Application of machine-learning techniques for the characterization of urban micro-environments to assess spatial variability of air pollutants

Several epidemiological studies have highlighted the link between short and long term exposure to air pollutants and their impact on the public health. Increase the spatial resolution of the concentration of air pollutants can help to improve the results of exposure modelling and to reduce their impact. The aim of this work is to characterize the spatial variability of air pollutants in urban context identifying the main factors that driven the concentration of air pollutants. This is done by performing GIS processing to split the study area with a homogeneous micro-climate and micro-environment conditions, and comparing the land classification with the experimental data coming from the official monitoring stations (OMS) of Edinburgh. The authors identify 3 main factors that can influence the micro-environment conditions for air pollutants that are the land use, the urban canyons and roads classification and proximity. The land use has been extracted from aerial photography splitting the image in homogeneous areas (segment) and combining three different machine learning techniques to classify the segments in the 4 selected categories. The urban canyons are assess computing the skyview factor and the aspect ratio of the building as suggest by Eeftens et al. 2013. The roads are classified in 5 different classes, that are based on the Open Street Map classification, to take into account the different level of traffic emissions for each class. These three main factors are compared with the measurements collected by the OMS to verify which factors driven the spatial variability of the air pollutants. The presented research is still a work in progress and not the whole process has been completed yet. Concerning the segment classification the accuracy is 96%, using a subset of 100 segments to train over 160 verified segments. Combining these different factor maps with the OMS measurements we expect to identify common spatial pattern of air pollutants.