There is now an interesting super-resolution code for S2:
The authors (not me) have trained a CNN with L1C data to super-resolve the 20 and 60m bands to 10m. Myself and colleagues are very interested, and we have implemented the GitHub code on our system. But the fact that the neural net has been trained with L1C data means that we still need to process an att corr for the super-resolved data. I use Sen2Cor from the command line but I don’t know it’s inner workings. I am therefore looking for a method to perform atmospheric correction of the super-resolved L1C data. What does this forum think of the options below:
1- Produce a duplicate .SAFE folder where the image data has 10m resolution. Leave the XML data intact and just run Sen2Cor as normal and try to fool it into processing 10m data as it would 20 and 60.
2- Produce a duplicate .SAFE folder but change the image format to geoTIFF to avoid the proprietary nature of Jp2 drivers. Change the image format in the XML file to tif. Run Sen2Cor, again trying to fool it.
3- Retrain the CNN with L2A data because the options above will simply not work, Sen2Cor needs the actual L1C data un-altered to run properly?