Interesting paper with an approach to surface water detection globally.  In general, the method works and generates results that are compelling enough to consider.  However there are many significant holes in the logic that were not tested (or not proven) by the authors which could have significant impacts on their results and conclusions.  Overall, much more detail is needed in the descriptions of how conditions (I noted several specific things in my comments below) were handled and tested for validity.  While I accept the premise that frequent MODIS observations are an advantage compared to frequency products using Landsat, which was stated by the authors in the introduction, it is still necessary to compare the frequency results from GLOBMAP to one or more of the frequency maps from Landsat (Pekel or Pickens at the very least).  The comparisons that were done were with other MODIS derived products.  This is ok for a first look but if you are trying to claim that you have a better approach than the Landsat products you must test this and show the results so the reader can decide for themselves.  For this paper to be published in context these evaluations must be performed and reported.  Beyond that it is important to clarify for the reader how the following things were handled so that the reader can trust the results.  How did you delineate the oceans?  Where did you cut off rivers where they meet the oceans?  How did you handle extensive burned areas globally which would effect your low NIR values?

Line        Comment

50           “Surface water was also mapping” needs revision for English grammar

99           determinate should be determine

101         why did you not use the finer resolution GMTED which was designed for use with MODIS data?

116         the sentence starting with “The cloud, ice…” appears to end abruptly or otherwise be an incomplete sentence

136         all of your validation sites are in the tropics, this is not a best practice.  For a global product you need to have validation from northern latitudes as well as mid latitudes to assess performance everywhere

173         If you use MOD09A1 you have a total of 46 possible observations in a year.  In most cases at least half (probably more) are not usable due to clouds or other data problems.  Using the “six lowest NIR” values, could be that you only have six total observations for a pixel.  This is a questionable method for a global product.

194         less than 15 percent or less than 15 count?  What do you do in the frequent case where Nland is < 15?  In northern latitudes you won’t get that many snow free observations, the method seems to ignore the typical case of snow surrounding ice covered lakes.

230         does your definition of intermittent include the fact that high latitude lakes are frozen for a large part of the year?  Much more clarity is needed on your definitions.

308         this statement confirms my earlier comments.  Many of your assumptions about the availability of clear sky observations are invalid for many places in the world.  Unless you provide a companion product describing the per pixel reliability (based on the number of observations available) users are likely to draw incorrect conclusions in many cases.

352         should be Carroll not Carrell

378         for these evaluations to be understood it is essential to know how many clear observations there were in each year.  How can the reader know that the variation you are reporting is not simply due to differences in the number of observations for a given year?

395         largest variation by total area or by percent change?  Total area would limit this to only very large lakes…

The dataset provides a MODIS-derived annual surface water frequency dataset, which can help analyze the dynamics of surface water. The manuscript is well structured and written! My major comments are:

I agree with the authors that the daily MODIS observations have value in capturing the variations of surface water; however, its limitation is also obvious. The 500-m resolution is too coarse for capturing the abundant small water bodies as well as the subtle changes of surface water. Moreover, the Sinusoidal projection of MODIS caused a considerable distortion in high latitudes, worsening this omission, particularly in North America and Eastern Russia, where a large portion of the global small water bodies are located. It seems that the authors have been mainly focused on China and a few low-mid latitudes, but did not assess the performance in high latitudes, which seems to require more attention during validation.

The authors reported the areas of global inland surface water, including permanent and maximum areas; however, I think the numbers could be biased by failing to capture the small water bodies as commented in the above paragraph. As many much finer surface water datasets have already been produced, I would suggest the authors clarify the conditions of these reported areas, such as water bodies larger than a certain size; otherwise, the areas would not be valid.

I am not convinced why the authors did not compare the results to the global water dataset produced by Pekel et al. and the GLAD (Pickens et al., 2020), which all provide permanent and seasonable water cover that can be comparable to this dataset.

Specific comments:

I would suggest removing “for change analysis of inland water bodies” in the title. 

Line 172, why six observations? 

Line 179, does the slope criteria also remove water in a sloppy area that is outside of shadow? Also, did the variation of solar angles along latitudes and seasons considered in estimating shadows?

Line 200, please clarify what resample method was adopted.

Line 427-448, I am not fully agreeing with the novelty of the method as mentioned here. The method still identifies water cover as explained in the methodology, so the statement of the advancement here does not seem to be a valid point. Also, the method seems to be a very simple one without considering calculating water index or machine learning-based models. I am honestly surprised that it was robust enough for producing a reasonable global result.

Figure 8 is hard to interpret because most of the pixels showing positive trends also show negative trends. I think that the authors need to come up with a better way of presenting the results.