Examining the Linear Regimes of the Community Earth System Model (CESM)

Abstract:

In ensemble prediction, Gilmour et al. (2001) proposed measures of relative nonlinearity to quantify the duration of the linear regime from “twin” pairs of ensemble members. The duration of the “linear regime” is useful in forming and interpreting ensembles in numerical weather prediction. Here this method is applied to the state-of-the-art climate model CESM, focusing on how its linear durations will change as the perturbations imposed on one location differ spatially and temporally. The spatial and temporal propagations of point perturbations provide insights into model physics and facilitate interpretation of model projections in future climate scenarios. They provide insight into chaos-like behavior on short time scales, and an indication of the sensitivity and saturation (mixing) times of CESM. Starting from the same initial state, we add relatively small “twin” perturbations (that is, positive and negative perturbations of the same magnitudes) to surface variables, with the locations of the perturbations spanning from the tropics to the poles. As the location changes, the model evolves differently in terms of how the point perturbation extends out of its origin and spreads globally, indicating that different physical mechanisms have played roles in different cases. Repeating the same set of experiments by changing only the perturbation magnitudes insures the linear regime is sampled without constructing an adjoint. Further, how uncertainty growth varies with location in the model state space can be explored by repeating the experiment for different initial states. We compare the responses of the linear regime durations in terms of locations, initial states and magnitudes of the perturbations systematically, and the implications for ensemble experiments and sensitivity studies are discussed. This work is a first step towards treating state-of-the-art climate models with the tools of nonlinear dynamics.