Multi-objective optimization of biogas plant siting using GIS-based suitability mapping, XAI-driven machine learning and network-constrained logistics

The site selection for biogas plants is a multifaceted decision-making problem, particularly in regions with rugged topography, that necessitates balancing raw material accessibility, environmental suitability, and logistics costs. This study presents an integrated decision-support framework aimed at identifying optimal locations for biogas plants utilizing livestock manure (both large and small ruminant) in Bitlis Province, where logistical planning is of critical importance due to challenging geographical conditions despite the region's rich livestock potential. The proposed framework integrates GIS-based spatial variables, suitability modeling using CatBoost machine learning, explainable AI analyses based on SHAP and ICE, logistics cost calculations based on real-world road network topology, and Pareto-based multi-objective Grey Wolf Optimization (MO-GWO) into a single workflow. The ROC-AUC = 0.93 ± 0.06 and AP = 0.89 ± 0.10 values obtained via Spatial Block Cross-Validation demonstrate that the model produces a reliable and generalizable suitability surface despite the strong spatial dependence in mountainous terrain. SHAP analysis revealed that animal density is the dominant variable in the suitability decision, while distance from the road network acts as a physical constraint that reduces the likelihood of suitability beyond an approximate 2 km threshold. The optimization results indicate that environmental suitability functions as a minimum feasibility condition, and the vast majority of selected sites cluster around suitability values above 0.98. The Pareto front analysis systematically revealed the trade-off between suitability and total system cost; it determined that marginal increases beyond the approximate 0.98 suitability threshold require increasingly disproportionate cost increases, and that the M3 solution occupies the most balanced operational position in terms of both criteria. When compared to network-based benchmarking solutions, the MO-GWO M1 solution reduced the annual total system cost from 20.39 million EUR to 15.37 million EUR; it was determined that the primary source of this savings was a dramatic reduction in transportation costs, from 10.45 million EUR to 4.17 million EUR. Winter resilience scenarios confirmed that the M3 configuration maintains its balanced position even under seasonal conditions. The proposed hybrid framework offers a transparent, repeatable, and implementable approach for energy system planning in regions with topographic constraints.