Assessing the Performance of Decision Tree Model in Predicting Precipitation in Kermanshah Synoptic Station

Rainfall is one of the main components of the water balance and its prediction is useful in managing agricultural water supply and managing water resources in reservoirs. Decision tree model as being one of the prediction models, has several functions in rainfall modelling and results in law making. In this study, to evaluate the performance of the decision tree model for the precipitation to be predicted in Kermanshah synoptic stations, and the algorithm CART (Classification and regression tree), being a kind of the regression decision trees, was used to predict the rainfall for the next 30 months. The data used in this study collected from the monthly rainfall stats, evaporation, relative humidity, maximum temperature, average temperature and wind speed in the statistical period of 1970 to 2010. To assess the created trees in this study, different statistical criteria were used. Finally, the results show that in Kermanshah synoptic stations, the regression decision tree is a relatively efficient model in predicting rainfall in which the use of moving average leads to a significant increase in the performance of the model than other modes. And in the case of modification in the range of changes in the input data, it is able to precisely estimate the rainfall 30 months prior to its occurrence, which in the simulations done, whenever the average five-year movement is used to reinforce the data, the combination of the previous rain and the maximum temperature is identified as the most proper status.