Ecological risk assessment of oilfield soil through the use of machine learning combining with spatial interaction effects

Shihan Wang; Mingxia Zheng; Yushan Tian; Hongyu Ding; Lina Yan; Beidou Xi; Yuanyuan Sun; Jing Su

doi:10.1016/j.ecoenv.2025.118527

Ecological risk assessment of oilfield soil through the use of machine learning combining with spatial interaction effects

Ecotoxicol Environ Saf. 2025 Jun 21:302:118527. doi: 10.1016/j.ecoenv.2025.118527. Online ahead of print.

Authors

Shihan Wang¹, Mingxia Zheng², Yushan Tian³, Hongyu Ding⁴, Lina Yan⁵, Beidou Xi⁶, Yuanyuan Sun⁷, Jing Su⁸

Affiliations

¹ State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: wangshihan22@mails.ucas.ac.cn.
² State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: zhengmx@craes.org.cn.
³ State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: truoshan1998@163.com.
⁴ State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: dinghy524@126.com.
⁵ State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: ylina_yan@foxmail.com.
⁶ State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: xibd@craes.org.cn.
⁷ State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: sunyy@craes.org.cn.
⁸ State Key Laboratory of Environmental Criteria and Risk Assessment, and State Environmental Protection Key Laboratory of Simulation and Control of Groundwater Pollution, Chinese Research Academy of Eiknvironmental Sciences, Beijing 100012, China. Electronic address: sujing@craes.org.cn.

PMID: 40544546
DOI: 10.1016/j.ecoenv.2025.118527

Abstract

With the intensification of oil extraction activities, total petroleum hydrocarbons (TPHs) and toxic elements contamination in soil around oil wells have become severe environmental problems. This paper proposed a novel method based on machine learning (ML) and remote sensing (RS) to predict concentrations of TPHs and toxic elements in the soil around 1252 oil wells for pollution investigation and monitoring in the Huabei Oilfield in China. RS can obtain variables that are closely related to soil pollution, such as the fractional vegetation cover (FVC), soil type and topographic factors. which can help reveal the pollution driving mechanism combining with ML. ML was used to predict pollutant concentrations, with predictors such as the distribution of oilfield capacity facilities interpreted by RS imagery. Combining RS and ML helps uncover pollution driving mechanisms. The potential ecological risk index (RI) method was utilized to assess ecological risks, and spatial autocorrelation analysis was conducted to determine the spatial distribution characteristics of the pollutants. The results indicated that the Gradient Boosting Machine (GBM) model exhibited strong performance in predicting concentrations of TPHs (R²=0.7730), As (R²=0.8614), Pb (R²=0.8678), Ni (R²=0.7539), Cd (R²=0.7447), and Hg (R²=0.6270) in soil. Oil extraction activities, land use, and soil properties are the dominant factors influencing the accumulation of TPHs and toxic elements. The ecological risk assessment combined with bivariate LISA mapping identified priority areas for risk control, of which 22.73 % were with no risk, 18.18 % were with combined TPHs and toxic elements risk, 8.67 % were with toxic elements risk, 6.61 % were with TPHs risk, and 26.04 % were with uncertain risk. The results can be applied to provide technical support for soil risk management and industrial site planning in the oilfield and surrounding area.

Keywords: Automated machine learning; Ecological risk assessment; Oilfield; Spatial bivariate correlation analysis; TPHs; Toxic elements.