Jan 2, 2026 · 45m · catalyst
How AI is changing weather forecasting
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Catalyst, host Shail Khan interviews Google DeepMind's Peter Battaglia to explore how advanced artificial intelligence architectures and large-scale atmospheric datasets are revolutionizing weather prediction and clean energy management.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Shayle holds 21% of the talking time here. How this is scored →
Shayle as informed peer 4.0 Guest teaching 5.3 Guest disagreement 0.1 Shayle pushing back 0.6
speaking balance: gold is Shayle, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 33:08 Clarifying perceived physical determinism versus model stochasticity
Peter gently rejects the host's premise that weather is simply a deterministic modeling problem compared to language, explaining that unobservable sub-scale chaos renders it practically stochastic from the model's perspective.
Hardest push from Shayle ▶ 32:27 Challenging whether weather is harder or easier than LLMsShayle pushes back on the direct equivalence with LLMs by arguing that underlying physical equations have a singular ground truth whereas next-token language generation possesses no single correct answer.
Biggest teaching moment ▶ 28:53 Markovian physics versus sequential language modelingPeter educates Shayle on why transformer attention mechanisms are applied spatially across the globe in weather forecasting rather than temporally across historical steps as done in autoregressive text models.
Shayle holds their own ▶ 13:51 Framework of historical forecasting improvement driversShayle demonstrates strong domain grasp by breaking down the historical trajectory of numerical weather prediction into three distinct technical drivers: data assimilation, raw compute scaling, and algorithmic parameterizations.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Shayle as informed peer | Guest teaching | Guest disagreement | Shayle pushing back | Why |
|---|---|---|---|---|---|---|
| History and Institutional Architecture of Weather Forecasting | 2 | 5 | 0 | 0 | Shayle explicitly asks Peter to school him on the historical evolution and institutional architecture of weather forecasting. Peter provides a comprehensive overview of 19th-century origins, the 1970s creation of NOAA and ECMWF, and the pipeline split between base forecasts and downstream post-processing. | |
| Numerical Weather Prediction and Atmospheric Chaos | 4 | 6 | 0 | 1 | Shayle asks whether global forecasts act like foundational LLMs and highlights the exponential complexity of forecasting into the future. Peter explains Navier-Stokes fluid approximations, atmospheric chaos, the butterfly effect, and the distinction between current-state estimation and future prediction. | |
| Drivers of Forecasting Progress and Early Machine Learning | 5 | 4 | 0 | 0 | Shayle structures an insightful tripartite hypothesis regarding whether historical accuracy improvements stem from better data, more compute, or algorithmic tricks. Peter agrees all three matter, detailing ECMWF spatial resolution upgrades and early statistical post-processing. | |
| Mid-Roll Sponsor Advertisements | 3 | 5 | 0 | 0 | Following sponsor advertisements, Shayle probes where the persistent technical gaps lie, specifically questioning precipitation modeling. Peter explains why smoothly varying variables like temperature are straightforward while localized phenomena like violent thunderstorms require finer spatial resolutions. | |
| Deep Learning Architectures in Modern Meteorology | 4 | 6 | 0 | 0 | Shayle questions what modern AI unlocks compared to supervised learning methods from five years ago. Peter explains how graph neural networks and transformers enable direct, long-range spatial connectivity across image and global scale dimensions. | |
| Physical Markov Dynamics vs. Language Models | 6 | 7 | 1 | 3 | Peter explains that unlike language, physical dynamics are Markovian, meaning transformers are applied across space rather than past temporal sequences. Shayle counters with a nuanced argument about physical determinism versus language ambiguity, prompting Peter to clarify that unobserved sub-scale phenomena make weather functionally stochastic to the model. | |
| Meteorological Training Datasets and Alternative Sensor Streams | 5 | 5 | 0 | 1 | Shayle suggests weather forecasting should benefit from vast historical sensor records and distributed mobile data. Peter explains the strengths of ECMWF's ERA-5 reanalysis dataset while detailing data limitations and the potential of unconventional sources like rain-sensing car wipers and video doorbells. | |
| Future Horizons for Energy, Logistics, and Disaster Mitigation | 3 | 4 | 0 | 0 | Shayle asks for future projections on what AI-driven meteorology will enable across industries. Peter details downstream impacts on electrical grid operations, renewable integration, logistics, tropical cyclone tracking, and early wildfire intervention. |
Statements from this episode (10)
Battaglia: Mobile weather apps rely on intermediaries to process NOAA data
“When you look at the app on your phone and you see, you know, the chance of precipitation, that's not coming directly from
[458] Peter Battaglia: NOAA, that's coming from other intermediaries that are, like, taking, you know, local weather station data and oth…”
Battaglia: Early ML in weather improved downstream calibration, not base models
“So actually, the first advent of like, AI and machine learning in weather forecasting, or at least some of the earliest, was not, like, trying to overhaul the whole weather forecast process itself, but making, you know, using more and more, like, statistical m…”
Battaglia: Precipitation and wind are harder to forecast due to fine-scale dynamics
“So that type of very high resolution, complex, you know, patterns of precipitation, for example, Wind as well. Those are much harder to predict because you're effectively predicting a lot more information. You can't just sort of summarize it by saying, oh, eve…”
DeepMind primarily uses traditional supervised learning for AI weather models
“What we use in our weather forecasting models, and a lot of folks out in the community are using it as, you know, as this field is advancing and this AI-based weather forecasting is developing, we're still mostly using fairly traditional machine learning, supe…”
Battaglia: Data pipelines matter more than neural network architectures today
“I don't feel that neural network architectures these days tend to be the sort of exciting part. It's usually more of like the training and the sort of data how you handle the data and that kind of thing.”
Battaglia: Weather is Markovian unlike text sequences in language models
“Weather is actually different from text in a fundamental way. In fact, most physical processes are. They are what's called Markov, in that the most recent state of the system determines the subsequent state.”
DeepMind uses transformers to model spatial interactions in weather, not time
“We actually use transformers not to model the spatial, the interactions in weather over time, like the sequence of text, but in space.”
Battaglia: AI weather models leverage macroscopic spatial ranges unlike traditional simulations
“In a traditional model, the way it simulates that is it, in very fine detail, it kind of figures out, like, what's the pressure, and the temperature, and the wind, and the moisture, and what are those things? What's going to happen next is determined strictly …”
Battaglia: Advanced weather forecasting has massive headroom for electric grid optimization
“But I have a feeling that there's a lot of headroom, a lot more to be gained in how we, you know, plan our, you know, how to operate our electrical grids, how we predict what the electrical demand is going to be.”
Battaglia: Logistics and supply chains have barely tapped AI weather forecasting
“I think that supply chains and logistics and, you know, even just like lots of choices that, you know, driving and these type of things, I think, really could be better informed by better weather forecasts, and I don't think we've even begun to kind of get int…”