A method for identifying urban flood inundation based on single-temporal remote sensing imagery

The extracted flood extents were further compared with those obtained using two conventional approaches and were generally consistent across the three study areas. Agreement F1-scores ranged from 0.694 to 0.870 relative to pre- and post-flood change detection based on the modified normalised difference water index, and from 0.603 to 0.866 relative to newly inundated area extraction using the European Commission Joint Research Centre permanent surface-water mask.

On an experimental platform equipped with an NVIDIA Jetson AGX Orin, processing the Porto Alegre, Manville, and Zhengzhou scenes required 24.3, 2.68, and 19.8 s, respectively. These quantitative results indicate that the proposed method can identify major urban flood inundation areas under diverse geographical and hydrological conditions. Requiring only a single post-event multispectral image, no model training, and no high-performance computing, the method provides a lightweight solution for rapid urban flood inundation mapping and resource-constrained applications, including onboard satellite processing.</p>