EGU2020-11243, updated on 12 Jun 2020
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

The decadal climate prediction skill with focus on the North Atlantic region

Shuting Yang and Bo Christiansen
Shuting Yang and Bo Christiansen
  • Danish Meteorological Institute, Research and Development, Copenhagen, Denmark (

The skill of the decadal climate prediction is analyzed based on recent ensemble experiments from the CMIP5 and CMIP6 decadal climate prediction projects (DCPP) and the Community Earth System Model (CESM) Large Ensemble (LENS) Project. The experiments are initialized every year at November 1 for the period of 1960-2005 in the CMIP5 DCPP experiments and 1960-2016 for the CMIP6 DCPP models as well as the CESM LENS decadal prediction. The CMIP5/6 ensemble has 10 members for each model and the CESM ensemble has 40 members. For the considered models un-initialized (historical) ensembles with the same forcings exist. The advantage of initialization is analyzed by comparing these two sets of experiments.

We find that the models agree that for lead-times between 4-10 years little effect of initialization is found except in the North Atlantic sub-polar gyre region (NASPG). This well-known result is found for all the models and is robust to temporal and spatial smoothing. In the sub-polar gyre region the ensemble mean of the forecast explains 30-40 % more of the observed variance than the ensemble mean of the historical non-initialized experiments even for lead-times of 10 years.

However, the skill in the NASPG seems to a large degree to be related to the shift towards warmer temperatures around 1996. Weak or no skill is found when the sub-periods before and after 1996 are considered. We further analyze the characteristics of other climate indicators than surface temperature as well as the NAO to understand the cause and implication of the prediction skill.

How to cite: Yang, S. and Christiansen, B.: The decadal climate prediction skill with focus on the North Atlantic region, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-11243,, 2020


Display file

Comments on the display

AC: Author Comment | CC: Community Comment | Report abuse

displays version 1 – uploaded on 05 May 2020
  • CC1: Comment on EGU2020-11243, Leon Hermanson, 07 May 2020

    Very interesting study, as I am sure you know there was an argument a few years ago about whether hurricanes are predictable on decadal time scales and the mid-1990s shift was cited as the only source of skill. I also wanted to comment that looking at a difference in correlation may not be the best way to consider the impact of initialization, please consider this paper for another method: Smith et al (2019)

    • AC1: Reply to CC1, Shuting Yang, 07 May 2020

      Thank you for the comments and the reference. It is a good point that correlation might not be the best measure for prediction skills. We also started to apply other methods. Signle-to-noise is certainly one of them.

      I must say I don't know about the study regarding the decadal prediction skill on hurricants you memtioned. It is very much appreciated if you have the references.