Item

Abstract

Convection parameterization schemes make the assumption that a large number of clouds is contained in each grid box and hence, given a certain profile, they produce always the same average response. However, since model resolutions have significantly increased, that assumption is no longer valid and random variability from the (unpredictable) distribution of the individual clouds becomes apparent near the grid scale. A scale-adaptive stochastic convection scheme for simulating this variability has first been introduced by Plant and Craig in 2008. Recently an approximation to the Plant-Craig concept has been suggested by Machulskaya and Seifert (2019), which applies two stochastic differetial equation for cloud number and the total mass flux and eliminates the need to track and store individual clouds. This approach has been implemented into ICON and is used to randomly rescale mass fluxes and tendencies from the Tiedtke-Bechtold convection scheme. We provide the results of first tests of the new scheme and a basic evaluation. We demonstrate that averaging of the input fields is required to get a stable closure mass flux and that the mass flux distribution produced by the scheme still closely follows the theory of Craig and Cohen. We further provide a basic evaluation with respect to forecast skill and an assessment of ensemble spread generation, where we use a convection-permitting ensemble as a reference.