Assessment of dynamic linear and non-linear models on rainfall variations predicting of Iran

Abstract

The main research aims to detecting the linear and nonlinear variability modeling in analyzing the variability patterns of rainfall series. For rainfall linear and nonlinear variability modeling, the ARIMA models and ARCH family models has been used for predicting the monthly and annual rainfall series extracted from IRIMO during 1975-2014 within 140 stations in Iran. Several ARIMA and ARCH (six models) models have been used and their validity has been confirmed by evaluating different accuracy indicators, using the hybrid model for the variability modeling. The analysis of ARIMA and GARCH selective models indicates existence of random and non-random in the rainfall time series. The combination model of (1, 0, 0) and GARCH (1, 1) is applied for the estimate and prediction of monthly rainfall. With careful valuation of the hybrid model, the ARIMA (1,0,0) and GARCH(1,1)  is recognized as the significant acceptable model by determines of different accuracy indicators similar to mean squared error (77025.34); root mean squared error (277.53); mean absolute error (167.68); mean absolute percentage error (79.68); and Theil’s U coefficient (0.365). However, the results showed that hybrid model, as a variability model is more efficient in forecasting the rainfall variability and underlying this model can be used as a variability forecast model and chaos phenomena in Iran. In addition, a nonlinear model (ARCH family, especially GARCH1, 1) provides a quantitative-analytical method to distinguish between a particular random and non-random model for rainfall variability in Iran.

Keywords: Linear models, Non-linear models, variability severity and rainfall variability.