Forecasting meteorological time series using soft computing methods: an empirical study

The interest of researchers in different fields of science towards modern soft computing data driven methods for time series forecasting has grown in recent years. Modeling and forecasting hydrometeorological variables is an important step in understanding climate change. The application of modern methods instead of traditional statistical techniques has lead to great improvement in past studies on meteorological time series. In this paper, we employ Support Vector Regression (SVR) and automatic model induction by means of Adaptive Gene Expression Programming (AdaGEP) for modeling and short term forecasting of real world hydrometeorological time series. The investigated time series datasets cover annual, respectively monthly data, on temperature and precipitation, measured at several meteorological stations in the Black Sea region. Two performance measures were used to assess the efficiency of the models obtained for forecasting, alongside statistical testing of the goodness of fit via the Kolmogorov-Smirnov test. Based on the results of rigourous experiments, we conclude that the models obtained by the AdaGEP algorithm are more competent in forecasting the time series considered in this paper than the models produced with the SVR algorithm.