A Calibrated Physics Ensemble (CPE) to Represent Climate Model Uncertainty and Guide Development of Observing Systems

Perturbed parameter ensembles (PPEs) are an increasingly-used tool enabling process understanding and uncertainty quantification in Earth System Models (ESMs). However, it is typical to generate these ensembles via combinations of (often microphysical) parameters with little consideration of how well the emergent macroscale state compares to the diversity of global observations available. We present a special type of PPE termed a "Calibrated Physics Ensemble" (CPE) where parameter combinations are generated so as to be consistent with satellite observation of the Earth's climate. We use the latest NASA GISS ESM and a suite of satellite-based atmospheric observations for CPE generation, leveraging machine learning and Bayesian inference to generate an ensemble of parameter vectors featuring high-dimensional parametric diversity, but that still produce base-state climates similar to Earth. Two key aspects of CPEs include: 1. their member diversity is distinctly shaped by the suite of observations and observational uncertainties utilized, and 2. members closely reproduce observed base-state climates, and thus are probabilistically plausible candidates for projections of future climate. Because the CPE framework makes this connection between observational information content and subsequent estimates of climate model uncertainty (e.g. via climate projection envelopes), we can use this framework to quantify the gain or loss in climate predictive capability resulting from the addition or subtraction of observations. In essence, the CPE framework is also a framework for a climate model Observing System Simulation Experiment (i.e. "climate OSSE"). We show demonstration of how a hypothetically improved observation constrains uncertainty in the climate model.