Abstract
Floods are among the world’s most costly and deadly natural disasters. Through decades of research and collaboration, many capable continental and global flood monitoring and forecasting systems (e.g., the National Water Model [NWM] and the Global Flood Awareness System [GloFAS]) have been developed, leveraging modeling, remote sensing, and deep learning approaches. Nonetheless, many inherent limitations remain, hindering society’s ability to prepare for floods, respond to emergencies, assess damage, and implement long term mitigation. Three prevalent limitations include a lack of incorporation of the ground’s true condition (i.e., soil moisture), limiting ground truth for models’ state validation, and cloud or dense vegetation obstruction, which lead to omission errors in satellite based flood extent maps. For the first limitation, we evaluated a new 400-meter downscaled satellite soil moisture product from the SMAP and SMOS missions during the prolonged 2022 floods in Sindh province, Pakistan. The study found that satellite soil moisture correlated much more strongly with Sentinel 1 flood extent than modeled soil moisture and effectively captured how inundation built up and receded. In the second study, to overcome the lack of in situ soil moisture observations, which hinders validation of both satellite derived and modeled soil moisture, we explored a proxy evaluation using rain gauges and the Soil Moisture to Rain (SM2RAIN) algorithm. Despite the inherent limitation of SM2RAIN, the proxy evaluation was moderately successful for certain climate zones and annual rainfall classes (i.e., temperate, higher annual rainfall). Lastly, to overcome the omission errors in satellite-based flood extent maps from optical images due to cloud contamination and dense vegetation, new terrain-based methods called FLEXTH and RS-FloodXdepth were applied to leverage topography information for flood enhancement. The study focused on the 2025 Midwest flooding in Louisville, Kentucky and 2017 Hurricane Harvey in Houston/Sugar Land and use high quality hand-label flood map from NOAA aerial photos for benchmarking purposes.
