Lim et al. (2026) UrbanFloodBench: Bridging AI and hydrology through benchmarking of coupled 1D–2D urban flood surrogate models

Identification

• Journal: Environmental Modelling & Software
• Year: 2026
• Date: 2026-09-05
• Authors: Jia Yu Lim, Herath Mudiyanselage Viraj Vidura Herath, Sanka Rasnayaka, Lucy Amanda Marshall, Hui Zou, Abhishek Saha, Jobayer Hossain, Jing Yen Tong, Ing Zhen Lee, Yukiya Tsukada, Shrey Gandhi, Katsutoshi Matsumaro, Sajay Raj, Artyom Mazur, Andrey Khlopotnykh, Dmytro Cheshenko
• DOI: 10.1016/j.envsoft.2026.107154

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, Australia
• Delft Institute of Applied Mathematics, Delft University of Technology, Delft, The Netherlands
• Individual Researchers
• Department of Electrical and Computer Engineering, College of Design and Engineering, National University of Singapore, Singapore
• Faculty of Computer Science and Information Technology, Universiti Malaya, Malaysia
• SI Research Institute, Japan
• Woxsen University, India
• National Research University Higher School of Economics, Russia
• Faculty of Applied Mathematics and Information Technologies, Oles Honchar Dnipro National University, Dnipro, Ukraine

Short Summary

This paper introduces UrbanFloodBench, a new open benchmark for coupled 1D–2D urban flood surrogate modelling. The benchmark combines a multi-catchment dataset, a variance-aware evaluation framework, and a global AI challenge to enable large-scale comparison of AI approaches.

Objective

• Develop an open benchmarking ecosystem for coupled 1D–2D urban flood surrogate modelling.
• Evaluate the performance of various machine learning (ML) approaches on a controlled, reproducible dataset.
• Identify effective computational methods for predicting water levels in fully coupled 1D–2D urban flood systems.

Study Configuration

• Spatial Scale: Urban catchments with varying sizes and complexities.
• Temporal Scale: Rainfall events with different intensities and durations.

Methodology and Data

• Models used: HEC-RAS hydrodynamic modelling software (version 6.7 Beta 4a).
• Data sources: Synthetic rainfall data, generated using a range of intensities and durations.

Main Results

• The UrbanFloodBench dataset comprises four urban flood models that span a broad range of drainage network sizes, surface topographic complexity, and hydraulic behaviour.
• The benchmark evaluates controlled HEC-RAS simulator emulation, where data-driven models are trained to reproduce HEC-RAS-generated hydraulic responses under prescribed rainfall forcing.
• The results show that multiple modelling approaches can achieve strong performance, while the 1D drainage component remains the primary unresolved challenge.

Contributions

• UrbanFloodBench establishes a reproducible foundation for advancing AI-based surrogate modelling and future evaluation of data-driven flood-emulation methods.
• The benchmark provides a common framework through which AI approaches for coupled urban flood modelling can be systematically evaluated, reproduced, and compared.

Funding

• NUS-USYD Ignition Grant
• Australian Bureau of Meteorology Design Rainfall Data System (Australian Bureau of Meteorology)
• National Oceanic and Atmospheric Administration (NOAA) Precipitation Frequency Data Server (PFDS)

Citation

@article{Lim2026UrbanFloodBench,
  author = {Lim, Jia Yu and Herath, Herath Mudiyanselage Viraj Vidura and Rasnayaka, Sanka and Marshall, Lucy Amanda and Zou, Hui and Saha, Abhishek and Hossain, Jobayer and Tong, Jing  Yen and Lee, Ing  Zhen and Tsukada, Yukiya and Gandhi, Shrey and Matsumaro, Katsutoshi and Raj, Sajay and Mazur, Artyom and Khlopotnykh, Andrey and Cheshenko, Dmytro},
  title = {UrbanFloodBench: Bridging AI and hydrology through benchmarking of coupled 1D–2D urban flood surrogate models},
  journal = {Environmental Modelling & Software},
  year = {2026},
  doi = {10.1016/j.envsoft.2026.107154},
  url = {https://doi.org/10.1016/j.envsoft.2026.107154}
}