Neural network model for maximum ozone concentration prediction

Abstract

A neural network dynamic model was used for predicting maximum ozone (O3) concentration at Santiago de Chile. Learning and test data were collected during summer and springtime periods of 1990, 1992 and 1993. A neural network having O3 t, Tt+1 (maximum air temperature) and Tt as inputs for predicting O3 t+1 was chosen because of its low test error. This neural network model greatly reduces the error coming from a pure persistence model when applied to the generalization set of data (1994). Long-term predictions results confirm the good concordance obtained between the observed and forecasted values thus showing the adequacy of neural networks to model the dynamics of this complex environmental phenomena.