Research map / Video / Video segmentation and tracking
STORM: Synergistic Cross-Scale Spatio-Temporal Modeling for Weather Forecasting
Qihe Huang, Zhengyang Zhou, Yangze Li, Jiaming Ma, Kuo Yang, Binwu Wang, Xu Wang, Yang Wang
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 |
|---|---|---|---|
| add temporal evolution encoder (ablation) | STORM · ERA5 global, 10 variables | RMSE | Improved |
| add multi-scale modeling instead of single scale (ablation) | STORM · ERA5 global, 10 variables | RMSE | Improved |
| add spatial encoder (ablation) | STORM · ERA5 global, 10 variables | RMSE | Improved |
| add cross-scale message passing (ablation) | STORM (3 scales) · ERA5 global | ACC: 0.976 -> 0.984 | Improved |
| add cross-scale message passing (ablation) | STORM (5 scales) · ERA5 global | RMSE: 16.879 -> 15.757 | Improved |
| use different temporal scales across branches instead of the same (ablation) | STORM (3 scales) · ERA5 global | ACC: 0.948 -> 0.984 | Improved |
| use different temporal scales across branches instead of the same (ablation) | STORM (2 scales) · ERA5 global | RMSE: 16.547 -> 16.093 | Improved |
| increase number of scales from 1 to 5 branches (d_model 64 to 256-512) (ablation) | STORM · ERA5 global | ACC: 0.941 -> 0.988 | Improved |
| increase number of branches from 4 to 5 (d_model 256) (ablation) | STORM · ERA5 global | ACC: 0.987 -> 0.988 | Improved |
| increase d_model from 256 to 512 at 4 branches (ablation) | STORM · ERA5 global | ACC: 0.987 -> 0.986 | Hurt |
| use multi-scale STORM instead of Pangu | STORM vs Pangu · Efficiency, inference | seconds per 100 samples: 5.11 -> 0.87 | Improved |
| use STORM instead of Triton on regional forecasting at 10 days | STORM vs Triton · South America 1deg, 10-day | ACC: 0.251 -> 0.499 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SimVP — SimVP: Simpler Yet Better Video Prediction (CVPR 2022) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ChaosBench: A Multi-Channel, Physics-Based Benchmark for Subseasonal-to-Seasonal Climate Prediction (NeurIPS 2024 · similarity 0.87)
- Scaling Storm-Resolving Atmospheric AI Simulation to the Entire Planet (NeurIPS 2026 · similarity 0.86)
- Scaling transformer neural networks for skillful and reliable medium-range weather forecasting (NeurIPS 2024 · similarity 0.85)
- OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales (NeurIPS 2025 · similarity 0.85)
- Integrating Multi-Source Data for Long Sequence Precipitation Forecasting (AAAI 2025 · similarity 0.85)
- On the Integration of Spatial-Temporal Knowledge: A Lightweight Approach to Atmospheric Time Series Forecasting (NeurIPS 2025 · 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.
