National Key Laboratory on Electromagnetic Environmental Effects and Electro-Optical Engineering, PLA University of Science and Technology, Nanjing, China
Corresponding author: S. Qiu, National Key Laboratory on Electromagnetic Environmental Effects and Electro-Optical Engineering, PLA University of Science and Technology, Hai Fuxiang Street No. 1, Nanjing 210007, China. (email@example.com)
 A single-station-based lightning discharge channel reconstruction system by combining a two-dimensional (2D) VHF broadband interferometer and a three-dimensional (3D) acoustic lighting mapping system has been developed and used for lightning observations. Two cloud-to-ground (CG) flashes with highly branched leaders recorded by the system are analyzed and presented in this paper. VHF radiation could well delineate the development of simultaneous leader branches, while acoustic emissions mainly located on the main channel which was traversed by return stroke (RS) process. Localizations by VHF and acoustic emissions agree well with each other. The mapping results confirm that audible acoustic emission of lightning discharge is mainly associated with high current process like RS. Leaders could generate detectable acoustic signals, with amplitude at least an order weaker than ensuing RS, but they are hard to identify except in closer ranges than the main channel. As a significant phenomenon, this paper provides the first 3D locations associated with sources of tearing sounds, which are inferred to be generated by downward negative leaders when they approach ground. The synchronized observation enable VHF interferometer locate lightning development in spatially quasi 3D, and three stepped leaders, five dart leaders and two dart-stepped leaders are identified, with the 3D velocity (1.3–3.9) × 105 m/s, (1.0–2.9) × 107 m/s and from (1.0–1.3) × 107 m/s to (2.4–2.6) × 106 m/s, respectively. In addition, the application of this approach in improving the accuracy of thunder ranging is discussed.
 Lightning discharges produce both radio frequency (RF) and acoustic emissions. Using spatially separated receivers, lightning channel could be reconstructed from both VHF and acoustic domains [Oetzel and Pierce, 1969; Few, 1970; Zhang et al., 2012]. From a physical perspective, VHF radiation is produced by lightning breakdown processes. If the receiver could record VHF signals with sufficient time resolution, a sequence of time-resolved RF pictures of lightning processes could be reproduced. In the past decades, various VHF mapping systems, such as VHF interferometer and Lightning Mapping Array (LMA), have been developed, and basic feature and evolvement of both cloud-to-ground (CG) and intracloud (IC) lightning have been revealed [Proctor, 1971; Rhodes et al., 1994; Shao and Krehbiel, 1996; Rison et al., 1999; Dong et al., 2001; Liu et al., 2012]. However, VHF technique has its inherent drawback either. For example, it is not suitable to locate high current events (such as return strokes). In contrast, audible thunder (about 20 Hz to 20 kHz) is generally associated with rapidly heated lightning channels, so thunder imaging could be used to locate high current events; moreover, it is suggested that not only return strokes (RS) but also other impulsive currents, such as in a leader step or a K-process could emit acoustic signals [Uman, 2001]. Therefore, thunder imaging may be used to complement VHF results to depict a more comprehensive lighting picture.
 However, more attention was given to VHF mapping technique [Proctor, 1971; Shao et al., 1996; Rison et al., 1999; Yoshida et al., 2010] in the past, and there were few reports in the literature about the two synchronized observational results. Few and Teer  once conceived a comparison between acoustic and VHF observations, but it didn't put in practice due to the tedious data processing work in VHF techniques in that period [Proctor, 1971]. Recently, with the development of VHF techniques, a joint observation with acoustic localization re-aroused much concern.Arechiga et al. reported the acoustic localization of triggered lightning using low-frequency (3.3 Hz–500 Hz) acoustic arrays, and the accuracy was validated by comparing with LMA locations.Johnson et al.  further explored thunder imaging techniques of both infrasound and audible thunder using array coherence mapping, and discussed some misfit with LMA results.
