Evaluation of NMME models in forecasting of monthly rainfall (Case study: Sefidrood Basin)

Monthly rainfall forecasting plays a major role in the water resources management and agroclimatic studies. The main purpose of this study is to assess the accuracy of NMME (North American Multi-Model Ensemble) in forecasting monthly rainfall in Sefidrood basin, North of Iran. For this purpose, the historical predicted data of NMME models for the period 1982 to 2017 were retrieved from, University of Columbia website, and compared with observed data obtained from the Iranian Meteorological Organization. The accuracy of NMME models predictions was evaluated by comparing them with the observed data, using statistical indices. The results showed that the single NMME model is not accurate, where the average value of determination coefficient (R2) was equal to 0.6. The models combination improved the accuracy of predictions, such that the determination coefficient increased to 0.7. Furthermore, for evaluation of the precipitation uncertainty, seventy-eight ensembles of the prediction models were investigated. The results of this evaluation showed that the models overestimated rainfall upto 80%. In addition, the uncertainty analysis of prediction models showed that the combination of models may reduce the uncertainty range.