10 Apr 2024
 | 10 Apr 2024
Status: this preprint is open for discussion.

Do data-driven models beat numerical models in forecasting weather extremes? A comparison of IFS HRES, Pangu-Weather and GraphCast

Leonardo Olivetti and Gabriele Messori

Abstract. The last few years have witnessed the emergence of data-driven weather forecast models able to compete and in some respects outperform physics-based numerical models. However, recent studies question the capability of data-driven models to provide reliable forecasts of extreme events. Here, we aim to evaluate this claim by comparing the performance of leading data-driven models in a semi-operational setting, focusing on the prediction of near-surface temperature and windspeed extremes globally. We find that data-driven models outperform ECMWF’s physics-based deterministic model in the average prediction of 10 m windspeed and 2 m temperature, and can also compete with the physics-based model in terms of extremes in most regions. However, the choice of best model depends strongly on region, type of extreme and sometimes even lead time. Thus, we conclude that data-driven models may already now be a useful complement to physics-based forecasts in those regions where they display superior tail performance, but that some challenges still need to be overcome before widespread operational implementation can take place.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Leonardo Olivetti and Gabriele Messori

Status: open (until 06 Jun 2024)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on egusphere-2024-1042', Anonymous Referee #1, 22 Apr 2024 reply
Leonardo Olivetti and Gabriele Messori
Leonardo Olivetti and Gabriele Messori


Total article views: 889 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
564 313 12 889 11 6
  • HTML: 564
  • PDF: 313
  • XML: 12
  • Total: 889
  • BibTeX: 11
  • EndNote: 6
Views and downloads (calculated since 10 Apr 2024)
Cumulative views and downloads (calculated since 10 Apr 2024)

Viewed (geographical distribution)

Total article views: 881 (including HTML, PDF, and XML) Thereof 881 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
Latest update: 28 May 2024
Short summary
Data-driven models are becoming a viable alternative to physics-based models for weather forecasting up to 15 d into the future. However, it is unclear whether they are as reliable as physics-based models at forecasting weather extremes. Here, we evaluate their performance in forecasting near-surface cold, hot and windy extremes globally. We find that data-driven models can compete with physics-based models, and that the choice of best model mainly depends on region and type of extreme.