An investigation of the effectiveness of advanced modeling tools on the forecasting of daily PM10 values in the Greater Athens area

Abstract

The present paper discusses the effectiveness of advanced modeling tools as a means of forecasting daily PM10 values, presenting as a case study the Greater Athens Area. The effectiveness of the described approaches is judged against mature forecasting approaches (e.g. linear regression and neural networks) in terms of absolute prediction error metrics and the confusion matrix, which shows the ability to correctly predict exceedances of the regulatory limit. Finally, a new approach is presented that incorporates the short-term dynamics of the PM10 time series into the forecasting model as explanatory variable. © Springer-Verlag Berlin Heidelberg 2009.