General comments:

This paper presents a comprehensive study of negative downward lightning flashes based on high-speed video camera recordings of negative cloud-to-ground lightning in several regions around the globe. This study presents solid statistics that help improve the current lightning protection standard (change from flash density to ground-strike point density). The subject is suitable for this journal.  Several comments follow. I recommend this paper be accepted after minor revisions.

Specific Comments:

1. Did the authors include upward lightning in South Africa dataset? If yes, I think those upward lightning contradicts your title (negative downward flashes). If not, please state so in the paper.
2. “Note that in Austria two flashes are observed whereby a new GSP is created by the tenth stroke in the flash, while the channel belonging to the previous GSP was used four and seven times, respectively.” It would be interesting to know the interstroke interval preceding the 10^th
3. Flash characteristic studies solely relying on high-speed cameras have limitations. I hope the authors could discuss those limitations and how those limitations could possibly influence the statistics presented. Two limitations that I can think of: (1) strokes creating a new termination could be missed by the camera (e.g., the stroke can occur at the back of cameras or simply out of view). (2) It is likely camera record length is not long enough to cover the entire flash.  I see that length for SA dataset is only 1s with manual trigger setup (not sure what’s the pre-trigger and post-trigger during manual trigger setup), maybe this partially explains why most SA flashes are single-stroke flash.  Simultaneous electric/magnetic field measurements/LLS data might help mitigate some of those limitations. They could be used to see if there are additional strokes in the vicinity but outside the field of view of camera or outside the duration of the camera records.
4. “It follows that the channel creating a GSP is re-used by a factor of 2.3” I think the word “re-used” is ambiguous. Sounds like the termination created by a previous stroke will be re-struck by 2.3 subsequent return strokes on average. Your statement “A ground contact point is struck 2.35 times on average” in Line 166 is more accurate.

Minor editorial suggestions:

1. Line 65: “Hence, the role of high-speed camera observations.” This Is not a complete sentence.
2. Line 70, enable us to determine
3. L127, 150 m

The authors provide a nice, concise summary of high-speed video observations of negative CG flashes from four locations around the world. My only pre-publication suggestions would be perhaps to consider augmenting the paper with some additional information that apparently should be in the data sets. This includes:

• lines 90-92: statistics on the occurrence of forked strokes, and on the occurrence of and duration of continuing currents (at locations other than the US site, as noted)
• lines 160-161 and other places where the data from South Africa are discussed: comparisons of upward-initiated flashes from the Sentech and Hillbrow towers in Johannesburg with observations from other towers, such as Gaisberg, Peissenberg, and Santis in the Alps, or the CN tower in Toronto. The comparison with the towers in the Alps might be particularly interesting insofar as so much of their lightning occurs in winter, if I am not mistaken, whereas the Johannesburg site is noted in this manuscript as having essentially no winter lightning.
• also related to lines 160-161: is the fact that the South African data set has a much larger percentage of single-stroke flashes than the others partly influenced by the population of upward-initiated flashes in that data set? It would be interesting to know, one way or the other

The only other comment is about references in lines 48-49 – at least some are outdated, and should be updated where possible; the Cummins et al. reference on the NLDN is especially old, and in fact, there’s a brand new paper in the March 2021 Journal of Atmospheric and Oceanic Tech. about the 2013++ NLDN.