WeatherNext 2

WeatherNext 2 is a global ensemble weather forecasting model developed by Google DeepMind and Google Research, based on the Functional Generative Network (FGN) architecture. It generates a full 64-member probabilistic ensemble in a single forward pass 8× faster inference than the diffusion-based WeatherNext Gen, while delivering superior accuracy across 99.9% of target variables and lead times. WeatherNext 2 was also extended for operational tropical cyclone forecasting, developed in collaboration with the National Hurricane Center, CIRA, and the UK Met Office. For more, see the announcement blog post.

At a glance

Attribute Details
Coverage Global
Spatial Resolution 0.25° (~25 km)
Temporal Resolution 6 hours
Forecast Horizon 15 days
Initialization Frequency Every 6 hours (00, 06, 12, 18 UTC)
Ensemble Members 64 members
Architecture Functional Generative Network (FGN) — graph transformer with noisy weights (paper)
Training Data ERA5 / HRES-fc0
Initialization Inputs HRES-fc0
Historical Forecasts 2022 – present

Supported variables

Surface variables

Variable Name Description Units
2m_temperature 2 metre temperature K
10m_u_component_of_wind 10 metre U (eastward) wind component m/s
10m_v_component_of_wind 10 metre V (northward) wind component m/s
100m_u_component_of_wind 100 metre U (eastward) wind component m/s
100m_v_component_of_wind 100 metre V (northward) wind component m/s
mean_sea_level_pressure Mean sea level pressure Pa
total_precipitation Total precipitation (6-hour accumulated) m
sea_surface_temperature Sea surface temperature K

Atmospheric variables (pressure levels)

These variables are available at 13 pressure levels: 50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000 hPa.

Variable Name Description Units
{level}_geopotential Geopotential height m2/s2
{level}_specific_humidity Specific humidity kg/kg
{level}_temperature Temperature K
{level}_u_component_of_wind U (eastward) wind component m/s
{level}_v_component_of_wind V (northward) wind component m/s
{level}_vertical_velocity Vertical velocity Pa/s

Replace {level} with the pressure level in hPa. For example, 500_geopotential for geopotential at 500 hPa.

See the Glossary for definitions of meteorological terms.

How to access

WeatherNext 2 forecast data remains available on BigQuery, Earth Engine, and Google Cloud Storage for existing workloads. For new projects and operational workloads, we recommend WeatherNext 3. To request access, fill out the WeatherNext Data Request form.