Datasets:

meteolibre-dev
/

weather_france_radar_satellite_gs

Modalities:

Formats:

Size:

Tags:

Libraries:

Dataset card Data Studio Files Files and versions

xet

Community

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Only members of EUMETSAT can access this dataset.

This dataset is a ready-to-use fusion of multiple datasets for the France region, including:

Ground station data: Sourced from data.gouv.fr with a 1-hour resolution. It includes 7 weather KPIs: "RR1", "FF", "DD", "T", "U", "PMER", and "VV".
Radar imagery: Radar rainfall accumulation images from the Météo-France open data initiative.
Geospatial data: Land cover and ground height information from EarthEnv and OpenTopography.
Satellite imagery: Sourced from the EUMETSAT platform, containing a subset of channels: vis_04, vis_09, nir_13, and nir_16.

Time Period

The data covers the period from January 2025 to July 2025.

Resolution and Projection

All data have a 1km spatial resolution (using EPSG:32630) and a 30-minute temporal resolution.

Data Description

The final dataset is structured in .zip files as follows:

ground_height_image/: Contains ground height information as NumPy .npz files (256x256 images).
groundstation/: Contains ground station readings as arrays of shape (9, 256, 256, 7), where 9 is the number of time steps and 7 is the number of KPIs.
index.json: A JSON file that indexes the entire dataset.
landcover/, landcover_image/: Contain land cover data (256x256 images).
radar/: Contains radar rainfall accumulation imagery as arrays of shape (9, 256, 256), where 9 is the number of time steps.
satellite/: Contains satellite imagery as arrays of shape (9, 256, 256, 4), where 9 is the number of time steps and 4 is the number of selected channels.

Each subdirectory contains .npz files for different timestamps and locations, all indexed by index.json.

Data Visualization

Here's a quick visualization of the data:

We extracted full images of France and segmented them into smaller patches for easier ingestion into a training pipeline.

Quick start (remote)

import numpy as np
from datasets import load_dataset

# The command is exactly the same!
# The library handles the sharded files behind the scenes.
ds = load_dataset("meteolibre-dev/mtg_meteofrance_256", streaming=True)

# You can iterate over it as if it were a single file
for example in ds["train"]:

    # Get the byte string
    satellite_bytes = example['satellite']
    # Convert bytes back to a NumPy array
    # You need to specify the correct dtype and shape
    satellite_array = np.frombuffer(satellite_bytes, dtype=np.int16).reshape(9, 4, 256, 256)

    hour = example['hour']
    minute = example['minute']
    datetime = example['datetime']
    radar =  np.frombuffer(example['radar'], dtype=np.float32).reshape(9, 256, 256)
    groundstation =  np.frombuffer(example['groundstation'], dtype=np.float32).reshape(9, 256, 256, 7)
    ground_height =  np.frombuffer(example['ground_height'], dtype=np.float32).reshape(256, 256)
    landcover =  np.frombuffer(example['landcover'], dtype=np.float32).reshape(256, 256, 4)

Quick start (local)

First the command :

hf download meteolibre-dev/mtg_meteofrance_256 --repo-type dataset --local-dir data/

Downloads last month: 30