Precise spatiotemporal simulation of on-road PM2.5 concentration based on mobile monitoring

As the main air pollution indicator, PM2.5 concentration often comes from monitoring data of fixed environmental monitoring stations and remote sensing image data. The spatial and temporal accuracy is generally insufficient, which makes it difficult to reveal the spatial and temporal distribution of PM2.5 in urban interior at microscale. In this study, using the mobile monitoring method of cycling, the typical working day (November 27, 2017) was selected to collect PM2.5 concentration data of roads in the main urban area of Guangzhou at a time and space granularity of 1 m•s. The machine learning method is utilized to simulate the refined spatiotemporal distribution pattern of on-road PM2.5 during the morning and evening peak hours. The results show that the average spatial range of PM2.5 concentration values close to each other in the morning peak hours is 24 m, which is larger than that in the evening peak hours of 15 m. There was a microscale spatial and temporal heterogeneity of PM2.5 concentration. The fitting degrees of morning and evening peaks' PM2.5 models constructed by Multilayer Perceptron (MLP) reached 0.70 and 0.68, respectively, which is obviously superior to the traditional Ordinary Least Square (OLS) linear regression model. The model reveals that the average concentration of PM2.5 in the whole road network of the main urban area was 30.19 μg/m3 in the morning peak, and reached 44.55 μg/m3 in the evening peak, with the maximum up to 94.82 μg/m3. The spatial distribution characteristics of "high in the west and low in the east" are significant. The refined mapping method of PM2.5 concentration proposed in this paper has a spatial accuracy of 1 m and can better describe the spatial heterogeneity. The method is proved to be feasible and can provide reference for public health travel and targeted pollution prevention. © 2023, Science Press. All right reserved.