Observations of Flash Floods
What the records map shows, where the numbers come from, and how to read them when weighing where new weather radars would matter most.
Every point on this site traces back to an official flash flood report filed by a local National Weather Service office. We collected eleven years of these reports, 2015 through 2025, for the lower 48 states and DC, kept only floods caused by weather, and added them up two ways: by county, which is how the reports are filed, and by HUC8 watershed, which is how water actually moves. The result is a picture of where flash flooding keeps happening, who it hurts, and what it costs.
Radar coverage is not uniform across the country. In places where the nearest radar beam sits high above the ground, rainfall estimates are less certain and warnings lean on less information. A map of repeated flash flood impacts is one of the inputs a decision maker needs when asking where an additional radar, or a gap filling radar, would pay for itself in warnings, property, and lives.
One whole storm system: a tropical storm, a line of severe thunderstorms, a slow frontal system. The NWS gives it a single episode number, and it can span several days and many counties. Counting episodes answers: how many distinct storms caused flash flooding here?
One flash flood report for one place, filed as part of an episode. A single episode often produces several events. Counting events answers: how many separate flash floods were recorded here? Fatalities, injuries, and damage are attached to events, so we sum them from the event records.
Example: a squall line crosses a metro area overnight. That is one episode. It drops flash floods in four counties, two of them twice. That is six events. The county view credits each county with its own events; all of them share the same episode.
The Storm Events Database records a cause for each flash flood. We keep the causes that come from weather, at any time of year: heavy rain (42,256 events), heavy rain with a tropical system (1,519), heavy rain on a burn scar (604), heavy rain with snowmelt (155), and ice jams (38). We removed 94 reports whose cause was a dam or levee failure, because those floods say nothing about where better rainfall observation would help.
Counties are the reporting unit of record, so the county view is the most faithful to the database and the most familiar to local officials. Watersheds are the unit floods obey: rain that falls in one county can flood the next one downstream. The HUC8 view assigns each event to the watershed containing the reported start point of the flooding, using the USGS Watershed Boundary Dataset. HUC8 basins are typically a few thousand square kilometers, the scale of a large river basin section.
| Item | Detail |
|---|---|
| Source | NOAA NCEI Storm Events Database, annual details files, 2015 through 2025 |
| Selection | EVENT_TYPE equals Flash Flood, county-type reports, lower 48 states and DC |
| Cause filter | Dam and levee failures removed (94 reports); all weather causes kept |
| Episodes and events | Counted from EPISODE_ID and EVENT_ID; episode counts are distinct storms touching that county or watershed |
| Fatalities and injuries | Direct plus indirect, summed from event records |
| Damage | Property and crop damage parsed from NWS entries such as 10.00K or 2.5M into dollars |
| County geometry | NWS AWIPS county shapefile, dissolved to one shape per county FIPS |
| Watershed geometry | USGS Watershed Boundary Dataset, HUC8, CONUS |
| Result | 44,572 events in 12,184 episodes; 1,044 fatalities; 291 injuries; $61.2B property and $0.8B crop damage |
The scripts that rebuild every number on the records map from the
raw NCEI files are in engine/records of the RUNOFF repository.
Damage figures are estimates made by forecasters shortly after each event, not audited losses, and large events dominate the totals. Injuries from flash floods are known to be under-reported. One storm episode can touch several counties, so summing episode counts across counties would overcount distinct storms; the totals panel on the map counts each episode once. Reporting practices also vary somewhat between offices and improve over time, so treat small differences between neighboring places as noise and large, repeated patterns as signal.
Episode and event wording follows the NCEI Storm Events bulk CSV format documentation: an episode is an entire storm system and can contain many different types of events; an event is an individual type of storm event within an episode. See the Storm Events Database for the database itself and its documentation page for the format description.