Zhouyayan Li, Bekir Zahit Demiray, Marian Muste, Ibrahim Demir

Floods rank among the most devastating natural hazards globally. Unlike many other natural calamities, floods typically occur in densely populated regions, resulting in immediate and long-term adverse impacts on communities, including fatalities, injuries, health risks, and significant economic and environmental losses annually. Traditional flood models, while useful, are constrained by simplifying assumptions, numerical approximations, and a lack of sufficient data for accurate simulations. Recent advancements in data-efficient Digital Elevation Model (DEM) and Digital Terrain Model (DTM) based flood models show promise in overcoming some of these limitations. However, these models' reliance on DEM or DTM data renders them sensitive to the dynamic nature of the Earth's surface. This study investigates the effectiveness of remote sensing imagery for flood inundation mapping, focusing on the role of high-resolution commercial optical PlanetScope images in data-limited scenarios. To address early-stage reflectance issues attributed to the lack of on-board calibration in PlanetScope constellations, we introduced a novel post-processing workflow, the Quantile-based Filling and Refining (QFR). Our results indicate that the initial flood extent maps produced using the widely adopted Normalized Difference Water Index (NDWI) were inferior to manual delineations and comparable to those generated using only the Near-Infrared (NIR) band, which also suffers from reflectance flaws. However, flood maps generated using NIR band data processed with the QFR significantly outperformed manual delineations. This research demonstrates the potential of commercial remote sensing imagery for precise flood inundation mapping, particularly at smaller scales, such as urban areas. Additionally, it underscores the QFR post-processing workflow's effectiveness in enhancing prediction accuracy, offering a streamlined and scalable method for improving flood modeling outcomes.