10,000 unique geographic locations comprising 10,000 km² of paired high-resolution SPOT 6/7 and multi-temporal Sentinel-2 satellite imagery. The collection features 450,000 image patches sampled across diverse biomes and socio-economic contexts globally to facilitate Earth observation research.
Use Cases
- Train super-resolution neural networks to reconstruct high-frequency details in Sentinel-2 imagery using SPOT 6/7 as the target
- Evaluate model generalization across different geographic regions using the stratified location metadata
- Perform multi-temporal analysis and cloud mitigation using the time-series of Sentinel-2 revisits associated with each static high-resolution patch
Strengths
- 10,000 km² of 1.5m/pixel pan-sharpened SPOT 6/7 imagery used as ground truth
- Multi-temporal Sentinel-2 L1C and L2A revisits provided for every high-resolution site
- Stratified sampling based on World Settlement Footprint and ESA WorldCover classes
- 450,000 image patches provided in 160x160 (SPOT) and 64x64 (Sentinel-2) pixel dimensions