A stochastic exposure model integrating random forest and agent-based approaches: Evaluation for PM2.5 in Jiangsu, China

This research proposes an Activity Pattern embedded Air Pollution Exposure Model (AP2EM), based on survey data of when, where, and how people spend their time and indoor/outdoor ratios for microenvironments. AP2EM integrates random forest and agent-based approaches to simulate the stochastic exposure to outdoor fine particulate matter (PM2.5) along with indoor and in-vehicle PM2.5 of outdoor origin. The R-2 of the linear regression between the model's calculations and personal measurement was 0.65, which was more accurate than the commonly-used aggregated exposure (AE) model and the outdoor exposure (OE) model. The population weighted PM2.5 exposure estimated by the AP2EM was 36.7 mu g/m(3) in Jiangsu, China, during 2014-2017. The OE model overestimated exposure by 54.0%, and the AE model underestimated exposure by 6.5%. These misestimate reflect ignorance of traditional studies on effects posed from time spent indoors (-85%) and doing low respiratory rate activities (-93%), problems of biased sampling, and neglecting low probability events. The proposed AP2EM treats activity patterns of individuals as chains and uses stochastic estimates to model activity choices, providing a more comprehensive understanding of human activity and exposure characteristics. Overall, the AP2EM is applicable for other air pollutants in different regions and benefits China's air pollution control policy designs.

Automated summary

This research proposes an Activity Pattern embedded Air Pollution Exposure Model (AP2EM) that accurately estimates people's exposure to outdoor fine particulate matter (PM2.5) by integrating survey data and stochastic estimates. The AP2EM outperforms commonly-used models and reveals the influence of factors often neglected in other studies on exposure estimation.