BNEF analyst Ryan Ward explains how Google DeepMind's AI model matched or surpassed traditional physics-based hurricane forecasting in the 2025 Atlantic season, predicting storm tracks within 60 miles up to three days ahead. While not replacing traditional models, AI offers faster, bespoke forecasts for critical applications like energy infrastructure evacuation and agriculture.
Summarized by Podsumo
AI models (e.g., Google DeepMind) matched or beat NOAA’s official forecasts for Atlantic hurricanes in 2025, outperforming decades-old physics-based models.
AI predicted hurricane tracks within ~100 km (60 miles) up to three days in advance—critical for evacuating energy assets like oil rigs and LNG facilities.
Even Hurricane Melissa, the strongest Atlantic hurricane on record, was forecast effectively by AI despite being outside its training data.
AI serves as a complement to traditional models, which aggregate multiple data sources including AI; bespoke models for farmers or energy companies are on the horizon.
"You have a computer that just solves those equations and tells you what the weather is going to be tomorrow. An AI weather model is just a fancy way of saying we're not going to use those equations. — Ryan Ward"
"The DeepMind model did basically on par with NOAA's official forecast... better than all the traditional physics-based models that have existed for decades. — Ryan Ward"
"These AI models are really fast and probably can be pretty good and cheap to use for specific purposes... that ability to be deployed across a variety of contexts is going to be really revolutionary. — Ryan Ward"