The initial stage of cloud lightning imaged in high-resolution
O. Scholten, B.M. Hare, J. Dwyer, C. Sterpka, I. Kolmašová, O. Santolík, R. Lán, L. Uhlíř, S. Buitink, A. Corstanje, H. Falcke, T. Huege, J.R. Hörandel, G.K. Krampah, P. Mitra, K. Mulrey, A. Nelles, H. Pandya, A. Pel, J.P. Rachen, T.N.G. Trinh, S. ter Veen, S. Thoudam, T. Winchen
mmanuscript submitted to
JGR: Atmospheres
The initial stage of cloud lightning imaged inhigh-resolution
O. Scholten , , , B. M. Hare , J. Dwyer , C. Sterpka , I. Kolmaˇsov´a , ,O. Santol´ık , , R. L´an , L. Uhl´ıˇr , S. Buitink , , A. Corstanje , ,H. Falcke , , , , T. Huege , , J. R. H¨orandel , , , G. K. Krampah , P. Mitra ,K. Mulrey , A. Nelles , , H. Pandya , A. Pel , J. P. Rachen ,T. N. G. Trinh , S. ter Veen , S. Thoudam , T. Winchen University Groningen, Kapteyn Astronomical Institute, Landleven 12, 9747 AD Groningen, TheNetherlands University Groningen, KVI-Center for Advanced Radiation Technology, P.O. Box 72, 9700 ABGroningen, The Netherlands Interuniversity Institute for High-Energy, Vrije Universiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium Department of Physics and Space Science Center (EOS), University of New Hampshire, Durham NH03824 USA Department of Space Physics, Institute of Atmospheric Physics of the Czech Academy of Sciences,Prague, Czechia Faculty of Mathematics and Physics, Charles University, Prague, Czechia Department of Astrophysics/IMAPP, Radboud University Nijmegen, P.O. Box 9010, 6500 GL Nijmegen,The Netherlands Astrophysical Institute, Vrije Universiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium NIKHEF, Science Park Amsterdam, 1098 XG Amsterdam, The Netherlands Netherlands Institute of Radio Astronomy (ASTRON), Postbus 2, 7990 AA Dwingeloo, TheNetherlands Max-Planck-Institut f¨ur Radioastronomie, P.O. Box 20 24, Bonn, Germany Institut f¨ur Kernphysik, Karlsruhe Institute of Technology(KIT), P.O. Box 3640, 76021, Karlsruhe,Germany Erlangen Center for Astroparticle Physics, Friedrich-Alexander-Univerist¨at Erlangen-N¨urnberg,Germany DESY, Platanenallee 6, 15738 Zeuthen, Germany Department of Physics, School of Education, Can Tho University Campus II, 3/2 Street, Ninh KieuDistrict, Can Tho City, Vietnam Department of Physics, Khalifa University, PO Box 127788, Abu Dhabi, United Arab Emirates.
Key Points: • Our new LOFAR imaging procedure can locate over 200 sources per millisecondof flash with meter-scale accuracy. • The Primary Initial Leader breaks up into many (more than 10) negative leadersof which only one or two continue after 30 ms. • Some negative leaders propagate from the positive charge layer back to get closeto the initiation point.
Corresponding author: Olaf Scholten,
[email protected] –1– a r X i v : . [ phy s i c s . a o - ph ] J u l anuscript submitted to JGR: Atmospheres
Abstract
With LOFAR we have been able to image the development of lightning flashes withmeter-scale accuracy and unprecedented detail. We discuss the primary steps behindour most recent lightning mapping method. To demonstrate the capabilities of ourtechnique we show and interpret images of the first few milliseconds of two intra-cloudflashes. In all our flashes the negative leaders propagate in the charge layer below themain negative charge. Among several interesting features we show that in about 2 msafter initiation the Primary Initial Leader triggers the formation of a multitude (morethan ten) negative leaders in a rather confined area of the atmosphere. From theseonly one or two continue to propagate after about 30 ms to extend over kilometershorizontally while another may propagate back to the initiation point. We also showthat normal negative leaders can transition into an initial-leader like state, potentiallyin the presence of strong electric fields. In addition, we show some initial breakdownpulses that occurred during the primary initial leader, and even during two ”secondary”initial leaders that developed out of stepped leaders.
One of the key open questions in lightning science concerns the understandingof the processes that are fundamental to the initiation and early development of alightning flash. In particular, it is not known what processes lead to the creation ofthe primary initial leader (PIL) channel and how that channel propagates. In recentyears, the use of lightning mapping arrays (Rison et al., 1999; Edens et al., 2012)and VHF radio interferometers (Rhodes et al., 1994; Yoshida et al., 2010; Stock etal., 2014) augmented with fast antennas and optical measurements (Hill et al., 2011;Montany`a et al., 2015; Qi et al., 2016; Tran & Rakov, 2016) have led to the generalpicture that lightning initiation begins with an ionization event (Stolzenburg et al.,2020), which could be in the form of a powerful narrow bipolar event (Rison et al.,2016) or much weaker VHF source as seen in (e.g. Marshall et al., 2019; Lyu etal., 2019). This is followed by an initial leader, as imaged very nicely in (Lyu etal., 2016) in VHF. The initial leader propagation usually involves a series of largepreliminary breakdown pulses (see for example (Kolmaˇsov´a et al., 2014, 2018)), afterwhich normal negative stepped leader propagation is observed. The transition fromthe initial leader to a negative stepped leader has been observed in (Stolzenburg et al.,2020) with high-speed video and in electric field change data. Why negative leadersinitially propagate in a different mode than the normal leader stepping seen later inthe flash is not understood. Furthermore, the positive leader is often not observedduring the initiation process and only appears in radio data much later on after thenegative leader is well developed.With the present work we add very accurate images, obtained using LOFAR,showing the dynamics of Dutch thunderstorms. LOFAR (van Haarlem et al., 2013) is asoftware-phased array consisting of several thousand simple antennas that is primarilybuilt for radio astronomy, see Section 2.1. Thunderstorms we have observed in thegeneral area of the Dutch LOFAR stations (Dutch thunderstorms) differ from thethunderstorms seen in the US by the fact that all the flashes we have observed initiate atthe bottom of the main negative charge layer, and then propagate down into the lowerpositive charge layer. For many flashes, including the two discussed more extensivelyin this work, we observe an extensive network of negative leaders which very rarelyresult in a ground stroke. It thus appears that the negative leaders become “trapped”in the potential well of the lower positive charge layer. To improve the insight inthe dynamics of the lightning discharge immediately after initiation we have improvedour imaging technique over our earlier procedure (B. Hare et al., 2019) by greatlyincreasing the number of located VHF sources. In Section 2.2 we elucidate on themain improvements of our present imaging procedure. –2–anuscript submitted to
