Multiscale Mining Pressure and Machine Learning-Based Susceptibility of Protected Areas in Ecuador

This study evaluates the spatial relationship between Ecuador’s mining cadastre and protected areas and investigates whether multiscale mining pressure can explain and predict direct mining-cadastre overlap. The objective was to move beyond conventional binary overlay analysis by identifying protected areas whose surrounding mining configurations indicate elevated overlap susceptibility. A CRISP-DM-based geospatial machine-learning framework was applied to 6,275 mining entities and 97 protected areas. Direct geometric intersections were quantified, followed by multiscale proximity analysis using 1, 5, 10, and 20 km distance thresholds and exclusive rings. Spatial autocorrelation was assessed, and multiscale frequency, area, and density indicators were used as predictors. Logistic Regression, Random Forest, and XGBoost were evaluated using spatial block cross-validation to reduce geographic information leakage. Model performance was assessed using ROC-AUC, PR-AUC, Brier score, and complementary classification metrics. Direct cadastral overlap affected 14 protected areas (14.43%), comprising 31 intersection pairs and approximately 90.91 km2 of cumulative overlap. Mining presence increased substantially with distance, occurring around 52.58%, 86.60%, 93.81%, and 97.94% of protected areas within 1, 5, 10, and 20 km, respectively. XGBoost achieved the best pooled spatially out-of-fold performance (ROC-AUC = 0.901; PR-AUC = 0.688; Brier score = 0.093). The most informative predictors were concentrated within the immediate 0–1 km surroundings. Direct overlap alone underrepresents the broader spatial context of mining pressure around Ecuador’s protected areas. Integrating multiscale geospatial analysis with spatially validated machine learning provides a reproducible framework for identifying and ranking protected areas according to mining-cadastre overlap susceptibility, while avoiding interpretation of these probabilities as evidence of active mining or environmental degradation.