LankaFlood AI

Predict · Stages 5 and 6

Let AI imagine the future.

A small learned model will predict the next water state from the current one, then repeat. No model exists yet, so this page shows the design and no predictions.

Draft model contract

The inputs and output are fixed early so the simulator, training and browser code all agree. The guide explains the mathematics.

  1. sₜStateWater depth on each grid cell at time t.
  2. zTerrainElevation, plus editable obstacles and drainage paths. Fixed during a run.
  3. aₜWeather (optional)Rainfall, cloud and upstream inflow, if this version of the model uses them.
  4. ŝₜ₊₁PredictionWater depth one step later. It feeds back in as the next input.
Rollout: each prediction becomes the next state, so the model imagines 5, 10 or 20 steps ahead and its errors compound. That is what this project will measure.

With and without weather

The same model will be trained three ways, so any difference comes from the inputs. Results are planned, and none exist.

M0

No weather

Water state and terrain only. How far can terrain and the current flood explain what happens next?

M1

Rainfall

M0 plus rainfall and upstream inflow. How much does knowing the water coming in improve the rollout?

M2

Rainfall and cloud

M1 plus cloud cover. Does cloud add anything once rainfall is already known?

The score to beat

On the real Kelani flood, simply assuming nothing changes between 30 November and 3 December scores a critical success index of 0.27, counting only ground analysed on both dates. A world model has to beat that to have learned anything useful about this transition.

Rollout

Steps to imagine (available once a model is trained)

Not built · arrives in stage 6

No model loaded

The trained model will download as a small ONNX file and run on your device with ONNX Runtime Web. Weights that have not been trained will never appear here.