Spherical DYffusion: AI Accelerates Century-Scale Climate Projections by 99%
Researchers at UC San Diego and the Allen Institute for AI developed Spherical DYffusion, a hybrid generative AI and physics-based model that compresses 100 years of climate pattern simulations into just 25 hours. This is achieved by integrating diffusion models with spherical data representations, enabling fast yet reliable forecasting over global scales.
This breakthrough demonstrates the power of combining domain knowledge (physics) with generative AI to overcome traditional computational bottlenecks in climate science. It encourages AI users to consider hybrid modeling approaches for complex, data-intensive tasks that require both speed and scientific rigor.
UC San Diego’s AI and Climate Science teams alongside the Allen Institute for AI are actively developing and validating Spherical DYffusion, showing promise in significantly accelerating climate scenario testing without sacrificing accuracy.
Step 1: Explore open-source diffusion model frameworks such as OpenAI’s guided diffusion or Hugging Face’s diffusion libraries. Step 2: Adapt these models to spherical data formats, using climate datasets like CMIP6 for training. Step 3: Validate the model by comparing generated projections against traditional physics-based simulations. Expected outcome: Fast generation of century-scale climate projections within hours, drastically reducing compute time. https://today.ucsd.edu/story/nine-breakthroughs-made-possible-by-ai