Nature has published an Author Correction to a paper titled «A spectral test of the butterfly effect and physical consistency in the diffusion-based GenCast's ensembles». The correction concerns a study proposing a spectral test to assess the butterfly effect and the physical consistency of forecast ensembles generated by GenCast, a diffusion-based weather prediction model.

GenCast belongs to the new generation of deep-learning models competing with traditional numerical prediction systems from the ECMWF and other operational centres. These models generate forecast ensembles through a diffusion process, and one open question is whether they faithfully reproduce uncertainty propagation and the spectral properties of the atmosphere, that is, whether they respect the expected physical consistency across scales.

The text available for the correction does not include technical details of the amendments, the affected data or the corrected figures. In Nature, an Author Correction indicates that the authors themselves have identified errors in the original paper and that the journal has published the corresponding amendment; the specific scope of the changes should be consulted in the full correction.

The correction is relevant to the community evaluating machine-learning weather prediction models, where spectral validation and the physical consistency of ensembles are key criteria for judging whether these systems can be used operationally. NewsPhysics will update this story if more details on the affected figures are published.