 In this work, a compact synchronized observation system combined VHF broadband interferometer with broadband acoustic arrays is designed, and preliminary observations are conducted with the system. Two widely branched natural CG flashes are presented, and discrepancies from the two mapping techniques are analyzed. An initial motivation of the work is to investigate different VHF and acoustic behaviors in lightning processes and study the physical mechanism involved in the generation of acoustic and VHF emissions. Since VHF interferometer and acoustic imaging techniques in this system operate with the same baseline geometry configuration and the similar mapping manner, the comparison between the two imaging results would be explicit. Furthermore, tearing sound in close thunder signature is analyzed, whose sources are first located by the synchronized observation system and hence the mechanism of tearing sound is also inferred. In addition, two applications of the synchronous observations are introduced, which would be used to improve the capabilities of both VHF interferometer and thunder ranging.
2. Instrumentation and Methodology
 The proposed synchronized short-baseline VHF and acoustic lighting mapping system is illustrated inFigures 1 and 2. A perfectly matched VHF and acoustic receiver array configuration is deployed, which consists of three horizontally separated units. The three units aligned at three apexes of a square with the baseline's length of 12.5 m. Each unit is composed of a VHF broadband omnidirectional antenna and a high-gain broadband capacitor microphone. Broadband VHF radiations (30 MHz–300 MHz) are filtered to remove FM interference (88 MHz–108 MHz) and then digitized at a rate of 1GS/s with 8 bit resolution by a 4-channel LeCroy digital storage oscilloscope (DSO). The DSO operates in sequence mode, which divides the whole memory into a number of adjustable segments and the dead time between segments is less than 1 μs. The circuit of microphone is specially designed for receiving thunder signals, with a wide dynamic range and a broad band from several hertz to about 20 kHz. Three microphones and an electric field (E-field) change meter with decay time constant of 1 ms (also referred as fast antenna [Kitagawa and Brook, 1960]) are connected to a PCI data acquisition card, with sampling rate at 100 KS/s and duration of 30 s. The trigger signal is generated by the DSO when the VHF radiation exceeds the threshold, which is then sent to the external trigger channel of the PCI card to start an acquisition. A PC-based program is designed to ensure all data are transferred into the hard disk before a new data acquisition, so all records are synchronized in time.
 In order to reconstruct lightning channel from VHF and acoustic emissions, broadband interferometry and acoustic ray tracing techniques are employed, which will be described briefly in this section.
 Suppose A and B in Figure 3 are two VHF or acoustic receivers in Figure 1, with the baseline length d, the incident angle (θ) of the distant VHF or acoustic plane wave could be resolved by equations (1) and (2), respectively:
where c and v represent the speed of light and sound in normal atmosphere, respectively, f is the available frequency component in VHF range, Δφ(f) is the corresponding phase difference spectrum, and Δt is the differential time of arrival (DTOA) of thunder wave between the two acoustic sensors. Δφ(f) and Δt could be obtain by Fourier transform by a couple of VHF radiations[Dong et al., 2001] and cross-correlation between acoustic signals [Few, 1970], respectively. In order to suppress random phase interference, the phase filtering algorithm [Qiu et al., 2009] based on translation- invariant wavelet denoising is employed.
 Since a single pair of receivers could only provide incident angle (θ) in one dimension, using an additional receiver to constitute a perpendicularity baseline, the azimuth and elevation (as denoted Az and El in Figure 1) could be resolved as follows:
 From the above description, it is explicit that procedures in resolving the incident angle by both VHF interferometry and acoustic ray tracing techniques are similar. The major difference in signal processing is that VHF broadband interferometry examines the phase difference spectrum in frequency domain (equation (1)) and the acoustic ray tracing technique directly calculates the time difference by cross-correlation (equation (2)). In addition, acoustic ray tracing technique could provide the range of the source by measuring the DTOA between thunder signals and corresponding E-field changes. Since the same geometry configuration is used in both systems, localization discrepancy arisen from array installation would be minimized. In order to exclude electromagnetic interference (EMI) effectively, VHF records with magnitude exceeding certain threshold are used in localization, and the threshold is selected according to the background noise. Shifted time window of 100 ms with 95 ms window overlapped is used in cross-correlation calculations of acoustic signal processing. In practice, the retrieved thunder sources are classified by their correlation coefficients, and sources with very low coefficients are excluded.
