基于空间随机森林的矿产资源定量预测——以河北大庙—红石砬钒钛磁铁矿带为例OA北大核心

Existing mineral prospectivity mapping(MPM)methods pay insufficient attention to spatial heterogeneity and spatial autocorrelation,which limits their effectiveness.To address this,we have developed a spatial random forest(SRF)method that integrates K-means clustering with traditional random forest(RF)for MPM in the Damiao-Hongshila Fe-V-Ti ore belt,Hebei Province,China.The SRF method first constructs spatially heterogeneous subsets via K-means clustering,establishing heterogeneous RF models as base learners.The predicted probabilities are subsequently aggregated using inverse distance weighting according to the distances between the target center locations and the corresponding cluster centroids.Our method outperforms traditional RF,with an improved area under the curve score of 6.83%,an improved accuracy of 8.62%,and an improved F1 score of 8.52%while maintaining high interpretability.Based on the SRF model and a concentration-area fractal analysis,five prospective targets were delineated for mineral exploration in the study area.In summary,SRF provides an effective framework for modeling both spatial heterogeneity and spatial autocorrelation in MPM.