Atmospheric Research 135136 (2014) 285305

Contents lists available at ScienceDirect

Atmospheric Research
journal homepage: www.elsevier.com/locate/atmos

High-speed video observations of natural cloud-to-ground

lightning leaders A statistical analysis
Leandro Z.S. Campos a,, Marcelo M.F. Saba a, Tom A. Warner b, Osmar Pinto Jr. a,
E. Philip Krider c, Richard E. Orville d
a
b
c
d

Instituto Nacional de Pesquisas Espaciais, So Jos dos Campos, SP, P. O. Box 515, 12227-010, Brazil
Department of Atmospheric Sciences, South Dakota School of Mines and Technology, Rapid City, SD 57701, USA
Institute of Atmospheric Physics, University of Arizona, Tucson, AZ 85721, USA
Department of Atmospheric Sciences, Texas A&M University, College Station, TX 77843, USA

a r t i c l e

i n f o

Article history:
Received 16 February 2012
Received in revised form 23 October 2012
Accepted 21 December 2012
Keywords:
Lightning
Stepped leader
Dart leader
Positive leader
Recoil leader
High-speed camera

a b s t r a c t
The aim of this investigation is to analyze the phenomenology of positive and negative
(stepped and dart) leaders observed in natural lightning from digital high-speed video
recordings. For that intent we have used four different high-speed cameras operating at frame
rates ranging from 1000 or 11,800 frames per second in addition to data from lightning
locating systems (BrasilDat and NLDN). All the recordings were GPS time-stamped in order to
avoid ambiguities in the analysis, allowing us to estimate the peak current of and the distance
to each flash that was detected by one of the lightning locating systems. The data collection
was done at different sites in south and southeastern of Brazil, southern Arizona and South
Dakota, USA. A total of 62 negative stepped leaders, 76 negative dart leaders and 29 positive
leaders were recorded and analyzed. From these data it was possible to calculate the
two-dimensional speed of each observed leader, allowing us to obtain its statistical distribution
and evaluate whether or not it is related to other characteristics of the associated flash. In the
analyzed dataset, the speeds of positive leaders and negative dart leaders follow a lognormal
distribution at the 0.05 level (according to the ShapiroWilk test). We have also analyzed how the
two-dimensional leader speeds change as they approach ground through two different
methodologies. The speed of positive leaders showed a clear tendency to increase while negative
dart leaders tend to become slower as they approach ground. Negative stepped leaders, on the
other hand, can either accelerate as they get closer to ground or present an irregular development
(with no clear tendency) throughout their entire development. For all the three leader types no
correlation has been found between the return stroke peak current and the average speed of the
leader responsible for its initiation. We did find, however, that dart leaders preceded by longer
interstroke intervals cannot present speeds of the order of 107 m s1. Finally, we have analyzed
the impact of recoil activity during positive leaders over their average speed and the return stroke
peak current. Even though the analysis considering the leader speed was not conclusive it was
possible to show that there is no apparent minimum or maximum peak current value for return
strokes preceded by leaders with or without recoil activity (considering the most common range
of values in literature, 2080 kA).
2013 Elsevier B.V. All rights reserved.

1. Introduction
Corresponding author. Tel.: +55 123 916 4253.
E-mail addresses: leandro.zanella@gmail.com,
leandrozscampos@outlook.com (L.Z.S. Campos).
0169-8095/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.atmosres.2012.12.011

In negative cloud-to-ground (CG) flashes, first strokes and

subsequent strokes that follow a new channel to ground are
preceded by stepped leaders. During its development it presents

286

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

very faint individual steps, responsible for its terminology. In

early works using streak-cameras (Boys, 1926, 1929), these
steps could only be observed through an intensification of the
negative (e.g., Schonland et al., 1935; Schonland, 1956; Berger
and Vogelsanger, 1966). The length for each step may vary from
3 to 200 m (typically 50 m) with step interval ranging from 5 to
100 s (Schonland, 1956; Berger, 1967; Berger and Vogelsanger,
1966; Yokoyama et al., 1990; Chen et al., 1999). Previous studies
have also showed the presence of electric field pulses prior
to the return stroke associated with the luminous step
formation of the leader phase (e.g., Appleton and Chapman,
1937; Chapman, 1939; Pierce, 1955; Kitagawa, 1957; Krider
and Radda, 1975; Krider et al., 1977; Beasley et al., 1982,
1983). Studies based on various methods have reported
stepped leader speeds ranging from 0.8 10 5 to 19 10 5
m s 1 (e.g., Schonland, 1938, 1956; Schonland et al., 1938a,b;
Berger and Vogelsanger, 1966; Nagai et al., 1982; Orville and
Idone, 1982; Beasley et al., 1983; Thomson et al., 1985; Proctor
et al., 1988; Mazur et al., 1995; Shao et al., 1995; Chen et al.,
1999), in good agreement with the results obtained for long air
gap discharges in laboratory scale (Les Renardires Group,
1981). Electric field measurements have given leader durations
ranging from a few milliseconds to tens of milliseconds, with
median values varying from 8 to 40 ms from different studies
(e.g., Schonland et al., 1938a; Pierce, 1955; Clarence and Malan,
1957; Kitagawa and Brook, 1960; Beasley et al., 1982; Rakov
and Uman, 1990b). In a more recent paper, Hill et al. (2011)
documented the occurrence of space stems ahead of the
stepped leader tip with the help of a digital high-speed camera
working with temporal resolution of 300,000 frames per
second. They have found an average interstep interval of
16.4 s and step lengths with an average of 5.2 m.
Subsequent return strokes in CG flashes that follow a
channel developed by a previous stroke are preceded by either
dart or dart-stepped leaders. Dart leaders seem to have a
continuous development towards the ground (Schonland et al.,
1935; Schonland, 1956; Jordan et al., 1992, 1997), in opposition
to the step-like nature of stepped leaders (described above). On
the other hand, in some cases the dart leader develops in a
stepped fashion during the final stages of its propagation. Such
events are called dart-stepped leaders. Optical measurements
of natural lightning have provided dart leader mean speeds
ranging from 5.5 106 to 19 106 m s1 (Schonland et al.,
1935; McEachron, 1939; Winn, 1965; Orville and Idone, 1982;
Jordan et al., 1992; Mach and Rust, 1997). The results obtained
for triggered lightning are very similar, with mean values
ranging from 1.1 107 to 2.0 107 m s1 (Hubert and Mouget,
1981; Idone et al., 1984; Jordan et al., 1992; Mach and Rust,
1997). As pointed out by Rakov and Uman (2003, p. 165), all
studies cited above provided mean values between 1 107 and
2 107 m s1 with the exception of Schonland et al. (1935),
who obtained 5.5 106 m s1 from 55 measurements made in
South Africa. Schonland et al. (1935), Table 8) have also pointed
that slower dart leaders tend to be preceded by longer
interstroke intervals. Previous works seem to diverge about
whether dart leader speeds increase or decrease as it approaches the ground; Schonland et al. (1935) and Orville and
Idone (1982) have observed both situations and Jordan et al.
(1992) noted a decrease tendency on average. When studying
rocket-triggered lightning flashes using the ALPS digital optical
imaging system (detailed by Yokoyama et al., 1990), Wang et al.

(1999) observed a speed increase in both cases they have

analyzed (one dart and one dart-stepped leaders). On the other
hand, Mach and Rust (1997) have not found any significant
change in the speed in a dataset of 35 natural and 26 triggered
dart leaders.
While negative leaders are deeply known, data on positive
leaders are still very sparse. The first speed measurement was
made by Berger and Vogelsanger (1966) using a streak camera.
The observed case increased its speed from 3.6 105 to 24 105
m s1 as it approached ground and had its polarity identified
from its luminosity, similar to what had been observed in
positive upward leaders initiated in towers. Later, laboratory
scale experiments made at Les Renardires Group (1977)
obtained speeds ranging from 0.12 105 to 0.42 105 m s1,
slower than the case observed by Berger and Vogelsanger
(1966) by two orders of magnitude. This discrepancy has
affected directly the conclusions made by Mazur et al. (1998)
and is eventually presented as an example of polarity
asymmetry in lightning physics (e.g., Williams, 2006). From
the studies made by Berger and Vogelsanger (1966) and Les
Renardires Group (1977) until recent works based on highspeed video (made by Saba et al., 2008, and Kong et al., 2008)
only indirect measurements (based on channel length and
leader duration) were reported in literature (e.g., Brook and
Ogawa, 1977; Fuquay, 1982). Kong et al. (2008) have observed
only one natural case in China with a camera set to 1000 frames
per second and obtained a speed increase from 0.1 105 to
3.8 105 m s1. Saba et al. (2008), based on recordings made in
Brazil and the USA at frame rates ranging from 1000 to 8000
frames per second, obtained speeds of 9 natural positive leaders
varying from 0.23 105 to 13.0 105 m s1 with a mean of
2.5 105 m s1. Even when including the results obtained by
Berger and Vogelsanger (1966) in their statistics they showed
that positive and negative (stepped) leader speeds are essentially within the same range. For 8 out of the 9 cases reported by
Saba et al. (2008) it was possible to obtain multiple speed
measurements. That has allowed them to observe an increase in
the speed of positive leaders as they approach the ground by a
factor ranging from 1.1 to 6.5. Even though Les Renardires
Group (1977) have observed that laboratory scale positive
leaders develop continuously, some authors (e.g., Laroche et al.,
1988; Idone, 1992; Biagi et al., 2011) reported that the upward
positive leader initiated at the top of the grounded wire of a
triggering rocket develops in a stepped manner. There have been
no reports of observations of natural cloud-to-ground positive
leaders made at temporal resolutions above tens of thousands of
frames per second that could confirm if they have a stepped
nature as well.
Finally, a number of researchers (e.g., Ogawa and Brook,
1964; Richard et al., 1986; Rakov et al., 1992) have reported the
occurrence of K changes or recoil streamers associated with
intracloud discharges and the in-cloud development of CG
flashes during the time between strokes. Shao et al. (1995)
observed these K discharges with the help of VHF imaging
systems, estimating their propagation speeds (between 106 and
107 m s1) and noticing their relationship with the initiation of
M components during the continuing current period of a CG
flash. More recently, Saba et al. (2008) identified the occurrence
of bright, short duration recoil leaders (following the terminology suggested by Mazur, 2002) throughout the development of
some downward positive leaders to ground recorded with the

