Figure from An Overview of CMIP5 and the Experiment Design discussed below
In recent posts, I have alerted readers to candid comments on the systematic bias in climate model predictions; e.g. see
As I report in the post
I have reviewed a paper with long-term climate predictions that states
“A global climate model that does not simulate current climate accurately does not necessarily imply that it cannot produce accurate projections”
There is a new article that perpetuates this myth that the longer term climate predictions do not require accuracy. The article is
Karl E. Taylor, Ronald J. Stouffer, Gerald A. Meehl, 2012. An Overview of CMIP5 and the Experiment Design. Bulletin of the American Meterological Society. April 2012.
The abstract of the article reads
The fifth phase of the Climate Model Intercomparison Project (CMIP5), now underway, promises to produce a freely available state-of-the-art multimodel dataset designed to advance our knowledge of climate variability and climate change.
The section of the article with respect to model bias is reproduced below [with highlight added]
Climate drift and bias correction. Below the seasonal thermocline, the ocean requires thousands of years to fully adjust to any change in external forcing. This means that the typical several-hundred-year CMIP5 control simulation, which attempts to determine the equilibrium climate for preindustrial conditions, is generally too short to eliminate residual drift (toward an eventual equilibrium). The drift may or may not significantly affect analysis of any particular aspect of the CMIP5 runs, but users should not prima facie assume the drift is inconsequential. If, for example, one were interested in examining how much historical warming has occurred according to some model, then the period in the control run that corresponded to the historical period would need to be examined to see if climate drift might explain part of the simulated trend. The usual approach is to assume that drift in the control run is also identically present in the corresponding period of the historical run, so simple subtraction yields trends without the artifact of residual drift. The drift will likely be most evident in variables linked to deep ocean conditions, but users should not assume the drift is negligible anywhere.
In the decadal prediction runs, a similar, but likely more significant, problem is expected. Because climate models are not perfect, their simulated equilibrium mean climate states will differ somewhat from the observed. When these models are initialized from observations, they will initially be forced away from their equilibrium states to match the observations. The model will subsequently tend to drift back toward its natural, but “biased,” equilibrium state. With the exception of the deep ocean, the initial drift will be much more severe than in the long-term runs, and this will be confounded with the climate evolution that is being predicted. In contrast to the long-term simulations, drift in the near-term simulations will in complicated ways almost certainly affect nearly all variables considered. Consequently, it will be essential to correct for drifts by applying a more sophisticated “bias correction” than for the long-term runs. There is no single, accepted approach for doing this (see, e.g., CMIP–WGCM–WGSIP Decadal Climate Prediction Panel 2011). Most users will find it difficult to bias correct the decadal prediction runs; it is therefore recommended that analysis of the near-term simulations be limited to the four variables that the modeling groups themselves plan to bias correct: near-surface air temperature, surface temperature, precipitation rate, and sea level pressure.
The statement that in “the decadal prediction runs, a similar, but likely more significant, problem is expected”, perpetuates the myth that is is easier to skillfully predict on multi-decadal time scales than on decadal time scales. That this is not correct should be obvious, but is not to these modellers. Indeed, this flawed perspective of climate model bias perpetuates the erroneous view that
“A global climate model that does not simulate current climate accurately does not necessarily imply that it cannot produce accurate projections.”
that I report on in my post