Sensitivity to the representation of precipitating ice in CRM simulations of deep convection

Abstract

A Cloud resolving model (CRM) is a useful tool for providing a proxy for observed data, against which parameterizations of convection in global and regional models can be compared. The parameterization of microphysics in CRMs has been shown to be crucial for the simulation of the evolution of heat and moisture profiles during deep convection. The aim of the study described herein is to validate the microphysics scheme in the U.K. Met. Office's Large Eddy Model as a prelude to future work where simulations of deep convection will be compared to a single column version of the Met. Office's Unified model. Three observed cases of convection are simulated each with different precipitation production mechanisms, as deduced by the multiparameter radar at Chilbolton. Four simulations are carried out of each case, using different representations of precipitation. The aim is to determine whether a flexible microphysics scheme is capable of modelling the evolution of the precipitation in each of the cases without recourse to tuning coefficients for case-specific conditions. The simulations are validated against radar observations by comparing spatial distribution of radar reflectivity and the type of precipitation at the melting layer. A number of systematic errors occur in simulations using a `single-moment' microphysics scheme (where graupel and snow mass concentrations are each represented with one variable and the number concentrations are prescribed). Whereas a `double-moment' microphysics scheme (that predicts both the mass and number concentrations of snow and graupel) produced simulations consistent with radar observations for all three cases. The sensitivity of important quantities relating to the parameterization of convection in GCMs to the microphysical parameterization has been examined. The total precipitation at the ground and the amount of cloud ice are compared between simulations that differ only in their representation of precipitation. The total precipitation is found to vary by up to 40% and the total cloud ice by up to 200% between simulations of the same case.