HPC Support for Weather, Climate, and Ocean Dynamics Digital Twins Experiment

Earth Science Focus Area(s): Ecosystem Change; Extending and Improving Weather and Air Quality Forecasts; Reducing Climate Uncertainty and Informing Societal Response. Project Objectives: Provide accelerated high performance computing (HPC), machine learning models and model construction workflows, analysis, and visualization support for demonstration projects that seek to build weather, climate and ocean dynamics focused modeling components that can rapidly infer system state based on user parameters and requirements to enable Earth System Digital Twins (ESDT) efforts. State-of-the-Art: The NASA Advanced Supercomputing (NAS) facility is well positioned to provide the scalable, accelerated computing and data required to build large-scale, complex weather, climate, and ocean dynamics surrogate models useful in ESDT efforts. In addition to providing a 300-square-foot, 128-screen hyperwall for scientific discovery, analysis, and diagnostics, the NAS hosts computing platforms such as the new 350-node GH200 and 128-node Cabeus general purpose graphical computing unit (GPGPU) systems. The facility has significant earth science-oriented data holdings [1], such as the estimating the Circulation and Climate of the Ocean (ECCO) model outputs, a wide array of Weather Research Forecast (WRF) high resolution numerical model simulations, observational data from the Surface Water and Ocean Topography (SWOT), and Modern-Era Retrospective analysis for Research and Applications (MERRA-2) model outputs to name a few. Additionally, the NAS has the technical expertise to run computationally intensive machine learning training cost-effectively, enabling teams to construct surrogate models on select fields within the ”traditionally simulated” data holdings. While the current capability is note-worthy, groups who strive to build both data-drive and physics informed surrogate models that can be used in digital twin modeling scenario environments (DTE) remain challenging. Users have expressed interest in the rapid prediction and generation of fields in the weather domain (winds, precipitation, surface temperatures, clouds fields), climate (surface temperature, surface humidity, and winds), ocean dynamics (sea surface height, UV horizontal speed, and vorticity) at regional scales, which may be used as building blocks for component models to enable DTEs. Technical Approach: The team will combine high-performance accelerated computing, additionally leveraging the current NAS data holdings, and work with subject matter experts to build out a general-purpose data-driven surrogate model constructor for selected WRF, Climate, and ECCO fields, such as temperature, pressure over the GOES-WEST (18) Mesoscale region 1, for training models. The workflows will not be specifically tied to this region, but this region will be used to test the approach. For the data-driven modeling approach, the group will build upon existing transformer model architectures for component models and, pending success, review and document the feasibility of a semi-automatic physics-informed surrogate model construction workflow on select fields within the data holdings. All models developed will be placed on the NAS data portal (data.nas.nasa.gov). Additionally, workflows will be “wrapped” in an agent-like architecture with a rudimentary API (internal to the NAS environment), which may be called from the hyperwall console and/or integrated into a Digital Twin Environment. Technology Impact: Providing teams with workflows for training data-driven surrogate models, and a potential approach to physics informed surrogate models on NAS GPGPU systems will significantly reduce time for model construction. Constructed models can improve fidelity and information content of Digital Twins Environment, leading to better science, decision making, and understanding of system dynamics. The capability seeks to be independent and general enough so that it can integrate into Science Mission Directorate (SMD) projects that can benefit from rapidly build and deploying surrogate models in “raw” or agent form elsewhere. Follow-on Funding and Infusion: The NAS team may support any Digital Twin team deploy versions of the machine learning-based surrogate models, constructed on the NAS, for inference on modest Earth Science systems in the field. The model agent(s), and the workflows used to build them, will maintained on the HPC system and may be used for follow-up research on future surrogate model development.  

Advance Earth system science knowledge through the identification, development, and demonstration of innovative information systems technologies

This is early/mid-stage (TRL 4) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.