Machine Learning-Enhanced Prediction for Soil-to-Air VOC Emission and Environmental Impact Pertaining Contaminated Fractured Aquifers.

How to scientifically and efficiently quantify the impact and hazards of volatile organic compounds (VOCs) pollution and volatilization from complex groundwater systems on surface air environments is a critical environmental issue. This paper employed an integrated modeling approach, incorporating numerical simulations, statistical analyses, and machine learning to address this issue. We comprehensively accounted for the different driving mechanisms, along with the various migration and transformation processes of groundwater VOCs. This investigation identified 11 key factors influencing surface pollutant flux. The data-enhanced statistical surrogate models and sampling-fusion-based support vector machine (SVM) surrogate models were established for appropriate generic modeling applications in which the high computation burden and difficulty could be avoided of the complicated numerical modeling. Those models would enable accurate prediction of surface fluxes and reliable classification of environmental risks. Notably, the pollutant fluxes through the soil-air interface over a short period could be sufficient to cause slow-airflow space air concentrations to exceed acceptable levels. Particularly, the established generic statistical surrogate models and SVM surrogate models have significant implications in efficiently and rapidly assessing the VOCs surface fluxes and environmental risk with meaningful quantified uncertainties for specific site conditions.