基于多源环境数据和随机森林模型的农田土壤砷空间分布模拟  

Prediction of arsenic for farmland soil based on multi source environmental data and random forest model

作　　者：史广[1] 刘庚 赵龙[3] 苏迎庆 毕如田[1] SHI Guang;LIU Geng;ZHAO Long;SU Yingqing;BI Rutian(School of Resources and Environment,Shanxi Agricultural University,Taigu 030801;Research Center for Scientific Development in Fenhe River Valley,Taiyuan Normal University,Jinzhong 030619;State Key Laboratory of Environmental Criteria and Risk Assessment,Chinese Research Academy of Environmental Sciences,Beijing 100012)

机构地区：[1]山西农业大学资源环境学院,太谷030801 [2]太原师范学院汾河流域科学发展研究中心,晋中030619 [3]中国环境科学研究院环境基准与风险评估国家重点实验室,北京100012

出　　处：《环境科学学报》2020年第8期2993-3000,共8页Acta Scientiae Circumstantiae

基　　金：国家自然科学基金项目(No.41401236);国土资源部公益性行业项目(No.201411007);山西省高等学校科技创新项目(No.2019L0777);国家重点实验室开放基金项目(No.NEL⁃SRT201708);山西省软科学项目(No.2018041004⁃6);太原师范学院校级1331创新团队项目(No.院办字[2017]34号)。

摘　　要：不同模型对土壤污染物空间分布预测精度具有重要影响,针对现有方法不能较好模拟土壤污染物较强的空间变异特征以及缺乏对影响污染物空间分布的关键环境因子识别,本研究基于随机森林(RF)模型,通过融合多源环境要素,开展了某冶炼厂周边农田土壤砷含量空间分布预测研究,并与反距离加权(IDW)和逐步线性回归模型(STEPREG)相比较.结果表明,研究区农田土壤砷污染范围较广,污染严重区域主要分布在研究区南部,3种模型模拟的砷污染空间分布虽总体趋势相似,但局部区域差异明显,IDW和STEPREG模型不能很好地反映研究区土壤污染的强空间变异特征,RF模型模拟结果较好的表达局部高污染区域的细部变化.不同环境要素对农田土壤砷含量空间分布影响的重要性不同,研究区环境变量和地形变量是影响土壤砷含量空间分布的关键环境因子.交叉验证结果表明,RF模型相对IDW和STEPREG模型具有最小的均方根误差(RMSE)、平均绝对误差(MAE)、平均误差(ME)和最大的R^2,RF模型的RMSE、MAE、ME较IDW模型分别降低了10.8%、5.5%和88.1%,较STEPREG模型分别降低了17.8%、18.4%和94.7%,表明采用RF模型对研究区农田土壤砷含量预测精度最高,取得了最优的预测效果.本研究结果能够为土壤重金属污染空间分布制图提供方法学参考.Different models have important effects on the prediction accuracy of spatial distribution of soil pollutants. The existed methods have limitations in simulating spatial variation and distribution of soil pollutants and identifying the key environmental factors. With the random forest(RF) model of integrating multi-source environmental factors, the spatial distribution of arsenic content was studied in farmland soil around a smelter, and compared with the inverse distance weighted(IDW) and stepwise regression(STEPREG) models. The results showed that arsenic pollution was widely distributed and the south areas were seriously polluted in farmland soil. The overall spatial distribution of arsenic pollution simulated by the three models was similar, but the IDW model and STEPREG model showed less spatial variation of soil pollution, while RF model simulated better spatial variability of heavy arsenic polluted area. The environmental factors were of different influences on the spatial distribution of arsenic content. Environmental and topographic variables in the study area were the key environmental factors influencing the spatial distribution of arsenic in soil. The cross-over studies have shown that the RF model has the smallest RMSE, MAE, ME and the largest R^2. Compared to the IDW and STEPREG models, the RMSE, MAE and ME of the RF model decreased by 10.8%~94.7%, respectively. Thus, the RF model has more accuracy for predicting spatial distribution of arsenic content in this study, providing the reference of cartographic methodology.

关 键 词：污染评价 随机森林 土壤砷 多源环境数据 

分 类 号：X53[环境科学与工程—环境工程]