JGR: Atmospheres
Some detailed images of the initial stage of the lightning discharge are shownin Section 3, partially to demonstrate the capabilities of our present imager. Weshow images for two flashes, one from 2018 and one from 2019. The 2018 flash showsthe typical initial development we have seen in all our imaged flashes, about ten innumber. The flash starts with a very small pulse in VHF, barely recognizable evenwith our sensitive LOFAR antennas. This develops into a Primary Initial Leader, asalso imaged in (Lyu et al., 2016). During its development we detect rapidly increasingVHF (30 – 80 MHz) activity reaching a maximum about 2 ms after initiation. Forthe 2019 flash we also have data recorded from a broadband magnetic-loop antennaduring this time, showing significant low-frequency emissions at particular stages ofthe PIL development. After descending down from the negative to the positive chargelayer the PIL initiates a plethora of negative leaders almost simultaneously in anarea of about 1 km . Of the original multitude of negative leaders, only one or twocontinue to propagate after about 30 ms to form the main part of the flash which maycover distances of 10 km or more. The 2019 flash is interesting since we see therea PIL and even two secondary Initial Leaders (SIL). The Initial Leaders are clearlydistinguishable from a negative leader through their propagation speed, a relativelylow density of imaged sources and powerful VHF emission.Some suggestions for a possible interpretation of our observations are presentedin Section 4. Our mapping procedure of pulses detected by LOFAR basically follows the struc-ture outlined in (B. Hare et al., 2019). Arrival-time differences for pulses coming fromthe same source in different antennas are extracted from the data using the maximain the cross correlations. The present procedure incorporates important improvementsthat are mainly due to an improved procedure, inspired by that of the Kalman filter,to follow the pulse from the same source across many different antennas that may bemany tens of kilometers apart. For completeness we outline here the procedure wefollowed (Scholten, 2020).
LOFAR (van Haarlem et al., 2013) is a radio telescope consisting of several thou-sands antennas. These antennas are spread over a large area with a dense core (acircular area with a diameter of 300 m), the Superterp, near Exloo, the Netherlands,and with remote stations spread over Europe, reaching baselines in excess of 1000 km.The signals from these antennas can be added coherently to make this effectively op-erate as a gigantic radio dish, primarily used for radio astronomy. For our lightningobservations we confine ourselves to the LOFAR stations in the Netherlands, reachingbaselines of the order of 100 km, see Fig. 1. We use the Low Band Antennas (LBAs) op-erating in the frequency range from 30 – 80 MHz. The LOFAR antennas are arrangedin stations. Each station has 96 dual polarized antennas with an inverted V-shape.The signals are sampled at 200 MHz (5 ns sampling time). For our observations we useabout 12 antennas (6 for each polarization) per station. For the lightning observationsthe circular memory (called Transient Buffer Board, TBB) is used that can store 5 sof data per antenna. Upon an external trigger, taken from (
Blitzortung.org , n.d.), thedata on the TBBs are frozen and read out for later processing. In this read-out pro-cess we experience some data loss (due to missed hand-shaking during the downloadfrom the antenna field) which, thanks to our large number of antennas, does not affectthe image quality. Per 5 s recording we store close to 1 TB of data for later off-lineprocessing. The antennas have been calibrated on the galactic background radiation(Mulrey et al., 2019). –3–anuscript submitted to
JGR: Atmospheres
Figure 1: Layout of the Dutch LOFAR stations, adapted from (B. M. Hare et al., 2018).The core of LOFAR is indicated by the red (cid:76) sign, the position of the broadband mag-netic loop antenna SLAVIA is indicated by the red (cid:74) , while the green boxes show thegeneral location of the 2018 and 2019 flashes that are discussed in this work.
The basis of our imaging procedure is described in (B. Hare et al., 2019). Wechoose a reference antenna which usually is located in the core at the Superterp. Foreach pulse for which we want to search for the source location we select a relativelysmall section of the time trace in the reference antenna. The arrival times of pulses fromthis source in other antennas are calculated from the cross correlation of the selectedtrace with the traces in the other antennas. The source position is determined from achi-square fit of these arrival times.Imaging a flash starts with RFI mitigation, see Section 2.2.1. The time calibra-tion of all participating antennas, discussed in Section 2.2.2 is the step that requiresmost attention since we want to reach an accuracy of 1 ns for all antennas. Findingthe source positions is the third step which is through a fully automatized pipeline,see Section 2.2.3. For the final image we select those sources that obey certain qualityconditions, see Section 2.2.4, where one has to balance keeping a sufficient numberof sources with limiting the scatter. Full details of the new procedure are given in(Scholten, 2020), here we will outline the main aspects.
Since the LOFAR core is situated in a rather densely populated part of theworld there are many radio and TV transmitters that interfere with our observations.Because of these our frequency range is limited to 30 – 80 MHz. At lower and higherfrequencies there is too much RFI to excise it. The few narrow-frequency lines in ourdetection window we mitigate by software notch filters.
