Mapping maximum floodwater footprints from sustained flood traces.

Remote-sensing-based flood extent mapping is essential for flood management. However, existing approaches that target standing floodwater often underestimate Maximum Flood Footprints (MFF) because inundation can expand and recede between satellite revisits. Here, we address this limitation by developing a flood-trace-based MFF extraction framework that uses sustained post-flood traces as delayed evidence of maximum inundation. To make flood traces reliably detectable, the framework introduces a trace-oriented anomaly detection strategy that uses Normalized Difference Water Index (NDWI) anomalies as the primary indicator and CCDC-based historical baselines to separate flood-related signals from normal seasonal variability and spatial heterogeneity. Validation on the once-in-a-century 2023 Hai River Basin (HRB) flood event shows that the framework achieves an average F1-score of 81.45%, whereas four widely used standing-floodwater-based approaches attain F1-scores of only 28.24% to 36.41%. At the catchment scale, the proposed method reduces MFF underestimation caused by missed flood-peak observations. Compared with our estimate, MFF derived from Sentinel-1 alone, Sentinel-2 alone, multi-source SAR imagery (Sentinel-1 + Gaofen-3 + Lutan-1), and fusion of Sentinel-1 and Sentinel-2 is 43.3%, 48.0%, 34.2%, and 20.2% smaller, respectively. This work presents an MFF extraction approach that does not depend on the coincidence between satellite acquisitions and flood peaks. Collectively, these findings contribute to more complete flood mapping and underscore the need to account for systematic underestimation in existing flood-extent datasets.