We are thrilled by the feedback, which we receive from the users of our services. We had no idea, in how many ways Sentinel-2 data can be used. But there is one prevailing message, which always pops up - quality of data. "Can you adjust the colour balance so that the images will be more natural and that scene borders will not be that obvious?" and "Why are there so many clouds?".
In the "old times" these issues were usually solved by manual fine-tuning and sometimes even photoshopping. These approaches do obviously not work on a global scale, even less so if new data are rolling in tens of millions pixels every second.
We had to look out of the box and come up with some innovative solutions:
- Atmospheric corrections
Atmospheric correction (sometime referred to as radiometric correction) is the process by which images are corrected to account for atmospheric contaminants, sensor viewing angle, and sun position. Correcting for atmospheric contamination is important because it converts pixel values to “actual” surface reflectance. In general this also leads to great contrast and depth in the values in the image, which is important when mosaicking or combining multiple images.
This was especially tricky, as we could not afford to run ESA-provided SenCor on each and every product (too costly, both from time and resources perspective). So we had to implement a simplified solution, which still does a pretty good job. Take a look at some examples.
Significant differences between the uncorrected image and the corrected image (reduced haze, more vibrant colors).
- Cloud replacement
Sentinel-2, with its orbit 786 km above the sea, flies just a bit above the clouds, usually only a couple of km above the land. And since wavelengths, to which S-2's sensors are adjusted to, do not penetrate the clouds, it makes it difficult for them not to become part of the data. But not many people show interest in them. We therefore did whatever possible to remove them. To fill the holes, we take the next best possible scene at that location. If that one also has clouds, we have no choice but to iterate as far in the past as possible, until we fill each and every pixel.
The process is not perfect yet as some small clouds remain undetected. Cloud shadow also messes things a bit. However, by adjusting the buffer around the clouds, it is usually able to generate practically cloud-less image.
Cloud-corrected image (Bahamas).
- Custom visualization/processing
Different use-cases require different earth observation products - some are happy with true colour and NDVI, others ask for EVI, SAVI, NDWI, etc. The experts ask for even more. There are limitless number of combinations of sensors to point out some important land features. Creating a new product would, in the old days, require re-processing of the whole archive, which would take weeks or even months. By introducing on-the-fly processing, we avoided that. However, there was still some work needed to change the software and deploy the service. Now, this is not needed any more. Users can, within WMS configurator, define their own specific Eo vizualizations, combining any band in any way. As soon as they save the configuration, the data appear in user's GIS client, connected to Sentinel Hub web services. Scripting language is simple yet powerful enough to satisfy most of the needs.
Second Modified Soil Adjusted Vegetation Index - Blue/Red visualization (Noord Brabant, the Netherlands).
Check out the showcase video: