Research map / Architectures and efficiency / RNNs and normalization
Compressing multidimensional weather and climate data into neural networks
Langwen Huang, Torsten Hoefler
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 |
|---|---|---|---|
| Replace original training data with NN-compressed data | WeatherBench CNN · Dataset 3, geopotential 500 hPa | weighted RMSE: 632.9 -> 637.3 (+0.7%) | Hurt |
| Replace original training data with NN-compressed data | WeatherBench CNN · Dataset 3, temperature 850 hPa | weighted RMSE: 2.906 -> 2.944 (+1.3%) | Hurt |
| Replace original training data with NN-compressed data | WeatherBench CNN · Dataset 4, geopotential 500 hPa | weighted RMSE: 688.8 -> 697.3 (+1.2%) | Hurt |
| Replace original training data with NN-compressed data | WeatherBench CNN · Dataset 4, temperature 850 hPa | weighted RMSE: 2.834 -> 2.888 (+1.9%) | Hurt |
| Add output scaling (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 818 -> 420 | Improved |
| Add XYZ transform to output scaling (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 420 -> 395 | Improved |
| Add Fourier features to output scaling and XYZ transform (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 395 -> 347 | Improved |
| Add skip connection and XYZ transform to output scaling (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 420 -> 292 | Improved |
| Add GELU to output scaling, XYZ transform, and skip connection (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 292 -> 254 | Improved |
| Add batch norm to output scaling, GELU, XYZ transform, and skip connection (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 254 -> 158 | Improved |
| Add Fourier features to output scaling, GELU, XYZ transform, and skip connection (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 158 -> 78 | Improved |
| Add XYZ transform to output scaling and Fourier features (ablation) | NN Compression · ERA5 geopotential, Dataset 1 | weighted RMSE: 346 -> 347 | No clear effect |
Closest papers by meaning
- Scaling Storm-Resolving Atmospheric AI Simulation to the Entire Planet (NeurIPS 2026 · similarity 0.84)
- DiffDA: a Diffusion model for weather-scale Data Assimilation (ICML 2024 · similarity 0.84)
- OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales (NeurIPS 2025 · similarity 0.84)
- Scaling transformer neural networks for skillful and reliable medium-range weather forecasting (NeurIPS 2024 · similarity 0.83)
- HR-Extreme: A High-Resolution Dataset for Extreme Weather Forecasting (ICLR 2025 · similarity 0.83)
- AstroCompress: A benchmark dataset for multi-purpose compression of astronomical data (ICLR 2025 · similarity 0.83)
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.