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

help of high-speed digital cameras. It was possible to observe

that at least some of the recoil leaders (RLs) retraced channel
segments that were previously ionized by the positive leader.
According to Saba et al. (2008, p. 3), in some of the positive
leader events that they analyzed the RLs began 45 to 120 ms
before the occurrence of the return stroke and most of them are
visible for only one frame (135250 s); they also appear as
bright channel segments (compared to the faint positive
leaders). As exemplified in their Fig. 4d,e and f (Saba et al.,
2008, p. 4), cases of RLs that lasted more than one frame could
have their minimum speed estimated (about 410 6 m s1, in
agreement with the works by Mazur, 1989, 2002, and Shao et al.,
1995, who reported RLs speeds on the order of 107 m s1) and
made it possible to observe that they retrace parts of the path of
the positive leader in a retrograde way, i.e., towards the leader
origin. Recoil leaders have also been identified in upward
leaders (apparently positive) initiated by tall towers, as reported
by Mazur and Ruhnke (2011) and Warner (2012), with a very
similar phenomenology. Finally, Mazur et al. (2011) and Warner
et al. (2012) have provided evidences that recoil leaders are
bipolar with bidirectional development, with the negative end
propagating towards the origin point of the flash.
2. Instrumentation and dataset
2.1. High-speed cameras
The data presented in this work were provided by five
different high-speed cameras (Photron FASTCAM 512 PCI, Red
Lake MotionScope 8000S and Vision Research Phantom v7.1,
v310 and v12.1) with temporal resolutions and exposure times
varying from 83 s (11,800 frames per second) to 1 ms (1000
frames per second). They were used in various campaigns
aiming to study CG flashes in southern and southeastern Brazil
(Ballarotti et al., 2005; Saba et al., 2006; Campos et al., 2007,
2009), southern Arizona (Saba et al., 2009, 2010; Saraiva et al.,
2010) and western South Dakota (Saba et al., 2008; Warner,
2012; Warner et al., 2011), USA, between February 2003 and
September 2009. All three cameras are GPS synchronized and
provide time-stamped images with no frame-to-frame brightness persistence.
The high-speed cameras used have a triggering system
based on a signal from an external source. For the present
study this signal came from a button pressed by the camera
operator. It is possible to set the pre- and post-trigger time
within the total recording time of approximately 2 s, which
has proven to be long enough to prevent the first strokes to
be missed and allow the complete recording of the lightning
flash considering its total duration (Saraiva et al., 2010).
As each event was observed by a single, standalone
camera, the leader speed estimates we have obtained was
two-dimensional (2-D), considering only the branching and
tortuosity that occurs on the plane of the image. Considering
that the channel morphology details that are manifested
perpendicularly to the plane of the image are unknown, the
2-D speeds are an underestimate of the real three-dimensional
(3-D) speeds through which the leaders develop towards
ground. One possible way to minimize the discrepancy between
the estimated 2-D and real 3-D speed is to apply a corrective
factor considering a cylindrically uniform distribution of channel
tortuosity based on the results by Hill (1968) and Idone and

287

Orville (1988). However, instead of using a fixed tortuosity

factor (defined as the ratio between the total length of the curve
to the distance between its extreme points) to do such
correction (e.g., as previously done by Nesbitt et al., 2000), we
decided to present the uncorrected, underestimated 2-D speeds.
Therefore, future works presenting a 3-D reconstruction of the
lightning channel from multi-station camera arrays will be able
to either confirm or refute some of the results of the present
paper.
In-depth discussions on the accuracy of high-speed
cameras for the determination of lightning parameters are
presented on previous works by Ballarotti et al. (2005), Saba
et al. (2006), Campos et al. (2007, 2009) and Campos and
Saba (2009). One recurring discussion presented in their
studies is the issue of processes whose occurrence and
duration are shorter than the exposure time of an individual
frame. In such cases all that can be determined is the
maximum duration additionally to occurrence-related parameters (as in the case of the M component analysis made
by Campos et al., 2007, 2009). For the present study, the
understanding of this limitation has allowed us to exclude
dart leaders that were so fast that could only be visible in a
single frame, which would allow only the estimate of the
minimum propagation speed. For non-statistical studies this
parameter can still be relevant, as in the case of the recoil
leader minimum speeds presented by Saba et al. (2008).
2.2. Lightning location systems
For the determination of two-dimensional speeds of lightning leaders it is necessary to know the geometric characteristics
of the camera and the lenses used, and the distance between the
observation site and the flash. This last parameter, as well as
stroke polarity and return stroke peak current estimate, was
obtained through data provided by lightning locating systems
(LLS). The stroke matching between high-speed camera and LLS
data was done based on the GPS synchronization (Saba et al.,
2006). All the observation sites used to obtain the data presented
in this work were located in regions that are well covered by LLS:
BrasilDat for Brazil (Pinto et al., 2006, 2007) and the NLDN for the
USA (Biagi et al., 2007; Cummins and Murphy, 2009). We
estimate that the errors produced by LLS uncertainties (Jerauld et
al., 2005; Biagi et al., 2007) in the calculated 2-D speeds range
from 3% to 30%.
2.3. Dataset overview
In the present paper we have analyzed a dataset of 62
negative stepped leaders, recorded at frame rates that ranged
from 1000 to 8000 frames per second in southeastern Brazil
(So Jos dos Campos, So Paulo) and southern Arizona
(Tucson). Additionally, 76 dart leaders have been recorded at
either 4000 or 8000 frames per second at the same observation
sites as the stepped leaders. Finally, we have analyzed 29 cases
of positive leaders observed in southeastern (So Jos dos
Campos, So Paulo) and southern (So Martinho da Serra, Rio
Grande do Sul) Brazil, southern Arizona (Tucson) and western
South Dakota (Rapid City), USA. These cases were observed at
temporal resolutions that ranged from 1000 to 11,800 frames
per second and include the dataset presented previously by
Saba et al. (2008, 2010).

288

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Considering all observation sites, the analyzed leaders

occurred at distances that ranged from 1.5 to 53 km from the
camera. Fig. 1 shows the distribution of distances considering
all 167 events that occurred at all four sites. The modal range
(the range of channel length values with the highest number
of observed cases) is 10 to 15 km and more than 90% of the
dataset touched the ground at points closer than 35 km from
the cameras.
We also present estimates of visible channel lengths
derived from the all the leader analyses. Table 1 summarizes,
for each observation site, the minimum and maximum
lengths, the arithmetic mean (AM), median, geometric
mean (GM) and modal range of each dataset. For leaders
that occurred at extremely close range from the cameras or
that were highly obscured by the opaque region of the
thunderstorm cloud had very small visible channel lengths,
sometimes below 500 m. On the other hand, in some events
the leader became visible after leaving the thunderstorm
cloud through its side, therefore presenting itself as being
longer than the cloud base. Such events were responsible for
the maximum length values observed on each site.
So Jos dos Campos, So Martinho da Serra and Rapid
City had similar AM, GM and median values (roughly
between 1300 and 1700 m) for the visible estimated channel
lengths for the leaders observed in those locations. Tucson,
however, allowed us to observe considerably longer channels, with its AM, GM and median lying roughly between
2400 and 2600 m.
3. Results and analyses
In the present work we will use the terminology previously
defined by Saba et al. (2008), who has termed partial 2-D

speeds those estimated between two or more consecutive

frames of the video record and average 2-D speeds the ratio
between the total channel extension and the time spent by the
leader to cover it.
3.1. Negative stepped leaders
3.1.1. Two-dimensional speed statistics
As mentioned previously, a dataset of 62 negative stepped
leaders were observed and analyzed. From these events it was
possible to obtain 371 partial 2-D speed estimates. Table 2
summarizes the overall statistical parameters obtained from
this analysis. Considering that the slower cases would present a
greater number of partial speed values, we did not present
averages, medians and geometric means for them in order to
avoid a statistical bias towards lower values. The same decision
was made concerning the distributions: Fig. 2 presents only the
histogram of the average speeds of the 62 analyzed cases. As
seen in the following sections, this care was also taken on the
other types of leaders.
We also present the dataset of stepped leaders for each
observation site in a compact form with the help of Table 3. It
is not possible to determine if the events recorded in So Jos
dos Campos are indeed faster than those observed in Tucson
due to the fact that the latter accounts for less than 7% of the
complete dataset. However, it is possible to notice that all the
range of speeds observed in Tucson fit into the range of the
larger So Jos dos Campos dataset.
When we have applied the statistical test developed by
Shapiro and Wilk (1965) to both samples of speeds (partial
and average) it was possible to conclude that they were not
significantly drawn from a lognormal distribution at the 0.05
level. More than 90% of the sample of average speeds

Fig. 1. Distribution of distances from the camera to the leader strike point, as estimated with the help of lightning location systems. The complete dataset was
included (167 events, from all four observation sites).