Since we want to achieve meter-scale resolution the relative timing of the anten-nas has to be calibrated at the nanosecond level. To achieve this over distances of –4–anuscript submitted to
JGR: Atmospheres
100 km we use a few selected pulses emitted during the flash in a bootstrap procedure.Regularly spread over the duration of the flash a small number (order of five) blocksof data (one block is 32k of 5 nanosecond time samples) are taken for the referenceantenna, which is taken in the dense core of LOFAR on the Superterp.Here, and later in the imaging procedure, we minimize, using a Levenberg-Marquardt algorithm, the root mean square time difference (
RM S ) between the calcu-lated arrival times and measured arrival times for all antennas to find the position of asource. The calculation uses the travel time of a signal from the source to the antenna.The measured arrival times are obtained from the peak position in the absolute valueof the cross correlation between a small part around the pulse in the reference antennaand the spectrum in the antenna.In each block, up to the four strongest pulses are identified as candidate calibra-tion pulses. The pulses from the same candidate sources are selected in the nearbystations. A candidate is eliminated from the calibration procedure if there is an am-biguity in selecting the correct pulse in the adjacent antennas. The known LOFARtiming calibrations are sufficient at this stage which for the core has an accuracy betterthan 5 ns. The
RM S is minimized to find the locations of the candidate calibrationsources. The Dutch antennas are separated in rings with increasing diameter. In iter-ations that follow, a larger ring of antennas is included. For each iteration the sourcelocations found in the previous iteration is used to make an educated guess of the pulsetimings for all antennas from these calibration sources. In a chi-square fit, the optimalstation timing calibrations (the same for all pulses) are determined while at the sametime updating the source locations. A station is excluded from the procedure for aparticular calibration source when the pulses attributed to this source show a largedifference with the actual arrival times for all antennas in this station.The fitting procedure is repeated increasing the radius of the ring around thereference antenna until all stations are included. Frequent visual inspection of the crosscorrelation spectra is important to guarantee that the used pulses are correctly assignedto the correct calibration source. When there is doubt, the candidate calibration sourceis eliminated from the procedure.In the final stage, the antennas in each station are calibrated (allowing for dif-ferences between antennas in the same station) by fitting simultaneously the antennatimings as well as the source locations of the remaining high-quality calibration sources,taking the previously obtained results as an initial guess.
The general source-finding stage is usually run as a standalone process. This is incontrast to the calibration stage, which requires human inspection. The time trace forthe whole flash is divided into overlapping blocks (32k of 5 nanosecond time samples)for each antenna and further processing is done on each block. The overlap is chosensuch that the pulses from sources anywhere in the general area of the flash can berecovered in all antennas.The block of the reference antenna is searched for candidate pulses to be imaged.The overlap regions are excluded so that the same source is not imaged twice. Can-didate pulses are the strongest ones that differ in peak position by more than about100 ns (the exact time difference depends on the width of the pulse) in the referenceantenna and are seen in the two antenna polarizations (dual polarization). Withineach block the candidate pulses are ordered in decreasing peak amplitude.For each candidate pulse in the reference antenna, a section of the time tracearound the pulse is taken for the calculation of the cross correlation with other anten- –5–anuscript submitted to
JGR: Atmospheres nas. Imaging (source finding) starts by performing a grid search over source locationsfor minimizing the
RM S for antennas on the Superterp. The thus obtained sourcelocation is taken as the initial guess for a chi-square search to find the optimal sourcelocation by minimizing the
RM S for all antennas within a certain distance from thecore. This fitting is repeated for an increasing number of antennas by increasing thecircle of included antennas. Peaks in the cross correlations are searched for within atiming window that is calculated from the covariance matrix that was obtained fromthe previous chi-square fit. If the peak in the cross correlation deviates by more thantwo standard deviations the antenna is flagged as excluded from the fit. An antenna isalso excluded when the width of the cross correlation, defined as the integral dividedby the peak value, differs by more than 60% from that of the self correlation in thereference antenna. The reason for excluding antennas is that it may happen that twopulses are close, or even interfere, for a particular antenna. It may also happen thatthe pulse is ’hidden’ in the noise. Not capturing this may derail the search for thesource location.The procedure of finding the source by gradually including more antennas whilelimiting the search window is inspired by that of the Kalman filter however is moreaccurate than even the extended Kalman filter, see (Pel, 2019) for an implementationof the Kalman filter for lightning imaging.After finding the source location the corresponding locations in the trace of eachantenna is set to zero and the following candidate pulse is taken.The inverted V-shaped LOFAR antennas have two possible orientations, SW-NEand SE-NW, and are thus sensitive to different polarizations of the incoming radiation.We notice that the pulse-shape may differ for the two polarizations, see Section 3.1.3for an example. For this reason we have organized our imaging algorithm such thatonly one of the antenna polarizations or both can be used in imaging where dual (both)is the default.
We observe that the imaging accuracy of a source is poorly reflected by thecovariance matrix that is obtained from minimizing the
RM S . The reason is thatselecting a wrong pulse in a series of antennas may still yield a reasonable fit but willresult is a source that is mislocated. We have observed that the obtained value forthe
RM S combined with the number of excluded antennas, N ex , appear to be goodadditional indicators of the image quality supplementing the diagonal element of thecovariance matrix corresponding to the error on the height, σ ( h ) , which is usually thelargest.Antennas are excluded from the fitting procedure when there is no clear peak inthe part of the spectrum that was searched. This could be due to the fact that thepulse is simply too weak to be seen but it could also be that the peak lies outsidethe search window. The latter is obviously problematic and should have contributedto the RM S . The setting of the imaging quality indicators ( σ ( h ), RM S , and N ex )is dependent on the location of the flash with respect to the core. Additionally thecriteria tend to be subjective, balancing a large number of sources with a minimum ofmislocated sources. In many cases the mislocated sources appear to be displaced by50 meters or more along the radial direction with the core of LOFAR at the center. –6–anuscript submitted to JGR: Atmospheres
In September 2018, the broadband magnetic loop antenna SLAVIA (ShieldedLoop Antenna with a Versatile Integrated Amplifier) has been installed by the De-partment of Space Physis, Institute of Atmospheric Physics of the Czech Academy ofSciences at a site about 10 km east from the LOFAR core near the village Ter Wisch,marked with the red (cid:74) in Fig. 1. The antenna has a surface area of 0.23 m andmeasures the time derivative of the magnetic field. The obtained waveforms are thennumerically integrated. The sampling frequency is 200 MHz, the frequency band islimited by a first order high-pass filter at 4.8 kHz and by a 13 th order low-pass filterat 90 MHz. The sensitivity of the recording system is 6 nT/s/ √ Hz corresponding to1 fT/ √ Hz at 1 MHz. At the Ter Wisch site, the signal is unfortunately affected bystrong man-made interferences, some of which cut through our high pass filter, andthe waveform had to be cleaned by 19 narrow band-rejection filters with bandwidths18-30 Hz at interference frequencies between 2 and 10 kHz, and at 18 kHz.
