P. B. Duffy, Curt Covey
Lawrence Livermore National Laboratory
Jason Bell
U.C. Santa Cruz
The climate simulated by coupled ocean-atmosphere general circulation climate models often drifts away from the present-day climate, even under constant present-day forcing. Flux adjustments (a.k.a. flux "corrections") are unphysical sources of heat, moisture, and momentum used in coupled GCMs to prevent this problem. The possibility that flux adjustments inadvertently distort other features of model simulations is a subject of vigorous debate [1 - 2]. Do flux adjustments suppress not only unrealistic climate drift but also natural climate variability in models? We aim to examine this question by looking for a relationship between the strength of flux adjustments and the interannual variability in the CMIP climate model control runs.
If flux adjustments do suppress climate variability in models, then -- all else being equal -- the models that use stronger adjustments should have less climate variability. Of course many other factors affect variability in climate models, and these factors differ among models. The CMIP data base, however, includes a large population of both flux-adjusted (9) and non-flux-adjusted (8) models. Any significant systematic influence of flux adjustment should be apparent if enough models are analyzed.
Although several CMIP subprojects [3] and other studies [e.g., 4] are actively investigating climate model variability, to our knowledge no other group is studying its relationship with flux adjustment in a large population of models.
We propose to look for a relationship between the magnitudes of heat flux correction and surface air temperature variability in the CMIP climate model "control" runs. We will focus on interannual-to-decadal and longer timescale variability. (Since most flux-adjusted models use adjustments that vary seasonally, it is not expected that flux adjustments will suppress the seasonal cycle. This supposition is verified in work by one of us [Covey] on the seasonal cycle in CMIP1 models.)
In preliminary work, we examined the relationship between variance of globally and annually averaged surface air temperature, and the presence or absence of flux adjustment, for the 17 CMIP models mentioned above. The average of surface air temperature variance is 0.008 K2 for flux-adjusted models and 0.011 K2 for non-flux-adjusted models. It is not clear that this small difference between the two classes of models is significant. However, taking a global mean before any other analysis is done has the effect of suppressing a great deal of the variability in the models. We propose to use a more sensible (and compute-intensive) measure of variability such as the global mean of interannual grid-point variance. We also propose to similarly measure the magnitude of heat-flux adjustment for each model, and to correlate this quantity with the amount of surface air temperature variability.
We believe that either way our results turn out -- i.e., whether or
not there is a substantial correlation between flux adjustment and
interannual variability in the CMIP population of models -- we will
reach a significant conclusion about modern coupled GCMs.
[1] Sausen et al., Climate Dynamics 2: 145 (1988)
[2] Kerr, Science 265: 1528 (1994)
[3] https://pcmdi.llnl.gov/mips/cmip/cmip_subprojects/project_list.html
[4] Stouffer et al., J. Climate, in press