Research map / Probabilistic and scientific modeling / Molecules and proteins
Transforming Weather Data from Pixel to Latent Space
Sijie Zhao, Feng Liu, Xueliang Zhang, Hao Chen, Tao Han, Junchao Gong, Ran Tao, Pengfeng Xiao, Xinyu Gu, 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 |
|---|---|---|---|
| increase the codebook size from 2^16 up to 2^128 (ablation) | WLA · ERA5 temperature | reconstruction quality | Improved |
| increase the codebook size (ablation) | WLA · ERA5 temperature | compression ratio | Hurt |
| increase codebook size beyond 2^128 (ablation) | WLA · ERA5 temperature | reconstruction quality | No clear effect |
| increase the number of input pressure levels from 6 to 25 (ablation) | WLA · ERA5 temperature | compression ratio: 150.2 -> 625.9 | Improved |
| increase the number of input pressure levels (ablation) | WLA · ERA5 temperature | reconstruction quality | Hurt |
| use WLA instead of VAEformer on w500 | VAEformer · ERA5 w500 | weighted RMSE: 0.117 -> 0.076 | Improved |
| use WLA instead of VAEformer on q1000 | VAEformer · ERA5 q1000 | weighted RMSE: 0.00035 -> 0.00028 | Improved |
| use WLA instead of VAEformer on tp6h | VAEformer · ERA5 tp6h | weighted RMSE: 0.80 -> 0.47 | Improved |
| use WLA (upper-air) instead of VAEformer on compression ratio | VAEformer · ERA5 upper-air | compression ratio: 323.1 -> 625.9 | Improved |
| use WLA instead of VQVAE on compression ratio | WLA · ERA5 upper-air | compression ratio: 1100.0 -> 625.9 | Hurt |
| run forecasting in latent space (Window-ViT-latent) instead of pixel space | Window-ViT · ERA5 T850 SEDI | SEDI | Improved |
| run forecasting in latent space | Window-ViT · ERA5 T850 5-day | RMSE | No clear effect |
Closest papers by meaning
- Compressing multidimensional weather and climate data into neural networks (ICLR 2023 · similarity 0.87)
- Probabilistic Precipitation Nowcasting with Rectified Flow Transformers (CVPR 2026 · similarity 0.86)
- OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales (NeurIPS 2025 · similarity 0.85)
- EMFormer: Efficient Multi-Scale Transformer for Accumulative Context Weather Forecasting (ICML 2026 · similarity 0.85)
- WeatherGFM: Learning a Weather Generalist Foundation Model via In-context Learning (ICLR 2025 · similarity 0.85)
- Scaling transformer neural networks for skillful and reliable medium-range weather forecasting (NeurIPS 2024 · similarity 0.84)
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.
