24-Hours Ahead Forecasting of PM10 Concentrations Using Artificial Neural Networks in the Greater Athens Area, Greece

Abstract

The study of atmospheric concentration levels at a local scale is one of the most important topics in environmental sciences. Multivariate analysis, fuzzy logic and neural networks have been introduced in forecasting procedures in order to elaborate operational techniques for level characterization of specific atmospheric pollutants at different spatial and temporal scales. Particularly, procedures based on artificial neural networks (ANNs) have been applied with success to forecast concentration levels of PM10, CO and O3. The present study deals with the development and application of ANN models as a tool to forecast daily concentration levels of PM10 in five different regions within the greater Athens area (GAA). Modeling was based on mean daily PM10 concentration, the maximum hourly NO2 concentration, air temperature, relative humidity, wind speed and the mode daily value of wind direction from five different monitoring stations for the period 2001–2005. Model performance showed that the ANN models could successfully forecast the risk of daily PM10 concentration levels exceeding certain thresholds. In addition, despite the limitations of the models, the results of the study demonstrated that ANN models, when adequately trained, could have a high applicability to predict the PM10 daily concentration 1 day ahead within the GAA.