HLS is not ARD
Address comment by Dr. Jordi Inglada (CESBIO).
Unless I am mistaken, you did not include a discussion about these issues in the paper. I feel that it is a pity, since I may not be the only reader puzzled by the results. I would have expected that reviewers of a serious journal as JSTARS ask you to make that explicit.
I was also wondering whether the metrics used, although usually accepted, are appropriate for this kind of analysis. Looking at the images, I don't have the impression that the original S2 data is closer to the L8 target than the HLS product.
It may be interesting to compute the metrics ignoring the outliers to see if you get less surprising results (i.e. S2 < HLS < GAN).