WeatherNext models

WeatherNext is a family of global, medium-range atmospheric models developed by Google DeepMind and Google Research, leveraging machine learning to significantly improve forecast accuracy and efficiency.

We have released two generations of WeatherNext models:

  • WeatherNext 2: Our state-of-the-art weather model that improves upon WeatherNext 1 with improved temporal resolution and better accuracy.
  • WeatherNext 1: Represents our original Graph and Gen models, which have been shown to be more skillful than ECMWF's HRES and ENS models.

We recommend starting with WeatherNext 2 for all new projects. It is our state-of-the-art model; WeatherNext 2 surpasses WeatherNext Gen on 99.9% of variables (e.g. temperature, wind, humidity) and lead times (0-15 days). The WeatherNext 1 models are considered legacy but remain available as a valuable reference and for users who need to compare results against our original published research.

You can learn more about core WeatherNext use cases on the Use Cases page.

WeatherNext 1 WeatherNext 2
WeatherNext Graph WeatherNext Gen
Historical Data Coverage 2020 - present 2020 - present 2022 - present
Architecture Graph Neural Network (GNN) implemented as a "encoder-processor-decoder" system Conditional Diffusion model incorporating GNN coupled with graph transformer Functional Generative Network (FGN) uses a graph transformer framework optimized with noisy weights
Spatial Resolution 0.25° (~30km at the equator) 0.25° 0.25°
Temporal Resolution 6 hours 12 hours 6 hours
See Vertex AI for access to 1hr timestep experimental capabilities
Forecast Initialization Frequency Times (UTC) Every 6 hours (00, 06, 12, 18 UTC) Every 6 hours (00, 06, 12, 18 UTC) Every 6 hours (00, 06, 12, 18 UTC)
Lead Times (Forecast Horizon) 10 days 15 days 15 days
Locations Global Global Global
Training data ERA5/HRES-fc0 ERA5/HRES-fc0 ERA5/HRES-fc0
Initialization Inputs for Generating Forecasts HRES-fc0 HRES-fc0 HRES-fc0
Forecast Type Deterministic forecast Ensemble forecast (50) Ensemble forecast (64)
Variables Atmospheric:
  • Geopotential
  • Specific humidity
  • Temperature
  • U component of wind
  • V component of wind
  • Vertical velocity
  • Each at pressure levels: 50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000 hPa
Surface:
  • 2 metre temperature
  • 10 metre u wind component
  • 10 metre v wind component
  • Mean sea level pressure
  • Total precipitation
 Atmospheric:
  • Geopotential
  • Specific humidity
  • Temperature
  • U component of wind
  • V component of wind
  • Vertical velocity
  • Each at pressure levels: 50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000 hPa
Surface:
  • 2 metre temperature
  • 10 metre u wind component
  • 10 metre v wind component
  • 100 metre u wind component
  • 100 metre v wind component
  • Mean sea level pressure
  • Total precipitation
  • Sea surface temperature
 Atmospheric:
  • Geopotential
  • Specific humidity
  • Temperature
  • U component of wind
  • V component of wind
  • Vertical velocity
  • Each at pressure levels: 50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000 hPa
Surface:
  • 2 metre temperature
  • 10 metre u wind component
  • 10 metre v wind component
  • 100 metre u wind component
  • 100 metre v wind component
  • Mean sea level pressure
  • Total precipitation
  • Sea surface temperature

See Glossary for more information on terms.