Small‐Scale Discharges Observed Near the Top of a Thunderstorm

We have used the LOw‐Frequency ARray (LOFAR) to image a few lightning flashes during a particularly severe thunderstorm. The images show an exceptional amount of VHF activity at altitudes above 10 km. Much of this is in the form of small‐scale discharges, not exceeding a few hundred meter, occurring seemingly randomly around the centers of active storm cells. To emphasize the incidental nature of these small‐scale discharges or sparks we refer to them as “sparkles.” A detailed investigation shows evidence that these sparkles are indicative of positive leader channels and that they are equivalent to the needle activity seen around positive leader tracks at lower altitudes.

investigate their detailed structure. We conclude that at least some sparkles are related to (VHF-quiet) positive leader activity much in the same way as needles are seen around positive leader tracks at lower altitudes. In most cases the track-structure is difficult to reconstruct due to the lower density of these sparkles as well as a probably more complicated leader structure in the turbulent top region in these clouds that have a rather complex charge structure. In the few cases where dart leaders show the track structure it is seen that they indeed connect many sparkles. Throughout this work we use the term dart leaders to for the fast re-ignition of leader channels. In the vast literature several different terms are used (see also Jensen et al. (2021) for a more extensive discussion) such as "K-changes" or "Recoil leaders" and where sometimes "Dart leaders" have been reserved for cloud-to-ground (CG) discharges. We feel that "dart" describes well the fast propagation where, if necessary, one may distinguish intra-cloud or CG leaders.
LOFAR (van Haarlem et al., 2013) is a radio telescope consisting of several thousands antennas. These antennas are spread over much of Europe. For the observations presented here we use the antennas in Dutch stations only operating in the 30-80 MHz VHF-band where the observations are imaged using the procedures described in Text S2 in Supporting Information S1.

Data
The 18 June 2021 storm was exceptionally strong with constant flashes for several hours with many hundreds of detected flashes, see (Text S1 and Figure S1 in Supporting Information S1). Since, for technical reasons, LOFAR records only a 1.5 s of data every 20 min, a total of nine recordings by LOFAR are available, each showing multiple flashes with sparkles at the higher elevations. This storm was the most severe one hitting the Netherlands during a round of severe storms in Europe lasting between 16 and22 June 2021 (MKWeather.com, 2021).
All recorded lightning flashes for this storm have a very similar structure where each LOFAR recording shows multiple lightning flashes across a few active thunderstorm cells. Each flash has many negative leaders at altitudes below 7 km altitude, many positive leaders between 7 and 10 km altitude showing the typical needle activity (Hare et al., 2019Pu & Cummer, 2019;Saba et al., 2020;T. Wu et al., 2019;B. Wu et al., 2022), and high-altitude negative leaders (Scholten et al., 2021a) as well as sparkles (the main subject of this work) at higher altitude. For this work we focus on two (out of six) thunderstorm cells in the seventh LOFAR recording in this series, 21C7 (at 18 June 2021, 19:37:28 UTC).
In Figure 1 we show the observations for the two thunderstorm cells obtained with the impulsive imager , see (Text S2 in Supporting Information S1). We have restricted the figure to altitudes exceeding 7 km to focus on what we call sparkles, relatively short (in time and space) clusters of VHF sources that seem not connected to lightning channels and show at seemingly random spots over comparatively long time periods (more than 1 s). The sparkles can be seen in the top panel of Figure 1 as a diffuse background of VHF sources at altitudes between 9 and 13 km. These sparkles show as two multi-colored "clouds" in the lower panels around the cores of the two thunderstorm cells centered at (N, E) = (30, −5) and at (20, 3) km, each with a horizontal diameter of the order of 10 km and with main density at 12 km altitude and marked by black ellipses. As shown by the color coding representing time, the multi-colored aspect of the clouds expresses that these sparkles occur at random times. The spottiness (at this scale) expresses that they are isolated and well separated. This in great contrast to negative leaders (HANLs at these altitudes) and dart leaders that show as thick and thin single-colored lines. By limiting to altitudes above 7 km we have eliminated the negative leaders at lower altitudes.
Analysis of radio soundings launched at De Bilt (approximately 140 km to the southwest of the storm location) at 18 June 2021 12:00 and 19 June 2021 00:00 UTC show that the height of the tropopause is around 12.5 km. The tropopause height indicates the height at which the vertical temperature gradient changes sign and hence air naturally stops rising at this level. Only storms with large internal dynamics are able to significantly rise above the tropopause, see Figure S1 in Supporting Information S1. Any storm that shows activity above the tropopause will exhibit large updrafts, which cause significant charge separation and are associated with strong turbulence. Note that the sparkles that we observe do not exclusively occur at levels above the tropopause but are also abundant below this level.

