Model Card for ESFM/ESFM_s_wm_ci

Final released CMIP6-initialized ablation checkpoint after masked ERA5 training. It tests whether pretraining on heterogeneous climate simulations improves ESFM initialization and transfer.

Checkpoint selection: Use for initialization and transfer ablations involving CMIP6 pretraining.

Model Details

Developed by: The ESFM research team, with the full contributor and author lists linked below.
Shared by: ESFM on Hugging Face
Model type: Final CMIP6-initialized comparison checkpoint (ESFM_s,ci); modified 3D Swin-UNet encoder-decoder
Model size: Approximately 115 million parameters
Masking protocol: Variable, pressure-level, and spatial masking
Forecast lead time: 6 hours
License: MIT
Repository: https://huggingface.co/ESFM/ESFM_s_wm_ci

Model Sources

Code: https://github.com/swiss-ai/ESFM
Paper: https://arxiv.org/abs/2605.00850
Project page: https://swiss-ai.github.io/ESFM/

The paper is currently available as an arXiv preprint.

Uses

Direct Use

Use for initialization and transfer ablations involving CMIP6 pretraining.

Downstream Use

Research on heterogeneous climate-model pretraining and adaptation to reanalysis or held-out climate simulations.

Out-of-Scope Use

Not the default ESFM checkpoint; use ESFM_s_wm for the manuscript's knowledge-distilled default model.

Bias, Risks, and Limitations

The model reflects biases from both CMIP6 simulations and ERA5. It is not rollout-specialized and is not an operational forecasting system.

All ESFM checkpoints are research artifacts. Users should validate forecasts for their variables, regions, seasons, lead times, missingness pattern, and decision context. Do not use the model as the sole basis for safety-critical decisions.

How to Get Started

The checkpoint is not packaged as a Hugging Face Transformers from_pretrained model. Construct the ESFM architecture with the matching repository config, then load the state dictionary. The released notebook contains the complete download, model-construction, normalization, and inference workflow.

git clone https://github.com/swiss-ai/ESFM.git
cd ESFM
# Open notebooks/inference_ESFMs_on_ERA5.ipynb

In the notebook, set:

EXPERIMENT_NAME = "ESFM_s_wm_ci"

To download the weights directly:

from huggingface_hub import hf_hub_download

model_name = "ESFM_s_wm_ci"
weights_path = hf_hub_download(
    repo_id=f"ESFM/{model_name}",
    filename=f"{model_name}.safetensors",
)
print(weights_path)

Set EXPERIMENT_NAME = "ESFM_s_wm_ci" in the released inference notebook, or use configs/config_ESFM_s_wm_ci.yaml.

Training Details

Training Data

Eight CMIP6 simulations for initialization, followed by WeatherBench2 ERA5 at 0.25-degree resolution.

Dataset preprocessing and the exact variable registry are documented in the ESFM repository and preprint.

Training Procedure

Continues from the 100,000-step ESFM_s_wm_ci_pre checkpoint with another 10,000 steps of masked ERA5 training. This experiment was run on 16 GPUs.

Training objective: Six-hour forecast learning, as specified above
Nominal architecture: ESFM small, approximately 115M parameters
Software environment: PyTorch/Lightning in the released NVIDIA PhysicsNeMo 25.03 container; lightning==2.5.1 is pinned in the Dockerfile
Training regime: Lightning precision="32-true" with FP32 parameters and optimizer state; selected model forward operations use CUDA BF16 autocasting through torch.autocast(dtype=torch.bfloat16).

Evaluation

The manuscript reports that CMIP6 initialization improves over a random start but does not match the knowledge-distilled initialization in the main six-hour ERA5 comparison. It also studies CMIP6 pretraining for transfer; current details are linked rather than duplicated here.

The manuscript uses held-out temporal data and reports task-appropriate metrics: latitude-weighted MAE and Pearson correlation for gridded deterministic forecasts, relative MAE for MODIS comparisons, station metrics for station models, and CRPS for ensembles. Detailed values are intentionally not copied into this card.

Technical Specifications

ESFM retains Aurora's 3D Swin-UNet backbone and adds variable-specific tokenization, axial attention across variables, perceiver aggregation across variables and pressure levels, learnable NaN tokens for missing patches, resolution-specific tokenizers where configured, and a decoder queried at target pressure levels. The small configuration uses a 256-dimensional embedding and approximately 115M parameters.

Environmental Impact

Hardware type: NVIDIA GH200 systems with four GPUs per node. This experiment was run on four nodes, totaling 16 GPUs.
Total training time: 22 hours
Compute location: Training used CSCS Alps infrastructure.

Citation

@misc{ozdemir2026esfm,
  title={Earth System Foundation Model (ESFM): A unified framework for heterogeneous data integration and forecasting},
  author={Firat Ozdemir and Yun Cheng and Salman Mohebi and Fanny Lehmann and Simon Adamov and Zhenyi Zhang and Leonardo Trentini and Dana Grund and Oliver Fuhrer and Torsten Hoefler and Siddhartha Mishra and Sebastian Schemm and Benedikt Soja and Mathieu Salzmann},
  year={2026},
  eprint={2605.00850},
  archivePrefix={arXiv},
  primaryClass={physics.ao-ph},
  url={https://arxiv.org/abs/2605.00850}
}