In this work we concentrate on imaging the initial development of two lightningflashes, where we almost randomly selected one from 2018 and another one from 2019.The 2018 flash shows features we see in all our imaged flashes (about 10 in total).A PIL is initiated at the lower side of a negative charge layer. This Initial Leaderpropagates with a velocity of about 10 m/s downward to the positive charge layerwhere it simultaneously initiates many negative leaders of which one or two continueto propagate over large distances. The 2019 flash shows a more complicated patternwhich can be understood as a PIL initiating in the usual way several negative leadersof which two convert into an initial leader again after a few milliseconds. This secondgeneration of initial leaders we have named Secondary Initial Leaders. In Section 3.2we will also present data for the 2019 flash from a broadband magnetic loop antenna(Kolmaˇsov´a et al., 2018). The complete 2018 flash, shown in Fig. 2, is a typical example of a flash imagedwith our techniques. The lightning flash (D20180813T153001) occurred on August 13,2018 at 15:30 at a distance of about 50 km from the core of LOFAR, see Fig. 1. Toobtain this image we have used antennas for both polarizations and set the limits onthe source quality as σ ( h ) < . RM S < N ex <
10 from a total of about265 antennas. This leaves 14267 imaged sources over the whole duration of the flashof 0.3 s. To give some idea of the effect of these limits we have relaxed the limit onthe
RM S to RM S < –7–anuscript submitted to
JGR: Atmospheres
Figure 2: Image of the 2018 flash showing sources with σ ( h ) < . RM S < N ex <
10 resulting in 14267 imaged sources. The top panel shows height v.s. timeof the sources where we have applied an off-set to put initiation close to t = 0. The samesources, with the same coloring, are shown in the other panels giving height and distancesnorth and east from the LOFAR core. To visualize the dynamics of the leader development after initiation, Fig. 3 showsthe time development in chronological frames. Time frame A shows the Primary InitialLeader starting at an initial height of 5.25 km developing downward. For the first0.5 ms no progression is observed but then it accelerates and progresses downwardalong a slanted path at about 1 . × m/s, somewhat faster than observed in (Lyuet al., 2016). After a kink in the path at 5 km altitude, (visible most clearly in thenorth v.s. altitude plot) the Primary Initial Leader fans out, initiating a multitudeof negative leaders at an altitude of about 4.5 – 4.0 km where the downward motionstops (see time frames B and C). Then distinct, almost horizontal, leaders develop inthe same fashion as negative leaders do with a speed of about 10 m/s, i.e. ten timesslower than the initial leader. It is impressive that the formation of negative leaders –8–anuscript submitted to JGR: Atmospheres
Figure 3: The early part of the 2018 flash is imaged for three subsequent time periodsto emphasize the dynamics in the initial stage. Shown are sources with σ ( h ) < . RM S < N ex <
20 resulting in 113, 2849, 2814 imaged sources respectively forthe three different time frames. After initiation at an height of 5.2 km one observes a fastdownward progression (A). At an height of about 4 km, the downward motion stops andnegative leaders develop in seemingly arbitrary directions at multiple places over an areain excess of 1 km (B). At the end only one continues to grow (C). 3 time frames, notpanels; figures need some work, we could show also subsequent time frames.starts simultaneously at multiple places over an area in excess of 1 km (B). Each ofthe new leaders appears to develop in seemingly arbitrary directions, some inward,some outward. Close inspection shows that they rather seem to cover the surface of aspatial structure. One also notices that there is some VHF activity along the path ofthe initial leader at 6 and 8 ms and 5.2 – 5.3 km height. Since this particular region inspace is observed to play a rather special role in the evolution of the flash we namedit the neck. The observed VHF emission indicates that current is flowing throughthe neck, even though there is no visible activity of a positive leader yet. Only after13 ms the positive leader starts to show at an angle w.r.t. the Primary Initial Leader.The neck will serve as the connection point between the upper negative and the lowerpositive charge for the whole duration of the flash. Time frame C shows a pronouncedpositive leader with ample twinkling activity along several needles (B. Hare et al.,2019; Pu & Cummer, 2019). The neck itself shows no needle activity. It is interestingto see that one negative leader propagated to within 500 m horizontal distance and atthe same altitude of the place where the neck showed a kink and started to fan out.Eventually there is only a single negative leader that continues to spread away fromthe initiation point, all others appear to have stalled. Fig. 4 shows the Primary Initial Leader with the located sources together andthe recorded power. The power is calculated as the square of the measured VHF-signal(the same as used in imaging) from on one of LOFAR’s core antennas and averaged over4 µ s. It is interesting to see that the VHF activity increases rapidly after initiation,reaching a peak at the time the Initial Leader creates negative leaders. This takesplace 2 ms after initiation. At later times the VHF activity decreases again.Making use of our high resolution, Fig. 5 shows a zoom in of the initial leaderto provide a better look at its properties. From the very first source, marked with a –9–anuscript submitted to JGR: Atmospheres
Figure 4: The VHF power in arbitrary units in 4 µ s bins, is compared to imaged sourcesfor the first 6 ms of the 2018 flash.star, the initial leader appears to fan out reaching a diameter of about 100 m in heightand less in horizontal directions. In the first part of the Initial Leader developmenta few different stages can be distinguished. The first stage ranges from initiation, t = 0 .
55 ms, till t = 0 .
94 ms. The VHF trace shows almost no power above background.The imager finds several good quality sources towards the end of this period. At thisfirst stage the leader moves over a distance of 20 m horizontally and 50 m downwardat a speed of 1 . × m/s. In the subsequent second stage, lasting from t = 0 .
