Weather Forecasting Revolution: Can AI Outshine Traditional Models?

A new chapter in weather forecasting is unfolding as major tech companies deploy machine-learning models that challenge traditional physics-based forecasts. These models could transform how we predict the ever-changing British weather, critical for daily life and disaster preparedness. The economic stakes are high; in the U.S., severe weather disasters caused nearly $182 billion in damages in just 2024, highlighting the value of accurate forecasts.

Traditional forecasting relies on supercomputers, like the British Met Office’s which operates at 60 quadrillion calculations per second and requires immense resources. Despite their accuracy in predicting large weather patterns, these models struggle with fine details such as local showers, often missing crucial weather events. In contrast, machine-learning models, which have emerged in recent years, can process data in mere seconds on standard laptops by drawing from 40 years of historical weather data, potentially offering quicker and more efficient forecasts.

Early verification data shows a mixed performance of these AI models; while some have outperformed traditional ones in predicting atmospheric pressure patterns, others still lag behind. Both model types share a limitation: their accuracy decreases the further into the future they predict, meaning forecasts beyond ten days remain unreliable.

Although machine-learning models excel in certain areas, they still require traditional models as a foundation. They perform well on a large scale but fail to catch smaller-scale features crucial for localized weather events. Furthermore, concerns about their predictive ability during extreme or rare weather events, or in a rapidly warming climate, remain. Experts suggest that the future of forecasting will likely see a hybrid approach, uniting the strengths of both AI and traditional physics-based methods for more localized and rapid forecasts. As these models evolve, the potential for an improved and more efficient weather prediction landscape looks promising.