Predicting the Fate and Source of Groundwater PFAS in the Pearl River Delta Region Based on Machine Learning

Extensive investigations into the increasingly severe contamination of perfluoroalkyl and polyfluoroalkyl substances (PFAS) in groundwater are currently causing high costs and long duration. Machine learning provides useful tools for predicting the occurrence of PFAS in groundwater and for tracing pollution sources. It also supports prioritizing investigations and implementing subsequent decisive prevention and control of PFAS pollution. This study combines the advantages of machine learning in predicting the occurrence and traceability of PFAS and employs Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Extreme Gradient Boosting (XGB) to thoroughly evaluate and predict the key factors thereof occurrence and source identification of PFAS in groundwater (n = 52) for the Pearl River Delta region. Findings reveal that groundwater chemical and spatial geographic information successfully predicts the PFAS occurrence, with industrial land, DOC, well depth, and soil saturation being the key factors. The contribution of mixed sources was successfully determined by using PFAS compositional profiles from industrial and agricultural sources. Industrial sources have a higher contribution; only 16.7% of mixed sources are attributed to agricultural sources. This study provides a reliable technical method and data support for a rapid assessment of the contamination status of groundwater PFAS in the Pearl River Delta region.