A Chronos 1.4 (CR14-1.0-16M) camera was mounted in a waterproof SolidBox 69200. The box was covered with a plexiglass dome, 200 mm Duradom as per the manufacturer's designation, and is depicted in Fig. .

The video resolution of the camera is 928 pixels × 928 pixels with 1612.33 frames per second and a constant shutter speed from 4.9 to 34 µs during daytime thunderstorms and a maximal shutter speed of 614.6 µs during nighttime thunderstorms. The shutter speed is set by the instrument operator depending on current weather conditions. The video length saved by the camera was set to 2 or 3 s. The format of the videos saved was H.264, resulting in MPEG-4 (.mp4) video, which sacrifices some quality for better compression. The main possible loss is gamma-encoded brightness, which possibly alters the absolute values of the real brightness of recorded lightning.

The camera lenses that we used were FE185C057HA-1 . Therefore, the high-speed camera was only sensitive to visible light. This spectral baseband was selected to minimize the absorption of light generated by lightning in the atmosphere.

Different radio frequency receivers were used to record the radio emissions generated by lightning activities. The main objective was to cover a wide radio frequency band. The very low frequency (VLF) radio frequency band is known to be sensitive mainly to large (size) and high-current lightning channels, such as the return stroke. On the contrary, the ultrahigh-frequency (UHF) radio frequency band involves radiation almost exclusively associated with the fine structure of lightning .

During the whole measuring campaign, NaI(Tl) ionizing radiation detector sensors and a silicon photomultipliers (SiPM) were used. Details of the AIRDOS-C gamma spectrometer used can be found in . The energy deposition range of this device is 200 keV to 40 MeV. The time resolution between two incident ionizing radiation events is 100 µs for particles with deposited energy above 1 MeV. For lower energies, only a 15 s integration of events is provided. The NaI(Tl) crystal with the SiPM and preamplifier is shielded in a metallic box.

The electric field is measured using the Kleinwächter EFM 115 electric field mill (EFM) from which the analog signal output leads directly to the data logger that contains a GPS receiver for timestamping the record . The logging time resolution is 110 ms.

In order to determine the meteorological situation in the vicinity of the cars during the measurement, the cars were equipped with several meteorological sensors: a distrometer , an anemometer , and a thermometer with a barometer .

Figure  shows an example of ionizing radiation measurements using the AIRDOS-C gamma spectrometer. In Fig. a, a storm approaching a parked measuring car is displayed using data from the Blitzortung.org network. The vertical red lines mark the times when lightning was registered by the STP antenna. Figure b shows the number of particles registered by the ionizing radiation detector every 15 s. In this particular example, the lightning activity ceased just after 20:30 UTC​​​​​​​. The graph shows an approximate 30 % increase in the ionizing radiation flux. This increase is related to a radon progenies washout from the atmosphere caused by rain that started at 19:45 UTC.

Figure  shows a short time period of ionizing radiation measurement at the time when the storm was closest to the car. Individual particles of ionizing radiation registered versus detected lightning are shown. Only particles that have passed on energies higher than 2.4 MeV to the detector are displayed.

During the measurement, the camera is pointed toward the zenith and its lens allows it to capture the entire sky from horizon to horizon. For the illustration of the camera view, a plain camera frame is shown in Fig. . Because lightning discharges often happen inside clouds and lightning channels are not directly visible, we converted the videos of the recorded lightning to luminosity in time. The integral values of the illuminations were calculated using a script from the video recordings of the high-speed cameras. The calculations were carried out over the entire image area by summing up the values of all pixels on each image. We call these luminosity curves. The results can be seen on the graphs in Fig.  that depict the course of light flux over time.

To maintain clarity, luminosity curves were added to videos, which were converted from black-and-white (BW) recordings captured by cameras to false colors in order to make the details of the lightning (that have high luminosity dynamics) visible.

