(Eduard M. Bazelyan, Yuri P. Raizer) Lightning

Lightning Physics and
Lightning Protection
E M Bazelyan
and
Yu P Raker
IOP
Institute of Physics Publishing
Bristol and Philadelphia
Copyright © 2000 IOP Publishing Ltd.

IOP Publishing Ltd 2000
All rights reserved. No part of this publication may be reproduced, stored in a

retrieval system or transmitted in any form or by any means, electronic,
mechanical, photocopying, recording or otherwise, without the prior permission of
the publisher. Multiple copying is permitted in accordance with the terms of
licences issued by the Copyright Licensing Agency under the terms of its agreement
with the Committee of Vice-Chancellors and Principals.
British Library Cataloguing-in-PublicationData
A catalogue record for this book is available from the British Library.
ISBN 0 7503 0477 4
Library of Congress Cataloging-in-Publication Data are available
Publisher: Nicki Dennis

Commissioning Editor: John Navas
Production Editor: Simon Laurenson
Production Control: Sarah Plenty
Cover Design: Victoria Le Billon
Marketing Executive: Colin Fenton
Published by Institute of Physics Publishing, wholly owned by The Institute of

Physics, London
Institute of Physics Publishing, Dirac House, Temple Back, Bristol BS1 6BE, UK
US Office: Institute of Physics Publishing, The Public Ledger Building, Suite 1035,
150 South Independence Mall West, Philadelphia, PA 19106, USA
Typeset in 10/12pt Times by Academic + Technical, Bristol

Printed in the UK by J W Arrowsmith Ltd, Bristol

Contents
Preface ix
1 Introduction: lightning, its destructive effects and protection 1
1.1 Types of lightning discharge 1
1.2 Lightning discharge components 5
1.3 Basic stages of a lightning spark 6
1.4 Continuous and stepwise leaders 9
1.5 Lightning stroke frequency 11
1.5.1 Strokes at terrestrial objects 11
1.5.2 Human hazard 11
1.6 Lightning hazards 12
1.6.1 A direct lightning stroke 12
1.6.2 Induced overvoltage 17
1.6.3 Electrostatic induction 18
1.6.4 High potential infection 19
1.6.5 Current inrush from a spark creeping along the
earth’s surface 20
1.6.6 Are lightning protectors reliable? 21
1.7 Lightning as a power supply 23
1.8 To those intending to read on 24
References 26
2 The streamer-leader process in a long spark 27

2.1 What a lightning researcher should know about a long spark 28
2.2 A long streamer 32
2.2.1 The streamer tip as an ionization wave 32
2.2.2 Evaluation of streamer parameters 34
2.2.3 Current and field in the channel behind the tip 39
2.2.4 Gas heating in a streamer channel 42
V

vi Contents
2.2.5 Electron-molecular reactions and plasma decay in

cold air 45
2.2.6 Final streamer length 48
2.2.7 Streamer in a uniform field and in the ‘absence’ of
electrodes 53
2.3 The principles of a leader process 59
2.3.1 The necessity of gas heating 59
2.3.2 The necessity of a streamer accompaniment 61
2.3.3 Channel contraction mechanism 64
2.3.4 Leader velocity 66
2.4 The streamer zone and cover 67
2.4.1 Charge and field in a streamer zone 67
2.4.2 Streamer frequency and number 70
2.4.3 Leader tip current 71
2.4.4 Ionization processes in the cover 73
2.5 A long leader channel 75
2.5.1 Field and the plasma state 75
2.5.2 Energy balance and similarity to an arc 79
2.6 Voltage for a long spark 81
2.7 A negative leader 83
References 88
3 Available lightning data 90

3.1 Atmospheric field during a lightning discharge 91
3.2 The leader of the first lightning component 94
3.2.1 Positive leaders 94
3.2.2 Negative leaders 96
3.3 The leaders of subsequent lightning components 98
3.4 Lightning leader current 100
3.5 Field variation at the leader stage 102
3.6 Perspectives of remote measurements 107
3.6.1 Effect of the leader shape 107
3.6.2 Effect of linear charge distribution 111
3.7 Lightning return stroke 115
3.7.1 Neutralization wave velocity 116
3.7.2 Current amplitude 117
3.7.3 Current impulse shape and time characteristics 122
3.7.4 Electromagnetic field 126
3.8 Total lightning flash duration and processes in the
intercomponent pauses 129
3.9 Flash charge and normalized energy 131
3.10 Lightning temperature and radius 132
3.1 1 What can one gain from lightning measurements? 134
References 136

Contents vii
4 Physical processes in a lightning discharge 138

4.1 An ascending positive leader 138
4.1.1 The origin 138
4.1.2 Leader development and current 141
4.1.3 Penetration into the cloud and halt 144
4.1.4 Leader branching and sign reversal 148
4.2 Lightning excited by an isolated object 150
4.2.1 A binary leader 150
4.2.2 Binary leader development 152
4.3 The descending leader of the first lightning component 158
4.3.1 The origin in the clouds 158
4.3.2 Negative leader development and potential transport 161
4.3.3 The branching effect 166
4.3.4 Specificity of a descending positive leader 168
4.3.5 A counterleader 169
4.4 Return stroke 171
4.4.1 The basic mechanism 171
4.4.2 Conclusions from explicit solutions to long line
equations 175
4.4.3 Channel transformation in the return stroke 181
4.4.4 Return stroke as a channel transformation wave 185
4.4.5 Arising problems and approaches to their solution 190
4.4.6 The return stroke of a positive lightning 194
4.5 Anomalously large current impulses of positive lightnings 195
4.6 Stepwise behaviour of a negative leader 197
4.6.1 The step formation and parameters 197
4.6.2 Energy effects in the leader channel 199
4.7 The subsequent components. The M-component 202
4.8 Subsequent components. The problem of a dart leader 207
4.8.1 A streamer in a ‘waveguide’? 207
4.8.2 The non-linear diffusion wave front 209
4.8.3 The possibility of diffusion-to-ionization wave
transformation 212
4.8.4 The ionization wave in a conductive medium 213
4.8.5 The dart leader as a streamer in a ‘nonconductive
waveguide’ 215
4.9 Experimental checkup of subsequent component theory 217
References 219
5 Lightning attraction by objects 222

5.1 The equidistance principle 223
5.2 The electrogeometric method 226
5.3 The probability approach to finding the stroke point 228
5.4 Laboratory study of lightning attraction 232

Contents
..I
Vlll
5.5 Extrapolation to lightning 236

5.6 On the attraction mechanism of external field 239
5.7 How lightning chooses the point of stroke 24 1
5.8 Why are several lightning rods more effective than one? 241
5.9 Some technical parameters of lightning protection 249
5.9.1 The protection zone 249
5.9.2 The protection angle of a grounded wire 25 1
5.10 Protection efficiency versus the object function 252
5.11 Lightning attraction by aircraft 255
5.12 Are attraction processes controllable? 259
5.13 If the lightning misses the object 263
References 264
6 Dangerous lightning effects on modern structures 265
6.1 Induced overvoltage 267
6.1.1 ‘Electrostatic’ effects of cloud and lightning charges 267
6.1.2 Overvoltage due to lightning magnetic field 270
6.2 Lightning stroke at a screened object 272
6.2.1 A stroke at the metallic shell of a body 272
6.2.2 How lightning finds its way to an underground cable 274
6.2.3 Overvoltage on underground cable insulation 277
6.2.4 The action of the skin-effect 283
6.2.5 The effect of cross section geometry 285
6.2.6 Overvoltage in a double wire circuit 290
6.2.7 Laboratory tests of objects with metallic sheaths 29 1
6.2.8 Overvoltage in a screened multilayer cable 294
6.3 Metallic pipes as a high potential pathway 296
6.4 Direct stroke overvoltage 300
6.4.1 The behaviour of a grounding electrode at high
current impulses 300
6.4.2 Induction emf in an affected object 305
6.4.3 Voltage between the affected and neighbouring
objects 307
6.4.4 Lines with overhead ground-wires 3 14
6.5 Concluding remarks 318
References 320

Preface
Today, we know sufficiently much about lightning to feel free from the mystic
fears of primitive people. We have learned to create protection technologies
and to make power transmission lines, skyscrapers, ships, aircraft, and space-
craft less vulnerable to lightning. Yes, the danger is getting less but it still
exists! With every step of the technical progress, lightning arms itself with
a new weapon to continue the war by its own rules against the self-confident
engineer. As we improve our machines and stuff them with electronics in an
attempt to replace human beings, lightning acts in an ever refined manner. It
takes us by surprise where we do not expect it, making us feel helpless again
for some time.
We do not intend to present in this book a set of universal lightning
protection rules. Such a task would be as futile as advertising a universal anti-
biotic lethal to every harmful microbe. The world is changeable, and today’s
panacea often becomes a useless pill even before the advertising sheet fades.
Technical progress has so far failed to take lightning unawares. Improvement
and miniaturization of devices increase our concern about the refined
destructive behaviour of lightning, but no prophet is able to foresee all of
its destructive effects.
We d o not plan to discuss in detail all available information on light-
ning. There are already some excellent books providing all sort of reference
data, among them the two volumes of Lightning edited by R H Golde and
Lightning Discharge by M Uman. Our aim is different. We think it important
to give the reader some clear, up-to-date physical concepts of lightning
development, which cannot be found in the books referred to. These will
serve as a basis for the researcher and engineer to judge the properties of
this tremendous gas discharge phenomenon. Then we shall discuss the
nature of various hazardous manifestations of lightning, focusing on the
physical mechanisms of interaction between lightning and an affected
construction. The results of this consideration will further be used to estimate
ix

X Preface
the effectiveness of conventional protective measures and to predict technical

means for their improvement. We give, wherever possible, technical advice
and recommendations. Our main goal, however, is to help the reader to
make his own predictions by providing information on the whole arsenal
of potentionally hazardous effects of lightning on a particular construction.
We have often witnessed situations when an engineer was trying hard to
‘impose’ this or that protective device on an operating experimental structure
which resisted his unnatural efforts. Ideally, the designer must be able to
foresee all details of the relationship between lightning and the construction
being designed. It is only in this case that lightning protection can become
functionally effective and the protective device can be made compatible
with the construction elements.
If an engineer is determined to follow this approach, both expedient and
well-grounded, he will find this book useful. It is a natural extension of our
previous book Spark Discharge, published by CRC Press in 1997, which dealt
with streamer-leader breakdown of long gas gaps. The streamer-leader
process is part of any lightning discharge when a plasma spark closes a
gigantic air gap. Although the destructive effect of lightning is primarily
due to the return stroke which follows the leader, it is the leader that
makes the discharge channel susceptible to it. This is why we give an overview
of the streamer-leader process, focusing on extrema1 estimations and
presenting some new ideas. We hope that the second chapter will prove infor-
mative even for those familiar with our book of 1997.
Some results of the lightning investigation run in the Krzhizhanovsky
Power Institute are used in the book. The authors would like to thank
D r B N Gorin and Dr A V Shkilev who kindly allowed us to use the originals
of lightning photographs. We are also grateful to L N Smirnova for
translation of this book.

Chapter 1
Introduction: lightning, its

destructive effects and protection
If you want to observe lightning, the best thing to do is to visit a special light-
ning laboratory. Such laboratories exist in all parts of the globe except the
Antarctic. But you can save on the travel if you just climb onto the roof of
your own house to give a good field of vision. Better, fetch your camera.
Even an ordinary picture can show details the unaided human eye often
misses. You might as well sit back in your favourite armchair, having
pulled it up to a window, preferably one overlooking an open space. The
camera can be fixed on the window sill, There is nothing else to do but
wait for a stormy night.
There is enough time for the preparations to be made because the storm
will be approaching slowly. At first, the air will grow still, and it will get
much darker than it normally is on a summer night. The cloud is not yet visible,
but its approach can be anticipated from the soundless flashes at the horizon.
They gradually pull closer, and the brightest of them can already be heard as
delayed and yet amiable roaring. T h s may go on for a long time. It may
seem that the cloud has stopped still or turned away, but suddenly the sky is
ripped open by a fire blade. This is accompanied by a deafening crash, quite
different from a cannon shot because it takes a much longer time. The first
lightning discharge is followed by many others falling out of the ripped
cloud. Some strike the ground while others keep on crossing the sky, competing
with the first discharge in beauty and spark length. This is the right time to start
observations: remove the camera shutter and try to take a few pictures.
1.1 Types of lightning discharge
The above recommendation to remove the camera shutter should be taken

literally. Much information on lightning has been obtained from photographs
taken with a preliminarily opened objective lens. It is important, however, that

2 Introduction: lightning. its destructive effects and protection
Figure 1.1. A static photograph of a lightning stroke at the Ostankino Television

Tower in Moscow.
no other bright light source should be present within the vision field of the
camera lens. The film can then be exposed for many minutes until a spark
finds its way into the frame. After this, the lens should be closed with the shut-
ter and the camera should be set ready for another shot. Experience has shown
that at least one third of pictures taken during a good night thunderstorm
prove successful.
All lightning discharges can be classified, even without photography,
into two groups - intercloud discharges and ground strikes. The frequency
of the former is two or three times higher than that of the latter. An inter-
cloud spark is never a straight line, but rather has numerous bends and
branchings. Normally, the spark channel is as long as several kilometres,
sometimes dozens of kilometres.
The length of a lightning spark that strikes the ground can be defined
more exactly. The average cloud altitude in Europe is close to three kilo-
metres. Spark channels have about the same average length. Of course,
this parameter is statistically variable, because a discharge from a charged
cloud centre may start at any altitude up to 10 km and because of a large
number of spark bends. The latter are observable even with the unaided
eye. In a photograph, they may look strikingly fanciful (figure 1.1). A photo-
graph can show another important feature inaccessible to the naked eye - the
main bright spark reaching the ground has numerous branches which have
stopped their development at various altitudes. A single branch may have
a length comparable with that of the principal spark channel (figure 1.2).
Branches can be conveniently used to define the direction of lightning
propagation. Like a tree, a lightning spark branches in the direction of

Tipes of lightning discharge 3
Figure 1.2. A photograph of a descending lightning with numerous branches.
growth. In addition to descending sparks outgrowing from a cloud toward

the ground, there are also ascending sparks starting from a ground construc-
tion and developing up to a cloud (figure 1.3). Their direction of growth is
well indicated by branches diverging upward.
In a flat country, an ascending spark can arise only from a skyscraper or
a tower of at least 100-200m high, and the number of ascending sparks
grows with the building height. For example, over 90% of all sparks that
strike the 530-m high Ostankino Television Tower in Moscow are of the
ascending type [l]. A similar value was reported for the 410-m high
Empire State Building in New York City [2]. Buildings of such a height
can be said to fire lightning sparks up at clouds rather than to be attacked
by them. In mountainous regions, ascending sparks have been observed
for much lower buildings. As an illustration, we can cite reports of storm
observations made on the San Salvatore Mount in Switzerland [3]. The
receiving tower there was only 70 m high but most of the discharges affecting
it were of the ascending type.
Skyscrapers and television towers are, however, quite scarce on the
Earth. So the researcher has a natural desire to construct, in the right
place and for a short time, a spark-generating tower of his own. For this,
a small probe pulling up a thin grounded wire is launched towards a

4 Introduction: lightning, its destructive effects and protection
Figure 1.3. A photograph of an ascending lightning.
storm cloud [4]. When the probe rises to 200-300m above the earth, an
ascending spark is induced from it. A discharge artificially induced in the
atmosphere is often referred to as triggered lightning. To raise the chances
for a successful experiment, the electric field induced by the storm charges at
the ground surface are measured prior to the launch. The probe is triggered
when the field strength becomes close to 200 V/cm, which provides spark
ignition in 60-70% of launches [5].
The value 200 V/cm is two orders of magnitude smaller than the thresh-
old value of E = 30 kV/cm, at which a short air gap with a uniform field is
broken down under normal atmospheric conditions. Clearly, no spark
ignition would be possible without the local field enhancement by electric
charges induced on the probe and the wire. Below, we shall discuss the
triggered discharge mechanism in more detail.
A field detector on the Earth’s surface (it might as well be placed on the
window of your own room) can easily determine the polarity of the charge
transported by a lightning spark to the ground. The polarity of the spark
is defined by that of the charge. About 90% of descending sparks occurring
in Europe during summer storms carry a negative charge, so these are known
as negative descending sparks. The other descending sparks are positive. The

Lightning discharge components 5
proportion of positive sparks has been found to be somewhat larger in

tropical and subtropical regions, especially in winter, when it may be as
large as 50%.
There is no special name for lightning sparks generated by aircraft
during flights, when they are entirely insulated from the ground. Such dis-
charges arise fairly frequently. A modern aircraft experiences at least one
lightning stroke every 3000 flight hours. Almost half of the strokes start
from the aircraft itself, not from a cloud. This often happens in heap
rather than clouds carrying a relatively small electric charge. The reason
for a discharge from a large ground-insulated object is principally the same
as from a grounded object and is due to the electric field enhancement by
surface polarization charge. This issue will be discussed after the analysis
of ascending sparks in section 4.2.
1.2 Lightning discharge components
An observer can notice a lightning spark flicker which, sometimes, may

become quite distinct. Even the first cinematographers knew that the
human eye could distinguish between two events only if they occurred with
a time interval longer than 0.1 s. Since lightning flicker is observable, the
pause between two current impulses must be longer than 0.1 s.
A current-free pause can be measured quite accurately by exposing a
moving film to a lightning discharge. With up-to-date lenses and photo-
graphic materials, one can obtain a good 1 mm resolution of the film, In
order to displace an image by 1 mm over a time period of 0.1 s, the film
speed must be about 1 cmjs. It can be achieved by manually moving the
film keeping the camera lens open (alas, an electrically driven camera is
unsuitable for this). Then, with some luck, one may get a picture like the
one in figure 1.4. The spark flashes up and dims out several times. Unless
the pause is too long, a new flash follows the previous trajectory; otherwise,
the spark takes a partially or totally new path.
A lightning spark with several flashes is known as a multicomponent
spark. One may suggest that the channel of the first component formed in
unperturbed air differs in its basic characteristics from the subsequent chan-
nels, if they take exactly the same path through the ionized and heated air.
The formation of subsequent components is considered in sections 4.7 and
4.8. Note only that multicomponent sparks are usually negative, both
ascending and descending. The average number of components is close to
three, while the maximum number may be as large as thirty. Generally, the
average duration of a lightning flash is 0.2 s and the maximum duration is
1-1.5s [6], so it is not surprising that the eye can sometimes distinguish
between individual components. Positive sparks normally contain only one
component.

