2.1 Lightning Observations

This study is principally based on the use of lightning and electromagnetic field measurements made at both near and long range of the thunderstorm. For short-range magnetic-field measurements we used the magnetic-field sensor SLAVIA (Shielded Loop Antenna with a Versatile Integrated Amplifier). The antenna loop plane was oriented vertically, at an azimuth of 235° (clockwise from the geographic North). A downward current at this azimuth with respect to the antenna induces the maximum positive response of the cosine antenna pattern. The sensor was coupled with a broadband receiver (5 kHz–37 MHz) sampling at 80 MHz. This receiver was initially developed for the TARANIS spacecraft (Blanc et al., 2007), and later adapted to ground-based measurements (Kolmašová et al., 2018, 2020). The voltage at the receiver input is proportional to magnetic field time derivative for a source at an azimuth . The waveform records are numerically integrated over time to estimate the variations of the horizontal magnetic field component . Duration of recorded waveforms was set to 420 ms including a pre-trigger time of 3.5, 7, or 10.5 ms. The measurements were conducted in a quiet electromagnetic environment of an external measurement site of the Laboratoire Souterrain à Bas Bruit (LSBB), located on the summit of La Grande Montagne (1,028 m, 43.94°N, 5.48°E) close to Rustrel, France. For electric-field measurements further away from thunderstorm, we used a vertical whip antenna mounted on a mast. Received signals are sampled at 12.5 MHz, and analyzed in the frequency range 3 kHz–5 MHz. Duration of recorded waveform was 30 ms including a pre-trigger time of 6 ms. The electric-field analyzer was located at a distance of 321 km northwest from Rustrel (at 560 m, 46.1°N, 2.8°E). A detailed description of the instrumentation is given in Farges and Blanc (2011). The same receiver was used in the study of Fullekrug et al. (2013) and Kolmašová et al. (2014). The time alignment was realized by the GPS receivers connected to both receivers. Before data analysis narrow-band interferences caused by radio transmitters were removed from electromagnetic recordings by applying several narrow band-pass filters.

The European lightning location network EUCLID (www.euclid.org) provided times of occurrence, locations, polarities, peak currents, and types of discharges, which were analyzed in this study. Individual EUCLID sensors report the angle of incidence of the electromagnetic field (method called magnetic direction finding) and the time of field arrival at the sensor. An optimization algorithm is used to find a position that fits best all the available measurements. The median location accuracy for CG discharges is about 100 m (Schulz et al., 2016) as there is a clearly defined striking point, where the lightning channel attaches to the ground and the return stroke current wave starts propagating upwards. In case of IC discharges which do not have ground striking points, EUCLID provides ground level positions that fit best the measured times received from the contributing sensors. Thus the coordinates of the IC discharges mainly result from optimization algorithm finding the best fit to times and angles reported by different sensors. In case of a perfect vertical IC channel, which could be projected as a single point at the earth surface, the location accuracy of IC is the same as for CG. In any other case it will be somewhat larger, depending on the geometry of the IC lightning channel. The detection efficiency for IC discharges was relatively low as there were not sufficient number of sensors in 2013 to localize typically weaker IC discharges.

To characterize the type of the discharge, the lightning location networks identify the CG/IC differences either in the peak-to-zero times, or in combinations of waveform parameters, or in the altitude of the discharge (Kohlmann et al., 2017). The machine learning techniques (Wang et al., 2020; Zhu et al., 2021) rely on the differences in the pulse shapes. To differentiate the “IC-type” pulses, PB pulses, and the “RS-type” pulses in our study, we used our measurements of the horizontal magnetic field components and a criterion based on visual classification of their pulse waveform shapes, as described below in Section 4.1.

2.2 Other Sources of Data

CTTs are issued from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) onboard the Meteosat Second Generation (MSG) satellite operated by the European Organization for the Exploitation of Meteorological Satellites (EUMETSAT). The radiometer SEVIRI scans the Earth disk to provide images in 12 spectral bands every 15 min at a spatial resolution of 0.027°, which corresponds to 3 km at the subsatellite point and about 4.5 km at the latitude of the study area (around 43°). The study area is therefore scanned four times in one hour, around 10, 25, 40, and 55 min of each hour. The thermal infrared band (IR) at ∼11–13 μm allows estimating the CTT with an accuracy generally better than ∼1°C.

