Statistical post-processing of the air pollution model SIMAIR

Abstract: Air pollution is a serious problem today, both from an environmental and from a health point of view. Especially in cities, particles smaller than 10 µm in aerodynamic diameter (PM10) can reach high concentrations. These particles are dangerous, even at low concentrations, since they are small enough to enter the lungs. In order to estimate the concentration of air pollutants, different measurements and air pollution models can be used. A combination of model data and measurements allows for the assessment of air pollution concentration over larger areas with a lower degree of uncertainty. Statistical post-processing is one approach to combining model data and measurements. SIMAIR is a Swedish system of models that uses meteorological data, emission data and dispersion models on different geographical scales to calculate the concentration of air pollutants on regional, urban and local levels. The aim of this Master’s Thesis is to study different statistical post-processing methods and to examine their adequacy with regards to dealing with air quality models. One method, Support Vector Regression, is implemented and analysed based on the results from the SIMAIR model. The compound that is examined is PM10. The statistical post-processing method is developed based on data from Hornsgatan in Stockholm from the year 2007 to 2009. This method is then validated using data from Västra Esplanaden in Umeå and Gårda in Gothenburg. The results are promising for all three sites; improvements are seen for almost all statistical indicators used to examine model performance.