Figure 1.4. The image of a multicomponent lightning in a slowly moving film.
1.3 Basic stages of a lightning spark
The affinity of lightning to a spark discharge was demonstrated by Benjamin

Franklin as far back as the 18th century. Historically, basic spark elements
were first identified in lightning, and only much later were they observed in
laboratory sparks. This is easy to understand if one recalls that a lightning
spark has a much greater length and takes a longer time to develop, so
that its optical registration does not require the use of sophisticated equip-
ment with a high space and time resolution. The first streak photographs
of lightning, taken in the 1930s by a simple camera with a mechanically
rotated film (Boys camera), are still impressive [7]. They show the principal
stages of the lightning process - the leader stage and the return stroke.
The leader stage represents the initiation and growth of a conductive
plasma channel - a leader - between a cloud and the earth or between
two clouds. The leader arises in a region where the electric field is strong
enough to ionize the air by electron impact. However, it mostly propagates
through a region in which the external field induced by the cloud charge
does not exceed several hundreds of volts per centimetre. In spite of this it
does propagate, which means that there is an intensive ionization occurring
in its tip region, changing the neutral air to a highly conductive plasma. This
becomes possible because the leader carries its own strong electric field
induced by the space charge concentrated at the leader tip and transported
together with it. A rough analogue of the leader field is that of a metallic
needle connected with a thin wire to a high voltage supply. If the needle is
sharp enough, the electric field in the vicinity of its tip will be very strong

Basic stages of a lightning spark 7
even at a relatively low voltage. Imagine now that the needle is falling down
on to the earth, pulling the wire behind it. The strong field region, in which
the air molecules become ionized, will move down together with the needle.
A lightning spark has no wire at its disposal. The function of a conductor
connecting the leader tip to the starting point of the discharge is performed by
the leader plasma channel. It takes a fairly long time for a leader to develop -
up to 0.01 s, which is eternity in the time scale of fast processes involving an
electric impulse discharge. During this period of time, the leader plasma
must be maintained highly conductive, and this may become possible only if
the gas is heated up to an electric arc temperature, i.e. above 5000-6000K.
The problem of the channel energy balance necessary for the heating and com-
pensation for losses is a key one in leader theory. It is discussed in chapters 2
and 4, as applied to various kinds of lightning discharge.
A leader is an indispensable element of any spark. The initial and all
subsequent components of a flash begin with a leader process. Although its
mechanism may vary with the spark polarization, propagation direction
and the serial number of the component, the process remains essentially
the same. This is the formation of a highly conductive plasma channel due
to the local enhancement of the electric field in the leader tip region.
A return stroke is produced at the moment of contact of a leader with
the ground or a grounded object. Most often, this is an indirect contact: a
counterpropagating leader, commonly termed a counterleader, may start
from an object to meet the first leader channel. The moment of their contact
initiates a return stroke. During the travel from the cloud to the ground, the
lightning leader tip carries a high potential comparable with that of the cloud
at the spark start, the potential difference being equal to the voltage drop in
the leader channel. After the contact, the tip receives the ground potential
and its charge flows down to the earth. The same thing happens with the
other parts of the channel possessing a high potential. This ‘unloading’ pro-
cess occurs via a charge neutralization wave propagating from the earth up
through the channel. The wave velocity is comparable with the velocity of
light and is about 10’ mjs. A high current flows along the channel from the
wave front towards the earth, carrying away the charge of the unloading
channel sites. The current amplitude depends on the initial potential distribu-
tion along the channel and is, on average, about 30 kA, reaching 200-250 kA
for powerful lightning sparks. The transport of such a high current is accom-
panied by an intense energy release. Due to this, the channel gas is rapidly
heated and begins to expand, producing a shock wave. A peal of thunder
is one of its manifestations.
The return stroke is the most powerful stage of a lightning discharge
characterized by a fast current change. The current rise can exceed
10’ A/s, producing a powerful electromagnetic radiation affecting the
performance of radio and TV sets. This effect is still appreciable at a distance
of several dozens of kilometres from the lightning discharge.

Current impulses of a return stroke accompany all components of a

descending spark. This means that the leader of every component charges
the channel as it moves down to the earth, but some of the charge becomes
neutralized and redistributed at the return stroke stage. Prolonged peals of
thunder result from the overlap of sound waves generated by the current
impulses from all subsequent spark components.
An ascending spark is somewhat different. The leader of the first compo-
nent starts at a point of zero potential. As the channel travels up, the tip
potential changes gradually until the leader development ceases somewhere
deep in the cloud. There is no fast charge variation during this process; as
a result, the first component has no return stroke. However, all subsequent
spark components starting from the cloud do develop return strokes and
behave exactly in the same way as a descending spark.
Of special interest is the return stroke of an intercloud discharge. Its
existence is indicated by peals of thunder as loud as those of descending
sparks. Clearly, an intercloud leader is generated in a charged region of a
storm cloud, or in a storm cell, and travels towards an oppositely charged
region. The charged region of a cloud should not be thought of as a con-
ductive body, something like a plate of a high voltage capacitor. Cloud
charges are distributed throughout a space with a radius of hundreds of
metres and are localized on water droplets and ice crystals, known as
hydrometeorites, having no contact with one another. The formation of a
return stroke implies that the leader comes in contact with a highly con-
ductive body of an electrical capacitance comparable with, or even larger
than, that of the leader. It appears that the role of such a body in an
intercloud discharge is played by a concurrent spark coming in contact
with the first one.
Measurements made at the earth surface have shown that the current
impulse amplitude of a return stroke decreases, on average, by half for
about lOP4s. This parameter variation is very large - about an order of
magnitude around the average value. Current impulses of positively charged
sparks are usually longer than those of negatively charged ones, and the
impulses of the first components last longer than those of the subsequent
ones.
A return stroke may be followed by a slightly varying current of about
100 A, which may persist in the spark channel for some fractions of a second.
At this final stage of continuous current, the spark channel remains electri-
cally conductive with the temperature approximately the same as in an arc
discharge. The continuous current stage may follow any lightning compo-
nent, including the first component of an ascending spark which has no
return stroke. This stage may be sporadically accompanied by current over-
shoots with an amplitude up to 1 kA and a duration of about s each.
Then the spark light intensity becomes much higher, producing what is gen-
erally termed as M-components.

Continuous and stepwise leaders 9
1.4 Continuous and stepwise leaders
This introductory chapter contains no theory, and this makes the discussion
of leader details a very complicated task. So we shall mention only its
principal features which can be registered by a continuously moving film.
Continuous streak photographs show lightning development in time. One
needs, however, a certain skill and experience to be able to interpret them
adequately. Suppose a small light source moves perpendicularly to the
earth at a constant velocity. It may be a luminant bomb descending with a
parachute. A film moving horizontally, i.e. in the transverse direction, at a
constant speed will show a sloping line (figure lS(u)). Given the film speed
(the display rate), one can easily calculate the light source velocity from
the line slope. A uniformly propagating vertical channel will leave on a
film a sloping wedge (figure 1.5(b)) rather than a line. From its slope, too,
one can find the channel velocity, or its propagation rate. The higher the
rate of the process in question, the higher must be the display rate in
streak photography. The highest display rates can be obtained using an
electron-optical converter, in which an image is converted to an electron
beam scanned across the screen by an electric field. A conventional photo-
camera registers the displayed electronic image from the screen onto an
immobile film. Electron-optical converters have provided much information
on long sparks, but their application in lightning observations has been
limited. The main results here have been obtained using mechanical streak
cameras. We described this technique and analysed streak pictures in our
book on long sparks [8].
Figure 1.6(a) shows the leader of an ascending lightning spark going
up from the top of a grounded tower in the electric field of a negatively
charged cloud cell. The leader carries a positive space charge and, therefore,
it should be referred to as a positive leader. One can clearly see the bright
trace of the channel tip, which looks like a nearly continuous line. This
kind of leader is known in literature as a continuous leader. The changing
trace slope suggests that the leader velocity changes during its propagation.
These changes are, however, quite smooth, not interrupting the tip travel up
to the cloud.
0 t o t
a b
Figure 1.5. The analysis of an image of a vertically descending light source in a

horizontally moving film (image display in streak photography): (a) point source,
(b) elongating channel.

Figure 1.6. A schematic streak picture of a positive ascending (a) and a negative
descending (b) lightning leader.
An essentially different behaviour is exhibited by the leader shown in

figure 1.6(b). The channel grows in a stepwise manner, covering several
dozens of metres in each step. Hence, this kind of leader is termed as a
stepwise leader. The new step in the photograph is especially bright; its
appearance makes the whole channel behind it also a little brighter. The
step length varies between 10 and 200m with an average of 30m. The time
lapse between two steps is 30-90 ps [9]. The stepwise pattern is characteristic
of negatively charged leaders. Positive leaders, both ascending and descend-
ing, usually grow in a continuous manner. When averaged over the total time
of development, the velocity of stepwise and continuous leaders prove nearly
the same, 105-106m/s, with an average of about 3 x 105m/s.
If the leader of the next component moves along the hot track of the
first one, it always develops continuously. The new process, termed a dart
leader, differs from the first one exclusively in a high leader velocity, about
(1-4) x 107m/s. It does not change much along its trajectory from the
cloud to the earth. Streak photographs clearly show the bright head of a
dart leader, while the channel light intensity is much lower. If the next
component takes its own path, its leader behaves in the same way as that
of the first component, i.e. it develops more slowly and often in a stepwise
pattern.
Dart leaders have not had a fair share of attention from researchers.
There is neither theory nor laboratory analogue of this type of gas discharge.
Still, it is a most fascinating form of discharge developing record high leader
velocities. The contact of a dart leader with the earth produces the fastest
current rise, which can reach its amplitude maximum within s. This is
the source of record strong electromagnetic fields which exert one of the
most hazardous effects on modern equipment. An attempt at a theoretical
treatment of the dart leader will be made in section 4.8.

Lightning stroke frequency 11
1.5 Lightning stroke frequency

1.5.1 Strokes at terrestrial objects
Experience shows that lightning most frequently strikes high objects,
especially those dominating over an area. In a flat country, it is primarily
attracted by high single objects like masts, towers, etc. In mountains, even
low buildings may be affected if they are located on a high hill or on the
top of a mountain. Common sense suggests that it is easier for an electrical
discharge, such as lightning, to bridge the shortest gap to the highest
object in the locality. In Europe, for example, a 30m mast experiences, on
the average, 0.1 lightning stroke per year (or 1 stroke per 10 years), whereas
a single lOOm construction attracts 10 times more lightnings. On closer
inspection, the strong dependence of stroke frequency on the construction
height does not look trivial. The average altitude of the descending discharge
origin is about 3 km, so a lOOm height makes up only 3% of the distance
between the lightning cloud and the earth. Random bendings make the
total lightning path much longer. One has to suggest, therefore, that the
near-terrestrial stage of lightning behaviour involves some specific processes
which predetermine its path here. These processes lead to the attraction of a
descending leader by high objects. We shall discuss the attraction mechanism
in chapter 5.
Scientific observations of lightning show that there is an approximately
quadratic dependence of the stroke frequency NI on the height h of lumped
objects (their height is larger than the other dimensions). Extended objects
of length I , such as power transmission lines, show a different dependence,
NI N hl. This suggests the existence of an equivalent radius of lightning
attraction, Re, h. All lightnings displaced from an object horizontally at
N
a distance r 6 Re, are attracted by it, the others missing the object. This
primitive pattern of lightning attraction generally leads to a correct result.
For estimations, one can use Re! RZ 3h and borrow the stroke frequency
per unit unperturbed area per unit time, nl, from meteorological observa-
tions. The latter are used to make up lightning intensity charts. For example,
the lightning intensity in Europe is nl < 1 per 1 km2 per year for the tundra,
2-5 for flat areas, and up to 10 for some mountainous regions such as the
Caucasus. A tower of h = lOOm is characterized by Re, = 0.3 km with
NI = n17rR&M 1 stroke per year at the average value of nl = 3.5 kmP2year-’.
This estimation is meaningful for a flat country and only for not very high
objects, h < 150m, which do not generate ascending lightnings.
1.5.2 Human hazard

It has long been proved that Galvani was wrong suggesting a special ‘animal
electricity’. A human being is, to lightning, just another sticking object, like a
tree or a pole, only much shorter. The lightning attraction radius for humans

is as small as 5-6m and the attraction area is less than lop4 km2. If a man
had stopped alone in the middle of a large field two thousand years ago,
he might have expected to attract a direct lightning stroke only by the end
of the third, coming millennium. In actual reality, however, the number of
lightning victims is large, and direct strokes have nothing to do with this.
It is known from experience that one should not stay in a forest or hide
under a high tree in an open space during a thunderstorm. A tree is about
10 times higher than a man, and a lightning strikes it 100 times more fre-
quently. When under the tree crown, a man has a real chance to be within
the zone of the lightning current spread, which is hazardous.
After a lightning strikes the tree top, its current ZM runs down along its
stem and roots to spread through the soil. The root network acts as a natural
grounding electrode. The current induces in the soil an electric field E = pj,
where p is the soil resistivity and j is the current density. Suppose the current
spreads through the soil strictly symmetrically. Then the equipotentials will
represent hemispheres with the diagonal plane on the earth’s surface. The
current density at distance r from the tree stem is j = IM/(27rr2)the field is
IMp/(27rr2)and the potential difference between close points r and r + Ar
+
is equal to A U = (ZMp/27r)[r-’ - (Y AY)-’] x E(r)Ar.If a person is stand-
ing, with his side to the tree, at distance r = 1 m from the tree stem centre and
the distance between his feet is Ar x 0.3 m, the voltage difference on the soil
with resistivity p = 200 f2/m will be A U x 220 kV for a moderate lightning of
ZM = 30 kA. This voltage is applied to the shoe soles and, after a nearly inevi-
table and fast breakdown, to the person’s body. There is no doubt that the
person will suffer or, more likely, will be killed - the applied voltage is too
high. Note that this voltage is proportional to Ar. This means that it is
more dangerous to stand with one’s feet widely apart than with one’s feet
pressed tightly together. It is still more dangerous to lie down along the
radius from the tree, because the distance between the extreme points
contacting the soil becomes equal to the person’s height. It would be much
safer to stand still on one foot, like a stork. But it is, of course, easier to give
advice than to follow it. Incidentally, lightning strikes large animals more
frequently than humans, also because the distance between their limbs is larger.
If you have a cottage equipped with a lightning protector, take care that
no people could approach the grounding rod during a thunderstorm. The
situation here is similar to the one just described.
1.6 Lightning hazards

1.6.1 A direct lightning stroke
In the case of a direct lightning stroke, the current flows through the
conducting elements of the affected object, with the hot channel contacting
the construction element which has received the stroke.

