Flood depth maps used for training & testing in "Subgrid informed neural networks for high-resolution flood mapping" & "Flood-LDM: Generalizable latent diffusion models for rapid and accurate zero-shot high-resolution flood mapping" papers [Dataset]

Abstract

This dataset contains coarse-grid flood maps, fine-grid flood maps, and Digital Elevation Model (DEM) images with a spatial resolution of 512 × 512 pixels, which were used to train and evaluate deep learning models in the studies “Subgrid informed neural networks for high-resolution flood mapping” (https://doi.org/10.1016/j.jhydrol.2025.133329) and “Flood-LDM: Generalizable Latent Diffusion Models for Rapid and Accurate Zero-Shot High-Resolution Flood Mapping” (https://openaccess.thecvf.com/content/WACV2026/html/Neo_Flood-LDM_Generalizable_Latent_Diffusion_Models_for_rapid_and_accurate_zero-shot_WACV_2026_paper.html). Flood simulations for both coarse and fine computational grids were generated using the HEC-RAS hydraulic modelling software developed by the US Army Corps of Engineers (https://www.hec.usace.army.mil/software/hec-ras/). Simulations were conducted for three Australian catchments: Wollombi, Burnett, and Chowilla. DEM datasets for the Wollombi and Burnett catchments were obtained from the ELVIS – Elevation and Depth – Foundation Spatial Data portal (https://elevation.fsdf.org.au/), while the DEM for the Chowilla floodplain was sourced from the dataset provided by Niels Fraehr (https://doi.org/10.26188/21235782). Rainfall and inflow forcing data used to drive the hydrodynamic simulations were obtained from the Bureau of Meteorology Water Data Online portal (http://www.bom.gov.au/waterdata/). The dataset includes paired coarse- and fine-resolution flood depth maps together with corresponding DEM inputs, enabling the development and benchmarking of machine learning models for rapid high-resolution flood mapping. Flood depth values are provided in centimetres (cm), while DEM elevations are given in metres (m). Due to file size limitations, the dataset is distributed in multiple parts. Further methodological details and guidance on dataset usage can be found in the associated publications.