3. Observations and Results
 The synchronized observation using the system illustrated above was first conducted during the summer of 2011 in the city of Nanjing, China. Receiver arrays were located at the top platform on a 6-floor building, with a relatively height of 20 m or so, which was assigned as the origin point of the mapping locations in the following results. Two normal multistroke CG flashes will be presented in this section. In this paper, theE-field changes follow atmospheric electricity sign convention.
3.1. Negative CG 075744
 The first CG 075744 occurred on July 26, 2011. It was a negative CG with three intensive leader-return strokes (labeled ‘L1–L3’ and ‘RS1–RS3’ inFigure 4a). The ground contact of RS2 and RS3 was about 1.5 km southeast apart from that of RS1. The flash originated overhead of the site, with only four VHF segments triggered in the first 5.5 ms. Both the elevation of these VHF source and E-field changes decreased with time, as denoted by arrows inFigures 4b and 4c, indicating a downward negative breakdown in the initial stage. Thereafter, leader L1 developed northwest, and progressed to ground in about 11.5 ms with considerable branches. The final part of L1 divided into three distinct branches, and only the middle branch contacted ground, which was supported by several scattered up-moving VHF sources in RS1. About 44 ms later, widely branched leaders L2 initiated and break a new path to ground in nearly opposite direction with L1 (shown inFigures 4d and 4e). About 30 ms after, leader L3 traversed the main channel of the L2-RS2 process rapidly, and initiated the ensuing RS3 (shown inFigures 4f and 4g).Thunder sources marked with black circles are overlaid on VHF results in Figures 4h and 4i and shown in 3D format in Figure 5b. Heights of some selected thunder sources are also labeled in Figure 4h. The thunder sources extended to 8 km in height, and had obvious horizontal components in the upper portion. Rare VHF sources were located at these regions. In the lower portion, thunder sources showed two distinct channels (labeled ‘T1’ and ‘T2’ in Figures 4h, 4i, and 5b), which were well matched with main channels established by L1-RS1 and L2-RS2 in VHF results, with only slightly shifts (less than 3.5°, corresponding to 92 m or so in distances) between the two mapping results. Pronounced ‘clap’ sounds appeared in the first several seconds of thunder signatures, and the first clap was characterized by the ‘N’ type pressure pulse. Thunder sources of claps came from nearly vertical channels below 1 km. The two ground terminations T1 and T2 were estimated at 657 m northeast and 872 m southwest from the site, respectively. Note that sources in T1 channel only located below 0.42 km, because acoustic emission from the T2 channel arrived at the observation site and dominated the acoustic signal after a fraction of a second.
 An interesting phenomenon in this thunder signal was that there were some weak acoustic emissions prior to the start of the loud claps. The weak acoustic signal lasted about 0.4 s and the sounds were like tearing of clothes. We refer to this weak acoustic signal as ‘tearing sounds’ and label it in Figure 5a. The tearing sounds were about an order weaker than the ensuing initial ‘N’ type pressure pulse, and the signal coherence between different microphones was low. Fortunately, we could retrieve the strongest acoustic pulse in the tearing sounds, which was about 0.1 ms before the first clap, with the location clustering around point ‘P’ in Figures 4h and 5b. The location of point ‘P’ was under the ensuing clap sources, with a height of about 120 m, and its projection on x-y plane was about 20 m closer to observation site than other closest sources.