Ligk tning hazards 13
Figure 1.7. Traces of lightning strokes at the steel tip of the Ostankino Television
Tower in Moscow.
Thermal effects of lightning are most hazardous at the site of contact of

a high temperature channel with combustible materials. This often leads to a
fire which becomes most probable when the continuous current stage has a
long duration. A return stroke is unlikely to cause a fire even in the case of
a powerful lightning discharge, because the strong shock wave produced
blows off the flames and combustion products. In combustible dielectric
materials a lightning stroke contacts on its way may first be broken down
by the strong electric field of the leader tip and then, in the return stroke
and continuous current stages, they may be melted through at the site of
contact with the hot spark. A burn-through or a burn-off often occurs at
the point where the spark contacts a metallic surface several millimetres
thick. The holes and burn-offs are usually of the same size. The photograph
in figure 1.7 demonstrates the traces of numerous lightning strokes on the
steel tip of the Ostankino Television Tower. Slight faults cannot disturb
the mechanical strength of a massive metallic construction. Normally, the
hazards of burn-offs and fuses are associated with the melted metal in-flow
into an object which may contain inflammable and explosive materials or
gas mixtures. Incidentally, not only is a burn-through of a metallic wall
dangerous but also the local overheating when the temperature of the
inner metal surface may go up to 700-1000°C. Unfortunately, the surface
often acts as a lighter.

14 Introduction: lightning,its destructive effects and protection
Thermal damage of conductors, through which lightning current flows,

occurs fairly rarely. It is characteristic of miniature antennas and various
detectors mounted on the outer construction surfaces. The probability of
emergency increases if lightning current encounters bolted or riveted joints.
The electric contact thus formed always has an elevated contact resistance
which may cause a local overheating. This results in the metal release and
rivet loosening, disturbing the mechanical strength of the joint. Mobile
joints (hinges, ball bearings, etc.) are subject to a similar damage. The site
of a sliding contact becomes locally overheated to produce cavities which
hamper the motion of mobile parts. In extreme conditions, they may
become welded.
Electrodynamic effects of lightning current rarely become hazardous.
Mechanical stress arising in electrically loaded and closely spaced metallic
structures or in a single structure with an abruptly changing direction of
the current is not appreciable and lasts less than 100 ms (it is the attenuation
time of a current impulse). However, lightning current has been repeatedly
observed to narrow down thin metallic pipes, to change the tilt of rods and
to strain thin surfaces. Such effects are not vitally dangerous in themselves
but, under certain conditions, may lead to an emergency. As an illustration,
imagine the situation when the lightning-affected pipe is part of an aircraft
speed control. What will happen if the crew take the readings for granted
and do not receive corrections from a ground air traffic controller?
Electrohydraulic effects of lightning are much more hazardous than
those discussed above. Modern machines have parts made from a variety
of composite materials. These may include, along with plastics, superthin
metallic films (both outer and inner), nearly as thin metallic meshes, and min-
iature conductors monolithic with a dielectric wall. Under the action of light-
ning current, these metallic parts evaporate, the arising arcs contacting the
plastic making it decompose and evaporate. A shock wave appears which
splits and bloats the composite wall. A similar effect arises when a lightning
spark partially penetrates through a narrow slit between vaporizable plastic
walls (most plastics possess gas-generating properties). No one questions a
great future of composite materials, but their peaceful coexistence with light-
ning is still a challenge to the engineer.
Direct stroke overvoltage represents a hazardous rise of voltage when
a lightning current impulse propagates across the construction elements.
We shall analyse this very dangerous effect of lightning with reference to
a power transmission line, because engineers first encountered the phenom-
enon of overvoltage in such lines. Moreover, the problem of electric
insulation for a transmission line can be stated most clearly. Figure 1.8
shows schematically a metallic tower with a ground rod (the grounding
resistance is Rg)and a high voltage wire suspended by an insulator string.
Above the wire, there may be a lightning conductor attached right to the
tower. It stretches along all the line and is to trap lightning sparks aimed

Lightning hazards 15
Figure 1.8. Lightning current as an overvoltage source on a power transmission line

during a stroke at the power wire (a) and a grounded tower (b).
at the line wires. A rigorous solution to this problem is given later in

this book. Here, we should only like to explain the nature of overvoltages
phenomenologically.
Suppose at first that the lightning conductor has proved unreliable, and
a discharge has struck the wire (figure 1.8(a)). At the point of the stroke, the
current will branch to produce two identical waves of the amplitude ZM / 2 ,
where ZM is the lightning current amplitude. The two waves will run towards
the ends of the line with a velocity nearly equal to vacuum light velocity,
c = 3 x 10sm/s. Until the end-reflected waves return, the wire potential
relative to the ground will rise to U , = Z,2/2. The wave resistance
Z = (L1/C1)1/2 in this expression is defined by the inductance L1 and the
capacitance C1 per unit wire length; it varies slightly, between 250 and
350R, with the height and the wire radius. With this wave resistance, the
average lightning current with an amplitude ZM = 30 kA will raise the wire
potential up to U , = 3750-5250 kV. The tower potential will practically
remain unchanged and equal to zero, so the insulation overvoltage will be
close to the calculated value of U,. This will be clear if we compare U ,
with the operating line voltage which does not exceed 1000 kV even in high
power lines but normally is 250-500 kV.
In reality, the distance to the line ends I is as large as many dozens of
kilometres. The time it takes the reflected wave of the opposite sign cutting
down the overvoltage to arrive back at the stroke point is At = 21/c, or
many hundreds of microseconds. This time is much longer than the strong
current duration in the return stroke (loops). For this reason, reflected
waves, which become strongly attenuated, do not normally have enough
time to interfere with the process so that the overvoltage acts as long as a
lightning current impulse. Practically, any lightning stroke at a wire
represents a real hazard: the insulation will be broken down to produce
short-circuiting. The power line in that case must be disconnected.
Suppose now that lightning has struck a tower. More often, this is
actually not a tower but rather an overhead grounded wire connected to it.

The lightning current will flow down the metallic tower to the ground
electrodes to be dissipated in the earth. Let us take point A at the height
of the insulator string connection. Due to the lightning current i(t), the
potential at this point, pA,will differ from the zero potential of the earth
by the voltage drop in the grounding resistance R, and in the tower induc-
tance L, between the tower base up to the point A:
di
p = R,i+ L,-
dt
However, the power wire potential will practically remain the same (in this
qualitative description, we ignore all inductances between the power wires,
tower and grounded wire). The power wire potential is due to the operating
voltage source of the power line: qw = Uop.Then, the insulator string voltage
will be
di
U = pw - (FA = Uop - R,i - Ls-,
dt
Note that the lightning current and operating voltage may have different
polarities. As a result, the overvoltage U may prove to be the sum of the
three terms in equation (1.2).
The inductance component of the overvoltage, L,di/dt, has a short
lifetime: it acts about as long as the lightning current rises. For a current
impulse with an average amplitude IM 30kA and an average rise time
tf = 5 p , the inductance voltage at L, 50pH will be about 300kV. The
resistance component U , at a typical grounding resistance R, = 10R will
have about the same value but will act an order of magnitude longer, i.e.,
as long as the lightning current flows. For this reason, this component
makes the principal contribution to the insulation flashover.
The emergency situation just described is not as bad as a direct stroke at
a power wire when the same lightning current can induce an order of magni-
tude higher voltage. The insulation of a ultrahigh voltage line can withstand
short overvoltages up to 1000- 1500kV and seldom suffers from lightning
strokes at a tower or a lightning protection wire. To produce a harmful
effect, the lightning current must be 3-5 times the average value. Lightning
strokes of this power do not occur frequently, making up less than 1% of
all strokes. Quite different is the effect of a direct stroke at a power network
with an operating voltage of 35 kV and lower. The insulation system will
suffer equally from a stroke at a power wire or a tower. It is no use protecting
such a line with grounded wire.
Insulation flashover due to the tower potential rise is referred to as
reverse flashover. This name does not imply the definite direction of the
discharge development but only indicates the direction from which the
potential rises, i.e., the grounded end of the insulator string rather than
the power wire.

The above illustration of overvoltages on the line transmission insula-

tion demonstrates, to some extent, a variety of mechanisms of direct light-
ning current effects. In actual reality, such mechanisms are much more
diverse. It is important to remember that in modern technologies, overvol-
tages are not always measured in hundreds of kilovolts, as for high-voltage
transmission lines. Short voltage rises of only 100-1OV may be hazardous
to microelectronic devices. Of special interest in this connection are situa-
tions when lightning current flows across solid metallic jackets with electric
circuits inside. These problems are discussed in chapter 6.
1.6.2 Induced overvoltage

Induced overvoltage is the most common and dangerous effect of lightning
on electric circuits of modern technical equipment. This effect is brought
about by electromagnetic induction. The current flowing through the
lightning spark and the metallic structures of an affected object generates
an alternating magnetic field which can induce an induction emf in any of
the circuits in question. The procedure of estimating induced overvoltages
is quite simple. If BaV(t)is the magnetic induction averaged over the circuit
cross section S , the induction emf is expressed as
When the length of the current conductor inducing the magnetic field is much
longer than the distance to the circuit, rc, and when the width of the circuit
normal to the magnetic field is much smaller than rc, we have
poS di
EemfM --
27rrc dt
where po = 47r x lop7Him is vacuum magnetic permeability. The order of
magnitude of the induction emf amplitude is defined as
where A,, is the maximum rate of the current impulse rise equal to 10" A/s
for the subsequent components of a powerful lightning flash. A circuit of area
S = 1 m2 located at a distance r, = 10 m from a lightning current conductor
may become the site of induced overvoltage with an amplitude up to 20 kV.
This value is only an arbitrary guideline, because induced overvoltage may
vary with the circuit area, its orientation and distance from the lightning cur-
rent. Circuits with an area of hundreds and thousands of square metres may
be created by large industrial metallic constructions and power transmission
lines. The distance between the circuit and the current flow may also vary
greatly. For such diverse parameters of a system, the problem will be more com-
plicated in the case of fast current variations along the spark and in time. It

cannot then be approached as a quasi-stationary problem, but one must take

into account the law of current wave propagation along the spark channel
and the finite velocity of the electromagnetic field in the space between the chan-
nel and the circuit. Solutions to such problems are illustrated in chapter 6.
There is another class of situations associated with electromagnetic
induction in screened volumes. Of special interest is the situation when the
lightning current i flows across a solid metallic casing and the circuit in
question is inside it. Unless the casing is circular, an emf-inducing magnetic
field gradually appears inside the casing. It is remarkable that the time
variation of the emf is not at all defined by dildt. The magnetic field going
through the circuit is affected more by the relatively slow current re-
distribution along the casing perimeter than by the time variation of the light-
ning current. The problem of pulse induction in aircraft inner circuits or in
screened multiwire cables ultimately reduces to the problem above. Some
approaches to its solution will be considered in chapter 6.
1.6.3 Electrostatic induction

Benjamin Franklin felt the effect of electrostatic induction when he raised his
finger up to a lifted wire during a thunderstorm. The electric field of a storm
cloud had polarized the wire by separating its electric charges. The strong
electric field of the polarization charge had broken down the air gap between
the thin wire end and the explorer’s finger and carried the charge through his
body to the earth.
Electrostatic induction induces a charge in any grounded conductor or a
metallic object. Suppose it is a vertical metallic rod of length 1 located in an
external vertical field Eo. When insulated from the earth, the rod would take
the potential of the space at its centre, pc = E01/2, which follows from the
symmetry consideration. The grounded rod potential is zero; hence, the
external field potential is compensated by the charge q1 induced by this
field on the rod. The charge can be estimated from the rod capacitance C,
as q1 = Crpc= E01C,/2.
The production of the charge q1implies the existence of current through a
ground electrode of the object. This is a low current, because it takes several
seconds for the cloud charge, creating the field Eo, to be formed. As much
time, At, is necessary for the charge -ql to flow down into the earth, leaving
behind the induced charge q1on the conductor. If the field Eo is largely created
by the leader charge of a close Lightning discharge, the exposure time of the
induced charge reduces to At x 10-3-10-2 s. But in this case, too, the current
through the ground electrode is low. For example, at Eo x 1 kV/cm character-
istic of close discharges, I = 10m, and C,. = lOOpF,t the average current is
t Approximately, C, = 27rq,l/ In h / r , where r is the rod radius, h is an average distance between
the rod and the earth (h = l / 2 ) , z0 = 8.85 x F/m is the vacuum dielectric permittivity. At
1 = 10m and r = 2cm, C, = 100pF.

ii x qi/At = EolCr/2Atx 0.5-0.05 mA. Even if the grounding resistance is as

high as R, x 10 kR (we deal here with a damaged ground electrode when the
connection with the grounding circuit is made across the high contact
resistance of the break), the induced charge current will change the rod poten-
tial relative to the earth by the value Ap = iiRg= 5-0.5 V. This potential rise
can be ignored in any situation.
When a lightning channel reaches the earth and the process of leader
charge neutralization begins in the return stroke, the field at the earth, Eo,
rapidly drops to zero, eliminating the charge qi. The same charge now
flows back through the grounding resistance for a much shorter time,
about 1 ps, and the current ii increases to about 0.5 A. At the same resistance
R, x 10 kR the voltage will rise to 5 kV. In practice, it may rise even higher,
producing a spark breakdown at the site of poor contact. The breakdown
may become very dangerous if there are explosive gas mixtures nearby,
since the spark energy is sufficiently high to set a fire.
There is another mechanism of igniting sparks in an induced charge field,
which may be hazardous even in the case of perfect grounding of a metallic
construction. Suppose that a grounded rod of length I and radius r is in the
leader electric field Eo of a nearby lightning discharge. The charge induced
on the rod will enhance the field at its top approximately by a factor of I / r .
With I >> r, this is sufficientto excite a weak counterpropagating leader process.
Of course, if this leader is only about 10cm long, it will have no effect on the
lightning trajectory. Its energy is, however, large enough to ignite an inflam-
mable gas mixture, if there is any in the vicinity, since the channel temperature
is close to 5000 K and its lifetime is as long as that of a lightning leader.
1.6.4 High potential infection

This unsuitable term has long been used in Russian literature on lightning
protection. It means that the surface and underground service lines, which
get into a construction to be protected, may introduce in it a potential
different from the zero potential of the construction metalwork connected
to earth connection. This may become possible if a service line is not
linked to the grounding of the construction but connected or passes close
to the earth connection of another construction loaded by lightning current
during a stroke (figure 1.9). This may also be a natural earth connection
formed at the moment of lightning contact with the earth due to an intense
ionization in it. If the introduced potential is high, it causes a spark break-
down between the service line and a nearby metallic structure of the
object, whose potential is zero owing to the earth connection. The scenario
of the emergency that follows has been described above.
To avoid sparking induced by high potential infection, all metallic
service lines subject to explosion zooms are linked to the earth connection
of the construction. All metalwork potentials are equalized. The connection,

Figure 1.9. Schematic input of high potential from remote lightning strokes.
however, becomes loaded by additional current, which finds its way there
from a remote lightning stroke, using the service line as a conductor.
When the earth connection resistance is low and the service line goes through
the ground with a high resistivity so that the current leakage through the side
surface is not large, nearly all of the lightning current arrives at the con-
nection from the stroke site. This situation appears to be somewhat similar
to a direct lightning stroke. Sometimes, special measures must be taken to
restrict the infection current. A detailed treatment of the problem of current
and potential infections will be offered in chapter 6.
1.6.5 Current inrush from a spark creeping along the earth’s surface
This phenomenon is familiar to all communications men who have to repair
communications cables damaged by lightning. The damaged site can be
detected easily, because it is indicated by a furrow in the ground extending
far away from the stroke site. A furrow may be as long as several dozens
of metres, or 100-200m in a high resistivity ground. Such a long gap can
be bridged by a spark because of the electric field created by the spark current
spreading out through the ground. The mechanism of spark formation along
a conducting surface differs from that of a ‘classical’ leader propagating
through air. A creeping spark can develop in low fields and have a very
high velocity.
Underground cables are not the only objects suffering from creeping
spark current. Similarly, it can find its way to underground service lines
and to the earth connections of constructions well equipped by lightning
protectors. But a protector palisade cannot stop lightning. When the conven-
tional way from the earth surface is blocked, it breaks through from beneath,
making a bypass in the ground. Lightning thus behaves very much like a
clever general in ancient times, who ordered his soldiers to make a secret
underground passageway instead of attacking openly the impregnable
castle walls. It is reasonable to suggest that the contact of a creeping spark
with combustible materials is as frequent a cause of a fire as a direct lightning
stroke.
The details of the creeping discharge mechanism have been unknown
until quite recently. They are analysed in chapter 6.

