Spatiotemporal dynamics of surface water and eco-environmental responses in Thac Ba Reservoir, Vietnam: a remote sensing and machine-learning approach

Abstract

This study analyses the spatiotemporal dynamics of surface water and shoreline environmental responses in Thac Ba Reservoir, Vietnam, using multi-source remote sensing and Random Forest classification. Landsat, Sentinel-1, Sentinel-2, terrain variables and hydrometeorological data were integrated for the benchmark years 2015, 2020 and 2025 under low-water and high-water conditions. Classification performance was consistently good, with overall accuracy ranging from 0.850 to 0.920 and Kappa values from 0.77 to 0.88. Under low-water conditions, water bodies expanded from 13,830.3 ha to 16,916.9 ha, terrestrial ecosystems declined slightly and semi-inundated ecosystems changed non-linearly. Vegetation patterns differed between hydrological states, indicating reorganization of the littoral transition zone. The results show that reservoir change involved not only variation in water extent, but also spatial adjustment of shoreline environments.