Quantifying uncertainty for climate change and long range forecasting scenarios with model errors . Part II : Non - Gaussian models with intermittency

As shown in Part I of this series, information theory provides a concise systematic framework for measuring climate consistency and sensitivity for imperfect models of a much more complex natural system. Here, we extend this analysis to non-Gaussian systems and discuss implications of such important effects as intermittency and coarse-graining on the prediction skill of non-Gaussian imperfect models. A suite of increasingly complex nonlinear models, some with intermittent hidden instabilities and with time-periodic features mimicking seasonal cycle, are utilized to illustrate a number of important issues in contemporary climate science. These include the role of model errors due to coarse-graining, moment closure approximations, and the memory of initial conditions in producing short, medium and long range forecasts. Importantly, we show that the predictive skill of the considered imperfect nonlinear models and their sensitivity to external perturbations is improved by assuring their climate consistency via appropriate inflation of the stochastic forcing. Furthermore, the discussed link between climate fidelity and sensitivity via the fluctuationdissipation theorem opens up an enticing prospect of developing techniques for improving imperfect model sensitivity based on specific tests carried out in the training phase of the unperturbed climate.