סמינר בחוג לגיאופיזיקה: Learning Entertainment Rates for Convective Parametrizations for the Next Generation of Climate Model

Abstract: 

The recent Paris agreement, in which 193 countries committed to prevent the glob from a 2oC warming above the pre-industrial climate, posed a great challenge to the scientific community: What CO2 concentration should be avoided? The spread between models in the predicted “point of no return” for this target warming is 130ppm, which amounts for two decades in current CO2 emissions. The imperfections of the representation of sub-grid scale (SGS) process accounts for a large portion of the model spread.

Two caveats exist current SGS parameterizations in climate models: A. There are different parameterizations for boundary layer turbulence, shallow convection and deep convection. Thus, even though these are part of a continuous spectrum of an underlying physics, their functioning in climate model is discontinuous and controlled by tuning parameters. B. As model resolution increases the quasi-equilibrium assumption on which the parameterization is founded becomes invalid (i.e. the convective grey zone). 

The eddy diffusivity/mass flux (EDMF) parameterization offers a framework for a unified parameterization of sub-grid scale turbulence and convection by decomposing subgrid motions into (possibly multiple) updrafts and an environment. This offers a pathway to deal with the first deficiency mentioned above. The recently developed Extended EDMF framework (Tan et al., 2017, JAMES) has prognostic equations for updrafts and environmental turbulent kinetic energy, with a consistent (conservative) decomposition of energies between updrafts and the environment. This prognostic modification to the steady EDMF parameterization is essential in order to deal with the second deficiency mentioned above. However, the adequacy of the EDMF framework depends on a correct representation of the interaction between updrafts and the environment through entrainment and detrainment. Thus these closure parameters become essential for correct representation of convection and for the functioning of climate models. 

In this work, entrainment and detrainment rates are learned from LES simulations in order to be implemented in the extended EDMF parameterization. To this end, LES simulations of various convective regimes are designed to reproduce observational test cases, and using passive tracers the entrainment and detrainment rates are calculated directly and compared with updraft properties. Finally, a first step toward a machine learning application of the functional form of entrainment and detrainment (using a Markov chain Monte-Carlo algorithm) is presented.

מארגן הסמינר: פרופ' איל חפץ