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

289

Table 1
Summary of the visible channel lengths estimated at each observation site for the whole dataset (all leader types). Min stands for minimum, Max for maximum,
AM for the arithmetic mean and GM for the geometric mean.
Location

Visible leader channel length (m)

So Jos dos Campos (southeastern Brazil)

So Martinho da Serra (southern Brazil)
Tucson (southern Arizona, USA)
Rapid City (western South Dakota, USA)

Sample

Min

Max

AM

Median

GM

Modal range

97
16
39
15

353
319
804
481

4964
3097
5509
3718

1514
1709
2612
1617

1346
1418
2443
1084

1301
1421
2387
1352

10001500
10001500
15002000
5001000

Table 2
Statistical parameters of the 62 negative stepped leaders. Min stands for
minimum value, Max for maximum value, AM for arithmetic mean and GM
for geometric mean.
2-D speed ( 105 m s1)

Partial
Average

Sample

Min

Max

AM

Median

GM

371
62

0.26
0.90

19.8
19.8

3.30

2.24

2.68

presented values below 6.0 10 5 m s 1, coherent with the

range of values presented by other researchers (Schonland,
1956; Berger and Vogelsanger, 1966). The lowest speed
found (2.6 10 4 m s 1) is very close to the one reported in
literature, 3 10 4 m s 1, presented by Proctor et al. (1988)
from VHF observations. Beasley et al. (1983) observed one
case whose development speed was 3.9 10 6 m s 1 at
approximately 100 m from the ground, almost twice as fast
as the highest speed found in our sample of partial speeds
(1.98 10 6 m s 1).
Previous works using high-speed digital cameras developed by Qie and Kong (2007), who studied a single case with
four grounded branches, and Lu et al. (2008), who studied

four cases, presented results that ranged from 0.7 10 5 to

23 10 5 m s 1. Lu et al. (2008) also analyzed three cases of
negative stepped leaders that propagated towards the ocean
and two of them presented speeds of the order of 10 6 m s 1.
They state that this is quite different from the land CG whose
speed is mostly at the order of 10 5 m s 1, suggesting that
they believe that there is a possibility that a great mass of salt
water can benefit the propagation of faster leaders. We
believe that the analysis of a considerably larger dataset is
required to see if there is indeed a statistical difference
between these cases and those that develop over a strictly
continental area (discussed in the present work).
The cases included in the present statistical analysis were
coherent with previous studies based on different techniques
(discussed at the Introduction of the present paper).
3.1.2. Stepped leader speed change with height
We have analyzed how the partial 2-D speed of the 62
stepped leaders change as they get closer to ground. Two
distinct approaches were used: in the first one we present
the distribution of speed values for all the cases within given
ranges of heights; and on the second one each case was
categorized individually in three possible classifications,
whose occurrence was analyzed statistically.

Fig. 2. Histogram of average 2-D speeds of 62 negative stepped leaders.

290

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Table 3
Same as Table 2 but including only average speeds and considering each
observation site (So Jos dos Campos and Tucson) separately.
Location

Average 2-D speed ( 105 m s1)

Sample Min

So Jos dos Campos

58
(southeastern Brazil)
Tucson (southern Arizona,
4
USA)

Max

0.90 19.8
1.36

AM

Median GM

3.40 2.25

2.76

2.39 1.87 1.86

1.81

For the first approach each case has provided us only

one speed value per each pre-determined range of heights
(e.g., from 0 to 500 m, 500 m to 1000 m, 1000 m to 1500 m,
and so on). In order to avoid statistical bias towards lower
values we have calculated an average speed of each case for
each range of heights, allowing us to have an equal contribution from all the events included in the dataset. Fig. 3
presents the results of this analysis, showing a scatterplot of
speed versus height for all 62 stepped leaders. The average
speed of each case is represented by the + symbol and the red
circles correspond to the average value of all the average
speeds of individual cases that were available for each given
range of heights. On one hand, as shown in Fig. 3a (which
includes all the considered dataset), there is a subtle tendency
for the maximum speeds at each range of heights to increase as
the leaders get closer to ground. On the other hand, there is
also an accumulation of cases presenting lower speeds, below
the average, throughout all the height ranges analyzed. This
behavior can be seen in greater detail with the help of Fig. 3b,
which presents a zoom into the range of speeds below
6 105 m s 1 (which includes 90% of the observed average
speeds, as shown in Section 3.1.1). It is possible to notice that
the average speeds for each range of heights seem to oscillate
around 2.5 10 5 m s1 and even though individual cases can
accelerate (as discussed later in the present section) it seems
that there is no universal tendency within the analyzed dataset
that suggests whether a pronounced increase or decrease in
the speeds of stepped leaders as they get closer to ground.
In the second approach we have analyzed each case
individually, plotting propagation speed versus time graphs
and categorizing them into three categories: (i) accelerated, for those cases that presented a clear acceleration,
concluding their visible propagation with a partial speed
higher than its average speed; (ii) decelerated, for those
cases with a final speed lower than the initial and the average;
and (iii) irregular, for those cases in which we have observed
a number approximately equal of partial speeds below and
above the average speed throughout the entire propagation,
with no clear tendency. Fig. 4 presents an example of each
category with the first example (Fig. 4a, which accelerated)
being a positive leader and the other two negative stepped
leaders.
As five cases presented only one partial speed measurement,
this analysis could be done on the remaining 57 negative
stepped leaders that are included in our dataset. Table 4 presents
a summary of the results obtained from the categorization of
each case. It can be noted that the occurrence of cases that
decelerated is very low (approximately 9% of the sample). On

the other hand, the amount of cases which accelerated and

presented an irregular development were very close (42% and
49%, respectively), indicating that, in general, the acceleration of
negative stepped leaders is not as pronounced as suggested by
previous works based on smaller samples (e.g., Nagai et al.,
1982; Qie and Kong, 2007; Lu et al., 2008).
3.1.3. Average stepped leader speed versus estimated
peak current
For 51 of the 62 analyzed cases it was possible to relate their
average 2-D speed to the estimated peak current of the return
stroke initiated by each leader. These values were estimated
through the data provided by the lightning location systems.
For the speed estimates of the 11 events that were not included
we have used the strike distance estimated for subsequent
strokes that followed the same channel to ground. Fig. 5
presents a scatterplot of these two parameters. A brief analysis
of this graph shows that the majority of the cases whose return
stroke peak current was below 20 kA were initiated by stepped
leaders with average speeds below the arithmetic mean of the
whole dataset (3.3 105 m s 1, as shown in Table 2). However, the existence of extremely fast cases at this range of currents
(i.e., events with peak currents close to 20 kA and average
speeds of the order of 10 6 m s 1) shows that there is no
well-defined cutoff regions in this graph for the available
dataset. Future works would add to this analysis if a greater
number of cases whose peak current is above 50 kA are
analyzed, which would allow us to know if stepped leaders that
initiate return strokes with such peak currents always present
speeds above 5 105 m s1, as suggested by the cases we were
able to study. It is worth mentioning that we have analyzed the
errors of the extreme cases (related to uncertainties in the
strike point distance) and in only one of them it was greater
than 15%. As we did not find any similar analysis in the
literature review made for this paper it is not possible to
compare it to previous results. Further analysis would be
greatly improved if data from LMA (Lightning Mapping Array),
a 3-D total lightning location system (Rison et al., 1999), is
available in addition to the high-speed camera records. Such
observational setup would allow us to estimate the total
channel length, making it possible to analyze whether it has
any influence over the peak current and compare it to the effect
of leader speed. This expected correlation is based on the
results presented by Proctor et al. (1988), who have observed
that the leader charge is proportional to the channel length
estimated with the help of a VHF imaging system.
3.2. Dart leaders
3.2.1. Two dimensional speed statistics
Our sample of dart leaders consist in 76 cases observed in
So Jos dos Campos, Tucson and So Martinho da Serra. From
their analysis it was possible to obtain 207 partial speed
measurements (following the terminology presented previously by Saba et al., 2008). Table 5 presents a statistical summary of
this dataset. For the same reason discussed for negative stepped
leaders, we have only presented the minimum and maximum
values for the partial speeds. It is important to mention that, in
the lack of electric field data and cameras with temporal
resolutions high enough to resolve individual leader steps, it

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

291

Fig. 3. (a) Variation of leader 2-D speeds with height for 62 negative stepped leaders; (b) zoom into the region of speeds below 6 105 m s1.

was not possible to know whether a given leader was either a

dart or a dart-stepped leader.
A brief evaluation of the dart leaders according to the
observation site is presented in Table 6. Even though the
fastest case of the whole dataset was observed in So Jos dos
Campos, the values of the arithmetic mean, median and
geometric mean suggest that the dart leaders recorded in
Tucson were considerably faster. The considerably smaller
dataset of So Martinho da Serra (8% of the complete sample)

makes it difficult to compare its events to those observed at

the other sites. It is worth noticing, however, that its slowest
case was faster than the minimum values of So Jos dos
Campos and Tucson by factors of 4.5 and 2.7, respectively.
Fig. 6 shows the histogram of the average speeds obtained for
the 76 cases. After using the statistical test developed by Shapiro
and Wilk (1965) over the dataset we have concluded that both
partial and average speed measurements were significantly
drawn from a lognormal distribution at the 0.05 level.

292

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Fig. 4. Examples of categorization of individual leaders into (a) accelerated, (b) decelerated and (c) irregular. The vertical line indicates the average speed for each case.

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Table 4
Statistics of the individual analysis of the speed variation with height of 57
negative stepped leaders.
Behavior

Number of events

Percentage

Accelerated
Decelerated
Irregular

24
5
28

42
9
49

When comparing the statistical distribution of our dataset

with the literature we can note that there is a great similarity
with the results obtained by Schonland et al. (1935), from
streak camera observations, who had obtained an arithmetic mean of 5.5 10 6 m s 1. On the other hand, as stressed
by Orville and Idone (1982), other researchers had obtained
higher values than those reported by Schonland et al.
(1935), with arithmetic means of the order of 10 7 m s 1
(e.g., McEachron, 1939; Winn, 1965; Berger, 1967; Hubert
and Mouget, 1981), indicating that the limited temporal
resolution of our dataset (4000 and 8000 frames per second)
might have induced a bias towards obtaining records of
slower cases. However, two cases presented speeds above the
upper limit of the histogram presented by Orville and Idone
(1982) in their literature review, whose maximum values were
between 21 and 23 106 m s1 when compiling the results of
all the researchers mentioned before. Idone et al. (1984)
presented five cases that were faster than the extreme case in
our dataset (2.95 10 7 m s 1), however, all of them were
observed during rocket-triggered lightning events. Considering
the lower end of the speed range, we have observed cases
whose speeds were below the minimum values reported in
literature, with a little less than 10% of the cases with speeds
lower than 1 10 6 m s 1, the lower limit of the review
histogram compiled by Orville and Idone (1982).

293

Table 5
Statistical parameters of the 76 dart leaders. Min stands for minimum value,
Max for maximum value, AM for arithmetic mean and GM for geometric
mean.
2-D speed ( 105 m s1)

Partial
Average

Sample

Min

Max

AM

Median

GM

207
76

1.91
3.33

295
295

46.1

28.5

27.6

3.2.2. Dart leader speed change with height

We have analyzed the variation of the 2-D speed with
height of dart leaders following the same approaches used for
negative stepped leaders (described in Section 3.1.2). Fig. 7
shows the variation of speed with height for the 76 analyzed
events. There is a clear tendency for an accumulation of lower
speed values as we observe lower heights. Along with this
tendency it is possible to notice that the average values for
each range of heights become progressively lower as it gets
closer to ground. One can argue that this behavior is caused
by a statistical bias, considering that the slower cases allowed
us to obtain speed estimates throughout a greater number of
ranges of heights. This possibility increases the importance of the
second approach (in which we analyze each case individually, as
described in Section 3.1.2) for the case of the dart leaders dataset.
For the second approach, in which we have evaluated the
individual behavior of the leaders as they got closer to
ground, only 55 of the 76 cases could be considered due to
the fact that for the remaining 21 cases we were able to obtain a
single partial speed measurement. Table 7 shows the incidence
of each considered behavior (accelerated, decelerated and
irregular). Even though approximately one third of the cases
have accelerated throughout their propagation, more than half
of the dataset is composed of dart leaders that decelerated as
they got closer to ground, a coherent result compared to what
was obtained through the first approach.

Fig. 5. Average 2-D speeds versus return stroke estimated peak current of 51 negative stepped leaders.

294

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Table 6
Same as Table 5 but including only average speeds and considering each
observation site (So Jos dos Campos, So Martinho da Serra and Tucson)
separately.
Location

Average 2-D speed ( 105 m s1)

Sample

So Jos dos Campos

(southeastern Brazil)
So Martinho da Serra
(southern Brazil)
Tucson (southern
Arizona, USA)

38
6
32

Min
3.33
14.9
5.55

Max

AM

Median

GM

295

41.7

16.7

20.8

27.5

23.6

25.3

54.8

34.9

39.3

46.4
242

This result is similar to the behavior observed by Schonland

et al. (1935) and Orville and Idone (1982). The latter researchers also presented two hypotheses about what could
cause this apparent deceleration: (i) the observed variations
are real and consequence of differences in the physical
parameters in the channel that is being re-ionized by the dart
leader (e.g., electric conductivity, channel radius, temperature,
particle density, etc.); (ii) the speed reduction is not real, being
caused by the inherent differences when a three-dimensional
channel is analyzed two-dimensionally. In the latter hypothesis
they speculate that a leader whose speed is approximately
constant could present an apparent acceleration or deceleration depending on the geometry of the channel it is retracing.
Considering that the conductivity regime of the reminiscent
plasma channel of the previous return stroke must not change
appreciably during the time scale of the dart leader propagation (a few milliseconds or even less) after a typical interstroke
interval (as the results of the model by Uman and Voshall,
1968, suggest), we believe that the possibility of the real
three-dimensional speed being constant is physically plausible.

Orville and Idone (1982) have also made a preliminary analysis

on the geometrical hypothesis by comparing the speed profile
with height for two flashes for which two or more dart leaders
have propagated through the same channel. These profiles
were very similar, indicating that either the second hypothesis
is true or channel geometry has more impact on the twodimensional speed measurements. Another possible factor that
could explain this behavior is how altitude may affect the
plasma channel characteristics and, as a consequence, the dart
leader that propagates through it. If this altitude-related
variation does not change appreciably over typical interstroke
times, one of its consequences would be the similarity of the
speed profile of dart leaders of the same flash. We were able to
reproduce this analysis on a single flash that presented five
strokes that followed the same channel, allowing us to obtain
profiles for four dart leaders with an identical geometry,
presented in Fig. 8. It is possible to notice that the leaders
that initiated strokes 2 and 3 presented their lowest speed
approximately at the same height, however, they have only
provided us three partial speed measurements, making it
difficult to consider this comparison conclusive. On the other
hand, the dart leaders that initiated strokes 4 and 5, whose
propagation could be analyzed in greater detail, presented
their minimum speed at very close heights, similarly to what
has been reported by Orville and Idone (1982). A more
detailed analysis can only be done from instruments that
allow a three-dimensional reconstruction of the dart leader
channels, making it possible to test more conclusively both
hypotheses.
3.2.3. Average dart leader speed versus other parameters
The impact of two important parameters over the average
dart leader speed has been analyzed: the estimated peak
current of the return stroke initiated by each leader and the
elapsed time since the previous return stroke.

Fig. 6. Histogram of the average speeds of the 76 dart leaders that were analyzed.

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

295

Fig. 7. (a) Variation of leader 2-D speeds with height for 76 dart leaders; (b) zoom into the region of speeds below 7 106 m s1.

Table 7
Statistics of the individual analysis of the speed variation with height of 55
dart leaders.
Behavior

Number of events

Percentage

Accelerated
Decelerated
Irregular

17
31
7

31
56
13

Fig. 9 shows a scatterplot of average dart leader speeds versus

the peak current of the following return stroke, estimated with
the help of lightning location systems. Due to the dependence
of the availability of such data, only 34 (out of the 76) cases
could be considered in the analysis. It is possible to notice that
the analyzed sample needs to be enhanced in order to include a
greater number of cases with average speeds is of the order of

296

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Fig. 8. Comparison of the variation of propagation speed with height of four dart leaders that followed the same channel to ground.

