Dan Tian, Hongxing Liu, Lei Wang, Sagy Cohen, T. Mandal
Abstract Remote sensing imagery is widely utilized for mapping flood extents. However, sensor limitations and environmental conditions often reduce mapping capability. Factors such as obscuring clouds, terrain shadows, and dense tree canopies frequently lead to significant information gaps and omission errors, particularly in forests and urban areas. Moreover, most remote‐sensing approaches estimate only flood extent and neglect water depth, even though depth information is critical for damage assessment and hydrologic modeling. This study introduces RS‐FloodXDepth, a novel hydrologically guided region‐growing method that enhances flood maps derived from remote sensing by integrating high‐resolution Digital Elevation Models (DEMs). Flood pixels identified in imagery are grouped into objects, serving as flood seeds. For each seed, its local water level is estimated from DEM‐derived surface elevation along its boundary. A seeded region‐growing algorithm, governed by hydrological principles of surface water flow, is then applied to propagate flood extent from the identified seeds. This method effectively detects previously missed flooded areas across a range of land‐cover types, including forests and buildings, and fills data gaps caused by clouds, shadows, and other occlusions, substantially improving map accuracy and reliability. Beyond extent, RS‐FloodXDepth concurrently produces flood depth estimations. The method was applied to the devastating flood in Goldsboro, North Carolina, during Hurricane Matthew (2016), using PlanetScope multispectral images and a high‐resolution DEM. Results demonstrate a significant reduction in omission error from 41.41% to 1.82% and a substantial increase in flood extent, particularly in forested and urban areas. RS‐FloodXDepth markedly improves the quality and informational content of flood inundation products, offering crucial data for emergency response, damage assessment, and flood risk management.