The multiple lightning from the two towers occurred at 0602:05 UTC 27 September 2012. Figure 2 shows the time-integrated high-speed camera image starting from the leader inception until the end of the lightning. Abundant recoil leaders appeared during the development of both upward leaders from these two towers, indicating that both were of positive polarity, which was also supported by the measured atmospheric electric field. The lightning lowered negative charge from cloud to ground, without any subsequent dart leader-return strokes after the current cutoff of the initial continuous currents at both towers. Since the lightning occurred in the daytime with relatively strong background light and the upward leaders exhibited weak luminosity in the very initial stage, it was not possible to determine the exact start time of both upward leaders, with the initiation order of them being unknown. Nevertheless, based on the channel lengths when the leaders were distinguishable, the time difference between the initiations of these two upward positive leaders was estimated to be within ±4 ms under an assumption of a 2D propagation speed of about 10⁵ m s^-1 of both leaders ( Jiang et al., 2013). By checking the electric field data, this multiple tower lightning was found to be self-initiated, without any preceding occurrence of CG or inter-cloud (IC) lightning flashes in the nearby area, and it did not induce any adjacent discharge processes.

Figure 2

	Figure Option ViewDownloadNew Window
	Figure 2 Time-integrated high-speed image from the leader inception until the end of the lightning.

Lu et al. (2009) observed that a tower-initiated flash with an upward positive leader triggered another tower-initiated flash with a negative upward leader. The horizontal distance between the towers was 375 m in their study, which was much closer than the distance involved in this paper. The discharge with the triggering leader lasted about 130 ms and the triggered one with the opposite polarity lasted only 14 ms. However, the durations of both discharges in this paper were longer than 250 ms. In Lu et al. (2009), the initiation of the triggered leader lagged 45.8 ms behind the triggering leader-considerably longer than in the present study. Since the upward leaders in this study were of the same polarity, there would have been no triggering relationship between them, for the occurrence of an upward leader may theoretically weaken the ambient electric field, and, consequently, produce a negative effect on the initiation of another leader with the same polarity. This is consistent with the relatively short time difference studied here. As in Warner (2012), four upward leaders in response to the approaching negative breakdown of a +CG flash were also found to be of the same polarity, which initiated almost simultaneously. It is worth noting that their case was other- triggered, while the multiple lightning in this study was self-initiated. In the dataset of Warner et al. (2012), the longest time difference between the first and last leader was 10.2 ms.

As shown in Fig. 2, abundant recoil leaders occurred during the lightning development. These recoil leaders transiently started in the remnant channel sections (with the luminosity below the lower detection limit of the optical observation), which were generated by the upward leader and propagated toward the trunk channel flowing with continuous current. Most of the recoil leaders died out before reaching the trunk channel and exhibited very short durations, while some of them succeeded in the junction and continued to propagate down through the already luminous channel. Figure 3 shows the evolution of a recoil leader during the lightning discharge of the old television tower. The recoil leader first appeared at 0 ms, and extended to the right, toward the main channel with a faint luminosity that indicated a continuous current. At 0.2 ms, the recoil leader had connected to the main channel, and then it continued to propagate downward, during which period the upper extremity of the leader channel exhibited a reduced luminosity. The downward portion of the leader reached the tower tip between 0.3 ms and 0.4 ms, and induced a reflection that traversed back though the channel to the upper extremity. The reflection wave finally realized an outward channel extension, as shown in Fig. 3f. Different from the observation of Warner et al. (2012) showing a bidirectional development of the recoil leader Fig. 6 in Warner et al. 2012), the image frames in this study demonstrated a unidirectional propagation, and the tailing end of the leader channel exhibited a weakening luminosity before the reflection at the tower tip.

Figure 3

	Figure Option ViewDownloadNew Window
	Figure 3 Evolution of a recoil leader initiated in the channel remnant. The recoil leader propagated toward the main discharge channel and succeeded in the connection. It continued to develop downward and induced a reflection when reaching the tower tip.

Figure 4a shows the frequency of recoil leaders during the period of upward lightning, with the number of recoil leaders having been counted every 5 ms. Twenty nine and 317 recoil leaders occurring in the discharge channels of the meteorology tower and the old television tower, respectively, were captured with the high-speed camera. The recoil leaders firstly occurred at about 10 ms after the initiation of the lightning, exhibiting an increasing frequency as the lightning developed, and reached the frequency peak of 26 recoil leaders per 5 ms at about 75 ms. It is interesting that although the lightning discharges from both towers initiated almost simultaneously, their associated recoil leaders were not similar, with the recoil leaders over the old television tower being greater than those over the meteorology tower. The reason for this phenomenon is not clear. It may be partly due to the lightning channels of the meteorology tower not being totally captured, as it was nearer to the camera. It also seems logical to attribute such a result to the different tower heights, as the meteorology tower is two times higher than the old television tower. Nevertheless, no concrete evidence has been found to support this inference.

The recoil leader shown in Fig. 3 lasted more than 0.6 ms, which is a considerably long duration among the recoil leaders in this flash. Most of the time, the recoil leaders were only captured by one to two frames, indicating they were short in duration (within 0.3 ms, theoretically). Figure 4b shows the numbers of recoil leaders with different durations. For simplicity, the duration of 0.1 ms was considered to correspond to those recoil leaders captured by only one frame (the camera was operated at a speed of 10 000 fps), and so forth. Although such an approximation is not sufficiently precise, the figure nevertheless roughly demonstrates an exponential declining trend in recoil leader occurrence versus duration.

Figure 4

	Figure Option ViewDownloadNew Window
	Figure 4 Statistical characteristics of recoil leaders: (a) occurrence frequency during the period of lightning, counted every 5 ms; (b) histogram of recoil leaders with different durations.

A thunderstorm's evolution and the associated charge structure may exert important effects on the lightning activity ( Qie et al., 2005). Figure 5 shows a vertical cross section of the radar echo reflectivity when the concurrent upward lightning flashes occurred. The area of the towers was controlled by the thundercloud with a radar echo intensity of no more than 45 dBZ, indicating non-severe convective activity at that moment. Overall, two strong radar echo centers of the whole thunder system were both located more than 50 km away from the towers, with a maximum intensity of 60 dBZ. Although the radar echo over the tower area was not very strong, and the associated thundercloud covering the towers exhibited inactive lightning activities, tower-initiated lightning was relatively frequent. Besides the concurrent flashes studied in this paper, five other upward lightning flashes initiated from the meteorology tower during a short period of about 20 minutes. As shown in Fig. 5, the height of the strong radar echo center was only about 2-3 km. Consequently, as the cloud was carrying some negative charge, the associated charge center may also have been relatively low. This could have been the cause of the upward lightning from the towers without any preceding nearby lightning discharge.

It is generally considered that a tower with an effective height of lower than 100 m only involve downward lightning; and as tower height increases, the probability of initiating upward lightning increases, by up to 100% if the tower is taller than 500 m ( Rakov and Uman, 2003). Based on this understanding, the old television tower, with its height of 112 m, would barely initiate upward lightning. The observations in this study indicated that the charge structure of the thunderstorm with its radar echo reflectivity center being very low is conducive to the occurrence of upward lightning from towers or grounded tall structures. As indicated in Fig. 5, the towers were separately underneath the two radar echo centers, which were not far from each other, and the region above the old television tower exhibited a larger area of relatively strong echo intensity than above the meteorology tower. This may be a possible reason for the large disparity in the number of recoil leaders over these two towers.