107 m s1 and/or followed by strokes with peak current above

40 kA. Even though previous works presented more cases with
higher speeds (Hubert and Mouget, 1981; Idone et al., 1984;
Jordan et al., 1992), their samples were also limited in terms of
peak currents, e.g., Jordan et al. (1992) have presented a single
case whose peak current was greater than 40 kA. There is no
clear upper limit for the average speeds of dart leaders that
initiate low current return strokes (20 kA or less), considering
that the three fastest cases presented peak currents between

5.7 and 16.7 kA. Similarly, there is no apparent upper limit for
the peak current of strokes initiated by slower leaders: the most
intense event (83.4 kA) was preceded by a dart leader that
propagated at approximately 3.0 106 m s1. As reported by
Jordan et al. (1992), we did not observe the behavior described
by Idone et al. (1984), who presented a strong linear correlation
(R=0.84) between dart leader speed and return stroke peak
current of 32 rocket triggered lightning. Another similar
analysis, but based on different parameters, was conducted by

Fig. 9. Average 2-D speeds versus return stroke estimated peak current of 34 dart leaders.

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

297

Fig. 10. Average 2-D speed of 76 dart leaders versus the preceding interstroke interval.

Rubinstein et al. (1995). They have studied leader-return stroke

sequences in rocket-triggered lightning (which are intrinsically
similar to dart leaders in natural flashes) and plotted the leader
half-peak width (which can be associated with the leader
speed: the faster leaders would present shorter widths) versus
the maximum leader field change. Even though they expected
that larger leader field changes would be related to shorted
half-peak widths (i.e. leaders with larger linear charge density
propagates faster), no such correlation was found, similarly to
our results.
Considering that the dart leaders propagate through a
region that has been previously ionized by one or more
previous return strokes, on which the electrical conductivity
is relatively higher than the neighboring region, another
parameter that might be related to its propagation speed is
the elapsed time since the previous return stroke. After
current cutoff in the lightning plasma channel it begins to
cool down, a process that occurs along with ionic recombination, which causes a progressive reduction on the electrical
conductivity of the channel in that region. It is expected that
the dart leaders propagate faster when the conductivity is
higher and slower when the channel is closer to neutral air in
terms of conductivity regime. In this context we expect to
observe an upper limit for the speeds of dart leaders that
occur after longer time intervals since the previous return
stroke. Fig. 10 presents a scatterplot between these parameters for all 76 cases included in our dataset. As expected, it is
possible to notice that no case that has occurred after
interstroke intervals longer than 100 ms presented speeds
equal to or higher than 1 10 7 m s 1. For cases preceded by
intervals between 200 and 600 ms this upper limit is even
lower (5 10 6 m s 1), indicating that the recombination
process that occurs in the channel is highly influential over
the development of dart leaders. Similar analysis have been
made previously for both natural (Schonland et al., 1935;
Jordan et al., 1992) and triggered (Hubert and Mouget, 1981;
Idone et al., 1984) flashes with a reduced dataset of cases

preceded by intervals longer than 150 ms. Our results are in

good agreement with what has been reported by them, with
the exception of Idone et al. (1984), who have observed a
linear dependence between the interstroke interval and the
dart leader speed. But this result does not contradict the
other studies because the range of intervals considered is
very small (up to 30 ms) and the correlation coefficient was
very low (R = 0.43). As mentioned at the beginning of
Section 3.2.1, it was not possible to know if some of the
cases that were preceded by longer interstroke intervals
were dart-stepped leaders instead of dart leaders. The results
presented by Rakov and Uman (1990a) indicate that the
occurrence of dart-stepped leaders (inferred from simultaneous electric field and TV records) after interstroke intervals
shorter than 100 ms is considerably low (below 15%) for
stroke orders higher than two; however, the analysis did not
include events preceded by intervals above that value.
Furthermore, Davis (1999), whose work is discussed in detail
by Rakov and Uman, 2003) showed that dart-stepped leaders
typically occur after interstroke intervals longer than 140 ms
for greater stroke orders. Under the light of these previous
studies, caution must be taken when considering events
preceded by intervals longer than 200 ms of Fig. 10 as they
are likely to be dart-stepped leaders. On the other hand, one
can also argue that the group of cases preceded by intervals
ranging from 100 to 200 ms will be composed of both dart
and dart-stepped events, and they are still limited to speeds
below 1 10 7 m s 1.
Additionally, our result is in agreement with the dart
leader model introduced by Borovsky (1995) at the same
time that it is somewhat conflicting with the considerably
different model suggested by Rakov (1998). Borovsky (1995)
described the dart leader process as an electric guided wave
propagating in a channel treated as a conducting cylinder.
The model results suggest that the guided wave velocity
depends on the channel temperature, which can be related to
the previous interstroke interval (as considered in our

298

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

speed measurements. Table 8 presents the statistical parameters related to these events and, as discussed in previous
sessions, we have presented only the maximum and minimum
values for the partial speeds.
Table 9 breaks the positive leader dataset in terms of the
observation site, presenting the statistical parameters of the
average 2-D speeds. Given the small sample size, the events
observed in So Jos dos Campos and Tucson cannot be
compared individually to the others. However, it is clear that
the leaders observed in Rapid City are slower than those
recorded in So Martinho da Serra. The dataset obtained in
the latter site presented speeds that are compatible with
those reported for stepped leaders (Tables 2 and 3, Fig. 2).
When a larger dataset is available, further investigations will
be necessary to evaluate this difference in greater detail.
The distribution of the average speeds of the 29 events is
presented in the histogram of Fig. 11. The statistical test
developed by Shapiro and Wilk (1965) shows that both
average and partial speeds samples follow a lognormal distribution at the 0.05 level. A detailed analysis of the histogram
shows that approximately 25% of the dataset presented average
speeds of the order of 104 m s1, a large fraction compared to
negative stepped leaders (among which less than 2% of the cases
presented speeds at this range of values). On the other hand,
more than 90% of the cases presented speeds lower than

Table 8
Statistical parameters of the 29 positive leaders. Min stands for minimum
value, Max for maximum value, AM for arithmetic mean and GM for
geometric mean.
2-D speed ( 105 m s1)

Partial
Average

Sample

Min

Max

AM

Median

GM

449
29

0.08
0.24

16.2
11.8

2.76

1.80

1.81

analysis). An opposing view is provided by the modeling

performed by Rakov (1998), who considered the dart leader
as a traveling ionizing wave developing on a lossy transmission
line. In this case, the model indicated that the propagation is
mainly dictated by the breakdown process that occurs at the
front of the leader wave and not by the properties of the
channel that lies ahead of it.
3.3. Positive leaders
3.3.1. Two-dimensional speed statistics
We have analyzed a dataset of 29 positive leaders
recorded in So Jos dos Campos, So Martinho da Serra,
Tucson and Rapid City with the help of high-speed cameras.
From their recordings it was possible to obtain 449 partial

Table 9
Same as Table 8 but including only average speeds and considering each observation site (So Jos dos Campos, So Martinho da Serra, Tucson and Rapid City)
separately.
Location

So Jos dos Campos (southeastern Brazil)

So Martinho da Serra (southern Brazil)
Tucson (southern Arizona, USA)
Rapid City (western South Dakota, USA)

Average 2-D speed ( 105 m s1)

Sample

Min

Max

AM

Median

GM

1
10
3
15

1.58
1.59
0.24

0.98
11.8
5.21
5.66

4.57
3.60
1.50

2.88
4.00
1.54

3.63
3.21
1.06

Fig. 11. Histogram of average speeds of 29 positive leaders.

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

299

Fig. 12. (a) Variation of leader 2-D speeds with height for 29 positive leaders; (b) zoom into the region of speeds below 6 105 m s1.

6.0105 m s1, similarly to what we have observed for the

negative stepped leaders. This fact indicates that both leader
types have essentially the same range of possible speeds, with
the positive ones presenting a greater tendency towards lower
values and a smaller minimum possible speed. The only

previous works that presented statistics on the development

speed of natural positive leaders was presented by Saba et al.
(2008, 2010), who considered part of the dataset of the present
work. Our analysis agrees with their conclusion, which
consisted in acknowledging that the speeds measured by Berger

300

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Fig. 13. Comparison of the average leader speed change by range of heights for negative stepped and positive leaders. The number near each point represents the
quantity of measurements available at each range of heights.

and Vogelsanger (1966), of the order of 106 m s1, and Les

Renardires Group (1977), of the order of 104 m s1, represent
extreme cases inserted in a wide range of possible values.
3.3.2. Positive leader speed change with height
The changes of the development speed of positive leaders
with height were analyzed using the same two approaches
described previously and applied to negative stepped
(Section 3.1.2) and dart leaders (Section 3.2.2). Fig. 12 presents
the distribution of speeds in different ranges of heights for the
29 cases of positive leaders analyzed in the present work. As we
analyze lower and lower heights it is possible to notice a
progressive increase in the maximum speed at each interval
along with a simultaneous concentration of points at regions of
higher speeds. The average values also present a clear increase
as the height decreases. On Fig. 12b, in which the region of
values below 6.0 10 5 m s 1 is shown in detail, this behavior
becomes more evident. It is possible to observe that the
average of the speeds measured below 500 m from ground is
one order of magnitude greater than those measured above
1500 m from ground.
When comparing Figs. 3 and 12 it is possible to notice a
considerable difference between the behaviors of negative
stepped and positive leaders. Fig. 13 presents the average
speed value at each range of heights for both leader types
Table 10
Statistics of the individual analysis of the speed variation with height of 28
positive leaders.
Behavior

