Corresponding author: M. Stolzenburg, Department of Physics and Astronomy, University of Mississippi, PO Box 1848, University, MS 38677, USA. (firstname.lastname@example.org)
 Time correlated high-speed video and electromagnetic data for 15 cloud-to-ground and intracloud lightning flashes reveal bursts of light, bright enough to be seen through intervening cloud, during the initial breakdown (IB) stage and within the first 3 ms after flash initiation. Each sudden increase in luminosity is coincident with a CG type (12 cases) or an IC type (3 cases) IB pulse in fast electric field change records. The E-change data for 217 flashes indicate that all CG and IC flashes have IB pulses. The luminosity bursts of 14 negative CG flashes occur 11–340 ms before the first return stroke, at altitudes of 4–8 km, and at 4–41 km range from the camera. In seven cases, linear segments visibly advance away from the first light burst for 55–200 µs, then the entire length dims, then the luminosity sequence repeats along the same path. These visible initial leaders or streamers lengthen intermittently to about 300–1500 m. Their estimated 2-D speeds are 4–18 × 105 m s−1 over the first few hundred microseconds and decrease by about 50% over the first 2 ms. In other cases, only a bright spot or a broad area of diffuse light, presumably scattered by intervening cloud, is visible. The bright area grows larger over 20–60 µs before the luminosity fades in about 100 µs, then this sequence may repeat several times. In several flashes, a 1–2 ms period of little or no luminosity and small E-change is observed following the IB stage prior to stepped leader development.
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.
 The initial or “B” portion [after Schonland, 1956; Clarence and Malan, 1957] of lightning flashes can be defined as the time when a succession of breakdown events occurs, before or during the start of initial leader development. Initial breakdown (IB) has been described in terms of fast electric field change (E-change) measurements [e.g., Kitagawa and Brook, 1960; Weidman and Krider, 1979; Beasley et al., 1982; Nag et al., 2009] as a series of bipolar pulses occurring for several milliseconds at the beginning of a flash, with the largest pulses typically lasting 10–80 µs. IB pulses are observed at the start of both intracloud (IC) and cloud-to-ground (CG) flashes, having opposite polarity in the two types, and they can have relatively large amplitudes of several V m−1 when observed at close range. Using single E-change sensors, Nag and Rakov  found an IB pulse “train” at the beginning of 18% of 327 negative CG flashes, while Schulz and Diendorfer  found IB pulses in 81% of 194 negative CG flashes. In our usage, “IB pulses” are likely the same as “characteristic pulses,” which Beasley et al.  identified in CG flashes as marking the transition (of a few milliseconds) between initiation and stepped leader. Based on the appearance of the earliest E-change pulses from many ground flashes, Clarence and Malan  hypothesized that the initial breakdown process “must be entirely different from the stepped leader process,” while other researchers [e.g., Weidman and Krider, 1979; Beasley et al., 1982; Proctor et al., 1988] have concluded that similar processes occur throughout time after initiation and prior to the first return stroke.
 In their study using VHF time-of-arrival data to map lightning flashes, Proctor et al.  determined “that ‘B’ portions, which are initially rapid changes in electric field, are caused by vigorous and heavily branched streamers that are driven by relatively high electric fields, near the origins of the flashes.” Some stepped leaders in their large data set “were preceded or accompanied by cloud streamers, which started at, or very near, the same origin” as the stepped leaders. Proctor et al.  concluded that the “B” process seen in their flashes was essentially the start of the leader, with no distinguishing characteristics separating VHF sources of their first streamers from those of stepped leaders. Speeds of first streamers of IC flashes were estimated by Proctor  to be 0.9–2.1 × 105 m s−1. Rhodes and Krehbiel , using an interferometer, showed that in one CG flash there was 100 ms of preliminary VHF activity prior to the stepped leader. The activity “consisted of a sequence of faster, impulsive breakdown events or bursts along each channel,” and each burst was “50–100 µs in duration and a few hundred meters to a kilometer in length.” During this period, two horizontal channels developed, to estimated lengths of 1–1.5 km, and at average speeds of 3 × 104 and 1 × 105 m s−1. Shao and Krehbiel  used interferometer data to derive upward streamer speeds of 1.5–3.0 × 105 m s−1 for the earliest 10–20 ms of the IB stage in two IC flashes.
 Although there are many electromagnetic observations of IB, there have been only very few indications that light is emitted at the initiation of lightning flashes. Using a photomultiplier tube and fast antenna together, Brook and Kitagawa  show one example where increased luminosity is coincident with the initial E-change and “the development of a stepped leader to ground.” (While Brook and Kitagawa  note only the abrupt increases in luminosity associated with stepped leader pulses, examination of their Figure 5 indicates the photomultiplier output begins to increase slightly at about the time the first IB pulse is visible in the E-change trace.) Also, according to Rakov and Uman , Sourdillon  determined that “most initial leaders emit very faint luminosity which was difficult to detect” and that some initial leaders were stepped while others extended continuously. To our knowledge, no more recent data of optical emissions detected coincident with flash initiation have appeared in the literature.
 In this paper, we describe time-correlated optical and electromagnetic data acquired during the IB interval of 14 flashes in Florida and one flash in South Dakota. Optical data are in the form of video imagery acquired at 50,000 or 54,000 frames per second for natural lightning at 4–41 km range from the camera. The primary electromagnetic data are fast E-change observations recorded with multiple sensors at 2, 5, or 10 sites; one E-change site was colocated with the camera, and all data were time-tagged to accuracy of 1 µs or better. Additional data were available for the Florida experiments from the VHF lightning mapping system at Kennedy Space Center (KSC) and from a VLF/LF network deployed in 2010. These observations show that bursts of luminosity are coincident with IB pulses and RF emission sources. Our data also indicate that all flashes have IB pulses. These findings may have important implications for understanding lightning initiation and initial development of lightning leaders.
2 Overview of Data Acquisition
 Data for this study were collected in three field campaigns with similar instrumentation. During two summers in Florida (July–August 2010 and 2011), we obtained time-correlated data for natural lightning around KSC using a suite of flat plate E-change antennas [e.g., Kitagawa and Brook, 1960], along with the Lightning Detection and Ranging II (LDAR) network [e.g., Murphy et al., 2008]. The E-change antennas were installed at 5 locations in 2010 and 10 locations in 2011, with a separation between sites of 6–104 km. (See Stolzenburg et al.  for a map of the 2010 sites; see Karunarathne et al.  for a map of the 2011 sites.) At each site, three instruments measured E-changes of lightning with decay time constants of 100 µs (Ch1, bandwidth 1.6–630 kHz), 10 s (Ch2, 0.016 Hz to 2.6 MHz), and 1 s (Ch3, 0.16 Hz to 2.6 MHz). As described in Stolzenburg et al. , the Ch1 data show fast changes (1–200 µs) due to events like IB pulses and stepped leader pulses; in addition to the fast changes, the Ch2 and Ch3 data also show relatively slow field changes (1–200 ms) due to events like leader extension and continuing current. Time resolution of the E-change data is 0.2 µs (i.e., sampled at 5 MHz) with GPS 1σ average timing accuracy of less than 2 ns, and data were digitized with 12 bit digitizer [see Karunarathne et al., 2012, for additional details]. The LDAR system at KSC uses the time of arrival to locate impulsive lightning events within part of the VHF (60–66 MHz) range; Murphy et al.  describe the capabilities of this system (also called 4DLSS). The flashes included in this study are within the 250 m median location accuracy for the LDAR network. For the 2010 experiment, a seven-station LIghtning NETwork (LINET) [Betz et al., 2004; 2008] was deployed in east central Florida. LINET uses the time of arrival of the magnetic field change due to lightning current surges at VLF/LF (5–200 kHz) to locate and discriminate between in-cloud and ground events. For LINET, the GPS-based timing errors are less than 0.1 µs, and locations are accurate to ~300 m corresponding to a 1 µs statistical timing error [Betz et al. 2004]. The third campaign from which this study uses data took place around Rapid City, SD, in June 2011. In this campaign, two similarly equipped E-change sites were operated, and no other electromagnetic data were collected.
 For part of the three campaigns, time-correlated data were acquired with a Vision Research Phantom™ “V12.1” high-speed video camera located at one of the E-change antenna sites. Images were captured at 18.5 or 20 µs image interval, with respective 18.1 or 19.6 µs image exposure time. Among other attributes [see Warner et al., 2011], the V12.1 camera has a CCD sensor array with photosite/pixel size of 20 µm and was used at 320 × 240 pixel image resolution. Camera data were recorded with timestamp derived from IRIG-B. In Florida, the camera was equipped with an 8 mm focal length lens. It was operated on 2 days in 2010 and 13 days in 2011 from three different sites, all near 0 km msl altitude. A total of 180 flashes (10 and 170, respectively) were captured during these two experiments. The intended target of most video captures was a CG flash within about 60 km range, although in several cases, an IC flash was also captured in the same file before or after the targeted flash. During the South Dakota campaign, imagery from multiple cameras was collected in a setup similar to that described in Warner et al. . In this study, we use data for one CG flash captured in SD with the V12.1 camera, which was equipped with 12.5 mm focal length lens and located near 1.2 km msl altitude.
 Additional data for the Florida campaigns were available from the 29-station Electric Field Mill network at KSC [e.g., Koshak and Krider, 1989] and were used in this study for calibration of the E-change data. Ground return stroke (RS) data from the KSC Cloud-to-Ground Lightning Surveillance System (CGLSS) network [e.g., Wilson et al., 2009] and, for South Dakota, from the National Lightning Detection Network (NLDN) [e.g., Cummins and Murphy, 2009] were used to estimate distance to the lightning events in order to derive spatial resolution of the video data.
 For this study, time alignment (to one video frame) with the E-change data is accomplished during postanalysis using return stroke times at the E-change sensor colocated with the camera. Because of the uncertainty in lining up the video data with the E-change data, “coincident” means “within 20 µs” herein. Throughout this paper, we use the physics sign convention for electric field (E), where a positive (upward) E exerts an upward force on a positive charge. Calibrated E changes given in the text for IB pulses and other events are not range-normalized unless so stated (in which case, they are range-normalized to 100 km). Times of discrete events are adjusted for propagation delays to the time these events would have arrived at the camera or relevant sensor site.
3 Analysis and Results
Table1 summarizes characteristics of the video data for the 15 flashes examined in this study. For these flashes, the first IB luminosity burst lasts one to five video frames (20–100 µs), and this first burst is followed by at least 3 and as many as 16 visible bursts through at least part of the IB stage. In the normal (i.e., nonhybrid) negative CG flashes, the time between the first IB luminosity burst and first RS varies from 11.5 to 87.8 ms. The three flashes that began as IC flashes are listed last in Table 1; two of these were “hybrid IC-negative CGs” in the terminology of Lu et al. , while the other was an IC flash.
Table 1. Characteristics of Luminosity Observed During Initial Breakdown (IB) in 12 CG Flashes and 3 That Began as IC Flashes (Listed Last)a
Distance is between camera and first LDAR or first LINET point detected during IB, except for 06/14/11 flash which lists distance to first NLDN ground stroke position. Pixel resolution given for video imagery is determined using the listed distance and the camera properties (lens and image size) applicable for each capture.
Altitudes are aboveground level estimated from the video imagery. Ground altitude in the Florida data is less than 100 m above sea level. For the 06/14/11 flash, ground altitude is 1.1–1.2 km msl. “na” denotes Class C cases where the luminosity burst altitude is not analyzable.
See text for definition of Classes; “ex.1,” etc., denote Examples 1–6 described in section 3.
 Among the 15 flashes, the luminosity visible at the time of IB pulses varies in appearance. Seven cases, labeled Class A in Table 1, show a distinct spot followed by an evolving linear feature. We will herein refer to these linear features as initial leaders for reasons we describe below (although they may be the same as the features called streamers by Proctor et al.  and others). The Class A examples are our best in terms of visibility and duration, and a few are described in detail first in section 3.1. Note that these are not simply the nearest flashes in the data set. Three cases (Class B in Table 1) show only a distinct spot of brightness, which is localized over a small region of a few hundred meters or less. In these, we can estimate altitude and determine whether the location of the luminosity changes each time it relights, although we do not see an initial leader develop. A few of these cases exhibit a clear transition to the stepped leader phase; we describe some of their features in section 3.2. Five additional cases (Class C) show diffuse brightness in a large area of the cloud at the time of the first IB pulses, but in these the bursts are so diffuse that we cannot reasonably determine a location of the IB luminosity. Section 3.3 highlights these cases, three of which begin with IC-type IB pulses. At their source, there may be no physical difference between these three varieties of visible luminosity. The differences we perceive in the imagery may be due either to differences in visibility of the source, i.e., its proximity to the camera and cloud edge, or to actual differences in the initiation.
 The flashes studied herein are reasonably typical of our entire data set in terms of their IB pulse E-change amplitudes and frequency. Many of our CG-type IB pulses fit the definition given by Nag et al.  for “classic” pulses, with pulse widths 10–80 µs. Like Nag et al. , we also see many IB pulses that are shorter in duration. Of the more than 14,000 flashes in our Florida data sets, we have closely examined 90 IC flashes and 127 negative CG flashes for IB pulses. All 217 of these flashes had IB pulses. The amplitudes of all the IB pulses (i.e., the entire IB pulse train) in some flashes were so small to only be detectable at a few of the sensors; thus, we would likely not have observed them without a 10 sensor array. These latter two findings confirm the speculation of Beasley et al.  that “these pulses are probably present in all cases…” although “they may sometimes be small and difficult to observe.”
 Since we are able to distinguish luminosity at the time of IB pulses in only a small fraction (~10%) of the total 180 flashes in our video data set, the remaining flashes either initiated outside the camera field of view or too deep within the cloud for the camera to detect any IB light. We guess that this relatively small fraction is due to visibility issues with the video imagery and not due to differences in the IB pulses themselves. This guess is based on the preliminary finding that, for flashes in which we do see IB luminosity, we see it with all the classic IB pulses regardless of their amplitude. Conversely, we have examined a few video flashes with especially large amplitude IB pulses, but no luminosity is visible in the video during the IB stage. Alternatively, the ~90% of our video flashes for which we do not see IB luminosity may not have emitted significant visible light during initial breakdown.
3.1 Luminosity Including Initial Leaders
3.1.1 Example 1
 The first flash we describe with visible luminosity during the initial breakdown pulses occurred on 26 July 2011 at 1909:01 UT. The earliest LDAR point associated with the flash was detected 54.8 ms before the first return stroke, at 3.9 km altitude and 9.6 km distant from the camera. The camera was operating at 50,000 frames/s, with 19.6 µs image exposure time. Image spatial resolution at the distance to the first LDAR source is 24.1 m per pixel. As we show in Table2, the full widths of the first six classic IB pulses range from 23 to 59 µs and the interpulse intervals range from 52 to 313 µs. These values are in good agreement with IB pulse parameters determined by Weidman and Krider  and Nag and Rakov .
Table 2. Time Widths and Interpulse Intervals for Beginning Pulses in Examples 1 and 2 (Data Shown in Figures 2b and 6b)
Classic IB Pulse No.
Begin Time (s)
Pulse Width (µs)
Interval to Next Pulse (µs)
26 Jul 2011, 1909:01 UT
8 Aug 2011, 2133:56 UT
 A distinct light burst is visible at 4.62 km altitude beginning at 1909:01.64495 UT, and the brightness and area of this burst increase in the second video frame (20 µs later, Figure1). (The increase in luminosity observed from the first to the second frame may be due to the spot being bright for only a portion of the first frame then bright during the full exposure of the second frame, or it may be that the spot actually gets brighter just after the beginning.) This luminosity burst is coincident in time with the first classic IB pulse (at 1909:01.64495 UT) in the E-change data (Figure2). At 10 µs before this first classic pulse, there is a short duration (3 µs), smaller amplitude pulse (so these pulses could be within a single video frame or in two adjacent frames). Over the following 95 video frames, i.e., for 1.9 ms, the luminosity pulse develops primarily downward, to 4.21 km, as a series of extending initial leaders or streamers. The extension is not continuous but rather occurs in a sequence of forward surges, each lasting about 80–100 µs. The earliest five surges in this example show a brighter “head” portion illuminating in the second frame after a dim extension. This brightest frame is immediately followed by a backward lighting of the entire “tail,” then a gradual dimming over the next three to four video frames. Then, the sequence starts again with a new, dim extension lighting to a lower altitude point.
 Five sequential video frames from the fourth of these initial leader surges (starting at 1909:01.645274 UT), are shown in Figures 1b–1f. The brightest frame (Figure 1c) coincides with the largest E-change in the fourth classic IB pulse in Figure 2b (also Table 2). This time coincidence is also the case for the other sequences, in which the brighter portion of each burst in luminosity is at the time of the largest E change in the classic IB pulse. A direct comparison of total luminosity per video frame to the E-change data during the IB interval of this flash is shown in Figure 2b. We note that, in this flash, the third classic IB pulse had the largest peak amplitude, about 0.34 V m−1 range-normalized to 100 km. For reference, the peak amplitude of the first return stroke range-normalized to 100 km was 1.22 V m−1; thus, the largest IB pulse was 28% of the amplitude of the first RS.
 The entire sequence of video frames, beginning at the first IB light burst and continuing for 2 ms of initial leader propagation, is shown in Clip 1. Figure3 shows the first 1 ms, along with the corresponding E-change data in detail. The sixth apparent downward burst shows branched luminosity for one frame, after which only the left (east) branch is seen. For the later luminosity surges in this example, after about 600 µs of progression (after 1909:01.645554), the brighter head is not seen first; instead, the entire, gradually lengthening initial leader lights up backward (relative to its progression direction), then dims over four to six frames. Thus, either the head is not as bright as each initial leader progresses further from the initial point or the head in this example has gone within more dense cloud by this time. Note that much smaller and less distinct IB pulses are present in the E-change data after about 1909:01:6455 (Figure 2b). During this time, several short branches are intermittently visible, upward near the initial point and downward along the initial leader, with each branch lit for only one or two frames.
 During the 2 ms these initial leaders are visibly advancing, their estimated 2D speed (or rate of lengthening) decreases from 4.6 to 4.2 to 2.6 to 1.8 × 105 m s−1 averaged over four sequential ~400 µs intervals. Intermittent and dimly lit sections of initial leader are then visible for another 0.6 ms. In the E-change data, the later IB pulses have smaller amplitudes than the first few pulses, consistent with previous work cited above. The second LDAR point of the flash, with an altitude of 4.38 km, occurs near the end time of the visible initial leader progression. Overall, the number of E-change pulses decreases markedly after about 1909:01.647 UT (Figure 2a).
 After luminosity of the last visible section of initial leader fades (at 1909:01.647394 UT), there are no bright luminosity events seen in the video for this flash until 4.5 ms later. At that time, there is a sudden relighting of the earliest, uppermost visible portion of the former initial leader, followed by what appears to be stepped leader-like progression from near the lower end. The stepped leader is mostly obscured by cloud until it becomes evident in the video at 1909:01.657174 UT or 12.2 ms after the first IB pulses were seen in the E-change and video data. Although much dimmer than the IB luminosity surges in this flash, this stepped leader then advances (for 42.7 ms) in two major branches (one downward, one westward and slightly upward) before the first return stroke (RS) occurs at 1909:01.699934. Figure 2c shows an overview of the E-change data from the beginning of the flash through the first RS, including the durations of the IB and stepped leader stages. The LDAR data in Figure 2c also indicate the presence of the two major branches visible in the video data. A portion of the video data for 0.5 ms of stepped leader propagation is shown in Figure4, with the intensity scale greatly enhanced compared to Figure 3. A video of these data is shown in Clip 2.
 The uppermost portion of the visible RS channel (Figure5) is in the same location as the initial leaders that were seen to develop in the 1.9 ms after the first IB pulses and first luminosity. Either the RS terminated at that point or the upper RS channel was obscured by cloud beyond that initial point. From the RS channel in Figure 5, we also see that the stepped leader (and presumably the initial leaders) effectively vanished into more dense cloud for a section between 4.2 and 3.1 km altitude. The other major branch of the stepped leader that propagated westward and upward near the cloud edge only illuminates a short distance during the RS but is not visible in the single frame shown in Figure 5. The CGLSS location of the RS is about 1.5 km south southeast of a point immediately below the first LDAR source location.
 Overall, the character of the linear feature (or initial leader) advance is different from that of stepped leaders seen in our video data (partly evident by comparing Figures 3 and 4, although this faint stepped leader is not our best example). The primary difference is that at each forward advance of an initial leader, the entire length lights “rearward” back to the initial point of the flash luminosity. In between forward surges, the entire initial leader goes dim. Stepped leaders, while also often brightest at the tip during a step, do not always (usually do not) relight the entire leader rearward after they step forward. In addition, compared to the stepped leader, the IB luminosity and initial leaders are much brighter over a longer length, at least after the first frame of each forward surge showing a new, dim extension. The third difference in the initial leaders is that the forward surges are not as frequent, having a longer time interval between them than is usually seen with a stepped leader. One final difference in the linear features is that branches are very rare and do not continue to advance more than one or two frames. As the examples below will reinforce, the IB stage of a flash often has a developing linear feature from which branches are typically absent. This characteristic of the IB bursts is another indication that they are unlike stepped leaders, which are typically described as branched (and relatively dim, as in alpha leaders) or highly branched (and bright, as in beta leaders) [e.g., Schonland, 1956; Beasley et al., 1982].
 Hence, we can tentatively state that in contrast to steps of stepped leaders, the surges of these initial leaders were longer, brighter, more separated in time, unbranched, and relit rearward along the entire developing feature back to the initial point of the flash. For these reasons, we refer to the luminous linear features observed during the IB stage as initial leaders. This use differs from the “streamer” term used in prior work [e.g., Proctor, 1981; Proctor et al., 1988; Rhodes and Krehbiel, 1989], although the features in our data may be the same as those interpreted from VHF data. The issue of whether the linear features shown herein are some type of leader rather than some type of streamer cannot be resolved with our data.
3.1.2 Example 2
 Another flash with visible luminosity during the IB pulses occurred on 14 August 2011 at 2133:56 UT. The earliest LDAR source of the flash was 16.2 ms before the first RS, at 4.9 km altitude and 11.8 km distance from the camera. The camera had the same setup as for Example 1; image resolution here is 29.6 m per pixel at the distance to the first LDAR source.
 In this flash, the first visible light burst is coincident with the first and largest IB pulse in the E-change data (at 2133:56.333780) and occurs about 170 µs before the first LDAR source (Figure6). This burst appears first as a faint spot, then diffuse luminosity extends over a larger area of the next frame. Altitude of the first spot in the video is 4.92 km. The IB pulse amplitude is about 35 V m−1 (peak-to-peak) at the nearest E-change site (Figure 6a). This peak amplitude is about 3.6 V m−1 range-normalized to 100 km or about 51% of the range-normalized amplitude of the first RS (7.1 V m−1). Several relatively diffuse light bursts are visible after this, and several relatively small IB pulses occur in the E-change data. The first LDAR source is nearly coincident with the second IB pulse of 3 V m−1 (0.3 V m−1 range-normalized). Full widths of the first nine classic IB pulses range from 30 to 51 µs, and the interpulse intervals are 16–184 µs (Table 2). These values are within the ranges found by Weidman and Krider  and Nag and Rakov .
 By 380 µs after the first IB pulse and light burst, a 230-m long vertical segment becomes faintly visible, extending downward below the earliest luminosity burst. The brighter lower end of a sequence of initial leaders is then seen intermittently (at intervals of 100–200 µs) over the next 740 µs, as they extend primarily downward to about 4.18 km altitude. During this interval, there are 10 large IB pulses, most of which are accompanied by luminosity of an initial leader (Figure 6b), as well as a few small IB pulses that are not coincident with any apparent luminosity. In a general sense, greater relative intensity within the (cropped) video frame is associated with larger amplitude in the classic IB pulses. However, we caution that a stronger correlation is not possible due to time resolution of the video data (19.6 µs exposure) for short pulses, to the cumulative intensity of longer luminous features later in the IB stage, and to the fact that the width of some IB pulses overlaps several video frames. The second LDAR source, at 5.4 km altitude, is coincident with a large IB pulse (18 V m−1, Figure 6a; 1.8 V m−1 range-normalized) and an initial leader segment. Another two LDAR sources occur during the IB stage, at 4.7 and 4.2 km, the latter of which is 40 µs after an initial leader is last visible at 4.18 km altitude. Over three sequential ~400-µs intervals, the 2D speed of advance of the initial leaders is estimated at 14.0 to 17.6 to 5.6 × 105 m s−1.
 Close examination of Figures 6a and 6b shows that just before the first large IB pulse, there is a relatively slow decrease in E-change that lasts about 20 µs and is indicative of a current moving negative charge downward or positive charge upward (or both). Similar slow E-change decreases (i.e., negative-going slopes) are seen before most of the IB pulses; three are marked in Figure 6b. The prepulse E-changes are coincident with the dim extension of the initial leader path (e.g., first frame in IB pulse sequence described for Example 1). Since the distance from the IB pulses to the sensor shown in Figure 6a is 13.1 km, the E-change is dominated by the induction term [e.g., Uman et al., 1975]. The induction term is proportional to the current; thus the prepulse E-change is caused by a current that presumably is the beginning development of the classic IB pulse.
 In Figure 6a, we also see that, after the first several large IB pulses, a continuous, negative-going E-change also occurs in the intervals between IB pulses, so that later IB pulses appear superimposed on this E-change. Two interpulse E-changes are marked in Figure 6a. These interpulse E-changes are also indications of a current flow associated with the induction term. Thus, the interpulse E-changes are probably caused by weak currents, driven by the ambient E, flowing in the decaying initial leader.
 After the earliest part of the IB stage (lasting about 1.14 ms and ending at about 2133:56.3349 UT), a few more bright luminosity pulses and short segments are visible over the next 2 ms, although their character begins to appear more like an advancing stepped leader. Most of the IB pulses are smaller after 2133:56.3350 UT (Figure 6a). The E-change data indicate the transition from IB to stepped leader probably occurs by about 2133:56.3360 UT, when the pulses become smaller and more frequent. For most of the 3.2 ms after 2133:56.3369 UT, no leader is apparent in the video data; after this time, the stepped leader is seen clearly as it progresses downward to the ground over the following 10 ms (500 video frames). This stepped leader has only three LDAR sources after the IB interval until the RS. The CGLSS position of the first RS is 11.5 km from the camera, about 0.8 km south of a point directly below the first LDAR source. As in the previous example, the initial leader portion which developed during the IB of this flash is part of the return stroke channel and illuminates during the RS.
3.1.3 Example 3
 On 8 August 2011, at 1644:15 UT, the camera captured a more distant flash with luminosity visible at the time of the IB pulses. The earliest LDAR source was over the ocean, 29.8 km distant from the camera and at 4.46 km altitude; it occurred 11.5 ms before the first RS. In this case, the earliest luminosity burst occurred simultaneously with the first large IB pulse and the first LDAR source (Figure7). The luminosity dimmed after the first frame, but was visible for two more frames (total of 60 µs). Photogrammetry indicates the burst (i.e., the brightest pixel) was centered at 5.23 km altitude. The coincident IB pulse has larger amplitude (17 V m−1 at the nearest site; 2.3 V m−1 range-normalized) than most other pulses during the 1.7 ms of IB (Figure 7a). The next luminosity burst occurs 120 µs later at 5.16 km (one pixel, 74.5 m, below the first burst), and it is coincident with the next IB pulse in the E-change data (Figure 7b). Another large luminosity burst and coincident IB pulse occur 60 µs later; this luminosity is again one pixel lower in altitude than the previous pulse. By 140 µs later (320 µs after the first bright spot appeared), the camera data indicate a short linear feature about 335 m long (Figure 7c). Initial leaders then extend intermittently over the next 65 frames in much the same way as described for Examples 1 and 2: a dimly lit extension appears, followed by a bright spot near the new end, followed by relighting of the entire length back to the starting point, then the initial leader dims for one to three frames. Figure8 and Clip 3 show 50 and 90 video frames from this flash for a portion of the initial leader progression. About 1.6 ms after the first bright spot was visible, the final visible initial leader has extended to about 1070 m long (5.23–4.26 km altitude) before it dims out of sight. Estimated 2D speeds of advance in this case vary from 10.0 to 7.1 to 5.9 to 5.1 × 10 5 m s−1 averaged over four sequential ~400 µs intervals. The second of two LDAR sources during the IB stage was detected near the end time of the visible initial leader at 4.26 km altitude.
 The data for this flash (Figures 7a and 7b) show clear examples of the prepulse and the interpulse E-changes discussed with Example 2. The distance to the sensor is 13.7 km, so (as in Example 2) the E-change is dominated by the induction term, and the prepulse and interpulse E-changes are indications of current flow (as in Example 2). In fact, similar E-changes are detected in all of our IB data for flashes close to our E-change sensors.
 The precise transition from initial leaders to stepped leader is not clearly seen in this case, as for the other flashes in our study. Within 100 µs after the last visible initial leader disappears, the first indications of stepped leader type luminosity become visible, beginning as relatively faint spots at various places along and ahead of the former initial leader. The timing of the transition fits well with the change in character of the E-change pulses (at about 1644:15.1655 UT), from larger and less frequent IB pulses to smaller and more frequent stepped leader pulses (Figure 7a). After this transition, there is intermittent and mostly faint video evidence of stepped leader development through the next 435 frames (8.7 ms) as the stepped leader progresses downward, presumably through more dense cloud. About 1.06 ms before the RS, the stepped leader becomes clearly visible as it exits apparent cloud base at 1.3 km altitude. As shown in Figure7d, this first RS relights the channel back to the initial spot, including the initial leader length traced out in the first few ms of the flash. As in Example 1, either the RS terminated near the initial IB burst location or the upper RS channel was obscured by cloud beyond that point. The CGLSS position of the RS is 30.2 km from the camera and 1.9 km south of a point directly below the first LDAR source.
3.1.4 Additional Cases
 As noted in Table 1, four other flashes in our data set show evidence of initial leader development. One case of IB luminosity bursts visible in a distant flash occurred on 28 July 2010 at 1818:46.36463 UT. For this flash, the first LINET source was 46.1 ms before the first RS at 6.6 km altitude and 31.9 km distance from the camera. (There were no LDAR sources detected until 0.4 ms later.) The video resolution at this range was 79.7 m, and the frame interval was 18.5 µs (18.1 µs exposure). At the time of the first IB pulse in the E-change data (Figure9a), there is a faint luminosity burst at 6.9 km altitude. Then, at the time of the second large IB pulse (240 µs after the first pulse) and coincident with the first LINET source, a distinct light burst is visible for 130 µs. During this second interval, the luminosity is visible between 6.6 and 6.9 km altitude (Figure9), and in the video it appears to lengthen upward (i.e., the brighter luminosity apparently extends upward). Thus, in this case, we may be seeing the initial leader develop upward from 6.6 km. Alternatively, after viewing the closer flash examples from 2011 and because the entire length in this case is faint, it is possible that we are seeing the downward-extending initial leader get brighter back up to the initial spot, similar to Example 1 (Figures 1 and 3). This latter interpretation fits better with the fact that the first luminosity in this case was higher, at 6.9 km. If we adopt this interpretation, we can estimate the initial leader speed of extension as about 10.7 × 105 m s−1 over its first 296 µs, which fits with the other cases. The first view of the initial leader dims after 130 µs. Over the next 185 µs, the linear feature appears twice more, faintly, and with the lower end one pixel lower each time. A second LINET source, at 5.7 km altitude, occurs with the third large IB pulse (Figure 9a). After this, there are several more pulses of diffuse luminosity in the same general area of the cloud until 1818:46.365777 UT, which agrees with the sequence of IB pulses (Figure 9a) during this time. The stepped leader first becomes apparent about 25.9 ms later, with only weak speckles of luminosity visible in the intervening 1400 video frames.
 Another flash from the 2010 experiment, captured at 1834:36 on 28 July 2010, is our most distant example in which IB light bursts and a linear feature are visible. In this case the first LINET source was 40.7 km distant and, at 6.3 km altitude, detected 210 µs after the first IB pulse. The earliest luminosity burst appears as spot of light at 5.3 km and is visible in three frames (56 µs), coincident with the second IB pulse. Beginning 74 µs later, a longer and brighter luminosity burst is visible, extending about 200 m lower (i.e., 2 pixels lower). This linear feature is visible again, much brighter and extending another 100 m lower, in the third IB luminosity sequence. This third burst is coincident with the largest IB pulse and the first LINET source. After this, the luminosity in the final burst is visible for nearly 300 µs and is much less distinct (as though the lower end entered thicker cloud). This series of light bursts ends at the time of the last large IB pulse and the second LINET source in the flash. Although we are unable to calculate the initial leader speed precisely or for long, our estimate of 13.4 × 105 m s−1 fits reasonably with the earliest portion of the other examples.
 The flash at 1635:34 on 8 August 2011 is near in both time and character to Example 3. In this case, the first large IB pulse is coincident with a faint burst in luminosity, centered at 5.3 km altitude, nearly coincident with the first LDAR source at 3.8 km altitude. A faint linear feature is first visible 280 µs after the first burst, then a series of initial leaders is visible at intervals for another 1.1 ms. The final time that an initial leader is brightly visible, with lower end near 4.6 km, coincides with the second LDAR source, at 5.5 km altitude. The first RS occurs about 20.4 ms later. (Later in this video, IB luminosity bursts of an IC flash were detected; see section 3.3).
 In the 2212:52 UT case from 14 August 2011, the IB light bursts start much higher in altitude (7.82 km) than in any of the other cases (Table 1). In this flash, the second LDAR source at 7.0 km altitude is nearly coincident with the first large IB pulse and first luminosity burst. Initial leaders propagate downward to about 6.5 km altitude at intervals beginning 260 µs after the first IB luminosity and ending 1.72 ms later. As evident in Table3, this advance is one of the fastest (starting at 13.4 × 105 m s−1) and longest (1365 m) among the seven cases studied. Despite its apparently high initiation altitude, this flash was a normal negative CG.
Table 3. Estimated 2-D Speed of Advance for Initial Leaders Observed During Initial Breakdown in CG Flashes, Using Video With 18–20 µs Image Interval (50,000–54,000 frames/s)
Flash Date, Time
Starting Altitude (km)
First Speed (×105 m s−1), Interval (µs)
Second Speed (×105 m s−1), Interval (µs)
Third Speed (×105 m s−1), Interval (µs)
Fourth Speed (×105 m s−1), Interval (µs)
Ending Length Visible (m)
Max. 2-D Speed Error (%)
Average speed Overall
7.6 all pts
 Summary characteristics for these four cases are included in Table 1, along with those of Examples 1–3. Speed estimates for the seven cases with initial leaders are given in Table 3 for several time intervals. (Time intervals are determined by when the feature is visible in the data, using a value near 400 µs when possible.) Note that the speed estimates are for the lengthening of the initial leaders over many video frames (i.e., through a few forward surges) and not the speed of the head, which is not resolved in our 18–20 µs imaging. We also note that these are two-dimensional speeds, and some of the differences among them are likely a function of the orientation of the progression relative to the image plane. There is no reason to expect, nor do we intend to imply, that the initial leaders move only in two dimensions; any motion toward or away from the camera would result in a slower estimate than the true speed. In addition, we have estimated average speeds based on the change in position at the end of the relevant video frame, so there is time uncertainty of one frame. Spatial accuracy is a function of the image pixel size, which is different for each case and listed in Table 1.
 Table 3 shows that, with one exception, the initial leaders have slower estimated speeds at later times in their visible development. Extreme values among the seven flashes range over an order of magnitude, from 1.8 to 17.6 × 105 m s−1. Most of the estimated speeds are in the range of 4–11 × 105 m s−1, which is a factor of about 2–10 faster than typical stepped leader speeds [e.g., Rakov and Uman, 2003]. The initial leaders are visible over apparent lengths of 331–1435 m in the earliest 241–1980 µs (13–99 frames) after the first IB luminosity burst.
3.2 Luminosity Without Visible Linear Development
 In our data, another type of IB light is characterized by a bright spot, typically appearing a few hundred meters across at maximum. The luminosity of the spot increases over two to three video frames, then dims over the next few frames. In one case, the same spot is visible several times during each large IB pulse. The other two cases show the brightest area in a slightly different location and more diffuse each time it lights up. We speculate that these examples may represent the same process as described above, except perhaps only the brightest “head” portion of the initial leaders is visible at the camera from deeper within the cloud.
3.2.1 Example 4
 During our Rapid City campaign, the video camera captured a CG flash with luminosity at the time of IB pulses on 14 June 2011, at 0233:20 UT. The NLDN ground location of the first RS was 11 km from the camera and 7 km from the other E-change sensor site. The earliest luminosity burst occurred simultaneously with the first large IB pulse (Figure10), 18.8 ms before the first RS. This burst got much brighter after the first frame, then it dimmed over two more frames. Unlike the cases in the previous section, this luminosity is not distinctly localized in a few image pixels; a relatively large area of the cloud near its base lights up. From Figure 10c and later images, we suspect that the source of the earliest IB light is within but not far above cloud base (at 1.54 km above the ridge, which is at an altitude of 1.1–1.2 km msl). The next IB luminosity burst, similar in appearance but brighter and longer than the first, is coincident with the next largest IB pulse. Based on the appearance of the E-change data and the fact that this second luminosity burst is visible in eight frames (160 µs), we might guess that either there is an initial leader developing or there are two or more overlapping luminosity bursts in this time window. Over the next 95 frames (1.9 ms), there are about 13 more light pulses, thereby accompanying most of the significant IB pulses in the E-change data (Figure 10b).
 After the IB luminosity bursts end at 0233:20.2692, there is about 1 ms of essentially no visible luminosity in the video data. Then (at 0233:20.2706) the same area of the cloud base lights up, increasing in intensity over two frames; this light burst (Figure 10d) is coincident with the first step-leader pulse in the E-change data. Within 60 µs a bright stepped leader emerges from what had been the brightest part of the cloud base during the earlier IB luminosity bursts. This highly branched stepped leader then makes its way to ground in 15 ms (750 frames). In the E-change data, these stepped leader pulses are comparable in amplitude to the IB pulses, which is unusual.
 Figure 10a shows a gradual decline in E-change after the end of the IB stage that continues until the stepped leader begins. This relatively slow, negative-going slope in the E-change may be indicative of weak currents continuing to flow in the initial leader.
3.2.2 Example 5
 Another example with distinct IB luminosity but no visible initial leaders is a flash that occurred at 1933:33 UT on 1 August 2011. In this case, the first LDAR source during the IB stage was 16.1 km distant and at 5.0 km altitude, although there were three LDAR points detected earlier, up to about 5 ms before the first IB pulse. (Earlier LDAR and E-change data indicate this flash was preceded by two attempted initiations 45 and 25 ms earlier in the same region of the cloud but higher in altitude; no luminosity was observed at those times.) The earliest luminosity burst appears as a “puff” of light in one frame coincident with the second IB pulse. About 120 µs later, a brighter distinct spot is visible, centered at 6.74 km (Figure11). This event is followed by a relatively long period with infrequent bursts of luminosity which are bright but diffuse over a small region centered at the same altitude as the spot and further left each time. In 1.56 ms, a time coincident with the largest IB pulse, the center of the luminous area has moved about 445 m left (west). Over the entire 2.64 ms period, it moves 714 m, and it is last visible just before the time of the first LDAR source. In view of the examples in section 3.1, one reasonable interpretation of these data is that only the brightest “head” portion of the initial leader is visible at each occurrence, and they are thus advancing at an average speed of 2.8–2.4 × 105 m s−1. Alternatively, because the altitude is quite high and the propagation is horizontal in this case, it could also be that we are seeing less bright luminosity from the upper (positive) end of a developing set of initial leaders in this flash.
 Similar to Example 4 and the others described, this flash shows a relatively long and dark transition between IB luminosity bursts and stepped leader propagation. The final IB luminosity is followed by about 1 ms of darkness; then faint indications of stepped leader development begin in locations to the left of and below the earliest distinct spot. Much of the early stepped leader propagation is not clear in this flash until the leaders exit cloud base. The RS occurs 28.4 ms after the first IB luminosity was seen in a location that shows up brilliantly at the top of the visible RS channel (Figure 11b). In addition, the early, apparently leftward-moving initial leader lights at the top of the visible channel during this RS. As in Figures 5 and 7, either the RS terminated near the initial IB burst location or the upper RS channel was obscured by cloud beyond that point.
3.2.3 Additional Example
 Our final case of localized but diffuse luminosity pulses during the IB stage is a complicated flash that occurred just after 00 UT on 14 August 2011; in Table 1, its time is given as 2401:26 UT on 13 August 2011 (for data management purposes). This hybrid CG flash began as an IC, with an initial set of IB pulses typical of IC flashes. There was no luminosity visible in the video data during this IC part of the IB, although 14 ms after the first IB pulse there is one bright luminosity burst coincident with a large, broad pulse in the E-change data and an LDAR source at 10.5 km. The LDAR data during the IC portion of this flash indicate the IB was about 1 km further away and 1 km higher than the later CG-type IB pulses. Hence, the IC-type IB may have been less bright or simply deeper in the cloud. (As we describe in section 3.3, two other hybrid flashes show luminosity with the IC-type IB pulses but not with the CG-type IB pulses.)
 About 150 ms after the start of the IC portion of the flash, a new set of CG-type IB pulses began. The first LDAR source during this sequence is at 7.7 km altitude and 28.9 km distance from the camera. The first five large IB pulses are not accompanied by evident luminosity. However, the sixth (and largest) IB pulse, about 3 ms after the first, is coincident with a luminosity burst centered at 6.1 km altitude. This burst is brightest in the first frame then dims over the following 80 µs. Five more bursts are visible over the next 2.6 ms; these are more diffuse and light the cloud up to about 6.8 km. A few more similarly diffuse luminosity bursts occur over the next 2.9 ms, then the bursts become much more dim and frequent, indicative to us of possible stepped leader development (though no leader channel is visible). The RS of this flash occurs 82.5 ms later or 87.8 ms after the first visible (CG-type) IB luminosity burst. Unlike in the previous examples, the RS light of this flash does not appear to reach the spot of the earliest luminosity burst until much later—700 µs after the RS begins. The RS channel is not visible there, but the same area lights up as during the earliest portion of the CG-type IB pulses.
 The LDAR source at the time of the first visible IB luminosity is 360 m lower and 2 km closer to the camera than the first LDAR point during this CG portion of the flash. Thus, we speculate that perhaps the visible initial leader got closer to the cloud edge than any of the earlier initial leaders. It is also possible that the first visible IB burst, which was coincident with the largest in E-change amplitude at the nearest sensor, was much brighter than the earlier ones had been. In either case, this example and the others in this section seem to have IB luminosity that is near the border of visibility for our camera. The IB pulses for these flashes are dimmer, deeper inside the cloud, and/or have a different character than those described in section 3.1.
3.3 Diffuse Luminosity Over a Large Region
 The final subset of IB luminosity is composed of five flashes in which there is a sudden luminosity increase over a relatively large area in the video data coincident with IB pulses in the E-change data. These recurring bright events typically last four to six frames (or about 100 µs) and are distinct in luminosity but not in location. We speculate that, in these cases, the initiation spot and possible initial leader progression are too deep inside cloud relative to the camera or outside the camera's field of view. Thus, the bright light emitted at initiation of these flashes is diffuse in the video data, having been effectively spread over a large area by the intervening cloud. Although we have included only five such examples in Table 1, we expect there are more in our data set, as these require a more careful examination of all the video data at the time of IB pulses than we have thus far accomplished.
3.3.1 Example 6
 On 8 August 2011, at 2341:26 UT, the video camera captured a hybrid CG flash that began with IC-type initial breakdown pulses starting 338 ms before the first RS. The first LDAR source during the IB stage was at 8.9 km altitude and 7.64 km from the camera. If we assume the image plane is at this distance, then the top of the video frame is at about 4.3 km altitude. Hence, the source of the IB activity in this case is likely to be above the camera's field of view.
 The first luminosity burst occurred simultaneously with the first IB pulse (Figure12) in this flash. (Note that IC-type IB pulses are different from the CG-type pulses shown elsewhere in this paper; these pulses are bipolar but deflect the E-change positive first, as shown by Weidman and Krider . They are typically fewer in number and farther apart in time than CG-type pulses. Also, in our Ch3 data, the slow E-change slope between IC-type IB pulses is positive-going rather than negative.) The luminosity burst appears as a sudden diffuse brightening (or increased contrast) in the layer between the cloud base and horizon, and it grows dimmer over the following three frames. The next IB luminosity burst (150 µs later), similar in appearance but visible for three frames longer, is coincident with the first LDAR point. Over the next 88 frames (1.76 ms), there are three more light bursts, each visible for one or two frames longer than the previous burst. These main bursts are coincident with the three large IB pulses (with interpulse intervals of about 470 and 610 µs) during this time period. The last two bursts are also near in time to two LDAR sources at 7.6 km altitude. After these five events, several more diffuse IB luminosity bursts are visible. The LDAR data indicate sources up to about 11.5 km over the IC portion of the flash. The CG portion of this hybrid flash starts about 250 ms after the beginning of the flash. There are possible CG-type IB pulses in the E-change data, but no coincident IB luminosity events are visible in the video data.
3.3.2 Additional Examples
 The other four cases with nonlocalized luminosity bursts at the time of IB pulses are similar to Example 6. The only IC flash in our data set for which we have identified IB luminosity bursts occurred on 8 August 2011, about 219 ms after the final part of a CG flash at 1635:34 UT (described in section section 3.1.4). In this IC flash, the earliest visible luminosity burst is coincident with the first IB pulse. Diffuse but bright light appears over a broad area in one frame, then it gets dimmer through the next four frames. This luminosity is also coincident with an LDAR source at 8.1 km altitude which is only 4.4 km distant from the camera site. At this range, the top of the video frame would be at about 2.4 km above ground; thus, we expect that the initiation of this IC flash is not within the camera's field of view. Similar but brighter and longer luminosity events occur twice more over the next 1.94 ms; the last event is visible in 11 frames (220 µs) and is at the time of a large and complicated IB pulse. In the 2.8 ms after the third luminosity burst, four more large E-change pulses are accompanied by light that appears in a different location, above the top of the previous CG flash. These pulses are not clearly identifiable as IB pulses, however, and no other bright luminosity is visible during the IC flash.
 As in Example 6, another hybrid CG flash in which the earliest, IC-type IB pulses are coincident with luminosity bursts in the video data occurred on 13 August 2011 at 2358:42 UT. Unlike Example 6, however, the earliest LDAR source for this flash was at 7.0 km altitude and 29.9 km distance from the camera, so the initiation of this flash was not above the camera's field of view (although it may have been off the side of the image). In this case, a diffuse but bright burst appears in four frames (80 µs), lighting a large area of the cloud. This first burst is 1.2 ms after the first IB pulse, although the coincident IB pulse is larger in amplitude than any of the previous pulses. Diffuse bursts of luminosity are visible twice more in the same general area over the following 2.2 ms. The CG-type IB pulses later in this flash are not accompanied by visible luminosity. The first RS occurs 156.8 ms after the first IB burst, and the RS channel light does not reach that general area until 540 µs later.
 Six luminosity bursts are visible at the time of six large IB pulses, including the first pulse, in a CG flash that occurred at 2322:07 UT on 8 August 2011. The first LDAR source during the IB stage was at 6.4 km altitude just 6 km from the camera; this source would have been about 3 km above the field of view. As in Example 6, the luminosity bursts in this case light a large area of the cloud base (for five or six frames each time), which fits with the light source being above the frame and deeper within the cloud. The first RS of this flash is 6.6 km from the camera and occurs 38.4 ms after the first IB luminosity.
 Our final case of diffuse luminosity occurring coincident with the earliest IB pulses was captured on 14 August 2011 at 2312:17 UT. The first LDAR source during the IB stage was at 5.9 km altitude but only 6.3 km from the camera, so the initiation spot was likely above the camera's field of view (top at about 3.5 km). Luminosity bursts, starting with the first IB pulse that occurs 28.1 ms before the first RS, appear as diffuse lighting over of a large part of the cloud. These bursts are visible with each of the largest IB pulses during the first 2.8 ms of the flash. After this IB stage ends, there is a relatively long period when the stepped leader is likely progressing downward but is not making sufficient light for it to be readily seen in the video. The stepped leaders become visible, appearing first at the top of the frame where the IB luminosity bursts had been brightest, about 9.7 ms before the first RS.
 In this section, we present a hypothesis to explain the sequence of events observed during the IB stage of a CG flash, including the transition to the stepped leader stage. The hypothesis is similar to, but expands upon, the hypothesis of Weidman and Krider , which was based solely on their E-change data. While our hypothesis is based on the data presented above, relevant portions of which we summarize here, it does not explain all aspects of those data.
 The high-speed video data show that bursts of light are coincident with IB pulses in the time-correlated E-change data. These luminosity bursts begin as early as the first IB pulse in some cases, and within the first 2 ms in all 15 flashes. Hence, we tacitly assume that we are seeing light emitted from a location at or near the flash initiation. We further assume that each IB luminosity burst and coincident IB pulse are caused by the same physical event. For 5 of the 15 flashes, including the IC flash and the two hybrid CG flashes that began with IC-type pulses, the IB bursts are apparently too deep within the cloud or outside the field of view of the camera, so these bursts are luminous but not localized in the imagery.
 In seven of the CG flashes, the video data show that the IB bursts include recurrent linear features. The linear features generally evolve in the following sequence: (frame 1) a dimly luminous visible segment extends from the spot of the first IB burst to a location below the previous luminosity burst, (frame 2) the linear feature becomes much brighter at the lowest end, (frame 3) less bright luminosity spreads upward on the entire segment to the site of the first luminosity burst, (frames 4+) the entire visible feature dims. Their apparent 2-D speeds of advance (lengthening) are about 4–18 × 105 m s−1 over the first few hundred microseconds and decrease by about 50% over the first 2 ms. Because they occur during the IB stage, we assume that the linear features seen in the video are initial leaders and not stepped leaders. (However, we remind the readers of our earlier statement that we cannot determine whether these features are leaders or streamers.)
 Each sequence of linear development accompanies a single IB pulse in the E-change data, and the sequence repeats with each large IB pulse. In the cases with best visibility to the camera, each new sequence lights up the entire length of the linear feature of the previous sequence and extends the length downward. Hence, we assume that each new luminous sequence is the visible manifestation of a new initial leader that begins at the same place as previous initial leaders, moves downward along the same path as previous ones, and extends the path downward. A few of the videos of CG flashes also show a 1–2 ms period without visible luminosity, after the IB light bursts end and before the stepped leader is seen.
 Immediately before each IB pulse, there is a small, negative-going, slow E-change. We assume that these prepulse E-changes are associated with weak currents flowing during the first, dim frame of an IB burst sequence. After the first several large IB pulses an additional interpulse negative-going E-change occurs and continues after the end of the IB stage until the stepped leader begins. We assume that these continuous slow E-changes indicate that weak currents are flowing in the initial leader path.
 Based on the above assumptions and the fact that normal negative CG flashes initiate in a large positive E [e.g., Stolzenburg et al., 2007; Stolzenburg and Marshall, 2009], we assume that the IB stage of a CG flash develops in a large, positive ambient E. For example, Coleman et al. [2003, 2008] observed that negative CG flashes initiated in the region just below the main negative charge and just above the lower positive charge, where E was large and positive. Hence, we further assume that the visible initial leaders are negative because they progress downward. There may also be upward-moving positive initial leaders, but we have no video data that unambiguously show these.
 Overall, the bipolar IB pulses seem to emanate directly from the flash initiation, are detected in the E-change data in the frequency band of 0.016 Hz to 2.6 MHz (with a high frequency characteristic wavelength of 100 m), emit detectable light along linear features with lengths of 300–1500 m, and are usually not located by the LDAR system (frequency band of 60–66 MHz and characteristic wavelength of 5 m). These data together suggest that an IB pulse is caused by a substantial current surge that is hundreds of meters long.
 Our hypothesis relies on two additional assumptions: (a) the initial leaders have only weak luminosity and (b) the brightest frames in the video data and the coincident IB pulses in the E-change data are caused by an impulsive breakdown event. This hypothesis can be stated as the IB stage of a normal negative CG flash consists of repeated initial negative leaders, each ending with an impulsive breakdown event at the new lower end of the initial leader, that together produce an ionized path sufficiently long and conductive to start the stepped leader. Under this hypothesis, the development from IB stage to stepped leader can be described as follows:
 The first initial leader moves downward from a point at or near the flash initiation location. In the video, it looks like a short, dim linear feature in the first frame of a sequence (e.g., Figure 3 and Clip 1). In the E-change data the initial leader causes a slow decrease starting less than 50 µs before the first large IB pulse (e.g., Figure 6b); the E-change is caused by negative charge moved downward in the initial leader.
 When the superposition of the initial leader's E due to charge in its “head” and the ambient E exceeds some (undefined) critical value, impulsive breakdown occurs at the lower end of the initial leader. In the video data, this impulsive event produces a bright spot at the lower end of the initial leader in the second frame of a sequence, then upward-moving brightness in the next video frame (e.g., Figure 3 and Clip 1). In the E-change data, the impulsive breakdown causes a large bipolar IB pulse with a “slowly varying” bipolar overshoot [Weidman and Krider, 1979] that may be related to the slowly dimming linear feature at the end of the sequence.
 After the first large IB pulse, there is little or no decrease in the E-change data for a few 100 µs, then shortly before the second large IB pulse, another slow decrease begins (e.g., Figure 6b). We hypothesize this E-change slope is due to development of a new initial leader that follows the path of the first initial leader and extends it farther downward.
 After the second initial leader has ended, the sequences described in (2) and (3) occur again for each large IB pulse. After the first few IB pulses, there is also an interpulse E-change decrease (e.g., Figure 6a) that occurs because of currents in the decaying initial leader channel driven by the ambient E.
 The IB stage ends after the last large IB pulse and is followed a nonluminous period of 1–2 ms. We hypothesize that the IB pulses stop because the previous initial leaders have moved enough charge to reduce the ambient E near the flash initiation to a value such that new initial leaders cannot be initiated.
 During the nonluminous period after the IB ends, there is still a slow decrease in E-change (e.g., Figures 2a and 10a) indicating continued charge transport, presumably driven by the ambient E in the weakly conducting initial leader path. We hypothesize that eventually, the negative charge accumulating at the lower end of the initial leader path will become big enough so that the stepping process can begin. A negative stepped leader has elsewhere been hypothesized to move forward because of the large charge it carries at its tip, so a large ambient E is then no longer needed [Bazelyan and Raizer, 2000]. We hypothesize that as the nascent stepped leader begins to extend, the current flow associated with its extension will heat the channel and thereby increase its conductivity. The idea that negative stepped leaders in CG flashes begin immediately after the IB pulses is not new [e.g., Beasley et al., 1982], but if the above hypothesis is correct, it provides a physical reason for this timing, namely that the IB pulses (and concurrent initial leaders) are needed to start the stepped leader.
 Although this hypothesis only invokes the development and extension of a negative, downward-moving initial leader, the initial leader may be bidirectional with an upward-moving positive end. We can imagine that such a bipolar initial leader would develop parallel to the ambient E, that opposite polarity charges would therefore build up at its ends, and that when E (due to the charges at the ends plus the ambient E) gets big enough in the region just beyond the initial leader ends, then an impulsive breakdown would occur at one or both ends.
5 Concluding Remarks
 Our fast electric field change data indicate that all flashes begin with initial breakdown pulses. The data presented herein show that luminosity, bright enough to be visible at distances up to 40 km or so through intervening cloud, accompanies the IB stage at the beginning of cloud-to-ground and intracloud lightning flashes. Bursts of light in high-speed video data are visible during the IB stage in 15 of 180 flashes examined with time-correlated observations, and these bursts are coincident with distinct IB pulses in the E-change data. The lack of IB bursts in the video for ~90% of our video-captured flashes is probably due to camera visibility issues (distance from camera, cloud depth to initiation, incorrect camera settings, IB location outside the field of view, etc.) and not necessarily to the lack of IB luminosity. Some of the IB luminosity bursts in the video imagery are also near in time and altitude to VHF (LDAR) or VLF/LF (LINET) sources. Some of the IB bursts are surprisingly bright and more intense than the early part of the stepped leaders seen in the same flashes. The more diffuse IB bursts we observe must also be bright at the source, since these illuminate relatively large portions of cloud in the imagery, despite substantial intervening cloud or despite their source being outside the camera's field of view. Although our sample size is small, the luminosity observed during IC-type IB pulses (of IC flashes or hybrid CG flashes that begin as IC flashes) looks generally similar to that observed during CG-type IB pulses of CG flashes.
 It is surprising to us to find distinct bursts of luminosity and visibly developing linear features so near in time and space to the presumed initiation of lightning flashes. The bright bursts and linear features are observed from within clouds, where the optical depth should be short. These results suggest that the sources of the luminosity (i.e., flash initiation and initial leaders or streamers) may be very bright and/or near the cloud edge. It is also surprising to note that, based on the video data, either the first RS terminates near the initial IB burst location or the uppermost part of the first RS channel is obscured by cloud beyond that point. The visible sequences of development of the linear features during the IB stage are quite unlike those of stepped leaders observed later in the same flashes: they are longer, brighter, more separated in time, light up the entire feature back to the initial point of the flash, advance 2–10 times faster than typical stepped leaders, and their speeds of advance decrease as they lengthen then disappear. In our data, there is usually a quiet period of 1–2 ms, with neither bright luminosity nor large E-change pulses, between the end of the IB and the beginning of the stepped leader. We offer a hypothesis regarding the physical processes occurring during the IB stage of CG flashes to explain some major features of these observations.
 This project was supported by the NASA/Mississippi Space Grant Consortium (grants NNG05GJ72H and NNX07AM36A) and the National Science Foundation (grants AGS-1016004 and AGS-1110030). Additional funding for the SD campaign came from the UM Physics Department. The camera was made available via NSF grant AGS-0813672. We particularly thank F. Merceret, J. Madura, J. Wilson, C. Maggio, C. Conn, J. Dwyer, M. Schaal, M. Bickett, and A. Detwiler for important help. All the sensor site hosts are also greatly appreciated: Florida Institute of Technology Department of Physics and Space Sciences (Melbourne), Hickory Tree Elementary School (St. Cloud), Brevard Community College Planetarium (Cocoa), Titusville-Cocoa Airport Authority, Wedgefield Golf Club, St. Luke's Lutheran School (Oviedo), Massey Ranch Airpark (Edgewater), and Fairfield Inn & Suites (Titusville). We also thank the anonymous reviewers for their many helpful comments to improve the paper.