event_id	string		Storm event.
geoid	string		County FIPS code.
county	string		County name.
state	string		State or territory.
baseline_radiance	number	nW/cm²/sr	Median pre-storm radiance.
baseline_nights	integer		Clear nights in the baseline.
baseline_source	string		event_window or prior_year.
post_obs_nights	integer		Clear nights observed after landfall.
median_valid_pixel_share	number		Median share of usable pixels after landfall.
days_to_50	integer	days	Days until light stayed at or above 50% of baseline. May be empty.
days_to_90	integer	days	Days until light stayed at or above 90% of baseline. May be empty.
censored	boolean		True if 90% recovery was not reached in the window.
censored_50	boolean		True if 50% recovery was not reached in the window.
min_light_ratio	number		Lowest light ratio after landfall.
post_days_window	integer	days	Length of the observation window.

Table county-daily

File county_daily.csv. One row per storm, county and clear night.

One row per event_id, geoid, date.

Field	Type	Unit	Description
event_id	string		Storm event.
date	date		Night of observation.
geoid	string		County FIPS code.
days_from_landfall	integer	days	Days since landfall (negative before).
radiance_nw_cm2_sr	number	nW/cm²/sr	County median radiance that night; empty when no usable pixels. May be empty.
valid_pixel_share	number		Share of usable pixels.
snow_excluded_share	number		Share of pixels excluded for snow.
light_ratio	number		Radiance divided by the county baseline; empty when not observed. May be empty.

Limits

• Night light is related to electricity service but is not the same thing. Generators can restore light before the grid does, and damaged or empty buildings can stay dark after power returns.
• Clouds and snow hide the ground. Counties with few clear nights carry a low observation count and a quality flag.
• Moonlight, seasonal lighting and changes in how people use light shift readings. The product is corrected for moonlight, and each value is compared with the county’s own baseline.
• Days to recovery are left blank and marked censored when recovery did not happen within the observation window.

Sources

Source	Role	Terms
NASA Black Marble VNP46A2 v2 (VIIRS on Suomi NPP), DOI 10.5067/VIIRS/VNP46A2.002	Measured night light, 2012 onward	NASA open data; acknowledge and cite
NASA Black Marble VJ146A2 v2 (VIIRS on NOAA-20)	Measured night light, 2018 onward	NASA open data; acknowledge and cite
NOAA National Hurricane Center HURDAT2 and operational best tracks	Storm timing and landfall	Public domain
FEMA disaster declarations (OpenFEMA)	Which storms and counties are included	Public domain
U.S. Census Bureau cartographic boundary files	County boundaries	Public domain
EAGLE-I county outage records, Oak Ridge National Laboratory (DOI 10.6084/m9.figshare.24237376)	Validation only	CC BY 4.0

Required notices:

• Night light data: NASA Black Marble (VNP46A2, VJ146A2), courtesy of NASA LAADS DAAC.

All datasets · API reference