Evaluation of the CFSv2 CMIP5 decadal predictions

Abstract

Retrospective decadal forecasts were undertaken using the Climate Forecast System version 2 (CFSv2) as part of Coupled Model Intercomparison Project 5. Decadal forecasts were performed separately by the National Center for Environmental Prediction (NCEP) and by the Center for Ocean-Land-Atmosphere Studies (COLA), with the centers using two different analyses for the ocean initial conditions the NCEP Climate Forecast System Reanalysis (CFSR) and the NEMOVAR-COMBINE analysis. COLA also examined the sensitivity to the inclusion of forcing by specified volcanic aerosols. Biases in the CFSv2 for both sets of initial conditions include cold midlatitude sea surface temperatures, and rapid melting of sea ice associated with warm polar oceans. Forecasts from the NEMOVAR-COMBINE analysis showed strong weakening of the Atlantic Meridional Overturning Circulation (AMOC), eventually approaching the weaker AMOC associated with CFSR. The decadal forecasts showed high predictive skill over the Indian, the western Pacific, and the Atlantic Oceans and low skill over the central and eastern Pacific. The volcanic forcing shows only small regional differences in predictability of surface temperature at 2m (T2m) in comparison to forecasts without volcanic forcing, especially over the Indian Ocean. An ocean heat content (OHC) budget analysis showed that the OHC has substantial memory, indicating potential for the decadal predictability of T2m; however, the model has a systematic drift in global mean OHC. The results suggest that the reduction of model biases may be the most productive path towards improving the model’s decadal forecasts.

Appendix: An issue with the CFSR coupled and its influence

In the course of the investigating the causes of the model biases discussed in Sect. 4, COLA scientists discovered a coding error in the coupler associated with the sea ice mask and the atmosphere–ocean coupling. This error is present in the versions of CFSv2 used by both COLA and NCEP in the results described here, as well as in the version of CFSv2 used for the CFSR data assimilation and CFS reforecasts (Saha et al. 2010). The result of this code error is that the fluxes provided by the atmosphere to the ocean in some regions are substantially different from the fluxes seen by the ocean. The locations of the inconsistency in the fluxes are dependent on domain decomposition and the number of processors used. Figure 13 shows the inconsistencies in net surface heat fluxes in the COLA runs and in a run made by COLA where the code error was corrected. The differences are computed from monthly mean atmospheric model and ocean model outputs, and points with nonzero sea ice are excluded.

Fig. 13

The annual mean difference (W m⁻²) between the diagnostic of the net surface heat flux supplied by the atmospheric model to the ocean model and the net surface heat flux received by the ocean model in CFSv2 for a the model run by COLA and b a run by COLA with the code error corrected. Points with nonzero sea ice are excluded

Considering the COLA version in Fig. 13a, there is inconsistency in the net surface heat flux in the annual mean in the eastern North Atlantic near 0°E 60°N, with the ocean net cooling calculated by the atmosphere on the order of 100 W m⁻². This negative inconsistency extends southward to the coast along Spain as well as westward into the North Atlantic and northeastward to 60°E. There is also a region of weaker negative inconsistency near the Bering Strait and in the northwestern North Atlantic. Similar features are also found in the NCEP version (not shown). When the code error is fixed (Fig. 13b), the regions of negative inconsistency disappear. The other surface fluxes (surface wind stress and fresh water flux) also show inconsistencies coincident with the negative heat flux inconsistencies in Fig. 13a, and these also disappear when the code error is fixed. Our explanation of the negative inconsistencies that are removed by fixing the code error is that the coupler is erroneously insulating the ocean from the atmosphere in these regions as if there were sea ice.

We might expect that the wind stress and heat flux inconsistencies would have seriously distorted the ocean circulation in the North Atlantic, including the AMOC and the subpolar gyre. Given that the AMOC variability has been postulated to be an important mechanism for decadal climate variability and predictability in the North Atlantic (e.g. Huang et al. 2011; Swingedouw et al. 2012), that mechanism is probably not realistically represented in the results reported here.