How Well can AI Weather Forecast Models Predict Extreme Events Like Storm Ciarán?

The development of weather forecasting models using Artificial Intelligence (AI) has become increasingly popular during the last few years. Unlike traditional weather forecast models which rely on the laws of physics, AI models rely on large amounts of historical data and establish patterns in the data to make predictions in the future.

A recent study led by scientists at the University Reading evaluated the performance of 4 AI models using Storm Ciarán as an example case study. Storm Ciarán was responsible for large damage across Europe in November 2023, including large hail and a tornado in Jersey.

In the recently published study, 4 AI weather forecast models developed by technology companies such as Google DeepMind and Huawei were evaluated. They showed that the models could accurately represent the structure of Ciarán showing key features of mid-latitude storms such as the storm’s air streams and cloud head.

The upper level winds relative to the storm’s location were also realistic. Nevertheless, all of the AI models underestimated the maximum wind speeds associated with Ciarán. This means if we had solely relied on these models during the Storm Ciarán event, then the intensity of winds would have been forecasted to be less severe than reality.

Why use AI models instead of physics-based models?

One of the key advantages of using AI models to predict weather is that once AI models have been developed it is much quicker to produce predictions forward in time, anywhere between 10–100 times faster than physics-based models.

In theory, with further development, they could also improve upon physic-based models in certain scenarios. Further assessment of AI models for additional case studies is crucial for enhancing their effectiveness.

Will physics-based weather forecast models be discontinued?

While the initial evaluation of AI models has shown some promising results, they seem to have limitations in representing extreme events. Furthermore, the AI weather models currently available require data from physics-based models as an input for their development and also to make predictions.

Therefore, it does not seem likely that we will be moving on from the traditional way to produce weather forecasts anytime soon. The most likely scenario is for AI models to be increasingly used alongside the traditional physics-based models, potentially improving the overall accuracy of weather forecasts.

News reference

Charlton-Perez, A.J., Dacre, H.F., Driscoll, S. et al. Do AI models produce better weather forecasts than physics-based models? A quantitative evaluation case study of Storm Ciarán. npj Clim Atmos Sci 7, 93 (2024).