The nonlinearity of a climate field's parameter dependence can be an important factor affecting model uncertainty, and its explicit form has not yet been thoroughly addressed in ensemble studies. One approach is to make assumptions about the form of nonlinearity and employ a metamodel to construct spatial patterns of a field's parameter sensitivity. An alternative method is to produce spatial patterns of that field's sensitivity first, and then to construct the nonlinearity of the parameter dependence associated with these patterns.
To produce spatial patterns of P uncertainty, empirical orthogonal function analysis and maximum covariance analysis is applied across two ensembles: the Climate Model Intercomparison Project phase 5 (CMIP5) ‘ensemble of opportunity' and a perturbed physics ensemble consisting of branch runs of the Community Earth System Model 1.1 Series (CESM1.1), in which parameters associated with the deep convection scheme are varied systematically. Comparing uncertainty patterns for the perturbed physics and CMIP5 ensembles, the magnitude is similar for P for both the historical climatologies and end-of-century changes. The highest amplitude of spread occurs in the tropics, and features common to both ensembles include uncertainty throughout the tropical Pacific and Indian Oceans, especially near convective margins and in the Intertropical Convergence Zone/South Pacific Convergence Zone (ITCZ/SPCZ) complex.
In addition, we use the resulting uncertainty patterns from the perturbed physics ensemble to visualize the behavior of P in parameter space. This exploration is conducted with an eye towards bridging the identification of important physical processes in perturbed physics ensembles with the uncertainty sampled in an ensemble of opportunity.