Identification of the primary pollution sources and dominant influencing factors of soil heavy metals using a random forest model optimized by genetic algorithm coupled with geodetector

作者：Tong Liu, Mingshi Wang, Mingshi Wang, Mingya Wang, Mingya Wang, Qinqing Xiong, Luhao Jia, Wanqi Ma, Shaobo Sui, Wei Wu, Xiaoming Guo · 发表于：Ecotoxicology and Environmental Safety · 年份：2025 · DOI：10.1016/j.ecoenv.2025.117731 · 被引用次数：15 · 研究领域：Geochemistry and Geologic Mapping、Heavy metals in environment

Identifying and quantifying the dominant factors influencing heavy metal (HM) pollution sources are essential for maintaining soil ecological health and implementing effective pollution control measures. This study analyzed soil HM samples from 53 different land use types in Jiaozuo City, Henan Province, China. Pollution sources were identified using Absolute Principal Component Score (APCS), with 8 anthropogenic factors, 9 natural factors, and 4 soil physicochemical properties mapped using Geographic Information System (GIS) kernel density estimation. Geodetector and a genetic algorithm optimized random forest model (GA-RF) were employed to quantify the dominant factors and precisely identify pollution sources. A Monte Carlo model was further applied to assess source-oriented health risk probabilities across age groups in the study area. The results revealed three principal components representing pollution sources, with contribution rates of 47.2 %, 33.3 %, and 19.5 %, respectively. For pollution source 1, industrial activities were dominant, with factory density (27.7 %) and distance from the factory (36.3 %) identified as the main factors. Cr, Cu, Mn, and Ni had high loads in this source. Pollution source 2, a combination of natural and transportation influences, was primarily affected by the normalized difference vegetation index (NDVI, 37.8 %), road network density (16.8 %), and proximity to roads (15.3 %). Pollution source 3 was linked to agricultural activities, with ...