Developing a New Class-based Probabilistic Hybrid Model for Monthly Precipitation Forecasting

High accuracy forecasting of monthly precipitation is one of the major challenges in hydrology and meteorology and is of great importance in water resources planning. In the current research a Class-Based Probabilistic Hybrid Model (CPHM) has been developed on the basis of a hybrid of classification methods and probabilistic kernel functions. Using this method, monthly precipitation (model output) can be forecasted more accurately for all months of a season according to seasonal precipitation (model input). The superiorities of this model over conventional monthly rainfall forecasting methods are on the one hand, its capability for monthly precipitation forecasting for a season such as autumn in Iran the previous months of which in summer have no precipitation, and on the other hand, the simultaneous prediction of precipitation for all months of a season which is valuable in terms of water resources management. In order to evaluate this model, it was applied to forecast autumnal monthly precipitation for Karkheh basin which includes Khuzestan fertile plain and its efficiency was compared to an optimized structural ANN model. Results revealed a high performance for the developed CPHM model while it was also superior to ANN model for its precipitation forecasts for all three months of autumn. The average accuracy of the model resulted from validation phase for three autumn months based on Nash-Sutcliff (NSE), Root Mean Square Error (RMSE), and Pearson correlation coefficient (PCC) indices were 0.7, 12, and 0.86, respectively.