Artificial neural networks applications for total ozone time series

Abstract

One of the main problems that arises when dealing with time series is the existence of missing values which have to be completed previously to every statistical treatment. Here we present several models based on neural networks (NNs) to fill the missing periods of data within a total ozone (TO) time series. These non linear models have been compared with linear techniques and better results are obtained by using the non linear ones. A neural network scheme suitable for TO monthly values prediction is also presented. © Springer-Verlag Berlin Heidelberg 2003.