Statistical methods for air quality model calibration and validation in urban areas

HA Statistics

Sim, Lauren Holmes ; Scott, Marian

University:University of Glasgow

Department:School of Mathematics and Statistics

关键词: HA Statistics;   

Others  :  http://theses.gla.ac.uk/7141/1/2016SimMSc.pdf

来源: University of Glasgow

【 摘 要 】

It is thought that air quality modelling is vital, this is because of the lack of monitoring sites and diffusion tubes in cities making it difficult to see spatially how air pollution is behaving. In this thesis, the Atmospheric Dispersion Modelling System (ADMS)-Urban is focussed on and it is of interest to see how well the modelled nitrogen dioxide (NO2) predictions and the monitoring site NO2 data are calibrated in Aberdeen over the year 2012. There are only six monitoring stations in Aberdeen and it will be highlighted in this thesis how close in space these monitoring stations are. To evaluate how comparable the modelled and monitoring data are, methods such as Deming Regression, Extreme Value Analysis, Functional Principal Components Analysis (FPCA) and Clustering and Functional Regression will be investigated. FPCA and clustering and Deming Regression highlight that the modelled and monitoring data appear not very well calibrated at Wellington Road, however these data are reasonably well calibrated at the other monitoring sites. FPCA and Clustering indicate that the roads appear to dominate and be the main cause of concern in Aberdeen in terms of air pollution. All methods suggest that between April 9th and July 18th the model and monitoring data appear not be well calibrated and this could be further explored to examine the potential causes. These analyses have identified how the relationships between the ADMS-Urban model output and the observed data may vary over time and space.