Results
To obtain a better insight in the nature of the sparkles we have investigated several of them in detail. Most of these sparkles can be resolved with our impulsive imager as a few individual VHF sources occurring at the same spot (typically within meters) over a time-span of milliseconds. Some of the larger ones may span a few hundred meters and contain many tens of sources as shown in the different panels of Figure 1. The chance that even a single source is due to an imaging effect can almost be excluded since in Figure 1 we only show those sources for which the pulse is fitted with an RMS of better than 2 ns for about 170 antennas with baselines reaching up to 100 km. Thus, having two or more sources within a few meters from each other excludes imaging artifacts. Some of the smallest sparkles show as single sources even when using our more sensitive TRI-D imager (Scholten et al., 2021b(Scholten et al., , 2022, see Text S2 in Supporting Information S1. The fact that even the TRI-D imager does not show much more structure is probably due to the very large amount of VHF emission all over the flashes making it impossible to image the weaker sources given the high background level. One of the most telling examples of the wide variety of sparkles we observe is shown in Figure 2 where we have zoomed-in on a tiny (in space and time) part of Figure 1. Figure 2a gives a zoom-in at a moderate scale showing a large number of sparkles all having about the size of a single dot, however when zooming in even further, as is done on Figures 2b and 2c it is clear that there is a large variety of sizes, some still having the size of a dot, some others with sizes of the order of a few hundred meter. An example of a larger sparkle is indicated by the orange circles in Figure 2c at t = 70 ms. It resembles a single corona-flash step of a HANL. It covers a vertical distance of only 200 m (including the very first source, 300 m) and has a horizontal extent of only 100 m. It is able to effectively discharge its volume as the subsequent HANL, starting at t = 175 ms, passes over and around the volume covered by the t = 70 ms discharge. This indicates that sparkles are negative discharges (see Text S4 in Supporting Information S1) where the necessary positive end is invisible in VHF. The general propagation   Figure 1 showing several dart leaders as well as some HANLs. In panel (b) only a limited time period is shown to indicate that many of the observed sparkles are lying on a track (indicated in gray) that conducts a dart leader at later times (t = 1,140 and 1,160 ms). See text for the meaning of the labels "S" and "R." The structure of the two negative discharges, marked as H 1 and H 2 are shown in more detail in panel (c). direction is upward for the structure at t = 70 ms as well as the HANL at t = 175 ms. An example of a smaller sparkle is indicated by the blue circle, also in Figure 2c, at t = 275 ms. Due only to the later dart leader at t = 360 ms it becomes clear that these sources are situated on a, hitherto unseen, positive leader track. The dart leader propagates upward, the same direction as the HANL.
The close-up on sparkles given in Figure 3a shows that they are very similar to those shown in Figure 2a. There are many small sparkles and also two larger structures, indicated by the black circles marked H 1 and H 2 . These larger structure are regarded as sparkles as they occur well before (400 ms) the development of the dart leaders in this area and because they are small (a length of less than 500 m) as compared to, for example, the dart leaders that cover many kilometers. In structure they are similar to the larger sparkles shown in Figure 3. These larger sparkles lie close (at a distance of less than 1 km) to the track taken by the dart leaders. Figure 3b, at the same spatial scale as Figure 3a but in time zoomed-in on some dart leaders, emphasizes the close relation between the sparkles and positive leaders. The several downward going dart leaders pass through the sparkle cloud and feed the negative leaders at lower altitudes (not shown). Fewer sparkles show here than in figure a simply because of the smaller time interval. In this figure we have indicated two regions "S" and "R" in the ground plane projection. The region labeled "S" shows intermittent and seemingly isolated VHF activity, while the region labeled "R" contains multiple downward dart leaders. The gray-line indicates a later dart leader (t = 1,140 and 1,160 ms). This shows that most of the activity in the "S" region is not truly isolated but is actually on a VHF-invisible positive leader. For example, just north of the "S" label is a long-thin negative discharge that, with LOFAR's resolution, is clearly some kind of discharge similar to needles. The dart leaders in region "R" of Figure 3b are not perfectly imaged by our impulsive imager, and show spots of VHF emission and empty holes. Thus, without the TRI-D imager, it is difficult to state unambiguously if the VHF emission is actually part of a dart leader or from sparkles. This indicated by the broad boundary between the sparkle region "S" and dart region "R." There is also sparkle activity around 12 km altitude in Figure 3b that is seemingly isolated, as we have no solid evidence of a positive leader connecting this activity to the main flash. A more detailed zoom-in on the longer sparkles among them showed that these propagate downward at an acute angle with the curve one would draw to connect them. Thus, even though the activity around 12 km altitude could be truly isolated from the main flash, we believe it is highly likely that the positive leader has actually propagated to 12 km altitude and simply has not been imaged due to the difficulty of locating positive leaders in VHF. Figure 3c gives a zoom-in on the larger sparkles H 1 and H 2 , showing that both are highly branched. As written earlier, they qualify as sparkles because they are relatively short (as compared to negative leaders) and occur well before a longer leader shows in their vicinity. The first one, H 1 , started at an altitude of 10.3 km and propagated downward over a distance of almost 1 km covering a horizontal distance of a little less than 300 m in about 10 ms, resulting in an average speed around 10 5 m/s. H 1 resembles a single corona flash of a HANL. The later one, H 2 , starts at an altitude of about 11.1 km moving upward over a distance of less than 300 m and horizontal over about 400 m, thus had an average speed of only 10 4 m/s. Therefore, despite both being highly branched negative leaders, H 1 and H 2 propagate at very different speeds. These two occur 500 ms before and less than 1 km eastward from where the dart leader passes at an altitude in between them. Their propagation structure implies that this region has a very complex charge structure.

