Research map / Probabilistic and scientific modeling / Dynamical systems
Improving Extreme Wind Prediction with Frequency-Informed Learning
Chenrui Xu, Xi Huang, Ying-Jun Angela Zhang, Jianwei Huang
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 gradient-penalized loss (ablation) | full model · ERA5 regional wind (next-frame) | Ex-RMSE: 0.2632 -> 0.1868 | Improved |
| add gradient-penalized loss (ablation) | full model · ERA5 regional wind (next-frame) | RMSE: 0.3351 -> 0.3287 | No clear effect |
| add frequency separation (ablation) | full model · ERA5 regional wind (next-frame) | Ex-RMSE: 0.2703 -> 0.1868 | Improved |
| add frequency separation (ablation) | full model · ERA5 regional wind (next-frame) | RMSE: 0.4199 -> 0.3287 | Improved |
| add NS-based structure (ablation) | full model · ERA5 regional wind (next-frame) | Ex-RMSE: 0.2363 -> 0.1868 | Improved |
| add NS-based structure (ablation) | full model · ERA5 regional wind (next-frame) | RMSE: 0.3754 -> 0.3287 | Improved |
| add body force and frequency modules to NS operator alone (ablation) | NS operator · ERA5 regional wind (next-frame) | RMSE: 0.7061 -> 0.3287 | Improved |
| use full model instead of PINN | PINN · ERA5, lead 1h | Ex-RMSE: 0.2541 -> 0.1868 | Improved |
| use full model instead of ConvLSTM | ConvLSTM · ERA5, lead 5h | RMSE: 1.0644 -> 0.9076 | Improved |
| use full model instead of CNN | CNN · ERA5, lead 5h | RMSE: 2.0757 -> 0.9076 | Improved |
| increase gradient penalty weight lambda to 0.15 or above (ablation) | gradient-penalized loss · ERA5 MFWPN baseline | convergence | Hurt |
| use small positive lambda (0.02) in gradient-penalized loss instead of MSE (ablation) | MFWPN baseline · ERA5 | Ex-RMSE | Improved |
Closest papers by meaning
- Fixing the Double Penalty in Data-Driven Weather Forecasting Through a Modified Spherical Harmonic Loss Function (ICML 2025 · similarity 0.83)
- Learning from Polar Representation: An Extreme-Adaptive Model for Long-Term Time Series Forecasting (AAAI 2024 · similarity 0.82)
- Explanation of Dynamic Physical Field Predictions using WassersteinGrad: Application to Autoregressive Weather Forecasting (NeurIPS 2026 · similarity 0.81)
- OceanCBM: A Concept Bottleneck Model for Mechanistic Interpretability in Ocean Forecasting (NeurIPS 2026 · similarity 0.81)
- Generalizing Weather Forecast to Fine-grained Temporal Scales via Physics-AI Hybrid Modeling (NeurIPS 2024 · similarity 0.81)
- ChaosBench: A Multi-Channel, Physics-Based Benchmark for Subseasonal-to-Seasonal Climate Prediction (NeurIPS 2024 · similarity 0.81)
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.
