An Efficient Advection Scheme for Cloud-resolving Models with Detailed Microphysics

Appearance of cloud resolving models (CRM) with explicit description of microphysical processes, which are able to properly describe cloud-aerosol interaction, dramatically increased the number of variables. Standard two moment bulk parameterization schemes contain more than 20 microphysical positive definite species such as mass contents and number concentrations of different hydrometeor type. The number of variables increases to several hundred when bin microphysical approach is used. This increase is caused by the fact that each hydrometeor type is described by a size distribution containing several tens of bins. High computational cost is the main obstacle hindering the wide application of CRM with detailed microphysics in large-scale prognostic and climatic models. Thus, decrease of the computational time without the loss of the accuracy in description of physical processes is of vital importance for CRM and climate models. Simulations of mesoscale convective systems using WRF with bin-microphysics show that the advection of these hundred variables takes about 80% of entire computer time. Thus, the first step to decrease the computer time of cloud-resolving models is the utilization of more computationally efficient advective schemes.

A necessary property of an acceptable advection scheme is that it should be positive definite as it advects size distribution functions. Most effective advection schemes use flux limiters which turn them non-linear. Therefore, they have to be computed separately for each advected field. It is therefore desired to develop a linear scheme which may be applied to all fields.

Taking this feature into account, in this study we propose a semi Lagrangian numerical scheme and test it by comparison with two high order monotonic nonlinear schemes by simulation of hail storms using the 2D Hebrew University cloud model and WRF. It is shown that a semi Lagrangian numerical scheme produces similar results requiring, however, 5-10 times less computer time.