AI Innovations: Project EAGLE Q&A: NOAA Physical Sciences Laboratory
News
AI Innovations: Project EAGLE Q&A
AI innovations: Q&A with PSL’s Sergey Frolov on NOAA’s Project EAGLE
Adobe Stock montage: BerkahStock; suteeda; gonin
A team of experts in
NOAA Research
, the
National Weather Service
, and the
Earth Prediction Innovation Center (EPIC)
recently
unveiled a
new, experimental environment
to rapidly test and demonstrate artificial
intelligence (AI) weather models in near-real time:
Project EAGLE
Project EAGLE stands for
xperimental
lobal and
imited-area
nsemble forecast system and is a notable advance in NOAA’s
AI weather prediction innovation capabilities.
Traditional weather models can take several years to develop and require expensive supercomputers to run.
Alternatively, AI models have significantly shorter development time and greatly-reduced computing cost.
However, before becoming part of a weather prediction system, AI models need to be tested thoroughly to see
if they are skillful or need more training to get there.
Project EAGLE, once fully developed, will provide the
fundamental platform and
infrastructure
for researchers both inside and outside of NOAA to run, test, and demonstrate
their AI weather prediction models against established NOAA forecasting systems in real time.
PSL has been heavily involved in the development of training datasets for AI models, such as the
UFS-Replay
dataset, the
20th Century Reanalysis
, and the
NNJA
observational archive, so the lab is uniquely positioned to be
part of Project EAGLE.
Sergey Frolov
, Chief Scientist for Project EAGLE and lead of
PSL’s Reanalysis and Data Assimilation team, answered a few questions to provide some additional
insights into this new initiative:
How did the idea for Project EAGLE come about?
“The existing research-to-operations pipelines at NOAA were not designed to handle the rapidly evolving
landscape of AI models. Several skillful NOAA AI models that had been already developed needed an easier
path to show this skill. An
AI
for Numerical Weather Prediction workshop
held in 2023 identified the
need for this path and a more organized effort.
“The Project EAGLE team drew inspiration for this new platform from our experience working with private
sector companies and existing open source projects which can deliver and demonstrate the skill of AI
products with significant higher development velocity compared to traditional numerical weather modeling
systems.”
What do you see as the major benefits of EAGLE?
“The research-to-demonstration-to-operations pipeline for AI weather models that EAGLE provides is
unprecedented in government. It should drastically increase the speed with which NOAA can develop,
streamline, and operationalize AI models and innovations to improve our weather forecasts.
“EAGLE will also provide a robust training and testing platform for scientific and weather enterprise
communities to start developing their research or commercial products from a well-supported and
understood AI platform that incorporates NOAA’s trusted metrics.”
What are Project EAGLE’s biggest challenges?
“Growing EAGLE into a sustained enterprise will require funding, collaboration, and trust-building across
NOAA and with external partners. It was relatively easy to establish this initial stage of EAGLE; now we
have to develop a process that will build on that to grow the capabilities of the platform and foster
continuous innovation and improvement of AI-powered forecasting.”
What’s next for Project EAGLE?
“The initial version of the platform consists of two test environments: Global-EAGLE-Solo for deterministic
models and Global-EAGLE-Ensemble for ensemble models. This allows researchers to test models that generate a
single forecast (deterministic) as well as models that generate multiple (ensemble). Within the year, we
will be integrating NOAA’s new
HRRR-Cast
, an AI-version of NOAA’s operational HRRR system developed by the
Global Systems Laboratory
that will greatly enhance the
high-resolution regional capabilities of both EAGLE environments.
“Further enhancements to EAGLE are on the way. PSL specifically is contributing to the future of EAGLE by
developing ocean and ice AI modeling capabilities. PSL is also developing an AI global model with grid
refinement over the U.S. that will further improve EAGLE’s level of detail of the Nation. The larger EAGLE
team led by
EPIC
also hopes to soon release a
separate, supported AI-training platform that will connect NOAA and U.S.-centric datasets to
Anemoi
, the leading AI for weather modeling library from the European Centre for
Medium-Range Weather Forecasts and several of its Member States. This will strengthen collaboration on AI
enhancements between NOAA Research labs and the National Weather Service, and between NOAA and peer weather
forecasting centers in Europe, leading to improved performance of EAGLE models and faster transition of
research to operations.
“Ultimately, EAGLE has a chance to become an integral platform within the weather enterprise in the U.S. This
will include not only a wide portfolio of AI weather and ocean forecast products and a robust way to
evaluate these AI models, but also a way for NOAA to effectively and efficiently share its massive data
holdings to AI-training pipelines across the U.S. weather enterprise.”
Learn more about Project EAGLE’s technical details:
Visit the EAGLE project page:
Posted: August 8, 2025
Back to PSL News
Signup for PSL email updates