Radar data used in this study were collected by the operational radar network ARAMIS (from French « Application Radar à la Météorologie Infra-Synoptique » which means Radar Application to the Meterology Infra-Synoptic). Radar reflectivity observations were composited following the approach described in Bousquet and Tabary (2014) to produce gridded reflectivity products at the spatio-temporal resolution of 1 km horizontally, 500 m vertically and 15 min from using plan position indicator (PPI) data collected during 15’ volumic radar cycles. In order to account for the non-simultaneity of the measurements during each 15’ period of sampling, all observations are synchronized with respect to the ending time of the period (the reader is referred to Bousquet et al., 2008 for more details about data analysis).

To analyze the meteorological conditions for the day of the storm that produced the PB pulses, we use the ERA5 reanalyzes provided by the European Centre for Medium-Range Weather Forecasts (ECMWF; available on the European Copernicus/ECMWF Data Center) to retrieve CAPE, geopotential (at 500 hPa), and horizontal wind velocity (at 1,000 and 500 hPa levels). ERA5 is based on the Integrated Forecasting System (IFS) Cy41r2 which was operational in 2016 (Hersbach et al., 2020). Furthermore, we use the NCEP/NCAR reanalyzes for the tropopause temperature which is a parameter available in the database produced thanks to a cooperation of the NCEP and the NCAR (Kalnay et al., 1996).

3.1 Weather Patterns at Synoptic and Regional Scales

On 19 June 2013, weather forecasts (https://www.estofex.org/) show a high probability of Deep Moist Convection (DMC) with very high CAPE values (exceeding 3,000 J kg⁻¹) over a large part of Central Europe. This was mainly due to the overlap of steep lapse rates of temperature along with high boundary layer moisture (dew points exceeding 20°C). Furthermore, an upper low initially centered over the Iberic Peninsula moved northwards above the Bay of Biscay, together with a surface cold front. It formed a cold-core low and forced a southeasterly flow over France (Figure 1). Figure 1a, which shows the altitude of the pressure level 500 hPa, highlights the cold-core low centered above northern Spain, with lower values around 5.5 km. Above the study area (1°−6°E; 41°–45°N), the altitude of the 500 hPa pressure was comprised between 5.6 and 5.7 km. A first consequence of this low is a lower altitude and a warmer temperature of the tropopause within the area, as indicated in Figure 1b. Indeed, the tropopause temperature had a maximum of −40°C within the center of the low and about −48°C above the study area. Another consequence is the air motion counterclockwise around the low, especially strong at 500 hPa level, as indicated by Figure 1c, which shows that the velocity of the geostrophic wind is maximum in the west part of the study area with about 45 m s⁻¹. The wind at this level is generally oriented toward the north in the study area. Figure 1d shows the wind at 1,000 hPa level is much weaker in the study area with a velocity lower than 5 m s⁻¹ in the western part of the study area and around 10 m s⁻¹ in its eastern part. The wind shear was therefore very high between the two levels, especially in the western part of the study area.

At the regional scale, thunderstorms were expected over Northern France because of large values of CAPE and steeper mid-level lapse rates of temperature. Furthermore, the low-level convergence and the slowly translating frontal boundary was expected to favor thunderstorm initiation. Figure 2a displays the distribution of the CAPE over France on that day at 00:00 UTC. The larger values around 3,000 J kg⁻¹ were mainly located over the north and the northeast of the map. This region of large CAPE values expanded over the northern part of France with values <2,000 J kg⁻¹, while another maximum was located over central France. Over southwestern France, weak mid-level lapse rates of temperature limited CAPE values around 500 J kg⁻¹ over the Pyrénées range and up to 1,000 J kg⁻¹ over Catalonia. The presence of very strong vertical wind shear, up to 40 m s⁻¹ between 1,000 and 500 hPa levels (∼0–6 km layer) in the western part of the study area, is expected to contribute to the development of thunderstorm systems of multi-cell type, which could maintain convective activity for a duration of a few hours (Markowski & Richardson, 2010).

