Wami–Ruvu ensemble flood forecast

The problem

The Wami and Ruvu basins drain into the Dar es Salaam region, Tanzania’s largest urban area. Flood warnings there need to arrive days ahead and come with an honest measure of uncertainty, so responders can decide whether to act now or wait.

Approach

• A distributed Wflow hydrological model of both basins, built with HydroMT from global open datasets.
• A historical run (IMERG + ERA5) gives model-consistent return-period thresholds.
• Each day the model runs on all ECMWF ensemble members. The alert level on each reach is the share of members exceeding the 2-, 5- and 20-year flows, as GloFAS does.
• Outputs follow Delft-FEWS PI / CF-NetCDF conventions, so the chain can be dropped into an operational FEWS system.
• Everything runs on GitHub Actions and publishes to a static site. There is no server to maintain.

Result

A public forecast viewer: map of reaches coloured by alert level, ensemble hydrographs per reach, and a ranked alert table.

Current status: the viewer is live and currently shows synthetic demonstration data. The modelling pipeline is written and will replace it with daily operational forecasts once the model build and historical calibration runs are complete.

Next step: forecast-based financing

Add a trigger layer that turns forecast probabilities into pre-agreed anticipatory actions (“release funds when P(> 5-yr flow) ≥ 60% at 3–5 days lead time”), and evaluate its hit and false-alarm rates on past events.

Limitations

Experimental, not an official warning service. Thresholds are model-derived rather than calibrated against observed impacts.