Tech

UC San Diego and Allen Institute’s Spherical DYffusion Model Forecasts Century of Climate in Hours

April 3, 2026 · From the archive, written in the Professor's voice before our October 2026 redesign from today.ucsd.edu

What happened

Researchers at UC San Diego and the Allen Institute for AI created Spherical DYffusion, a hybrid AI-physics model that compresses 100 years of climate projections into just 25 hours of computation. The approach integrates generative AI with physics-based datasets to enhance both speed and accuracy of long-term climate simulations. This represents a significant leap over traditional climate models that typically require weeks or months for similar forecasts.

Why it matters

This breakthrough teaches us that combining domain-specific physical laws with generative AI techniques can drastically accelerate simulations without sacrificing fidelity. For AI practitioners, it means hybrid models leveraging scientific principles can outperform pure data-driven approaches in complex forecasting. This should encourage incorporating physics constraints into your AI pipelines when modeling real-world phenomena.

Who's doing it

The team at UC San Diego and the Allen Institute for AI spearheaded this research, demonstrating the model’s ability to simulate century-scale climate dynamics more efficiently than standard methods. Their work is a benchmark in climate AI research, offering a scalable tool for environmental scientists.

Try it

1. Visit the UC San Diego AI research portal at https://today.ucsd.edu/story/nine-breakthroughs-made-possible-by-ai to access Spherical DYffusion resources.
2. Download the model codebase and climate datasets provided.
3. Run the model using their documented scripts to generate accelerated climate forecasts and analyze outputs—expect a drastic reduction in compute time compared to classical simulations.

Read the original at today.ucsd.edu