3.2. Negative CG 140349
 The second negative CG 140349 occurred on August 13, 2011. It consisted of 7 strokes with the entire dart leaders following the same main channel established in the initial leader stage. The ground termination of this CG lay about 2.56 km southwest. Figure 6 shows composite views of this flash. The flash originated at southwest of the observation site and then underwent the preliminary breakdown (PB) stage which lasted about 35 ms (marked in Figures 6a and 6c). Note that PB stage here refers to the in-cloud process that initiates the initiation of the downward-moving stepped leader; see also inRakov and Uman . The fact that no characteristic pulse [Nag and Rakov, 2008] could be recognized in E-field changes of PB stage (seeFigure 6a) is attributed to the low sensitivity of this fast antenna. Three branches labeled B1–B3 progressed simultaneously after PB stage. Branch B1 went all the way to ground, while B2 and B3 were highly branched in propagation. Finally branch B1 successfully attached ground and initiated the first RS, which could be confirmed by rapid up-moving VHF sources after RS onset. All leaders thereafter traversed channel B1, which suggested that all RS follow the same major channel. The third leader-RS process is selected for an example inFigures 6b and 6d.
 Similar to CG 075744, thunder sources of this flash are also overlaid on VHF results in Figures 6c and 6d. Most of the thunder sources located along the branch B1, which corresponded to the main channel of this flash, while almost no thunder sources located on B2 and B3 branches. Thunder sources below 3 km show excellent agreement with VHF locations; while in the upper portion, minor discrepancy exist within the range of 3° in angles and about 300 m in distances. Note that no similar tearing sounds could be discerned in the initial portion of the acoustic signal, as shown in Figure 7.
4.1. Physical Mechanisms of VHF and Acoustic Emissions
 Lightning channel could be well reconstructed by both VHF and acoustic techniques, and the results show well agreement with each other; moreover, since the two CG flashes presented here both behave highly branched leaders, they are quite suitable to illustrate the different physical mechanisms involved in both mapping results.
 In most natural negative CG flashes, downward leader phase consists of appreciable branches. Generally, only one branch contacts ground and establishes the main channel in the ensuing RS. In the presented two CG flashes, VHF radiation could well delineate the development of simultaneous branches, while acoustic emissions mainly located on the channel which was traversed by RS process. These observational results indicate that audible acoustic emissions are primarily generated by RS in CG flashes, which has a different mechanism with VHF radiations. In particular, the lower vertical RS channels could generate pronounced clap sounds in close ranges, which is consistent with the explanation by Few that sound intensity could be enhanced by perpendicular channel segments. Isolated thunder sources scattered in branched channels which did not attach ground were probably generated by the neutralization of charges when RS wave arrived at the branch. Besides, in each flashes, there were a number of well-organized thunder sources in upper portion of the channel, which were invisible by VHF interferometer. These regions often located beyond the origin of the flash identified by VHF interferometer and connected with the main channel. We speculate that initial breakdown processes must have existed in these regions, but they were so weak for interferometer to locate. These breakdown processes may arise from weak negative PB stages in the cloud whose VHF radiations did not exceed the triggering threshold, or they were positive leaders. The fact that they could be located by thunder imaging is probably attributed to the ensuing RS wave arrival at these channels.
 It is an interesting question that whether leaders produce detectable acoustic signals. The discussion in section 4.2 below indicates that leaders also generate acoustic emission, but the amplitude is about an order weaker than RS, or even smaller. Although leader processes generally lead RS about tens of milliseconds, it is hard to identify their signals due to the interference by acoustic signals of other RS segments whose acoustic emissions arrive at the same time with the leader. One exception is that leader acoustic signals arrive earlier than the first arrived RS acoustic emission, which means branched leader processes could be located if these branches are significantly closer to observation site than the minimum distance from main channel to observation site.
Johnson et al. recently proposed a coherence map to depict 4 Hz–40 Hz thunder regions, and they found higher frequency sound (e.g.,>10 Hz) was less good and tended to be poorly correlated with LMA RF sources. Spectrum analysis of the received thunder in this work indicates that they are mostly audible sound which ranges from several hertz to several kilohertz. Therefore, higher frequency sound (>10 Hz) is also well correlated with RF sources. The fact that some thunder sources differ slightly from corresponding VHF channels may be caused by the presence of atmospheric turbulence, wind shears and temperature gradients, which produce curved-ray acoustic path [Few and Teer, 1974]. In addition, located thunder sources usually clustered in groups like “string of pearls” (such as Figure 6c), which could be attributed to the interference effect of several sound radiators.
