This service recreates 10 spectral bands of Sentinel-2 without clouds, shadows, image artifacts, and to some degree, errors in atmospheric correction. This is done by combining data from multiple satellites (Sentinel-1, Sentinel-2, Sentinel-3, and Landsat 8/9) with state-of-the-art deep learning. It produces the most up-to-date Sentinel-2 data available online. Below you can find more information about the data generation process, facts about the data, and links to API code examples.
This deep learning technique can utilize partly clouded imagery, radar data, microwave data, different spatial resolutions, and different temporal frequencies. By combining these data sources we can deliver a cloudless Sentinel-2 product we update daily, to catch any ground cover changes as fast as possible. It’s simply about getting the most use out of every satellite that’s already orbiting the earth in a practical and cost-effective way.
This approach can utilize SAR radar data from Sentinel-1 to update our cloudless Sentinel-2 imagery with surprising detail because a deeper context about the ground cover has been provided by the optical satellites (eg. Landsat, Sentinel-2) and the AI’s temporal understanding. This approach will, furthermore, map any changes from Landsat 8/9 onto the cloudless Sentinel-2 data while keeping to the Sentinel-2 data format.
There is no difference between today’s new cloudless data and the historical data available as all data has been produced in the same way. This is to ensure consistency throughout our service. However, this also means historical data is produced without any future insights and it’s only backward-looking. It will not extrapolate into what we know it will become because it will not see future data even if we do have it available.
Our cloudless imagery is not limited by the size of Sentinel-2 tiles. We can create wall-to-wall cloudless imagery that fits your arbitrary size needs, as we update our imagery on days without Sentinel-2. We do that to include data from other satellites that have different orbits and data acquisition than Sentinel-2. That results in easy usability as you can increase your area of interest in all directions knowing there will be data to support your analysis.
This is developed for land-based monitoring and all open water bodies are removed after the images have been created. We do not store SAR data for open seas and it’s not recommended to use any unremoved water imagery that you might find on the platform. You can find lakes consistently in our imagery, however, all water data is considerably lower quality than our land-based imagery. You can read more about the data generation process and our cloudless imagery in our Cloudless ‘Synthetic’ Sentinel-2 Data and Multiple Satellites, One Optical Image articles.
492 ± 33
559 ± 18
664 ± 15
704 ± 8
740 ± 7
782 ± 10
Near Infrared (NIR)
732 ± 53
864 ± 11
1613 ± 47
2200 ± 92
All ten spectral bands are in the same 10-meter resolution meaning 20-meter resolution spectral bands have been upsampled during the deep learning data generation process. All spectral bands are in the same order and have the same band names as the Sentinel-2 MSI Instrument (ie. true color is [B3,B2,B1]).
The cloudless data has been corrected with Sen2Cor for bottom-of-atmosphere disturbance beforehand. Furthermore, all of our Sentinel-2 data is harmonized to the pre-January 25. baseline. This is to allow for more efficient time-series analysis across multiple years.
If you don’t mind working with individual tiles, the easiest, fastest, and cheapest way to acquire ClearSky Vision data is by buying and using our own standard tile size. The imagery is projected in UTM and saved in GEOTIFF as INT16. You can read more about the data specifications below for our 10-band multi-spectral imagery:
-32,767 to 32,767
No Data Value