NVIDIA AI Podcast – Ep. 266

NVIDIA’s Mike Pritchard Shares How Applying AI to Climate Simulation is Helping Forecast the Future
Date: July 23, 2025
Host: Noah Kravitz
Guest: Dr. Mike Pritchard, Director of Climate Simulation Research at NVIDIA

Episode Overview

This episode dives deep into the transformational impact of artificial intelligence (AI) on climate simulation and prediction. Host Noah Kravitz interviews Dr. Mike Pritchard, who leads climate simulation research at NVIDIA. Together, they explore how AI is revolutionizing the accuracy, resolution, and usefulness of weather and climate models. The discussion traverses Dr. Pritchard’s unique path into climate science, technical advances in simulation, real-world applications (including disaster planning and insurance), and the future of interactively approaching climate risks. The episode closes with a call to action for citizen engagement and open-source collaboration.

Key Discussion Points and Insights

1. Mike Pritchard’s Path to Climate Science

• Backstory: Mike entered science torn between jazz and chemistry, only to “land in an astrophysics program and learn about the cosmos” before traveling and realizing the urgent global impact of climate change.
• Turning Point: Exposure to the realities of sea-level rise in Bangladesh, and a fascination with radiative transfer in planetary atmospheres, drew him to climate physics ([01:10]).

“I realized there's this physics problem that really matters to the world and sort of haven't looked back since.”
— Mike Pritchard [01:22]

2. The Challenge of Climate Modeling

• Difference Between Weather and Climate Prediction:
  • Weather: Predicting specific events over days (using initial atmospheric conditions).
  • Climate: Predicting statistics/averages (e.g., how much will it warm on average globally in 50 years?) ([02:59]).
• Complexity: Many key variables, like cloud behavior, cannot yet be simulated explicitly at planetary scale due to computational limits.

“The trillion dollar question is how much will it heat up? Because all these hazards scale with the overall warmth of the planet. It's like a fever.”
— Mike Pritchard [03:27]

3. Enter AI: Accelerating and Transforming Climate Simulation

• Short-Circuiting Moore’s Law: AI is used to “learn the physics” of high-resolution regions and generalize them across global models, drastically speeding up simulation ([05:05]).
• A Decade of Progress in AI Modeling:
  • 2017: Hybrid models (“multi-scale simulators") incorporating explicit physical modeling and AI started breaking computational bottlenecks.
  • 2025: End-to-end AI weather and climate forecasting outperforms traditional physics-based models, with “huge ensembles for counterfactuals” and “interactive climate samplers” ([05:36]–[06:39]).

“I feel like AI is now disrupting things across the era. System modeling Stack.”
— Mike Pritchard [06:35]

4. Breakthroughs: Benchmarks, Emulation, and Super-Resolution

• Benchmarking/Competitions: Open-source APIs and Kaggle competitions have accelerated innovation in hybrid modeling ([08:10]).
• Full-AI Models: "Video prediction" style networks now auto-regress and predict atmospheric states, with results often outperforming physics-based forecasts ([08:35]).
• Super-Resolution with AI:
  • NVIDIA’s "cordif" tool brings blurry, low-resolution predictions up to the fine scale needed for real-world decision making ([09:55]).

“So now we have this thing called cordif. It does multivariate super resolution and new channel synthesis and it's being used in lots of different settings to take the pain out of generating high resolution state estimates of the atmosphere.”
— Mike Pritchard [10:32]

5. Quantifying Accuracy and Realism of AI Models

• Weather Forecasting: AI-driven forecasts now exceed traditional models in skill, marking a “quiet revolution” in the field ([11:13]).
• Climate Prediction: Verification is tougher; models must also conserve energy and mass for 50+ year horizons ([12:21]). AI models increasingly serve as emulators, making sharing and inference faster.

6. Real-World Applications and Societal Impact

• Extreme Weather and Insurance: AI enables “ensemble counterfactual generation” – thousands of realistic alternative weather/climate outcomes powering better risk assessment ([13:41]).
  • Example: Running “28,000 years worth of equivalent summer 2023” with AI to examine rare, extreme heatwaves ([14:14]).
  • Insurance companies are actively adopting these models.
• Interactive Tools for Policy and Disaster Planning:
  • New systems (“Climate in a Model,” “Climate in a Bottle”) let users “ask” for scenarios like major hurricanes, enabling more targeted preparation ([13:56], [26:05]).

