A Physics-Inspired Multi-Flow Graph Neural Network with Explicit Flow Allocation for Sustainable Rainfall-Driven Flood Simulation

Graph-based surrogate models offer an efficient alternative to hydrodynamic simulations, but accurately representing multi-directional runoff propagation remains challenging. This study proposes a physics-inspired multi-flow graph neural network with an explicit flow allocation mechanism for rainfall-driven surface runoff simulation. The method allows runoff information to propagate through multiple downstream pathways and dynamically regulates their relative contributions according to local water-surface elevation differences. The allocation coefficients are interpreted as relative information weights rather than volumetric fluxes, avoiding unverified mass-conservation claims. The framework was evaluated using flood simulation data from Harris County, Texas, and compared with a single-flow baseline and a multi-flow model without explicit allocation. The models were trained on nine watershed–event cases, validated on three cases, and evaluated on an independent test set of 51 independent watershed–event cases. Introducing multi-flow connectivity alone increased the root mean square error by 5.4% relative to the single-flow baseline, whereas the proposed model achieved the best performance, with a root mean square error of 0.164 m, a Nash–Sutcliffe efficiency of 0.880, and a Pearson correlation coefficient of 0.950. The results indicate that increasing graph connectivity is insufficient by itself; physics-inspired directional allocation is essential for representing multi-directional runoff propagation in graph-based flood surrogate models. By improving simulation efficiency and directional representation, the proposed method contributes to the goals of sustainable flood management and community resilience.

Authors

Institutions

Publication Details

Journal
Sustainability
Published
2026-11-12
DOI
https://doi.org/10.3390/su182211514
Primary Topic
Hydrological Forecasting Using AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Physics-Inspired Multi-Flow Graph Neural Network with Explicit Flow Allocation for Sustainable Rainfall-Driven Flood Simulation

Xinru Zhang, Fuxin Chai, Ying Xiao, Huimin Xu et al.
Sustainability
Hydrological Forecasting Using AI
article

A Physics-Inspired Multi-Flow Graph Neural Network with Explicit Flow Allocation for Sustainable Rainfall-Driven Flood Simulation

Xinru Zhang, Fuxin Chai, Ying Xiao, Huimin Xu, Ying Ding, Hongping ZHANG, Xi Yang
article en

Abstract

Graph-based surrogate models offer an efficient alternative to hydrodynamic simulations, but accurately representing multi-directional runoff propagation remains challenging. This study proposes a physics-inspired multi-flow graph neural network with an explicit flow allocation mechanism for rainfall-driven surface runoff simulation. The method allows runoff information to propagate through multiple downstream pathways and dynamically regulates their relative contributions according to local water-surface elevation differences. The allocation coefficients are interpreted as relative information weights rather than volumetric fluxes, avoiding unverified mass-conservation claims. The framework was evaluated using flood simulation data from Harris County, Texas, and compared with a single-flow baseline and a multi-flow model without explicit allocation. The models were trained on nine watershed–event cases, validated on three cases, and evaluated on an independent test set of 51 independent watershed–event cases. Introducing multi-flow connectivity alone increased the root mean square error by 5.4% relative to the single-flow baseline, whereas the proposed model achieved the best performance, with a root mean square error of 0.164 m, a Nash–Sutcliffe efficiency of 0.880, and a Pearson correlation coefficient of 0.950. The results indicate that increasing graph connectivity is insufficient by itself; physics-inspired directional allocation is essential for representing multi-directional runoff propagation in graph-based flood surrogate models. By improving simulation efficiency and directional representation, the proposed method contributes to the goals of sustainable flood management and community resilience.

SustainabilityVol. 18(22)
China Institute of Water Resources and Hydropower Research (CN)
Clean water and sanitation, Climate action
Hydrological Forecasting Using AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon