Research map / Time series / Long-term forecasting
STCast: Adaptive Boundary Alignment for Global and Regional Weather Forecasting
Hao Chen, Tao Han, Jie Zhang, Song Guo, Lei Bai
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 Spatial-Aligned Attention with dynamic global-regional boundary (ablation) | STCast regional · Regional forecast (East Asia), 10-day | RMSE: 0.7610 -> 0.4921 | Improved |
| initialize global-regional distribution with distance-decay prior instead of Xavier uniform (ablation) | SAA · Regional forecast (East Asia), 10-day | RMSE: 0.7192 -> 0.4921 | Improved |
| add Temporal Mixture-of-Experts instead of MLP block (ablation) | STCast global · Global forecast, 10-day (normalized) | RMSE: 0.6426 -> 0.5763 | Improved |
| add month embedding to MoE (ablation) | TMoE · Global forecast, 10-day (normalized) | RMSE: 0.6049 -> 0.5763 | Improved |
| use Gaussian month embedding instead of one-hot (ablation) | TMoE with 12 experts · Global forecast, 10-day (normalized) | RMSE: 0.5911 -> 0.5763 | Improved |
| increase number of experts from 4 to 6 to 12 (ablation) | TMoE · Global forecast, 10-day (normalized) | RMSE: 0.6433 -> 0.5932 -> 0.5763 | Improved |
| increase experts to 24 (biweekly) (ablation) | TMoE · Global forecast | parameters: 1164.7M parameters | Hurt |
| use Great Circle distance instead of Manhattan or Euclidean in SAA (ablation) | SAA · Regional forecast (East Asia), 10-day | RMSE: 0.4945 -> 0.4921 (Manhattan); 0.5088 -> 0.4921 (Euclidean) | Improved |
| use STCast with SAA instead of directly trained STCast / OneForecast | STCast regional · Regional forecast (East Asia), mean over variables | RMSE and ACC: RMSE decreases by 0.05 while ACC increases by 0.1 | Improved |
| ensemble forecasting with Perlin noise on STCast | STCast · Ensemble forecast, 10-day | RMSE: 0.5107 -> 0.5113 | No clear effect |
| ensemble forecasting for baselines such as Pangu and Fuxi | Pangu, Fuxi · Ensemble forecast, 10-day | RMSE: Pangu 0.6213 -> 0.5205; Fuxi 0.6968 -> 0.5498 | Improved |
Closest papers by meaning
- On the Integration of Spatial-Temporal Knowledge: A Lightweight Approach to Atmospheric Time Series Forecasting (NeurIPS 2025 · similarity 0.86)
- ChaosBench: A Multi-Channel, Physics-Based Benchmark for Subseasonal-to-Seasonal Climate Prediction (NeurIPS 2024 · similarity 0.85)
- EWMoE: An Effective Model for Global Weather Forecasting with Mixture-of-Experts (AAAI 2025 · similarity 0.85)
- ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data (AAAI 2025 · similarity 0.85)
- Integrating Multi-Source Data for Long Sequence Precipitation Forecasting (AAAI 2025 · similarity 0.85)
- High-Dynamic Radar Sequence Prediction for Weather Nowcasting Using Spatiotemporal Coherent Gaussian Representation (ICLR 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.