3.2 Description of the Storm Activity

Figure 2b displays the CG flash density calculated with a resolution of 0.05° × 0.05° using EUCLID data collected within the 6-hr period 01:00–07:00 UTC; nearly 15,000 CG lightning flashes were recorded over France and in the surrounding area during this time period. The flash number was obtained from using typical grouping criteria of time and distance between CG strokes (0.5 s and 10 km, respectively, Cummins et al., 1998). The larger values of the CG flash density are located over northern France, with several areas of activity and a local maximum value close to 2 flashes km⁻². The study area is indicated by a frame in Figure 2 and corresponds roughly to the most southern part of France. The CG lightning flash density was low in this area with values not exceeding 0.25 flash km⁻². This lightning activity was produced by several thunderstorm cells, generally organized in southeast northwest oriented lines. Several lines of convective cells were simultaneously observed with the French radar network ARAMIS. The first cells were initially small with a size around 10 km in diameter and progressively, new cells developed and clustered within these lines which moved northwards at a velocity of about 50 km hr⁻¹. The convective system (see description in Section 4.3) displays a multi-cell organization including cells with moderate reflectivity values (up to 45 dBZ at low altitude). It produced lightning activity during more than 5 hr with low CG flash densities (up to 0.25 km⁻²). The CTT of this convective system was not very cold since the minimum value detected during the period of lightning activity was −57°C. The coldest values of CTT were observed around 05:00 UTC. The properties of lightning discharges detected by EUCLID in the area of interest and their correspondence with our electromagnetic observations are described in Section 4.2.

4.1 Analysis of Electromagnetic Waveforms

As a base line, we investigated magnetic field waveforms, since the magnetic field receiver used a broader frequency band and was located closer to the storm, allowing to distinguish more waveforms details. The level of instrumental noise and external interferences allows identifying broadband pulses exhibiting peak-to-peak amplitudes larger than 0.4 nT. The individual waveforms were analyzed to identify trains of bipolar PB pulses, IC-type pulses, and RS-type pulses. We used a simple criterion based on their pulse waveform shapes (Wang et al., 2020). An isolated bipolar pulse was considered to belong to an IC discharge if the immediately following overshoot of opposite polarity exceeded one-half of the peak amplitude of the original pulse. A sequence of bipolar pulses with hundred microsecond long interpulse time intervals was considered as a group of PB pulses. A RS-type pulse generated by a CG discharge was defined by a smaller overshoot of the opposite polarity, not reaching one-half of the peak amplitude of the original pulse. For simplicity, we will use shorter terms: RS, IC, and PB pulse in the following text. As the receiver was operating in a triggering mode, the recordings of waveform captures were predominantly triggered by the RS pulses. Nevertheless, strong PB pulses were also able to activate the trigger.

Whenever a PB train consisting of at least five bipolar pulses was visually identified in the magnetic field waveform, we searched for the same lightning phenomena in the concurrently recorded electric field waveform snapshots, and checked visually the similarity of the pulse train patterns recorded by both receivers. Additionally, the 420-ms long magnetic-field waveforms were used to examine the pulse activity following the PB trains within a period exceeding the longest previously published pre-stroke time interval between the PB process and the first RS (383 ms; Marshall, Schulz, et al., 2014).

The resulting data set consists of 10 strong PB trains, observed between 3:46 and 5:56 UT. The selected PB pulse trains were followed by an IC activity in seven cases and by weak RS pulses in the remaining three cases. Note, that the criteria did not include any meteorological condition and did not take into account the observed flash rates.

In the following, we focus on properties of the 10 trains of PB pulses, which were selected using the criteria defined above. We labeled them by letters from A-J. Three examples of magnetic- and electric-field waveforms containing typical PB trains are plotted in Figures 3a–3c. Magnetic field and electric field waveforms are respectively represented by blue and red lines. The initial polarity of electric field pulses confirmed the motion of the negative charge downward, as in case of negative CG discharges (note that we use the atmospheric electricity sign convention for the measurement of electric field that is positive downward). The polarity of the magnetic field pulses is given by the relative position of the receiving antenna to the recorded discharge. The polarity of the magnetic field waveforms shown in Figure 3 was inverted to stress the similarity of both magnetic and electric field pulse patterns. All waveforms were shifted in time to have the PB pulse with the largest peak-to-peak amplitude at the time t = 0. The peak amplitudes of PB pulses within the trains in Figures 3a and 3b were nearly monotonically increasing with time; they reached a maximum in a few hundreds of microseconds, and then decreased again, forming a “diamond” shape of the observed peak amplitudes. The pulses were relatively regularly distributed within the trains. Toward the end of the trains, the pulses lost their clear bipolar shapes, and became rather chaotically spaced. The envelopes of the pulse peak amplitudes within seven trains (out of 10) had a shape of a diamond as in Figures 3a and 3b. The envelopes of three remaining trains were less regular, similar to Figure 3c. The envelopes of all three trains which were followed by a weak RS had a diamond shape. Nevertheless, envelopes of other three trains which were isolated also have a shape of a diamond, so we do not see any correlation between the shape of the envelope and the presence or absence of the RS.

