Predicting citywide distribution of air pollution using mobile monitoring and three-dimensional urban structure

Understanding relationships between urban structure patterns and air pollutants is key to sustainable urban planning. In this study, we employ a mobile monitoring method to collect PM2.5 and BC data in the city of Philadelphia, PA during the summer of 2019 and apply the Structure of Urban Landscapes (STURLA) methodology to examine relationships between urban structure and atmospheric pollution. We find that PM2.5 and BC vary by STURLA class, and some classes exhibit significant difference in pollution concentrations. We also find that the proportions in which STURLA components are present throughout the urban landscape can be used to predict the spatial distribution of urban air pollution. Among frequently sampled STURLA classes, gpl (grass, pavement, and low-rise buildings) hosted the highest PM2.5 concentrations on average (16.60 +/- 4.29 mu g/m(3)), while tgbwp (trees, grass, bare soil, water, pavement) hosted the highest BC concentrations (2.31 +/- 1.94 mu g/m(3)). Furthermore, STURLA combined with machine learning modeling was able to correlate PM2.5 (R-2= 0.68, RMSE 2.82 mu g/m(3)) and BC (R-2 = 0.64, RMSE 0.75 mu g/m(3)) concentrations with urban landscape composition and interpolate concentrations throughout the city. These results demonstrate the efficacy of the STURLA methodology in modeling relationships between air pollution and urban structure patterns.