Figure 1.10. This lightning has missed the teletower tip by over 200m.
1.6.6 Are lightning protectors reliable?

Lightning protectors are believed to be reliable, since their design has
changed but little over two and a half centuries. Nevertheless, the photo-
graph in figure 1.10 makes one question this judgement: the lightning
struck the Ostankino Television Tower 200 m below its top, i.e., the Tower
could not protect itself. This is not an exception to the rule. Most descending
discharges missed the Tower top more or less closely, contrary to what had
been expected. This is a serious argument against the vulgar explanation of
the major principle of protector operation that lightning takes a shortcut
at the final stage of its travel to the earth. There are also other arguments,
perhaps not as obvious but still convincing.
Breakdown voltage spread is registered in long gaps even under strictly
identical conditions. The breakdown probability 9 varies with the pulse
amplitude of test voltage U (figure 1.11). Deviations from the 50% proba-
bility voltage, Use%,, are appreciable and may be 10-15% either way.
Curve 2 in figure 1.11 shows the probability function !F( U ) for a shorter
gap. In certain voltage ranges, both curves promise breakdown probabilities
remarkably different from zero. This means that if two different gaps are
tested simultaneously, there is a chance that any of them (the smaller and
the larger gap) will be bridged. In general, this situation is similar to that aris-
ing when a lightning discharge is choosing a point to strike at. It does not
always take the shortest way to a protector but, instead, may follow a
longer path in order to attack the protected object.
For solving the lightning path problem, one has to treat a multielectrode
system consisting of several elementary gaps. For lightning, all elementary

Figure 1.11. Distributions of breakdown voltages in air gaps of various lengths with a
sharply non-uniform electric field.
gaps have a common high voltage electrode (the leader that has descended to
a certain altitude), while the zero potential electrodes are formed by the
earth’s surface with grounded objects and protectors distributed on it. The
problem of protector effectiveness thus reduces to the calculation of
breakdown probabilities for the elementary gaps in a multielectrode
system. The general formulation of this problem is very complex, since the
spark development in the elementary gap in real conditions cannot be
taken to be independent. The discharge processes affect one another by
redistributing their electric fields, which eliminates straightforward use of
statistical relations describing independent processes.
We cannot say that the spark discharge theory for a multielectrode
system has been brought to any stage of completion. But what has been
done, theoretically and experimentally, allows the formulation of certain
concepts of the lightning orientation mechanism and the development of
engineering approaches to estimate the effectiveness of protectors of various
heights (see chapter 5).
Investigation of multielectrode systems is also important from another
point of view: we must find ways of affecting lightning actively. It would
be reasonable to leave the discussion of this issue for specialized chapters
of this book, but they will, however, attract the attention of professionals
only, or of those intending to become professionals. It is not professionals
but amateurs who, most often, try to invent lightning protectors. They
have at their disposal a complete set of up-to-date means: lasers, plasma
jets, corona-forming electrodes for cloud charge exchange, radioactive

Lightning as a power supply 23
sources, high voltage generators stimulating counterpropagating leaders, etc.

That lightning management has a future has been confirmed by laboratory
studies on sparks of multimetre length. These experiments and their implica-
tions will be analysed below, so there is no point in discussing them here. Still,
it is hard to resist the temptation to make some preliminary comments
addressed to those who like to invent lightning protection measures.
When explaining the leader mechanism at the beginning of this chapter,
we noted that the leader tip carries a strong electric field sufficient for an
intense air ionization. It is very difficult to act on this field directly, because
it would be necessary to create charged regions close by, whose charge den-
sity and amount would be comparable with those in the immediate vicinity of
the tip. Pre-ionization of the air by radioactive sources is of little use because
of the low air conductivity after radiative treatment. The initial electron
density behind the ionization wave front in the leader process is higher
than 10l2~ m - and ~ , in a ‘mature’ leader it is at least an order of magnitude
higher. These and even much lower densities are inaccessible to radiation at a
distance of dozens of metres from the radiation source which must present no
danger to life. The same is true of a gradual charge accumulation due to a
slow corona formation between special electrodes. Besides, one cannot pre-
dict the polarity of a particular spark to decide which charge is to be
pumped into the atmosphere.
Quite another thing is plasma generation. In principle, we could create a
plasma channel comparable with the lightning rod height, thus increasing its
length. A high power laser could, in principle, be used as a plasma source. It
is clear that it should be a pulse source and the plasma produced should have
a short lifetime. It must be generated exactly at the right moment, when a
lightning leader is approaching the dangerous region near the object to be
protected. This is a new problem associated with synchronization of the
laser operation and lightning development, giving a new turn to the task of
lightning protection, which does not at all become easier.
Finally, we should always bear in mind that most lightning discharges are
multicomponent. In about half of them, the subsequent components do not
follow the path of the first component. In fact, these are new discharges
which would require individual handling. To prepare a laser light source for
a new operation cycle for a fraction of a second is possible but difficult techni-
cally. The cost of such protection is anticipated to be close to that of gold.
It is not our intention to intimidate lightning protection inventors. We
just want to warn them against excessive enthusiasm.
1.7 Lightning as a power supply
The question of whether lightning could serve as a power supply cannot be

answered positively, no matter how much we wish it to be one. Some authors

of science fiction books force, quite inconsiderately, their characters to har-

ness lightning in order to use its electric power. Even without this service,
lightning has done much for people by stimulating their thought. The
energy of a lightning flash is not very high. The voltage between a cloud
and the earth can hardly exceed l00MV even in a very powerful storm,
the transported charge is less than lOOC, and maximum energy release is
10'OJ. This is equivalent to one ton of trinitrotoluene or 2-4 ordinary
airborne bombs. A family cottage consumes more power for heating, illumi-
nation, and other needs over a year. Actually, only a small portion of the
lightning power is accessible to utilization, while most of it is dissipated in
the atmosphere.
Normally, a person lives through 40-50 thunder storm hours during a
year. All storms send to the earth an average of 4-5 lightning sparks per
square kilometre of its surface providing a power of less than 1 kW/km2
per year. In a country of 500 x 400km2, this is about 200MW, which is
a very small value compared with the electrical power produced by an indus-
trial country. Just imagine the immense net which would be necessary for
trapping lightning discharges in order to collect such a meagre power!
Other natural power sources, such as wind, geothermal waters, and tides,
are infinitely more powerful than lightning, but they are still not utilized
much. Clearly, we should not even raise the problem of lightning power
resources.
1.8 To those intending to read on
There will be no more popularized stories about lightning in this book. Nor
shall we mention ball lightning here. The next chapter will contain a
thorough analysis of available data and theoretical treatments of the long
spark, because we believe that without these preliminaries the lightning
mechanism may not become clear to the reader. Nature has eagerly employed
standard solutions to its problems, so lightning is quite likely to represent the
limiting case of the long spark. It would be useful for readers to familiarize
themselves with our previous book S p a r k Discharge, because it is totally
concerned with this phenomenon. But even without it, they will be able to
find here basic information on long sparks. We have tried to describe their
general mechanisms and to give predictions as to their extension to air
gaps of extrema1 length. Even for this reason alone, the next chapter is not
a summary of the previous book. Lightning is as complicated a phenomenon
as the long spark and is definitely more diverse. It is a multicomponent
process. Since its subsequent components sometimes take the path of an
earlier component, we must consider the effects of temperature and residual
conductivity in the spark channel on the behaviour of new ionization
waves.

To those intending to read on 25
Even a simple model should not treat a kilometre spark in terms of

electrical circuits with lump parameters. A lightning spark is a distributed
system. The time for which the electric field perturbation spreads along the
sparks is comparable with the duration of some of its fast stages. The
allowance for the delay can, in some cases, change the whole picture
radically. This requires new approaches to lightning treatments. Experimen-
tal data and theoretical ideas concerning the lightning leader and return
stroke are discussed together. First, there are not many of them. On the
other hand, we have tried to point out ideological relationships between
experiment and theory and to offer a more or less consistent physical
description.
Spark discharges in a multi-electrode system are the subject of a special
chapter. We present available data and analyse possible mechanisms of light-
ning orientation. This is, probably, the most debatable part of the book.
Field studies of lightning orientation are very difficult to carry out primarily
because constructions of even 100-200 m high are affected by descending
discharges only once or twice a year. The observer must have exceptional
patience and substantial support to be able to reveal statistical regularities
in lightning trajectories. From field observations, one usually borrows the
statistics of lightning strokes at objects of various height and, sometimes,
the statistics of strokes at protected objects, such as power transmission
lines with overhead grounding wire connections. This material, however, is
too scarce to build a theory. For this reason, one has to refer to laboratory
experiments on long sparks generated in 10-15m gaps. No one has ever
proved (or will ever do so) the geometrical similarity of sparks; therefore,
experimental data can be extended to lightning only qualitatively. Neverthe-
less theoretical treatments must be brought to conclusion when one develops
recommendations on particular protector designs. We analyse the reliability
of engineering designs, wherever possible.
The last chapter of the book discusses lightning hazards and protection
not only in terms of applications. Even the classical theory of atmospheric
overvoltages in power transmission lines required the solution of com-
plicated electrophysical problems. Thorough theoretical treatments are
necessary for the analysis of lightning current effects on internal circuits of
engineering constructions with metallic casings, on underground cables,
aircraft, etc. The range of problems to be considered is not limited to
electromagnetic field theory. We shall also discuss gas discharge mechanisms
of a spark creeping along a conducting surface, the excitation of leader
channels in air with the composition and thermodynamic characteristics
locally changed by hot gas outbursts, and the lightning orientation under
the action of the superhigh operating voltage of an object. These theoretical
considerations will not screen our practical recommendations concerning
effective lightning protection and the application of particular types of
protectors.

References
[l] Bazelyan E M, Gorin B N and Levitov V I 1978 Physical and Engineering Funda-
mentals of Lightning Protection (Leningrad: Gidrometeoizdat) p 223 (in Russian)
[2] McEachron K 1938 Electr. Engin. 57 493
[3] Berger K and Vogrlsanger E 1966 Bull. SEV 57 No 13 1
[4] Newman M M, Stahmann J R, Robb J D, Lewis E A et a1 1967 J . Geophys. Res.
72 4761
[5] Uman M A 1987 The Lightning Discharge (New York: Academic Press) p 377
[6] Berger K, Anderson R B and Kroninger H 1975 Electra 41 23
[7] Schonland B, Malan D and Collens H 1935 Proc. Roy. Soc. London Ser A 152 595
[8] Bazelyan E M and Raizer Yu P 1997 Spark Discharge (Boca Raton: CRC Press)
p 294
191 Schonland B 1956 The Lightning Discharge. Handbuch der Physik 22 (Berlin:
Springer) p 576

Chapter 2
The streamer-leader process in

a long spark
This chapter will deal with the spark discharge in a long air gap. We have
already mentioned in chapter 1 that this material should not be ignored by
the reader. But for the long spark, specialists would know much less about
lightning. Today, high voltage laboratories are able to produce and study
long sparks of several tens and even hundreds of metres long [l-31. Many
of the long spark parameters and properties lie close to the lower boundary
of respective lightning values. Most effects observable in a lightning
discharge were, sooner or later, reproduced in the laboratory. One exception
is a multicomponent discharge, but the obstacles lie in the technology rather
than in the nature of the phenomenon. It would be very costly to instal and
synchronize several high voltage power generators, making them discharge
consecutively into the same air gap.
As for the fine structure of gas-discharge elements, long spark research-
ers are far ahead of lightning observers. This could not be otherwise, since a
laboratory discharge can be reproduced as often as necessary, by starting the
generator at the right moment, within a microsecond fraction accuracy, and
strictly timing the switching of all fast response detectors. But with lightning,
the situation is different. It strikes every square kilometre of the earth’s sur-
face in Europe approximately 2 to 4 times a year. So, even such a high con-
struction as the Ostankino Television Tower (540m) is struck by lightning
only 25-30 times a year. Of these, only 2-3 discharges are descending,
while the others go up to a cloud. Normally, lightning observations have
to be made from afar, so that many details of the process are lost. The
gaps in the study of its fine structure must, of necessity, be filled in laboratory
conditions.
The long spark theory is far from being completed, and there is no
adequate computer model of the process. Still, there has lately been some
progress, primarily owing to laboratory investigations. It would be unwise
to discard these data and not to try to use them for the description of
21

28 The streamer-leader process in a long spark
lightning. In this chapter, we shall outline our conception of the basic

phenomena in a long spark. We shall present some newer data and ideas
which emerged after the book [4] on the long spark had been published.
We should like to emphasize again that many details of the spark physics
are still far from being clear.
2.1 What a lightning researcher should know about a

long spark
The key point is how a spark channel develops in a weak electric field, by 1-2
orders of magnitude lower than what is necessary to increase the electron
density in air (Ei x 30 kV/cm under normal conditions). Naturally, we
speak of a discharge in a sharply non-uniform field. Near an electrode
with a small curvature radius (suppose this is a spherical anode of radius
r, x 1-lOcm), the field is Ea(ra) Ea > E, at the voltage U x 50-500kV.
This is the site of initiation of a discharge channel. At a distance r = lor,
from the electrode centre, the channel tip enters the outer gap region,
where the initial value of E = Ea(r,/r)* is one hundredth of that on the
electrode. This weak field is incapable of supporting ionization. Nevertheless,
the channel moves on, changing the neutral gas to a well-ionized plasma.
There is no other reasonable explanation of this fact except for a local
enhancement of the electric field at the tip of the developing channel. The
enhancement is due to the action of the channel’s own charge. Indeed, a con-
ductive channel having a contact with the anode tends to be charged as much
as its potential U , relative to the grounded cathode. Current arises in the
channel, which transports the positive electric charge from the anode (more
exactly, from the high voltage source, to which the anode is connected). (In
reality, electrons moving through the channel toward the anode expose low
mobility positive ions.) Such would be exactly the mechanism of charging a
metallic rod if it could be pulled out of the anode like a telescopic antenna.
Then the strongest field region would move through the gap together with
the rod tip. We can say that a strong electric field wave is propagating through
a gap, in which ionization occurs and produces a new portion of the plasma
channel. We can also name it as an ionization wave, and t h s term is commonly
used.
The wave mechanism of spark formation was suggested as far back as
the 1930s by L Loeb, J Meek, and H Raether. The channel thus formed
was termed a streamer (figure 2.1). Experiments showed that the streamer
velocity could be as high as 107m/s. In lightning, this velocity is demon-
strated by the dart leader of a subsequent component. Even the mere fact
that these velocities are comparable justifies our interest in the streamer
mechanism. It is important to know what determines the streamer velocity
and how it changes with the tip potential. For this, we have to analyse

What a lightning researcher should know about a long spark 29
grounded
cathode
Figure 2.1. A schematic cathode-directed streamer: U, (x), external field potential;

