UrbanFloodBench Dataset

Abstract

This dataset, UrbanFloodBench, was developed to provide an open and reproducible benchmark for testing, evaluating, and comparing data-driven surrogate models for coupled urban flood systems.

UrbanFloodBench is an open dataset of coupled 1D–2D urban flood simulations generated using HEC-RAS version 6.7 Beta 4a, developed by the U.S. Army Corps of Engineers Hydrologic Engineering Center (https://www.hec.usace.army.mil/software/hec-ras/download.aspx). The dataset contains four rain-on-grid urban flood models representing coupled 1D drainage networks and 2D surface flow domains: Beaver Lake, Davis CA, New Orleans, and Coogee NSW. For each model, synthetic rainfall events were simulated and full event outputs are provided. The original HEC-RAS simulation outputs have been processed into an ML-friendly format. The repository includes dynamic simulation outputs, static node/link attributes, and geospatial shapefiles for each model.

Models 1 and 2 were used to host the international UrbanFloodBench Kaggle competition: https://kaggle.com/competitions/urban-flood-modelling. The competition dataset is also available at: https://www.kaggle.com/datasets/jiayulim/urbanfloodbench/data.

This dataset is accompanied by the preprint paper titled "UrbanFloodBench: Bridging AI and Hydrology through Benchmarking of Coupled 1D–2D Urban Flood Surrogate Models" (http://dx.doi.org/10.2139/ssrn.6900753).

For full details on dataset generation, model configuration, variables, units, file structure, and recommended usage, please refer to the accompanying README.pdf, the preprint paper, and the UrbanFloodBench Kaggle competition webpage.