Design and First Applications of Cleo, a Flexible Superdroplet Model for Warm-Cloud Physics on High-Performance Computers

Abstract

Eulerian bulk and bin models for cloud microphysics hinder our understanding of warm tropical clouds in ways the Super-Droplet Method (SDM) can overcome. Microphysics-related uncertainties across global models and Large Eddy Simulations (LES), as well as discrepancies from observations, afflict precipitation patterns, extremes, radiation budgets, and cloud feedbacks. They arise not only from insufficient knowledge of cloud microphysics, but also modelling choices which bulk and bin models make to represent hydrometeors and microphysical processes. SDM, being more fundamental and Lagrangian, offers many advantages including eliminating these choices and facilitating exploration of knowledge gaps. However, the computation costs of current SDMs preclude their application to large enough domains for elucidating the microphysics-mediated coupling between shallow cumulus organisation and precipitation. We create Cleo, a novel C++ SDM implementation for high-performance computers, seeking to make larger domain SDM computationally feasible. Cleo’s memory layout is designed to minimise memory allocations and optimise access patterns. We use Kokkos (shared memory) and Message Passing Interface (distributed memory) parallelism, and couple to fluid-dynamical drivers via Yet Another Coupler, which allows total independence of their grids and domain decompositions. Thus Cleo enables efficient resource allocation as well as portability across different high-performance computer architectures. Cleo shows ideal superdroplet-scaling and reasonable strong-scaling and is ready for its final stage of development for large-domain LES: higher efficiency particle transport across distributed memory. Equally important as the computational performance is that Cleo flexibly models warmcloud microphysics. In up to 3-D simulations Cleo can represent all the principal microphysical processes driving warm-clouds; and Cleo’s design enables users to easily compile an arbitrary number and combination of different processes and data outputs, as well as different options for uncertain details of microphysics, e. g. terminal velocity parametrisations and collision kernels. We offer two ways to model droplet collisions: the canonical SDM algorithm and an extended version which includes our novel representation of collisional breakup. Cleo is therefore a fully-fledged SDM, whose flexibility facilitates detailed study of warm-cloud microphysics. We apply Cleo to constrain uncertainty in precipitation relevant to shallow cumulus organisation. First, we examine the sensitivity of precipitation in the Kinematic Driver (KiD) framework to particular modelling choices in SDM. We find less sensitivity to the choices of superdroplet initialisation, collision efficiencies, and the number of superdroplets than previous studies supposed. Second, we constrain the amount of and controls on sub-cloud rain evaporation from shallow cumuli. In agreement with previous studies, microphysics rather the thermodynamic environment predominantly determines such rain evaporation. Furthermore, droplet collisions have a negligible impact on rain evaporation whereas the ventilation effect can increase it by up to a factor of 10. Both studies advance warm-cloud precipitation modelling and encourage using Cleo to better understand shallow cumulus organisation. This dissertation constitutes substantial progress for modelling warm-clouds through the development and application of a novel high-performance SDM. It opens the door to deciphering the role of microphysics in shallow cumulus organisation from a fundamentally different, process-based perspective. Beyond warm-clouds, Cleo’s design makes it easily extendable to other particle-based components of our climate system, to clouds on exoplanets, and to help bridge the gap between field observations and cloud microphysics modelling.