Hybrid Expert–Machine Learning Weighting for GIS-Based Photovoltaic Suitability Mapping: A Spatial Robustness Assessment under Uncertainty

Geographic Information Systems and Multi-Criteria Decision-Making (GIS–MCDM) workflows are widely used for photovoltaic (PV) site suitability mapping, but their outputs depend strongly on criterion weights, and many studies report deterministic suitability maps without testing whether priority zones remain stable under weight uncertainty. This study combines expert-derived weights from the Analytic Hierarchy Process (AHP, consistency ratio = 0.0142) with SHAP-based feature importance from an XGBoost model trained on 72 existing PV plants (ROC–AUC = 0.9391) to construct a hybrid weighting scheme (α = 0.7), applied through SAW, TOPSIS, and VIKOR to map PV suitability in the Province of L’Aquila, Italy. Spatial robustness is assessed using 1,000 Monte Carlo weight perturbations and Tornado-style sensitivity analysis. Results show a clear stability gradient across the aggregation methods, with mean coefficient of variation values of 0.046, 0.121, and 0.193 for SAW, TOPSIS, and VIKOR, respectively. The top-suitability probability maps identify spatially concentrated high-suitability cores that persist across the three methods, while the Tornado analysis shows that GHI, urban distance, and slope exert the strongest control on spatial stability: the same criteria for which expert-derived and data-driven weights disagree most. By distinguishing robust priority zones from sensitivity-prone areas and identifying the criteria that govern spatial stability, the framework provides a defensible, uncertainty-aware workflow for PV siting and spatial energy planning.