Tech

Spherical DYffusion Model Enables Century-Scale Climate Predictions in Hours

April 15, 2026 · From the archive, written in the Professor's voice before our October 2026 redesign from today.ucsd.edu

What happened

UC San Diego and the Allen Institute for AI have developed 'Spherical DYffusion,' a generative AI model that fuses deep learning with physics-based climate data to simulate 100 years of global climate patterns in just 25 hours. This hybrid approach leverages diffusion probabilistic models operating on spherical data representations to accelerate complex climate projections significantly.

Why it matters

This breakthrough exemplifies how combining domain knowledge with generative AI can compress traditionally time-consuming simulations into manageable timeframes. For AI practitioners, it underscores the power of hybrid models that respect physical constraints while exploiting AI's generative strengths, suggesting a new paradigm for scientific modeling workflows.

Who's doing it

The collaborative effort by UC San Diego and the Allen Institute for AI has yielded this innovation, dramatically reducing computational time for climate modeling without sacrificing scientific fidelity. Their approach has implications for faster policy-relevant climate analytics and decision-making.

Try it

1. Review the UCSD news release at https://today.ucsd.edu/story/nine-breakthroughs-made-possible-by-ai for technical details.
2. Access the Spherical DYffusion model code or related repositories if available, possibly hosted on GitHub by the Allen Institute.
3. Prepare domain-specific climate datasets formatted for spherical coordinates, then run the model to generate accelerated climate forecasts using a high-performance GPU environment.

Read the original at today.ucsd.edu