Improving the Parameterization of Cloud and Rain Microphysics in E3SM Using a Novel Observationally-Constrained Bayesian Approach
Motivation: Earth System Models (ESMs), including the Energy Exascale Earth System Model (E3SM), suffer from biases in cloud properties. To some extent these biases can be mitigated by tuning free parameters associated with microphysics or other cloud-related parameterizations. However, a critical challenge is the fact that not all sources of parameterization error are easily adjusted or tuned. That is, most parameterization schemes suffer from so-called ``structural errors'' that are the result of fixed approximations, assumptions, and uncertain mathematical formulations. If structural errors could be addressed, then tuning can be performed such that optimal parameter values represent true physical knowledge. In this case, so-called ``bottom-up'' approaches to ESM model development, wherein parameterizations are revised to incorporate advances in process-level knowledge, can be reconciled with ``top-down'' approaches, wherein tuning is performed on the full ESM by relying on ``emergent properties'' of the climate system to constrain parameterizations. For parameterizing cloud microphysics, uncertainties exist at all levels of scheme complexity and scales. It is challenging to quantify these uncertainties rigorously in current microphysics schemes because many of their structural elements are fixed and “hard-coded” into scheme formulations. To address this challenge, we have recently developed a new bulk parameterization of microphysics, the Bayesian Observationally-constrained Statistical-physical Scheme (BOSS). This approach allows for systematic adjustment of scheme structure and quantification of structural and parametric uncertainties via Bayesian inference. It provides a consistent framework for uncertainty quantification and constraint using both bottom-up (e.g., from detailed process-level microphysical modeling) and top-down (e.g., from global satellite observations) approaches. In this way, BOSS facilitates a hierarchical approach to parameterization development and tuning.
Proposed research: We will use BOSS to parameterize liquid (cloud and rain) microphysics in E3SM. Bottom-up and top-down experiments will be performed, using information from the former as prior probabilities for the latter — thus ensuring that the two are consistent. Our bottom-up approach will utilize column- and box-models with constraint of BOSS provided by a number of “reference” bin and bulk microphysics schemes, while our top-down tests will use BOSS integrated into the Morrison and Gettelman (MG2) microphysics parameterization in E3SM. For the top-down tests, we will employ a number of global satellite datasets as the constraint, representing a best estimate of climate- and cloud-relevant quantities. For all top-down experiments, microphysical parameters as well as those in other model parameterizations will be simultaneously perturbed. We will also compare a top-down ``control'' constraint of E3SM parameters with one where BOSS replaces the standard liquid microphysics parameterization, thus revealing any unique benefits of using BOSS.
Expected outcomes: The main objective is an improved treatment of microphysics in E3SM, rigorously informed by state-of-the-art microphysical process models and optimally tuned against a set of extensive global satellite-derived observations. This will improve the representation of microphysics in warm clouds, which impacts climate sensitivity and is critical for modeling aerosol indirect effects and the hydrological cycle. While we expect this effort to yield tangible improvement in E3SM climate simulations and reduced uncertainty, this will also demonstrate the hierarchical, combined bottom-up and top-down Bayesian approach as a viable method for parameterization development, testing, and tuning in ESMs more broadly.