Abstract

Recent epidemiological studies have reinforced the link between short and long-term exposure to air pollutants and adverse effects on public health especially over the weaker part of the population, like children and older adults. The creation of simple tools to locate sensible areas as well as of dedicated Spatial Decision Support System (SDSS) to improve the management of pollution risk areas system is strongly advised. The aim of this work is to develop a SDSS methodology, based on easy to find data and usable by decision makers, to assess and reduce the impact of air pollutants in a urban context. To achieve this goals I tested the exploitability of a set of low-cost sensors for outdoor air quality monitoring, I characterized the urban micro-environments and the spatial variability of air pollutants using remote sensing compared to field data and eventually I developed a SDSS to improve the public health designing and comparing different scenarios. The city centre of Edinburgh has been used as study case for the purposed methodology. To test the reliability and applicability of low cost sensors as proxies for remote sensed data, we conducted a measurements campaign to compare the observed data between an official measurements station (OMS) in Trento (Italy) and electrochemical and thick film sensors respectively of Carbon Monoxide (CO) and Ozone ($O_3$). Due to data quality and availability we decided to characterize the urban micro-environments of Edinburgh (Scotland, UK) in eight main classes (water, grass, vegetation, road, car, bus, buildings and shadow) combining the Geographic Object-Based Image Analysis (GEOBIA) with Machine Learning algorithms to process the high resolution (0.25m x 0.25m) RGB aerial ortho-rectified images. This land-use characterization combined with other geographical informations, like the classification of the roads and the urban morphology, were compared with 37 Nitrogen Dioxide (NO2) concentration data, collected using passive tubes during a six week campaign of measurements conducted by the school of Chemistry of the University of Edinburgh. I developed a new open-source GIS python library (PyGRASS), integrated in the stable release of GRASS GIS, to speed-up the prototyping phase and to create and test new GIS tools and methodologies. Different studies on SDSS were carried out to implement procedures and models. Based on these models and data all the factors (land-use, roads and geo-morphological features) were ranked to identify which are driving forces for urban air quality and to help decision makers to develop new policies. The sensor tested in Trento revealed an evident drift in measurement residues for CO, furthermore the measurements were also quite sensitive to external factors such as temperature and humidity. Since these sensors required frequent recalibration in order to obtain reliable results, their use was not as low-cost as expected. The characterization of urban land-use in Edinburgh with GEOBIA and machine learning provided an overall accuracy of 93.71\% with a Cohen's k of 0.916 using a train/test dataset of 9301 objects. The $NO_2$ data confirm the assumption that air concentration is strongly dependent on geographical position and it is strongly influenced by the position of the pollutant's source. Using the results of the tests and remote sensing analysis, I developed an SDSS. Starting from the current situation, I designed three scenarios to assess the effect that different policies and actions could have on improving air quality at on the local and district level. The outcomes of this work can be used to define and compare different scenarios and develop effective policies to reduce the impact of air pollutants in an urban context using simple and easy to find data. The GIS-based tool can help to identify critical areas before deploying sensors and splitting the study area in homogeneous micro-environments clusters. The model is easy to expand following different procedures.