|In essence the authors have taken the code and design principles of (Thielicke, 2014) and applied this to satellite imagery of glaciers. Therefore, implementations like "intensity capping" and "CLAHE" are present. The work of Thielicke (named "The flapping flight of birds: Analysis and application") has been tailored towards PIV data, and in his thesis work the design issues have been tested against others, so a justification can be made to a specific implementation, as presented in (Thielicke & Stamhuis, 2014). Fore-mostly, I think this straightforward copy needs to be emphasized more clearly in the manuscript, in this sense the toolbox is a similar implementation as SenDit, though (Nagy et al.2019) give clear credit to the feature tracking pipeline they have wrapped their toolbox around. This is in stark contrast to the presented manuscript.|
Secondly, the response by the authors is brief and no effort is put in backing-up their argument. The majority of answers include phrases like, "we find little difference", "we found no clear improvement", but results or information is lacking. Thus it is difficult to get convinced, especially since the points I have raised are backed up by references. One can not expect me to run the test for you, in my opinion the burden of proof needs to be at the side of the authors.
The authors have doubts themselves if the Cryosphere is the correct journal for this manuscript. I have the same but stronger doubts. There is no clear scientific finding present in the paper, nor is a new method developed. We also agree upon the other point: there is a lot of overlap between the intended readers and the audience of The Cryosphere, but I think exposure for the GIV MATLAB toolbox can also be achieved through Cryolist. For example, after the release of IMGRAFT announcements have been posted there and it is now a common toolbox in teaching and research. Hence the Cryosphere might not be the correct journal for this work and better journals exist.
- The authors claim to see no difference between 10 and 8 bit data, contradicting two other studies. They say the changes are "little", their assessment not based upon the individual building block. So, if isolation is done these pattern will likely emerge.
- "The visible bands are correlated, which is why we sum them into a single band", What is the rationale, I can't understand why this should be evident? You're now only turning my question into your answer....?
- There is a vague comment about the use of different bands, please see (Redpath et al.) for more detailed analysis.
- NCC vs.Freq. this seems to contradict to (Heid & Kaab, 2012), at least they hypothesise this to be the case.
- nlfilter, you seem to use small image sniplets and not whole Sentinel-2 or Landsat scenes. When this is the case, such functions do much better.
- It is still unclear what the authors envision with their toolbox. Their open source argument is for attracting others to improve upon their work. Though on contrary, they also "do not believe code commenting is an important factor". Why is this contradiction there?
- The argument that "Other good feature-tracking toolboxes" "can be challenging to use" is subjective, and not an argument why not improve upon them. Secondly, many of the listed toolboxes have a GUI and are used in courses.
- In the previous implementation the geo-referencing consisted of a simple multiplication of the pixelsize by 10. In the newer version the authors have included the possibility of reading geotiff-files, but it seems the same irregularities are also included in the code. For example, some parameters are hard coded: 'xtick',10 . When other bands of Sentinel-2 are used, this will break down. Similarly, when Landsat8 data is used. Secondly, the authors build on top of there strange angular-metric transformations, for example, by integrating functions like "coordtom" (based upon matlabcentral/fileexchange/38812-latlon-distance). This ratio might work for a single line, but for a high (Ant)arctic scene this ratio changes extensively over a scene. Though there is a mapping toolbox present in Matlab, why is this not used?
- Although the authors do not respond to my suggestion, I still believe a workflow plot might benefit this work, a screenshot of the guy is less informative.
- Matlab should be in the title, as IDL, Python and Julia are (decreasingly or increasingly more) populair.
Heid & Kaab, 2012. 10.1016/j.rse.2011.11.024
Nagy et al. 2019. 10.3390/rs11101151
Thielicke, 2014. hdl.handle.net/11370/31931f33-4aa0-4280-892e-93699af0e9b6
Thielicke & Stamhuis 2014. 10.5334/jors.bl