Discussion
Important for the present analysis is the ability of the LOFAR imagers to determine with meter precision the location of a large number of sources (see Ref. Scholten et al. (2022) for a more detailed discussion of the resolution). This is essential for the present study as this allows to locate several sources per sparkle, thus excluding the possibility of their being imaging artifacts. Another, equally important, aspect is that, due to the large number of well imaged sources, we can clearly resolve the tracks of dart leaders. With conventional LMA systems the density of sources along a dart leader may make it difficult to distinguish them from the surrounding sparkle activity. These high-resolution data, with examples shown in Figures 2c and 3b, show that several sparkles are located on the tracks of dart leaders. Some others appear to line-up with each other, strongly suggesting that they are in fact on the track of a positive leader not (yet) made visible by a dart leader. If all sparkles are indeed signatures of positive leader tracks, based on what is seen at lower altitudes, one would expect to have observed many more dart leaders. This may imply that not all sparkles follow our hypothesis that they exist on un-imaged positive-leader channels. However, it may also be that at these altitudes it is more difficult to initiate a dart leader, much like negative leaders at higher altitudes are observed to differ considerably from those at lower altitudes. It is important to note that the growth of intra-cloud positive leaders is invisible in VHF with very few exceptions (Idone, 1992;Kong et al., 2008;Visacro et al., 2017;Wang et al., 2016), see Text S4 in Supporting Information S1, and positive leaders show in our images by the needle activity or by dart leaders passing over their tracks. This leads us to the conjecture that many sparkles are thus not uncorrelated discharges, but in fact connected by an extended network of positive leaders. However, for most of the sparkles there is no clear evidence for a connecting positive leader structure.
When imaging the details of the sparkles it is clear that they show a large variety in their structure. Some may show as only a few sources, like the ones marked in blue in Figure 2c, other may span a few hundred meters, like H 1 and H 2 in Figure 3c and most are somewhere in between. In this work we have not attempted to distinguish between them as there seems to be a continuum of shapes and structures. Even the two larger ones differ greatly in propagation speeds. Sparkles can well be distinguished from extended leader-forming structures like the HANLs seen in Figure 1 with a close-up of two shown in Figure 2a. Likewise the brightness of the sparkles coves a wide range from being as bright as negative leaders to being barely visible.
As is evident from the right panel of Figure S1 in Supporting Information S1 the clouds reach very high altitudes (for storms in this climate zone) reaching well above the tropopause lying at around 12.5 km. At the highest altitudes, the region where the sparkles are observed, thus strong turbulence is expected. This is supported by detailed cloud-model calculations that are routinely being made by the Dutch weather service. This turbulence may explain why the charge structures, indicated by the sparkle clouds, see Figure 1, have an extent of several kilometers only. This complicated charge structure is also evidenced by the sparkles H 1 and H 2 where H 1 starts horizontal and turns vertical while H 2 exhibits a very planar structure, as shown in Figure 3c. It is known that negative discharges turn horizontal and branch when entering positive charge regions. Thus, there must be a complex mix of positive and negative charge in this region. Possibly a thin negative charge layer bordered by two thin positive charge layers.
In spite of the turbulence at these altitudes still rather long dart leaders can be distinguished. An example is shown in Figure 3b. There are several possible explanations such as, -turbulence occurs at a scale of several kilometers, -the turbulence is like a rolling wave where the length is along the leader, or-the dart leader develops at a much shorter time scale than the scale involved with the turbulent motion. Typical wind speeds are of the order of a few m/s while the time scale for a leader is generally less than 1 s. Turbulence is most probably driving the charge separation in the clouds which may thus be larger than usual.
