Data for: Can AI-based weather prediction models simulate the butterfly effect? The role of architecture and implementation.

The butterfly effect is a property of the atmosphere, where tiny-amplitude initial perturbations grow quickly in areas of convection and latent heat release, then spreading out and moving upscale, eventually affecting even the largest planetary scales. This process takes about two weeks and leads to the existence of an intrinsic predictability limit. In this study we investigate the ability of current state-of-the-art AI-based weather prediction models to reproduce six key characteristics of the butterfly effect and compare them to results from a conventional weather prediction model at various resolutions. We find that the AI models separate into two groups. The first group did not reproduce any of the six characteristics, while the second group did reproduce some, in particular fast initial uncertainty growth and indication of an intrinsic limit. However, the behavior was physically inconsistent and appeared to result from numerical noise that depended on implementation details such as whether the model was run on a CPU or GPU. It seems likely that the inability of AI models to simulate the butterfly effect results from limitations in the analysis data used for training, since their size, design and architecture turned out to be largely irrelevant.

Access to data