Number of events

Percentage

Accelerated
Decelerated
Irregular

23
2
3

82
7
11

within the same graph, allowing an easier comparison. While

the averages for the negative stepped leaders oscillate between
23 10 5 m s1 the positive leaders present a steady increase
by one order of magnitude (i.e., from 104 to 105 m s1). It is
important to stress, however, that throughout the majority of
their development the positive leaders present inferior speeds
when compared to negative stepped leaders, and this polarity
asymmetry may have consequences that manifest in other
aspects of cloud-to-ground lightning. According to Williams
and Heckman (2011), in the context of the bidirectional leader
model (Kasemir, 1960; Mazur, 2002) the fact that positive
leaders are usually slower than negative stepped leaders at
heights greater than 1500 m (and possibly inside the thunderstorm cloud) can account for the fact that positive ground
flashes present, usually, a single return stroke followed by long
continuing current while negative flashes may develop up to
tens of return strokes (e.g., Saba et al., 2006, 2010).
In the second approach only 28 cases would be considered
due to the fact that one of them have allowed us to obtain only
one partial speed measurement, making it impossible to analyze
its variation as the leader developed towards ground. Table 10
shows which fraction of the dataset presented each considered
behavior (accelerated, decelerated and irregular). This result
seems to have a good agreement with those provided by
the first approach: more than 80% of the sample accelerated
throughout their development towards ground, while the
other two possibilities manifested themselves in approximately equal proportions. We believe that these results
show that the acceleration of positive leaders is clearer and
more definite than in the case of negative stepped leaders.
3.3.3. Average positive leader speed versus other parameters
Similarly to what has been done for negative stepped and
dart leaders, we have also plotted the average 2-D speeds of

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

301

Fig. 14. Average 2-D speeds versus return stroke estimated peak current of 28 positive leaders.

positive leaders versus the estimated peak current value of

the following return stroke, presented on Fig. 14. Similarly to
what has been observed for both types of negative leaders,
there is no apparent exclusion regions that indicate upper or
lower limits for specific combinations of speeds and currents.
Even though all the cases whose peak values were greater
than 80 kA have presented speeds of the order of 10 5 or
10 6 m s 1, the sample size is still insufficient to confirm that
there is a minimum speed for leaders that initiate more
intense return strokes. As far as the authors know this is the
first analysis of this nature concerning positive leaders,
making it impossible to compare it to previous results presented
in literature.
As previously reported by Saba et al. (2008), part of the
positive ground flashes present the occurrence of a great
number of recoil leaders during the development of the positive
leader towards ground. At the time, however, no analysis had
been done relating the characteristics of the positive leaders or
of the return strokes to the presence of recoil leaders. We now
present a preliminary analysis relating the occurrence of visible
recoil activity with (i) the average positive leader speed and
(ii) the estimated peak current of the return stroke initiated
by the leader.
Fig. 15 presents a scatterplot of the average 2-D speeds of
all 29 positive leaders versus the presence (or absence) of
recoil leaders throughout its development towards ground.
Similarly to what was observed reported in other papers
(Saba et al., 2008; Mazur and Ruhnke, 2011; Warner et al.,
2011), these recoil leaders developed through previously
ionized positive channel branches. Also, for those cases for
which the direction of propagation could be determined, it
was possible to see that they tend to move towards the point
of origin of the flash. Almost all the cases that presented
recoil activity had average speeds below 3 10 5 m s 1 (with
only one exception) and the only positive leader with a speed
of the order of 10 6 m s 1 did not present any visible recoil

leader during its development. However, due to the low

luminosity emitted by most positive leaders during its early
stages, some events could only have their propagation speed
estimated with greater precision when it is closer to ground.
This behavior has been previously reported by Saba et al.
(2008), Section 3.1) and it is during the less luminous phase
that most recoil activity occurred. For this reason, the three
cases in which the partial speeds could be measured during
the earlier, weakly luminous phases of the positive leaders,
are presented separately. These events are shown as red
symbols in Fig. 15b, in which we have used a logarithm scale
in order to facilitate their visualization. All of them presented
average speeds below 1 10 5 m s 1. However, our limited
dataset size does not allow us to say whether or not there is
an upper limit of the positive leader speed for which the
occurrence of recoil activity may occur.
Finally, Fig. 16 shows the estimated peak currents of return
strokes preceded by positive leaders that did and did not present
recoil activity. One can notice that both possible behaviors
presented very similar ranges of peak current values (roughly
2090 kA for those preceded by recoil activity and 2080 kA for
those that were not) suggesting that there is no well-defined
minimum or maximum peak current for positive return strokes
preceded by either possible behavior. The range of peak currents
of return strokes preceded by recoil activity represents approximately 60% of the +CG flashes analyzed by Saba et al. (2010),
so the analysis of extremely intense events (with peak currents
above 100 kA) in future works is important to confirm if there is
a maximum.
4. Concluding remarks
A statistical analysis has been made in a dataset of highspeed video recordings of leaders that preceded cloud-toground lightning. It was possible to show that negative
stepped and positive leaders present average speeds within a

302

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Fig. 15. Average 2-D speeds of 29 positive leaders versus the occurrence of visible recoil leaders during their development. In (b) different colors were used to
show the cases in which the speed estimates were obtained afterwards (black) and during (red) recoil activity.

similar range of values, with the latter presenting a tendency

towards lower speeds by one order of magnitude. We have
also found that the majority of positive leaders increase their
propagation speed as they get closer to ground, while negative
stepped leaders present either a more modest acceleration or
speeds that vary in an irregular fashion. Even though the exact
microphysical reason for this polarity asymmetry remains
unknown, other researchers have been studying its consequences in problems of plasma channel stability of positive and
negative ground flashes.
We have also showed that more than half of the observed
dart leaders decelerated throughout their development towards

ground. It was not possible to confirm, however, whether this

behavior is real or was caused by the channel three-dimensional
geometry of the analyzed flashes. This open question can be
answered in the future with the help of a network of multiple
cameras located at different observation sites.
A very brief analysis of how leaders from different locations
behave. Due to the limited size of our dataset, however, it was
not possible to draw any significant conclusions. The only clear
difference that emerged concerns the ranges of positive leader
speeds from two of the observation sites: the events recorded at
So Martinho da Serra had speeds that were more compatible
with those estimated for stepped leaders, while in Rapid

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

303

Fig. 16. Scatterplot of the positive return stroke estimated peak current versus the presence (or absence) or visible recoil activity during the preceding positive
leader.

City the observed leaders had slower speeds, in agreement

with previous laboratory experiments. The authors encourage future investigations on the influence of the location
climatology over the leader behavior, and a subsequent comparison with physical and computational models of leader
development.
No relation has been found between the average speeds of
the three analyzed leader types (negative stepped and dart,
and positive) and the estimated peak current of the return
stroke they initiated. On the other hand, evidences suggest
that dart leader propagation speeds become lower when they
are preceded by interstroke intervals above 100 ms.We
believe that this is caused by the progressive ionic recombination that takes place in the lightning channel after current
cutoff, causing the electrical conductivity to decay with time.
This lower conductivity would limit the speed with which the
dart leader can develop. The transition to a stepped mode of
propagation for some events (dart-stepped leaders) after
interstroke intervals between 100 and 200 ms makes it very
important that future works address this relationship considering larger datasets either recorded at higher frame rates or
accompanied by electric field measurements.
Finally, we have studied the behavior of two characteristics of positive return strokes that did or did not present
recoil leader activity during the development of their leader.
At first we have analyzed how the average speed of the
positive leader is affected by the presence of recoil leaders
during its development. The data currently available is
insufficient to obtain a conclusive result on this matter,
considering that the recoil activity is mostly concentrated
during periods in which positive leader propagation is
barely detectable by the high-speed camera systems. On the
other hand, we believe that it can be said that there is no
clear maximum or minimum peak current value for return

strokes preceded by recoil leader activity, considering the

most frequent range of values reported in literature (between
20 and 80 kA).

Acknowledgments
We would like to thank K. Cummins for the coordination,
C. Weidman, S. Fleenor, W. Scheftic, J. Wilson, P. Shaw, C.
Jones, and P. Fleenor for the field work and A. Saraiva for the
data reduction of the campaign conducted in Tucson. Also,
the data collection in So Jos dos Campos would not have
been possible without the support provided by C. Schumann,
C. Medeiros, R. da Silva, S. Viegas and C. Lopes. This research
has been supported by CNPq and FAPESP through the projects
475299/2003-5 and 03/08655-4, respectively, and by the NASA
Kennedy Space Center, Grant NNK06EB55G. One of the authors
(L.Z.S.C.) is also grateful for the scholarship 2010/02716-5
provided by FAPESP throughout the development of the
research presented on this paper. We also appreciate the efforts
of the two anonymous reviewers and their helpful comments
that greatly improved the discussion and presentation of our
results.