94– 0.97 ms an increased VHF activity is visible. During this stage the leader movesover a distance of 80 m horizontally and 90 m downward with a speed of 4 × m/s,considerably faster than in the first stage. The third phase lasts till about 1.1 ms whereone sees a clear first burst of VHF intensity. The few sources (6) that are located at thisstage are lying on a continuation of the leader seen at the second stage or around theprevious leader. In the fourth stage from t = 1 . × m/s. At later stages the general downward motion stops,the propagation speed decreases, the VHF intensity continues to drop, and we are ableto map an increasing number of sources. From the located sources we observe thata multitude of negative leaders branch off from the PIL, and then propagate at thetypical speed of a negative leader, 10 m/s. To show that the flash activity already started at 0.55 ms we show in Fig. 6 apart of time trace in the vicinity of this first imaged source. Here the pulse standsout clearly and, because the spectrum is relatively clean, we can observe it in allantennas. Also some even smaller pulses can be imaged, but these do not pass thequality criteria used in making Fig. 3 and Fig. 5. It should be noted that the tracein the two polarization directions is rather different, signalling that this first imagedpulse is due to a complicated current pattern, with currents in multiple directions. Inaddition this pulse is considerably longer than the pulse response of our system (which –10–anuscript submitted to
JGR: Atmospheres
Figure 5: A zoom-in on the sources observed in the initial 1.1 ms of the 2018 flash. Thefirst located source, the initiation point, is marked with a black star.is 50 ns FWHM) which is additional evidence that its source is composite, not justa single short pulse. We observe that the sub-structure of the pulse does not changesignificantly for antennas at different orientations w.r.t. the source (taking into accountthe polarization). Based on the pulse response of the system of 50 ns (15 m lengthat the speed of light) we thus conclude that the spatial extent of the source must besmall, of the order or less than (10m) . The complete image of the 2019 flash we discuss in this paper, D20190424T213055,occurring on April 24, 2019 at 21:30 is shown in Fig. 7. The time is shifted such thatthe flash starts close to t = 0. Likely due to the close proximity to the core of LOFAR,we could image for this flash a larger density of sources per ms of the flash. The flash –11–anuscript submitted to JGR: Atmospheres s Figure 6: The Hilbert-envelope of the trace for each of the two polarization directionsas measured at the core of LOFAR. The vertical axis gives pulse amplitude in arbitraryunits, the horizontal axis time (in µ s) centered at the first located source (at 0.55 ms inFig. 5).shows the same charge-layer structure as was seen for the 2018 flash discussed in theprevious section. For this flash the positive charge layer extends from about 4 kmalmost to the ground. The negative charge layer does not extend above 7 km height.The currents from the lower positive to the upper negative layer all appear to flowthrough the neck that was formed at initiation. In Section 3.2.2 we will show thatwithin 100 m from the neck we observed a negative leader propagating to the positivecharge layer. The initiation phase of the 2019 event is shown in detail in Fig. 8. It has beenverified that there is no distinct peak visible in the time traces before the time of thefirst imaged source at t = 0 . × m/s rather straight downthis time. At an altitude of about 4 km it starts to fan out, producing a multitude ofnegative leaders over an area of about 1 km . This is qualitatively the same as wasobserved for the 2018 event. Time frame B shows that after 3 ms most of the negativeleaders stop propagating, while two of them show a fast motion covering a distanceof 2 km in 2 ms, ten times faster than the propagation speed of the negative leaders(10 m/s) and the same as that of the Primary Initial Leader. Another resemblanceis that the number of imaged sources on this leader is relatively low. For this reasonwe call them Secondary Initial Leaders. The bottom-left panel shows that when theyreach a height of 3 km one of the Secondary Initial Leaders repeats the process ofgenerating a multitude of negative leaders over an area in excess of 1 km , i.e. in thesame positive charge layer situated a kilometer below and 2 km eastward than fromthe positive layer seen in time frame A. The bottom-right panel shows that most ofthe negative leaders in the second phase have stopped propagating, and that some ofthe initial group are re-activated.At the time of initiation of this flash the magnetic loop antenna recorded data.The magnetic loop antenna is situated at a site some 10 km east from the LOFARcore, see Fig. 1. Fig. 9 shows the recorded magnetic loop antenna spectrum alignedwith the imaged sources and the VHF power (averaged over 2 µ s) as recorded byLOFAR. The figure shows that during the evolution of the Primary Initial Leader, inthe first 1.5 ms, the VHF power increases in steps, identical to what was observed –12–anuscript submitted to JGR: Atmospheres
Figure 7: Image of the 2019 flash showing sources with σ ( h ) < . RM S < N ex <
25 (out of a total of close to 400) are shown, resulting in 46523 imaged sources.in Fig. 5 for the 2018 flash. During this time the magnetic loop antenna recorded adozen strong initial breakdown pulses. The VHF power shows another strong increase(by a factor 2 or 3) at t = 3 ms when the Secondary Initial Leaders start to emerge.Around the same time the magnetic loop antenna also measures an enhanced densityof initial breakdown pulses. There is another phase of enhanced activity seen in theML antenna when the secondary initial leaders start to ignite the negative leadersaround t = 4 . –13–anuscript submitted to JGR: Atmospheres
Figure 8: The initiation of the 2019 flash. Setting the image quality at σ ( h ) < . RM S < N ex <
35 leaves 323,99,413, and 2332 sources for the different sequen-tial time frames. The black star in time frames B – D marks the position of the point ofinitiation.is about 0.01 – 0.02 ms. This is very close to the burst duration seen in (B. M. Hareet al., 2020) where it is associated with negative leader stepping. However, the timebetween bursts of VHF power is about 0.1 – 0.15 ms which is longer than the 0.05 msobserved in (B. M. Hare et al., 2020) for normal negative leader propagation.From Fig. 8 it can be seen that the initial leader propagates almost verticallydownward and the propagation speed can thus be deduced directly from Fig. 10. Thisindicates a constant acceleration of the PIL after it starts propagating, very similar towhat was observed for the 2018 flash.In a follow-up paper, (Scholten et al., 2020), the correspondence between thebroad band signal and the LOFAR image will be discussed in more detail. There we –14–anuscript submitted to
JGR: Atmospheres
