Hello everyone,
I am studying HydroGraphNet, a physics-informed, graph-based neural operator designed as a fast surrogate for HEC-RAS–style shallow water equation (SWE) solvers in flood forecasting.
From my understanding, HydroGraphNet:
-
Learns depth-averaged hydraulic dynamics (water depth, velocity, discharge)
-
Is trained on SWE-based numerical solvers (e.g., HEC-RAS 1D/2D)
-
Prioritizes real-time, basin-scale prediction
However, many real flood scenarios (e.g., debris-laden floods, mudflows, density-varying flows, strong vertical accelerations) violate SWE assumptions and are more accurately described by full 3D Navier–Stokes equations (NSE) or particle-based methods (SPH).