American Geophysical Union (AGU) 2025
December 15 – 19, 2025
Earth Prediction Innovation Center (EPIC) is excited for the American Geophysical Union (AGU) Annual Meeting, taking place December 15 – 19, 2025. This meeting is the most influential event in the world dedicated to the advancement of Earth and Space Sciences.
AGU25 will be held in New Orleans, LA, and online. More than 25,000 attendees from over 100 countries will convene to explore the theme, “Where Science Connects Us.” The meeting will host a diverse community of scientists, students, journalists, policymakers, educators, and organizations who are coming together to share, inspire, collaborate, and engage as a united community grounded to better understand our planet and environment, opening pathways to discovery, opening greater awareness to address climate change, opening greater collaborations to lead to solutions and opening the fields and professions of science to a whole new age of justice equity, diversity, inclusion, and belonging.
Abstracts
Zach Shrader
Presentation Date: Monday, December 15, 2025; 8:30 AM- 12:00 PM CDT
Location: Hall EFG (Poster Hall) (New Orleans Convention Center)
The Earth Prediction Innovation Center (EPIC) is dedicated to accelerating contributions to the Unified Forecast System (UFS) by engaging the community and providing community members with the necessary tools and knowledge to contribute innovations to our Nation’s forecasting and modeling systems. The EPIC Community Engagement (ECE) team supports EPIC’s innovation through several initiatives, including community training events and the annual Unifying Innovations in Forecasting Capabilities Workshop (UIFCW). It also publishes dynamic content on social media, the EPIC Community Portal (ECP), GitHub, and other platforms to meet the community’s evolving needs and to remove barriers to innovation.
As part of its role in promoting community work, ECE has supported outreach efforts for emerging initiatives such as the Experimental AI Global and Limited-area Ensemble forecast system (EAGLE-AI) by assisting with infographic design and helping publish blog content to the ECP, broadening the reach of early releases and updates. The User Support (US) team complements ECE’s efforts by updating UFS application documentation, compiling technical FAQs, and monitoring support requests. The US team provides ECE with immediate feedback on community engagement efforts, which allows ECE to adjust content and outreach strategies based on community needs and requests. This collaboration results in a responsive support environment that rapidly evolves to meet the community’s needs, ensuring that users and developers have the support they require to innovate within our national forecasting systems.
Keven Blackman & D. Alex Burrows
Presentation Date: Tuesday, December 16, 2025; 2:15 PM – 5:45 PM CST
Location: Hall EFG (Poster Hall) (New Orleans Convention Center)
The Earth Prediction Innovation Center (EPIC) is dedicated to accelerating advancements in the Unified Forecast System (UFS) by actively engaging with the community and equipping its members with the essential tools and knowledge necessary to drive innovations in our nation’s forecasting and modeling systems. The EPIC Community Engagement (ECE) team fosters innovation through various initiatives, including community training events and the annual Unifying Innovations in Forecasting Capabilities Workshop (UIFCW). Additionally, the ECE team disseminates dynamic content via social media, the EPIC Community Portal (ECP), GitHub, and other platforms to meet the community’s evolving needs and eliminate barriers to innovation.
The User Support (US) team complements the efforts of ECE by continuously updating UFS application documentation, compiling technical FAQs, and monitoring support requests. This team provides immediate feedback to ECE on community engagement efforts, enabling ECE to refine its content and outreach strategies based on community needs and requests. This collaboration results in a highly responsive support environment that evolves rapidly to meet the community’s needs, ensuring that users and developers have the necessary support to innovate within our national forecasting systems.
Furthermore, EPIC aims to create an inclusive and collaborative atmosphere where diverse perspectives are valued, and innovative ideas can flourish. By leveraging the strengths and feedback of the community, EPIC is committed to driving continuous improvement and excellence in weather prediction, ultimately enhancing the accuracy and reliability of forecasts that are vital to public safety and economic stability.
Alex Burrows
Presentation Date: Tuesday, December 16, 2025; 4:40 PM – 4:50 PM CST
Location: 275-277 (New Orleans Convention Center)
With the recent ubiquitous development of Artificial Intelligence/Machine Learning (AI/ML) algorithms to complement Numerical Weather Prediction (NWP), Earth Prediction Innovation Center (EPIC), along with other NOAA agencies such as the Environmental Modeling Center (EMC), the Physical Sciences Laboratory (PSL), and the Global Systems Laboratory (GSL), has been experimenting with global AI/ML forecast systems. To support this effort within NOAA, EPIC is considering multiple frameworks for forecast verification, including EMC’s Verification System (EVS) and GSL’s weather verification system (wxvx). EVS and wxvx are both wrapper workflows that support the Model Evaluation Toolkit (MET) and METplus.
While EVS is essentially hardcoded on the National Centers for Environmental Prediction’s (NCEP) Weather and Climate Operational Supercomputing System (WCOSS) supercomputer, which requires nearly a dozen spack-type modules installed, and runs 31-day and 90-day verification, wxvx was designed to be flexible, handling output from physically based and AI/ML-based forecast systems through the use of YAML configuration files. Additionally, since it provides pre-built MET/METplus executables for Linux systems (via its sibling met2go package), it eliminates the need for dozens of module loads, making it portable to NOAA, non-NOAA, and cloud provider machines. Additionally, through YAML configurations, the output statistics can be adjusted to accommodate the number of forecast cycles and forecast lengths.
Specifically, EPIC will use wxvx to verify model output from the ECMWF’s Anemoi framework. This talk will focus on verification with wxvx, describing the current state and future plans. The current capability of wxvx includes global and regional verification for grid-to-grid comparison of forecasts. In active development is the extension to grid-to-obs verification. In the future and in support of EPIC’s community framework, various verification statistics will be hosted at EPIC’s web portal (epic.noaa.gov) along with Anemoi forecast output.
