Seasonal and interannual surface water variability in the Okavango River Basin using multi-satellite data and machine learning

Surface water resources in arid regions are increasingly vulnerable to climate variability and growing demand, which calls for accurate, timely, and spatially explicitly monitoring approaches. This study developed a machine learning-based framework for quantifying seasonal and interannual surface water dynamics through multi-sensor fusion approach by combining radar and optical data. Sentinel-1 synthetic aperture radar (SAR), Sentinel-2 multispectral imagery, and Landsat-8 surface reflectance data were processed within the Google Earth Engine (GEE) platform for the period 2015-2021. The study used Random Forest (RF) and Gradient Tree Boosting (GTB) classifier, achieving a remarkable overall classification accuracy from Sentinel-1/2 (OA = 99.34-99.92%; Kappa = 0.8862-0.9983) and Landsat-8 (OA = 99.33-99.76%; Kappa = 0.9859-0.9949). Moreover, positive correlations were observed among water indices, particularly Modified Normalized Difference Water Index (MNDWI) and Normalized Difference Water Index (NDWI), whereas SAR backscatter variables (VV_median and VH_median) indicated a positive correlation with vegetation indices such as Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI). Meanwhile, Shapley Additive Explanations (SHAP) revealed that the MNDWI, NDWI, and SAR backscatter metrics were the most influential predictors of surface water occurrence, followed by vegetation indices and lastly single bands (B2-B8). Seasonal analyses presented greater surface water variability in October and November compared to drier months of May to September. Interannual variability analysis was conducted, indicating maximum surface water extent in 2016 (28,869 km 2 ), and minimum in 2019 (9743 km 2 ). Meanwhile, descriptive analysis with hydroclimatic variables showed that variations in precipitation and evapotranspiration (ET) corresponded with observed spatiotemporal patterns of surface water extent except in 2017. Overall, the proposed framework offers a scalable approach for monitoring surface water variability in data-scarce arid regions, supporting water resource management, drought monitoring, and climate adaptation.