Estimating the geographical patterns and health risks associated with PM2.5-bound heavy metals to guide PM2.5 control targets in China based on machine-learning algorithms

PM2.5 is the main component of haze, and PM2.5-bound heavy metals (PBHMs) can induce various toxic effects via inhalation. However, comprehensive macroanalyses on large scales are still lacking. In this study, we compiled a substantial dataset consisting of the concentrations of eight PBHMs, including As, Cd, Cr, Cu, Mn, Ni, Pb and Zn, across different cities in China. To improve prediction accuracy, we enhanced the traditional land-use regression (LUR) model by incorporating emission source-related variables and employing the best-fitted machine-learning algorithm, which was applied to predict PBHM concentrations, analyze geographical patterns and assess the health risks associated with metals under different PM2.5 control targets. Our model exhibited excellent performance in predicting the concentrations of PBHMs, with predicted values closely matching measured values. Noncarcinogenic risks exist in 99.4% of the estimated regions, and the carcinogenic risks in all studied regions of the country are within an acceptable range (1 × 10−5–1 × 10−6). In densely populated areas such as Henan, Shandong, and Sichuan, it is imperative to control the concentration of PBHMs to reduce the number of patients with cancer. Controlling PM2.5 effectively decreases both carcinogenic and noncarcinogenic health risks associated with PBHMs, but still exceed acceptable risk level, suggesting that other important emission sources should be given attention. [Display omitted] •The LUR model is improved with emission variables and machine-learning algorithms.•Emission-related variables contribute more than land cover and climate variables.•The reduction in PBHM health risk due to PM2.5 control alone is limited.