Sentinel-2 Deep Resolution (S2DR3, S2DR4) General Knowledge
Last update: -
1. Temporal availability
S2DR can super-resolve and cloud-mitigate Sentinel-2 scenes from 1 January 2018 onward with the same cadence (5-day revisit). The full scene catalog is available on Copernicus Hub website.
2. Spatial resolution
The output GSD is 1m/px in UTM projection. The actual effective resolution may vary depending on location and type of terrain. Textures are reconstructed down to 1m detail, but individual objects are reconstructed on the scale of 2-3m.
3. Min and Max Area Of Interest (AOI)
AOI definition is based on a bounding box or a polygon in GeoJSON. The minimum technical AOI size is 0.4 km2 (40 ha), as quality of results over smaller areas would be compromised. The maximum technical AOI size is 150 km2 (undergoing improvement).
4. Output Products of S2DR
S2DR output will contain 4 products:
- S2L2Ax10_T[MGRS]-[UID]-[DATE]_MS.tif – 10-band multi-spectal image
- S2L2Ax10_T[MGRS]-[UID]-[DATE]_TCI.tif – True colour RGB image
- S2L2Ax10_T[MGRS]-[UID]-[DATE]_NDVI.tif – Pseudo-colour NDVI image
- S2L2Ax10_T[MGRS]-[UID]-[DATE]_IRP.tif – Ifra-red preudo-colour image
5. Spectral Band Information
| Sentinel-2 L2A | S2DR3 / S2DR4 | |||
|---|---|---|---|---|
| Band | Description | Central Wavelength (nm) | Bandwidth (nm) | Band Index |
| B1 | Coastal Aerosol | 443 | 20 | – |
| B2 | Blue | 490 | 65 | 1 |
| B3 | Green | 560 | 35 | 2 |
| B4 | Red | 665 | 30 | 3 |
| B5 | Vegetation Red Edge 1 | 705 | 15 | 4 |
| B6 | Vegetation Red Edge 2 | 740 | 15 | 5 |
| B7 | Vegetation Red Edge 3 | 783 | 20 | 6 |
| B8 | NIR | 842 | 115 | 7 |
| B8A | Narrow NIR | 865 | 20 | 8 |
| B9 | Water Vapour | 945 | 20 | – |
| B10 | SWIR – Cirrus | 1375 | 30 | – |
| B11 | SWIR | 1610 | 90 | 9 |
| B12 | SWIR | 2190 | 180 | 10 |
Bands marked with “–” are not generated in S2DR3/4 outputs. S2DR3/4 produces 10 super-resolved spectral bands corresponding to Sentinel-2 bands B2–B8A, B11 and B12.
6. System Requirements — Python SDK Local Installation
Operating System
- Linux (Ubuntu 18.04 or later recommended)
- Python 3.12
Hardware
CPU
- Standard x86_64 processor
GPU (required)
- NVIDIA GPU with CUDA support
- Minimum 32 GB VRAM
- Tested on: T4, P100, V100, A100
RAM
- Minimum 16 GB
- 32 GB recommended
Dependencies
- NVIDIA CUDA drivers — developer.nvidia.com/cuda-downloads
- GDAL (gdal-bin)
- pip