Getting to the "Core" of Output Differences as Produced by Climate Models

An atmospheric model, whether it is a forecast model, or used to diagnose which processes drive the climate, are built on a foundation of three main conservation principles. These principles state that within the earth-atmosphere system, mass, energy, and momentum must be conserved. These basic physical principles are represented by a closed set of equations that we call the primitive equations. These equations form the "dynamic core" of computer general circulation or climate models. The equations also represent the motions of air within model. The model output is generally some variable representing mass (pressure), temperature, or both.

The primitive equations are too complex in their raw form to be represented adequately in these models and are often simplified following the concept of Occam's Razor (they are simplified as much as possible, and no more). One way to simplify these equations is to "linearize" them, thereby eliminating non-linearity that can make a computer model unstable. This non-linearity is one process by which the atmosphere behaves in a chaotic fashion, and is one reason (of many) why weather forecasting is not 100% reliable.

Kondrashov et al. (2011) use a simplified three-level atmospheric circulation model designed by Marshall and Molteni (1993) to show the evolution of the wind fields. It is a model that has a fairly coarse horizontal and vertical resolution by today's standards, but has been shown to represent the largest-scales of the atmosphere's climate in a fairly realistic way. The experiment here examines the tendencies of the 500 hPa mass field (streamfunction) due to linear and non-linear processes. The non-linear processes were divided into those that are resolvable by the model and those that are of smaller scale than can be represented by the model.

The model output was slightly different depending on which processes were included in the model run. A comparison of the full non-linear processes or interactions (Fig. 1a) gave slightly different results than when only non-linear resolvable processes (Fig. 1b). The model output also differed from that of two similar studies because of the strategies used to calculate these model "climates". The authors concluded the article by stating that "the manner of defining interactions between the resolved and unresolved modes plays a crucial role in the dynamical interpretation of the tendency-based statistical diagnostics."

The basic lesson to be learned from the author's closing statement is that there can be output differences produced within the same model core even when using the same input data, and depending on how the non-linear processes are represented. If the representation of climate can be different depending on how these core processes are represented, then producing model scenarios on the time scale of a century must be done very cautiously, and interpretation be done carefully.

Additional References
Marshall, J. and Molteni, F. 1993. Toward a dynamical understanding of atmospheric weather regimes. Journal of the Atmospheric Sciences 50: 1972-1818.