AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

NVIDIA's Mike Pritchard on AI-Driven Climate Simulation

· NVIDIA AI Podcast

In episode 266 of the NVIDIA AI Podcast, host Noah Kravitz speaks with Mike Pritchard, Director of Climate Simulation Research at NVIDIA, about how AI is transforming climate science and weather forecasting. Pritchard explains the distinction between weather prediction (days, initialized from observations) and climate prediction (decades, statistics of weather) and notes that traditional physics-based climate simulators struggle with resolution and the compute cost of running hundreds of multi-decade scenarios. He frames AI as a way to "short-circuit Moore's law," first through hybrid physics-AI models (20x-100x speedups) and now through fully AI atmospheric emulators that he describes as the most skillful weather forecasting systems available today. Pritchard details NVIDIA's work under the Earth2 initiative, including a tool called CorrDiff for multivariate super-resolution and new channel synthesis, and a research project called "Climate in a Bottle" that lets users query a generative model for specific weather events (e.g., tropical cyclones) using low-dimensional inputs like ocean temperature. He highlights that AI enables huge ensembles of counterfactual climate scenarios—running thousands of summers' worth of extremes to study rare events—capabilities he says are beyond traditional physical simulators. He also flags the emergence of ocean emulators and coupled atmosphere-ocean AI models as the next frontier, drawing a parallel to the 1980s when coupled physics models first enabled El Niño prediction. Looking ahead, Pritchard emphasizes sustained Earth observations as critical infrastructure for ML, points listeners to NVIDIA Earth2's open-source recipes, and calls for both policy engagement and ML researcher participation in unsolved climate benchmarks. He expresses optimism that interactive, queryable AI climate models will broaden public understanding of a warming planet whose committed warming from existing CO2 is already locked in for roughly a century.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.