Prediction of temperature variations using artificial neural networks and ARIMA model

Abstract

Hydrological processes modelling is difficult but an important affair. These processes are caused by the parameters' interaction effects for which the large number of parameters and also their interactions lead to a complicated system. To design water resource systems, it is needed to predict the hydrological factors such as temperature, river's flow, quantity of rainfall and air humidity based on the historical data in configuration of time and place series. Time series analysis is a mostly used prediction method, which is based on the stationarity and linearity assumptions. Artificial neural network is another prediction method which has shown its performance and reliability for the time series prediction. In this paper, the efficiency of two prediction methods, namely auto regressive integrated moving average (ARIMA) and artificial neural networks (ANNs) are compared with each other based on the prediction of temperature variations of Gorgan city in Iran, between 1970 to 2010. The results indicated that the ARIMA method provides better estimates than the ANN method.