CRA5: Compressed Climate Science Data from ERA5 via VAEformer

A compressed version of the ERA5 climate reanalysis dataset, created using a variational autoencoder transformer (VAEformer) model. The dataset is hosted on Hugging Face by author 'taohan10200' and was last updated in April 2026. The primary purpose is to address storage challenges associated with massive climate datasets through extreme compression.

Use Cases

• Benchmarking novel data compression techniques based on the described VAEformer architecture.
• Training machine learning models on compressed climate data representations for tasks like weather prediction.
• Conducting climate analysis where storage efficiency for large-scale reanalysis data is a constraint.

Strengths

• Dataset is derived from the widely-used ERA5 climate reanalysis, a standard reference in the field.
• The compression method, VAEformer, is described in an associated research paper, providing technical context.

Limitations

• Description metadata is limited; actual data quality, format, and structure require manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's size, row count, and specific file formats are unknown, which may limit suitability assessment.

Provenance

Source

Derived from the ERA5 climate reanalysis dataset.

Collection Method

Created via a dual-stage extreme compression process using a variational autoencoder transformer (VAEformer).

Freshness

Last updated 2026-04-17 15:07:58; freshness should be verified.