NASA Winter Modeling Experiments Push the
Boundaries of Simulation to Explore the Earth
Computer models are essential tools for understanding and predicting weather and climate. Together with observations they give a more complete picture of the changing Earth system. At NASA, models support everything from satellite instrument teams and field campaigns to research on timescales ranging from days to centuries. Model-based research can involve participating in international assessments and experiments. One of the latest such endeavors is the DYAMOND Initiative: DYnamics of the Atmospheric general circulation Modeled On Non-hydrostatic Domains.
The DYAMOND Initiative centers on comparing global storm-resolving models, which use grid boxes 5 kilometers (km) wide and below. Two phases of DYAMOND experiments have explored where simulations from different model groups agree and provide deeper insights into the Earth’s climate system, how sensitive the simulations are to a particular implementation, and what performance and analysis bottlenecks arise.
A team of scientists from the NASA Goddard Space Flight Center, NASA Jet Propulsion Laboratory (JPL), and Massachusetts Institute of Technology (MIT) formed one of 15 model groups contributing to Phase II of the DYAMOND Initiative. The global experiments focused on simulating northern hemisphere winter over the 40-day period January 20, 2020 to March 1, 2020. NASA High-End Computing (HEC) resources hosted several global storm- and eddy-resolving simulations for DYAMOND Phase II, also known as DYAMOND Winter.
“These simulations explicitly resolve atmospheric convection and Coriolis-induced turbulence in the ocean, producing better simulations of clouds and precipitation in the atmosphere, deep water formation and global currents in the oceans, and the fluxes at the interface of the oceans and atmosphere,” said William Putman, Associate Chief, NASA Goddard’s Global Modeling and Assimilation Office.
NASA’s DYAMOND Phase II/Winter simulations used the NASA Goddard Earth Observing System (GEOS) model for the atmosphere and the MIT general circulation model (MITgcm) for the ocean in three configurations:
- Coupled Atmosphere-Ocean: A coupled 6-km, 72-level atmosphere and 4-km, 90-level ocean with interactive, two-moment aerosol cloud microphysics (later extended to a 14-month simulation).
- Atmosphere+Carbon: A 3-km, 181-level atmosphere with single-moment, 6-phase cloud microphysics including 1-km global carbon emissions for chemistry transport.
- Atmosphere: A 1.5-km, 181-level atmosphere with simple parameterized chemistry.
The simulations ran on the NASA Center for Climate Simulation (NCCS) Discover supercomputer and the NASA Advanced Supercomputing (NAS) Division Pleiades and Aitken supercomputers, harnessing up to 39,400 processor cores. Computational usage and performance for each simulation configuration were as follows:
Configuration | Total Cores Supercomputer |
Throughput | Data Volume |
---|---|---|---|
Coupled Atmosphere-Ocean | 8,160 Intel Xeon Haswell NAS Pleiades and Aitken |
3 simulated days per wallclock day |
0.3 petabytes |
Atmosphere+Carbon | 39,360 Intel Xeon Skylake NCCS Discover |
7 simulated days per wallclock day |
2.0 petabytes |
Atmosphere | 39,440 Intel Xeon Skylake NCCS Discover |
1.5 simulated days per wallclock day |
1.3 petabytes |
“NASA High-End Computing platforms at NCCS and NAS are providing essential resources for preparing our modeling systems for emerging HEC systems and exploring new science developments,” Putman said. “These partnerships support our exploration of traditional hybrid parallelism approaches within GEOS today and prepare our models for future HEC platforms by exposing new depths of parallelism within our model components.”
On the atmospheric modeling front, NASA Goddard is applying similar modeling techniques to explore the interactions of more than 200 chemical species in the global composition forecasting (CF) system called GEOS-CF. Scientists are preparing GEOS-CF for 3-km simulations — the highest-resolution, complex, global composition forecasts to date.
With coupled atmosphere-ocean modeling, the NASA Goddard-JPL-MIT team is distributing hourly atmospheric and oceanic model output from the full 14-month simulation to the science community while continuing their own data analysis. They are also planning an even higher-resolution simulation that explicitly represents ice shelf cavities and the ocean carbon cycle.
Related Links
- GEOS DYAMOND Phase II, GMAO Research Site.
- Pushing the Boundaries of Simulation to Explore the Earth, NASA@SC21.
- A Global, Coupled Ocean-Atmosphere Simulation with Kilometer-Scale Resolution, NASA@SC21.
Jarrett Cohen, NASA Goddard Space Flight Center