Figure 9: The power recorded at a core station of LOFAR (re-binned over 2 µ s, middlepanel) is compared to the recording in the magnetic loop antenna (top panel, units of[nT]) and the height of the imaged sources shown for the first 7 ms in Fig. 8. All panelshave been aligned in time.will address beside the pulses seen in the broad band spectrum close to initiation alsothe pulses seen later during the flash. In the vicinity of the neck region, we have observed some interesting leaderpropagation that should be taken into account when reconstructing the structure ofthe charge layers in the vicinity of the initiation point labeled with a (cid:13) in Fig. 11.At about t = 10 ms a negative leader propagates up to the point (a) in the figure,a mere 500 m away from the initiation point horizontally. At t = 35 ms the samenegative leader activates a second time and approaches even to a distance of 300 m.At about the same time from point (b) at an altitude of 5 km a negative leader startsto propagate down, with no previous activity seen at its initiation point, towards theneck, to turn parallel to it at a distance of a mere 100 m where it stops propagating atthe top of the charge layer where the first negative leaders were observed. At about thesame time some VHF sources are observed along the positive leader. At t = 55 ms at(c) a bit higher than 5 km another negative leader starts propagating towards point (b)where it appears to connect to the channel of the earlier negative leader that startedfrom here. This very complicated leader structure seems to imply that around theinitiation point a fair amount of charge was present in a complicated charge structure.In the 2018 flash we have also observed that a negative leader propagated up from thelower positive charge layer to the vicinity of the neck, but the other features appearto be unique for the 2019 flash. –15–anuscript submitted to JGR: Atmospheres Magnetic Loop Signal [nT]VHF Power time [ms] H e i gh t [ k m ] Figure 10: The power recorded at a core station of LOFAR (re-binned over 0.5 µ s, mid-dle panel) is compared to the recording in the magnetic loop antenna (top panel) and theheight of the imaged sources shown for the first 0.5 ms in Fig. 8. All panels have beenaligned in time. We have improved the procedure used in (B. Hare et al., 2019) for imaging light-ning flashes using data from the LOFAR radio telescope. Our new imaging procedurehas a proven capability to locate over 200 sources per millisecond of flash with meter-scale accuracy.We investigated the fist few ms of some flashes over the Netherlands and observeinteresting leader structures, for example pointing to dense and relatively small chargeclouds (see presentation).All images of flashes we have made with LOFAR show a negative charge layeraround 5 km and below this a positive charge layer which is consistent with the atmo-spheric electric fields over the LOFAR core as were determined in (Trinh et al., 2020).We also observe an extensive structure of negative leaders in the lower positive chargelayer. Very rarely do we see negative leader activity propagating up into the upperpositive layer. This general structure is opposite to what is generally observed in USthunderstorms.In all our imaged flashes we observe a very similar initial development of the dis-charge where after initiation the discharge slowly grows (in fractions of a millisecond),picks up intensity (in the form of strong increase in emitted VHF power) as is typicalfor an initial leader. When the initial leader reaches the charge cloud which attractedit, the fast forward propagation stops and instead we observe the creation of many –16–anuscript submitted to
JGR: Atmospheres
Figure 11: An expanded view of the neck region for the 2019 flash. The letters are usedin the text to refer to particular parts of this image. The initiation point is labeled by the (cid:13) .more normal negative leaders that propagate at about one-tenth of the speed of theinitial leader.This naturally leads to the picture that the initiation at the bottom side of anegative charge layer was driven by a relatively small (order 1 km ) pocket of ratherdense positive charge. This created the strong electric field that is needed to causelightning initiation. In this strong field the elongating leader rapidly acquires chargeby induction leading to increased currents and heating which could explain the initialacceleration we observe in the PIL. An acceleration of the initial leader was also ob-served in (Cummer et al., 2015), although the acceleration was much smaller (about afactor 2 in velocity) than what we observe (close to an order of magnitude). Since theambient electric field extends up to the positive charge pocket, the fast propagation ofthe PIL stops as it reaches it and we start to observe the usual negative leaders thatpropagate along potential extremals (Coleman et al., 2008) and thus in a much weakerambient field.Our observations suggests that a relatively small positive charge pocket may bethe driver for the strong electric field in which the initiation of the lightning happens.The presence of large hydro meteors, see (e.g. Dubinova et al., 2015), or a large numberof small droplets, see (Kostinskiy et al., 2019), will be required as initiation sites. Theinitiation process is greatly facilitated by a high density of free electrons as createdby cosmic rays as suggested in (Rutjes et al., 2019). It is not clear if any of these –17–anuscript submitted to JGR: Atmospheres mechanisms can explain the initial accelerating propagation of the initial leader thatwe have observed.The large number of negative leaders in a very confined space, as initiated bythe PIL, are a strong indication that there must have been a pocket with a ratherhigh charge density. Once the charge in this confined charge cloud is collected themajority of the leaders stop propagating and only few remain. For the 2019 flash theremay have been another positive charge cloud in the vicinity that attracts a secondaryInitial Leader. As this is initiated from the tip of a propagating negative leader thereis sufficient charge available to immediately have a fast-propagating discharge.It has been suggested that turbulence is a very efficient mechanism in especiallythunderclouds for creating regions with large fields that even grow exponentially withtime (Mareev & Dementyeva, 2017). This could thus be the mechanism that createdthe relatively small pocket with high charge density.Another indication of a dense and complicated charge layer structure in thevicinity of the initiation point is the fact that in our images we see a negative leaderturning back towards the Neck, the initiation point of the flash. This happened forthe 2018 as well as for the 2019 flash.We observe a strong correlation between the signal of the broadband antennaand the emitted VHF power, as was already suggested in (Kolmaˇsov´a et al., 2018,2019). The relation between emitted VHF power, the signal of a broadband antennaand interferometry images for the initial leader has been investigated in (Krehbiel,2017) for New Mexico lightning flashes.
