What is a reasonable ensemble size for operational ensemble forecast?

The numerical prediction centers around the world face the same questions when they develop an ensemble forecast system. How many ensemble members do we need to have better representing forecast uncertainties with the limited computational resources? What is relationship between resolution and ensemble membership? This study will start from Lorenz model for various initial perturbed forecasts for over 1000 members, then testing NCEP global ensemble forecast system (GEFS) with different ensemble size and resolution. The skills of ensemble forecast generated by 5, 10, 20, 40, 60 and 80 members are compared from various verification methods during December 1st, 2009 to January 31st, 2010 for 500hPa, 1000hPa geopotential height and 850hPa temperature field over 6 regions. A preliminary result indicates that the performance of ensemble forecast is significantly affected by ensemble size and resolution. The further discussion will be follow up by using NCEP standard probabilistic verification method which includes RMS error of ensemble mean and ensemble spread, anomaly correlation of ensemble mean, continuous rank probability score (CRPS), resolution and reliability from Brier Score (BS) decomposition and relative operating characteristics (ROC).