Techniques for monitoring city air quality
Μέθοδοι επίβλεψης της ποιότητας του αέρα στο αστικό περιβάλλον
In this thesis, alarming events of air pollution in the Attica region are being investigated, using ARIMA models and statistical control charts. The research focused on estimating, fitting and forecasting suitable ARIMA models by identifying specific patterns in the time series of pollutants and by using statistical process control techniques to detect possible harmful exceedances. The main objective of this study was to analyze and plot the residuals (forecast errors) of air pollutants, taking nitrogen dioxide (NO2) and ozone (O3) daily mean concentrations from 2010-2013 and from 8 different stations located throughout Attica as a case study. The percentage of missing data for each annual time series was around 10% on average, thus multiple imputation techniques were used as an initial step. Corrective actions were taken to monitor such autocorrelated processes including differencing the time series in order to achieve stationarity and remove trend from the data. It was proved that NO2 and O3 time series were correlated and only the O3 time series showed regular peaks (seasonality). After implementing ARIMA models and checking the residuals for correlation and normality, one – step ahead forecasts were produced for NO2 and O3 concentrations. Forecast errors were studied and plotted in a ̅ chart / MR chart for individual data as an aid to detect outliers. We were mainly interested in the large positive differences between the observed and predicted values (positive forecast errors). Statistical analysis showed that successive large ―disturbances‖ were only occurred for O3 concentrations at Liosia station, indicating an event that we should pay special attention to, while in all other stations the outliers were significantly low in number, indicating a well – estimated model, close enough to the actual concentrations of 2013.