Google Research Flood Forecasting Grounding

AI River Flood Forecasting & Inundation Workbench

Simulate watershed rainfall runoff, AI hydrograph routing in ungauged river basins, and urban floodplain inundation. Test emergency flood defenses and calculate forecast lead time before river crest arrival.

T +42.0 h
River Stage Depth
7.42 m
Bankfull: 6.50 m (+0.92m)
Peak Discharge Q
3,480 m³/s
Peak at T +54h
Forecast Lead Time
36.0 hrs
Actionable evacuation window
Warning Category
WARNING
Inundation: 42.6 km²
Dynamic River Reach & Inundation Map
River Current
Overbank Inundation
Warning Threshold (6.5m)
Advisory Level (5.0m)
Flood Defense Levee
Streamflow Hydrograph Q(t) & Rainfall Hyetograph P(t) 7-Day Continuous Forecast
Rainfall Hyetograph (mm/h)
AI Simulated Streamflow (m³/s)
Severe Flood Threshold
Current Simulation Time marker
Simulation ready. Drag scrubber or hit Simulate to run streamflow routing.

How AI Predicts Floods in Ungauged Basins

Traditional hydrological models rely on physical calibrations against decadal stream gauge records. However, over 95% of river reaches globally are completely ungauged, leaving vulnerable communities without early warning systems.

The Google Flood Forecasting Initiative overcomes this by training deep Long Short-Term Memory (LSTM) networks on combined global static geophysical attributes (catchment drainage area, slope, soil composition, lithology) and dynamic ERA5 reanalysis and IMERG satellite precipitation.

Effective Runoff Q = P_excess · Catchment_Area / T_conc
Bankfull Crest Depth h(t) = (Q(t) / (B · (1/n) · S^(1/2)))^(3/5)

By routing upstream runoff down the river network through a digitized Digital Elevation Model (DEM), hydraulic inundation extent can be predicted days in advance.

Operational Workflow & Decision Matrix

Forecast Lead Time vs. Evacuation Windows

A forecast is only valuable if the lead time (time between alert issuance and flood threshold breach) exceeds community evacuation and temporary barrier setup windows. This simulator computes the actionable warning lead time in real time.

Levee Overtopping Mechanics

Once stage exceeds natural riverbanks or artificial levee crests, flow velocity drops while inundation footprint scales exponentially across flat floodplains. Activating the temporary levee parameter raises the retention crest by 1.5 meters.

Hydrological Model Variations

AI LSTM: Superior non-linear saturation threshold mapping and zero-shot transfer.
GR4J: Four-parameter lumped production and routing reservoirs.
Saint-Venant: 1D kinematic wave shallow water momentum equations.

Enjoy this tool? Build your own with Super