To estimate a range of widths of particular bipolar PB pulses detected during the thunderstorm, we have selected the first pulse with a peak-to-peak amplitude exceeding 1 nT and a pulse with the largest peak-to-peak amplitude within each magnetic-field pulse train. Two sets of 50-µs long waveform snapshots containing selected pulses were produced (Figures 3d and 3e). For each pulse, the half of the time interval between the negative and positive peaks was set as the time t = 0. Because of the different angles of arrival and traveled distance of individual events, which influenced the strength of the received signal, the waveforms were normalized by their positive peak maxima. Individual pulses are plotted by black dotted lines with exception of the strongest pulses from trains E and J, which exhibited unusual shapes having two positive or negative peaks and are displayed by solid lines (light gray, train E; dark gray line, train J, indicated by arrows) in Figure 3e. The pulses plotted by blue lines show an average small pulse (Figure 3d) and an average large pulse (Figure 3e) with a similar duration of the initial polarity half-pulse of about 12 µs, and a different duration of the opposite polarity half-pulse of about 15 and 23 µs. Note that the waveshapes were not affected by the arrival angles of the received signals.

The inter-pulse time intervals between clearly distinguishable bipolar pulses in all magnetic-field trains varied from 25 to 220 µs with a mean value of 78 µs. The time between the first and the largest PB pulse varied from 260 to 650 µs with a mean value of 450 µs. To estimate the duration of the trains, we have used 10-ms long waveform snapshots starting by the first PB pulse and divided them in 100-µs long parts. Mean values of the magnetic field and standard deviations for all these parts were then estimated. Taking into account the integrated noise of the analyzer (0.14 nT given in Kolmašová & Santolík, 2013) a threshold of 0.07 nT was chosen for the absolute value of the mean and a threshold of 0.21 nT for the standard deviation. The time, when both the mean value and the standard deviation dropped permanently below their thresholds, was considered as the end of a train. Overall PB trains were found to span over 2.4 ms on average with a duration varying from 1.7 to 3.4 ms. The RS-type pulses that were found to follow the PB pulse trains A, E and G, respectively occurred 15, 81, and 20 ms after the first detectable PB pulse.

4.2 Analysis of Relevant EUCLID Detections

The time development of the thunderstorm and the flash density as detected by EUCLID are illustrated in Figure 4. The selected area (1°−6°E; 41°–45°N) covers the whole thunderstorm, which was moving to the northeast along the Mediterranean coast from 01:00 to 07:00 UTC. The lightning flash activity was recorded between 01:18 UTC and 06:35 UTC, with 1,433 reported strokes, including 927 CG strokes and 506 IC strokes. Note that the detection efficiency of EUCLID for IC discharges was still very limited in 2013. Note also that EUCLID discriminates IC and CG discharges of both polaritites, and that PB pulses—if detected-should appear in the detection list as IC discharges. Nevertheless, the variability of both IC and CG pulse shapes being enormous, the classification algorithm might get confused by non-typical waveform characteristics of detected pulses (Kohlmann et al., 2017). EUCLID provided information for the strongest PB pulses occurring within 10 trains of PB pulses (Table 1). We labeled them by letters from A to J. There were two PB pulses detected by EUCLID within the PB train G. These two PB pulses occurred at the same location 112 µs apart, belonged to two strongest pulses, and we labeled them G_a and G_b. This time delay was confirmed also by electromagnetic recordings. The EUCLID classification of the type of the discharge in Table 1, sixth column is not taking into account in our analysis as we know from the analysis of electric and magnetic waveforms, that all these PB pulses have negative IC waveform signatures. Such PB detections were misclassified as CG strokes in the CG flash density and were not considered in Figure 4a. The remaining CG strokes were grouped into 604 CG flashes using the same criteria as in Figure 2b. The CG flash density reached about 0.25 flash km⁻².

