Opening photo: Jörg Blobelt, Dresden 2002 · CC BY-SA 4.0

From cloudburst to decision

One storm. Four layers of analysis. One decision.

Follow a single storm through the twin of Deurne. The maps below are real layers from the pilot, not mock-ups.

  1. 01The storm

    A cloudburst outruns the sewer. In minutes.

    Input
    A design cloudburst (here 59 mm in 3 h) with optional climate uplift, or an event observed by a VMM rain gauge.
    Why it matters
    Short, intense storms overwhelm drainage before any river rises. The damage is decided street by street.

    In FloodTwinEvery run stores its storm, so anyone can see what was assumed.

  2. 02The terrain

    Water follows the ground. To the centimetre.

    Layer
    Hydrologically conditioned DTM in Belgian Lambert 72, heights in Ostend datum (m TAW).
    Derived
    Slope, flow accumulation, wetness index and height above drainage, labelled as inference, never as depth.

    In FloodTwinEvery raster carries its source, licence and SHA-256 fingerprint.

  3. 03The physics

    The storm, solved. Not guessed.

    Solver
    Deltares SFINCS v2.5.0, a 2D shallow-water model, compiled from source. Depths downscaled to 1 m.
    Checked
    Against the official VMM pluvial map: CSI 0.36. Published with its limits, inside the product too.

    In FloodTwinNo AI invents a depth. Every number comes from the physics run.

  4. 04The exposure

    Every doorstep, counted. Not a colour on a map.

    Method
    Each GRB footprint is sampled along its facade against 5, 15 and 30 cm thresholds; roads by length.
    Withheld
    Damage in euros stays off until validated Flemish depth-damage curves exist.

    In FloodTwinExposure exports as GeoPackage and COG, in the Belgian grid.

  5. 05The decision

    Test the measure first. Then build it.

    Counterfactual
    Draw a basin, berm or green roof; it runs against the same storm and returns a depth-difference (Δh) map.
    Honest result
    A Rivierenhof basin alone lowers p95 depth by 0.13 m and clears 281 m of road, but protects no building.

    In FloodTwinKnowing that before the excavator arrives is the point.

Evidence · Antwerp-Deurne pilot

Numbers from a real run, not a brochure.

Measured on 17 November 2026 on the Groot Schijn catchment in Deurne: a 59 mm, 3-hour design storm on official Flemish terrain and building data.

buildings evaluated
2,428
GRB footprints in the 1 km² pilot
reached by 15 cm water
646
threshold-entry depth, T20 baseline
seconds to solve
8.3
SFINCS wall-clock, 5 m grid, 3 h storm
agreement with VMM map
0.36
critical success index at 5 cm

An honest accuracy figure. Against the official VMM Watertoets pluvial map the pilot scores a CSI of 0.36. That is useful for screening, not yet for design: the sewer network is not modelled and infiltration is uniform. We publish the figure and the limitations together, inside the product too.

Measures are not oversold. In the pilot, a Rivierenhof retention basin alone lowers the 95th-percentile depth by 0.13 m in the area it drains and clears 281 m of road below 15 cm, but protects no building on its own. The twin shows that before anything is built.

Solver: Deltares SFINCS v2.5.0 (GPL-3.0, built from source). Monetary damage is withheld until validated Flemish depth-damage curves are available.

FloodTwin workspace: simulated maximum flood depth over the Deurne pilot on a 1 m terrain

Trust

Built for decisions that will be questioned.

Water authorities answer to councils, auditors and residents. FloodTwin keeps the evidence chain intact from gauge to report.

Every layer says what it is

Official record, observed, model estimate or inference: the label travels with the number into every export.

Tamper-evident audit trail

Each run, review and adjustment is written to a SHA-256 hash chain you can verify yourself.

Licences enforced in code

Non-commercial or restricted datasets are refused at ingest, and attribution is attached automatically.

People decide

Gauge alerts wait for a human reviewer. Nothing is dispatched automatically.

Credits

Pay for physics, not for looking.

Plans are sold on the Sensifai Marketplace and include a monthly credit allowance. Credits are only used when the solver runs; a failed run is refunded automatically.

Credit cost per action
ActionCredits
Explore terrain, layers and the point inspector0
Baseline storm simulation (SFINCS 2D)10
Mitigation counterfactual with Δh map15

Civic pilot plan: 500 credits a month · Regional authority plan: 2,500 credits a month.

FAQ

Questions engineers ask first.

Short answers here; the guide has the detail.

Run the next storm before it arrives.

Open the Deurne pilot, or talk to Sensifai about a twin of your own catchment in Flanders.