4.2. Tearing Sounds
 Tearing sounds are the unique acoustic signal when lightning strikes a few hundred meters away [Uman, 2001]. Although many observers near CG flashes have reported these common sounds prior to the first loud clap, no detailed measurements have been reported to the best of our knowledge. The origin of tearing sound has been attributed (1) to a straight channel section of length of the same order as the distance to the observer [Hill, 1977] and (2) to a multitude of upward-going connecting discharges from earth [Malan, 1963].
 Observations presented here and also in other cases in our experiment confirm that these tearing sounds could be heard mostly in several hundred meters, consistent with previous observers. For example, tearing sounds were recorded in close CG 075744 (shown in Figure 5a), but they were obscure in CG 140349 (shown in Figure 7). Furthermore, as advancement in this work, source locations of tearing sounds in close CG lightning are retrieved (as denoted by point ‘P’ in Figures 4h and 5b). Considering the fact that the amplitude of tearing sounds was about an order weaker than the ensuing loud clap, the origin of these sounds should be attributed to other mechanisms (such as leaders) than the intensive return strokes. Since sources of tearing sounds in Figure 4h located at about 120 m in height, and this location corresponded to the downward negative leader process identified by VHF results, it was inferred that tearing sounds were probably generated by downward negative leaders. The duration of tearing sounds lasted in a fraction of a second before claps, which was primarily because some spatially separated leader branches were closer to observation site than the main channel in hundreds of meters, consistent with the discussion in section 4.1.
Colgate and McKee  once analyzed the sound formation mechanism from stepped leader process, and predicted that weak acoustic pulse could be produced by the electrostatic stress from stepped leaders. The electrostatic sound pulse was estimated roughly 1/300 of the subsequent main stroke. They noted that this weak pulse would be observed before the arrival of acoustic signal from return strokes. Our observation confirmed their theory, but the amplitude ratio of the weak acoustic pulse versus the ensuing loud claps was in the order of 1/10, much higher than the prediction made by Colgate and McKee . The discrepancy could be attributed to the fact that the current and charge density they used were much lower than those when the leader approaches ground. As summarized by Rakov and Uman , the mean leader current and charge density near ground were 1.3 kA and 3.4 C/km, respectively, while these values were 125A and 0.33C/km in the prediction model by Colgate and McKee . The higher current and charge density would result in the stronger acoustic emission. Taking into account of this difference, our observational results are reasonable compared with the model prediction.
Arechiga et al. once made a first-order estimation of the expected thunder signal using LMA sources. They assumed that each located LMA source emitted an acoustic signal when it occurred. Using the LMA source position, occurrence time and the propagating speed of thunder (340 m/s), the acoustic waveform received at the observational site could be reproduced. The estimated acoustic waveforms were compared with the measured records. However, the two results did not resemble each other. Considering LMA sources are most generated by leader processes, and the acoustic pulse of leader is about 10 times smaller than that from RS, it is evident that the estimated acoustic waveforms byArechiga et al.  only represent the sound of leaders, but not the sound of the whole flash. Therefore, the estimated acoustic waveform differed from the measured one.
4.3. Further Applications
 As stated above, since different mechanisms are involved in VHF and acoustic mapping results, a combination of two techniques could complement each other, and give a more comprehensive lightning picture, which could reflect both breakdown and high current processes. Two applications of the synchronous observations are further discussed in this section.
