submitted.news

Traditional Weather Models Outperform AI in Predicting Extreme Events

Written by

Submitted

in

Published on May 1, 2026

Weather forecasting has increasingly leaned on artificial intelligence, promising faster and more accurate predictions. However, recent research has revealed a worrying gap in AI capabilities, particularly when it comes to forecasting extreme weather conditions. Traditional physics-based models still demonstrate superior accuracy in these critical situations.

A study and his team from the University of Geneva tested various leading AI models, including GraphCast and Pangu-Weather, against a database of extreme weather events. Despite their advancements, the AI systems faltered in predicting severe incidents, such as the Siberian heat wave of 2020, where they underestimated temperatures and failed to account for the profound effects of climate change.

The failure stems from the training methodologies of AI models, which rely heavily on historical data that lacks record-breaking events. Engelke noted that AI attempts to predict future weather based on patterns observed in the past, limiting its ability to foresee unprecedented situations. In contrast, traditional models utilize complex mathematical equations that adapt more readily to emerging conditions.

While AI has shown promise in forecasting typical weather patterns and certain severe events, experts acknowledge the necessity of enhancing model accuracy for extreme conditions. As researchers seek methods to integrate more comprehensive training data, traditional forecasting methods are likely to remain a crucial tool in our weather prediction arsenal for the foreseeable future.

Related News

Related Articles