文献类型：期刊文献

英文题名：A Novel Four-Dimensional Prediction Model of Soil Heavy Metal(Loid) Pollution: Geographical Explanations Beyond Artificial Intelligence "Black Box"

作者：Wang, Qi[1] Li, Cangbai[2] Hao, Dongmei[3] Shi, Xuewen[3] Xu, Yafei[3] Liu, Tongxu[1] Sun, Weimin[1] Zheng, Zelong[4] Liu, Jianfeng[5] Li, Wanqi[3] Liu, Wengang[3] Zheng, Jiaxue[6] Li, Fangbai[1]

第一作者：Wang, Qi

机构：[1] National-Regional Joint Engineering Research Center for Soil Pollution Control and Remediation in South China, Guangdong Key Laboratory of Integrated Agro-environmental Pollution Control and Management, Institute of Eco-environmental and Soil Sciences, Guangdong Academy of Science, Guangzhou, 510650, China; [2] Key Laboratory for City Cluster Environmental Safety and Green Development of the Ministry of Education, School of Ecology, Environment and Resources, Guangdong University of Technology, Guangzhou, 510006, China; [3] School of Management, Lanzhou University, Lanzhou, 730099, China; [4] Guangzhou First People’s Hospital, School of Medicine, South China University of Technology, Guangzhou, 510180, China; [5] Research Institute of Forestry, Chinese Academy of Forestry, Beijing, 100091, China; [6] School of Data Science and Artificial Intelligence, Dongbei University of Finance & Economics, Dalian, 116025, China

年份：2023

外文期刊名：SSRN

收录：EI(收录号：20230093567)

语种：英文

外文关键词：Artificial intelligence - Heavy metals - Risk perception - Soil pollution - Soils

摘要：The existing artificial intelligence (AI)-based prediction approaches of soil pollutants are insufficient to estimate the geospatial source-sink processes and to juggle the interpretability and accuracy, resulting in poor spatial extrapolation and generalization. In this study, we developed and tested a geographically interpretable four-dimensional AI prediction model of soil heavy metal(loid) (Cd) contents (4DGISHM) in Shaoguan city of China from 2016 to 2030. The 4DGISHM approach characterized spatio-temporal changes in source-sink processes of soil Cd. It was operationalized by estimating spatio-temporal patterns and the effects of drivers and their interactions of soil Cd from local to regional scales using TreeExplainer-based SHAP and parallel ensemble AI algorithms. The results indicate that the MSE and R2 values of the prediction model were 0.012 and 0.938, respectively at 1km spatial resolution. The predicted areas exceeding the risk control values of soil Cd across Shaoguan from 2022 to 2030 increased by 22.920% at baseline scenario. In 2030, enterprise and transportation (SHAP values 0.23 and 0.12, respectively) were the major drivers. Soil Cd was only marginally influenced by the driver interactions. Our approach goes beyond AI "black box" to juggle the spatio-temporal source-sink explanation and accuracy, and advances geographically precise prediction and control of soil pollutants. ? 2023, The Authors. All rights reserved.