Evaluation of Different Data Stationary Methods Efficiency Using ARIMA Family Models

Forecasting rainfall depth is very important in water resources management. ARIMA family time series models have a widespread application in this context. The main objective of this study was to predict monthly rainfall using the best time series stationary method and the most suitable ARIMA family model. In this study, Ardabil synoptic station’s data was used. In the first step, the trend and seasonality terms of monthly rainfall data from 1990 to 2016 were removed using different methods and in the second step the efficiency of different ARIMA models for predicting monthly rainfall was investigated. Results showed that the stationary method using 12 period centered moving average and seasonal average in order to remove trend and seasonal variation, respectively, is the best stationary method with the highest correlation coefficient (r=0.8). Also, the SARIMA (1,0,1) (0,0,1)12 model with the highest correlation coefficient (r=0.8) and the lowest Akaike criterion (AIC=191.74) is the best prediction model for monthly rainfall at the studied station. Finally, the monthly rainfall of the next 3 years (2017-2019) forecasted using the optimized stationary method and the selected model. Results showed that the rainfall trends of Ardabil synoptic station will be decreased in the next three years.