Model Card for ESFM/ESFM_s_wm_modis

Final ESFM small checkpoint for six-hour dense global forecasting of four sparse MODIS precipitable-water products from Terra/Aqua swaths, with ERA5 atmospheric auxiliary inputs.

Checkpoint selection: Six-hour forecasting of the four documented MODIS precipitable-water products from sparse gridded swaths and the matching auxiliary ERA5 inputs.

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 deterministic sparse-satellite checkpoint; modified 3D Swin-UNet encoder-decoder
Model size: Approximately 115 million parameters
Masking protocol: Variable, pressure-level, and spatial masking
Ensemble members: 1
Forecast lead time: 6 hours
License: MIT
Repository: https://huggingface.co/ESFM/ESFM_s_wm_modis

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

Six-hour forecasting of the four documented MODIS precipitable-water products from sparse gridded swaths and the matching auxiliary ERA5 inputs.

Downstream Use

Sparse-satellite forecasting research and supervised adaptation to other satellite products with carefully matched preprocessing and validation.

Out-of-Scope Use

Not a general MODIS product generator, not valid without the auxiliary variables expected by the config, and not a substitute for calibrated observations or data assimilation in safety-critical use.

Bias, Risks, and Limitations

The model is trained against MODIS retrievals and can reproduce their biases. A closer match to MODIS than ERA5-derived PWV does not establish greater accuracy relative to the true atmosphere. Applicability to other sensors, retrieval versions, or preprocessing pipelines is unverified.

All ESFM checkpoints are research artifacts. Validate forecasts for the target variables, stations or 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 ESFM with the matching config and load the state dictionary. The released notebook demonstrates checkpoint download and architecture construction.

from huggingface_hub import hf_hub_download

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

The public ERA5 notebook can download this checkpoint but cannot supply valid MODIS inputs. Use configs/config_ESFM_s_wm_modis.yaml with locally preprocessed MODIS data and the released inference pipeline.

Clone the implementation first:

git clone https://github.com/swiss-ai/ESFM.git
cd ESFM

Training Details

Training Data

Hourly Terra MOD05 and Aqua MYD05 infrared and near-infrared total precipitable-water products, gridded to 0.25 degrees, plus ERA5 temperature and specific humidity. Training covers 2014-2020; the manuscript evaluates 2023-2024.

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

Training Procedure

Initialized from the masked ERA5 checkpoint, finetuned for 50,000 steps as ESFM_s_wm_modis_pre, then continued for 10,000 steps on 16 GPUs.

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 the final model reconstructs dense, physically plausible PWV fields from sparse swaths and remains stable in reported rollouts. Relative MAE and Pearson correlation details are maintained in the linked preprint.

Detailed numerical results are intentionally not copied into this card.

Technical Specifications

ESFM uses variable-specific tokenization, axial attention across variables, perceiver aggregation, a 3D Swin-UNet backbone, and a decoder queried at target pressure levels. Missing patches are represented by learnable NaN tokens. Resolution-specific tokenizers and station mapping are enabled for the relevant sparse-data configs. The ensemble checkpoint additionally applies member-conditioned AdaLN-Zero after the backbone.

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: 12 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}
}