Effect of the ensemble system size on the precipitation forecast accuracy

Numerical weather prediction (NWP) models are not completely accurate and error free, and there is always some uncertainty. The errors in weather forecasting stem from the limitations of human theoretical understanding of the atmosphere and the operational capacity to produce forecasts. It is necessary to make a forecast, along with an estimate of its uncertainty. This is accomplished by creating ensemble systems of weather forecasts differing in the initial conditions or physical formulation of NWP models. There are several methods for post-processing of ensemble forecasting, including Bayesian Model Averaging (BMA) and Ensemble Model Output Statistics (EMOS) that they are more popular because of higher efficiency and accuracy. In this research, first, an 18-member ensemble system is formed, which each member is an independent run of the WRF model with different physical configurations. BMA method was used to estimate the density function of predicting 24-hour cumulative precipitation. Due to some hardware limitations and access to an ensemble system with fewer number and more efficient members, the size of the ensemble system has been reduced to 7 members. Using the BMA method, a weight is assigned to each ensemble member. The size of the ensemble system is reduced by removing the members who had less weight. The probabilistic prediction verification obtained from the 7-member ensemble system in a test period from 15 January 2020 to 15 May 2020 has been checked using reliability diagram. The results show that the probabilistic predictions are sufficiently skilled for 24-hour cumulative precipitation.