Toward an Operational GNN-Based Multimesh Surrogate for Fast Flood Forecasting

arXiv cs.LG / 4/6/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper proposes a graph-neural-network (GNN) surrogate to accelerate operational flood forecasting that otherwise depends on slow high-fidelity 2D hydraulic solvers like Telemac2D.
Using a production-grade Telemac2D setup on a large unstructured mesh (~4×10^5 nodes) for the Lower Têt River (France), the authors build a learning-ready dataset from synthetic but operationally grounded flood events.
The surrogate uses a projected-mesh strategy to keep training feasible while retaining supervision fidelity, and a multimesh connectivity design to expand the spatial receptive field without deepening the network.
Experiments show that explicit conditioning on the discharge boundary feature Q(t) is essential, multimesh connectivity adds gains when conditioning is correct, and pushforward training improves stability for long autoregressive rollouts.
On the studied case, the learned surrogate delivers 6-hour predictions in ~0.4 seconds on a single NVIDIA A100 GPU versus ~180 minutes on 56 CPU cores, indicating practical suitability as a complement for operational flood mapping.

Abstract

Operational flood forecasting still relies on high-fidelity two-dimensional hydraulic solvers, but their runtime can be prohibitive for rapid decision support on large urban floodplains. In parallel, AI-based surrogate models have shown strong potential in several areas of computational physics for accelerating otherwise expensive high-fidelity simulations. We address this issue on the lower T\^et River (France), starting from a production-grade Telemac2D model defined on a high-resolution unstructured finite-element mesh with more than

4\times 10^5

nodes. From this setup, we build a learning-ready database of synthetic but operationally grounded flood events covering several representative hydrograph families and peak discharges. On top of this database, we develop a graph-neural surrogate based on projected meshes and multimesh connectivity. The projected-mesh strategy keeps training tractable while preserving high-fidelity supervision from the original Telemac simulations, and the multimesh construction enlarges the effective spatial receptive field without increasing network depth. We further study the effect of an explicit discharge feature

Q(t)

and of pushforward training for long autoregressive rollouts. The experiments show that conditioning on

Q(t)

is essential in this boundary-driven setting, that multimesh connectivity brings additional gains once the model is properly conditioned, and that pushforward further improves rollout stability. Among the tested configurations, the combination of

Q(t)

, multimesh connectivity, and pushforward provides the best overall results. These gains are observed both on hydraulic variables over the surrogate mesh and on inundation maps interpolated onto a common

25\,\mathrm{m}

regular grid and compared against the original high-resolution Telemac solution. On the studied case, the learned surrogate produces 6-hour predictions in about

0.4\,\mathrm{s}

on a single NVIDIA A100 GPU, compared with about

180\,\mathrm{min}

on 56 CPU cores for the reference simulation. These results support graph-based surrogates as practical complements to industrial hydraulic solvers for operational flood mapping.