In this study, an effective post-processing approach has been examined to improve skill of NMME precipitation forecasts. This method is based on the existence of a correlation between the historical raw forecast and observational data. In this respect, the Copula-Bayesian approach was used along with the Normal Kernel Density marginal distribution, kernel Copula function, and a novel approach to select final improved forecast data amongst the existing candidates on the calculated Conditional Probability Distribution Functions (CPDF). In this approach, called the Double Copula method, four input variables are effective for determining the improved NMME data. These are 1) the likelihood of an improved forecast (as a probable observation) for a given raw forecast (CPDFf) 2) the likelihood of raw forecast for the corresponding improved forecast (CPDFo) 3) the probability of occurrence of raw and 4) the probability of occurrence of improved forecast data (as PDF). The evaluation of the proposed method for improving the precipitation forecast by the NMME model has been performed in Karoon basin, Iran. Here, the data of 1982-2010 for the calibration period (hindcast) and 2011-2018 (forecast) to validate the results have been used. The results show that the improved forecast data is more reliable due to several achievements namely; 1) higher spatial and temporal accuracy and consistency are observed, 2) extreme values of precipitation are better detected, and finally, 3) during different length of time, the involved uncertainties have been reduced significantly in comparison with raw data.