By Armando Quesada Webb
At the beginning of this year, the European Centre for Medium-Range Weather Forecasts launched the Artificial Intelligence Forecast System (AIFS), the first fully operational AI-based weather forecasting system . Other experimental models, such as Google’s GraphCast, the California Institute of Technology’s FourCast, and Huawei’s Pangu-Weather, a Chinese company, are also seeking to take a step forward in this new revolution in meteorology . According to Pedram Hassanzadeh, associate professor of geophysical sciences at the University of Chicago, the AIFS offers significant advantages, as they are “cheap, easy to develop, accurate, and fast models that also reduce electricity bills.”
Hassanzadeh, however, knows that AI-based weather prediction models are still far from perfect. The scientist is part of a group of academics from the University of Chicago, the University of New York, and the University of Santa Cruz in California who recently exposed a major limitation of this technology : the ability to predict unprecedented weather events.
The research, published a month ago in the journal Proceedings of the National Academy of Sciences, suggests that these new models can make short-term weather forecasts with “surprising accuracy,” but fail when it comes to high-intensity events not found in the data used to train the artificial intelligence. This is because the neural networks that power the AI can only predict based on past patterns.