Datasets:

deinal
/

spacecast-data

license: cc-by-4.0
tags:
  - space
  - plasma
  - physics
size_categories:
  - 100K<n<1M
pretty_name: Vlasiator Dataset for Machine Learning Studies
citation: |
  @misc{vlasiator2025mldata,
    title        = {Vlasiator Dataset for Machine Learning Studies},
    author       = {Zaitsev, Ivan and Holmberg, Daniel and Alho, Markku and Bouri, Ioanna and 
                    Franssila, Fanni and Jeong, Haewon and Palmroth, Minna and Roos, Teemu},
    year         = {2025},
    publisher    = {Hugging Face},
    url          = {https://huggingface.co/datasets/deinal/spacecast-data},
    doi          = {10.57967/hf/7027},
  }

Vlasiator Dataset for Machine Learning Studies

The data is stored in Zarr.

It can be downloaded to a local data directory with:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="deinal/spacecast-data",
    repo_type="dataset",
    local_dir="data"
)

This will yield a local data folder that can be used with spacecast:

data/
├── graph/                 - Directory containing graphs for training
├── run_1.zarr/            - Vlasiator run 1 with ρ = 0.5 cm⁻³ solar wind
├── run_2.zarr/            - Vlasiator run 2 with ρ = 1.0 cm⁻³ solar wind
├── run_3.zarr/            - Vlasiator run 3 with ρ = 1.5 cm⁻³ solar wind
├── run_4.zarr/            - Vlasiator run 4 with ρ = 2.0 cm⁻³ solar wind
├── static.zarr/           - Static features x, z, r coordinates
├── vlasiator_config.yaml  - Configuration file for neural-lam
├── vlasiator_run_1.yaml   - Configuration file for datastore 1, referred to from vlasiator_config.yaml
├── vlasiator_run_2.yaml   - Configuration file for datastore 2, referred to from vlasiator_config.yaml
├── vlasiator_run_3.yaml   - Configuration file for datastore 3, referred to from vlasiator_config.yaml
└── vlasiator_run_4.yaml   - Configuration file for datastore 4, referred to from vlasiator_config.yaml

Preprocess the runs with mllam-data-prep, run:

mllam_data_prep data/vlasiator_run_1.yaml
mllam_data_prep data/vlasiator_run_2.yaml
mllam_data_prep data/vlasiator_run_3.yaml
mllam_data_prep data/vlasiator_run_4.yaml

This produces training-ready Zarr stores in the data directory.

Simple, multiscale, and hierarchical graphs are included already, but can be created using the following commands:

python -m neural_lam.create_graph --config_path data/vlasiator_config.yaml --name simple --levels 1 --coarsen-factor 5 --plot
python -m neural_lam.create_graph --config_path data/vlasiator_config.yaml --name multiscale --levels 3 --coarsen-factor 5 --plot
python -m neural_lam.create_graph --config_path data/vlasiator_config.yaml --name hierarchical --levels 3 --coarsen-factor 5 --hierarchical --plot

Citation

@misc{vlasiator2025mldata,
  title        = {Vlasiator Dataset for Machine Learning Studies},
  author       = {Zaitsev, Ivan and Holmberg, Daniel and Alho, Markku and Bouri, Ioanna and 
                  Franssila, Fanni and Jeong, Haewon and Palmroth, Minna and Roos, Teemu},
  year         = {2025},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/deinal/spacecast-data},
  doi          = {10.57967/hf/7027},
}