Assessing neighborhood variations in ozone and PM2.5 concentrations using decision tree method

Abstract Typical air pollution events involving ozone (O3) and PM2.5 occurred frequently in China, while the fine-scale pollution variation, especially at a neighborhood level (2 km*2 km), is complex and still not clear. To assess how urban form and meteorology influence neighborhood air pollution distribution, this study took the Minhang district in Shanghai, as experimental cases, and performed a neighborhood-scale investigation on O3 and PM2.5 by using mobile measurements. Both land-use regression model and decision tree model were used to examine the relationship between air pollutant concentration and influenced variables. As the decision tree model captured the linear and non-linear relationship between variables, it was demonstrated that explained more variations of O3 and PM2.5 concentrations than the LUR model. The results also showed that O3 concentrations were mainly affected by meteorological factors while PM2.5 concentrations were more heavily determined by background level and residential area. Both O3 and PM2.5 showed a significant correlation with air temperature, traffic volume, building height, and green space. Interestingly, green spaces were negatively correlated with the PM2.5 variations, which was almost the opposite to that of O3. With the superiority to the discrete observation, the decision tree model based concentration surfaces clearly revealed the heterogeneity of O3 and PM2.5 distributions. This study not only preliminarily identifies the impacts of land-use type and meteorological factors on the spatial patterns of O3 and PM2.5, but also provides a possible alternative method for assessing the neighborhood air pollution in the future.