[Prediction of Heavy Metal Concentrations in PM2.5 in the Agricultural Area of Yangtze River Delta Region Based on Machine Learning].

Abstract

Heavy metals in PM_2.5 can considerably impact air quality, human health, and the ecological environment. However, studies on heavy metals in PM_2.5 in agricultural areas are relatively limited. In this study, observational data on heavy metal concentrations in PM_2.5 in the Yangtze River Delta Region from 2000 to 2020 were collected. Three machine learning-based prediction models for heavy metal concentrations in PM_2.5 were constructed to predict and analyze the regional pollution characteristics of six heavy metal elements （Pb, Cu, As, Cd, Zn, and Ti） in PM_2.5 in agricultural areas of the Yangtze River Delta. The results showed that none of the three machine learning models, random forest （RF）, support vector machine （SVM）, or gradient boosting machine （GBM）, exhibited good prediction performance when individually predicting the concentrations of heavy metal elements in PM_2.5 （R² < 0.66 in nearly half of the models）. However, the performance improved significantly after integrating the three models with weighted averaging （R² > 0.66 in all models）, which achieved quantitative prediction capabilities for the concentrations of the six metal elements （RPD > 1.4）. The prediction results for the concentrations of heavy metals in PM_2.5 in agricultural areas of the Yangtze River Delta revealed that the average mass concentrations （ng·m^-3） of the six heavy metal elements were in the order of Zn > Pb > Cu/Ti > As > Cd, but significant differences were observed in their spatial-temporal distributions. The concentrations of Pb, Cd, As, and Zn in PM_2.5 decreased from 2015 to 2017, while the concentrations of Cu and Ti did not show significant temporal changes. Spatially, the concentrations of Pb, Cu, and Ti in PM_2.5 were higher in the northern areas of the Yangtze River Delta Region but lower in the south. The concentrations of As and Cd were higher in the mountainous areas of northern Anhui and western Zhejiang, while Zn concentrations were relatively high across all agricultural areas. These results provide an effective method for predicting regional heavy metal concentrations in atmospheric particulate matter and offer a reference basis for understanding the characteristics of atmospheric particulate matter pollution and regional pollution reduction efforts in agricultural areas of the Yangtze River Delta.