Figure 4c presents another representation of the time distribution of discharges detected by EUCLID including their types and peak currents. Blue crosses correspond to positive CG discharges, blue open circles show negative CG discharges, while IC discharges are represented by black dots. Although PB pulses were reported as IC, +CG and −CG in EUCLID data, we show them using red stars. Regardless of the fact that EUCLID assigned positive currents in four cases of +CG discharges related to PB pulses, we plot all of them below zero, as we were able to reveal their negative polarity from the initial polarity of PB pulses in the electric-field records. The peak currents related to PB pulses varied from 15 to 36 kA. The peak currents reported by EUCLID (indicated by arrows in Figures 3a–3c) were associated with the strongest PB pulses within trains with exception of the trains F and G. EUCLID reported the second strongest PB pulse occurring within train F and two strongest pulses occurring within train G. The peak currents belonging to three RS pulses, which were found to follow PB pulse trains, were weaker than the corresponding strongest PB pulses and reached 28, 13, and 9 kA. In three other events, EUCLID reported IC discharges following the detections related to PB pulses, in remaining four events the PB pulses detections were isolated in time with no other activity detected by EUCLID within 30 seconds. (Note that the absence of the RS or the presence of a weak RS is resulting from our selection criterium 2).

EUCLID information indicate that the magnetic and electric field PB pulse trains were respectively measured at distances 69–176 and 253–377 km from their source lightning discharges. From this distance measurements, as the signals propagated from their source discharges to the receivers, one can independently verify the correspondence of trains measured by both receivers by checking the difference in the times of arrival of the same pulses. We found that all the analyzed pulses corresponded to the speed of light propagation within 10 µs pulse timing precision. We were also able to calculate the peak-to-peak electric field amplitudes of the largest PB pulses normalized to the 100 km range assuming that the electric field amplitude is inversely proportional to the distance, without additional attenuation (Kolmašová et al., 2016). The amplitudes varied from 2.5 to 8.7 V/m reaching on average 5.9 V/m. The properties of observed PB events, and relevant EUCLID detections are summarized in Table 1.

4.3 Thundercloud Structure

In order to place our electromagnetic observations in a cloud structure context, we examine the CTT and the horizontal distribution of the radar reflectivity using horizontal cross section of the 3D reflectivity fields at a height of 3 km. These horizontal sections at 3 km altitude allow radar reflectivity values to be kept high enough to separate the convective cells which are not very developed that day, and associate PB events with the most likely cell. According to ERA5 reanalysis, the temperature at this altitude on that day is about 1° or 2°C. The vertical structure of the reflectivity field is also inferred from vertical cross-sections performed at selected locations. Figures 5-7 show these three representations at the time of occurrence of the observed strong PB pulses A, D, F, G, H, and J in the area of the convective system that produced all PB pulses. The scales of temperature and reflectivity are the same for all panels for a comparison. We superimpose in these graphs the location of the PB pulses together with those of the CG and IC discharges detected by EUCLID within a 10 min window to be as well as possible synchronized with both radar and cloud top temperature scans. The scan of the CTT above the study area is made at a given time, explicitly at 10, 25, 40, and 55 min of each hour and indicated in the CTT panels. The CG and IC discharges are represented in a time interval of 10 min straddling the times of the CTT panels, which corresponds also with the duration of the radar scan plotted in each case. In order to account for possible errors in the synchronization of composite radar data that could arise from the fast storm motion, Table 1 indicates the minimum CTT and the maximum radar reflectivity at 3 km within the cells associated with the PB events, and for the whole system, respectively.

Figure 5 shows the plots of four cases of PB pulses from a period during which seven PB pulses were recorded (03:45–05:00 UTC) with the 10 min of lightning activity and PB pulses, for each. The storm system was characterized by northwest-southeast oriented lines of convective cells moving northwards at a speed of about 50 km hr⁻¹. Observations indicate a reinforcement of the convection over time with CTT decreasing and radar reflectivity increasing. Figures 5a and 5b display the PB event A (red and black star) and the lightning discharges over 03:35–03:45 UTC. Few discharges are produced during this period. They are located within some developing cells with CTT a little colder than −50°C (panel a) and maximum radar reflectivity of about 40 dBZ (panel b). The event A is located within a convective line with a weak vertical development, with a minimum CTT of −39°C (Figure 5a), radar reflectivity of 32 dBZ at 3 km (Figure 5b) and of 20 dBZ below 10 km (Figure 5c). This discharge is relatively isolated and is an IC pulse according its waveform shape. It was produced by a cell that has a core of reflectivity at low altitude (<2.5 km) and another one between 4 and 5 km (Figure 5c). Such structure could be related to a low dipole of charge, according to the study by Salvador et al. (2021).

