Influence of a stochastic convection parameterization on the statistics of rainfall in the Community Atmosphere Model

Understanding the physical mechanisms responsible for extreme weather phenomena and their changing behaviour in a warming world is essential to predict such events. This task, however, is challenging because of the nonlinear behaviour of the climate system, the small scales and fast dynamics of extreme events and the inherent intermittency of our turbulent atmosphere and ocean. Recent work has shown that representing the inherent intermittency and chaotic nature of turbulent sub-grid scale processes, such as convection, via stochastic parameterizations can improve the performance of weather and climate models, including the interaction of sub-grid scale processes and large scale climate. However, the influence of stochastic parameterizations on the prediction of extreme events has not received much attention.

In this study we explore the influence of a stochastic parameterization of convection on the statistics of extreme rainfall events in the USA. The deep convection parameterization of the Community Atmosphere Model (CAM5), which is based upon the bulk mass-flux scheme of Zheng and McFarlane, is modified to have a stochastic entrainment rate in the mixing model for the calculation of dilute CAPE. The entrainment events are described by a stochastic Poisson process, based on cloud resolving model simulations by Romps and Kuang [2010]. This modification represents turbulent mixing in the atmosphere in a manner more consistent with large eddy simulations of convection.

We find that the stochastic scheme results in an increase in the amounts of both light and intense precipitation, which is in closer agreement with observed rainfall distributions compared to results from CAM5 with its usual deterministic parameterization of convection. The increase is due to changes in both the convective precipitation and the large scale precipitation, which suggests that stochastic entrainment alters not just the sub-grid scale processes but also their interaction with the large scale climate. We use a hindcast-based system developed by O’Brien et al. [abstract submitted to session #2018] to characterize the influence of stochasticity on the fidelity of simulated extremes.