ERA5 Patchify: Global Weather Reanalysis Data Tiled for ML Training

ERA5 Patchified Dataset contains patches extracted from the ECMWF ERA5 global atmospheric reanalysis on a 0.25-degree grid. The data is tiled into 128x128 non-overlapping patches with float16 normalized channels and stored in Parquet files. The dataset was created by meteolibre-dev and was last updated on 2026-04-20.

Use Cases

• Training machine learning models for weather prediction based on the tiled reanalysis data.
• Developing downscaling or super-resolution models based on the 128x128 patch structure.
• Benchmarking ML architectures for geospatial data processing based on the normalized float16 channels.
• Implementing a hybrid training-inference pipeline based on the described workflow with ERA5 and IFS HRES data.

Strengths

• Data is derived from the authoritative ECMWF ERA5 reanalysis.
• Patches are provided on a high-resolution 0.25-degree global grid.
• Data is structured into a consistent 128x128 patch size suitable for ML input.
• Values are stored as normalized float16, which can reduce storage and memory requirements.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, total size, and specific file formats beyond Parquet are unknown.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ECMWF ERA5 reanalysis.

Collection Method

Patches extracted and tiled from the global reanalysis grid.

Time Range

null

Freshness

Last updated 2026 04 20 15:50:29; freshness should be verified.

Geography

Global coverage on a 0.25-degree grid.