Acosta et al. (2026) Exploring graph neural networks for flood modeling: Challenges, opportunities, and future prospects

Identification

• Journal: Environmental Modelling & Software
• Year: 2026
• Date: 2026-08-04
• Authors: Carlo Malapad Acosta, Herath Mudiyanselage Viraj Vidura Herath, Abhishek Saha, Sanka Rasnayaka, Lucy Amanda Marshall
• DOI: 10.1016/j.envsoft.2026.107132

Research Groups

• Department of Computer Science, School of Computing, National University of Singapore, Singapore
• School of Civil Engineering, Faculty of Engineering, The University of Sydney, Sydney, New South Wales, Australia
• Delft Institute of Applied Mathematics, Delft University of Technology, Delft, The Netherlands

Short Summary

This review synthesizes 49 peer-reviewed studies on Graph Neural Networks (GNNs) applications in flood modeling, identifying key gaps in physics incorporation, scalability, and cross-domain generalization. It outlines strategies to overcome these limitations and highlights the potential of GNNs as next-generation flood models.

Objective

• To explore the application of Graph Neural Networks (GNNs) in flood modeling
• To identify the strengths and limitations of current GNN approaches
• To propose research opportunities for advancing GNN-based flood modeling

Study Configuration

• Spatial Scale: Unstructured spatial domains, including river networks, coastal regions, urban areas, and storm water drainage systems
• Temporal Scale: Short-term to long-term flood prediction and risk management

Methodology and Data

• Models used: Graph Neural Networks (GNNs), Physics-Informed Neural Networks (PINNs)
• Data sources: Historical hydrodynamic data, satellite imagery, observation datasets, reanalysis products

Main Results

• GNNs can handle unstructured spatial data and integrate physics-based priors for improved generalizability
• Current GNN approaches are limited by oversmoothing node representations and difficulty capturing global relationships
• Physics-informed architectures and loss functions can improve physical consistency and interpretability of GNN models

Contributions

• This review provides a comprehensive overview of GNN applications in flood modeling, highlighting their potential as next-generation flood models
• The study identifies key research gaps and limitations, including physics incorporation, scalability, and cross-domain generalization
• The proposed strategies for overcoming these limitations can enhance the development of GNN-based flood models

Funding

• This research was funded by the National University of Singapore (NUS), the University of Sydney, and the Delft Institute of Applied Mathematics.

Citation

@article{Acosta2026Exploring,
  author = {Acosta, Carlo Malapad and Herath, Herath Mudiyanselage Viraj Vidura and Saha, Abhishek and Rasnayaka, Sanka and Marshall, Lucy Amanda},
  title = {Exploring graph neural networks for flood modeling: Challenges, opportunities, and future prospects},
  journal = {Environmental Modelling & Software},
  year = {2026},
  doi = {10.1016/j.envsoft.2026.107132},
  url = {https://doi.org/10.1016/j.envsoft.2026.107132}
}