Research map / Probabilistic and scientific modeling / Molecules and proteins
ClimaX: A foundation model for weather and climate
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| pretrain on CMIP6 and finetune for downstream S2S prediction | ClimaX · ERA5-S2S weeks 3-4 and weeks 5-6 | RMSE: 1.96 -> 1.89 (T850 weeks 3-4); 1.96 -> 1.92 (T850 weeks 5-6); 1.79 -> 1.66 (T2m weeks 3-4); 1.90 -> 1.70 (T2m weeks 5-6) | Helped |
| freeze attention layers during ClimateBench finetuning (ablation) | ClimaX · ClimateBench temperature predictions | Total: 0.300 -> 0.297 (surface temperature); 11.952 -> 10.739 (diurnal temperature range) | Helped |
| freeze attention layers during ClimateBench finetuning (ablation) | ClimaX · ClimateBench precipitation predictions | RMSE: 0.554 -> 0.549 (precipitation); 1.575 -> 1.572 (90th percentile precipitation) | Mixed |
| use pretrained ClimaX rather than scratch-trained Cli-ViT (ablation) | ClimaX · ClimateBench | Total: 0.305 -> 0.300 (surface temperature); 15.792 -> 11.952 (diurnal temperature range); 3.430 -> 3.390 (precipitation); 4.44 | Mixed |
| use the pretrained model rather than scratch-trained Cli-ViT (ablation) | ClimaX · Downscaling MPI-ESM to ERA5 | RMSE: 811.61 -> 807.43 (Z500); 3.58 -> 3.49 (T850); 2.80 -> 2.79 (T2m); 4.01 -> 3.99 (U10); 4.07 -> 4.06 (V10) | Helped |
| predict all input variables in one finetuned model instead of finetuning each target separately (ablation) | ClimaX · Global weather forecasting | RMSE | Mixed |
| use iterative 6-hour forecasts rolled out to longer horizons instead of direct forecasts (ablation) | ClimaX · Global weather forecasting | RMSE | Hurt |
| condition on randomized lead time during finetuning (ablation) | ClimaX · Global weather forecasting, 6 hours to 7 days | RMSE | No clear effect |
| extend the lead-time-conditioned model beyond its finetuning range (ablation) | ClimaX-cont · Global weather forecasting, 2 weeks and 1 month | RMSE | Hurt |
| increase pretraining data and model size (ablation) | ClimaX · ERA5 3-day forecasting | RMSE | Mixed |
| increase data resolution from 5.625° to 1.40625° (ablation) | ClimaX · Global weather forecasting | RMSE | Helped |
Closest papers by meaning
- WeatherGFM: Learning a Weather Generalist Foundation Model via In-context Learning (ICLR 2025 · similarity 0.88)
- ChaosBench: A Multi-Channel, Physics-Based Benchmark for Subseasonal-to-Seasonal Climate Prediction (NeurIPS 2024 · similarity 0.86)
- Scaling transformer neural networks for skillful and reliable medium-range weather forecasting (NeurIPS 2024 · similarity 0.86)
- OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales (NeurIPS 2025 · similarity 0.86)
- EWMoE: An Effective Model for Global Weather Forecasting with Mixture-of-Experts (AAAI 2025 · similarity 0.86)
- ClimODE: Climate and Weather Forecasting with Physics-informed Neural ODEs (ICLR 2024 · 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.