Speeding up CRMs for cloud-climate interaction studies by acceleration of mean state tendencies

Cloud-resolving models (CRMs) are routinely used to simulate boundary-layer and deep convective cloud processes, aid in the development of moist physical parameterization for global models, study cloud-climate feedbacks and cloud-aerosol interaction, and as the heart of superparameterized climate models. CRMs are computationally demanding, placing practical constraints on their use in these applications, especially for long, climate-relevant simulations.

In many situations, the horizontal-mean atmospheric structure evolves slowly compared to the turnover time of the most energetic turbulent eddies. We use this time scale separation to accelerate the time-integration of a CRM, the System for Atmospheric Modelling. Our approach uses a large time step to evolve the horizontally averaged state variables, followed by a short time step to calculate the turbulent fluctuations about the mean state. Using this approach, we are able to accelerate the model evolution by a factor of 8 or more in idealized stratocumulus, shallow and deep cumulus convection without substantial loss of accuracy in simulating mean cloud statistics and their sensitivity to climate change perturbations. We show how to adapt the approach to challenges arising from rapidly falling precipitation and from advecting scalars with a variety of lifetimes.