SciGroveBeta
Climate

Scaling Storm-Resolving Atmospheric AI Simulation to the Entire Planet

Zeyuan Hu, Akshay Subramaniam, Noel Keen, Tao Ge, Jaideep Pathak, Mohammad Shoaib Abbas, Suman Ravuri, Karthik Kashinath, Naser Mahfouz, Peter Caldwell, Mike Pritchard, Noah Brenowitz

Featured July 4, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

This new AI model, STRATA, can predict global weather at a super-detailed, storm-level scale much faster and more efficiently than giant supercomputers, by cleverly breaking the world into small local pieces and then stitching them back together.

In depth
STRATA is an autoregressive AI emulator designed for global storm-resolving atmospheric dynamics. It leverages a tile-based training strategy and 3D local attention to efficiently simulate kilometer-scale weather phenomena, achieving significant energy efficiency gains over traditional physics models. A novel Stereographic Rotary Position Embedding (StereoRoPE) ensures grid-invariant position encoding, allowing the model to generalize across different grid topologies.

Key Takeaways

  • 1
    STRATA enables global storm-resolving AI simulation at 4.9-km resolution, a significant advance over previous regional or coarse-resolution AI models.
  • 2
    The model achieves 50 times better energy efficiency and 741 simulated days per wall-clock day compared to the SCREAM physics model.
  • 3
    Key innovations include tile-based training, 3D local attention, Stereographic Rotary Position Embedding (StereoRoPE), and a pixel-space de-aliasing decoder.

Conceptual Flow

HIGH LEVEL
1
How STRATA Simulates Global Weather

The model learns how small weather patches change over time, then stitches these learned changes together to predict global weather for the whole planet.

Global Weather Data
Stitch Patches Together
Global Weather Forecast
2
Faster, Greener Global Weather Prediction

This new method predicts global weather 50 times more efficiently than old supercomputer models, making detailed forecasts much more affordable.

STRATA AI Model
Uses Little Energy
Fast, Cheap Forecasts