How remote sensing can enhance flood mapping by leveraging a terrain-based flood inundation model

Flood mapping is essential for effective disaster management, requiring rapid and accurate delineation of flood extents and depth. Remote sensing, especially using satellite imagery, provides data for large spatial extents but faces challenges like cloud and canopy cover. This study explores a hybrid approach combining remote sensing data and a low-complexity flood inundation model (FLDPLN) to generate accurate flood inundation maps. A novel flood edge extraction algorithm was developed and implemented within Google Earth Engine (GEE). The edge pixels are used to estimate stream water depth/stage which serves as input to the FLDPLN model to estimate flood inundation extent and depth. The proposed methodology (named FLDSensing) was applied over the Verdigris River in southeast Kansas to map a flood event that occurred in late May 2019. A Sentinel-2 imagery was used to generate a flood map within the FLDSensing GEE App. FLDSensing results show a better match of flood inundation extent compared to a HEC-RAS 2D benchmark with performance metrics of CSI/F1-scores reaching 0.8/0.9 from 0.6/0.7 using remote sensing alone. Integrating remote sensing data and analysis with the FLDPLN model addresses issues using remote sensing imagery alone and provides a more complete and accurate flood mapping solution.