Our 1D–2D hydraulic surrogate is, in essence, a world model: a learned internal model that predicts how water levels, flows, and network-surface exchanges evolve given an initial state and a sequence of forcing inputs.
Instead of solving equations step by step, the model learns a latent representation of the hydraulic domain and a transition operator that advances state in time — preserving invariants like mass conservation.
The result: planning agents and early warning systems can 'imagine' thousands of alternative hydraulic futures in real time. What if rainfall doubles? What if a pump fails? What if we build a wall here?
Partner with us on this research →Deploy in days, not quarters. Bring your simulation engine, your sensors, your audience — we provide the operational backbone.