Two decades of surface water change in Lake Volta: A hybrid spectral index and machine learning approach (2004–2024)

Despite its socio-economic importance, the Volta Lake faces growing water stress due to climate variability, land use change, and limited hydrological monitoring. Existing surface water assessments in West Africa have often relied on single-index remote sensing approaches, which are prone to misclassification under complex conditions such as turbidity, vegetation encroachment, and seasonal variability. This study addresses this gap by developing a hybrid framework that integrates five spectral water indices (NDWI, MNDWI, AWEI, NWI, WRI) with a Random Forest classifier to map and analyze two decades (2004–2024) of surface water changes in Lake Volta. Results show a significant decline in surface water extent: NDWI and MNDWI values dropped by 88.4% and 84.3%, respectively, while AWEI decreased by 61.2%. The hybrid model achieved high classification accuracy, with AWEI yielding a 90.4% overall accuracy, a Kappa coefficient of 0.88, and an F1-Score of 0.91. Accuracy metrics confirm the model's robustness across heterogeneous land-water interfaces and changing climatic conditions. This study is the first to generate a 20-year, high-resolution water inventory for Lake Volta using a multi-index machine learning approach. It offers an operationally scalable tool for water resource monitoring in data-scarce regions. The findings have direct policy relevance, highlighting areas of persistent water loss and informing adaptive reservoir management, land use planning, and transboundary governance. By combining physical spectral logic with machine learning precision, this work advances remote sensing methodology and provides a replicable framework for sustainable water governance across West Africa.