HISTORICAL EVENT INTELLIGENCE · FLOOD
Flood Data
108,000+ global flood records back to 1947. US records carry USGS gauge peaks, road closures and storm reports; non-US records carry modeled population exposure. Licence the archive, or stream floods live by API.
What's in the dataset
The peril where naming the evidence tier matters most.
Stream gauges
Peak stage, peak time, and flood stage per gauge — a single Appalachian event in our archive carries 44 of them
Road closures
Location, status, and the narrative remark from the reporting office
Storm reports
Ground reports tied to the event, with time, place, and text
Flood class
Flash, areal, riverine, or coastal — the distinction that changes who is affected and how fast
Alert-zone geometry
The issuing office's polygon (not an inundation extent — see the FAQ)
Population & economic exposure
Modeled exposure inside the footprint, on roughly 40% of the archive
Warning, or water on the ground?
The question every flood dataset should answer about itself — so here is ours.
United States — 67,081 records
These begin as National Weather Service products — flash flood, areal, riverine and coastal warnings and watches — and are then enriched with what was actually measured: stream gauges with peak stage and peak time, road closures, and storm reports. The alert is the container; the gauge readings are observations. 95% of US records carry this enrichment, and because the sources are NWS, so does 98% of all flood enrichment in the archive.
International — ~41,000 records
Outside the US, records more often describe flooding already reported — fatalities, evacuations, damage — drawn from official and media reporting. There is no international equivalent of the US gauge and storm-report feeds, so these records do not carry that enrichment. About 41% do carry a modeled exposure assessment — population and economic value inside the alert area, which is an estimate of what was in the way, not a measure of observed impact.
One caution on that split: the US share reflects how densely flood alerts are issued and ingested, not where flooding happens. Read it as a map of alerting coverage, not of global flood risk. Counts are as of September 2026.
What we exclude: global flood model predictions are forecasts, not events, and our event pipeline filters them out rather than counting them. The distinction between an issued warning, a report, and a footprint is the whole discipline here, and why observed flood intelligence reads differently from a warning feed.
Flood data by place
Risk scores, current alerts, and event history for locations worldwide.
Watching something right now? Live flood activity shows what the feed is tracking this hour, and Flood Intelligence covers the detection-to-decision pipeline.
Two ways to get it
The same flood records, served for history or for right now.
Historical datasets
Bulk extracts filtered to your geography, flood class, and date range — CSV, GeoJSON, or NDJSON, by subscription or one-time purchase.
Historical Event Intelligence →Real-time flood API
Flood warnings and events as they are issued, with footprints and exposure — the live end of the same record structure.
Event Intelligence API →Flood Data — FAQ
Where can I download historical flood data?
Historical flood datasets are licensed through DisasterAWARE Historical Event Intelligence, filtered to your geography and date range and delivered as CSV, GeoJSON, or NDJSON. Free sample datasets across other perils are available with no signup, and we can stage a flood extract scoped to your area of interest on request.
Is this observed flood data or flood warnings?
Both, and we label which is which. US records — 62% of the archive — originate as National Weather Service flood products, then are enriched with observed data: USGS stream-gauge peak stage and peak time, road closures with narrative remarks, and storm reports. International records more often describe flooding that has already been reported, with impact narrative and population exposure attached. We do not present a warning as an observation, and we do not count model predictions as events at all.
What does the flood dataset contain?
As of September 2026, 108,000+ flood records with contiguous coverage from 1947 across 207 countries. Flood enrichment follows National Weather Service coverage: 63,900 of the 64,900 enriched records are US, so it is best understood as a US capability. Those records carry the issuing office and product type, flood class (flash, areal, riverine, coastal), stream gauges with peak stage and flood stage, road closures, storm reports, and alert-zone geometry. About 41% of non-US flood records carry a modeled exposure assessment — population and economic value inside the alert area, which is not a measure of observed impact.
Does the data include flood depth or inundation extent polygons?
Not yet. Records carry a depth reference indicating the modeling method where one applies (NWS CatFIM), but the inundation surface itself is not attached, and the geometry on a record is the alert zone rather than a mapped flood extent. Depth surfaces derived from gauge observations and CatFIM are on our roadmap; we would rather tell you what the data is today than imply a layer that is not there.
Is there a flood data API?
Yes — two. The Event Intelligence API streams flood warnings and events in real time with footprints and exposure, and Historical Event Intelligence serves the archive by query: by country, state, county, radius around a point, flood class, date range, or exposed population.