AI is good at weather forecasting. Can it predict freak weather events?

AI weather models, while highly efficient and accurate for typical conditions, fundamentally struggle to predict extreme "gray swan" events because they rely solely on historical patterns without underlying physical principles. This limitation poses significant risks in operational forecasting and long-term risk assessment, as these models tend to underestimate the severity of unprecedented storms, potentially leading to dangerous false negatives that compromise public safety and infrastructure resilience. The core issue stems from neural networks functioning as predictive tools based on past data rather than simulating atmospheric physics. Consequently, they cannot extrapolate beyond the intensity of events they have already witnessed, even if similar phenomena occurred in different geographical regions. This constraint highlights a critical gap in current AI applications for climate science, where the inability to anticipate rare, high-impact events limits their utility for comprehensive disaster prepared planning compared to traditional, physics-based supercomputer models. To address this, researchers advocate for hybrid approaches that integrate mathematical tools and atmospheric physics into AI frameworks. By combining the computational efficiency of machine learning with the rigorous logic of physical laws, scientists aim to develop models capable of predicting extreme anomalies. This evolution is vital for open data initiatives, as it ensures that shared datasets and AI tools evolve to safely handle the full spectrum of climate risks rather than just the average conditions.

Source: sciencedaily.com
Published on 2025-05-23

AI is good at weather forecasting. Can it predict freak weather events?

Related news