U ( x ) ,potential along the conductive streamer axis.
processes taking place in the streamer tip region where ionization occurs. It is
necessary to find out how the processes of charged particle production are
related to electron motion in the electric field, due to which the charged
region travels through the gap like the crest of a sea wave.
The specific nature of spark breakdown is not restricted to the ionization
wave front, because its crucial parameter is the channel tip potential U,. Its
value may be much smaller than the potential U , of the electrode, from which
the streamer has started, since the channel conductivity is always finite and
the voltage drops across it. Therefore, the analysis of streamer propagation
for a large distance will require a knowledge of the electron density behind
the wave front and the current along the channel in order to eventually
calculate the electric field in the travelling streamer and to derive from it
the voltage drop on the channel. Incidentally, the field and the current
preset the power losses in the channel. It will become clear below how
important this parameter is for spark theory.
The streamer creates a fairly dense plasma. Without this, it would be
unable to transport an appreciable charge into the gap. A quantitative
description of the ionization wave propagation provides the initial electron
density in the channel and defines its initial radius. Behind the wave front,
the streamer continues to live its own life. A streamer channel may
expand, through ionization, in the radial electric field of its intrinsic
charge, provided that the latter grows. The cross section of the current
flow then becomes larger. The channel continuously loses the majority cur-
rent carriers - electrons. The rates of electron attachment to electronegative
particles and electron-ion recombination strongly affect the fate of the
discharge as a whole. If the air through which a streamer propagates is
cold and the power input into the channel is unable to increase its tempera-
ture considerably (by several thousands of degrees), the process of electron
loss is very fast, since the attachment alone limits the electron average lifetime
to lop7 s. This is a very small value not only at the scale of lightning but

also of a laboratory spark, whose development in a long gap takes lop4-

lop3s. One must be able to analyse kinetic processes in the channel behind
the ionization wave front. Without the knowledge of their parameters, one
will be unable to define the conditions, in which a streamer breakdown in
air will be possible.
Here and below, we shall mean by a breakdown the bridging of a gap by
a channel which, like an electric arc, is described by a falling current-voltage
characteristic. The channel current is then limited mostly by the resistance of
the high voltage source. Such a situation in technology is usually called short
circuiting.
Current rise without an increase of the gap voltage inevitably suggests a
considerable heating of the gas in the channel. Due to thermal expansion, the
molecular density N decreases, thereby increasing the reduced electric field
E / N and the ionization rate constant (see [4]). Another consequence of the
heating is a change in the channel gas composition because of a partial
dissociation of 0 2 ,N2 and H 2 0 molecules and the formation of easily
ionizable NO molecules. The significance of many reactions of charged
particle production and loss changes. The importance of electron attachment
decreases, because negative ions produced in a hot gas rapidly disintegrate to
set free the captured electrons. The electron-ion recombination rate becomes
lower. But of greater importance is associative ionization involving 0 and N
atoms. The reaction is accelerated with temperature rise but it does not
depend directly on the electric field. This creates prerequisites for a falling
current-voltage characteristic.
Clearly, a researcher dealing with long sparks and lightning cannot
avoid considering the energy balance in the discharge channel, which deter-
mines the gas temperature. It is here that the final result is most likely to
depend on the scale of the phenomenon and the initial conditions. In the
laboratory, a streamer crossing a long gap seldom causes a breakdown
directly. A streamer propagating through cold air remains cold. It will be
shown below that the specific energy input into the gas is too small to heat
it. Even during its flight, the old, long-living portions of a streamer lose
most of their free electrons. In actual fact, it is not a plasma channel but
rather its nonconductive trace which crosses a gap. The researcher must pos-
sess special skills to be able to produce an actual streamer breakdown of a
cold air gap in laboratory conditions.
The situation with lightning may be different. Most lightnings are multi-
component structures. With the next voltage pulse, the ionization wave often
propagates through the still hot channel of the previous component. It is not
cold air but quite a different gas with a more favourable chemical composi-
tion and kinetic properties. Surrounded by cold air, the hot tract shows some
features of a discharge in a tube with a fixed radius and, hence, with a more
concentrated energy release. It seems that the mechanism of the phenomenon
known as a dart leader is directly related to streamer breakdown. One should

What a lightning researcher should know ahout a lorig spark 31
Figure 2.2. A photograph and a scheme of a positive leader.
be ready to give a quantitative description or make a computer simulation of

this process.
Long gaps of cold air are broken down by the leader mechanism. During
the leader process, a hot plasma channel (5000- 10 000 K) is travelling
through the gap. Numerous streamers start at high frequency from the
leader tip, as from a high voltage electrode, and form a kind of fan. They
fill up a volume of several cubic metres in front of the tip (figure 2.2).
This region is known as the streamer zone of a leader, or leader corona,
by analogy with a streamer corona that may arise from a high voltage elec-
trode in laboratory conditions. The total current of the streamers supplies
with energy the leader channel common to the streamers, heating it up.
The streamer zone is filled up with charges of streamers that are being
formed and those that have died. As the leader propagates, the zone travels
through the gap together with its tip. so that the leader channel enters a space
filled with a space charge, 'pulling' it over like a stocking. A charged leader
cover is thus formed which holds most of the charge (figure 2.2). It is this
charge that changes the electric field in the space around a developing
spark and lightning. It is neutralized on contact of the leader channel with
the earth, creating a powerful current impulse characteristic of the return
stroke of a spark. Thus, we can follow a chain of interrelated events,
which unites the simplest element of a spark (streamer) with the leader
process possessing a complex structure and behaviour.
All details of the leader development directly follow from the properties
of a streamer zone. In lightning, it is entirely inaccessible to observation
because of the relatively small size and low luminosity. Today, there is no
other way but to study long sparks in laboratory conditions and to extrapo-
late the results obtained to extremely long gaps. This primarily concerns a
stepwise negative leader, whose streamer zone has an exclusively complex

structure. It contains streamers of different polarities, starting not only from

the leader tip but also from the space in front of them.
The leader channel of a very long spark, let alone of lightning, is its long-
est element. An appreciable part of the voltage applied to the gap may drop
on this element. This is why one should know the time variation of the
channel conductivity. The channel properties mostly depend on the current
flowing through a given channel cross section. If the current is known, it is
not particularly important whether it belongs to a long spark or lightning.
The parameter that changes is the time during which one observes this
process: for lightning, it is one or two orders of magnitude longer than for
a spark. By analysing the self-consistent process of leader current production
in the streamer zone and its effects on plasma heating and conversion in the
channel, one can derive the conditions for an optimal leader development in
a gap of a given length. There are reasons to believe that these conditions are
realized in lightning when it develops in an extremely weak field. Nature
always strives for perfection, not because it is animated but because optimal
conditions most often lead to the highest probability of an event.
To conclude this section, long spark theory is of value in its own right to
specialists in lightning protection. Lightning current is the cause of the most
common type of overvoltage in electric circuits. The amplitude of lightning
overvoltages reaches the megavolt level. In order to design a lightning-
resistant circuit, one must be able to estimate breakdown voltages in air
gaps of various lengths and configurations. This can be done only with a
clear understanding of the long spark mechanism.
2.2 A long streamer

2.2.1 The streamer tip as an ionization wave
Let us consider a well developed ‘classical’ streamer, which has started from
a high voltage anode and is travelling towards a grounded cathode. The
main ionization process occurs in the strong field region near the streamer
tip. We shall focus on this region. The front portion of a streamer is
shown schematically in figure 2.3 together with a qualitative axial distribu-
tion of the longitudinal field E , electron density ne, and a difference between
the densities of positive ions and electrons, or the density of the space charge
p = e(n+ - ne) (the time is too short for negative ions to be formed).
The strong field near the tip is created mostly by its own charge. In front
of the tip where the space charge is small, the field decreases approximately as
E = E , ( T ~ / Y ) ’ ,which is characteristic of a charged sphere of radius .,Y Here,
E, is the maximum streamer field at the tip front point. In fact, the radius at
which the field is maximum should be termed the tip radius .,Y It approxi-
mately coincides with the initial radius of the cylindrical channel extending
behind the tip. The front portion of a conventionally hemispherical tip

A long streamer 33
Figure 2.3. Schematic representation of the front portion of a cathode-directed

streamer and qualitative distributions of the electron density ne, the density difference
n, - ne (space charge), and longitudinal field E along the axis.
should be called the ionization wave front. The streamer tip charge is primarily
concentrated in the region behind the wave front. The field there becomes low,
dropping to a value E, in the channel, small as compared with E,. The lines of
force going radially away from the tip in front of it become straight lines inside
the tip and align axially along the streamer channel.
Let us mentally subdivide the continuous process of streamer develop-
ment into stages. The strong field region in front of the tip is the site of
ionization of air molecules by electron impact. The initial seed electrons
necessary for this process are generated by the streamer in advance. Their
production is due to the emission of quanta, accompanying the ionization
process because of electronic excitation of molecules. In our case, highly
excited N2 molecules are active so that the quanta emitted by them ionize
the O2 molecules, whose ionization potential is lower than that of N2. The
radiation is actively absorbed, but its intensity is high enough to provide
an initial electron density M~ of about 105-106~m-3at a distance of 0.1-
0.2cm from the tip. Each of these electrons gains energy from the strong
field, giving rise to an electron avalanche. Since the number of avalanches
developing simultaneously is very large, they fill up the space in front of
the streamer tip to form a new plasma region. Owing to the electron outflow
towards the channel body, the positive space charge of the plasma becomes
exposed. Simultaneously, electrons that have advanced from the ahead
region neutralize the positive charge of the ‘old’ tip which turns to a new
channel portion, thereby elongating the streamer.
The gas in the wave front region must be highly ionized for the electron-
ion separation to produce an appreciable charge capable of creating a strong
ionizing field in front of the newly formed tip. For this reason, the region of

concentrated tip charge is somewhat shifted towards the channel body

relative to the intensive ionization site (figure 2.3). Normally, the electric
field is pushed out of a good plasma conductor, and the space charge (if
the conductor is charged) quickly concentrates near its surface as a ‘surface’
charge. The plasma of a fast streamer (‘fast’ in the sense that will be specified
below) possesses a fairly high conductivity, and these properties apply to
such a streamer. Therefore, the region of strong field and space charge in
the tip looks like a thin layer, as is shown in figure 2.3.
If the streamer length is I >> r,, its velocity and the tip parameters change
little during the time the tip travels a distance of its several radii. T h s means
that, depending on the time t and the axial coordinate x, all parameters
are the functions of the type E ( x . t) = E ( x - V,t), and what is shown in
figure 2.3 moves to the right as a whole, without noticeable distortions. The
picture changes only as the streamer velocity changes relatively slowly. This
kind of process represents a wave, in t h s case a wave of strong field and
ionization. The external parameter determining the wave characteristics (its
velocity V,, maximum field E,, tip radius r,, electron density behind the
wave n,) is the tip potential U,. It is indeed an external characteristic of the
tip, although it partly depends on the properties of the wave itself. The poten-
tial U , is equal to the anode potential U , minus the voltage drop on the
streamer channel. The channel properties, however, are initially determined
by the ionization wave parameters, so that the problem of streamer develop-
ment is, strictly speaking, just one problem. Still, it can be approximately
subdivided into two parts: the ionization wave problem and the problems of
voltage drop and current in the channel. Both parts will be related by the
dependencies of V,( U t ) and current il at the channel front on velocity V,.
2.2.2 Evaluation of streamer parameters

The formulas to be derived in this and subsequent sections of this chapter do
not claim high accuracy. The streamer and leader problems are very complex,
and a rigorous solution can be obtained only by numerical computation. But
a simplified analytical treatment may also be useful because it provides an
understanding of basic laws and relations among the process parameters.
In other words, one is able to get a general idea of the physics of the phenom-
enon under study and to estimate the order of values of its characteristics.
Let us consider a fast streamer, whose velocity is much higher than the
electron drift velocity in the wave. Streamers are fast in many situations of
practical interest. The calculation of electron production can ignore the
slight drift of electrons from a given site in space for the short time the
ionization wave passes through it. In this case, the ionization kinetics
along the streamer axis is described by the following simple equations:
3 = yn,.
at
-
nC
a0
= exp
s vidt =exp

A long streamer 35
Figure 2.4. Ionization frequency of air molecules by electron impact under normal
conditions (from the data on ionization coefficient cy and electron drift velocity V,
in [l 11).
where vi = v i ( E )is the frequency of electron ionization of molecules. Its time

integral has been transformed to the integral over the coordinate x along
the wave axis, according to the equality dx = V , dt corresponding to the
coordinate system moving together with the wave. Due to the sharp increase
of the ionization frequency with field (figure 2.4), the region where the field is
not much less than its maximum makes the largest contribution to the
electron production. This region in the wave is of the same order of
magnitude as the tip radius (figure 2.3). So we can write the approximate
expressions for the integral (2.1) and streamer velocity:
This type of formula was first suggested by Loeb [5] and has been used since
that time, in this or modified form, in all streamer theories [6-lo]. The
velocity of a fast streamer is weakly related to the initial no and final n,
electron densities and is determined only by the maximum field E, and the
tip radius r,.
The quantities E, and r , which determine V , are not independent. They
are interrelated by the tip potential U,. For an isolated conductive sphere with a
uniformly distributed surface charge Q', we have U = r,E, = Q ' / ~ T T T E ~ Y , ,
where E~ = ( 3 6 x~ lo")-' FZ 8.85 x 10-12F/m is vacuum permittivity. A
streamer looks more like a cylinder with a hemispherical rounded end (see
figure 2.3). We can show [4] that in a long perfect conductor of this shape,
one half of the potential at the hemisphere centre is created by charges
concentrated on the hemisphere surface and the other half by those on the
cylinder surface, so that the tip charge is Q = ~ T T T E ~The Y , Ufield
~ . at the tip
front point is, to good approximation, only one half of that in an isolated

sphere with the same potential, or

0; = 2E,r,.
The tip charge moves because of the electron drift under the field action.
The electron density in the wave plasma and the respective plasma conduc-
tivity must provide the charge transport with the same velocity as that of
the wave. This permits estimation of the plasma density in the streamer
just behind its tip. With the same assumptions as those in (2.2),the electron
density in the strong field region on the streamer axis increases as
ne M no exp (vimt)for the time At M rm/Vs.During this period of time, the
electron density rises to its final value nc M no exp (vi,&) , and the electron
drift towards the channel with velocity V , = peE, (where pe is electron
mobility taken, for simplicity, to be constant) exposes the charge which
creates the field E, in the region of the new streamer tip.
The electron charge that flows through a unit cross section normal to the
axis in the wave front region over time A t is
At
q = ep,E,noSo exp(6,t) dt = PeEmnc
~
vim
It leaves behind it a positive charge of the same surface density q. It is this
charge that creates the field E,. We shall see soon that the effective thickness
of a positively charged layer is Ax << r,. In electrostatics, the field of such a
layer at the conductor surface is equal to E, M q / q ( t h s equality is
absolutely exact for the surface charge of a perfect conductor). By substituting
q from (2.4),we get an estimate for the density behind the ionization wave:
n, M Eovim/epCL,.
The plasma density n, is not related directly to the streamer velocity and is
essentially determined by the maximum field value which defines the ioniza-
tion frequency.
Let us make sure that the tip charge is indeed concentrated in the thin layer.
The effective time for the charge to be formed in the layer approximately is
(At - t ) exp(vi,r) dt exp(q,t) dt = vi;.

1
The space charge layer of thickness Ax moves to a new site at velocity

Ax/& which is equal to the streamer velocity V,, since the ionization
wave moves as a whole. Hence, using (2.2),we obtain
Ax = VsAt, M V , / q , M r,/ln(n,/no). (2.5)
The final plasma density is many orders of magnitude larger than the initial
density no, so that the logarithm in (2.5)is a large value. Therefore, we have
Ax << r,.
The formulas derived here claim for nothing more than an illustration of
functional relations among streamer parameters. Numerical factors allowing

A long streamer 37
the transition from an order of magnitude to a specific value have been deliber-
ately ignored. This is justified because we have simplified all initial conditions
and the derivation procedure in order to reveal the physics of the phenomenon
in question. The significance of a formula will increase if it is ‘equipped’ with
numerical factors, even approximate ones. Since we know the origin of these
factors, we can partly judge about the theory validity and meaningfully
compare the analytical results with computations and measurements.
A more rigorous treatment of a fast ionization wave, using one-
dimensional equations consistent with the streamer physical model [4,9],
yields the expressions
qmrm nm
v,= n, = nC
- = ln- (2.6)
(2k - 1) ln(n,/no) ’ kePe ’ nm no
where k is the power index from the approximate formula vi(E) N Ek and n, is
the electron density in the wave front at the point of maximum field (the
density is an order of magnitude smaller than the maximum achievable
density n,). In the field range characteristic of an air streamer, k = 2.5.
The issue of the streamer tip radius or maximum field represents the
most complicated and least convincing point in streamer theory. It is likely
that their values are established under the action of a self-regulation mechan-
ism related to the proportionality V , = q(E,) and to the rapidly increasing
(at first) and then slowly growing dependence of q on E (figure 2.4). If, at
constant tip potential, the tip radius turns out to be too small and the field
E, respectively too high, corresponding to the slow growth of vi(E), the
channel front end will not only move forward quickly but it will expand as
fast under the action of a strong transverse field. The value of r , will rise
while that of E,, according to (2.3), will fall.
Suppose, on the contrary, that the radius ,Y is too large and the field E,
is too low, corresponding to a rapid growth of vi(E).Any slight plasma pro-
trusion in the tip front will locally enhance the field. The ionization rate will
greatly increase there, and the protrusion will run forward as a channel of a
smaller radius. Some qualitative considerations of this kind were suggested in
an old work of Cravath and Loeb [12], but these authors discussed a lightning
channel obeying other mechanisms, because lightning develops via the leader
process. These ideas were used in [6,7] to formulate an approximate streamer
theory. A semi-qualitative criterion was suggested for choosing the maxi-
mum field feasible in the streamer tip. According to [6,7], E, corresponds
to the saturation point or bending in the function vi(E). This criterion was
refined in [13] by establishing a quantitative relation of E, to the slope of
the q ( E ) function and to the charge and normal field distributions over
the streamer tip surface. It was shown that the field E, at the tip front
point is established such that the normal field on its lateral surface
corresponds to the point of transition from the rapid to the slow growth of
the ? ( E ) function (figure 2.4).

