Contamination of organophosphate esters in soil surrounding paint factories: Machine learning-based distribution prediction and risk prioritization assessment.

Organophosphate esters (OPEs) are widely used in paint industry. However, their contamination and risk in paint industry-related areas remain insufficiently characterized. In this study, 33 OPEs, including traditional and novel OPEs, and their diester and hydroxylated transformation products were quantified in 48 surface soil samples collected around paint-related factories in eastern China. The total OPE concentrations ranged from 12.7 to 419 ng/g, and were dominated by chlorinated and novel OPEs. Spatial analysis using Kriging interpolation revealed a distinct point-source distribution pattern, with elevated concentrations in soils adjacent to paint factories reaching 334 to 419 ng/g. Additionally, four machine learning (ML) models were developed to predict OPE distribution in study region, using spatial coordinates, soil properties, and molecular descriptors as predictors. Among them, the tree-based model of gradient boosting regression tree achieved the highest predictive accuracy (testing set, R2 = 0.61, MAE = 0.43, MSE = 0.33), and outperformed Kriging interpolation based on external validation. This demonstrates the feasibility of ML-based approaches for predicting OPE distribution patterns. Furthermore, risk assessment results indicated that novel OPEs tris(2,4-di-tert-butylphenyl) phosphate (AO168=O) and tris(4-nonylphenyl) phosphate (TNPP) posed notably high risks, with risk quotient values of 3.60-1200 and ToxPi prioritization scores of 0.71-0.80. Overall, this study provides new insights and methodological references for understanding the contamination and risks of emerging pollutants associated with the paint industry.