Figure  captures one frame from a video (https://doi.org/10.5446/60389, ) that clearly shows a flash of lightning as well as the leaders and recoil leaders. During the thunderstorm recording, the camera gain was decreased in order to capture lightning that has high brightness dynamics. As a result, lightning is not visible in the video at the same time as details of clouds and surrounding terrain. The processed video has the position of the horizon marked with a green circle with an inscribed designation of the cardinal directions. In the upper part of the video, there is a white graph showing the luminosity curve with a green pointer marking the position of the current frame over time. On the right, a color palette is visible that corresponds to the brightness recorded by each camera pixel with a depth of 8 bits. The bottom line shows the number of the current frame over the total number of frames recorded, Sg = 1/1 (information on which part of the camera's internal memory was used for recording), and the time of the current frame “T =” in seconds relative to the trigger.

The resulting luminosity curves contain similar parts (sharp peaks or slow changes in luminosity). Thus, we can compare the video recordings of luminosity curves with a similar progression, and we can choose those parts from several recordings that have similar luminosity curves and a visible lightning channel (not covered by clouds). This process helps us to see what lightning really looks like.

One of the advantages of measuring lightning using magnetic loop antennas is that the observation is not disturbed by optically nontransparent clouds or, as is the case for storms occurring during the day, sunlight scattered in the atmosphere. On the other hand, when we use a coil to measure the magnetic component of the electric field, we only see its changes in time or specifically a change in the current that flows through the lightning channel. If a constant current flows through the lightning channel, it is not possible to detect it using a magnetic loop antenna. Figure  shows an example of lightning recorded by an STP antenna and a camera at the same time. In the camera recording, a change in light conditions is clearly visible. The sky was brighter before the lightning than after it occurred. Slow changes in brightness, and thus slow changes in current, are not visible on the antenna recording. On the other hand, we see clusters of fast pulses that can give information about the development of lightning channels, which are not visible on the camera recordings because they are probably obscured by light emanated by a constant current in the main channel. Pulses that are visible on both camera and antenna recordings are recoil leaders. The most prominent pulse is the CG (cloud-to-ground) return stroke.

We have tried to compare the detection of lightning using the STP antenna with its detection using the Blitzortung.org network. First, we have to note that not every lightning event detected by the STP antenna was recorded, as it takes several minutes to store lightning data from the oscilloscope to the data storage device. Second, not every lightning event detected by the STP antenna was detected by the Blitzortung.org network. Figure  shows that the STP antenna detected lightning at a different time or at a distance of more than 120 km away when the storm was closest to the car according to the Blitzortung.org network. On the contrary, as shown in Fig. , when the storm was located tens of kilometers from the observation site according to the Blitzortung.org network, we see perfect conformity with the data from the STP antenna. In both figures, an interval of ±1 s is marked around the vertical lines corresponding to the detection times.

Lightning at 18:24:48 UTC was detected by a high-speed camera (see video, https://doi.org/10.5446/60389, ). According to the video, one of the lightning channels occurred almost directly above the measurement car, but the nearest lightning event detected by Blitzortung.org was at least 70 km away. The positioning accuracy of the Blitzortung.org network is in the order of kilometers. Blitzorting.org detected discharges at a distance of 70–80 km simultaneously. Thus, we can deduce that this lightning event was more than 80 km long or that a synchronous discharge occurred 80 km away.

Figures  and show a comparison of the vertical electric field measurements at the car measurement site with those from the camera recordings. The EFM on the car is not grounded; thus, it measures the gradient of the electric potential between the sensor and the car body. As can be seen in Fig. , the electric discharges occur at times of large electric field changes. However, if we look at the details of some discharges recorded by the camera (Fig. ), we can see that the changes in the electric field do not correspond directly to individual discharges. Please note that EFM measures the vertical component of the electric field, which gives the integrated value over a large area of clouds, and the field is also deformed by the presence of a car.

