Research map / Image generation / Super-resolution
Pushing the Limits of High-Resolution Weather Forecasting through Data Scaling
Yang Zhao, Peisong Niu, Tian Zhou, Ziqing Ma, Guanlong Ma, Rong Jin, Huiling Yuan, Liang Sun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train 0.1 deg model from scratch on synthetic-plus-real data instead of fine-tuning a 0.25 deg model (coarse-to-fine) (ablation) | BaguanHR · 2025 analysis data | RMSE: up to 20% RMSE reduction | Improved |
| double training data volume via SR-synthesized 0.1 deg data (7 to 18 years) (ablation) | BaguanHR · 72h forecast | RMSE: -4.6% RMSE | Improved |
| increase training data from 7 to 18 years (ablation) | BaguanHR · 120h forecast | RMSE: -4.9% RMSE | Improved |
| use per-variable super-resolution instead of full-variable SR (ablation) | SR data generator · surface variables (sp, t2m, msl) | RMSE: 90.7% for sp, >20% for t2m and msl; 1.6%-4.7% upper-air | Improved |
| add lead-time-aware loss weighting (ablation) | BaguanHR replay buffer · MSL 72h | RMSE: 145.32 -> 144.57 | Improved |
| add stochastic buffer replacement (ablation) | BaguanHR replay buffer · MSL 72h | RMSE: 147.41 -> 144.57 | Improved |
| add stochastic buffer replacement and loss weighting (ablation) | BaguanHR · T2M 72h | RMSE: 1.12 -> 1.09 | Improved |
| add lead-time-aware loss weighting (ablation) | BaguanHR · TCC 72h | RMSE: 0.289 -> 0.287 | No clear effect |
| native 0.1 deg forecasting instead of post-hoc Swin2SR upscaling of 0.25 deg forecasts | BaguanHR vs Baguan+swin2sr · power spectrum / extremes | short-wavelength power, extreme TS: TS gains exceeding 10% in most cases | Improved |
| compare BaguanHR to IFS-HRES at 10-15 day lead times | BaguanHR · long lead times | RMSE | No clear effect |
Closest papers by meaning
- Precipitation Downscaling with Spatiotemporal Video Diffusion (NeurIPS 2024 · similarity 0.86)
- HR-Extreme: A High-Resolution Dataset for Extreme Weather Forecasting (ICLR 2025 · similarity 0.85)
- Scaling Storm-Resolving Atmospheric AI Simulation to the Entire Planet (NeurIPS 2026 · similarity 0.85)
- Satellite Observations Guided Diffusion Model for Accurate Meteorological States at Arbitrary Resolution (CVPR 2025 · similarity 0.85)
- DAWP: A framework for global observation forecasting via Data Assimilation and Weather Prediction in satellite observation space (NeurIPS 2025 · similarity 0.85)
- Scaling transformer neural networks for skillful and reliable medium-range weather forecasting (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
