Riverine Early Warning Simulation
AI River Flood Forecasting & Inundation Workbench
Inspired by the Google Flood Forecasting Initiative with scientist Deborah Cohen. Model how satellite precipitation, antecedent soil moisture, and AI streamflow transfer predict crest timing, ungauged river discharge ($m^3/s$), and downstream floodplain inundation.
River Discharge ($Q$)
1,480 m³/s
118% of Bankfull Cap
Peak Crest ($Q_{max}$)
1,895 m³/s
Expected at t = +32 hrs
Inundated Floodplain
42.4 km²
Peak Overbank: +1.85 m
AI Forecast Lead Time
34 hrs
94% Confidence (Ensemble)
RIVER REACH & INUNDATION MAP
Riverbed
Inundated Area
Settlement Risk
Click or drag along the river reach to inspect localized cross-sections and populated zones.
72-HOUR AI STREAMFLOW HYDROGRAPH
Hydrograph
Bankfull Threshold
Solid teal curve displays routed streamflow; shaded blue bars show rainfall hyetograph ($mm/h$).
How AI Solves the Global Ungauged Basin Problem
Traditional river forecasting relied on physical stream gauges installed inside river channels. In developing regions and remote watersheds, gauges are sparse or nonexistent. Google's Flood Forecasting Initiative, co-led by Deborah Cohen, leveraged deep learning (Long Short-Term Memory networks) trained on global hydrological features to predict flood risk even where zero physical stream gauges exist.