Ecological changes resulting from climate conditions can severely affect human societies especially in the area of economy and safety. Climate catastrophes may cause social and economic tension. Forecasting such changes accurately can help the government to control the disasters and to achieve possible benefits (such as water supply in flood). Weather forecasting is the application of science and technology to predict the state of the atmosphere for a given location. Rate of raining is a very important factor in weather forecasting. Different forms of weather forecasting models represent different stochastic processes. Three broad classes of time series modeling in practice are the autoregressive (AR) models, the integrated (I) models, and the moving average (MA) models. These models represent the linear dependence on previous observations. Cyclic variation known as periodic fluctuation or seasonality (S) might be dealt with in time series analysis by using a sinusoidal model. A less completely regular cyclic variation might be considered by using a special form of an auto regressive integrated moving average.
In this paper, a hybrid approach based on seasonal auto regressive integrated moving average (SARIMA) method and Locally Linear Model Tree (LoLiMoT) is proposed for forecasting rate of raining. A neural network based on local linear models weighted constructed by a tree algorithm is applied in this research. Training of this network is divided into a structure and a parameter optimization part. A recursive least-squares algorithm is used for training the network since the network is linear in its parameters. A two phase model is developed based on data gathered in Zabol Synoptic Station from 1939 to 2011. In the first phase, the SARIMA model is implemented to predict the raining rate. In the second step neural network based on locally linear model tree is applied to residuals to improve the prediction result. Finally, the proposed model is compared to Sin-Cos model; Result obtained confirm the efficiency of this approach as a practical tool for forecasting the rate of raining.