References
Appleton, E.V., Chapman, F.W., 1937. On the nature of atmospherics, 4. Proc.
Roy. Soc. (Lond.) A158, 122.
Ballarotti, M.G., Saba, M.M.F., Pinto Jr., O., 2005. High-speed camera observations
of negative ground flashes on a millisecond-scale. Geophys. Res. Lett. 32,
L23802. http://dx.doi.org/10.1029/2005GL023889.
Beasley, W., Uman, M.A., Rustan Jr., P.L., 1982. Electric fields preceding cloudto-ground lightning flashes. J. Geophys. Res. 87, 48834902.
Beasley, W.H., Uman, M.A., Jordan, D.M., Ganesh, C., 1983. Simultaneous
pulses in light and electric field from stepped leaders near ground level.
J. Geophys. Res. 88, 86178619.

304

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Berger, K., 1967. Novel observations on lightning discharges: results of

research on Mount San Salvatore. J. Franklin Inst. 283, 478525.
Berger, K., Vogelsanger, E., 1966. Photographische Blitzuntersuchungen der
Jahre 19551965 auf dem Monte San Salvatore. Bull. Schweiz. Elektrotech.
Ver. 57, 599620.
Biagi, C.J., Cummins, K.L., Kehoe, K.E., Krider, E.P., 2007. National Lightning
Detection Network (NLDN) performance in southern Arizona, Texas, and
Oklahoma in 20032004. J. Geophys. Res. 112, D05208. http://dx.doi.org/
10.1029/2006JD007341.
Biagi, C.J., Uman, M.A., Hill, J.D., Jordan, D.M., 2011. Observations of the
initial, upward-propagating, positive leader steps in a rocket-and-wire
triggered lightning discharge. Geophys. Res. Lett. 38, L24809. http://
dx.doi.org/10.1029/2011GL049944.
Borovsky, J.E., 1995. An electrodynamic description of lightning return
strokes and dart leaders: guided wave propagation along conducting
cylindrical channels. J. Geophys. Res. 100, 26972726.
Boys, C.V., 1926. Progressive lightning. Nature 118, 749750.
Boys, C.V., 1929. Progressive lightning. Nature 124, 5455.
Brook, M., Ogawa, T., 1977. The cloud discharge. In: Golde, R. (Ed.), Lightning.
Physics of Lightning, 1. Academic Press, London, pp. 191230.
Campos, L.Z.S., Saba, M.M.F., 2009. Comment on Computer simulations
on sprite initiation for realistic lightning models with higherfrequency surges. In: Asano, T., Suzuki, T., Hiraki, Y., Mareev, E., Cho, M.G.,
Hayakawa, M. (Eds.), J. Geophys. Res., 114, p. A12324. http://dx.doi.org/
10.1029/2009JA014335.
Campos, L.Z.S., Saba, M.M.F., Pinto Jr., O., Ballarotti, M.G., 2007. Waveshapes of
continuing currents and properties of M-components in natural negative
cloud-to-ground lightning from high-speed video observations. Atmos.
Res. 84, 302310. http://dx.doi.org/10.1016/j.atmosres.2006.09.002.
Campos, L.Z.S., Saba, M.M.F., Pinto Jr., O., Ballarotti, M.G., 2009. Waveshapes
of continuing currents and properties of M-components in natural
positive cloud-to-ground lightning. Atmos. Res. 91, 416424. http://
dx.doi.org/10.1016/j.atmosres.2008.02.020.
Chapman, F.W., 1939. Atmospheric disturbance due to thundercloud discharge, 1.
Proc. Phys. Soc. 51, 876894.
Chen, M., Takagi, N., Watanabe, T., Wang, D., Kawasaki, Z.I., Liu, X., 1999. Spatial
and temporal properties of optical radiation produced by stepped leaders.
J. Geophys. Res. 104, 27,57327,584.
Clarence, N.D., Malan, D.J., 1957. Preliminary discharge processes in lightning
flashes to ground. Q. J. R. Meteorol. Soc. 83, 161172.
Cummins, K.L., Murphy, M.J., 2009. An overview of lightning locating
systems: history, techniques, and data uses, with an in-depth look at
the U.S. NLDN. IEEE Trans. Electromagn Compat. 51 (3), 499518. http://
dx.doi.org/10.1109/TEMC.2009.2023450.
Davis, S. M., 1999. Properties of lightning discharges from multiple-station
wideband electric field measurements. Ph.D. Thesis, University of Florida,
Gainesville, 228 pp.
Fuquay, M.D., 1982. Positive cloud-to-ground lightning in summer thunderstorms. J. Geophys. Res. 87, 71317140.
Hill, R.D., 1968. Analysis of irregular paths of lightning channels. J. Geophys.
Res. 73, 18971906.
Hill, J.D., Uman, M.A., Jordan, D.M., 2011. High-speed video observations of a
lightning stepped leader. J. Geophys. Res. 116, D16117. http://dx.doi.org/
10.1029/2011JD015818.
Hubert, P., Mouget, G., 1981. Return stroke velocity measurements in two
triggered lightning flashes. J. Geophys. Res. 86, 52535261.
Idone, V.P., 1992. The luminous development of Florida triggered lightning.
Res. Lett. Atmos. Electr. 122, 2328.
Idone, V.P., Orville, R.E., 1988. Channel tortuosity variation in Florida triggered
lightning. Geophys. Res. Lett. 15, 645648.
Idone, V.P., Orville, R.E., Hubert, P., Barret, L., Eybert-Berard, A., 1984. Correlated
observations of three triggered lightning flashes. J. Geophys. Res. 89,
13851394.
Jerauld, J., Rakov, V.A., Uman, M.A., Rambo, K.J., Jordan, D.M., Cummins, K.L.,
Cramer, J.A., 2005. An evaluation of the performance characteristics of
the U.S. National Lightning Detection Network in Florida using rockettriggered lightning. J. Geophys. Res. 110, D19106. http://dx.doi.org/
10.1029/2005JD005924.
Jordan, D.M., Idone, V.P., Rakov, V.A., Uman, M.A., Beasley, W.H., Jurenka, H.,
1992. Observed dart leader speed in natural and triggered lightning.
J. Geophys. Res. 97, 99519957.
Jordan, D.M., Rakov, V.A., Beasley, W.H., Uman, M.A., 1997. Luminosity
characteristics of dart leaders and return strokes in natural lightning.
J. Geophys. Res. 102, 22,02522,032.
Kasemir, H.W., 1960. A contribution to the electrostatic theory of a lightning
discharge. J. Geophys. Res. 65, 18731878.
Kitagawa, N., 1957. On the electric field change due to the leader processes and
some of their discharge mechanism. Pap. Meteorol. Geophys. (Tokyo) 7,
400414.

