A library of large-eddy simulations with ClimateMachine.jl

Abstract:

Advances in high-performance computing capabilities have enabled large-eddy simulations (LES) of turbulence, clouds, and convection, but their potential to improve parameterizations in global climate models (GCMs) has not yet been harnessed. The ClimateMachine is a newly developed climate model that can run GCM simulations and limited-area LES, and it is built to run efficiently on Graphics Processing Units (GPUs). We use the ClimateMachine LES driven by large-scale forcings from a GCM at different locations to simulate clouds across dynamical regimes in both the current and warmer climates. A library of several hundred LES are run on GPUs on the Google Cloud Platform. The LES library is enabling systematic development and improvement of GCM parameterizations with data assimilation and machine learning techniques. It also provides an opportunity to analyze cloud feedbacks in LES with dynamically consistent changes in the large-scale dynamics.