 One application is the realization of quasi 3D VHF localization using single station. In some extent, single-station-based lightning image by acoustic arrays is like a stationary 3D picture, which could only depict the stationary frame of 3D lightning channels; while that by VHF interferometer is like a 2D video, which could provide the time-resolved localization of lightning breakdown processes. Though leader propagation does not generate significant thunder, it is responsible for forming an ionized channel that subsequently carries the predominant current events such as return strokes. Through a combination of the two mapping results, VHF 2D results having the same azimuth and elevation with acoustic mapping results will add the ‘range information’ (shown inFigures 4d and 5d), which would upgrade single station VHF interferometer results from 2D to quasi 3D and give a better understanding of VHF mapping results. It is worth noting that the term ‘quasi 3D’ means some of the points in VHF mapping results would have 3D positions. If the located VHF or thunder channels themselves overlaid in the azimuth and elevation format, it is hard to match VHF and corresponding thunder sources, and hence 3D VHF localization is not available. Usually, 3D VHF localization of the downward leader portion of CG flashes is easily realized due to its spatial structures.
 Using the quasi 3D VHF localization, characteristics of leader development could be described by single station detection. In the presented two CG flashes, three stepped leaders and five dart leaders were identified. The three stepped leaders, as illustrated in Figures 4 and 6, leaders L1, L2 in CG 075744 and leader L1 in CG 140349 lasted 16.5 ms,13.8 ms and 18.5 ms, respectively, with 3D velocity in the range of (1.3–3.9) × 105 m/s during the final several kilometers. The 3D velocity of five dart leaders L3–L7 in CG 140349 ranged from (1.0–2.9) × 107 m/s. In addition, L3 in CG 075744 and leader L2 in CG 140349 were considered to be dart-stepped leaders, which behaved as obvious velocity decreases from (1.0–1.3) × 107 m/s to (2.4–2.6) × 106 m/s as they progressed to ground. These calculated parameters of the three type leaders are coincided with previous studies determined by optical and electric approaches [Schonland, 1956; Krider et al., 1977; Orville and Idone, 1982; Jordan et al., 1992; Rakov and Uman, 1990; Wang et al., 1999; Davis, 1999; Biagi et al., 2010; Hill et al., 2011].
 Another application is the improvement in thunder ranging. Traditionally, in the process of determining DTOA between thunder and E-field changes (or optical signals, etc.), it is impossible to correlate the specificE-field waveform (or optical signals, etc.) with its thunder signal (the lightning discharge is considered to be instantaneous), so the time difference usually has an uncertainty of about 1 s, which would give rise to a range errors about 350 m in acoustic localizations. Since VHF interferometer could correlate the mapping results with the specific lightning process, it bridges a gap between thunder and corresponding electric signals and thus eliminates the uncertain time, which is especially useful for localizing striking points of multistroke CG flashes.
 Synchronized VHF broadband interferometer and acoustic mapping results are first reported in this paper. Since same array configuration and similar principle in source localization is employed in both techniques, comparisons between two results are more distinctive.
 Mapping results by VHF and acoustic emissions agree well with each other; furthermore, discrepancies of the two results are also illustrated. VHF subsystems could locate appreciable leader branches, while thunder locations are concentrated on the main channel, which confirms the differences in physical mechanisms that VHF radiation is generated by breakdown processes and audible thunder is produced by rapidly heated channels. Leaders could generate detectable acoustic signals, but they are hard to identify except that they are in closer ranges than the main channel. As a significant phenomenon, tearing sounds are specially analyzed, and it is inferred to be generated by downward negative leaders when they approach ground. Finally, two applications of the synchronized observation technique are discussed. One application is for upgrading VHF interferometer results from 2D to quasi 3D, and hence the 3D properties of lightning development could be obtained by a single station. Three types of leader processes are identified by the analysis. The calculated 3D velocity for the stepped leader, dart leader and dart-stepped leader are (1.3–3.9) × 105 m/s, (1.0–2.9) × 107 m/s and from (1.0–1.3) × 107 m/s to (2.4–2.6) × 106 m/s, respectively. Another application is for improving the ability of thunder ranging, which could help to eliminate the uncertain time in DTOA estimations.
 Since the observation reported here is conducted in the city, high EMI is suffered in VHF observations; in addition, only limited results are obtained in preliminary observations. Efforts would be made to improve the two mapping techniques and to accumulate more data to reveal fine structure differences in both CG and IC in the future.
 This research was supported by National Natural Science Foundation of China (grant 60971063) and NCET under grant 07-0383.