SCREAM, the Simple Cloud Resolving E3SM Atmosphere Model, is the first global atmospheric model capable of using graphics processing unit (GPU) architectures to run efficiently at a global resolution of 3 kilometer grid cell length. Incorporating state-of-the-art methods for representing fluid dynamics, microphysics, moist turbulence, and radiation, SCREAM is a full-featured atmospheric general circulation model developed for very fine-resolution simulations.
Thanks to SCREAM’s computationally-efficient, performance-portable design, the model has become—as far as its development team is aware—the first nonhydrostatic global atmospheric model with resolution finer than 5 kilometers (km) to run on an exascale supercomputer, to run at scale on both NVIDIA and AMD GPU systems, and to exceed one simulated-year-per-day of throughput. Nonhydrostatic effects increase model accuracy but also computational time; as a result, they are not always included in global cloud-resolving models.
Conventional climate models use grid cells from 25 to 100 km to a side. One key advantage of higher resolution is the ability to capture fine-scale extreme weather events such as deep convective thunderstorms, which provide a large fraction of tropical rainfall and drive global circulations, leading them to be a leading cause of climate change uncertainty. SCREAM also captures atmospheric rivers, which bring heavy precipitation to the west coasts of North America and northern Europe and are therefore of major interest to the US government. Beyond storms, higher resolution is also important for accurately resolving coastlines and mountains, which is important for accurate simulation of winds and precipitation. Modeling of the timing of precipitation and the distribution of light versus heavy rainfall is also greatly improved.
SCREAM is one of only a few global atmosphere models to be ported to GPUs and the first to run on nearly all nodes of an exascale system. In 2023, the SCREAM model completed some of the world’s first year-long climate simulations at cloud-resolving resolution, making use of Oak Ridge National Laboratory’s 1.2 exaFLOP (1.2 quintillion computing operations per second) Frontier machine. In recognition of this achievement, a group of SCREAM scientists was awarded the inaugural 2023 Gordon Bell Prize for Climate Modeling for “innovative parallel computing contributions toward solving the global climate crisis.”
Coming soon
In addition to continuously improving SCREAM’s accuracy and efficiency, the team is exploring the feasibility of using machine-learning emulation techniques to achieve fine-resolution output more rapidly and with lower computational demand.
