Pushing the Limits of High-Resolution Weather Forecasting through Data Scaling
AuthorsYang Zhao, Peisong Niu, Tian Zhou, Ziqing Ma, Guanlong Ma, Rong Jin, Huiling Yuan, Liang Sun
Resources
By turning abundant coarse weather reanalysis into high-resolution training data, this work shows that better data scaling—not just better models—can substantially improve global forecasts.
Key results
BaguanHR’s overall RMSE improvement against IFS-HRES within the evaluated 72-hour window.
Average RMSE reduction against IFS-HRES at 24-hour lead time.
Average RMSE reduction against IFS-HRES at 72-hour lead time.
RMSE reduction when training data expands from 7 to 18 years for 120-hour forecasts.
Training data-loading speedup from SSD partitioning and hybrid storage.
Number of NVIDIA A800 GPUs used for the BaguanHR training pipeline.
What the paper found
BaguanHR tackles the central problem in 0.1° global weather forecasting: long high-resolution records are scarce, while ERA5 offers decades of data only at 0.25°. Instead of transferring a coarse-grid forecasting model, the framework transfers data. Variable-wise Swin2SR models downscale ERA5 fields into 0.1° synthetic data, which is combined with real high-resolution analysis and used to train a Swin Transformer forecasting model from scratch. The approach exploits super-resolution’s lower conditional entropy and greater robustness to noisy inputs than temporal forecasting, while hierarchical weather embedding, lead-time-aware loss weighting, and stochastic replay-buffer replacement stabilize autoregressive rollouts. On WeatherBench-style evaluation, BaguanHR outperforms IFS-HRES across more than 85% of lead times within 72 hours, reducing RMSE by 4.0% overall, 5.8% at 24 hours, and 9.7% at 72 hours. Scaling the training set from 7 to 18 years reduces RMSE by 4.6% at 72 hours and 4.9% at 120 hours, revealing a power-law data-scaling effect. The system uses 32 NVIDIA A800 GPUs and improves data-loading throughput by 4× through SSD partitioning and hybrid storage. Compared with coarse-to-fine baselines, it reduces RMSE by up to 20%, preserves more short-wavelength atmospheric structure, and improves extreme-event prediction, including moisture, thermal, and low-pressure-system extremes. The main conclusion is that high-resolution AI weather forecasting is primarily data-limited, and synthetic data construction can be more scalable than architectural model transfer.
Original abstract
The development of 0.1$^{\circ}$ global weather forecasting models based on machine learning (ML) is constrained by the limited availability of high-resolution data, as decades of reanalysis are only available at 0.25$^{\circ}$ resolution. While existing approaches fine-tune 0.25$^{\circ}$ forecast models on limited 0.1$^{\circ}$ samples, we show that this transfer is hindered by the irreversible information loss inherent in coarse-resolution forecasting. Therefore, we propose BaguanHR, a framework that shifts the focus from transferring models to transferring data. We first show that super-resolution (SR) has lower conditional entropy and input amplification than forecasting, making it a more robust vehicle for resolution transfer. By leveraging this advantage through variable-wise SR, we synthesize extensive 0.1$^{\circ}$ data from ERA5. BaguanHR's performance on the synthetic-plus-real dataset exceeds both ML-based methods and IFS-HRES, achieving superior performance across over 85% of the lead times within 72 hours. Furthermore, our findings highlight a power-law scaling effect, as a twofold increase in data reduces RMSE by 4.6% for 72-hour forecasting and 4.9% for 120-hour forecasting. Our results demonstrate that scaling high resolution ML-based forecasting is primarily a data bottleneck, and that variable-wise super-resolution provides a simple yet general solution to unlock long coarse-resolution reanalyses for high-resolution training.
Read the original paperMore in AI for Science
Browse all 43 papers →AI-guided high-throughput discovery of iridium- and ruthenium-free palladium-oxide catalysts for durable acidic oxygen evolution
Ken J. Jenewein, Faezeh Habib Zadeh, Xiaoxiao Wang, Gustavo Malkomes, Huafan Zhang, Natalie Page, Jae Jin Bang, Peter J. Santiago, Karla V. Contreras, Katherine K. Li, Allison Perna, Lorena M. Britton, Fahrettin Kilic, Kevin J. Cruse, Armin Taheri, Krishnanand Mallayya, Harley Quinn, Rebecca A. Durr, Peter A. Beaucage, John M. Gregoire, Rafael Gómez-Bombarelli
An AI-guided robotic lab discovered palladium-based catalysts that could make acidic water electrolysis more durable while reducing dependence on scarce iridium and ruthenium.
Discovery of radio emission from the exoplanet $β$ Pictoris b
Kevin N. Ortiz Ceballos, Edo Berger, Yvette Cendes
Astronomers have detected radio auroras from β Pictoris b, revealing that this distant giant planet has a magnetic field at least 1.25 kilogauss strong.
EurekaBench: Measuring Agentic Ability to Discover New Scientific Insights
Jiayi Geng, Zhengxuan Wu, Kevin S. Chen, Seungone Kim, Joseph Janssen, Zora Zhiruo Wang, Bhupalee Kalita, Runtian Gao, Aaron Ho, Andrew Oakleigh Nelson, Olexandr Isayev, Francisco Villaescusa-Navarro, Ching-Yao Lai, Howard Chen, Graham Neubig
EurekaBench tests whether AI agents can move beyond accurate prediction to uncover mechanisms and insights that genuinely advance scientific understanding.