Predicting removal efficiency of organic pollutants by soil vapor extraction based on an optimized machine learning method

Shuai Zhang; Jiating Zhao; Lizhong Zhu

doi:10.1016/j.scitotenv.2024.172438

Predicting removal efficiency of organic pollutants by soil vapor extraction based on an optimized machine learning method

Sci Total Environ. 2024 Jun 1:927:172438. doi: 10.1016/j.scitotenv.2024.172438. Epub 2024 Apr 12.

Authors

Shuai Zhang¹, Jiating Zhao¹, Lizhong Zhu²

Affiliations

¹ College of Environmental and Resource Sciences, Zhejiang University, Hangzhou, Zhejiang 310058, China; Zhejiang Provincial Key Laboratory of Organic Pollution Process and Control, Hangzhou, Zhejiang 310058, China.
² College of Environmental and Resource Sciences, Zhejiang University, Hangzhou, Zhejiang 310058, China; Zhejiang Provincial Key Laboratory of Organic Pollution Process and Control, Hangzhou, Zhejiang 310058, China. Electronic address: zlz@zju.edu.cn.

PMID: 38614354
DOI: 10.1016/j.scitotenv.2024.172438

Abstract

Soil vapor extraction (SVE) was a cost-effective technology for remediating volatile and semi-volatile organic contaminated soils. Many factors, including SVE parameters, soil properties, and contaminant characteristics, significantly influenced the remediation efficiency of SVE. The optimal conditions for organic pollutants removal efficiency were site-specific and varied among studies. Therefore, a generalized model was needed to predict the remediation efficiency of SVE in organic contaminated soils. This study employed machine learning to predict the removal efficiency of organic pollutants by SVE. The model's development was based on a trainset, and its predictive capabilities were evaluated using a testset. An XGBoost (XGB) model was derived from literature data (R² = 0.9728). Time, pollutant type, and temperature were identified as the three most important features affecting SVE remediation efficiency. The accuracy (R² = 0.9799) and universality of the model were enhanced through an optimization scheme. The developed XGB model demonstrated the ability to predict the removal efficiency of organic pollutants by considering all collected influential factors. The mechanism of multi-factor interaction on remediation efficiency was clarified. Overall, this study would contribute to evaluating the remediation potential of SVE for specific organic contaminated soils, aiding in maximizing the removal efficiency of organic pollutants under optimal conditions.

Keywords: Machine learning models; Organic pollutants; Soil remediation; Soil vapor extraction.