Comparison of Subseasonal Predictive Skill of the CFS and GEFS Operational Systems at NCEP and the Impact of Bias Correction

Abstract For subseasonal predictive skill, the relative merit of atmospheric model configuration and surface–atmosphere coupling has not been fully explored. The Climate Forecast System, version 2 (CFSv2), and Global Ensemble Forecast System, version 12 (GEFSv12), are two operational forecast systems used at the National Centers for Environmental Prediction for subseasonal prediction. CFSv2 incorporates a fully coupled system, whereas GEFSv12 utilizes simplified one-way surface boundary forcings to drive its atmospheric component. Nevertheless, GEFSv12 features a more advanced atmospheric model than CFSv2. This raises the question of how the atmospheric-only model and the surface–atmosphere coupling influence subseasonal forecast skill. Therefore, a comprehensive comparison of deterministic forecast skill between these two systems is conducted to better understand the relative merit of these two factors in subseasonal forecasting. Our evaluation reveals that, prior to bias correction (BC), GEFSv12 generally exhibits higher predictive skill than CFSv2 across most variables and regions. However, this performance contrast is largely reversed following BC, with CFSv2 outperforming GEFSv12 in the majority of cases. Diagnostic analyses suggest that the superior pre-BC skill of GEFSv12 is primarily attributable to its smaller systematic biases, probably stemming from its more advanced atmospheric model configuration. In contrast, the post-BC results indicate that CFSv2’s relative strength lies in its more accurate representation of surface–atmosphere coupling, which enhances its subseasonal forecast skill despite its less advanced atmospheric component. These findings imply that further improvements in subseasonal prediction may require not only enhanced simulation of coupling processes but also continued reduction in biases associated with atmospheric model deficiencies. Significance Statement The Climate Forecast System, version 2 (CFSv2), and Global Ensemble Forecast System, version 12 (GEFSv12), are two subseasonal systems differing in their atmospheric model configuration and handling of surface–atmosphere interactions. However, the relative merits of atmospheric model configuration versus coupling processes of these two systems remain unexplored. This study provides a thorough and quantitative comparison of the subseasonal predictive skill of CFSv2 and GEFSv12. The main results reveal that GEFSv12, with a more advanced atmospheric model, outperforms CFSv2 before bias correction, while CFSv2 reversely displays overall higher subseasonal predictive skill than GEFSv12 after removing larger systematic bias. This underscores that advancing subseasonal prediction skill will likely require both improved representation of surface–atmosphere coupling processes and continued mitigation of biases arising from deficiencies in atmospheric model components.