A locally conservative transport scheme is designed for next-generation computing platforms.
Tracer transport is a fundamental process in any fluid model, including the Dept. of Energy's E3SM Atmosphere Model (EAM). In EAM v1, tracer transport is the most computationally expensive component of the model. With algorithms designed to map well to parallel distributed computing platforms, this work achieves a 2.2x speedup over the previous transport scheme without sacrificing any of its key traits such as conservation, accuracy, and shape preservation.
Climate models are routinely required to solve the tracer transport equation for approximately 40 to 100 separate tracer species, depending on specific science applications. The number of tracers makes transport the most computationally expensive part of the climate model. In addition to simply solving the transport equation, a transport scheme must also ensure that several properties such as global mass conservation and local bounds preservation are maintained, in order to assure physical realism. In this work, we develop a locally conservative transport scheme based on combining semi-Lagrangian time-stepping strategies with the compact, high order stencils of spectral element methods. This combination enables long time steps and local computations, which translate into improved computational efficiency and parallelism. Additionally, most of the algorithm is independent of the number of tracers, which makes it less expensive to perform simulations that require many tracer species. We demonstrate the scheme in the E3SM Atmosphere Model and show a 2.2x speedup over the model's current transport scheme.