Optimization of Geological Feature Distribution Mapping based on Remote Sensing Data with Random Forest Algo-rithm for Potential Geothermal Field

Conventional geological mapping in volcanic geothermal fields is hindered by rugged topography, dense vegetation, and limited site accessibil- ity, resulting in sparse and inconsistent geological data. To address these lim- itations, this study proposes a remote-sensing-based geological reconstruction framework utilizing Random Forest classification to enhance subsurface fea- ture interpretation in inaccessible areas. Regional geological maps are inte- grated with remote sensing–derived physical property datasets to improve pat- tern recognition, hierarchical feature learning, and grid-based spatial recon- struction. Pre-processing includes data normalization, value standardization, and pixel resolution adjustment to ensure computational stability and model re- liability. Using an open-source Random Forest Classifier, decision-tree ensem- bles are trained to categorize geological units and detect spatial anomalies. The resulting reconstructed model reveals detailed geological structures, including previously unrecognized lithological inclusions and possible intrusive bodies that conventional mapping fails to capture, which generally relies on coarse- scale interpolation. Model predictions indicate widespread inclusion stratigra- phy, highlighting the structural and lithological complexity of the geothermal system, with an overall classification accuracy of 72.82%. The findings demon- strate the value of machine-learning-assisted geological modeling for volcanic environments. Future work should incorporate field validation, refine uncer- tainty quantification, explore alternative algorithms (e.g., XGBoost, CNN-based models), and expand applications toward geothermal monitoring and resource assessment.