More time before the flood.

The human problem

Many communities lack the observations needed for reliable flood forecasts.

The prior constraint

Sparse measurements limited forecasting in ungauged watersheds.

AI’s actual role

A model learned from hydrological data across watersheds.

The documented result

At up to five days ahead, reliability matched or exceeded same-day forecasts from the comparison system for extreme riverine events.

Why it may matter

Earlier information could help communities prepare; the study did not measure lives saved.

Limitations

Forecast skill varies by basin. Warnings still require local delivery, trust, and capacity to act.

Unresolved questions

Evaluate forecast performance and warning access in more communities.