Scaling Laws of Global Weather Models

Yuejiang Yu, Langwen Huang, Alexandru Calotoiu, Torsten Hoefler
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:150406-150426, 2026.

Abstract

Data-driven models are revolutionizing weather forecasting. To optimize training efficiency and model performance, this paper analyzes empirical scaling laws within this domain. We investigate the relationship between model performance (validation loss) and three key factors: model size ($N$), dataset size ($D$), and compute budget ($C$). Across a range of models, we find that Aurora exhibits the strongest data-scaling behavior: increasing the training dataset by 10$\times$ reduces validation loss by up to 3.2$\times$. GraphCast demonstrates the highest parameter efficiency, yet suffers from limited hardware utilization. Our compute-optimal analysis indicates that, under fixed compute budgets, allocating resources to more total training data yields greater performance gains than increasing model size. Furthermore, we analyze model shape and uncover scaling behaviors that differ fundamentally from those observed in language models: weather forecasting models consistently favor increased width over depth. These findings suggest that future weather models should prioritize wider architectures and larger effective training datasets to maximize predictive performance.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-yu26r, title = {Scaling Laws of Global Weather Models}, author = {Yu, Yuejiang and Huang, Langwen and Calotoiu, Alexandru and Hoefler, Torsten}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {150406--150426}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/yu26r/yu26r.pdf}, url = {https://proceedings.mlr.press/v306/yu26r.html}, abstract = {Data-driven models are revolutionizing weather forecasting. To optimize training efficiency and model performance, this paper analyzes empirical scaling laws within this domain. We investigate the relationship between model performance (validation loss) and three key factors: model size ($N$), dataset size ($D$), and compute budget ($C$). Across a range of models, we find that Aurora exhibits the strongest data-scaling behavior: increasing the training dataset by 10$\times$ reduces validation loss by up to 3.2$\times$. GraphCast demonstrates the highest parameter efficiency, yet suffers from limited hardware utilization. Our compute-optimal analysis indicates that, under fixed compute budgets, allocating resources to more total training data yields greater performance gains than increasing model size. Furthermore, we analyze model shape and uncover scaling behaviors that differ fundamentally from those observed in language models: weather forecasting models consistently favor increased width over depth. These findings suggest that future weather models should prioritize wider architectures and larger effective training datasets to maximize predictive performance.} }
Endnote
%0 Conference Paper %T Scaling Laws of Global Weather Models %A Yuejiang Yu %A Langwen Huang %A Alexandru Calotoiu %A Torsten Hoefler %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-yu26r %I PMLR %P 150406--150426 %U https://proceedings.mlr.press/v306/yu26r.html %V 306 %X Data-driven models are revolutionizing weather forecasting. To optimize training efficiency and model performance, this paper analyzes empirical scaling laws within this domain. We investigate the relationship between model performance (validation loss) and three key factors: model size ($N$), dataset size ($D$), and compute budget ($C$). Across a range of models, we find that Aurora exhibits the strongest data-scaling behavior: increasing the training dataset by 10$\times$ reduces validation loss by up to 3.2$\times$. GraphCast demonstrates the highest parameter efficiency, yet suffers from limited hardware utilization. Our compute-optimal analysis indicates that, under fixed compute budgets, allocating resources to more total training data yields greater performance gains than increasing model size. Furthermore, we analyze model shape and uncover scaling behaviors that differ fundamentally from those observed in language models: weather forecasting models consistently favor increased width over depth. These findings suggest that future weather models should prioritize wider architectures and larger effective training datasets to maximize predictive performance.
APA
Yu, Y., Huang, L., Calotoiu, A. & Hoefler, T.. (2026). Scaling Laws of Global Weather Models. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:150406-150426 Available from https://proceedings.mlr.press/v306/yu26r.html.

Related Material