NVIDIA’s Mike Pritchard Shares How Applying AI to Climate Simulation is Helping Forecast the Future - Ep. 266

What It Covers

Mike Pritchard, NVIDIA's director of climate simulation research, explains how AI accelerates climate modeling by 20-100x, enables interactive weather prediction, and generates thousands of counterfactual extreme weather scenarios for improved planning and resilience.

Key Questions Answered

• •AI Speed Advantage: AI climate models run 1,000x faster than physics-based simulators, enabling generation of 28,000 years of equivalent weather data to study rare heat waves and extreme events with statistically meaningful sample sizes for insurance and planning applications.
• •Resolution Breakthrough: Traditional climate models operate at county-scale resolution (25 kilometers), missing critical cloud and storm physics. AI hybrid models achieve 20-100x faster simulations while incorporating high-resolution atmospheric processes that standard supercomputers cannot handle at planetary scale.
• •Interactive Climate Prediction: Climate in a Bottle technology allows users to request specific weather scenarios by inputting three parameters (time of day, season, ocean temperature) plus event preferences, generating 600 megabytes of atmospheric data from 200 kilobytes of input for targeted hurricane and extreme weather analysis.
• •Observational Data Integration: AI models utilize satellite and station observations directly without human-written physics assumptions, opening pathways to incorporate previously unusable Earth system data and potentially extending atmospheric predictability limits beyond the traditional one-week forecast horizon into multi-week timescales.