line 163: “statistically robust”: I would interpret that to mean “can you randomly subset the towers and end up with the same general result?” Based on the spatial distribution in figure 6, it looks like the answer is “yes”. At the point of discussion of figure 6, it might be worth making an explicit comment in reference to those words from line 163 

line 190: the question of an adaptive value of eps is a good one, but you could even further support the choice of 0.1 km by referring to figure 5a, where the ground-truth data from the 25-year history of Gaisberg tower validation and the DBSCAN analysis agree very nicely. That also adds more support to the question of "statistically robust". 

line 193/ figure 3: are the cluster sizes based on the entire period 2012-2025? I assume so, but it would probably be worth mentioning explicitly. 

figure 7 / lines 276-294: I think that the "count" on the vertical axis of figure 7 is the number of structures, not the number of lightning events, but this probably should be made clear. Following from the discussion about the near-zero median and the definition of the 50% ellipse, the fact that the median offset is only -10 m says that almost exactly 50% of the towers have a larger median SMA than the DBSCAN results and 50% have smaller median SMA. That further supports the usefulness of the ellipse as a location accuracy metric where/when direct validation is not available. A potential "future work" analysis based on this paper, if one really wanted to dig deep into verifying the representativeness of the error ellipses, could be to look at the individual events that are included in the DBSCAN clusters at a few of the towers, comparing their ellipses to their position offsets from the tower. 

It appears there are references missing in lines 75 through 79, or at least a note to reference the following paragraph where the references exist. 

Is there logic as to why the threshold for structure height was 150 m? 

Regarding the sentence in line 195, how many clusters are there total? Are event clusters comprised of strokes? 

Some acronym definitions are repeated, please check for first appearance of acronyms appropriately. 

Is the EUCLID data filtered to only be reported CG strokes? Or are IC discharges included? What method is used to exclude ICs if ICs are excluded? 

For structures located on the edge of the sensor network, should there be a separate consideration due to the worse DE as represented by the larger median location accuracy in Figure 6? Additionally, it is odd that the latter years exhibit a worse median location accuracy for some structure locations (between 450m and 500m). Are these outliers towers that are on the edge of the network?

A few case studies using known structures, such as from video camera observations of lightning or documented structure information, may provide further validation to the method.