It has long been recognized that the representation of “model physics” (e.g., clouds, convection, radiation, and turbulence) is an important source of uncertainty in numerical models that range from large eddy simulation to general circulation model scales. A key source of uncertainty is specification of parameters that control the rates and/or characteristics of physical processes. In contrast to errors in forecast initial conditions, the characteristics of model physics uncertainty are not well understood, hence model physics error is not typically included in assimilation systems. With increasing use of ensembles of simulations for data assimilation and probabilistic forecasting, it is desirable to perturb not only the initial conditions, but also model physics parameters (current “physics ensembles” are a crude example of this). To do so requires knowledge of which parameters have the greatest effect on model results, as well as the characteristics of the relationship between model output and changes to parameters.

In this paper, we demonstrate use of numerical methods that map the functional relationship between changes in model physics parameters and changes in model output. We examine cloud microphysics and radiation packages from a cloud resolving model that are similar to schemes used in modern regional and general circulation models. We identify which parameters have the most significant effect on various model output fields, highlight the role of nonlinearity in the parameter-state relationship, and examine how changes in the characteristics of observations (changes in observed variables and their distribution in time) affect the model state. We demonstrate the utility of nonlinear mapping techniques for exploring optimal parameter perturbation strategies, as well as for highlighting necessary areas for improvement in model physics routines.