Impact of Ensemble Size and Horizontal Resolution on the Performance of the NCEP GEFS

Juhui (Jessie) Ma


Numerical Weather Prediction (NWP) centers around the world face the same questions when they develop (or upgrade) an ensemble forecast system. How many ensemble members do we need to better represent forecast uncertainties with limited computational resources? What is the relationship between resolution and ensemble size? This investigation starts from the Lorenz 96 model by using the ensemble transform with rescaling (ETR) initialized perturbations for up to 320 ensemble members. The results are contrasted with tests based on the NCEP Global Ensemble Forecast System (GEFS) from different ensemble sizes and resolution. The impact of various ensemble sizes is studied by applying a set of verification measurements for the period of December 1st, 2009 to January 31st, 2010. The several variables (such as 500hPa geopotential height) are verified over the Northern Hemisphere (NH) and Southern Hemisphere (SH) extra-tropics. Results indicate that increasing ensemble size is beneficial to improve skill of ensemble mean for small ensemble members (especially less than 40-member). However, the skill of probabilistic forecast will be significantly improved with further increasing ensemble members. Meanwhile, the relative benefits of T126L28 resolution with 70 members and T190L28 resolution with 20 members (the same model) which have equivalent computing cost are also compared. The comparison of the two configurations indicates that increasing model resolution is more (less) beneficial than increasing ensemble size for short (long) lead times.