Kitagawa, N., Brook, M., 1960. A comparison of intracloud and cloud-toground lightning discharges. J. Geophys. Res. 65, 11891201.
Kong, X., Qie, X., Zhao, Y., 2008. Characteristics of downward leader in a
positive cloud-to-ground flash observed by high-speed video camera
and electric field changes. Geophys. Res. Lett. 35, L05816. http://
dx.doi.org/10.1029/2007GL032764.
Krider, E.P., Radda, G.J., 1975. Radiation field wave forms produced by
lightning stepped leaders. J. Geophys. Res. 80, 26532657.
Krider, E.P., Weidman, C.D., Noggle, R.C., 1977. The electric fields produced
by lightning stepped leaders. J. Geophys. Res. 82, 951960.
Laroche, P., Eybert-Berard, A., Barrett, L., Berlandis, J.P., 1988. Observations of
preliminary discharges initiating flashes triggered by the rocket and
wire technique. Paper presented at the 8th International Conference on
Atmospheric Electricity. International Commission on Atmospheric Electricity, Uppsala, Sweden.
Les Renardires Group, 1977. Positive discharges in long air gaps at Les
Renardires, 1975 results and conclusions. Electra. 53, 31153.
Les Renardires Group, 1981. Negative discharges in long air gaps at Les
Renardires, 1978 results. Electra. 74, 67216.
Lu, W., Zhang, Y., Li, J., Zheng, D., Dong, W., Chen, S., Wang, F., 2008. Optical
observations on propagation characteristics of leaders in cloud-toground lightning flashes. Acta Meteorol. Sinica 22, 6677.
Mach, D.M., Rust, W.D., 1997. Two-dimensional speed and optical risetime
estimates for natural and triggered dart leaders. J. Geophys. Res. 102,
13,67313,684.
Mazur, V., 1989. Triggered lightning strikes to aircraft and natural intracloud
discharges. J. Geophys. Res. 94, 33113325.
Mazur, V., 2002. Physical processes during the development of lightning
flashes. C. R. Physique 3, 13931409.
Mazur, V., Ruhnke, L.H., 2011. Physical processes during development of
upward leaders from tall structures. J. Electrost. 69, 97110.
Mazur, V., Krehbiel, P.R., Shao, X.M., 1995. Correlated high-speed video and
radio interferometric observations of a cloud-to-ground lightning flash.
J. Geophys. Res. 100, 25,73125,753.
Mazur, V., Shao, X.M., Krehbiel, P.R., 1998. Spider lightning in intracloud and
positive cloud-to-ground flashes. J. Geophys. Res. 103, 19,81119,822.
Mazur, V., Ruhnke, L.H., Warner, T.A., Orville, R.E., 2011. Discovering the
Nature of Recoil Leaders. Paper presented at the 14th International
Conference on Atmospheric Electricity. International Commission on
Atmospheric Electricity, Rio de Janeiro, Brazil.
McEachron, K.B., 1939. Lightning to the Empire State Building. J. Franklin
Inst. 227, 149217.
Nagai, Y., Kawamata, S., Edano, Y., 1982. Observation of preceding leader and
its downward traveling velocity in Utsunomiya district. Res. Lett. Atmos.
Electr. 2, 5356.
Nesbitt, S.W., Zhang, R., Orville, R.E., 2000. Seasonal and global NOx production
by lightning estimated from the Optical Transient Detector (OTD). Tellus
52B, 12061215.
Ogawa, T., Brook, M., 1964. The mechanism of the intracloud lightning
discharge. J. Geophys. Res. 69, 51415150.
Orville, R.E., Idone, V.P., 1982. Lightning leader characteristics in the
Thunderstorm Research International Program (TRIP). J. Geophys. Res.
87, 11,17711,192.
Pierce, E.T., 1955. Electrostatic field-changes due to lightning discharges. Q. J.
R. Meteorol. Soc. 81, 211228.
Pinto Jr., O., Nacaratto, K.P., Pinto, I.R.C.A., Fernandes, W.A., Neto, O.P., 2006.
Monthly distribution of cloud-to-ground lightning flashes as observed
by lightning location systems. Geophys. Res. Lett. 33, L09811. http://
dx.doi.org/10.1029/2006GL026081.
Pinto Jr., O., Pinto, I.R.C.A., Nacaratto, K.P., 2007. Maximum cloud-to-ground
lightning flash densities observed by lightning location systems in the
tropical region: a review. Atmos. Res. 84, 189200. http://dx.doi.org/
10.1016/j.atmosres.2006.11.007.
Proctor, D.E., Uytenbogaardt, R., Meredith, B.M., 1988. VHF radio pictures of
lightning flashes to ground. J. Geophys. Res. 93, 12,68312,727.
Qie, X., Kong, X., 2007. Progression features of a stepped leader process with
four grounded leader branches. Geophys. Res. Lett. 34, L06809. http://
dx.doi.org/10.1029/2006GL028771.
Rakov, V.A., 1998. Some inferences on the propagation mechanisms of dart
leaders and return strokes. J. Geophys. Res. 103, 18791887.
Rakov, V.A., Uman, M.A., 1990a. Some properties of negative cloud-to-ground
lightning flashes versus stroke order. J. Geophys. Res. 95, 54475453.
Rakov, V.A., Uman, M.A., 1990b. Waveforms of first and subsequent leaders
in negative lightning flashes. J. Geophys. Res. 95, 16,56116,577.
Rakov, V.A., Uman, M.A., 2003. Lightning: Physics and Effects. Cambridge
Univ, Press, New York.
Rakov, V.A., Thottappillil, R., Uman, M.A., 1992. Electric field pulses in K
and M changes of lightning ground flashes. J. Geophys. Res. 97,
99359950.

L.Z.S. Campos et al. / Atmospheric Research 135136 (2014) 285305

Richard, P., Delannoy, A., Labaune, G., Laroche, P., 1986. Results of spatial and
temporal characterization of the VHF-UHF radiation of lightning. J. Geophys.
Res. 91, 12481260.
Rison, W., Thomas, R.J., Krehbiel, P.R., Hamlin, T., Harlin, J., 1999. A GPS-based
three-dimensional lightning mapping system: initial observations in
Central New Mexico. Geophys. Res. Lett. 26, 35733576.
Rubinstein, M., Rachidi, F., Uman, M.A., Thottappillil, R., Rakov, V.A., Nucci,
C.A., 1995. Characterization of vertical electric fields 500 m and 30 m
from triggered lightning. J. Geophys. Res. 100, 88638872.
Saba, M.M.F., Ballarotti, M.G., Pinto Jr., O., 2006. Negative cloud-to-ground
lightning properties from high-speed video observations. J. Geophys.
Res. 111, D03101. http://dx.doi.org/10.1029/2005JD006415.
Saba, M.M.F., Cummins, K.L., Warner, T.A., Krider, E.P., Campos, L.Z.S., Ballarotti,
M.G., Pinto Jr., O., Fleenor, S.A., 2008. Positive leader characteristics from
high-speed video observations. Geophys. Res. Lett. 35, L07802. http://
dx.doi.org/10.1029/2007GL033000.
Saba, M.M.F., Campos, L.Z.S., Krider, E.P., Pinto Jr., O., 2009. High-speed video
observations of positive ground flashes produced by intracloud lightning.
Geophys. Res. Lett. 36, L12811. http://dx.doi.org/10.1029/2009GL038791.
Saba, M.M.F., Schulz, W., Warner, T.A., Campos, L.Z.S., Schumann, C., Krider,
E.P., Cummins, K.L., Orville, R.E., 2010. High-speed video observations of
positive lightning flashes to ground. J. Geophys. Res. 115, D24201. http://
dx.doi.org/10.1029/2010JD014330.
Saraiva, A.C.V., Saba, M.M.F., Pinto Jr., O., Cummins, K.L., Krider, E.P.,
Campos, L.Z.S., 2010. A comparative study of negative cloud-to-ground
lightning characteristics in So Paulo (Brazil) and Arizona (USA) based
on high-speed video observations. J. Geophys. Res. 115, D11102.
http://dx.doi.org/10.1029/2009JD012604.
Schonland, B.F.J., 1938. Progressive lightning, 4, The discharge mechanisms.
Proc. Roy. Soc. (Lond.) A164, 132150.
Schonland, B.F.J., 1956. The lightning discharge. Handb. Phys. 22, 576628.
Schonland, B.F.J., Malan, D.J., Collens, H., 1935. Progressive lightning, 2. Proc.
Roy. Soc. (Lond.) A152, 595625.
Schonland, B.F.J., Hodges, D.B., Collens, H., 1938a. Progressive lightning, 5. A
comparison of photographic and electrical studies of the discharge
process. Proc. Roy. Soc. (Lond.) A166, 5675.
Schonland, B.F.J., Malan, D.J., Collens, H., 1938b. Progressive lightning, 6.
Proc. Roy. Soc. (Lond.) A168, 455469.

305

Shao, X.M., Krehbiel, P.R., Thomas, R.J., Rison, W., 1995. Radio interferometric
observations of cloud-to-ground lightning phenomena in Florida. J. Geophys.
Res. 100, 27492783.
Shapiro, S.S., Wilk, M.B., 1965. An analysis of variance test for normality
(complete samples). Biometrika 52, 591611.
Thomson, E.M., Uman, M.A., Beasley, W.H., 1985. Speed and current for
lightning stepped leaders near ground as determined from electric field
records. J. Geophys. Res. 90, 81368142.
Uman, M.A., Voshall, R.E., 1968. Time interval between lightning strokes and
the initiation of dart leaders. J. Geophys. Res. 73, 497506.
Wang, D., Takagi, N., Watanabe, T., Rakov, V.A., Uman, M.A., 1999. Observed
leader and return-stroke propagation characteristics in the bottom
400 m of a rocket-triggered lightning channel. J. Geophys. Res. 104,
14,36914,376.
Warner, T.A., 2012. Observations of simultaneous upward lightning leaders
from multiple tall structures. Atmos. Res. 117, 4554. http://dx.doi.org/
10.1016/j.atmosres.2011.07.004.
Warner, T.A., Orville, R.E., Marshall, J.L., Huggins, K., 2011. Spectral (6001050
nm) time exposures (99.6 s) of a lightning stepped leader. J. Geophys.
Res. 116, D12210. http://dx.doi.org/10.1029/2011JD015663.
Warner, T.A., Saba, M.M.F., Orville, R.E., 2012. Characteristics of Upward
Leaders from Tall Towers. Paper presented at the 22nd International
Lightning Detection Conference. Vaisala, Broomfield, USA.
Williams, E.R., 2006. Problems in lightning physics-the role of polarity
asymmetry. Plasma Sources Sci. Technol. 15, S91S108.
Williams, E., Heckman, S., 2011. Polarity Asymmetry in Lightning Leader
Speeds: Implications for Current Cutoff and Multiple Strokes in Cloudto-Ground Flashes. Paper presented at XIV International Conference on
Atmospheric Electricity. International Commission on Atmospheric Electricity, Rio de Janeiro, Brazil.
Winn, W.P., 1965. A laboratory analog to the dart leader and return stroke of
lightning. J. Geophys. Res. 70, 32653270.
Yokoyama, S., Miyake, K., Suzuki, T., Kanao, S., 1990. Winter lightning on
Japan Sea Cost development of measuring system on progressing
feature of lightning discharge. IEEE Trans. Power Deliv. 5, 14181425.