Using observations to design Earth System Model Perturbed Parameter Ensembles

Perturbed Parameter Ensembles (PPEs) are an increasingly-used tool enabling process understanding and uncertainty quantification in Earth System Models (ESMs). Typically, the combinations of (often microphysical) parameters characterizing any one PPE member are determined without consideration of how well the emergent macroscale state compares to the diversity of global observations available. In this presentation, we discuss the advantage of deriving parameter combinations in a way that ensures macroscale states do not exhibit large discrepancies with respect to a large collection of observations. This sampling process, which relies on use of machine learning, leads to a special type of PPE that we coin a “calibrated physics ensemble”, or CPE. By definition, a CPE has high-dimensional parametric diversity coincident with similar emergent mean-state climatologies.

Using the latest NASA GISS ESM as a testbed for generation of a CPE, we discuss the role of observational uncertainty in the development of such an ensemble, and delve into the question of whether use of an expanded ‘mean-state’ definition vs. consideration of processes/sensitivities results in different members comprising a CPE. One overall advantage of a CPE versus a PPE is that all CPE members more closely reproduce observed base-state climates, and thus are equally plausible candidates for projections of future climate. This refined selection of plausible candidates, run into the future, can be one strategic way that climate projection envelopes can be established. We conclude the presentation with discussion of how a CPE can be used as a tool in guiding future observations that can refine the next generation of CPEs. Because the CPE’s members are all equally viable model configurations, this refinement can be directly mapped to a quantitative change in the projection envelope (i.e., a ""climate OSSE"" proof-of-concept).