An analysis of linear and nonlinear time series for rainfall amount of some provinces

This study comprises the obtaining and comparison of the forecasting for climate time series via Box-Jenkins method - a linear time series method - and via artificial neural networks for non-linear time series. This comparison has been made for the time series of monthly rainfall amount between 1960 and 2012 in Ankara, İstanbul, Antalya, Erzurum and Samsun. The results, carried out according to the Box-Jenkins models and artificial neural networks, have been compared based on 12-step ahead forecasting. In artificial neural network model of each province the number of neurons are the same for input layer and, correspondingly, the same for output layer. The neural networks in which the different number of neurons and different activation functions for hidden layer were tried have been put into account. According to the analysis results, it can be deducted that the rainfall amount series show non-linear structure; as a result, when compared to Box-Jenkins, artificial neural networks have more noticable forecasting values.