This is a well written paper describing a 3m water extent dataset that can be used for validating 10-20 m resolution water extent data sets. Unfortunately, rather than just focusing on this validation dataset, the authors also introduce a new Sentinel-1 algorithm called the Equal Percent Solution (EPS) for surface water extent delineation. My main concerns are:

• The water extent validation dataset derived from 3m PlantScope images covers 90 sites worldwide. It is certainly a valuable dataset, but still limited when applied for validating global water body data sets. For example, there are very few sites in arid and semi-arid environments where water body mapping with both optical and radar data can be very problematic.
• Validation datasets do not just require sites with water bodies. Also areas with no water cover should be included to arrive at representative accuracy statistics.
• The EPS method is essentially a threshold-based classifier for images characterized by a bi-modal distribution. This method - and its limitation – are well known. A major problem is that it does not work if there is no bi-modal distribution, i.e. it fails when in many instances, e.g., when the water extent is small or there is no water. Also, it does not work if there are many water-look-alike area.

My recommendations to improve this work are: (i) extent the validation data set to cover more diverse environmental conditions, i.e. also areas with no water bodies, (ii) rather than validating their own Sentinel-1 dataset, the authors can demonstrate the value of their validation dataset by using it to validate several published water body dataset.

The EPS method is in my view not publishable within the context of this paper. The authors should consider a dedicated paper where they work out the innovation of their algorithm compared to published algorithms in much more detail. Furthermore, the method needs much more testing and critical examination. From my knowledge of the Sentinel-1 backscatter data, I expect it to fail in many situations, e.g., when trying to apply it to area with no water or trying to map water in arid environments.

MINOR COMMENTS

Lines 46ff: Also the Copernicus Emergency Management Services offers global near-real-time flood maps based on Sentinel-1

Line 53: How does the sentence starting with Wieland et al. (2023) logically connect with the previous sentence?

Line 59: “can” instead of “could”

Line 76: Delete “of the true distribution”

Line 81: Delete “urban regions have a higher density of built-up environments" (that is obvious, isn’t it?)

Line 193: Do the authors believe that muddiness is the reason for the different Senintel-1 signals? If yes, then much more evidence is necessary to support this claim.

3m hand-labeled water product should be useful as most existing ones are >10m. However, I think authors have not fully justified their motivation in creating a new dataset, and how this intercompared with existing ones. For example, authors mentioned they used Wieland et al. (2023) has a 1m product. How does your product compare with theirs? Why bother having a new coarser resolution ones? Authors mentioned “multiple independent evaluations are needed”, but does your product adds another layer of subjectivity? Or do you have other justifications to strengthen this part?

How is the temporal information in the Planet Images influence the representativeness of the labeling results? Can the authors expand on this, provide cautions to users and also maybe write down such information with the data to better inform users?

Authors claim they used a novel EPS method. However, it seems the details of the EPS method were not given. How is it accomplished? Why is it novel?

Given the above consideration, I’d suggest major revisions.