SSL4EO-S12-v1.1

Attention: The Zarr Chunk file version of SSL4EO-S12-v1.1 was moved to embed2scale/SSL4EO-S12-v1.1-Zarr. This repository contains data that can be used with webdataset.

The dataset includes 246,144 locations with four timestamps each from the modalities S2L1C, S2L2A, S2RGB, S1GRD, LULC, DEM, and NDVI. We refer to our technical report for details.

The samples are stored in as Zarr Zip files (zarr version 2) with the metadata directly aligned as additional data variables. The samples are randomly grouped in tar shards and can be loaded easily via webdataset. Because we want to keep the loading of specific modalities flexible, we differ from the standard and save each modality in separate folders. Therefore, we share custom code for data loading via webdataset in our GitHub repository.

Sentinel-2 and Sentinel-1 time series examples with four seasonal images:

Modality examples in SSL4EO-S12 v1.1:

Sentinel-2 L2A example, loaded as xarray dataset:

<xarray.Dataset> Size: 7MB
Dimensions:      (time: 4, band: 12, y: 264, x: 264)
Coordinates:
  * time         (time) datetime64[ns] 32B 2020-11-16T02:30:01 ... 2021-08-08...
  * band         (band) <U3 144B 'B01' 'B02' 'B03' 'B04' ... 'B09' 'B11' 'B12'
  * y            (y) float64 2kB 4.728e+06 4.728e+06 ... 4.725e+06 4.725e+06
  * x            (x) float64 2kB 7.149e+05 7.149e+05 ... 7.175e+05 7.175e+05
    sample       <U7 28B '0080717'
    spatial_ref  int64 8B 0
Data variables:
    bands        (time, band, y, x) int16 7MB 463 457 451 445 ... 686 663 635
    center_lat   float64 8B 42.66
    center_lon   float64 8B 125.6
    cloud_mask   (time, y, x) uint8 279kB 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0
    crs          int64 8B 32651
    file_id      (time) <U38 608B '20201116T023001_20201116T022955_T51TYH' .....
    sample_id    (time) <U9 144B '0080717_0' '0080717_1' '0080717_2' '0080717_3'

Sentinel-1 GRD example, loaded as xarray dataset:

Dimensions:      (time: 4, band: 2, y: 264, x: 264)
Coordinates:
  * time         (time) datetime64[ns] 32B 2020-11-27T21:46:46 ... 2021-08-30...
  * band         (band) <U2 16B 'vv' 'vh'
  * y            (y) float64 2kB 4.728e+06 4.728e+06 ... 4.725e+06 4.725e+06
  * x            (x) float64 2kB 7.149e+05 7.149e+05 ... 7.175e+05 7.175e+05
    sample       <U7 28B '0080717'
    spatial_ref  int64 8B 0
Data variables:
    bands        (time, band, y, x) float16 1MB -15.13 -17.44 ... -18.64 -19.33
    center_lat   float64 8B 42.66
    center_lon   float64 8B 125.6
    crs          int64 8B 32651
    file_id      (time) <U67 1kB 'S1B_IW_GRDH_1SDV_20201127T214646_20201127T2...
    sample_id    (time) <U9 144B '0080717_0' '0080717_1' '0080717_2' '0080717_3'

Download

You can download the dataset via the Hugging Face CLI (pip install huggingface_hub). Please note that the full dataset requires 2.3TB of storage.

hf download embed2scale/SSL4EO-S12-v1.1 --repo-type dataset --local-dir data/SSL4EOS12

If you like to download only a subset of the data, you can specify it with --include.

# Only download val data
hf download embed2scale/SSL4EO-S12-v1.1 --repo-type dataset --include "val/*" --local-dir data/SSL4EOS12

# Only download a single modality (e.g., S2L2A)
hf download embed2scale/SSL4EO-S12-v1.1 --repo-type dataset --include "*/S2L2A/*" --local-dir data/SSL4EOS12

For development, webdataset supports data streaming and does not need any local data (depending on your internet connection). Code for data loading via webdataset is provided in the GitHub repository.

Metadata

We provide train_metadata.parquet and val_metadata.parquet files with the following columns:

tar                                     ssl4eos12_shard_000001.tar
zarr                  ssl4eos12_val_seasonal_data_0000001.zarr.zip
sample_id                                                  0080717
split                                                          val
center_lon                                              125.637763
center_lat                                               42.659342
crs                                                          32651
bounds                  [714870.0, 4725040.0, 717510.0, 4727680.0]
geometry         POLYGON ((125.65334657340223 42.64709631345207...
S2_time_0                                      2020-11-16 02:30:01
S2_time_1                                      2021-03-04 02:36:09
S2_time_2                                      2021-05-18 02:35:51
S2_time_3                                      2021-08-08 02:25:49
S1_time_0                                      2020-11-27 21:46:46
S1_time_1                                      2021-02-26 21:38:39
S1_time_2                                      2021-05-26 21:46:46
S1_time_3                                      2021-08-30 21:46:51
cloud_cover_0                                                  0.0
cloud_cover_1                                                  0.0
cloud_cover_2                                                  0.0
cloud_cover_3                                               0.0011

The sample_id is similar to the v1 dataset. S2_time_{i} is shared between S2L1C, S2L2A, S2RGB, and NDVI. LULC is augmented with the cloud and ice/snow mask from this timestamp.

Citation

If you use this dataset in your work, please cite:

@article{blumenstiel2025ssl4eos12,
  title={{SSL4EO-S12} v1.1: A Multimodal, Multiseasonal Dataset for Pretraining, Updated},
  author={Blumenstiel, Benedikt and Ait Ali Braham, Nassim and Albrecht, Conrad M and Maurogiovanni, Stefano and Fraccaro, Paolo},
  journal={arXiv preprint arXiv:2503.00168},
  year={2025}
}

This dataset is an updated version of:

@article{wang2022ssl4eo,
  title={{SSL4EO-S12}: A large-scale multimodal, multitemporal dataset for self-supervised learning in Earth observation [Software and Data Sets]},
  author={Wang, Yi and Ait Ali Braham, Nassim and Xiong, Zhitong and Liu, Chenying and Albrecht, Conrad M and Zhu, Xiao Xiang},
  journal={IEEE Geoscience and Remote Sensing Magazine},
  volume={11},
  number={3},
  pages={98--106},
  year={2023},
  publisher={IEEE}
}