Using ML and Bayesian Statistic to Constrain Microphysics Parameters across Scales, from LES to Esms

Bulk microphysics schemes are a pervasive source of uncertainty in global atmospheric models and are thus a critical target for improvement, in part because the representation of clouds strongly affects simulated global climate. We present work that makes strides towards constraining structural and parametric uncertainties in bulk parameterization of warm microphysics across a spectrum of scales. These questions are investigated using the Bayesian Observationally-constrained Statistical-physical Scheme (BOSS), a structurally flexible data-driven warm bulk microphysics parameterization, as well as using other methods for identifying possible structural errors. We consider a broad spectrum of frameworks for constraint of microphysics, from process-level insights gained via detailed comparison to "reference" microphysics schemes such as bin and Lagrangian approaches, for example in box- and 1D-models. We also show how this differs from constraint of BOSS parameters in high-resolution, limited domain, large eddy simulations. We then address how these insights scale to Earth System Models (ESMs) such as NASA GISS ModelE, CESM, and E3SM, where microphysics and cloud distributions both are parameterized. An important component of this latter effort is the joint perturbation of microphysics and cloud scheme parameters, to obviate issues related to compensation of errors between these model components. Here, single-column versions of ESMs serve an important role to bridge the gap between the high-resolution process and low-resolution global scales. We show results from recent implementation of BOSS into global ESMs, compare constraint of parameters in LES and in single-column versions of global models, and discuss prospects for the future; including how detailed microphysical insights can be reconciled with global perturbed-parameter ensemble tuning of global atmospheric models.