Integrating Landsat 8 remote sensing and machine learning for water quality assessment in a tropical river basin

Artisanal and small-scale gold mining (ASGM) has degraded river water quality in sub-Saharan Africa, while conventional monitoring remains constrained by cost and logistical challenges. This study integrates Landsat 8 OLI multispectral imagery with machine-learning regression to predict pH, dissolved oxygen (DO), electrical conductivity (EC), total dissolved solids (TDS), total suspended solids (TSS), and turbidity across 15 stations in Ghana’s Ankobra River Basin. Six OLI bands and ten spectral indices were extracted using Google Earth Engine, and Random Forest (RF), Support Vector Machine (SVM), k-Nearest Neighbour, and Gaussian Process Regression were evaluated using repeated 5-fold cross-validation. RF achieved strong performance across parameters (R2 = 0.73–0.93), whereas SVM provided the highest turbidity accuracy (R2 = 0.94). All stations had DO below WHO guidelines, while EC exceeded 400 µS/cm at 12 stations. Kriging identified major turbidity and EC hotspots, supporting satellite-based ML for cost-effective water-quality monitoring.