Inspiration
After a flood or debris-flow event in mountainous terrain, responders need answers quickly, yet several obstacles stand in the way: • Satellite products are hard to act on. A flood mask is a raster; decision-makers need to know what it means for people, roads and settlements • Ground access is uncertain. Damaged roads and bridges can isolate communities, and field reports arrive slowly. • Optical imagery is unreliable in monsoon conditions. Cloud cover limits what can be observed, so radar must carry much of the analysis. • Steep terrain produces false positives. Radar shadow, layover, hillsides and wet snow are easily mistaken for water. • Coarse resolution cannot confirm structural damage. At 10 m, a mask can indicate exposure but not whether a bridge or building survived.
This project closes these gaps by linking hazard mapping directly to infrastructure exposure and road connectivity, while stating clearly what the data can and cannot support
What it does
Trishuli Flood Response is an end-to-end geospatial pipeline that combines Sentinel-1 radar, Sentinel-2 optical imagery, the Copernicus GLO-30 DEM and pre-event OpenStreetMap (OSM) data. It maps flood hazard and translates it into operational questions: which buildings, roads and bridges overlap the hazard, and which settlements may have lost access to hospitals and towns
How we built it
Hazard mapping A classical baseline (radar change, optical information where usable, terrain filtering) and a trainable flood segmentation model. A separate debris/scour mask is produced alongside the water mask. Exposure analysis Intersects the hazard with pre-event OSM buildings, roads and bridges to estimate exposure
