How does collocation tool in SNAP work?

Hello everybody,

I’m doing a colllocation of Landsat 8 and Sentinel 2 images, which is needed later on for a temporal analysis. I did this using collocation tool in SNAP and want to see what is the difference in the algorithm compared to another algorithm I’m using (AROP package). The only information I found was through the github source code which I’m not able to understand currently, as well as this link from the current forum where @marpet elaborates on it. However, the latter case seems vague to me as it is not explained precisely how the collocation algorithm works.
So, I was wondering, is there any reference where I can read about the algorithm or someone that can explain further?

Kind Regards,

Have you seen this description in the Help?

It is not easy to tell in detail how the collocation works, because it depends on user settings and the products which are used.
Simply said, for a pixel in the master product the corresponding pixel is looked up (by geolocation) in the slave product. If found, the data might be resampled, depending on the user setting and then transferred into the target product.

Thank you @marpet , I’ve already read the Help’s description, although I cannot completely understand as I do in AROP package’s reference for example.
Regarding your answer, is it possible to name a few potential user settings? Can I change those settings as an end-user (apart from e.g. resampling technique)?
Also, under what criteria does the algorithm decides which pixel of the master corresponds to a given pixel of the slave? From the Help, I cannot tell how the brightness values get into the game. Am I missing anything?

The criteria to find the corresponding pixel is always latitude and longitude.
One of the settings is resampling, another setting which is defined by the slave product is the kind of GeoCoding. How to locate a pixel by lat/lon is different for pixel-based GeoCoding or GeoCoding using a CRS.

The brightness values do not influence the collocation. They are just treated as values which need to be transferred. But due to the resamplling the values can change a bit.

@marpet Thank you very much for your response Marco. This information is not 100% what I needed but is fairly adequate to know that the technique has nothing to do with feature- or area-based coregistration.