Sample-Adjusted Threshold Calculation (SATC) for Optimizing Flood Mapping From a Trustworthy and Explainable Perspective

Near-real-time flood maps are vital for efficiently coordinating emergency responses during flooding events. Synthetic aperture radar (SAR) satellite remote sensing, lauded for its capacity to capture data day and night in diverse weather conditions, emerges as a leading tool for acquiring flood mapping information. However, in cross-regional flood monitoring, challenges in accurately detecting floodwater pixels arise from interference factors affecting SAR backscatter, which are inherently present in various geo-environments (e.g., wind waves, vegetation, thick clouds, and high-relief terrain). These factors include shadow and layover effects, as well as radar response areas that resemble water surfaces and land cover. To address these challenges, our study proposes the sample-adjusted threshold calculation (SATC) approach, which not only quantifies the impact of these interference factors but also enhances algorithm generalization and interpretability, effectively overcoming the limitations of regional specificity in flood mapping. SATC applies three layers of constraints, including sample space distribution, sample proportion, and algorithm threshold space, to comprehensively calculate a flood segmentation threshold satisfying varied conditions involved in cross-regional flood monitoring. The experimental results of applying SATC to various scenarios indicate that these SATC-optimized algorithms achieve a notable 10–30% improvement in accuracy while concurrently reducing false and omission rates to <20% . Furthermore, incorporating SATC into the knowledge-driven flood intelligent monitoring (KDFIM) framework yields optimal results, achieving a flood mapping accuracy >95% and a Kappa value >0.95. The reached KDFIM(SATC) facilitates heightened robustness and enables the rapid diagnosis and quantification of flood mapping details, achieving flood detection in just 0.3795 s/100 km². Overall, KDFIM(SATC) proves to be robust as a pivotal tool in emergency response efforts during flooding events.