The streamer’s choice of maximum field was manifested during a

numerical simulation of short fast streamers within the framework of a
complete two-dimensional formulation [ 14- 161. The mechanism of auto-
matic establishment of E, was demonstrated in [13] in the calculation of
long streamer development with arbitary initial conditions and considerably
simplified equations (see also [4 Suppl.]). No one has been able yet to
calculate a long streamer within a complete two-dimensional model. We
can conclude from these results that the field at the front point of an air
streamer propagating at atmospheric pressure and room temperature
seems to be E, M 150-170kV/cm. The tip radius varies with the tip poten-
tial, approximately satisfying (2.3).
We shall give a numerical example as an illustration. At E, = 170 k V / m
in air (vim M 1.1 x 1011 s-1 ,pe M 270 cm2/Vs) and r , = 0.1 cm (corresponding
to U, = 34 kV), the streamer velocity for no M lo6cmP3from formula (2.6) is
close to 1.7 x lo6 mjs and the electron density in the newly born channel
portion is n, M 9 x 1013~ m - Within
~ . 20% accuracy, these values coincide
with the results of integration of unreduced equations in the one-dimensional
model illustrated in figure 2.3 [lo]. They show not more than a 2- or 3-fold
disagreement with numerical simulations of streamers in a two-dimensional
model developed by different workers. This type of computation is extremely
complicated and not particularly advanced. So the simple formulas (2.6) are
useful since they can provide rough working estimations. They are also
applicable to a gas of lower density, in whch similarity laws are operative.
Since v, M N f ( E / N ) , where N is the number of molecules per 1 cm3, f is a
function of the type given in figure 2.4, and pe M N - ‘ , we have
E, - N, r, -u,/N. v, U,. n, N ~ . (2.7)

The streamer velocity is independent of N and n, of the tip potential U,.
The latter fact opens up a tempting but yet unused possibility to test
experimentally the theoretical concept of the maximum tip field E, being
constant. For this, it is sufficient to measure the electron density right
behind the tip of a fast streamer, in which U, = U, - EJ rapidly decreases
with the channel length 1 (Eavis the average channel field). The constant data
on n, will indicate the constant values of E,. On the other hand, the decrease
in electron density with decreasing streamer velocity could become a strong
argument to support the hypothesis of constant tip radius r,, which still has
advocates.
From (2.6) and (2.3), the velocity of a fast streamer is proportional to its
tip potential, because its radius is proportional to the potential at
E, = const. The velocity V , becomes lower than the electron drift velocity
V,(E,) RZ 4 x lo5mjs at U, M 5-8 kV. At lower voltages, the streamer
moves slower than the drift electrons, so that the formulas become invalid.
The analysis of equations presented in [lo] shows that the streamer velocity
drops with further decrease in U, at V , < Vem. The electron density at the

A long streamer 39
wave front decreases, all electrons are drawn out of the tip, and space charge
fills it up. However, the final plasma density behind the ionization wave does
not decrease. The tip radius becomes very small, and the streamer stops at
low U,. This result agrees with experiments: no streamers with a velocity
less than (1.5-2) x lo5mjs have ever been observed in air under normal
conditions.
2.2.3 Current and field in the channel behind the tip

As the streamer develops, its channel is under high potential which changes
from the anode potential U, at the starting point to a certain value U, at the
front end, close to the tip potential U, (the difference between U , and U, of
about E,Ax << U, is due to a small potential drop in the tip). The channel
is electrically charged, since the potential at any point x along it is higher
than the unperturbed potential of space U o ( x )created by electrode charges
in the absence of streamer. Current must flow through the channel to
supply charge to the new portions of the growing streamer. When setting our-
selves the task of estimating this current and the current through the external
circuit (this is the streamer current to be measured), we must first find the
channel charge, for it is the time variation of this charge that produces the
current. Suppose a streamer has started from an anode of small radius P , .
Let us examine the stage when the streamer length becomes 1 >> ra (I is
much larger than the channel radius r ) . We can then neglect the time
variation of the anode charge, because its capacitance is small, and take
the external current to be close to the current i, entering the channel through
its base at the anode. Besides, a streamer conductor can be regarded as being
solitary, and the unperturbed potential U, far from the anode can be ignored.
Assume first that the channel is a perfect conductor. From a well-known
electrostatics formula, the capacitance of a long solitary conductor is
~ ) . charge is Q = CU, because a perfect conductor is
C = 2 7 r ~ ~ l / l n ( l / Its
under only potential U . Introduce now the concept of capacitance per unit
length of the conductor, C1, which is frequently used in electro- and radio-
engineering to analyse long lines. The average capacitance per unit length
C 27rq - 5.56 xlo-”
c1=-=--
I ln(I/r) ln(I/r)
has a nearly constant value which only slightly varies with I and r. Calcu-
lations show that the local capacitance C1(x)practically coincides with the
average value from (2.8) along the whole length of a long conductor,
except for its portions lying close to its ends. But even at the ends, the
local capacitance is less than twice the average value. This, however, does
not concern capacitances of the free ends which are much larger (see below).
As an approximation justifiable by calculations, we shall use the capaci-
tance per unit length from (2.8) and apply it to a real streamer channel. If a

channel possesses a finite conductivity, then it must have a longitudinal

potential gradient and U = U ( x ) , when current flows through it. The
charge per unit channel length has the form
27r&O[U(X)- U()]
.(x) % C , [ U ( x )- Uo(x)]
=
Wllr)
which allows for the fact that the local charge of a channel raises its potential
relative to unperturbed potential. Uo(x). A similar refinement should be
introduced in formula (2.3), which generally looks like
U, - UO(l) = 2Emrm. (2.10)
as well as in (2.7). If the channel radius r varies along its length, a character-
istic value may be substituted into (2.9), because the capacitance varies with r
only logarithmically.
Now turn to channel current. When a channel elongates by dl, its new
portion acquires charge ~ [ d lindex
; E will denote parameters of the front
channel end, x = 1. This charge is supplied directly by local current ir over
time dt = dl/V,. Therefore, at any stage of streamer development, we have
(2.11)
The current at the tip is defined mainly by the tip potential and streamer
velocity. At the anode, the current is
/ 1
1,. = dQ
dt, Q= J’ ~ ( xdx) = s o C1 [ U ( x )- Uo(x)]dx
0
(2.12)
where, Q is the total channel charge. Strictly, Q should be supplemented by

the tip charge Q, = 2 7 r ~ ~ r ~-[ U U,([)],
, but it is relatively small in a long
streamer.
Currents i, and ii at the opposite ends of a streamer channel do not
generally coincide. Of course, their values may be very close or differ con-
siderably, depending on particular conditions. For example, if the electrode
voltage is raised during the streamer development, the potential and charge
distributed along the channel increase. Some of the anode current is used to
supply an additional charge to the old channel portions, so that only the
remaining current reaches its front end: i, > il. But if the electrode voltage
is decreased, the ‘excess’ charge of the old channel goes back to the supply
through the anode surface, so that the current decreases nearer to the
anode (positive current is created by charges moving away from the
anode): i, < if.
A long streamer can develop at constant voltage when the electric field in
the channel, E ( x . t ) , does not vary much with time. The potential at any point
Jt
of the existing channel U ( x ) = U, - E ( x )dx and ~ ( xvary ) slightly with
time, which means that current does not branch off on the way from the

A long streamer 41
anode to the channel tip. In this case, the anode current is close to the end
current defined by (2.11) including potential U, which may be much lower
than U,. Many experiments have shown that the average channel field
must exceed a certain minimum value of about 5 kV/cm for air in normal
conditions (see sections 2.2.6 and 2.2.7) to be able to support a long positive
streamer. For instance, if U , = 600 kV at the anode and the streamer length
is 1 1 m, nearly all voltage drops in the channel and U, << U,, but currents
i, and il still do not differ much.
To get a general idea about the orders of magnitudes, let us consider a
variant which seems quite realistic and is manifested by some calculations
(section 2.2.6). This is the variant with constant applied voltage and
slowly varying average field in the channel, when the current along it
changes little and, therefore, can be evaluated from (2.1 1). For example, at
I = 1 m, r = 0.1 cm, V , = 1.7 x lo6 m/s, and Ul = U, 34 kV, as in the
illustration in section 2.2.2 (with U, > 500 kV), we have ln(l/r) = 6.9,
C1 = 8 x 10-12F/m, r1= 2.7 x lOP7C/m, and i, = i, = 0.46A. Streamer
currents of this order of magnitude (as well as much higher or much lower
currents) have been registered in many experiments. These values can also
be obtained from calculations with the account of possible streamer velocities
from lo5 to lo7 m/s in air, which have been found in some experiments to be
even higher [4].
In a simple model of potential and current evolution in a developing
streamer channel; the latter can be represented as a line with distributed
parameters: the capacitance C1 and resistivity R1 = (xr2epene)-' per unit
length. The electron density n,(x. t ) should be calculated in terms of the
plasma decay kinetics (see section 2.2.5). Estimations show that self-induction
effects are not essential in streamer development [4]. Then, the process is
described by the following equations for current and voltage balance:
d r di dU .
-+--0, -- = zR1, 7 = C1(U - Uo). (2.13)
at dx dX
A boundary condition in the set of equations (2.13) at x = 1 is the equality
4 = c1[U1 - UO(4l Vs (2.14)
equivalent to (2.1 1). Formula (2.12) automatically follows from (2.13) and
(2.14). Another boundary condition may be the setting of anode potential,
since U ( 0 ,t ) = U a ( t ) .Equations (2.13) and (2.14) will be used in the next
section to evaluate the heating of a streamer channel. Illustrations of
streamer development calculations will be given in sections 2.2.6 and 2.2.7
after a discussion of the plasma decay mechanism. A complete set of
equations for a long line, generalized by taking self-induction into account,
will be applied in section 4.4 to the treatment of a lightning return stroke.
Equality (2.11) allows evaluation of longitudinal field E, in the channel
behind the streamer tip, where the electron density is still as high as that

created by an ionization wave. The current behind the tip is conduction

current il = 7rrfnencpeEc.By equating this expression to (2.11) and using
(2.6) and (2.10) with U , = U,, we find
For a 1 m streamer, the product of logarithms in the denominator of (2.15) is

close to 100. Therefore, the field in the front end of a streamer channel in
normal density air is E, M 4.2 kV/cm (E, = 170 kV/cm from section 2.2.2).
Within the theory accuracy, this value does not contradict the average
measured channel field of 5kV;cm necessary to support the streamer.
There is no ionization in such a weak field, therefore electrons are lost in
attachment and electron-ion recombination processes.
Current il near the channel end is lower than that of the tip adjacent to
the channel, because the charge per unit tip length Tt = Q J r , is larger than T
in the channel. This is a typical consequence of end effects for long conduc-
tors, well-known from electrostatics. The surface charge density at the free
end of a conductor is much higher than on its lateral surface. In our
simple model, in which a channel tip has been replaced by a hemisphere of
radius ,Y and charge Q, written after formula (2.12), the average charge
per unit length is T, M 2 7 r ~ ~ -
[ UU~O ( l ) ]It. is In (l/rm) times larger than T,
at the channel end (see (2.9)). The tip current i, much exceeds il. This does
not affect the total charge balance, because the charge Q of a long channel
is much larger than the tip charge Q,.
Note that current perturbation in the tip region has a local character. It
cannot be detected by current registration from the anode side. The streamer
here makes use of its own resources - the charge of the ‘old’ tip has moved on
into the gap with the elongating streamer. It is the charge overflow that
creates current it. If a current detector were placed at the site of a newly
born portion of the channel, it would register current i M it for a very
short period of time At = rm/Vs M lop9s; then the current would decrease
to il and evolve as the solution of equations (2.13) and (2.14) indicates.
2.2.4 Gas heating in a streamer channel

A streamer process is accompanied by current flow and, hence, by Joule heat
release. As was mentioned above, the viability of a plasma channel depends
primarily on temperature, so this issue is of principal importance. The initial
heating of a given gas volume occurs when a streamer tip with its high current
and field passes through it. As the channel develops, the gas is heated further
by streamer current flowing through it. Let us evaluate both components of
released energy.
The energy released in 1 cm3 per second is j E = aE2,where j = aE is the
current density and a is the plasma conductivity in a given site in a given

A long streamer 43
moment of time. The energy released in 1 cm3 as a result of ionization wave
s2 s
passage is
W = a E dt = aE2dx/Vs (2.16)
where the integrals are formally taken from --x to +xbut actually over the
ionization wave region. The principal contribution to energy release is made
by a thin layer behind the wave front where the electron density and field are
high. The integral of (2.16) was found rigorously to be ~ ~ E i / using 2,
equations for this wave region [4]. This value has the physical meaning of
electrical energy density at maximum field. The contribution of the region
before the wave front is In (nm/no)times, or an order of magnitude, smaller
than this value. Although the field there is as high as that behind the front,
the electron density is of the order of n, and the conductivity 0 is
In (nm/no)times smaller (section 2.2.2). Therefore,
W x e o E k / 2 x 2.6 x J/cm3 (2.17)
where the numerical value corresponds to E, = 170 kV.
The fact that the density of energy release in a gas is of the same order of
magnitude as the energy density of the electric field is quite consistent with
electricity theory. When a capacitor with capacitance C is charged through
resistance R to voltage U of a constant voltage supply, half of the work
QU = CU2 done by the supply is stored by the capacitor as electrical
energy, and the other half is dissipated due to resistance, irrespective of its
value. The value of R determines only the characteristic time of the capacitor
charging, RC. Something like this is valid for the case in question but, of
course, without both energies being rigorously equal to each other, because
this situation is much more complicated. Indeed, according to the results
of section 2.2.2, the tip capacitance is C, = Q / U , % 27re0rm, volume
V, x 4rrLI3, and field E , E Ut/rm, so that the energy dissipation per unit
tip volume is W FZ CtU:/2Vt z eOEk (we have ignored the unessential
term Uo(l)).
Joule heat is released directly in a current carrier gas, or an electron gas.
Then electrons give off their energy to molecules in collisions. An appreciable
portion of electron energy (even most of it in a certain range of E I N ) is used
for the excitation of slowly relaxing vibrations of nitrogen molecules. Some
energy is used for ionization and electron excitation of molecules, about
U’ = 100 eV per pair of charged particles produced, i.e., n , ~ ’= J/cm3 at
n, FZ 1014cmP3.But even without the account of these ‘losses’, the gas tempera-
ture rise in the wave front region appears to be negligible: AT < Wlcv = 3 K.
Here, cv = qkBN = 8.6 x loP4J/(cm3/K) is the heat capacity of cold air and
kB is the Boltzmann constant.
Let us see what subsequent gas heating can provide by the moment it
is somewhere in the middle of a long streamer channel. We multiply the

second equation of (2.13) by i and integrate over the whole channel length,
assuming, for simplicity, that constant voltage U, is applied to the anode.
After taking a by-part integral in the left side of the equation, we substitute
ailax from (2.13) and i(1) i, from (2.14), followed by simple transforma-
tions. As a result: we have
U,i, =
So
1
I
i RI dx + - -d x
dt
+ [y - Ci U I U O ( / )Vs
] (2.18)
which describes the power balance in the system; here, Uo(1) is unperturbed
potential of the external field at the streamer tip point x = 1. The input power
Uai, into a discharge gap is used for Joule heat release in the channel (the first
term on the right), for increasing the electric energy stored in its capacitance
(the second term), and for the creation of new capacitance due to the channel
elongation (the third term). Joule heat associated with the ionization wave is
not represented here. The field burst and the tip impulse current that make up
W calculated above are absent from equations (2.13) and (2.18). Having
integrated equality (2.18) over the period of time from the moment of
channel initiation to the moment t the channel has acquired length I , we
get the equation for the energy balance in the system at the moment t :
where charge Q is given by (2.12). The energy input into the channel, U,Q, is
used to create capacity (the last term on the right), partly stored in this
capacity (the integral) and partly dissipated (&Is). The braces ( ) t indicate
the time averaging of the process. In case of a long channel, much of the
applied voltage drops across its length, so the tip potential U , is small
most of the time, as compared with average channel potential U,, of about
U,. Then, the last term in (2.19) can be neglected.
If we compare the left side of (2.19) with the substituted expression for Q
from (2.12) and the integral in the right side of (2.19), we can conclude that
the difference between these values cannot be much smaller than their own
values but rather have the same order of magnitude. Therefore, the energy
KdlSdissipated in the channel is equal, in order of magnitude, to the gained
electrical energy, which is in agreement with a similar situation discussed
above.
The average energy dissipated per unit channel length is W,, x CI Uiv/2
and the average energy contributed per unit channel volume is
(2.20)
where rav is the average channel radius. With the formation of every new
portion of the channel, its radius was approximately proportional to the

A long streamer 45
tip potential owing to the fact that the maximum tip field remained approxi-
mately constant. So we have Uav/r,, M E,. Substituting this expression and
(2.8) into (2.20), we find
One can see that subsequent heating of the channel gas adds little to the initial
heating by an ionization wave passing through the particular channel site.
To conclude, gas heating due to streamer development is negligible if the
gap voltage remains constant. Higher voltage does not change the situation
because the energy dissipated in the channel grows in proportion with the
channel cross section and the air volume to be heated. Specific heating
remains unchanged, since it is determined by a more or less fixed volume
density of electric energy.
2.2.5 Electron-molecular reactions and plasma decay in cold air

Electron loss in cold air is due to attachment to oxygen molecules and
dissociative recombination. The main attachment mechanism in dry air at
moderate fields is a three-body process
0 2 +e +0 2 + 0; +0 2 , (2.21)
k,, = (4.7 - 0.257) x cm6/s, y = E / N x 10'6V.cm2
where kat is the rate constant at T = 300K. In higher fields, the dominant
process is dissociative attachment O2 + e -+ 0- + 0 with the rate constant
-9.42 - 12.717 at y < 9
log k, = (2.22)
-10.21 - 5.7,'~ at 7 > 9.
In not excessively high fields of E < 70 kV/cm at 1 atm, air is ionized at the
rate constant ki = q / N
logki = -8.31 - 12.7,'~ at y < 26. (2.23)
Since the rate of electron loss through attachment is proportional to electron
density y1, and that through recombination is proportional to y12, the latter is
unimportant at the beginning of ionization. The equality ki = k , valid at
y M 12 determines the minimum field mentioned above, which is necessary
to initiate the growth of electron density in unperturbed air; Ei E 30 kV at
p = 1 atm and room temperature.
Oxygen molecules possessing a lower ionization potential than N2 are
ionized in fields not much exceeding the ionization threshold. Electrons
recombine with 0; at the rate constant 0, usually termed a recombination
coefficient:
0; + e -+ 0 + 0, /? M 2.7 x 10-7(300/Te)'/2cm3/s (2.24)

where T, is electron temperature in Kelvin degrees. However, complex ions

are more effective with respect to electron-ion recombination. The most
important ions in dry air are 0; ions, while in atmosphere saturated by
water vapour, as in thunderstorm rain, H 3 0 + ( H 2 0 ) 3cluster ions are more
important. For these, the recombination coefficients
0; + e + O2 + 02. D = 1.4 x 10-6(300/T,)'/2 cm3/s (2.25)
H 3 0 + ( H 2 0 ) 3+ e ---f H + 4H20, B 6.5 x 10-6(300/T)'12 cm3/s

(2.26)
are an order of magnitude larger than for simple ions.
Complex 0; ions are formed from simple ions in the conversion
reaction
0; +0 2 +0 2 --f 0; + 02, k = 2.4 x 10-30(300/Te)'/2cm6/s. (2.27)

Chains of hydration reactions lead to the production of H30'(H20)3 ions. A
typical chain looks like this:
0; + H20 -+ O;(H20) + 0 2 > k = 1.5 x lop9 cm3/s

O i ( H 2 0 ) + H20 H30' + O H + 0 2 ?
+ k = 3.0 x lo-'' cm3/s
H30+ -tH20 + (M) H30+(H20) + (M),+ k = 3.1 x cm3/s
H30+(H20) + H20 -t (M) H30+(H20)2 + (M). k = 2.7 x lop9 cm3/s
+
H3O+(H20)2 -t H 2 0 i-(M) -+ H 3 0 + ( H 2 0 ) 3-t (M), k = 2.6 x cm3/s

(2.28)
(M is any molecule, k correspond t o p = 1 atm, T = 300 K); here, a hydrated
ion replaces a 0; ion.
Another, similar chain begins with the production of an H 2 0 + ion in
ionization of water molecules by electron impact. Then comes the conversion
reaction
H 2 0 ++ H 2 0 -+ +
H 3 0 + OH, k = 1.7 x lop9 cm3/s
producing an H30+ ion, followed by the reaction chain of the type (2.28).
The production of complex ions is accompanied by their decay. For an
0; ion, this is the reaction
0; +0 2 + 0; i-0 2 -t0 2 ,
k = 3.3 x 10p6(300/T)4exp(-504O/T) cm3/s. (2.29)
It is greatly accelerated by gas heating, but in cold air the reaction effect
is negligible. The same is true of other complex ions, including hydrated ions.

A long streamer 41
In a cold streamer channel, simple positive ions turn to complex

ions very quickly, for the time T,, x 10-8-10-7s. It is these ions that
determine the rate of electron-ion recombination in cold air, except for a
very short initial stage with t 6 r,,,,.
If the ionization rate is too low and if the detachment-decay of negative
ions is slow, as in a cold streamer channel, the plasma decay is described by
the equation
3
dt
= -vane - pn,2 (2.30)
where va is electron attachment frequency. Its solution at initial electron

density equal to the plasma density behind the ionization wave, n,, is
(2.31)
where the time is counted from the moment the streamer tip passes through a
particular point of space.
Accordingto(2.15), wehaveE x 4.2kV/cmandE/N x 1 . 7 10-16V/cm ~
for a streamer channel just behind the tip at p = 1 atm. The electron
attachment frequency from (2.21) is va x 1.2 x lo7 sC1 and the characteristic
attachment time is 7, = v;’ x 0.8 x Over this time, most simple 0;
ions in dry air turn to complex 0; ions. Electrons recombine with them
with the coefficient ,8 x 2.2 x lop7cm3/s corresponding to electron tempera-
ture Te x 1 eV = 1.16 x lo4K at the above value of E/N. The initial electron
density n, x 1014cmP3 is so high that the parameter @nc/vax 2 determining
the relative contributions of recombination and attachment is larger than
unity. This means that at an early decay stage with t < ra x lop7s, electrons
are lost primarily due to recombination, with attachment playing a lesser role.
Later, at t > 2ra x 2 x lO-’s, the electron density decreases exponentially, as
is inherent in attachment, but as if starting from a lower initial value
nl = nJ(1 + @ n c ~ ax) 0.3n,; ne x nl exp (-vat).
The plasma conductivity decreases by two orders of magnitude, as
compared with the initial value, over t x 3 x s. At the streamer velocity
V , x lo6 m/s, this occurs at a distance of 30 cm behind the tip. A micro-
second later, the conductivity drops by six orders of magnitude. The streamer
plasma in cold humid air decays still faster because of a several times higher
rate of recombination with hydrated ions and due to the appearance of an
additional attachment source involving water molecules. These estimations
indicate a low streamer viability. It is only very fast streamers supported
by megavolt voltages that are capable of elongating to 1 x 1 m in cold air
without losing much of their galvanic connection with the original electrode.
This is supported by experiments with a single streamer and a powerful
streamer corona [4].

Note that a streamer plasma has a longer lifetime in inert gases, where
attachment is absent and recombination is much slower. This makes it
possible to heat the plasma channel by flowing current for a longer time
after the streamer bridges the gap (the estimations of section 2.2.4 do not
extend to these conditions). Such a process sometimes leads to a streamer
(leader-free) gap breakdown [ 171. Still, the formulation of the streamer
breakdown problem is justified for hot air and is related to lightning (see
section 4.8 about dart leader).
2.2.6 Final streamer length

When a streamer starts from the smaller electrode (anode) of radius Y,, to
which high voltage U , >> Elyais applied, it propagates in a rapidly decreasing
external field. It is first accelerated but then slows down after it leaves the
region of length Y, where it senses a direct anode influence. If the voltage is
too low, the streamer may stop in the gap, without reaching the opposite
electrode (say, a grounded plane placed at a distance d). The higher is U,,
the longer is the distance the streamer can cover; at a sufficiently high voltage,
it bridges the gap. In order to estimate the sizes of the streamer zone and
leader cover in a long spark or lightning - a task important for their
theory - we need a criterion that would allow estimation of maximum
streamer length under different propagation conditions. No direct measure-
ments of this kind have been made for single long streamers in air, because
there is always a burst of numerous streamers. This, however, is quite
another matter (see below). So we shall use indirect experimental results
and invoke physical considerations, theory, and calculations.
It has been established experimentally that streamers comprising a
streamer burst are able to cross an interelectrode gap of length d only if the
relation E,, = U,/d exceeds a certain critical value E,, which varies with the
kind of gas and its state. Under normal conditions in air, t h s critical value
is E,, e 4.5-5kV in a wide range of d = 0.1-10m. The data spread does
not exceed the measurement error. Bazelyan and Goryunov [18] recommend
the value E,, = 4.65 kVjcm for positive streamer, averaged over various
measurements. Therefore, the voltage necessary for a streamer to bridge a
gap of length d is U,,, = E,,d or more. For example, a gap of 1 m length
requires about 500 kV (Ec,e 10 kV/cm for negative streamer in air).
At the moment of crossing a gap, all voltage U , is applied to the
streamer, so Ea" is also the average field in the streamer. If a gap is long
enough, E,, can be identified with the average channel field. Indeed, in criti-
cal conditions with E,, = Ecr,a streamer crosses a gap at its limit parameters.
It approaches the opposite electrode at its lowest velocity corresponding to
the minimum excess of the tip potential U, e U , over the external potential,
AUl = U , - Uo(d)= 5-8 kV, below which the streamer practically stops. In
the case of a grounded electrode, U o ( d )= 0. If a gap is so long (say, 1 m) that

A long streamer 49
A U , = U, << U,, nearly all applied voltage drops in the channel. Therefore,
E,, can be treated as the lowest field limit, at which a streamer is still capable
of propagating.?
This interpretation remains valid when a streamer does not bridge a gap
but stops somewhere on the way. Indeed, according to (2.6) and (2.10), the
streamer velocity, and hence its ability to move on, is determined only by
the tip potential excess over the external potential, A U , = U , - Uo(I),and
is independent of the latter. No matter where a long streamer stops, we
shall have A U , << U,, though the external potential value at this point,
Uo(lmax),may be high. Generally the average field limit in the channel and
the streamer length at the moment it stops, ,I are interrelated as
In order to be able to use these relations in practice, we must know not only
the easily registered gap voltage U , but also potential Uo(Imax)inaccessible to
measurement. In most cases, it is hard to estimate even by calculation. For a
particular streamer, the external field is determined, in addition to the anode
charge, by the whole combination of charges that have emerged in the gap
and its vicinity. Especially important is the charge of all other streamers
that were formed together with the one under study. Consequently, the
field U o ( x ) ,in which the streamer is moving, represents a self-consistent
field. An outburst of hundreds of streamer branches is characteristic of air;
they fill up a space comparable with l,,,. It is this maximum length, rather
than the small anode radius, that will determine the external field fall
along the gap length. Estimations of a self-consistent field U o ( x )involve
considerable difficulties and errors (we shall come back to this when
evaluating the size of the streamer zone of a leader). So in reality, critical
field E,, can be evaluated only from experimental data that relate to a
situation with streamers bridging a discharge gap. Then, the potential
Uo(I,,,) is known reliably because it coincides with the potential of the
electrode, usually the grounded one: Uo(lmax)= U o ( d )= 0. But if it is
known that Uo(lmax)<< U,, as in the case of a long streamer moving in a
sharply non-uniform field, the criterion of (2.32) for a definite streamer
length will be extremely simplified: I, x U,/Ecr.
The existence of critical field E,, has a rather clear physical meaning. The
reason for the appearance of a minimum average field in a channel is its finite
t The average quantities E,, and E,, describe, to some extent, the actual field strengths in the
channel even when the external field is extremely non-uniform, changing by several orders of
magnitude along the gap far from the conductor. For example, if we close the gap with such a
thin wire that short-circuiting current does not change the electrode voltage, a short time
later, after the current along the wire is equalized, the actual gap field will become constant
along its length and exactly equal to Eav.

resistance. A channel must conduct current necessary to support the motion

of the streamer tip. This is the current which supplies charge to a new portion
of the channel produced at its tip front. The nature of the streamer process is
such that the current il just behind the tip is proportional to its velocity (see
formulas (2.1 1) and (2.14)). Local field E, necessary to support this current is
defined by (2.15) derived from the channel resistivity per unit length for a still
dense plasma. The value of E, in (2.15) slightly depends on varying streamer
parameters, such as length, tip radius, and velocity, and is largely determined
by the gap gas composition, which predetermines maximum field E, at the
tip and the slope of the v,(E) curve (the latter was taken into account in
(2.15) by the k parameter). The calculated value of E, = 4.2kV/cm for air
appeared to be surprisingly close to the measured value E,, = 4.65 kV/cm.
One should not give too much importance to this coincidence of the
values, but the agreement in the order of magnitude is definitely not
accidental.
Because of the plasma decay and conductivity decrease, the current
support in other channel portions may require a stronger field than E,.
For this reason, decay processes appreciably affect the value of Ecr, as is
indicated by experiments. An important mechanism of electron loss in cold
air in a relatively low field E,, is the attachment in three-body collisions
(section 2.2.5). Here, the attachment frequency is v, M N 2 , so the conven-
tional similarity principle E N for field E,, is violated: the reduced field
N
E,,/N does not remain constant and the value of E,, decreases more rapidly
than density N [19]. When a streamer propagates through heated air, the
critical field becomes lower not only due to a lower density but as a result
of a direct temperature action. This was found from measurements at various
p and T , up to 900 K [ 19,201. The reason is clear: on gas heating, the action of
attachment and recombination becomes weaker (section 2.2.5). In electro-
positive gases, in which there is no attachment, the value of E,, is lower
than in cold air, other things being equal. For instance, in nonpurified
nitrogen with an oxygen admixture up to 2%, the field is E,, M 1.5kV/cm
at p = 1 atm. In inert gases where the attachment is absent and the recombi-
nation has a lower rate than in molecular gases, E,, is much lower, about
0.5 kV/cm [21,22].
The channel field does not vary much in time along its length because of
the compensation due to countereffects. On the one hand, the conductivity in
an old channel portion is lower than in a new one because of the plasma
decay. On the other, that old portion was produced by a faster ionization
wave at a higher tip potential corresponding to a larger channel cross section.
As a result, the resistivity per unit length R1 = ( ~ r ~ e p ~ ndoes , ) - ~not vary
much along the channel. Of course, it grows in time because of electron
loss, but at the same time, the streamer velocity decreases together with the
channel current. For this reason, the time variation of the channel field
E ( x ,t ) = R I (x,t ) i ( x .t ) is much slower than that of any of the cofactors.

