Artificial Intelligence for Numerical Weather Prediction (AI4NWP) Workshop
On November 28 and 29, 2023, the National Oceanic and Atmospheric Administration (NOAA) hosted a workshop in Boulder, Colorado, to identify emerging opportunities in machine learning tools for numerical weather prediction (NWP). Attendees were key partners in NOAA line offices, the private sector, and peer NPW centers.
The workshop was sponsored by NOAA’s Physical Sciences Laboratory (PSL) and Global Sciences Laboratory (GSL), the Cooperative Institute for Research In Environmental Sciences (CIRES ), and NOAA National Weather Service (NWS).
Goals
The workshop aimed to:
- Identify a research and development roadmap and priorities for integrating emerging AI tools for NWP into the NOAA production pipeline.
- Match priorities with existing datasets for training, and identify gaps in training datasets.
- Identify potential partnerships between NOAA and external collaborators.
- Define the scope and focus of investment areas required to sustain a minimally viable research initiative in this area (e.g. HPC, reanalyses, etc.).
More on AI4NWP
Over the last 18 months, a number of revolutionary machine learning tools have emerged in the private industry that are poised to fundamentally alter the economics of numerical weather prediction . This includes the GraphCast model from Google DeepMind, FourCastNet from NVIDIA, and PangGu Weather from Huawei.
All of these models have been trained on a publicly available ERA5 reanalysis dataset from European Centre for Medium-Range Weather Forecasts (ECMWF) and are now shown to produce competitive forecasts to the headline HIRES model from ECMWF. More on machine learning and weather forecasting is available on the ECMWF Science Blog .
Day 1 Presentations
| Presentation Title | Presenter | Materials |
|---|---|---|
| AI Models at ECMWF | Matthew Chantry, ECMWF | |
| NOAA Operations Suite Introduction | Brian Gross, NOAA NWS | |
| NOAA Available Data Sets | Sergey Frolov, NOAA PSL | |
| Research to Operational Cycle: Industry Perspective - Google DeepMind | Peter Battaglia, Google DeepMind | |
| Research to Operational Cycle: Industry Perspective - Microsoft Weather | Matt Corey, Microsoft | |
| Research to Operational Cycle: NOAA Perspective - R2O Process Overview | Jeff Whitaker, NOAA PSL | |
| Research to Operational Cycle: NOAA Perspective - Testbeds and Proving Grounds | Adam Clark, NOAA National Severe Storms Laboratory | |
| Breakout Session: Opportunities and Barriers for Integration of AI in the R2O Process | All participants |
Day 2 Presentations
| Focus on Skill Improvement Presentation | Presenter | Materials |
|---|---|---|
| Verification/Tests Beds | Ligia Bernardet, NOAA GSL | |
| Data Assimilation | Daryl Kleist, NOAA NWS | |
| Short Range, Convective Allowing Forecasts | Curtis Alexander, NOAA GSL | |
| Global Modeling MRW/S2S | Jeff Whitaker, NOAA PSL | |
| Reanalysis | Sergey Frolov, NOAA PSL | |
| HFIP | Kevin Garrett, NOAA NWS | |
| Breakout Session: Opportunities for Forecast Skill Improvement (Breakouts by Application) | All participants |
Workshop Report
A report from the workshop will be posted here when it becomes available.