The world of weather forecasting is undergoing a quiet revolution, and at the heart of it is artificial intelligence (AI). While AI has the potential to revolutionize the field, it also presents unique challenges, particularly when it comes to predicting unprecedented weather events. This is the story of how AI is changing the way we forecast the weather, and why it's more perilous than we think.
The Promise of AI in Weather Forecasting
AI models are faster, cheaper, and require less computational infrastructure than traditional physical models. When it comes to predicting typical weather patterns and events, their accuracy is comparable and improving rapidly. For example, during the 2025 Atlantic hurricane season, Google DeepMind's model outperformed nearly every physical model on storm track and intensity. This is particularly exciting for regions that lack traditional forecasting resources, as AI models can provide valuable insights to communities on the frontlines of climate change.
The Gray Swan Problem
However, the promise of AI in weather forecasting is not without its pitfalls. The 'gray swan' problem is a major concern. Gray swan weather extremes are physically plausible but so rare that they are poorly represented in training datasets. Climate change is leading to more first-of-their-kind weather extremes, and AI models are struggling to keep up. For example, the 2021 Pacific Northwest heatwave was so severe that it would have been virtually impossible without climate change, yet AI models failed to predict it.
The Case for AI Forecasting
Despite these limitations, meteorologists are rapidly adopting AI forecast models. This is because AI models are faster, cheaper, and require far less computational infrastructure than physical models. When it comes to predicting typical weather patterns and events, their accuracy is comparable and improving rapidly. In fact, since 2023, leading AI models such as GraphCast, Pangu-Weather, and the ECMWF's AIFS have matched or outperformed the best physical models on medium-range forecasting metrics.
The Need for Rigorous Testing
However, rapidly adopting these models without addressing the risks would be dangerous, especially in parts of the world highly vulnerable to the impacts of climate change. Researchers are calling for more rigorous testing of AI forecast models before public agencies widely adopt them. One proposed framework, the AI Retraining Without Iconic Events (AIRWIE) protocol, would deliberately withhold a designated set of extreme events from the training dataset to assess the models' ability to extrapolate unprecedented weather extremes.
The Way Forward
Efforts to understand and address the limitations of AI forecasting will be critical, because there's no turning back now. AI is already reshaping the way we predict the weather, and as the climate becomes increasingly volatile, meteorologists will need every tool in their arsenal to be sharp and reliable. Despite their current limitations, there is much to gain from continuing to push these systems forward and figuring out how to best integrate them with physical forecasting. In my opinion, the future of weather forecasting lies in a hybrid approach that combines the strengths of both AI and physical models.
Conclusion
The story of AI in weather forecasting is a fascinating one, filled with both promise and peril. As we continue to push the boundaries of what's possible, it's important to remember that the future of weather forecasting lies in a hybrid approach that combines the strengths of both AI and physical models. Only then can we hope to truly harness the power of AI to predict and prepare for the weather of the future.
