This paper is the result of geothermal geological survey engineering.
Objective This study aims to develop new technical means to enhance comprehensive evaluation capabilities, as traditional methods relying on geological exploration, GIS, and empirical judgment are often insufficient for a thorough assessment of site suitability for geothermal power plants.
Methods A random forest model has been constructed by using a global spatial database of 18 environmental parameters at a resolution of 0.5°×0.5°, which including geological, topographic, hydrogeological, and geothermal features. The optimal parameter combination was identified through classification metrics, model validation, comparison with previous prediction results, and consistency with planned geothermal power plant locations. A global suitability map for geothermal power plant development was generated. Feature importance was evaluated with scikit-learn, and the dominant controls on site selection were further examined using SHAP analysis.
Results The optimized parameter set achieved an AUC value of 0.9768. High suitability scores were observed in southern Europe; eastern and northern Africa; the southwestern (Himalayan belt), eastern coastal, and northeastern regions of China; Indonesia; eastern Russia; southern Australia; and along the circum−Pacific geothermal belt in both Oceania and the Americas. Terrestrial heat flow, lithospheric thickness, Moho depth, volcanic density, and rock thermal conductivity are key factors influencing site suitability.
Conclusions The random forest model demonstrates strong capability in integrating multi-source spatial data to evaluate global geothermal power plant site suitability. The findings provide valuable references for industry planning, policy formulation, and construction assessment in the geothermal sector. Moreover, the study highlights the broader applicability of random forest–based resource evaluation and prediction in other fields.
DownLoad: