Corresponding author: M. Stolzenburg, Department of Physics and Astronomy, University of Mississippi, PO Box 1848, University, MS 38677, USA. (email@example.com)
 Correlated data for natural cloud-to-ground lightning flashes were obtained with a high-speed video camera (54000 frames per second), five stations of electric field change antennas, and a seven-station VLF/LF LIghtning NETwork (LINET), along with operational lightning and electric field networks at Kennedy Space Center, Florida. Four observed flashes at 25–55 km distance had return-stroke-like upward illumination (UI) from the ground just after (within 2 ms of) a return stroke (RS), but in a separate channel connection to ground (0.7 to 2.5 km apart). The UI events were visible for short duration (0.5 ms or less) and short length (to 1.7 km altitude or less), and they had slow propagation speed (1.6–2.6 × 107 m s−1) and no apparent connection to the RS channel. In one case, the RS was no longer visible to the ground when the UI began 1.30 ms later, although mid-level channel was still distinct. In another case, the main RS had dimmed over the 0.52 ms interval before the UI but was still visible. The remaining two cases had the entire length of the prior RS channel no longer visible (for 40 and 240 μs) before the UI began (1.32 and 1.87 ms later). Electric field change data for the UI events resemble those of weak return strokes. These events were not located by the operational ground-strike and VHF detection systems; all four UI events were detected by LINET as negative ground strokes with small (less than 7 kA) peak current magnitude.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Several prior studies have also established that a large fraction (up to 51%) of multistroke CG lightning flashes produce more than one channel to ground [e.g., Kitagawa et al., 1962; Rakov et al., 1994; Willett et al., 1995; Valine and Krider, 2002; Qie et al., 2005; Saba et al., 2006]. One particular class of multichannel flashes has been described by Guo and Krider as “double-ground” in which the two terminations are separated by several tens of microseconds.Willett et al. discuss three double-ground cases, separated in time by only 6.7, 8.4, and 9.9 μs, among the 32 natural CG flashes in their study. Using wideband electric field recordings and standard video data (30 frames per second), Willett et al. determined that both terminations in the double-ground cases radiate first-stroke-type field and time derivative (dE/dt) waveforms. Their optical data (available for two of the three cases) also show the two double-ground terminations were visible at the same time and were not connected below cloud base.Rakov and Uman , using wideband electric field and television video (17 ms frame exposure) data, found much longer time separations (287–596 μs) between the double terminations in four of their eight events for which the double-field waveforms could be distinguished. For seven strokes with multiple terminations,Thottappillil et al.  give separation distances of 0 to 5 km between ground points.
 More recently, Ballarotti et al.  have data for six examples of “forked lightning” captured with high speed video and fast electric field recordings (video time resolution and exposure of 1 ms). The time separation between the two strike points was as small as 31 μs but in all cases was less than 2 ms. Their data indicate that both terminations of the channel were connected to ground simultaneously, for 1 to 5 ms. The forks illuminate with two contact points in the same or sequential video frames, and the later termination appears while the first one is still illuminated. Like Guo and Krider , Ballarotti et al. suggest that the double-ground connections may result from competing branches of a single stepped leader reaching the ground at nearly the same time. The discussion byRakov and Uman , involving probable RS propagation speeds and height of visible branch points, describes why this is a reasonable explanation only if the time separation between events is less than about 100 μs in their Florida data. Rakov and Uman hypothesize that their double-ground events (with 287–596 μs between them) are “two sequential return strokes of the same flash, each initiated by its own leader, rather than by the double branching of a single leader.”
 This study focuses on four events we term “upward illuminations” (UIs) recorded in high- speed video and multistation electric field change measurements during a 2010 field campaign at NASA Kennedy Space Center (KSC), Florida. As we will show, these UI events appear in the electric field change data as very weak strokes occurring less than 1.9 ms after a normal RS of a natural, negative cloud-to-ground lightning flash. We herein define a UI based on the high speed video data (recorded at 54,000 frames per s and 18.1 μs exposure time) as having the following characteristics: (a) a new ground connection, within 2 ms of a normal RS, that follows a different channel to ground from any prior strokes in the flash; (b) no apparent visible connection to the preceding or concurrent RS channel; (c) only the lowest channel portion, 1–3 km above ground, illuminates, and the luminosity moves upward, as with a return stroke but without visible channel at higher altitudes; (d) the main part of the UI's upward path follows a stepped leader or branch that was moving toward the ground simultaneously with the successful leader (i.e., the one that results in the return stroke). Later we will discuss whether the UIs should be deemed weak return strokes. Here we mention only that the UIs differ from typical return strokes because the upward return current appears to ‘return’ only a short distance up the path ionized by the preceding downward stepped leaders. The UI events may be the same type of event previously studied by Rakov and Uman , Ballarotti et al. , Saba et al. [2006, Figure 10], and others. While the time resolution of our optical data is at least 50 times faster than in the aforementioned studies, it still is too slow to allow us to resolve the Willett et al.  or Guo and Krider  type events. With our coincident electric field change and video data of individual leaders and return strokes, the goal of this study is to describe the physical characteristics of the UI phenomenon.
 In July and August of 2010, we obtained time-correlated data for natural lightning flashes using a suite of flat-plate electric field change antennas, a VLF/LF LIghtning NETwork (LINET), and three KSC operational networks: Lightning Detection and Ranging 2 (LDAR2), Cloud-to-Ground Lightning Surveillance System (CGLSS), and Electric Field Mills. The flat-plate antennas [e.g.,Kitagawa and Brook, 1960] were installed at five locations on and near KSC, with a separation between sites of 8–34 km (Figure 1). These instruments, discussed in more detail in the following paragraph, measured electric field changes (E) of lightning. A seven-station LINET system was deployed 16 June–15 August 2010 around east-central Florida (Figure 1). As described in Betz et al. [2004, 2008], LINET uses time-of-arrival of the magnetic field change at VLF/LF (5–200 kHz) frequencies to locate and discriminate between in-cloud events and ground strokes. The nine-station LDAR2 lightning mapping system locates impulsive sources in the VHF (60–66 MHz) frequency range;Murphy et al.  describe the capabilities of this recent upgrade to KSC's earlier LDAR network. The CGLSS network [e.g., Wilson et al., 2009] is a VLF/LF system using medium-gain IMProved Accuracy from Combined Technology (IMPACT) sensors like those applied at high-gain in the National Lightning Detection Network (NLDN) [e.g.,Cummins et al., 1998, 2006; Biagi et al., 2007] operated throughout the contiguous U.S. For this study, both CGLSS and NLDN yield data on all RS positions, times, and peak currents for cloud-to-ground flashes detected in the KSC region. Data from the calibrated 29-station Electric Field Mill network at KSC [e.g.,Koshak and Krider, 1989] were also available and were used in this study for calibration of the field change data.
 During most of our 2010 field experiment at KSC (including the days investigated herein, July 28–29), we had five electric field change sensor sites (Figure 1). Three sites (K02, K14, and K24) were located within a few meters of one of the KSC electric field mills. The other two sites were at KSC Weather Station B (WSB) and at the Brevard Community College Planetarium (BCC). At each site we had essentially three flat-plate sensors (called Ch1, Ch2, and Ch3) with different decay time constants, different relative gains, and different bandwidths, as shown inTable 1. The antenna data were time-tagged to GPS pulse-per-second with 1 μs accuracy and calibrated (to within 3% to 10% depending sensor type and location) by comparing to the KSC field mill data. The Ch2 and Ch3 data show relatively slow field changes (1–200 ms, due to events like leader extension and continuing current) and also the fast field changes (1–200 μs, due to events like initial breakdown pulses, stepped leader pulses, or return strokes). The Ch1 data essentially show only the fast changes. We use the physics sign convention for electric field, where positive values at the ground indicate negative charge overhead. Calibrated electric field changes given in the text have not been range normalized.
Table 1. Electric Field Change Sensor Characteristics
Decay Time Constant
1.6 kHz to 630 kHz
0.016 Hz to 2.6 MHz
0.16 Hz to 2.6 MHz
 On July 28–29, 2010, additional time-correlated data were acquired with a Vision Research Phantom V12.1™ high-speed video camera positioned at the WSB electric field change antenna site on KSC (Figure 1). We captured 54,000 optical images per second (18.5 μs image interval), with 18.1 μs image exposure time and 8 mm focal length lens. The V12.1 camera features a CCD array with individual sensor size of 20 μm and is highly suitable for lightning imagery [e.g., Warner et al., 2011]. Optical data were obtained between 1415 and 1620 Local Time on both days, for a total of ten flashes comprised of 33 observed return strokes, at 25–55 km range from the camera site. Times of discrete events presented herein (i.e., return strokes, LDAR2, CGLSS, and LINET sources) have been adjusted for propagation delays to the time these events would have arrived at the camera site (to facilitate comparisons with the camera data).
3. Observations and Analysis
3.1. Hybrid CG Flash on 28 July
 The first flash considered in this study was a negative CG flash that occurred at 1836:21 UT on 28 July 2010. Relative to the camera site, the LINET ground strike location was 9.4 km north and 34.0 km west (35.3 km distant), as depicted in Figure 1. (The CGLSS ground strike point is within 0.2 km of the LINET location.) Several frames from the video data are shown in Figure 2. The video camera was pointing nearly west at the time, so the image plane in Figure 2is approximately south-to-north. At the range of this RS, the full image covers 28.1 km horizontally and 21.1 km vertically (with 88 m pixel resolution). Despite the distance, direction, and daylight conditions, portions of horizontal and vertical leader development were distinctly visible in the video data for 21.2 ms before the RS occurred at 1836:21.53 UT.
 Based on the LDAR2, LINET, and E data shown in Figure 3, the flash initiated at 1836:21.26 just above 6 km altitude. It began developing as an intracloud (IC) flash, with a typical IC electric field change record [e.g., Shao and Krehbiel, 1996] for the first 160 ms or so (Figure 3a). The LDAR2 and LINET sources show the initial negative leader propagation was upward, to about 9.5 km, followed by nearly 200 ms of extensive VHF activity at 6–8 km and intermittently at higher altitudes. After the early IC character, the field change data show a rollover to CG character about 100 ms before the RS. Thus, this flash is a hybrid CG [e.g., Rison et al., 1999; Maggio et al., 2009] in the sense that it initiates as an IC flash but eventually connects to ground. Development of the leader to ground is apparent in the LDAR2 data (Figure 3a) beginning at about 1836:21.51, only 20 ms before the RS.
 Prior to the RS, substantial horizontal leader propagation at altitudes of 4.3–5.1 km (negative leaders, with distinct stepping) and at 6.3–7.5 km (positive leaders, with no apparent stepping) was visible in the optical data for 15.54 ms, beginning about 283 ms after flash initiation. Eventually, nearly 300 ms after initiation, two long branches of a negative stepped-leader were seen to advance downward (Figure 2a); the first of these was visible below the apparent cloud base (1.8 km altitude) for 3.35 ms before the RS. The second, more northerly of these two stepped-leaders was visible below cloud base for 0.78 ms before it connected to ground as the main RS channel at 1836:21.53016 UT (Figure 2b). The main RS luminosity reached its full vertical extent of 6.55 km (Figure 2c) about 204 μs (11 frames) later, propagating mostly upward for five frames, then more horizontally (near 5.4 km altitude) for the next four frames, then upward again. During the first five frames (93 μs) after its start, the approximate two-dimensional speed of the RS luminosity front was 6.6 × 107 m s−1. (Note here and below: our RS and UI speeds are the average value over a substantial vertical extent, 6.1 km in this case. Schonland  showed that branched RSs, such as the ones discussed in this paper, slow down as they propagate upward, from 10 × 107 m s−1 near the ground to 4 × 107 m s−1 near the top of the visible channel. This decline in speed provides a plausible explanation for why the average speed we estimate over a large vertical extent is smaller than the peak RS speeds often quoted, up to 2.4 × 108 m s−1, measured with streak cameras over the lowest 0.3–1.3 km of a RS [e.g., Idone and Orville, 1982]. The speeds we report herein, which are for branched RSs, are in good agreement with the branched-RS speeds measured at KSC byIdone and Orville : namely, 2.9–13 × 107 m s−1 with a mean of 6.6 × 107 m s−1.) Portions of the RS channel were visible in the video for about 3.76 ms after the first frame with the visible ground connection, although the lower part of the channel to ground disappears from view after about 0.8 ms. The CGLSS peak current estimate for the RS was −33.1 kA, the NLDN value was −40.6 kA, and the LINET value was −48.5 kA. (Note: we do not discuss the differences in peak current estimates from the LINET, NLDN, and CGLSS systems. These techniques are described in Betz et al. , Cummins et al. , and Wilson et al. .)
 In this hybrid flash, the upward illumination from the ground was first seen at 1836:21.53146 UT, 1.297 ms after the beginning of the main RS. The bright portion of the UI reached its full (above-horizon) vertical extent of 1.5 km in 93 μs (five frames), and the approximate 2D propagation speed was 1.7 × 107 m s−1 for the UI luminosity. A few frames later, much dimmer luminosity (similar to that of a leader) extended upward to about 1.7 km (Figure 2d). In total, the UI channel was visible for 0.47 ms. As shown in Figure 2, the UI is in a separate channel, approximately 2.5 km to the south (as viewed from the camera site) of the main RS. The channel section of the UI appears in the same location as the first of the two downward leader branches visible before the main RS (c.f. Figures 2d and 2a). Although it was visible nearly to the horizon, this stepped leader had not obviously connected to ground earlier during the RS. It is clear in Figure 2c that the RS illuminated only the upper portion of the branch subsequently used by the UI, while Figure 2dshows that the UI illuminated only the lowest, ground-connected portion of this branch. These two facts suggest that the lower part of the leader used by the UI had cut off electrically from the rest of the flash's stepped leader branches before the RS.
 When the UI luminosity begins at the ground, the channel of the main RS is no longer visible all the way to the ground, although a portion of the main channel is still clearly seen at mid-levels in the cloud. (This mid-level portion of the main RS channel is visible for an unusually long time, 3.76 ms, after the beginning of the RS.) Although it is difficult to verify from the presentation inFigure 2, brightness-enhanced images indicate that the UI channel did not visibly connect with the main RS channel, in the sense that the (uppermost) portion of the channel between the UI and the main channel did not illuminate. Also, there was no subsequent apparent increase in luminosity of the main RS channel above the UI channel, as one would expect with a branch component, for example, or if the current of the UI connected up to the main channel.
 The estimated peak current of the LINET ground stroke associated with the UI is −5.2 kA. LINET gives the location of the UI as 2.4 km south and 1.1 km east of the RS, which agrees well with the video data. Neither CGLSS nor NLDN detected the UI event as a return stroke. There were no LDAR2 sources until 1.52 ms after the RS, near the end of the UI; this LDAR2 source was at 6.9 km altitude and 6 km south-southeast of the return stroke's ground location.
 The Ch3 antenna data from the site nearest to the flash and Ch1 antenna data from the camera site are shown in Figure 3b for 4.0 ms beginning just before the RS. The RS field change of −30 V m−1 is followed by numerous large oscillations, corresponding to times when the luminosity wave and current pulse reach major branches, sections, and junctions along the channel [e.g., Rakov and Uman, 2003]. The gradual downward slope of the Ch3 data in Figure 3b indicates that current continued to flow in the channel for nearly 4 ms after the RS; this observation fits with the video data of continued luminosity for 3.76 ms in part of the channel. Beginning about 0.70 ms after the RS, there is a series of small field change pulses which steadily increase in amplitude to reach a maximum of about 4 V m−1over 0.10 ms, then decrease and end after another 0.23 ms. Associated with these pulses, one LINET source, with −2.2 kA peak current at 4.2 km altitude, was detected 0.8 ms after the RS. During this interval, the main activity visible in the video data is occurring high and to the south of the mid-level kink in the channel; step-like bright spots and diffuse cloud light are seen intermittently there, along with numerous re-illuminations of the long horizontal channel to the north centered near 4.7 km altitude. The UI from the ground occurs about 0.26 ms after this pulse train ceases.
 The initial UI light pulse is accompanied by a relatively large (−5 V m−1, or about one-sixth the amplitude of the preceding RS) unipolar negative-going field change pulse (Figure 3c). This initial pulse rises to its peak (negative) value in 2 μs; for comparison, the risetime of the RS was 4 μs (in reasonable agreement with the average 10–90% risetimes of 2.6 μs for first return strokes [Master et al., 1984]). The initial pulse is followed by several smaller pulses, then by another large pulse with a slower risetime (5 μs) and then by several more smaller pulses; the overall duration of the UI was about 0.13 ms. Thus the electric field change data for the UI resemble those of a relatively weak negative RS, not only in the overall shape, but also in the initial fast risetime.
 Although we have noted above that the UI channel did not visibly connect to the main RS channel, the data do indicate that this weak ground stroke may have influenced (or possibly led to) subsequent leader development beyond the opposite end of the channel. Within 0.26 ms after the UI (and 1.56 ms after the main RS), extensive and complex branching activity became visible near the cloud top and south of any previously seen channels. This newly visible activity (Figures 2e and 2f) appears as gradual, step-like advancement of first one channel (at 9.5–9.9 km) and then a second channel (at 9.1–7.5 km) apparently near the cloud's edge. Each channel is seen to extend and then dim, visible over 23–25 video frames (about 460 μs). The E change data (Figure 3c) have many small step-type pulses indicative of leader development beginning 0.08 ms before this first cloud top channel is visible and continuing until just before the second channel reaches its apparent full extent. These pulses begin near the time of an LDAR2 source that is at 6.9 km altitude and horizontally very near the initiation location of the flash. Several individual step-type pulses in the Ch3 data are closely time correlated (within the 18.1 μs image exposure) with distinct bright steps visible in a few individual video frames. Meanwhile, the UI channel to ground is still visible (for 0.41 ms), fading out 0.2 ms after the cloud top channels became visible. The central portion of the main RS channel (centered near 6.3 km) is also still visible, though it is gradually dimming throughout this period.
 After the upward illumination in this flash, there were two visible failed dart leaders and associated K-changes, starting at about 64 and 111 ms after the main RS. The second of these (at 1836:21.64 UT) appeared to nearly reach the ground as a dart-stepped leader, but ended at about 2 km altitude after 0.7 ms of visible, mid-level and sub-cloud development. A portion of this leader traveled along the same path as the original, earliest downward leader branch which preceded and was partly used by the UI. The K-change (ΔE of about −2 V m−1) associated with this failed dart leader is apparent in the Ch3 data in Figure 3a.
 Finally, for this flash (as well as the others presented below) we can state that close examination of available USGS satellite imagery and topographic maps in the vicinity of the RS and UI event locations has revealed no tall towers or other very tall structures in the area. (In the video images of Figure 2, there is a tree near the apparent UI strike point; this small tree is on Merritt Island and no further than 8 km from the camera.) The LINET, CGLSS, and NLDN locations for the preceding RS agree to within 0.5 km, and all three systems place the RS about 35 km away, just south of Lake Harney, FL (Figure 1). Both LINET and the video data indicate the UI event is about 2.5 km further south (relative to the camera's image plane), near Puzzle Lake along the Saint Johns River.
3.2. Five-Return-Stroke Flash on 29 July
 The second example we consider in this study was a powerful negative CG flash with five return strokes, the first of which occurred at 1917:04.23653 UT on 29 July 2010. Figures 4a–4c show the Ch2 antenna data from the site closest to the flash and Ch1 data from the camera site, with LDAR2, CGLSS, and LINET source altitudes for the flash. In this case the first LDAR2 source and the first LINET source were both at 5.9 km altitude and occurred only 14.3 ms before the first RS (Figure 4a). With the camera, three distinct leaders were sequentially visible after they extended below cloud base, and stepping was visible in the third of these channels for just 2.06 ms before it connected to ground. Although the location of the first RS was 21.7 km south and 13.3 km west of the camera site (25.4 km distant), this stroke was very bright in the camera data (Figure 5a). The first-stroke peak current was −78.9 kA from CGLSS, −88.4 kA from NLDN, and −75.6 kA from LINET. A second RS, following a different channel to ground (originally developed by one of the other two leaders seen earlier), occurred 34.8 ms after the first. The ground location was 1 km further south relative to the camera, and its estimated peak current was −11.9 kA from CGLSS and −15.7 kA from LINET. (This stroke was not detected by NLDN.) No leader stepping was seen in the camera data prior to the second RS, but several branches did light up (Figure 5b) during that RS.
 About 50 ms after the second RS, another sub-cloud leader became visible in the optical data. This leader appears in the same place as the earliest of the three seen before the first RS, though it did not connect to ground during that stroke. The leader extends below cloud base and becomes increasingly bright for 0.72 ms until it reaches ground for the third RS at 1917:04.32203 (Figure 5c). According to LINET, the location of this stroke is 22.3 km south and 12.0 km west of the camera site (25.3 km distant and near the highway intersection of I-95 and FL-528,Figure 1). (The CGLSS location is within 0.2 km of the LINET position for this stroke.) The CGLSS-estimated peak current was −27.6 kA (−33.5 kA with NLDN; −33.5 kA with LINET). In the five video frames (93 μs) immediately after the beginning of the third RS, two major branches illuminate nearly to ground, while the RS channel disappears into the cloud then reappears at 6.7–7.2 km altitude. This activity is followed by several frames (0.13–0.22 ms after the RS) showing increased sub-cloud leader development to the north of the third return stroke's ground location. Both these periods of visible activity are coincident with relatively large oscillations in the field change data (Figure 4b).
 The upward illumination in this flash began at 1917:04.32255 UT, 0.519 ms after the third RS, from a ground location about 0.7 km to the right (approximately northwest in the image plane). At the time, the third RS channel has dimmed but is still illuminated (and presumably still carrying current, for at least the 1.0 ms that it is visible) to the ground, as shown in Figure 5d; the gradual decline seen in the Ch2 field change data (Figure 4b) after the third RS also indicates current is continuing to flow in the channel. The UI reaches its apparent maximum vertical extent of about 1.1 km (above horizon) in four video frames (74 μs). Estimated 2D propagation speed of the UI luminosity wave is 1.6 × 107 m s−1, which is much slower than that of the preceding RS at lowest levels (4.2 × 107 m s−1 in this case.) The UI does not visibly connect to the main channel below cloud base (visible cloud base altitude is 1.4–1.6 km above the horizon), and the main channel continues to dim throughout this interval. During the UI, which is lit for 0.33 ms in the video data, two other branches connected to the UI channel also light briefly.
 Neither CGLSS nor NLDN detected a ground event at the time of this UI, and there were no coincident LDAR2 sources during the event. At the time of the UI, LINET detected a ground stroke with −3.2 kA peak current and location about 0.6 km west and 0.3 km south of the preceding RS. The Ch2 and Ch1 antenna data (Figures 4b and 4c) show a rapid field change (with a 2 μs risetime) similar to that of a weak negative RS coincident with the UI. Compared to the third RS field change of about −58 V m−1 (at the nearest site to the flash), the UI field change is only about one tenth as large. The risetime of the third RS was also 2 μs.
 It is interesting to note that the first three return strokes of this flash followed three different sub-cloud leaders and connected to ground in three different locations. The UI during the third RS occurred in another, fourth, ground location. Two additional subsequent return strokes were detected in this flash: the fourth (at 1917:04.36918) followed the second return stroke's channel to ground, with CGLSS estimated peak current of −14.5 kA (−19.0 kA at 2.4 km altitude with LINET, −16.0 kA with NLDN), while the fifth (at 1917:04.42278) followed the third return stroke's channel to ground, with estimated peak current of −17.6 kA (−21.6 kA with LINET, −20.9 kA with NLDN).
3.3. Two-Return-Stroke Flash on 28 July
 Our third case of upward illumination occurred in a normal, negative CG flash with two return strokes on 28 July 2010. The first LDAR2 source in this flash was located at 5.7 km altitude, 38.5 ms before the first RS time of 1820:24.23910 UT (Figure 6a). The leader to ground was visible in the video data for only 0.21 ms before the start of RS luminosity (Figure 7a). The RS luminosity front then propagates mostly vertically to 5.0 km altitude (for 130 μs), before bending northward and upward to its highest visible point at 5.8 km after 204 μs. The average 2D speed of the RS luminosity is 4 × 107 m s−1 over this time range, but it is slightly faster (5.6 × 107 m s−1) over the first five frames. Estimated peak current of the RS from CGLSS was −16.1 kA (−19.0 kA from LINET, −19.1 kA from NLDN), the electric field change risetime was 5 μs, and the LINET stroke location was 5.2 km north and 33.4 km west of the camera site (33.8 km distant, near Puzzle Lake in Figure 1, and just a few km south of the flash described in section 3.1).
 About 0.71 ms after the start of the first return stroke's luminosity and while it was still fully lit, a new channel suddenly became visible in the camera data; from the camera's perspective, this new channel was roughly parallel to and south of the main channel at mid-level in the cloud. This channel was visible for 0.42 ms total, its activity ending with obvious stepping below the cloud base over several video frames, although it did not reach the ground. All visible channel luminosity associated with the first RS ended about 140 μs later.
 The UI in this flash (Figure 7b) began at 1820:24.240413 UT, 1.318 ms after the first RS, from a ground location about 1.6 km to the north. In this case, the main RS channel disappeared from view only 56 μs (three video frames) before the UI began. The UI channel reached its apparent full vertical extent of about 1.5 km (the base altitude of an intervening debris cloud) in 93 μs (5 video frames). The estimated 2D propagation speed of this UI luminosity is 1.7 × 107 m s−1 (compared to 5.6 × 107 m s−1for the third RS). Along with several branches near the ground stemming downward from the middle of the UI channel, the entire UI event is visible for a total of 0.47 ms. Similar to the two previous flash examples, there is no visible channel connection between the UI and the main channel. Also, the main channel does not re-illuminate during or after the UI.Figure 7 shows the camera image with the maximum extent and luminosity of the UI (Figure 7b), along with images of the first (Figure 7a) and second (Figure 7c) full return strokes of this flash for comparison. Note that the two return strokes are visible much higher (into the cloud) compared to the UI.
 Neither CGLSS nor NLDN detected this UI event. LINET did detect the UI as a ground stroke, about 1.1 km north and 0.7 km east of the RS, with an estimated peak current of −4.6 kA. The LINET data also show a source 0.55 ms after the RS (0.76 ms before the UI), which is nearly time-coincident with the beginning of the new channel activity (described above) seen in the optical data to the south of the main RS channel. (This LINET source is not shown inFigure 6b because no altitude was determined.) The LDAR2 data include three sources within 2 ms after the RS (Figure 6b), including one before the UI event, but none of these is coincident with visible leader or UI activity. Meanwhile, the Ch3 antenna data in Figure 6b clearly show a fast field change (risetime of 2 μs) similar to that of a weak negative RS coincident with the UI. As with the earlier examples presented, the UI field change is smaller than that of the preceding RS (about one-third of the return stroke's −16 V m−1 field change). Also like the earlier examples, there is a gradual slope of the electric field change that extends through the time of the UI event and is indicative of current continuing to flow in part of the lightning channel.
 The second full RS of this flash occurred about 76.9 ms after the first RS (1820:24.3170 UT). Stepped leader development was visible with the camera about 0.76 ms before the second RS, not in the same place as the earlier RS channel, and multiple branches were apparent just above the ground within 170 μs of the second RS. As Figure 7 shows, only the extreme top of the visible channel is in the same location as that of the first RS. Relative to the camera site, the location of the second RS is 7.8 km north and 31.5 km west (32.4 km distant), or about 3 km closer than the first RS. The brighter luminosity of this stroke in the video data is also in agreement with the slightly larger estimated peak current values: −17.2 kA from CGLSS (−20.4 kA and −22.3 kA from LINET and NLDN, respectively). Another 76.1 ms after this second RS, a faint dart leader is visible in the optical data at middle to low levels in the cloud. This leader is visible for 368 μs, traveling along the path of a side branch that had been visible during the second RS, but this dart leader does not reach ground nor result in a RS. The Ch3 antenna data (Figure 6c) indicate a clear K-change coincident with this failed dart leader, and LDAR2 detected a source at 6.5 km altitude at the time of a fast pulse in the Ch3 data just before the K-change.
 As with the previous example, we note that the return strokes in this flash followed multiple channels to ground. The two normal return strokes, and the intervening UI event, each followed different leaders or leader branches that connected to ground in three different locations. Substantial stepped leader development and branching were visible with the camera before, during, and after the two return strokes at middle and low (sub-cloud) levels. However, the UI channel was not seen to connect to the main RS channel that immediately preceded it.
3.4. Distant Flash on 28 July
 The final case of upward illumination we consider in this study was part of a negative CG flash that occurred at 1935:10 UT on 28 July 2010. In fact, it is not clear from our data whether there was one flash or two at this time: in addition to the UI event, there were four return strokes, with the first and last possibly not connected to the intervening return strokes. The first RS occurred before the video data record began, while the last RS was seen several km north and much less distinctly than the two prior video-recorded return strokes. Although the first and last return strokes were 538 ms apart, their positions with both NLDN and LINET are close together, while the other return strokes were about 6 km further south and 2–4 km further west than the first and last. It is not obvious from LINET or LDAR2 data for the flash (Figures 8a and 8b) whether the second and third return strokes are connected to the first and last return strokes within the cloud. Nevertheless, for the purpose of this description, we assume these strokes were all part of one complex flash.
 The earliest LDAR2 sources detected in this flash were at 4.4–6.7 km altitude, about 54 ms before the first RS; after just three sources, no further LDAR2 points were located for 35 ms. The VHF activity re-starts with a source at 7.0 km detected 18 ms before the first RS at 1935:10.16417 UT (Figure 8a). The LINET-estimated peak current of this RS was −70.5 kA (−70.6 kA from NLDN, not detected by CGLSS). About 85 ms later, a faint sub-cloud leader became visible in the optical data 0.11 ms before the start of the second RS at 1935:10.24984 UT. (Visible cloud base is at 1.8 km altitude.)Figure 9a shows this RS at full luminosity, 74 μs after it began. The estimated 2D speed of the RS luminosity was 6.5 × 107 m s−1. LINET located this second ground stroke 25.6 km north and 48.4 km west of the camera site (54.7 km distant and near the town of Osteen, FL; Figure 1), and its estimated peak current was −26.1 kA (−26.6 kA from NLDN, −22.0 kA from CGLSS; CGLSS also placed the stroke within 0.6 km of the LINET location). The RS channel was visible in the video data for 1.63 ms.
 The UI in this flash (Figure 9b) began at 1935:10.251702 UT, 1.858 ms after the second RS, from a ground location about 0.8 km to the right (essentially northeast in the camera's image plane). As partly shown in Figure 9, there were a few branches faintly apparent, extending upward and downward from the top end of the UI as it reached it full vertical extent of about 1.4 km over 56 μs (3 video frames). The estimated 2D speed of the UI luminosity is 2.6 × 107 m s−1. The UI event was visible for 0.44 ms. Once again, there were no visible connections to the prior RS channel, and that channel did not appear to re-illuminate during the UI event.
 LINET detected this UI as a ground stroke about 0.5 km south and 2.5 km east of the second RS location (Figure 1), and estimated its peak current as −6.5 kA. As with the other three UI events, neither CGLSS nor NLDN located a ground stroke at the time. An additional LINET source with −2.5 kA peak current was detected at 8.0 km altitude in the time between the RS and the UI (0.85 ms after the RS, 1.0 ms before the UI). There is no coincident activity visible in the optical data, but the Ch3 antenna data (Figure 8b) indicate a significant bipolar pulse followed by a weak K-change. LDAR2 also detected one source 0.84 ms after the RS (and 0.01 ms before the LINET cloud stroke) at 5.6 km altitude and very near the horizontal position of the initial source of the entire flash. After this single point, there are no other LDAR2 sources associated with flash for about 20 ms.
 In the Ch3 antenna data (Figure 8b) for this event, a field change of −4 V m−1 (and a risetime of 5 μs) marks the time of the UI. The field change of the prior RS was about −12 V m−1 (with a risetime of 4 μs), so once again we find the UI event is much smaller, as for a weak negative RS. Weak current flow is also evident in the field change data after the second RS and through time of the UI event.
 The next full RS of this flash occurred at 1935:10.272 UT, 20.9 ms after the UI. Unlike the other examples, this subsequent RS came to ground in the same location as the UI; in the video data (Figure 9c), the third RS has a similar appearance as the UI but is not as bright, no branches are visible, and there is channel faintly visible higher in the cloud. The LINET locations for both events agree to within 30 m, and the estimated peak current of the third RS is slightly smaller (−5.3 kA versus −6.5 kA for the UI). The CGLSS location of this third stroke is also 25.9 km north and 46.4 km west of the camera site (53.2 km distant), within 0.9 km of the LINET location. This RS channel is re-illuminated three times by apparent M-components, the last of which occurred 40 ms after the start of the third RS. A slow field change associated with continuing current was also apparent in the Ch3 antenna data (Figure 8a) for at least 90 ms after the third RS. It is noteworthy that the third return stroke was visible to a much higher altitude (4.8 km) than the UI (1.4 km) even though the UI had a slightly larger peak current than the third RS and both followed the same path for the lowest 1.4 km.
 The fourth and final RS of this flash occurred 430 ms after (at 1935:10.702 UT) and about 6 km distant from the third RS. As mentioned above, this last RS was much less distinct in the optical data, with luminosity lasting only about 0.2 ms and not extending above the apparent cloud base. The estimated peak current from CGLSS was −18.7 kA (−23.0 from NLDN, −21.4 from LINET). Data for the final part of this flash (after 1935:10.4 UT) are not shown in Figure 8a because the electric field change sensors at the nearest site did not trigger on this stroke.
 While each case of upward illumination described above is different in some details, there are common features that are worthy of synopsis. To aid the discussion, we have collected in Table 2 some characteristics of the UIs along with those of the preceding RSs.
Table 2. Characteristics of Upward Illumination (UI) Events and Their Preceding Return Stroke (RS)a
Labels (3.1–3.4) indicate the section in which each flash was presented.
Distance from camera (km)
Time separation (ms)
Ground point separation (km)
2.5 (camera) 2.6 (LINET)
0.7 (camera) 0.8 (LINET)
1.6 (camera) 1.3 (LINET)
0.8 (camera) 2.6 (LINET)
Electric field change (V m−1)
Risetime to peak (μs)
Peak current (kA) (LINET)
Duration of vis channel (ms)
2D speed of luminosity (m s−1) [frames from 0th]
6.6 × 107 five
1.7 × 107 five
4.2 × 107 two
1.6 × 107 three
5.6 × 107 five
1.7 × 107 four
6.5 × 107 four
2.6 × 107 three
Max alt of vis luminosity (km)
Visible cloud base alt (km)
Time (UT), Date
1836:21, 28 Jul ‘10
1917:04, 29 Jul ‘10
1820:24, 28 Jul ‘10
1935:10, 28 Jul ‘10
 First, we note that there is video evidence in three of the four UI cases for multiple leaders developing toward ground prior to the main RS. If these are competing leaders [e.g., Ballarotti et al., 2005], then the ‘winning’ leader here gets the return stroke, and one of the ‘losers’ gets the upward illumination slightly later, during or after the return stroke. The clearest example of this in our data is the first flash discussed (section 3.1), where the main leader splits near 5 km altitude, and then multiple leader branches develop over many milliseconds within view of the camera (e.g., Figure 2a). Multiple faint leaders are apparent only below cloud base (about 1.5 km) in the second and third examples, so we cannot determine if these leaders are separate or branches from a main leader within the cloud. The stepped leaders in the fourth flash are difficult to discern due to distance and low cloud base. However, we conclude that the UI events presented herein are unlike the “forked lightning” events described by Ballarotti et al. , since the UI channels are not visibly connected to the main RS channel. In fact, in three of our four cases, the RS channel is no longer visible to the ground when the UI event occurs. (In other video data, including Figure 5c, we do see “forked lightning” events like those shown in Ballarotti et al. .)
 A second common feature of the flashes having UI events is that they each exhibit extensive branching and leader development after the main RS and before the UI event. Much of this activity is along and below cloud base; some of it occurs in places where leaders were evident before the RS. While this type of activity is not uncommon in CG flashes, to us, these particular flashes seem to have more extensive and more prevalent post-RS branching than usual. The RF data presented above also show evidence of post-RS leader activity: in three cases there was one LINET source detected between the times of the main RS and the upward illumination, and in two cases there was an LDAR2 source during this interval. In each case, Ch2 or Ch3 antenna data indicate current continues to flow in some of the lightning channels (although not necessarily to ground) in the few ms (or longer) after the RS until the UI event.
 A third feature common to all the cases is that each RS preceding a UI event is in a new ground location from any prior strokes in the flash. (Two cases followed first return strokes of flashes, while the other two followed subsequent return strokes to new locations.) A similar finding was noted by Rakov and Uman for their double-ground cases and byBallarotti et al.  in which both terminations of their forked strokes were new ground strike points for the flashes. The separation distances given in Thottappillil et al.  for seven strokes with multiple terminations (i.e., two at 0–1 km, three at 1–2 km, one at 3–4 km, one at 4–5 km) are similar to the distances between UI and RS in our smaller data set (Table 2). Rakov et al.  suggested that relatively poor channel conditioning may play a role in producing multiple ground contact points for negative CG flashes in general. Furthermore, each UI event also occurs in a new ground location. Thus, in each case, both the UI and the RS that immediately precedes it are likely due to stepped leaders propagating toward ground.
 The fourth common feature of the UI events is that none of the UI channels appeared to connect back to the main RS channel, as if the UI were a late, unseen leader branch connecting to ground. Instead, the UI luminosity appeared to stop at a low altitude rather than following the stepped leader path all the way back to the flash origin in the cloud. The first flash presented (section 3.1) is the best example of this feature, since the descending leaders (both the one associated with the RS and the one associated with the UI) are visible from a relatively high altitude, the RS channel is visible even higher into the cloud, and the UI vertical extent is considerably less than the leaders or the RS. In this regard, however, we note that our four examples are rather distant from the camera site (25–55 km), thus faint connections may not be visible in these video data.
 Of the three examples with return strokes occurring after the UI, later in the flash, one has a subsequent RS in the same location as the UI. The other case (the first flash presented, with no subsequent RS) has a subsequent, very long, failed dart leader that nearly connects to ground at the UI location. Overall, in our 2010 data set of ten CG flashes, seven are multistroke and six of these have multiple ground contact points. Evidently, poor channel conditioning, of the sort suggested by Rakov et al.  as important for multiple ground terminations, was common during the times the video camera was operating in our experiment.
 The fifth characteristic common to the UI events presented herein is that they are relatively weak strokes in several senses, as shown in Table 2. The UIs extend to lower maximum altitudes (1.1–1.5 km) than any of the preceding RS altitudes (4.8–7.2 km). The 2D propagation speeds of the UI events are much slower (1.6–2.6 × 107 m s−1) than is typical for branched RS speeds, both in the literature [e.g., Idone and Orville, 1982] as noted above, and in our data for the preceding RS in these cases (4.2–6.6 × 107 m s−1). The UI events had relatively small peak currents (−3.2 kA to −6.5 kA), much smaller that the peak currents of the preceding RSs (−19.0 to −48.5 kA). The relatively long time (of order ms) that had elapsed between the stepped leader propagation and the UI event may have allowed time for the leader channel to cool appreciably, which may explain the slower UI speeds and lower UI peak currents. In addition, the UI luminosity duration (0.33–0.47 ms) was shorter than the preceding RS luminosity (0.95–3.76 ms), indicating a shorter current duration [e.g., Diendorfer et al., 2003; Flache et al., 2008]. Electric field changes of the UIs (−4 to −6 V m−1) were also significantly smaller than the E changes of the preceding RSs (−12 to −58 V m−1). Taken together, the smaller UI peak currents, shorter UI luminosity durations, and smaller UI field changes indicate that smaller amounts of charge were transferred by the UI events than by the preceding RSs. All these observations also fit with the notion that the leader or leader branch within which the UI occurs was effectively cut off from the main leader and main RS channel.
 A related common feature of the UI events is that none were located by the two operational ground-stroke detection systems available at KSC. While NLDN is known to have lower detection efficiency for low peak current ground strokes [Jerauld et al., 2005; Cummins et al., 2006], it is not clear from our study why CGLSS does not detect these events as separate ground strokes. Wilson et al. show that CGLSS does detect many low peak current events in new ground contact points, although their analysis region is restricted to flashes within about 20 km of our camera site. For the 29 normal return strokes seen in all the camera data on 28–29 July 2010, stroke detection efficiencies were 85% for NLDN and 90% for CGLSS. Stroke detection efficiencies for CGLSS on four other days in 2010 were 78–94%, averaging 83% of 573 return strokes identified with the electric field change data. (During our 2010 field campaign, there were problems with three of the six CGLSS sensors that degraded the performance of CGLSS (W. P. Roeder, personal communication, 2011).) The LINET system, designed to locate both in-cloud events and ground strokes, did detect all four UI events, along with all 29 normal return strokes and 96–97% of the 450 return strokes identified on two other days analyzed during the experiment. Although not all the return strokes were identified by LINET as ground strokes (with 0 km altitude), primarily due to unfavorable sensor geometry in the 2010 network, this system appears better able to detect weak ground-connected events like the upward illuminations.
 Based on the analyses to date, it seems reasonable to suggest that conditions favoring flashes with multiple ground strike locations also favor UI events. As mentioned in section 1, numerous prior investigators have determined that up to about half of all multistroke CG flashes may have multiple ground contact points. It also seems likely that multibranched leaders, spreading across substantial horizontal distances and possibly connecting to ground in different locations, are important in the formation of UI events. Rakov and Uman found 15 of the 246 return strokes (6%) in their study attached to ground in two places. In this study, we have 4 out of 29 normal return strokes (14%) in 10 flashes showing subsequent UI events. Satellite imagery of the area where these four flashes occurred has revealed no tall towers or other significant tall structures that may have initiated long (unseen) upward leaders, although closer examples, with the strike-points well in view of the camera, would be needed to be certain that towers are not involved.
 Throughout this paper, we have purposefully used the words ‘apparent’ and ‘visible’ to specifically indicate what is documented in our data. We do recognize, as one anonymous reviewer pointed out, “that the viewing instrument has a threshold and that the viewing must be degraded for the relative large UI distances given of 25 to 55 km.” However, the fact that stepped leaders, dart leaders, and post-RS branch development are seen in our data for several of the UI cases means that we are not completely blind to low luminosity events, particularly possible channel connections, despite their distance. If unseen channel connections between the UI and preceding RS are present, then it seems reasonable to conclude that these connections carry only relatively low currents (e.g., less than the 100 A typical of stepped leaders, which are observed by our camera) upward to the cloud.
 We now consider whether the UIs are return strokes in the common meaning of a RS, which we take from Rakov and Uman : “The terms “stroke” or “component stroke” apply only to components of cloud-to-ground discharges. Each stroke involves a downward leader and an upward return stroke and may involve a relatively low level continuing current that immediately follows the return stroke.” The meaning of ‘return’ is given in the return stroke definition ofSchonland (who used the now out-of-date terms ‘leader streamer’ and ‘return streamer’): “Each separate stroke of a flash to ground is of a dual nature. A downward-moving weakly luminous process, the leader streamer, is followed, upon arrival near the ground, by an intensely luminous and much more rapid main or return streamer which traverses the leader channel in the reverse direction.” Neither definition discusses the vertical length of a return stroke. As shown above, the UIs have a downward stepped leader and an upward-moving luminosity that follows the preceding stepped leader path upward. As we have pointed out above, the electric field changes of the UIs look like small versions of typical return strokes' field changes. The UIs have currents similar to those of weak return strokes. Therefore, it may be reasonable to label the UIs as weak return strokes.
 Regardless of the term used to classify the UIs, there are two main visual differences between the UIs and typical RSs: namely, the UIs shorter vertical extent and their slower upward velocity. The shorter vertical extent of the UIs is seen in Figures 2, 5, 7, and 9 and in Table 2; the slower speed is documented in Table 2. One might ask how far up the preceding stepped leader a stroke must travel before it is classified as a “return stroke.” Is 1000 m enough, or 100 m? We admit that this aspect of the definition is not especially important to us. Rather than how to classify them, what we are more interested in determining is the mechanism of the UIs. To us, it seems the most important criterion in the occurrence of the UI event is that one stepped leader (or leader branch) must effectively cut off electrically from the leader that becomes the main RS channel. The disconnected leader beyond the cut-off portion would have an electric potential distribution distinct from the main leader, it would not conduct any of the main RS current or charge, and its subsequent propagation could be influenced by the charge rearrangement of the main RS. In situations where this cut-off leader has approached sufficiently close to ground, it may continue to propagate downward and connect as a UI event soon after the main RS. This hypothesis is consistent with the common features noted above and inTable 2, based on the data presented in section 3: the UI events observed thus far are preceded by multiple leaders (before the RS) and extensive branching (after the RS), they are short in height and duration, and they are weak in electric field change and peak current. Since we argued above that a UI is a type of return stroke with a short upward return path and since the short path seems to be due to a physical process not shared by most return strokes, we suggest referring to UIs as “UI-type return strokes.” Of course a better understanding of the full physical situation requires more and closer UI examples, with a more complete set of electric field change data. Higher spatial resolution in the video data is needed to specifically determine whether the UI channel luminosity connects back to the RS channel, either immediately before or during the event. Closer examples are also needed to fully address the question of whether the UI events always occur in branches off the main leader or sometimes in separate leaders.
 In this paper we describe four cases of short-length, ground-connected strokes that began 0.519–1.858 ms after and 0.7 to 2.5 km distant from a normal return stroke (RS) in four negative cloud-to-ground lightning flashes observed at 25–55 km distance. High-speed video data (18.5 μs image interval, 18.1 μs image exposure) and time-correlated electric field change data show these upward illumination events are not exactly like anything else documented in the literature. The UI events are relatively weak in terms of their luminosity, peak current, propagation speed, and field change relative to the preceding return strokes. The LDAR2, CGLSS, and NLDN data show no indication of a ground strike at the times of the four UI events; LINET detected all four as low peak current (−3.2 to −6.5 kA) ground strokes. In the video data the UI events are visible for relatively short durations (less than 0.5 ms) and short heights (less than about 1.7 km altitude above horizon), and their luminosity propagates at two-dimensional speeds of only about 1.6 to 2.6 × 107 m s−1(25–40% the speed of the preceding return strokes). Notably, they do not appear to connect back to the main RS channel or up into the cloud. In each case, the preceding RS and the UI itself are each to a new ground location for the flash (first RS in two cases, subsequent RS to a new location in two cases). Immediately before the RS and UI, multiple leaders are visible in three of the cases; extensive branching is visible during and after each RS and UI. Electric field change data indicate that current continues to flow after the RS and through the UI event, although the RS is no longer illuminated all the way to ground at the time of the UI in two of the four cases. In one case, the UI channel is re-used later in the flash by a subsequent RS. Since the four flashes with upward illumination are from a data set of only ten flashes acquired in east-central Florida on two days in July 2010, we further conclude that these events may not be rare in this region. Closer examples are needed to understand the physical mechanisms that may be responsible for the UI features, but the data presented are consistent with the hypothesis that a UI is caused by a downward stepped leader that has become electrically cut off from the main stepped leader. If this hypothesis proves correct, then one might call this phenomenon a UI-type return stroke.
 This project was supported by the NASA/Mississippi Space Grant Consortium (grant NNG05GJ72H), the National Science Foundation (grant AGS-1016004), the UM Department of Physics and Astronomy, and the UM Office of Research and Sponsored Programs. The V12.1 camera was made available via NSF grant AGS-0813672. We thank Mark Stanley, Richard Sonnenfeld, Frank Merceret, John Madura, Jennifer Wilson, Joseph Dwyer, Meagan Schaal, Gary Huffines, and Andy Detwiler for important help with various aspects of this study. Special thanks are due to Lauren Vickers for assisting with the camera operations and the electric field change data collection. The Florida LINET sensor hosts are greatly appreciated for their vital assistance: Florida Institute of Technology Department of Physics and Space Sciences in Melbourne, Hickory Tree Elementary School in St. Cloud, Brevard Community College Planetarium in Cocoa, Titusville-Cocoa Airport Authority in Titusville, Wedgefield Golf Club in Wedgefield, St. Luke's Lutheran School in Oviedo, and Massey Ranch Airpark in Edgewater. Any mention of copyrighted, trademarked, or proprietary products or services does not constitute endorsement thereof by the authors, their institutions, NASA, the NSF, or the U.S. Government. Such references are given solely for the purpose of informing readers of the resources used to conduct this study.