1Department of Forestry & Management of the Environment & Natural Resources, Democritus University of Thrace, 193 Padazidou st., 68200, Nea Orestiada, Greece
2Department of Production Engineering & Management, School of Engineering, Democritus University of Thrace, University Library Building, 67100, Xanthi, Greece
3Department of Physics, Laboratory of Meteorology, University of Ioannina, 45110, Ioannina, Greece
This manuscript presents the design and the development of an Artificial Neural Network (ANN) model that estimates the surface ozone concentrations when the values of other pollutant and meteorological parameters are already known for the case of Athens suburb Lykovryssi. This is the characteristic suburb which is very close to a major city and at the same time far from the city center. Thus the developed ANN can constitute a potential model for all suburbs with similar characteristics. The large amount of data records used as input and the good testing results show the generalization ability of the developed ANN. The Training has been performed for different numbers of iteration cycles in order to avoid over-Training. Principal component analysis and stepwise regression analysis were performed for the same area in the past. Comparing the results of the statistical analysis to the output of the designed optimal ANN, we have discovered that the Neural Network performs more accurately.
Artificial Neural Networks, Pollution of the Atmosphere, Tropospheric Ozone