Sparkles can be defined as spatially small (not exceeding a few 100 m) intermitted discharges near the top of the clouds where this activity is seen over extended periods lasting the full duration of a LOFAR recording. Because this relates to a relatively long term phenomena it can be said that in the top regions of the cloud the discharge process is taking place in a sparkling mode. In LMA observations (Calhoun et al., 2013;Emersic et al., 2011;MacGorman et al., 2017) very similar looking small-scale intermittent discharges were observed in the overshooting tops of strong subtropical thunderstorms lasting for prolonged time periods. There are similarities between needles and sparkles where both sometimes develop into larger negative discharges and both come at a large variety of lengths. Also like needles we observe that at least some sparkles are a signature of positive leaders. However there is also a large number of sparkles where there is only circumstantial evidence that they are related to positive leaders while at lower altitudes needle activity is most of the time followed by a dart leader passing through the track. Another difference is that needles often show a clock-work regularity while it is at best intermittent for sparkles. For these reason we prefer to keep needles and sparkles separate. It might be that due to differences in the atmosphere at these altitudes (density and/or humidity) the sparkles may not twinkle as regularly as needles and this may also be the reason that relatively few dart leaders show. As stated earlier, differences in the atmosphere between lower and higher altitudes is probably also the reason that negative leaders appear as HANLs rather than normal negative leaders as observed at lower altitudes. It may thus also be the reason that needles show as sparkles at high altitudes, assuming all sparkles are indeed negative discharges around positive leaders.
We certainly do not want to exclude the possibility that the majority of sparkles for which a clear relation to positive leaders is not seen might be discharging small-scale charge structures within thunderclouds. These small-scale structures may either be produced by thunderstorm electrification and charge transport (e.g., including effects of small scale turbulence) or by charges left behind by previous lightning leaders.
Since sparkles are probably not all one all one type of discharge, the larger of our sparkles may also be the same as blue corona discharges observed from space near the top of deep convective clouds with overshooting tops, as reported by several recent studies (Chanrion et al., 2017;Dimitriadou et al., 2022;Husbjerg et al., 2022;Soler et al., 2021). They are believed to be streamer dominated discharge phenomena as no strong signals are observed in the 777.4 nm signature spectral line of hot lightning leaders. The currently accepted generation mechanism of blue corona discharges is that they begin with electrical breakdown in virgin air between two cloud charge layers of opposite charge. Large sparkles such as H 1 and H 2 appear to have characteristics (kilometer scale and being at the top of the clouds) that approach those of the blue corona discharges and are likely to be categorized as such if being observed from space. From our analysis we find that not all are isolated discharges and several are electrically connected to previously established leader channels. This suggests a new generation process for the blue corona discharges. Furthermore, our study indicates that there can be many of those types of discharges in an intense thunderstorm with smaller spatial and temporal scales, posing a great challenge to detect them from space. See Text S3 in Supporting Information S1 for a discussion of the difference between sparkles and compact intra-cloud discharges (Antunes de Sá et al., 2021;T. Wu et al., 2011).
The observations show a multitude of blue, kilometer-scale, discharges at the cloud top layer at 18 km altitude and a pulsating blue discharge propagating into the stratosphere reaching 40 km altitude.

Summary
We have used LOFAR to image a, for Dutch standards, particularly severe thunderstorm where we observed many VHF emitting sources at altitudes above 10 km. Particularly intriguing is that much activity seems isolated from the main flash, which we refer to as sparkles, in the form of seemingly uncorrelated sources of small spatial extent, that are similar to what has been seen before in LMA observations (Calhoun et al., 2013;Emersic et al., 2011;MacGorman et al., 2017) in the overshooting tops of strong subtropical thunderstorms. With LOFAR we can resolve the structure of most of these sparkles and find that they are reminiscent of the needles seen around positive leaders at lower altitudes (Hare et al., 2019) in the sense that they are short (typically up to 100 m) negative discharges. Sometimes these sparkles develop into larger negative discharges very much like what is seen with needle activity at lower altitudes. Additionally we find that several sparkles are in fact located along positive leader channels made visible by dart leaders although for a large number of sparkles such a relation is not seen unlike is the case for needles. Another difference with needles is the sparkles re-activate much less frequently. The sparkles for which no clear relation to positive leaders is seen might also be discharging small-scale charge structures within thunderclouds related to the strong turbulence associated with the overshooting tops.
The sparkles are seen very prominently in the present particularly violent thunderstorm in the top region of each thunderstorm cell. It is very likely that these cells have overshooting tops, but, as the height of the tropopause in Dutch summers may exceed 12 km, the sparkles are not confined to the overshooting tops as was observed in