Spatiotemporally adaptive gaussian mixture modelling (ST-GMM) framework for SAR-based long-term surface water monitoring

Study region Bangladesh, located within the Ganges–Brahmaputra–Meghna (GBM) delta, is one of the world’s largest and most hydrologically dynamic monsoonal floodplain systems. Study focus This study develops a Spatiotemporally Adaptive Gaussian Mixture Model (ST-GMM) framework for monthly surface water monitoring using Sentinel-1 SAR imagery. The framework dynamically models land–water backscatter distributions to derive spatially and temporally adaptive classification thresholds under varying hydro-climatic conditions. New hydrological insights for the region The proposed ST-GMM framework provides an adaptive, training-independent approach for nationwide surface water monitoring by combining Sentinel-1 SAR observations with district-specific probabilistic threshold estimation, enabling reliable mapping under persistent monsoon cloud cover. Validation against 4310 verified 2025 ground-truth observations yielded an overall accuracy of 92.55% (κ = 0.85; balanced accuracy = 91.2%). In a separate optical cross-sensor benchmark on a common 10 m grid, ST-GMM achieved 97.9%-pixel agreement and IoU = 0.896 with Sentinel-2 NDWI, while comparison with Otsu showed a statistically significant performance difference (McNemar χ² = 142.76, p < 0.001). Application of the framework revealed a pronounced seasonal hydrological cycle, with surface water extent expanding from approximately 15,900 km² during the dry season to 37,500 km² during the monsoon. Despite substantial inter-annual fluctuations, no significant long-term national trend was detected during 2015–2025. The framework therefore provides a scalable methodology for long-term hydrological monitoring in cloud-prone monsoonal and deltaic environments.