Acknowledgments
The LOFAR cosmic ray key science project acknowledges funding from an Ad-vanced Grant of the European Research Council (FP/2007-2013) / ERC Grant Agree-ment n. 227610. The project has also received funding from the European Re-search Council (ERC) under the European Union’s Horizon 2020 research and in-novation programme (grant agreement No 640130). We furthermore acknowledge fi-nancial support from FOM, (FOM-project 12PR304). BMH is supported by the NWO(VI.VENI.192.071). AN is supported by the DFG (NE 2031/2-1). TW is supportedby DFG grant 4946/1-1. The work of IK and O.Santol´ık was supported by EuropeanRegional Development Fund-Project CRREAT (CZ.02.1.01/0.0/0.0/15-003/0000481)and by the Praemium Academiae award of the Czech Academy of Sciences AS. Thework of RL and LU was supported by the GACR grant 20-09671S. KM is supported byFWO (FWOTM944). TNGT acknowledges funding from the Vietnam National Foun-dation for Science and Technology Development (NAFOSTED) under Grant 103.01-2019.378. ST acknowledges funding from the Khalifa University Startup grant (projectcode 8474000237).This paper is based on data obtained with the International LOFAR Telescope(ILT). LOFAR (van Haarlem et al., 2013) is the Low Frequency Array designed andconstructed by ASTRON. It has observing, data processing, and data storage facilitiesin several countries, that are owned by various parties (each with their own fund-ing sources), and that are collectively operated by the ILT foundation under a jointscientific policy. The ILT resources have benefitted from the following recent majorfunding sources: CNRS-INSU, Observatoire de Paris and Universit´e d’Orl´eans, France;BMBF, MIWF-NRW, MPG, Germany; Science Foundation Ireland (SFI), Departmentof Business, Enterprise and Innovation (DBEI), Ireland; NWO, The Netherlands; TheScience and Technology Facilities Council, UK.The data are available from the LOFAR Long Term Archive (lta.lofar.eu), underthe following locations: –18–anuscript submitted to
JGR: Atmospheres
L664246_D20180813T153001.413Z_stat_R000_tbb.h5L703974_D20190424T213055.202Z_stat_R000_tbb.h5 all of them with the same prefix srm.grid.sara.nl/pnfs/grid.sara.nl/data/lofar/ops/TBB/lightning/ and where“stat” should be replaced by the name of the station, CS001, CS002, CS003, CS004,CS005, CS006, CS007, CS011, CS013, CS017, CS021, CS024, CS026, CS030, CS031,CS032, CS101, CS103, RS106, CS201, RS205, RS208, RS210, CS301, CS302, RS305,RS306, RS307, RS310, CS401, RS406, RS407, RS409, CS501, RS503, or RS508.
References
Blitzortung.org. (n.d.). http://LightningMaps.org .Coleman, L. M., Stolzenburg, M., Marshall, T. C., & Stanley, M. (2008). Horizon-tal lightning propagation, preliminary breakdown, and electric potential innew mexico thunderstorms.
Journal of Geophysical Research: Atmospheres , (D9). Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2007JD009459 doi: 10.1029/2007JD009459Cummer, S. A., Lyu, F., Briggs, M. S., Fitzpatrick, G., Roberts, O. J., & Dwyer,J. R. (2015). Lightning leader altitude progression in terrestrial gamma-ray flashes. Geophysical Research Letters , (18), 7792-7798. Retrievedfrom https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2015GL065228 doi: 10.1002/2015GL065228Dubinova, A., Rutjes, C., Ebert, U., Buitink, S., Scholten, O., & Trinh, G. T. N.(2015, Jun). Prediction of Lightning Inception by Large Ice Particles andExtensive Air Showers. Phys. Rev. Lett. , , 015002. doi: 10.1103/PhysRevLett.115.015002Edens, H. E., Eack, K. B., Eastvedt, E. M., Trueblood, J. J., Winn, W. P., Kre-hbiel, P. R., . . . Thomas, R. J. (2012). Vhf lightning mapping observationsof a triggered lightning flash. Geophysical Research Letters , (19). doi:10.1029/2012GL053666Hare, B., et al. (2019). Needle-like structures discovered on positively charged light-ning branches. Nature , , 360–363. doi: 10.1038/s41586-019-1086-6Hare, B. M., et al. (2018). LOFAR Lightning Imaging: Mapping Lightning WithNanosecond Precision. Journal of Geophysical Research: Atmospheres , (5),2861-2876. doi: 10.1002/2017JD028132Hare, B. M., Scholten, O., Dwyer, J., Ebert, U., Nijdam, S., Bonardi, A., . . .Winchen, T. (2020, Mar). Radio emission reveals inner meter-scale structureof negative lightning leader steps. Phys. Rev. Lett. , , 105101. Retrievedfrom https://link.aps.org/doi/10.1103/PhysRevLett.124.105101 doi:10.1103/PhysRevLett.124.105101Hill, J. D., Uman, M. A., & Jordan, D. M. (2011). High-speed video observationsof a lightning stepped leader. Journal of Geophysical Research: Atmospheres , (D16). doi: 10.1029/2011JD015818Kolmaˇsov´a, I., Marshall, T., Bandara, S., Karunarathne, S., Stolzenburg, M.,Karunarathne, N., & Siedlecki, R. (2019). Initial breakdown pulses ac-companied by vhf pulses during negative cloud-to-ground lightning flashes. Geophysical Research Letters , (10), 5592-5600. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2019GL082488 doi:10.1029/2019GL082488Kolmaˇsov´a, I., Santol´ık, O., Defer, E., Rison, W., Coquillat, S., Pedeboy, S., . . .Pont, V. (2018). Lightning initiation: Strong pulses of vhf radiation ac-company preliminary breakdown. Scientific Reports , (3650), 2045-2322.Retrieved from https://doi.org/10.1038/s41598-018-21972-z doi:10.1038/s41598-018-21972-z –19–anuscript submitted to JGR: Atmospheres
Kolmaˇsov´a, I., Santolik, O., Farges, T., Rison, W., Lan, R., & Uhlir, L. (2014).Properties of the unusually short pulse sequences occurring prior to the firststrokes of negative cloud-to-ground lightning flashes.
