Flood mapping

Basemap gallery

Flood classes

Flood water — new since baseline Water (post scene)

Cloud / no data

How it works

Study area — draw a rectangle or polygon with the map toolbar (max 1,000 km²; smaller areas run faster).

Scenes — set a baseline (pre-flood) and a flood window, then scan the public Sentinel-2 archive (AWS Open Data). Scenes are ranked by scene-level clear-sky share.

Run — your browser downloads the six needed bands, runs the WorldFloods v2 U-Net locally (WebGPU when available, WASM otherwise), and maps flood = water on pixels that were land at baseline.

Everything runs in your browser — no server, no account, no data leaves your machine. A run downloads roughly 50–120 MB of imagery and takes ~1–5 minutes depending on area size and hardware. Move the threshold slider after a run to re-threshold instantly. Clouds are blind spots for optical imagery; prefer clear scenes. Note: the WorldFloods model was trained on L1C top-of-atmosphere data and is applied here to L2A surface reflectance — results are close but validate before operational use.

mmflood (SAR) uses Sentinel-1 — no cloud dependency at all, works in any weather. Calibration (raw radar signal → physical backscatter) is computed entirely in your browser from public ground-control points, an approximation of the full terrain-correction pipeline a server would normally do. Water detection itself matches server-based processing closely (validated within ~1% on a real scene); flood-extent numbers specifically (which also depend on the pre-flood baseline) are more approximate as a result, and the model is validated as weaker on small or emerging floods — treat results as indicative, not precise, especially at low flood percentages.