Global coverage of the embeddings, coloured by the top-3 principal components of the 768-d UniverSat vectors (mapped to RGB). Distinct colours mark distinct backscatter regimes — dense tropical forest (magenta), arid and desert surfaces (green), boreal and temperate land (tan), open water and other smooth low-backscatter surfaces (blue).

Core-S1RTC-UniverSat 🛰️

Dataset	Modality	Number of Embeddings	Sensing Type	Embedding Dim	Source Dataset	Source Model	Size
Core-S1RTC-UniverSat	Sentinel-1 RTC	1,464,989	SAR (dual-polarisation VV/VH, radiometrically terrain-corrected)	768	Core-S1RTC	UniverSat	4.2 GB

This dataset provides a dense, global set of whole-image embeddings for the Major TOM Core-S1RTC collection. Each Sentinel-1 RTC fragment is encoded into a single 768-dimensional vector with UniverSat, a resolution- and modality-agnostic transformer backbone for Earth Observation. It is the Sentinel-1 counterpart of Core-S2L2A-UniverSat, and an embedding expansion of Major TOM in the spirit of the Major TOM Global Embeddings project.

Content

Field	Type	Description
embedding	array	Raw UniverSat whole-image embedding (768 × float32)
grid_cell	string	Major TOM grid cell (e.g. 874D_289R)
timestamp	string	Sensing timestamp of the source product (e.g. 20191105T144246)
product_id	string	ID of the original Sentinel-1 product

Rows are stored across 13 data-*.parquet shards (~113k rows each), keyed to the source fragment by (grid_cell, product_id, timestamp), so embeddings can be joined back to imagery, geometry, and other Major TOM expansions of Core-S1RTC.

Note that Sentinel-1 and Sentinel-2 observe the same grid cell on different dates, so these rows join to Core-S2L2A-UniverSat on grid_cell only, never on timestamp.

Input Data

• Sentinel-1 RTC fragments from Major TOM Core-S1RTC, read as 3 channels — VV, VH, and the derived polarimetric ratio VV/VH — converted from linear power to dB and normalised per channel with the standard Major TOM NORM_s1 statistics.
• Each fragment is encoded at its native 10 m grid (image size 1068 × 1068 px ≈ a 10.7 km tile), so one embedding summarises the whole fragment — no tiling, no overlap.
• Input patch size: 120 m (≈ an 89 × 89 patch grid); the encoder pools this into a single global descriptor per fragment.

The embeddings build on the Major TOM Core-S1RTC dataset.

Citation

If you use this dataset, please cite UniverSat:

@article{perron2026universat,
  title   = {UniverSat: Resolution- and Modality-Agnostic Transformers for Earth Observation},
  author  = {Perron, Yohann and Astruc, Guillaume and Gonthier, Nicolas
             and Mallet, Clement and Landrieu, Loic},
  journal = {arXiv preprint arXiv:2606.23503},
  year    = {2026}
}