Machine Learning Enhancements to NASA Weather and Climate Modeling
Numerical computations and simulations representing many complex and interrelated physical processes are significant drivers of modern scientific progress. In recent years, machine learning (ML) applications in weather and climate simulation and prediction have promised to accelerate the accuracy and computational processing power of Earth System Models (ESMs). NASA is applying ML applications agency-wide for projects supporting mission objectives. Efforts include building foundation models for weather and climate; developing radiative-transfer emulators; and studying planetary science and exploration, biogeochemical processes and oceanography, hydrology, and atmospheric science. A major focus at NASA Goddard Space Flight Center has been making a variety of ML enhancements to NASA’s Goddard Earth Observing System (GEOS) atmospheric model — a weather and climate ESM developed for analyses and weather forecasts of Earth’s atmosphere, climate simulations and predictions, and chemistry-resolving simulations.
NASA's high-performance computing resources have enabled the development of ML models that enhance the representation of cloud formation and atmospheric composition. By analyzing extensive observational and high-resolution simulation data, these models improve predictions of atmospheric chemistry and physics, as well as the associated impacts on effective radiative forcings, which are crucial for understanding and addressing climate change.
Katherine H. Breen,
NASA Goddard Space Flight Center
Quick Facts
- Researchers use GPU resources on the NASA Center for Climate Simulation’s Discover and Prism high-performance computing systems to develop, train, and deploy ML models integrated with legacy ESM systems.
- The team has published novel work that enhances and physically constrains the representation of cloud formation and atmospheric composition — components that are essential to predicting the effective radiative forcing of Earth’s atmosphere.
- Successfully integrating this work into GEOS and other NASA ESMs provides enhanced predictions based on sophisticated, physically constrained ML models.
- Future work will include using foundation models for downstream tasks related to NASA mission objectives and continuing enhancements to existing ESM systems.
Researchers
- Katherine H. Breen, NASA Goddard Space Flight Center Please enable Javascript to see email address.
- Donifan Barahona, NASA Goddard Space Flight Center Please enable Javascript to see email address.
