Multi-Sensor Machine Learning and Spatial-Statistical Analysis of Illegal Small- Scale Mining (Galamsey) Dynamics in Wassa Amenfi East District, Ghana, 2023–2025

Abstract Illegal small-scale gold mining (galamsey) continues to drive rapid environmental degradation in Ghana, yet operational multi-temporal monitoring remains limited. This study developed a multi-sensor machine learning framework integrating Sentinel-1 SAR and Sentinel-2 optical data to map and monitor galamsey in Wassa Amenfi East District between 2023 and 2025. Random Forest, XGBoost and Support Vector Machine classifiers were trained on 13 spectral and SAR features using 612 stratified reference points (70/30 split), achieving high accuracy (Overall Accuracy 0.945–0.962; Kappa 0.827–0.875) with no statistically significant difference between classifiers (McNemar's tests, all p > 0.05). The red band (B4), VH backscatter and the green band (B3) were the most influential predictors. An independent, basemap-verified validation of 300 points found comparable Overall Accuracy (0.967) but lower detection completeness than the original test set implied (Producer's Accuracy 0.813 vs 0.939), indicating in-sample accuracies represent an upper bound on real-world performance. Detected mining area increased from 5,443 ha in 2023 to 17,614 ha in 2025 (+ 224%), driven mainly by newly established sites (13,805 ha). Spatial statistics confirmed very strong clustering (Moran's I = 0.567, p < 0.001; Geary's C = 0.433, p < 0.001), with extensive, statistically significant hotspots along river corridors at the 95% and 99% confidence levels and no coldspots detected. The framework offers a practical, low-cost basis for multi-year monitoring that quantifies expansion, separates new mining from persistence, and localises high-intensity hotspots to support targeted enforcement and environmental management in Ghana and similarly affected regions.