Figures 5d and 5e show the discharges recorded between 04:20 and 04:30 UTC for the event D, which was detected at 04:23:28 UTC (red and black stars). These observations correspond to a more developed storm stage characterized with more discharges produced over 10 min. Most discharges were observed in the main convective line, with CTT of −54°C and reflectivity value at 3 km of 44 dBZ. However, the PB pulse D was detected alone in a small cell at the rear of the main convective line. Figure 5f shows that this cell has a weak development compared to the main line, with a CTT of about −45°C and reflectivity value of 32 dBZ at 3 km. The cell from the main line also displayed in Figure 5f at a latitude of 43.7°N is much more developed with the 20 dBZ at 11 km of height and large reflectivity values that reach the ground. The system produced even more discharges 15 min later as indicated in Figures 5g and 5h, when the PB event F was recorded. A large majority of lightning discharges are still produced within the main convective line that exhibits cells with −55°C for the coldest CTT and 45 dBZ for the maximum value of the reflectivity. The cell that produced the PB event F was significantly less developed with a CTT of −43°C and a reflectivity value of 35 dBZ at 3 km. The vertical cross section of reflectivity shown in Figure 5i shows that this PB pulse seems associated with the most recent cell on the edge. Figures 5j–5l display the PB pulse G for which the area of analysis was shifted of 0.5° to the east and a little enlarged to the south in order to include the most southern part of the system. The intensity of the storm was maximum during this period as was the lightning activity in terms of flash rate (Figure 4c). The cells are now essentially organized within a single convective line. At 04:40 UTC, the CTT is colder than −55°C and the radar reflectivity reaches 45 dBZ at 3 km, in the two regions of the storm (Figures 5j and 5k) that concentrate a large majority of the lightning discharges. The PB event G is located at the edge of a cell located in the most southern area, with reflectivity value of about 40 dBZ at 3 km. It exhibits a core of reflectivity at 2.5 km with values that reach about 45 dBZ and a 20 dBZ top above 10 km of height (Figure 5l).

The PB pulse H is plotted in Figure 6 with six panels in this case. Indeed, because the PB event was produced at 05:35:55 UTC, it is difficult to associate it with a horizontal cross section of radar reflectivity and we consider two at 05:30 and 05:45 UTC. Panel a displays the CTT of the storm with a minimum value of −56°C. However, the PB pulse is out of the coldest region of the cloud top. Figures 6b and 6e display the horizontal cross section of radar reflectivity at 05:30 and 05:45 UTC, respectively. Figures 6c, 6d and 6f display three vertical cross sections according to the dashed lines in Figures 6b and 6e. For both configurations, the PB pulse is located out of the most active region of the storm, rather at its edge on east side, which is confirmed by the vertical cross sections. Indeed, the largest reflectivity values and the lightning discharges are always located to the west of the longitude of the PB location. The vertical cross sections show low values of reflectivity (30–35 dBZ) at the location of the PB event, with maximum at a height of 3–4 km and a small horizontal extent.

Figure 7 displays the last case of PB event (J) detected at 05:55:58 UTC. In this case, we plot the CTT at 05:55 UTC (panel a), the horizontal cross section of radar reflectivity at 06:00 UTC (panel b) and four vertical cross sections (panels c–f). The location of the PB pulse is out of the main convective region according to panels a and b. The four vertical cross sections, two west-east (panels c and d) and two south-north (panels e and f), allow to compare the storm structures in the vicinity of the PB pulse and in the most convective region of the storm. We note also in this case, a weak development associated with the PB event, with no precipitation close to the ground and a low cloud top. It seems this event was produced by a weak cell in early development, or associated with very weak activity at the edge of the main storm system whose lightning activity is strongly decreasing at that time according to panel c of Figure 4.

The peak current estimated in EUCLID data varied from 22 to 36 kA for six displayed PB events, which is rather large for IC strokes. The negative polarity detected for these events suggests they may have been produced by a lower thunderstorm dipole, which consists of a negative charge layer above a positive one and tends to produce a larger proportion of IC flashes and large peak currents (Salvador et al., 2021).