Air pollution in terms of pollutants

researching the air pollution in terms of pollutants, their formation and movements as well as the regulations with which the air in Glasgow has to comply.


In my first year, I focused on researching the air pollution in terms of pollutants, their formation and movements as well as the regulations with which the air in Glasgow has to comply. In my second year, I worked more in depth with the NO2 concentrations for 2015 monitoring data (8 stations across Glasgow, http://www.scottishairquality.co.uk/air-quality/monitoring) provided by SEPA (my industrial partner). Initial analysis of the data (hourly nitrogen dioxide, NO2, measurements for 2015) was performed and it was checked with which EU regulations the air in Glasgow does not comply. The initial analysis includes using ACF and PACF plots to explore the temporal correlations at each station for the hourly observations. Furthermore, we fitted temporal and spatio-temporal models using the D-STEM software in MATLAB to check the spatial and time correlations between the stations. The temporal model and univariate (only NO2 as a response) spatio-temporal model were almost identical, whereas the multivariate (NO2 and NOx as responses) spatio-temporal model performed better in terms of predictive power. Nonetheless, all the models struggled by underestimating or overestimating the predictions for the hourly NO2 concentrations for a year.

However, the main focus of my work is on creating an emulator for the ADMS-Urban (http://www.cerc.co.uk/environmental-software/ADMS-Urban-model.html) simulation model used by SEPA. The ADMS-Urban model is used to simulate the air pollution (hourly measurements for different set of pollutants for different time periods) in the city under different set of conditions (meteorological data, emissions, traffic). In the thesis, I focus on the hourly NO2 pollutant concentrations for a year under different meteorological and emissions scenarios. The idea to use an emulator a more simplistic version of the simulation and also to be able to provide uncertainty measure for the predictions. An emulator is a statistical model used for predicting the output from a simulation model for untried sets of inputs. The emulator does not eliminate the use of the simulation model, just reduces the number of runs for it. Our work is extending the work of Finazzi et. al. (submitted), who use an emulator for the ADMS-Urban model predictions for Aberdeen. However, the emulator developed by Finazzi is for the annual average measure for the air pollution for every station. My work aims to be able to extend this work to the hourly scale for each station.