We attempted to determine the duration of lightning events by searching for local maxima in the illumination of the camera chip in time from peaks that have a prominence higher than 2 times that of noise. Then the first and last maxima were used for the determination of the first and last snap, and the lightning duration was calculated from the frame rate. First, 10 snaps of the camera record were omitted due to possible artifacts stemming from the video data compression. The amplitude of noise was established from the first 100 frames (from frame 10 to frame 110). Examples of how we determined the duration of some lightning events are shown in Figs.  and . The histogram of the duration of lightning events is shown in Fig. . This histogram comprises 107 lightning events measured during 12 thunderstorms.

To calculate the total length of a lightning event from the radio signal shown in Fig. , we used a methodology that treats signals greater than 4σ variance, calculated from the moving window of 10 ms, as the start of the lightning discharge. This algorithm is applied symmetrically from both ends of the recorded signal. Therefore, the recording of the length of the lightning begins at the first sample of the detected signal and ends at the last sample of the recorded signal.

The measured durations are usually in the order of hundreds of milliseconds, with shorter lightning events being rare. The median duration of lightning events is 524 ms. This result is significantly longer than values reported previous studies: 200–300 ms and 350 ms .

The processing of lightning recordings differs for day and night storm observations, as they require different high-speed camera settings. In particular, the shutter speed (exposition time) and the analog gain settings vary. This difference could affect the exact measurements of the total length of a discharge, resulting in a shortening of the estimated duration. As sunlight is scattered in the atmosphere during the daytime, weak discharges might have been omitted. Therefore, the extracted durations of lightning events could be underestimated.

From the following analysis of the lightning duration, we excluded two thunderstorms during which it was not possible to distinguish between individual lightning events (events for which there was less than 500 ms between the individual detected discharges). We considered lightning to be an “event” when the time between the individual discharges did not exceed 100 ms.

Data from the cameras reveal similar phases of lightning development. We provide examples of video recordings with the individual phases clearly visible and unobscured by clouds.

Figure  captures one frame from a video (https://doi.org/10.5446/60390, ) that shows (from T=+0.27 s to T=+0.40 s; https://av.tib.eu/media/60390#t=1:11,1:22, last access: 1 January 2023) the positive side of a leader. When the current flowing through the leader begins to weaken, recoil leaders start to appear at its end (T=+0.38 s; https://av.tib.eu/media/60390#t=1:21,1:30, last access: 1 January 2023). The term “recoil leader” was taken from the literature . Based on our observations, however, we cannot confirm that all the visible recoil leaders reuse a pre-established ionized channel.

Figure  captures one frame from a video (https://doi.org/10.5446/60388, ) that shows an invisible positive leader, but only blurred recoil leaders are visible. From T=+0.07 s (https://av.tib.eu/media/60388#t=1:36,1:44, last access: 1 January 2023), a negative side of a leader is visible. The negative leader branches abundantly, and its propagation is faster than the positive leader and contains hot luminous ends. The second visible negative leader starts at T=+0.25 s (https://av.tib.eu/media/60388#t=1:51,1:59, last access: 1 January 2023). The negative leaders do not generate recoil leaders.

Based on the recordings, in which some parts of the lightning are clearly visible, we can characterize other lightning events. In Fig. , an example of the luminosity curve and antenna data is shown. At the very beginning, the lightning usually starts with a faint peak with a fast-rising edge (at the time 0.2 s). Then there is an optically dark phase with low luminosity. However, this quiet period is not really quiet in the radio signal; it is a phase in which a leader develops. As the leader in the cloud connects to more and more charged regions, it becomes brighter (sometimes slowly, sometimes abruptly). During and after a decrease in the current in the positive leader, the recoil leaders appear (after 0.4 s). Some recoil leaders have high luminosity when they connect to the main channel. In the radio spectrum, we simultaneously observe peaks that correspond to short intense brightenings. In some cases the lightning results in CG discharge, which is accompanied by a high-intensity flash (at 0.5 s). However, this phenomenon does not occur very often. Using the described equipment, we registered CG return strokes in less than 10 % of cases.