A novel prediction framework for estimating high spatial resolution near-ground PM2.5 and O3 concentrations at street-level in urban areas

Abstract

National monitoring stations lack spatial coverage for reflecting microscopic changes in air pollution. Several studies have attempted to use locally measured data to develop prediction models to complement national stations. However, the lack of meteorological stations in urban areas makes it challenging to obtain temperature (TEM) and relative humidity (RH) with high spatial resolution; since these are important air pollution predictors, these models cannot be applied to entire urban areas. Here, we propose a new prediction framework that estimates near-ground high spatial resolution PM_2.5 and O₃ concentrations based on short-time and large-scale monitoring and multisource urban data. We conducted a mobile monitoring experiment in Wuhan using electric bicycles to collect PM_2.5 and O₃ concentrations and TEM and RH to train our models. First, we predicted the near-surface TEM and RH via mobile monitoring of the TEM, RH and other built environment data. Second, we used near-surface TEM and RH with other urban big data to predict the PM_2.5 and O₃ concentrations. The results revealed that the estimation performance of the proposed two-stage machine learning prediction models is high, with R² values above 0.95. Satellite top of atmosphere reflectance (TOA), land surface reflectance (LSR) and street view data were incorporated into the new framework to obtain higher-spatial-resolution (50 m) air pollution maps. Our results revealed that TEM and RH varied considerably between the near-surface and meteorological stations. Accurate near-ground TEM and RH are important for predicting near-ground PM_2.5 and O₃ concentrations. Furthermore, TOA and LSR are promising for predicting near-ground PM_2.5 concentrations.