“AI lets us say, you know, I'd like to find a hurricane, please. I'd like to find a hurricane in a future climate. ... That’s normally really hard to do with normal simulation technology, but AI. I totally see that’s going to be possible.”
— Mike Pritchard [14:53]

7. Digital Twins and Systemic Optimization

• Digital Twins in Climate: Integrating climate models with industrial digital twins could one day enable “optimization of resilience” across critical infrastructure ([18:17]).
• Coupling Models: Progress in combining atmospheric and ocean AI models (full emulation of Earth’s systems) is accelerating, reminiscent of breakthroughs that first enabled El Niño predictions ([26:04]).

8. The Data Revolution

• More Data, More Discovery:
  • AI enables unprecedented use of Earth observations, bypassing the limitations of human-written model assumptions ([19:43]–[21:44]).
  • There’s an ongoing “unification of AI for improving state estimation and prediction.”
• Synthetic Data’s Role: Past and simulated data help pre-train models, filling gaps in sparse observational records ([24:34]).

9. The Road Ahead: Oceans, Interactivity, Transparency

• Ocean Simulation: Major leap as ocean and atmosphere AI emulators begin to unite ([25:16]).
• Interactivity: New models handle “queryable time and informatics”—the ability to directly request specific events (like hurricanes), a paradigm shift for scientists and policymakers ([26:03], [27:59]).
• Transparency and Data Provenance: Ongoing work to ensure AI model outputs are as trustworthy and traceable as those from physics-based models, a must for government and industry adoption ([16:16]).

Notable Quotes & Memorable Moments

• “I almost feel like it's a bit like learning jazz, actually. It's like a never-ending enterprise, you'll never be done.”
  — Mike Pritchard, on climate modeling [04:38]
• “The most skillful forecasts in the world now come from AI systems that are inscrutable. We don't understand fully how they're making their predictions, but they're easy to score and convince yourself they're better.”
  — Mike Pritchard [09:29]
• “If you value the observing systems that have led and the weather modeling agencies that have led to the data sets that are leading to this amazing revolution in AI prediction quality, express that to the people in charge.”
  — Mike Pritchard, call to action [28:44]
• “Machine learning is nothing without the data it's trained on.”
  — Mike Pritchard [28:50]
• “A lot of the warming already happens. The average lifetime of a carbon dioxide molecule in the atmosphere is 100 years. And so the past 200 years of emissions will definitely be warming our planet in the future.”
  — Mike Pritchard [29:31]

Timestamps for Key Segments

• Dr. Pritchard’s Journey to Climate Science: [01:10]–[02:17]
• Difference: Weather vs. Climate Modeling: [02:59]–[04:46]
• AI’s Emergence & Game-changing Advances: [05:05]–[06:39]
• Resolution in Climate Models: [06:39]–[07:54]
• Hybrid & Full-AI Simulation Breakthroughs: [08:10]–[10:53]
• Weather/Climate Model Accuracy Trends: [11:13]–[13:00]
• Using AI Predictions for Extreme Event Planning (Insurance, Disaster Response): [13:39]–[16:16]
• Digital Twins in Climate Science: [17:45]–[19:27]
• TED Talk Messaging / Data Revolution: [19:43]–[22:53]
• Interactivity and “Climate in a Bottle”: [26:03]–[27:59]
• Call to Action / How to Get Involved: [28:44]–[30:08]
• Where to Learn More (NVIDIA Earth 2): [30:24]

Final Takeaways & Resources

• AI is not just accelerating simulations—it is reshaping how scientists and the public experience and prepare for climate risks, offering unprecedented resolution, interactivity, and access.
• The path forward includes open science (“Earth 2” initiative), further coupling across physical domains (ocean/atmosphere), and ongoing public engagement.

Get Involved:

• Advocate for Earth observation funding and data access.
• Explore open-source climate benchmarks and competitions (e.g., on Kaggle).
• Dive into NVIDIA’s open-source simulation tools at NVIDIA Earth 2.

“What a time to be alive.”
— Noah Kravitz & Mike Pritchard [31:11]