Examining Model Fidelity via Shadowing Time

Thursday, 18 December 2014
Leonard A Smith1,2 and Hailiang Du2, (1)London School of Economics, London, United Kingdom, (2)University of Chicago, Chicago, IL, United States
Fully fledged climate models provide the best available simulations for reflecting the future, yet we have scant insight into their fidelity, in particular as to the duration into the future at which the real world should be expected to evolve in a manner today's models cannot foresee. We know now that our best available models are not adequate for many sought after purposes. To throw some light on the maximum fidelity expected from a given generation of models, and thereby aid both policy making and model development, we can test the weaknesses of a model as a dynamical system to get an informed idea of its potential applicability at various lead times. Shadowing times reflect the duration on which a GCM reflects the observed dynamics of the Earth; extracting the shortcomings of the model which limit shadowing times allows informed speculation regarding the fidelity of the model in the future. More specifically, by identifying the reasons models cannot shadow we learn the relevant phenomena limiting model fidelity, we can then look at the time scales on which feedbacks on the system (which are not active in the model) are likely to result in model irrelevance. The methodology is developed in the “low dimensional laboratory” of relatively simple dynamical systems, for example Lorenz 95 systems. The results are presented in Lorenz 95 systems as well as GCMs. There are severe limits on the light shadowing experiments can shine on GCM predictions. Never the less, they appear to be one of the brightest lights we can shine to illuminate the likely fidelity of GCM extrapolations into the future.