Inland Excess Water (IEW) Monitoring Using Sentinel-1/2: A SplitClass Segmentation and Temporal Gap-Filling Approach

Yahya Ibrahim, Márta Belényesi, Chang Liu, Mátyás Richter-Cserey, Máté Simon, Tamas Sziranyi, Csaba Benedek; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 2875-2885

Accurate mapping of Inland Excess Water (IEW), encompassing waterlogging and water retention areas, is essential for advancing sustainable land management and large-scale water conservation efforts. Existing remote sensing methods often focus on Binary water-nonwater segmentations, struggle in wet-dry transitional zones, and face challenges from cloud-related data gaps. To address these limitations, we propose a novel approach that combines what we refer to as SplitClass segmentation with a temporal Gap-Filling framework. SplitClass segmentation is an adaptive approach that assigns either a single class or a pair of top-2 class labels to each pixel, depending on prediction confidence. Meanwhile, the framework fuses temporal optical (Sentinel-2) and radar (Sentinel-1) data, ensuring improved temporal consistency and mitigating cloud occlusions. To support this research, we present IEW-Seg, a new dataset comprising fine-grained waterlogging classes, multi-source satellite inputs, and field-validated ground truth. Quantitative evaluation demonstrates that our method significantly enhances IEW detection accuracy. The official website and more detailed information are available https://waterdetection.github.io/water_detection/ .