The Bayesian Observationally constrained Statistical-physical Scheme (BOSS), a novel microphysical parameterization framework that effectively leverages uncertain observational information

Description

In recent years, microphysical parameterization schemes have grown ever more sophisticated, with numerous prognostic hydrometeor categories, and either size-resolved (bin) particle size distributions, or multiple prognostic moments of the size distribution. However, the sophistication of these schemes has not yet resulted in a commensurate reduction in the uncertainty of model representation of microphysical processes and the related effects of microphysics on numerical simulation of climate and weather. We posit that this may be caused by unconstrained assumptions of these schemes, such as ad hoc parameter value choices and structural uncertainties (e.g., choice of a particular form for the size distribution). We present work on the development and observational constraint of a novel microphysical parameterization approach, the Bayesian Observationally constrained Statistical-physical Scheme (BOSS), which seeks to address these sources of uncertainty. Our framework avoids unnecessary a priori assumptions, instead relying on observations to provide probabilistic constraint of the scheme structure and sensitivities to environmental and microphysical conditions. Here, we show estimation of the optimal structural form and parameter values of BOSS; specifically, we present tests of BOSS’ ability to reproduce the microphysical behavior of a traditional two-moment microphysics scheme. BOSS is capable of varying levels of complexity, as demanded by observed behavior or computational expense considerations, and we here compare the performance of various versions of BOSS with different levels of structural complexity. In all cases, observationally constrained optimal parameter values are estimated within a Bayesian inference framework using a Markov Chain Monte Carlo sampler, which allows for robust estimation of parameter uncertainty. BOSS has the potential to improve ensemble-based probabilistic weather and climate forecast systems by observationally constraining and robustly estimating microphysical parameterization uncertainty to a degree impossible with traditional schemes.