Mariah Pope
Presentation Date: Wednesday, December 17, 2025; 8:30 AM – 12:00 PM CST
Location: Hall EFG (Poster Hall) (New Orleans Convention Center)
The proliferation of data-driven weather prediction has produced several models that rival the skill of traditional numerical weather prediction (NWP) at a fraction of the cost. However, retraining of the published models with custom (NOAA) data can be a cumbersome process due to hardware setup, staging data, and efficiently scaling the model. This burden creates a steep learning curve for users, which acts as a barrier to rapid scientific iteration and progress. We present a framework of community tools for developing machine learning weather models using ufs2arco from the NOAA Physical Sciences Laboratory (PSL) and Anemoi from the European Centre for Medium-Range Weather Forecasting (ECMWF) to guide users through training their models using a combination of NOAA and ECMWF datasets. First, the ufs2arco package enables users to quickly create large training datasets from existing data in the cloud, and is specifically designed to improve access to NOAA data. Anemoi then provides an easily customizable framework for developing and testing graph-based machine learning weather prediction models. Together, ufs2arco and Anemoi establish a comprehensive, data-driven framework that we can currently run using either on-prem or cloud-based resources. We guide users through environment setup, data preparation, model training, and inference execution. Finally, we demonstrate the efficiency and flexibility of this framework by providing benchmark metrics for computational speed and cost. Overall, this work makes NOAA data easily accessible for machine learning model development, and makes the ECMWF Anemoi framework more readily deployable.
Jong Kim
Presentation Date: Wednesday, December 17, 2025; 2:15 PM – 5:45 PM CST
Location: Hall EFG (Poster Hall) (New Orleans Convention Center)
The community Noah-MP land surface model is a critical component supporting the Unified Forecast System (UFS) applications, including the global scale medium-range and sub-seasonal to seasonal forecast applications. Currently, both within-component and between-component land model coupling strategies are applied in the UFS Weather Model system. Based on the Joint Effort for Data assimilation Integration (JEDI), a collaborative effort is currently underway to develop the UFS Noah-MP Land Data Assimilation (DA) System and to facilitate the research-to-operations (R2O) transition by using various workflow configuration tools aligned with the NOAA operational forecast system update. In this presentation, we discuss ongoing efforts to develop the UFS land model and DA evaluation framework, including the JEDI-based land DA and workflow capabilities, the land model coupling strategy, observation data processing, analysis baseline cross-comparison, configuration tools, and computational platform deployment. A successful UFS land collaboration and coordination effort will expedite community involvement in land model and DA development and contribute to further investigation of land-atmosphere interactions with other components in a coupled global and regional Earth system models.
Kris Booker & Anna Kimball
Presentation Date: Wednesday, December 17, 2025; 2:15 PM – 5:45 PM CST
Location: Hall EFG (Poster Hall) (New Orleans Convention Center)
The NOAA Earth Prediction Innovation Center (EPIC) program has enabled an accelerated pace of numerical weather prediction innovation through its collaboration with government, academic, and enterprise sectors. Utilizing a common modeling framework, the UFS (Unified Forecast System), has enabled once-siloed atmospheric modeling groups to collaborate effectively. However, the greater challenge has come from constraints that kept organizations from accessing high-performance computing platforms capable of running new modeling innovations at scale. EPIC has solved this dilemma by providing UFS community contributors with public access to detailed insights from these high-performance computing (HPC) platforms. Every contribution considered for integration is extensively tested through a CI/CD pipeline on various NOAA research and development high-performance computing systems (RDHPCS) platforms, which have performance capabilities similar to those of the Weather and Climate Operational Supercomputing System (WCOSS-2). Systematically capturing data such as walltime and core hours for each model run provides contributors with detailed insights into computational efficiency and resource usage. This enables the community to track the evolution of model updates over time, offering quick insight into how performance and efficiency improve successively. The design supports future integration of scientific forecast skill, allowing users to correlate computational costs with forecast accuracy. The data collected is processed and displayed to contributors through the EPIC Community Portal, creating more transparency between EPIC and the broader community. This openness empowers contributors, researchers, and developers to identify trends, compare model versions, and contribute to continuous model refinement. Ultimately, strengthening the feedback loop between operational teams and the user community accelerates improvements in forecasting quality and operational efficiency.
Ben Koziol
Presentation Date: Thursday, December 18, 2025; 2:15 PM – 5:45 PM CST
Location: Hall EFG (Poster Hall) (New Orleans Convention Center)
The Unified Forecast System (UFS) and Short-Range Weather App (SRW) combine NOAA’s operational weather forecasting capabilities with enhanced community support, documentation, and outreach. The UFS community recently incorporated UFS Fire and the associated Community Fire Behavior Model (CFBM) component as an optional atmosphere-coupled modeling configuration. UFS Fire can operate in one-way (atmosphere-to-fire) or two-way (atmosphere-fire feedback) modes. In addition, the SRW introduced the Rapid Refresh Forecast System’s Smoke/Dust module (RRFS-SD) as an optional one-way atmosphere-coupled forecast task. RRFS-SD provides forecasts of smoke and dust dispersion across North America, and uses data on wildfire emissions and dust sources to predict near-surface and vertically integrated concentrations. RRFS, also considered part of the UFS, is NOAA’s next-generation convection-allowing weather forecasting system currently under evaluation for release next year. This poster provides background on the UFS and SRW, with a focus on the new fire and smoke/dust modeling capabilities. Also included are technical descriptions, instructions on accessing the functionality, and plans for future component developments.
