Short Course at AMS 2026: NOAA’s Earth Prediction Innovation Center – AI in Numerical Weather Modeling
On January 25, 2026, the National Oceanic and Atmospheric Administration (NOAA)’s Earth Prediction Innovation Center (EPIC) hosted a hands-on Artificial Intelligence (AI) Short Course during the 106th American Meteorological Society (AMS) Annual Meeting at the George R. Brown Convention Center in Houston, Texas.
This hybrid workshop brought together students, researchers, NOAA scientists, private sector professionals, and international collaborators to explore scalable AI weather prediction frameworks, operational transition pathways, and verification tools supporting next-generation forecast systems.
The short course emphasized practical implementation, enabling participants to run AI models using community notebooks and open-source tools that are actively shaping NOAA’s operational AI strategy.
Event Overview
Date: January 25, 2026
Format: Hybrid (In-Person + Virtual)
Location: George R. Brown Convention Center
Sponsors: NOAA EPIC, University Corporation for Atmospheric Research (UCAR) Cooperative Programs for the Advancement of Earth System Science (CPAESS)
The course provided participants with:
- Background on AI/ML for numerical weather prediction
- Hands-on execution of AI forecast models
- Model visualization and verification workflows
- Insight into NOAA’s AI Research-to-Operations (R2O) pipeline
- Overview of Project EAGLE (Experimental AI Global and Limited-area Ensemble) and AI integration with physics-based models
- Exposure to deterministic and ensemble AI forecasting systems
- Experience with end-to-end workflows (data prep, training, inference)
- Familiarity with NOAA-aligned verification metrics and HPC/cloud approaches
All notebooks used in the workshop are publicly available via:
GitHub – NOAA-GSL/ai-notebooks
AMS 2026 Short Course Repository
Participants were able to leverage Google Colab and cloud-based workflows, lowering barriers to experimentation and enabling rapid innovation using freely available computational resources.
AI Training
Contributors: EPIC: Keven Blackman, Mariah Pope, Andrew Justin, Natalie Perlin, Brandon Selbig, Anna Kimball, Aaron Jones
Download Presentation Slides
(Introductory slides): EPIC – AMS26 AI EPIC Short Course – Intro
(Training content): Artificial Intelligence in Numerical Weather Modeling Short Course at AMS 2026
The NOAA EPIC AI in Numerical Weather Modeling Short Course at AMS 2026 delivered a fully hands-on, end-to-end machine learning workflow for developing, generating, and verifying AI-driven weather forecasts.
Rather than focusing only on model theory, the course walked participants through the complete operational pipeline, from NOAA data preprocessing to model training, inference, post-processing, and verification, using open-source tools aligned with NOAA workflows. Attendees were able to execute each stage themselves, reinforcing EPIC’s mission to lower barriers to AI experimentation while maintaining scientific rigor and operational standards.
The training pipeline integrates community-developed and NOAA-supported frameworks to demonstrate how AI can complement traditional NWP by enhancing efficiency, scalability, flexibility, transparency, and reproducibility.
Strategic Impact
The EPIC AI Short Course demonstrated:
High demand for AI-driven weather modeling tools,
Readiness of NOAA’s AI R2O infrastructure,
Importance of verification transparency, and
Value of open-source community frameworks through integration of:
Anemoi training framework,
ufs2arco data access tools,
Project EAGLE operational pathway, and
wxvx verification infrastructure,
NOAA is building a unified ecosystem for AI powered weather prediction.
Post Event Take-Aways and Analysis
Attendance Summary
Participant Survey Feedback:
A post-event survey was shared with attendees with responses indicating high satisfaction with session relevance, schedule balance, presentation organization, and technical facilitation. Participants noted that presenters handled technical issues well and provided clear, structured, and practical notebook walkthroughs, though some expressed a desire for more hands-on time, particularly with the second AI notebook (“Would have liked some time to play around with the second notebook”). Overall, feedback was positive, with appreciation for the guidance provided (“Thank you for walking us through the tutorial! Felt like the presenters handled tech issues and other’s questions well.”), alongside requests for additional time to experiment independently.
Take-aways and Future Plans
Key Take-aways
Explore larger workshop spaces at future AMS meetings
Communicate Google Colab free credit availability in advance
Encourage participants to set up their Google accounts in advance
Action Items
Engage with Google regarding expanded Colab credits
Begin planning AI content for Unifying Innovations in Forecasting Capabilities Workshop 2026 (UIFCW26)
Improve pre-event technical guidance
Continuous Development Efforts at EMC
Ongoing enhancements of NOAA EAGLE include:
Deterministic System Improvements
Two major updates were introduced to the loss function:
- Spherical harmonic representation for loss function to improve hurricane intensity forecasts
Revised pressure-level weighting and varable scaling to better refine vertical atmospheric structure
Ensemble Expansion
- Adopt Continuous Ranked Probability Score (CRPS) loss function to increase ensemble spread (this is for global EAGLE ensemble)
Enhanced ensemble diversity and probabilistic forecasting capability
AI Training Framework Modernization
- Adoption of the Anemoi ML training framework
Shared architecture alignment with other operational centers
Streamlined Machine Learning Weather Prediction (MLWP) development
Coupled Global ML System Development
Work is underway to develop a global coupled MLWP system for sub-seasonal to seasonal forecasting:
- Atmosphere and ocean components use Spherical Fourier Neural Operators (SFNO)
Air-sea coupling mechanism is being designed
Target: extend AI forecasting skill into longer timescales
Project EAGLE represents NOAA’s most comprehensive effort to institutionalize AI weather modeling within its operational ecosystem.
Future development will focus on improving forecast performance and transitioning EAGLE into formal operational implementation.