Geophysical ResearchLetters , (14), 5316-5324. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2014GL060913 doi: 10.1002/2014GL060913Kostinskiy, A. Y., Marshall, T. C., & Stolzenburg, M. (2019). The mechanism ofthe origin and development of lightning from initiating event to initial break-down pulses. (arXiv:1906.01033). Retrieved from https://arxiv.org/abs/1906.01033 Krehbiel, P. (2017, Aug). Studies of lightning initiation. In (p. 1-2). doi: 10.23919/URSIGASS.2017.8105171Lyu, F., Cummer, S. A., Lu, G., Zhou, X., & Weinert, J. (2016). Imaging light-ning intracloud initial stepped leaders by low-frequency interferometric light-ning mapping array.
Geophysical Research Letters , (10), 5516-5523. doi:10.1002/2016GL069267Lyu, F., Cummer, S. A., Qin, Z., & Chen, M. (2019). Lightning initiation processesimaged with very high frequency broadband interferometry. Journal of Geo-physical Research: Atmospheres , (6), 2994-3004. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2018JD029817 doi:10.1029/2018JD029817Mareev, E. A., & Dementyeva, S. O. (2017). The role of turbulence in thunder-storm, snowstorm, and dust storm electrification. Journal of Geophysical Re-search: Atmospheres , (13), 6976-6988. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2016JD026150 doi: 10.1002/2016JD026150Marshall, T., Bandara, S., Karunarathne, N., Karunarathne, S., Kolmasova, I.,Siedlecki, R., & Stolzenburg, M. (2019). A study of lightning flash initiationprior to the first initial breakdown pulse. Atmospheric Research , , 10 -23. Retrieved from doi: https://doi.org/10.1016/j.atmosres.2018.10.013Montany`a, J., van der Velde, O., & Williams, E. R. (2015). The start of light-ning: Evidence of bidirectional lightning initiation. Scientific Reports , (1), 15180. Retrieved from https://doi.org/10.1038/srep15180 doi:10.1038/srep15180Mulrey, K., Bonardi, A., Buitink, S., Corstanje, A., Falcke, H., Hare, B., . . .Winchen, T. (2019). Calibration of the lofar low-band antennas using thegalaxy and a model of the signal chain. Astroparticle Physics , , 1 - 11.Retrieved from doi: https://doi.org/10.1016/j.astropartphys.2019.03.004Pel, A. (2019). Imaging lightning with the extended kalman filter (Master Thesis,University of Groningen, Faculty of Science and Engeneering, KVI-CART).Retrieved from http://fse.studenttheses.ub.rug.nl/id/eprint/19751
Pu, Y., & Cummer, S. A. (2019). Needles and lightning leader dynamics imagedwith 100 – 200 mhz broadband vhf interferometry.
Geophysical Research Let-ters , (22), 13556-13563. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2019GL085635 doi: 10.1029/2019GL085635Qi, Q., Lu, W., Ma, Y., Chen, L., Zhang, Y., & Rakov, V. A. (2016). High-speed video observations of the fine structure of a natural negative steppedleader at close distance. Atmospheric Research , , 260 - 267. doi:https://doi.org/10.1016/j.atmosres.2016.03.027Rhodes, C. T., Shao, X. M., Krehbiel, P. R., Thomas, R. J., & Hayenga, C. O.(1994). Observations of lightning phenomena using radio interferometry. Journal of Geophysical Research: Atmospheres , (D6), 13059-13082. doi: –20–anuscript submitted to JGR: Atmospheres
Nature Communications , (1), 10721. Retrieved from https://doi.org/10.1038/ncomms10721 doi:10.1038/ncomms10721Rison, W., Thomas, R. J., Krehbiel, P. R., Hamlin, T., & Harlin, J. (1999). Agps-based three-dimensional lightning mapping system: Initial observations incentral new mexico. Geophysical Research Letters , (23), 3573-3576. doi:10.1029/1999GL010856Rutjes, C., Ebert, U., Buitink, S., Scholten, O., & Trinh, T. N. (2019). Gen-eration of seed electrons by extensive air showers, and the lightning in-ception problem including narrow bipolar events. Journal of Geophysi-cal Research: Atmospheres , (13), 7255-7269. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2018JD029040 doi:10.1029/2018JD029040Scholten, O. (2020). A practical guide to lightning imaging with lofar (internalreport). Kapteyn Institute, University of Groningen, NL. Retrieved from
Scholten, O., et al. (2020). Broadband radio pulses and lofar imaging. to be submit-ted to JGR .Stock, M. G., Akita, M., Krehbiel, P. R., Rison, W., Edens, H. E., Kawasaki, Z., &Stanley, M. A. (2014). Continuous broadband digital interferometry of light-ning using a generalized cross-correlation algorithm.
Journal of GeophysicalResearch: Atmospheres , (6), 3134-3165. doi: 10.1002/2013JD020217Stolzenburg, M., Marshall, T. C., & Karunarathne, S. (2020). On the transitionfrom initial leader to stepped leader in negative cloud-to-ground lightning. Journal of Geophysical Research: Atmospheres , (4), e2019JD031765.Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2019JD031765 (e2019JD031765 2019JD031765) doi: 10.1029/2019JD031765Tran, M. D., & Rakov, V. A. (2016). Initiation and propagation of cloud-to-groundlightning observed with a high-speed video camera. Scientific Reports , (1),39521. Retrieved from https://doi.org/10.1038/srep39521 doi: 10.1038/srep39521Trinh, T. N. G., Scholten, O., Buitink, S., Ebert, U., Hare, B. M., Krehbiel,P. R., . . . Winchen, T. (2020). Determining electric fields in thun-derclouds with the radiotelescope lofar. Journal of Geophysical Re-search: Atmospheres , (8), e2019JD031433. Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2019JD031433 (e2019JD031433 10.1029/2019JD031433) doi: 10.1029/2019JD031433van Haarlem, M. P., et al. (2013). LOFAR: The LOw-Frequency ARray. A&A , ,A2. doi: 10.1051/0004-6361/201220873Yoshida, S., Biagi, C. J., Rakov, V. A., Hill, J. D., Stapleton, M. V., Jordan, D. M.,. . . Kawasaki, Z.-I. (2010). Three-dimensional imaging of upward positiveleaders in triggered lightning using vhf broadband digital interferometers. Geophysical Research Letters , (5). doi: 10.1029/2009GL042065(5). doi: 10.1029/2009GL042065