The authors combine LOFAR radio imaging of lightning channels with weather radar data to assess the possible meteorological causes of the small scale “sparkle” discharges observed near the top of several mixed-mode multicellular/supercellular thunderstorms in the Netherlands. They make admirable use of the limited operational weather radar resolution at upper altitudes to say as much as they can about the storm context for sparkles.

Overall, these novel observations are a welcome addition to the handful of studies that have examined these small lightning discharges, especially because LOFAR provides substantial additional sensitivity to these small discharges. Furthermore, the relatively high latitude of the storm observations nevertheless contains some supercellular storm structures, making this study novel in comparison to the US Great Plains and Florida storms that have more frequently been studied, and shows the value of studying lightning in a variety of environments globally.

Below, I have asked for clarification in a few major areas, but none require a significant revision to the basic framing or conclusions of the study.

Major concerns:

1. Section 2.2 describes the sparkle classification algorithm, which seems reasonable, and per the LOFAR images in the manuscript and appendix seems to have been conservative in not classifying every sparse VHF source as a sparkle. However, to better understand what LOFAR is detecting, I wanted to ask about the nature of these sparse sources, both sparkle and non-sparkle. The clarifications fall into three categories.

(1) radial location errors along an azimuth (in LMA data, these usually point back to the network centroid and are more common at longer ranges): green sources in Fig. C1(c) and C2(c). These seem strange since the other nearby flashes don’t show this effect.

(2) potentially mislocated sources, such as the two sources west of the main orange cluster in the center-right of Fig. C2c; perhaps some of the sparkles and other single sources are similarly mislocated sources throughout the dataset, and in reality occurred along one of the better-imaged main channels?

(3) In Fig. C2, a semi-circular area of sparkles and a few other sparse sources, mostly in the absence of other extensive lightning, between 6.7 and 6.8 east and 53.2. and 53.3 north. Is this indicative of an electrified updraft region where otherwise large discharges were unluckily not detected in the short (<2 s) LOFAR imaging window?

Could the authors confirm my interpretation or speculate as to other causes for these sparse sources?

2. Line 62: Another relevant point of comparison is to Bruning et al. (2010), which isolated lightning associated with overshooting surges, identified charge where possible, and placed it in the context of radar data. For example, see the discussion on p. 3747, third paragraph in section 4b. Admittedly, they did not go as far as your other references in quantifying the radar statistics of these overshooting lightning surges. Also, the correlation of large upper-level reflectivity values to bursts of small VHF sources is known as far back as Lhermitte + Krehbiel (1979). Ushio et al. (2003) also made some early observations of the overshooting top lightning.

Bruning, E. C., W. D. Rust, D. R. MacGorman, M. I. Biggerstaff, and T. J. Schuur, 2010: Formation of charge structures in a supercell. Mon. Wea. Rev., 138 (10), 3740–3761, doi: 10.1175/2010MWR3160.1.

Lhermitte, R., and P. R. Krehbiel, 1979: Doppler radar and radio observations of thunderstorms. IEEE Trans. Geosci. Electronics, GE-17 (4), 162–171

Ushio, T., S. J. Heckman, H. J. Christian, and Z.-I. Kawasaki, 2003: Vertical development of lightning activity observed by the LDAR system: lightning bubbles. J. Appl. Meteor., 42, 165–174.

3. Line 45-47: The authors seem critical of the possibility that heterogeneous charging and turbulence in the updraft core (independently of screening layer mixing) might be responsible for the small-scale structure of charge, and they use evidence from Calhoun et al. (2014) to support their point. However, the charge pockets illustrated in Calhoun et al 2014 are fundamentally limited in size by the grid spacing of the model used, which was dx=1 km in the horizontal. As Bryan et al. (2003) show, the resolved structures in cloud models are filtered below 6*dx, i.e., 6 km in Calhoun et al. With dx=125 m, Brothers et al. (2018, Fig. 4) show charge structures that are much closer to the scale of the sparkles illustrated herein. That figure also showed that the charge in and near the overshooting top tends to remain smoother than deep in the updraft. Later, in the discussion section, leading up to line 382-3, the alternatives are discussed more completely, and in a way that seems fair to the observations and all of the possibilities, so my recommendation here is to adjust the wording to be more neutral to the various hypotheses.

