8.1.2 Methods of Evaluation
A climate model is a very complex system, with many components. The model must of course be tested at the system level, that is, by running the full model and comparing the results with observations. Such tests can reveal problems, but their source is often hidden by the model’s complexity. For this reason, it is also important to test the model at the component level, that is, by isolating particular components and testing them independent of the complete model.
Component-level evaluation of climate models is common. Numerical methods are tested in standardised tests, organised through activities such as the quasi-biennial Workshops on Partial Differential Equations on the Sphere. Physical parametrizations used in climate models are being tested through numerous case studies (some based on observations and some idealised), organised through programs such as the Atmospheric Radiation Measurement (ARM) program, EUROpean Cloud Systems (EUROCS) and the Global Energy and Water cycle Experiment (GEWEX) Cloud System Study (GCSS). These activities have been ongoing for a decade or more, and a large body of results has been published (e.g., Randall et al., 2003).
System-level evaluation is focused on the outputs of the full model (i.e., model simulations of particular observed climate variables) and particular methods are discussed in more detail below.