Deterministic models and Neural Nets: a successful methodology for the air dispersion models

In this work is presented the development of an integrated model composed by a dispersion model and a neural net (NN). The neural net model uses the concentrations predicted by an air dispersion model (ADMD) as input variables of the net. This methodology was tested in the case of a releases from an elevated emission source using the urban data set of the Indianapolis field study. We also compare the performance of the dispersion model alone (ADMD) with the integrated model (ADMD-NN). The comparison shows an improvement of all the main statistical index when the neural network is added downstream to the dispersion model. Tests conducted on the integrated models reveal the system is able to reproduce the expected behaviour of pollutant concentration with the downwind distance and stability of the atmosphere.