3a. Before moving on from this topic: another factor that is less favorable toward a screening layer mechanism is that the depth of the sparkles extends several km below cloud top.

Brothers, M. D., E. C. Bruning, and E. R. Mansell, 2018: Investigating the relative contributions of charge deposition and turbulence in organizing charge within a thunderstorm. J. Atmos. Sci., 75 (9), 3265–3284, doi: 10.1175/JAS-D-18-0007.1.

Bryan, G. H., J. C. Wyngaard, and J. M. Fritsch, 2003: Resolution requirements for the simulation of deep moist convection. Monthly Weather Review, 131 (10), 2394–2416.

4. An overview of the basic meteorological parameters used in severe thunderstorm forecasting would increase the value of this study for those interested in the environments that make lightning in the Netherlands, and would also help in understanding the storm modes observed. A thermodynamic sounding (skew-T) including a hodograph, plus calculation of the 0-1, 0-3, and 0-6 km vertical wind shear and CAPE would be especially helpful; the MetPy or SharpPy packages are among the readily available tools that can do this. A quick examination (from the University of Wyoming sounding archive) of 12 UTC data from Meppen, DE (very near the LOFAR site) showed there was probably at least moderate CAPE (especially if the sounding profile were modified to account for afternoon warming near the surface) and a fair amount of low-level directional wind shear, perhaps consistent with a supercell environment. An ERA5 reanalysis sounding from 18 UTC at the same site could also be useful. The hodograph and shear measures would be useful in diagnosing whether bowing linear or supercellular modes were more favorable.

5. Line 185 and following, convective system B and C: from the presentation in Fig. 5 (Fig. 3 is too zoomed out for me to say more) these do not look to me like bow echo structures, since those tend to be symmetric about their axis of propagation, and often feature counter-rotating bookend vortices. This looks to me more like a complex multicellular setup that develops supercellular characteristics around the time the B and C regions merge with one another. The discussion of Fig. 10 in section 3 suggests there mid-level rotation consistent with a mesocyclone at these two times, further favoring a transient supercellular structure. As such, I disagree with the conclusion (“first known case of a bow echo”) on line 409-410. The semantics might not much matter; as the authors show, it is the presence of a deep, turbulent updraft that matters, and such updrafts are likely in a variety of convective modes.

6. Second paragraph of section 3.3: I found the discussion of the convex shapes in the graphs hard to understand. I think the authors’ conclusion is that sparkles have a systematically larger value of reflectivity and spectrum width, and that this is true at all values of moderate to high reflectivity. A more direct way to confirm this finding would be with a formal test to the shift of the maxima of the 2D histograms, which could be shown by a simple 2D Kolmogorov-Smirnov test. A 1D test could also be conducted within each normalized bin.

Minor concerns:

7. 20: suggest, “of point sources in the VHF radio band”

8. Fig. 1, annotation 1, and text near line 30: Is there a negative leader corresponding to the positive leader segment? I would expect a bidirectional tree.

9. Section 2.4 The radar processing methods are well-described in this and previous sections and take the required care with the processing of the polarimetric variables, storm advection, etc. However, the vertical resolution limitations described in appendix B should be highlighted in this section. There are clearly unresolved vertical gradients that might bias the radar statistics at the lightning locations.

10. Section 2.4 and related radar figures: In addition to altitude, the temperature profile of the atmosphere is also a key parameter. It would be useful to know the altitudes of the 0°C and -40°C isotherms, which constrain the altitudes of active primary noninductive charge separation, and to indicate these on vertical cross section figures where appropriate.

12. 221: "convergence (upward motion on the ground)” would be more precisely stated as “near-surface convergence forcing upward motion”