Ensemble Size and Skill of the NCEP CFSv2

The impact of ensemble size on the performance of the NCEP CFSv2 is analyzed. The skills of
ensembles generated using 4 members as it is done in the current operational setting and of the
enhanced 20 members in the experimental setting are compared for representative periods of
months after 2011. The low resolution of the CFSv2 warrants that this enhancement is cost
limited compared to current high-resolution models. The enhanced ensemble generated every
cycle consists of the 4 ocean states of the operational system and a 20-member subset of the
operational 80-member EnKF perturbations of the operational NCEP data assimilation system.
Results are based on the analyses of the 2mT and Precipitation over North America and global.
Various measures of performance are applied: spread-skill relationship, Brier score, reliability
diagrams, relative operating characteristics and outlier statistics. Consistent with past analysis of
the hindcast, in general, probabilistic skill scores improve with a larger ensemble. By design, the
most benefit of this ensemble enhancement is obtained at the medium range but the skill
improvement extends to longer lead times. Overall, increasing the ensemble size reduces the
sampling error to statistically adjust the real-time ensemble spread to the full forecast probability
distribution function benefitting users whose products require all the information available in a
model.