The Study of Comparative Among Time Series Models Applied to Monthly Rainfall Forecasting: A Case Study of Ilam Province, Iran

The aim of this study is to compare the performance of different time series models for the purpose of modeling as well as monthly rainfall forecasting in Ilam province. In this regard, monthly rainfall data were collected from 8 rainfall station from 1993 to 2011. After preparation of data, R software was used to predict the height of rainfall in 8 rainfall station (2004 to 2011) using monthly height of rainfall data (1993 to 2004). In this study, 5 models (AR, MA, ARMA, ARIMA, and SARIMA) with 12 different structures were examined. Due to the fact that the factors of trend, seasonal and jump are deterministic components in nature, there would be no need for the insertion of modeling. However, the modeling of random component is considered to be highly essential in modeling and water resource management. Therefore, the time series of rainfall was decomposed and the random component modeled accordingly. The results indicates that in 37.5% of data ARMA(1,2), in 25% of data ARMA(2,1) and in 12.5% of data for each of the models of AR(1), ARIMA(1,1,2) and SARIMA(1,1,1)(1,1,1)[12] the best performance occurs in monthly rainfall forecasting. In other words, any given change in data might dramatically trigger the volatility in best time series. So, it is extremely important, first, to assess the all-time series models for any area as well as any hydrological parameters and then, to select the best model for the purpose in question.