This study analyses data from the network of pyranometers with a focus on cloud radiation enhancement. The network was installed at the ARM-SGP site, well-known for its cloud measuring capacities, resulting in a unique data set for detailed cloud enhancement analyses. Along with statistics, the authors investigate effects of macro- and microphysical cloud parameters on enhancement. I find this study interesting; below are my comments.

General comments:

To me, the study lacks quantitative information about the effects. Many violin plots are provided visualizing orange vs blue data, but how close or different those are?  Are the differences orange-blue statistically significant? For example, L 291 mentions ‘significant’: in which sense significant? Also comparison with other studies: when similarities are mentioned, is it possible to say if they are quantitatively similar?  How big effects can be expected?

I would appreciate a better explanation of the analyses of transition signatures with regards to cloud shadow mask/with regards to flow (Sec. 4.1 vs Sec. 4.2). 

Specific comments:

L 20-21: Plant photosynthesis and feedbacks belong to the previous sentence about land-atmosphere interactions, and photovoltaic power generation is more of an engineering application.

L 46: I would remove ‘Building on this network’, it does not anything to the sentence if the same data set was used.

L 88: ‘high concentrations of aerosols from the atmospheric boundary layer’: change ‘from’ to ‘in’

L 95-96: I do not understand why ‘however’ is used, and why not just report cloud shadow speeds, their direction and cloud cover in one sentence.

L 105-106: these instruments, data from which is not used in this study, why they need to be mentioned? instead you could give a bit more information including producers and precision of the instruments that were used for the purposes of this study.

Caption to Fig. 4 ’shadow chord lengths represent arbitrary cross-sections of the shadow shape due to the stationary nature of the measurements’ – this point is not completely clear to me.

L 195: ‘-60m towards shadow and 600m towards…’ It is confusing when speaking about time series in the previous sentence, one uses length as limits. Can you explain better, how you come to these limits?

Fig 5, y-axis: ‘normalized irradiance’ – is not it transmittance? also T is used. I think, Lennard-Jones in the caption should be changed to Buckingham (quite innovative use of potential functions, I found it amusing). I also suggest to clarify definition of epsilon; current is ɛ = Tpeak/clearsky (%) but from the figure it looks like CE in % should be simply (Tpeak-1)*100?

L 225: could authors show an example of fitting in Fig. 5?

Intro to Section 4: side-escape, forward-escape and albedo-enhancement mechanisms: would be good to have a short discussion how those mechanisms could be separated from each other or how they manifest themselves in the current framework. It would also benefit a reader, if the authors could name there all the cloud variables to be considered.

Caption to Fig. 6: ‘distance to the respective cloud (orange) or shadow (blue) mask edge’. Only shadow is mentioned in x-axis label. Also probably I missed it, but how were the masks (distance) collocated with GHI analyses (time series)?

Section 4.1 title does not really say anything; cloud vs shadow mask?

L 271-272: reference is needed about 200 m distance

L 287: ‘is more pronounced on the sunlit side of the cloud, also depending on the cloud depth’. I would appreciate illustration of sunlit vs dark side somewhere in the figures. Then about this effect: how strong effect was found in this study?

L303: ’ shadow irradiance near the sunlit side showed a slight increase, highlighting the notable influence of the albedo-enhancement’ - I do not see this from the figure. What also puzzles me is that optically thickest clouds (b) show the same measured-to clear sky GHI ratio as all cases (a), about 40%. why is that? I would expect thickest clouds should have lower transmittance.

L 337: ‘radiation enhancement, approximately 2% above the existing enhancement level near the shadow’s edge’ – enhancement over enhancement level is quite confusing.

About x_e and epsilon: I think there is not much information about how much these vary, but results of correlation with different variables. It would be interesting to have these linear correlations visualized in the appendix. 

Fig. 11: when feature importance is discussed, what method was used? Is it machine learning based? Which model? It would be good to add this information in Methods.

Fig. 11: It is interesting that epsilon and x_e show correlations of opposite sign with the same variable (e.g., solar zenith angle, droplet effective radius). Since both epsilon and x_e characterize the strength of CE, is it possible to make some kind of estimate of 'best' combinations? Are there optimal combinations of variables?  

This paper addresses an interesting topic, which is the radiative effects produced at cloud edges. More specifically, how solar radiation changes rapidly (in time and space) in the transition of a cloud-free measurement to a cloud-shadow measurement. This topic is important and of current interest, as the best description of cloud effects, including those within scattered clouds regions, is necessary for many reasons such as the consequences on the management of solar energy plants and the impact on meteorological and climate modelling.

The observational data set is robust, and quite particular. Probably, there are no other campaigns with so many instruments for solar radiation measurements with the spatial density of this present case. Methods are also well designed and applied, in my opinion.

So I don’t have so much to add, as I concur with many comments included in the first review; in particular, some results, as provided in figures 6-9 could be explained a little more clearly and/or more quantitatively. Anyway, I think this is quite a pioneering work that can be a first step for further analysis of the same data set, which could be devoted to better explain causal relationships between the already identified features and the measured radiative effects.

Technical details:

Figure 1. Panel (b) shows so many things. I would suggest using this panel to show only the position of PyrNet stations, and reproduce a similar figure as an additional panel of Figure 2, which would contain the transmittance measurements,  the cloud mask, the cloud shadow and so… This would make much easier the comparison of the (suggested) two panels in Figure 2.

Projected valid area and focus area seem to be separated by 0.005 degrees, not 0.01 as you say in the caption of Fig. 2 and also in line 146. I might be missing something, but this is what it seems to be according with dashed lines in Fig. 2.

Figure 3 caption. Authors say that the blue line is “the mean of all observations” which is not clear to me. “All” means observations in 1h time interval? Or a running mean?

The definition of epsilon (ε) is not clear to me, and actually, is provided only in Figure 5 and its caption. If T is already a kind of normalized irradiance, why do you need to divide it again by clearsky? Or is ε only the same as T but in %?

In Eq (4) and Eq (5) you use the expression “exp(1)”.  I understand that this results in the Euler number (e), so why don’t you write simply “e”?