See the attached PDF.

This paper discusses different definitions of fractal dimensions for cloud fields. Its detailed exploration of what different fractal dimensions actually mean, and how they behave across different cloud fields is a worthwhile addition to the literature, and it should be published. The paper is clearly written, although sometimes a bit long and could serve from a small additional round of editing (e.g., I’m not sure that it is necessary to cite the Feynman lectures here). I have the following comments for the authors to consider, hopefully further improving the paper:

I am not convinced about the framing of this paper as an objective classification of cloud fields. For that to be the case, I would expect some kind of relevant difference to occur between clouds in different classes of D, beyond the relatively trivial relationship with other metrics. I could guess that there are certain distinct differences (e.g., certain values correspond roughly with cumulus, stratus, gravel, fish, etc – but in a more precise way), but the paper is not actually that case. I recommend either making that case (in which case the paper becomes likely quite different, for instance by applying De to Janssens’ cloud botany), or reframing the introduction and conclusions to simply focus on fractal properties.

Around line 260, and further on around line 280, I would like the authors to be more specific about their methods (e.g., “ignoring the portion the distribution that is dominated by large clouds that extend beyond the measurement domain”) , and the biases their choices may or may not introduce. For instance, in the method section it was stated that clouds over land were not included, but that could easily bias against larger clouds.Likewise, the exclusion on L 260 does something similar.

A few more minor comments:

L 63: Please give some citations

L78: What is the pixelsize/resolution of the satellite images?

L86: 72 images is not a whole lot to process into powerlaws. Is it possible to generate a larger dataset? What is the resulting number of clouds that are analyzed, and the margin of error in some of the later analysis as a result?