Sentinel-2 Super-Resolution (all bands at 10m): Snap plugin now available!

Can you please show a screenshot?
SNAP makes a difference between the physical data type and the one which is interpreted by the GUI based on the metadata.

Respected Braun screenshots of steps which I did during super-resolution of bands?

I meant of how you determined the changing data type

Without knowing the implementation of the resampling tool, I think I can sheed some light on it.

Original S2 data is stored as uint16 along with a scaling factor which scales it to the geophysical float32 value.
When the super-resolution does the resampling, it works with these geo-physical float32 values and stores them int the end. A scaling back to unit16 is not applied.

If you want to go back to unit16, you can devide the values by 10000, the so-called QUANTIFICATION_VALUE. You find it in the MTD_MSIL1C.xml file.
Then you can store the result.
You can use the Convert Data operator for this.

I determined changing data type by extracting the pixel values.

Respected marpet thank you i will do as you suggested.

Ah wait!
What I said it not correct. Sorry. The values will be wrong afterwards.

You can use the Band Maths operator via gpt.

The expression would be e.g. B2 /10000
and you set the data type uint16 and as scaling factor 10000.
This way it should work.

Respected marpet thank you for clarifying the answer. I have another question super-resolution plugin gave a raster with 12 bands as an output. Is there a possibility to resample just a few bands not all and get each resample band separately as an output, not a single raster?

That is not an option of the super-resolution plugin.
But you can add a band subset and export only the bands you need.

I’ve once provided an example for doing this.

Thankyou respected marpet.

Please could someone shed some light on the progress of the sen2res Snap plugin.

I rather foolhardily tried to run it on an entire Sentinel image. The bottom left progress dialogue switches between different bands /484 tiles. I presumed this meant I had 9/14 bands to super resolve 484 tiles. However, it has been running happily 24/7 for 6 days now and new bands seem to be appearing. It has certainly started more than three 60 m -> 20 m band resolutions of 484 tiles, so I am now confused as to what a single tile is, and therefore how long this will process. Can anyone advise?

Secondly, as it looks increasingly like I will have to interrupt the process. I would ideally like to save the processing that has been completed this week even if it is partial. Is it simply a case of saving the file from Snap mid processing?

Hi GWH, may I ask did you figure out this? I have exactly the same situation as you. The Sen2Res has been running 5 days for an entire sentinel image and I have no idea how long this processing will take. How long it took in your case?

Do you have any suggestion how to save the partial completed results, for example, I already get SRB1 and SRB5, but I did not find the file in the output directory. Thank you.

Dear all,

I am using the Sentinel 2 super-resolution plugin to downscale Red Edge band 6 to 10 m from Level 2A products. The result band is in digital level (dl) according to the Pixel Info tool. Why it is not dimensionless, as the L2A is corrected at BOA so it has reflectance values? Nevertheless, pixel values make senses (0 to 1). I have read in your article that it preserves the reflectance.

Thank you very much,

The unit ‘dl’ means dimensionless.
See page 32 of S2 MPC Level 2A Input Output Data Definition (esa.int)

1 Like

I have a simple question Sir… I am trying to make super-resolution Sentinel-2 imagery. And by using your Sen2res, it was possible to create all bands 10m resolution images. And then, I made RGB and NIR bands to 5m resolution by using Deep Neural Network. Then, is it possible to use these 5m bands again to create other 10m bands to 5m enhanced resolution? Many thanks!!

Oh, yes, and you can repeat this process to any resolution you like… but there is no magic. The original 20m → 10m algorithm only assumes that objects that are physically present in the scenary have different reflectance in different bands. Hence, it tries to preserve the geometrical details (extracted from all 10m bands), while preserving the reflectance of the 20m bands (if you average the result super-resolved image, you should get back your original 20m band). There are no 5m details in the original picture. All that your neural network is doing is a kind of interpolation from similar images in a gigantic database. The 5m bands you get are a mixture of original scene information, together with many random pieces taken from unspecified images that have more or less similar content. This can be visually very good, and might even be useful for statistical detection applications for example. But these are completely irrelevant for many other applications. So, yes, you could try to super-resolve that again, and again, down to any resolution you want. And, progressively, at each iteration, incorporate more and more spurious details from random images into your scenary. And all the while, with nice looking pictures at each step. A bit like, if you erase a tourist on the grass and fill that missing portion of a picture with an in-painting software, you could get many nice looking photos of grass as a result, some of them with flowers, some of them with other details like small white stones, etc, all these completely irrelevant to what was actually behind the tourist in the original picture.

2 Likes

Hi there, I am attempting to download the Super-Resolution SNAP plugin from the links here http://nicolas.brodu.net/common/recherche/sen2res/sen2res-1.0.nbm and here http://step.esa.int/thirdparties/sen2res/1.0/sen2res-1.0.nbm but it seems that the links are broken. Is there a chance that they are able to be fixed? Thanks in advance

Both links work for me. Maybe you try again.

Also note that the plugin was made for SNAP 5. There is no guarantee that it is still working.
Sen2Res – STEP (esa.int)

Trying the Microsoft Edge browser on my PC was able to make the link work, however Google Chrome on PC and Brave on mobile both refused to download.

1 Like

use edge browser instead of chrome. Worked for me