Are Weather Emulators Pushing the Frontier of Subseasonal Heat Extremes Forecasting?

Cas Decancq, Thomas Mortier, Diego Miralles

Track: Type E (Late-Breaking Abstracts)

Keywords: Weather Emulators, Weather Forecasting, Deep Learning, Climate Extremes, Subseasonal Forecasting

Abstract:

Predicting climate extremes such as droughts, heatwaves, and heat stress episodes remains a critical challenge in Earth system sciences. Current state-of-the-art methods often fail to deliver reliable forecasts, especially at subseasonal-to-seasonal (S2S) timescales (i.e., from two weeks to two months in advance). As global climate variability continues evolving, the need for advanced, trustworthy, data-driven forecasting methodologies has never been more pressing.

Extended numerical weather prediction systems, such as those led by the European Centre for Medium-Range Weather Forecasts (ECMWF), remain the primary method for S2S prediction [1]. While recent deep learning approaches have demonstrated remarkable competitive performance [2], proposed models predominantly focus on global-scale average weather predictions, overlooking critical local-scale extreme events [3], [4]. Moreover, creating accurate probabilistic forecasts conditioned on the initial state remains a significant challenge within the scientific community.

Here, we evaluate two baseline methods (climatology and persistence) and three dynamical weather forecasting systems (ECMWF-IFS, CMA, NCEP) [1], together with six of their recently proposed deep learning counterparts — Pangu-Weather [5], FuXi [6], ArchesWeather [7], ECMWF-AIFS [8], GraphCast [9], and Aurora [10] — on their ability to predict both global surface temperature and six extreme heatwave events across continents at a 14-day lead time. Results show that deep learning emulators are already pushing the frontier of subseasonal temperature forecasting without being trained to do so, though they do not yet show clear improvements for extremes prediction — that is where dynamical models become competitive. All models, however, manage to beat climatological and persistence baselines for heat extremes prediction. Forecast characteristics depend greatly on model architecture and training setup, and few models manage to achieve both high deterministic skill and correct variability over spatial and temporal scales. Overall, current science lacks methods that can deliver reliable and calibrated probabilistic forecasts. This analysis not only highlights the limitations of current deep learning and dynamical weather forecasting systems, but also establishes a practical benchmark for future research.

References

[1] F. Vitart and A. W. Robertson, “The sub-seasonal to seasonal prediction project (S2S) and the prediction of extreme events,” npj Climate and Atmospheric Science, vol. 1, p. 3, 2018, doi: 10.1038/s41612-018-0013-0.
[2] L. Olivetti and G. Messori, “Do data-driven models beat numerical models in forecasting weather extremes? A comparison of IFS HRES, Pangu-Weather and GraphCast,” EGUsphere, preprint, Apr. 2024. [Online]. Available: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1042
[3] P. A. G. Watson, “Machine learning applications for weather and climate need greater focus on extremes,” Environmental Research Letters, vol. 17, no. 11, p. 111004, Nov. 2022, doi: 10.1088/1748-9326/ac9d4e.
[4] O. C. Pasche, J. Wider, Z. Zhang, J. Zscheischler, and S. Engelke, “Validating deep-learning weather forecast models on recent high-impact extreme events,” arXiv preprint arXiv:2404.17652, 2024.
[5] K. Bi, L. Xie, H. Zhang, X. Chen, X. Gu, and Q. Tian, “Pangu-Weather: A 3D high-resolution model for fast and accurate global weather forecast,” arXiv preprint arXiv:2211.02556, 2022.
[6] L. Chen, X. Zhong, F. Zhang, Y. Cheng, Y. Xu, Y. Qi, and H. Li, “FuXi: A cascade machine learning forecasting system for 15-day global weather forecast,” npj Climate and Atmospheric Science, vol. 6, no. 1, p. 190, 2023.
[7] G. Couairon, R. Singh, A. Charantonis, C. Lessig, and C. Monteleoni, “ArchesWeather & ArchesWeatherGen: A deterministic and generative model for efficient ML weather forecasting,” arXiv preprint arXiv:2412.12971, 2024.
[8] S. Lang et al., “AIFS – ECMWF’s data-driven forecasting system,” arXiv preprint arXiv:2406.01465, 2024.
[9] R. Lam et al., “Learning skillful medium-range global weather forecasting,” Science, vol. 382, no. 6677, pp. 1416–1421, 2023.
[10] C. Bodnar et al., “A foundation model for the Earth system,” Nature, pp. 1–8, 2025.

Submission Number: 90