A long streamer 51
We shall illustrate this by giving a particular analytical solution to the

set of equations (2.13), (2.14), (2.6), (2.10), and (2.31), which is not far
from the actual result (see below). Assume the channel field and capacitance
per unit length to be constant, with the current along the channel being the
same: E(x: t ) = const, Cl(x,t ) = const, and i(x, t ) = i(t). Neglect potential
Uo(x), inessential to a long streamer in a sharply non-uniform field, and
suppose that the plasma at point x decays exponentially starting from the
moment t, of its production (as is inherent in attachment without recombi-
nation). We shall have
U ( X )= U , - E X , Ui = U, - E l , Vs = A U , ,
i = C,V,U, = CIAU:
where the nearly constant coefficient A is, according to (2.6) and (2. lo), equal
to
vim
A=
2(2k - l)EmIn (n,/no)
= const. (2.33)
The integration of dl = V, dt yields I x lmax[l - exp ( - A E t ) ] , I, = Ua/E,

t, = t(1) at x = 1. But the requirement R1(x, t ) = R I ( t ) involved in the initial
assumptions can be met only for one value of the channel field:
E = va/2A = (AUt)min/2Vem~a x 1.2kV/cm. Here, V,, 4 x 105m/s is
electron drift velocity at maximum tip field E,, (Aut),, x 8 kV and
r, = vi' x 0.85 x lop7s is the attachment time. The relation for E can be
interpreted as follows. The potential difference (A U,),,, necessary to
provide the minimum streamer velocity Vsminx V,, must be gained in
field E along the plasma decay length V,,T,. A similar treatment of the
streamer process will be offered in the next section when discussing the
streamer motion in a uniform field. The order of magnitude of the 'critical'
field E is correct. Therefore, the assumptions underlying the particular
solution are not meaningless, so the solution illustrates the main idea.
The existence of a critical field has been confirmed quantitatively by
numerical models of the streamer process. Let us discuss the calculations
obtained from a simple, evident model. We mean the above set of equations
(2.13) and (2.14) supplemented by expressions (2.6), (2. lo), and (2.33), which
define the streamer velocity and local channel radius, together with (2.3 1) for
the plasma decay, in which the time is counted off from the moment t, of its
production at point x. The streamer development in air from a spherical
anode of radius ra = 5 cm at U, = 500 kV is demonstrated in figure 2.5.1
The calculations were made with 7, = 0.85 x s and recombination
coefficient /3 = 2 x cm3/s. The general tendency in the behaviour of
principal parameters is quite consistent with the qualitative picture above.
t The numerical simulation was made in cooperation with M N Shneider. This type of equation,
but with a constant channel radius, was solved in [23].

5M)
400 2 - 23.37 ns
3- 53.5ns
4- IOOns
2 300 5- 310ns
-
.-m
s
L
2 200
v
'E
0
100
0
20 40 60 80 1
x , cm x , cm
U -t 4
0.1
0 20 40 60 80 1 0
x , cm
Figure 2.5. Streamer propagation in air from a spherical anode of 5cm radius at
500 kV. The distributions of potential U , current I , field E , and electron density ne
along the channel at various moments of time until the streamer stops.
Note that the nonmonotonic character of some current distributions when

the potential at a given point x grows with time is associated with a slight
time decrease in capacitance (2.8) of the elongating channel. The streamer
acquires its maximum velocity lo7 mjs very soon, over s; then it steadily
decelerates and stops at I, = 0.94 m.
It was found from (2.32), with the account of U,,(lmax)and (AUt),in,
that the actual average field in the channel at the moment of zero velocity

A long streamer 53
was E,, = 4.9 kV/cm. From a simplified criterion, it was found to be

E,, = U,/I,,, = 5.3 kV/cm. The agreement with the experimental value of
4.65 kV/cm is quite satisfactory. Calculation with U , = 250 kV yielded
I, = 0.39m and at U , = 750 kV it was 1.42m. Equation (2.32) is satisfied
at the same E,, = 4.9 kV/cm, i.e., the constant value of the average channel
field is confirmed by calculations of the final stage of a streamer process
for various streamer lengths. t When one takes into account only recombina-
tion, a streamer elongates much more, to 1.25 m; with no account of electron
losses, it elongates to 3 m (Ec, = 1.7kV/cm), as the qualitative picture
suggests.
2.2.7 Streamer in a uniform field and in the ‘absence’ of electrodes

So far, we have dealt with a streamer which starts from a high-voltage elec-
trode, to which it remains galvanically connected, and is supplied by current
from a voltage source through the electrode. Such are typical experimental
designs and, partly, conditions in the streamer zone of a positive leader, in
which the leader channel and its tip possessing a high positive potential act
as the electrode. However, a situation may arise when a streamer is initiated
in the body of a gas gap, where the external field is sufficiently high. This kind
of streamer develops without a galvanic connection with a high-voltage
source. Such streamers seem to be present in negative leaders. Note that
lightning propagating from a cloud down to the earth most often carries a
negative charge, while that going up from an object on the earth is positive.
In some situations, a streamer may take its origin from the electrode vicinity
and begin its travel being connected to it, but later it may break off because of
the plasma decay in an old channel portion. If the external field is still strong
at some gap space length, the streamer will move on, having ‘forgotten’ about
its former connection with the electrode. This behaviour is characteristic of
the streamer zone of a leader.
Consider a simple case when there is a uniform electric field Eo at some
distance from the electrode and a fairly long conductor of length 1 and
vector Eo along the x-axis. This may be a metallic rod in laboratory
conditions, or a plane or rocket going up to charged clouds, or a dense
plasma entity created in this way or other. The conductor is polarized by
the external field to form a charged dipole. The vectors of the dipole and
external fields are summed. The total field E,,, in the body of a perfect
conductor drops to zero, since an ideal conductor is always equipotential.
In the symmetry condition, all its points take the potential of the external
unperturbed field at the middle point of the conductor. Sometimes, the
field in the conductor body is said to be pushed out into the external
t The calculation using a simplified formula without the account of Uo(lmax)at l,, = 0.39m
gives an error: 5.7 instead of 4.9 kV cm.

Figure 2.6. The potential distribution along a conductive rod in a uniform electric
field. Broken line, the potential in the absence of a conductor.
space. The dipole charge enhances the field (E,,, > Eo) at the ends of the
polarized body (figure 2.6).
The problem of field redistribution by polarization charge can be solved
rigorously by numerical methods for any geometry, but simplified evalua-
tions are also possible. In the close vicinity of a charged dipole ‘tip’ of
radius Y << 1, the longitudinal field varies nearly in the same way as the
field of a sphere of identical radius. Therefore, the external field perturbation
by polarization charges is attenuated at a distance Y from each of the two
conductor ends. Let us take the conductor middle point to be the coordinate
origin. The end potentials of a polarized conductor differ from that of
an unperturbed one, U. = -Eox at the same points by AU x Eol/2. The
absolute strengths of the total field at the conductor ends rise to
E, x A U / r x Eol/2r, and the field increases with increasing l/r. This
estimate fits fairly well the numerical evaluation in [24].
At 1 >> r, ionization processes and streamers may arise at the ends of a
polarized conductor even at a relatively low external field Eo (figure 2.7(a)).
Ionization waves run in both directions, leaving plasma channels behind, in
much the same way as with a streamer starting from a high-voltage electrode.
If their velocity V , appreciably exceeds the electron drift velocity, the condi-
tions of streamer travel from the positive and negative ends will not differ
much. The total charge of developing streamers is zero at any moment of
time. This could not be otherwise, because none of them is connected with
the electrode and, through it, with the high-voltage source. Charges do not
escape the gap but are only redistributed by the streamer current. A streamer
Figure 2.7. Excitation of streamers of both signs from the ends of a conductor in a
uniform field. The charge distributions per unit channel length are shown schemati-
cally: ( a ) with active plasma, (b) with plasma decay in the older channel portions.

A long streamer 55
travelling along the vector Eo is charged positively, while a counterpropa-

gating streamer acquires a negative charge. Naturally, the current and the
charge separation occur due to electron drift. As in a streamer starting
from an electrode, the work done for plasma production and charge separa-
tion is done at the expense of the power source creating external field Eo. If it
is a conventional high-voltage source, current flowing through the streamer
during its propagation is due to the variation in the charge induced on the
electrode surfaces when the value and distribution of polarization charges
in the gap bulk change. This current takes away some of the source power
which is eventually used for the streamer development.
As streamers develop, the total length of a polarized conductor
increases, increasing the potential difference AU pushed out of the plasma
channels. On the other hand, the plasma in the old, central channel decays,
so that the charge overflow from one half of the conductor to the other
becomes more difficult. Finally, a streamer cannot elongate any more,
because the gain in length at the tip is lost due to the plasma decay at the
‘tail’. What one observes now is a pair of detached plasma sections of limited
length going away in both directions. At the front ends, they have a limited
potential difference AU with respect to the external potential. The process is
stabilized. It may probably go on until there is the external field.
We should like to emphasize that this issue is of principal importance. A
streamer needs the external field for charge redistribution in the created
plasma, i.e., behind the tip but not in front. The streamer creates its own
field that contributes to the gas ionization in the tip region, while the channel
field provided by an external source is necessary to support the current to the
tip, without which the streamer could not move on.
When the streamer plasma in the old channel near the starting point has
decayed completely, two galvanically disconnected streamers continue to
move in opposite directions. Now the polarization effects of each of the
conductive sections are added to the earlier polarization effects of the
whole channel. As a result, there are four charged regions with alternating
polarity (figure 2.7(b)). Nevertheless, there are still only two ionization
waves moving only forward, away from the channel centre, trying to elongate
the streamer. No return ionization waves arise in its central portion which
has lost conductivity because of a smooth charge distribution towards the
ends. The charge is ‘smeared’ along more or less extended ‘semiconducting’
regions and cannot create a sufficiently strong field to initiate ionization.
High-voltage engineers are familiar with this phenomenon. They can some-
times decrease an electric field, alternating in time, at the site of its local
rise between a sharp metallic edge and an insulator, by coating the dielectric
at their boundary with semiconducting material.
The parameters of fast ‘electrodeless’ streamers are described by the
same formulas (2.6), (2.10), and (2.14). The role of U, - U. here is played
by the quantity AU x Eo1/2, where 1 is the length of a channel section with

preserved conductivity. Here, A U also represents the excess of the streamer tip
potential over the external one. In particular, as the length I increases and the
field at the conductor ends becomes as high as E, M 150-170kV/cm for
normal density air (section 2.2.2), the growth of E, ceases. As I and A U
increase further, the streamer tip radius Y,, not E,, increases because
A U z E,Y,. This is accounted for by the self-regulation mechanism discussed
in section 2.2.2.
For the understanding of the streamer process in the streamer zone of
a leader, where the field is nearly uniform and the leader channel acts as a
high voltage ‘electrode’, it is essential that the ‘electrode’ and ‘electrodeless’
situations should be strictly equivalent, provided that the positive and the
negative streamers are identical and the external field is uniform. Let us
mentally cut, at the centre, a plasma conductor developing in both directions
from this centre and discard, say, the negatively charged half. Let us now
replace it by a plane anode under zero potential and assume a negative
potential to be applied to a remote plane cathode. A cathode-directed
streamer produced at the anode by a local inhomogeneity, whose field
initially supported ionization, will be identical to a positive streamer in the
electrodeless case. Indeed, in both cases, the conductor potential U coincides
with the external potential, U ( 0 ) = Uo(0). The charge pumped from the
negative half into the positive one will now be supplied by the source current
from the anode. Here, the principle of mirror reflection in a perfectly
conducting plane, well-known from electrodynamics, reveals itself in every
detail. According to this principle, the distributions of charge, current and
field in half-space do not change if the plane is replaced by the mirror
reflection of half-space charges.
These considerations were used in the calculations and representation of
results on streamer development in a uniform field U, = -Eox from the
point x = 0 towards lower potential (figures 2.8 and 2.9).t The solution
was derived from the same set of equations (2.13), (2.14), (2.6), (2.10),
(2.31) and the same plasma decay characteristics as in section 2.2.6. The
calculations show that a streamer does not develop if the external field is
lower than a certain minimum value. The values of Eommdo not differ
much from the critical channel field E,, which determines the streamer
length in a non-uniform field calculated with (2.32): Eo M 7.7 kV/cm. One
may probably use for estimations the experimental value E,, x 5 kV/cm as a
realistic Eo (section 2.4.1). If the uniform external field slightly exceeds
the minimum, the excess tip potential is small, the streamer velocity is low,
and the channel field is close to the unperturbed external field Eomn.This
situation is illustrated in figure 2.8.
If Eo is appreciably higher than Eo however, the tip potential is much
higher than the external potential, and the streamer develops a high velocity
t Numerical simulation was made in cooperation with M N Shneider.

A long streamer 57
-3001 ' ' . ' . ' . ' ' " 1

0 5 IO 15 20 25 30 35 40
x, cm x, cm
' 2
0 0 5 10 15 20 25 30 35 0
x, cm x, cm
Figure 2.8. An air streamer in a uniform field Eo =7.7 kV/cm, slightly exceeding the
critical minimum, with calculated distributions of potential U , current I , field E and
electron density ne. Dashed line, applied field potential counted from the streamer
origin. The oppositely charged streamer running in the opposite direction is not
shown.
(figure 2.9). The current in well-conducting portions a t the tip is high but
decreases towards the channel centre. Owing to the high current, much
positive charge is pumped into the tip region even at an early stage; in the
tail, however, where the conductivity has decreased, the current is low. As
a result, positive charge pumped out of it is not reconstructed; moreover,

--_ ---I_. h
5 .4
-800 3 - 482.1 ns \
4- 576.6ns
5 - 653.1ns
-1000
0 20 40 60 80 100
x, cm x, cm
2.5 7
x, cm x, cm
Figure 2.9. An air streamer in a uniform field Eo =10 kV/cm with the charge distribu-
tion 7 instead of the ne curves similar to those in Figure 2.8.
the tail becomes negatively charged. The calculated charge distributions in

figure 2.9 were found to be exactly as those represented schematically in
the right-hand side of figure 2.7(b) in terms of double polarization of the
whole channel and each of its conducting sections individually. At
Eo zz Eomin and low current, the polarization effect of the conducting section
is very weak (figure 2.8).

The principles of a leader process 59
2.3 The principles of a leader process
This section is a key one for the understanding of a long spark and the first
lightning component. We shall try to answer the question why a simple struc-
tureless plasma channel has no chance to acquire a considerable length in cold
air of atmospheric pressure. The reader will see what is necessary for a spark to
become long and have a long lifetime and how Nature realizes this possibility.
2.3.1 The necessity of gas heating

Section 2.2 dealt with the development of a simple plasma channel - a
streamer - which has no additional structural details. The theoretical con-
siderations concerning the streamer process are, in general, supported by
experiment, indicating that the streamer gas is cold and the channel field is
too low for ionization to occur. In these conditions, the plasma produced in
the tip by an ionization wave decays later. Electrons are lost due to recombi-
nation (it exists in any gas) and attachment inherent in air as an electronegative
gas. Losing its conductivity and, hence, the possibility to use current from an
external source, a streamer eventually stops its development, unless it encoun-
ters a strong field on its way (section 2.2.7). Sometimes, the streamer lifetime
can be made longer by a steady voltage rise, but this possibility is, naturally,
limited. Even at a megavolt voltage of a laboratory generator, an air streamer
can become only several metres long. Voltages of a few dozens of megavolts
inducing lightning discharges are, at best, capable of increasing the streamer
length to several tens of metres but not to the kilometre scale characteristic
of lightning. At high altitudes, however, the air density is low and a streamer
may cover a longer distance. This probably accounts for vertical red sprites
above powerful storm clouds dozens of kilometres above the earth’s surface
[25], which were found to travel downwards.
The only way of preventing or, at least, slowing down air plasma decay
in a low electric field is by increasing the gas temperature in the channel to
several thousands of Kelvin degrees and, eventually, to 5000-6000 K or
more. In a hot gas, electron loss through attachment is compensated by
accelerated detachment reactions, and recombination slows down. The
mechanism of associative ionization comes into action, and electron
impact ionization is enhanced because the gas density decreases on heating.
These processes make it possible for a plasma channel to support itself, or, at
least, to approach this condition, in a relatively low field. A hot spark looks
like a hot arc or a glow discharge column after contraction [26]. We shall not
discuss here the details of these processes (for this, see section 2.5). It suffices
to take for granted the statement that gas heating does maintain plasma
conductivity, making the spark viable.
It follows from section 2.2 that an increase of potential U at the streamer
front does not contribute to gas heating. Total energy release grows as U 2 ,

but the streamer cross section 7rrk also increases as U 2 .So the released energy
density proportional to ( U / r m ) 2 , which defines the heating, remains low. To
increase the channel temperature considerably, it is necessary to accumulate
a much higher energy in a much narrower plasma column. For this, the func-
tional relation providing the low U / r m ratio must be violated. This is
impossible in a primary ionization wave but becomes possible in a differently
organized channel development. Let us try to approach this problem by
considering the final result and estimate voltages and plasma channel radii,
at which the gas temperature would become sufficiently high. This can be
done in terms of the general energy considerations discussed in section 2.2.4.
Suppose there is a charged space with characteristic size R in the front
region of a developing plasma channel. Its capacitance is C x m o R with
C1 x x0per unit length along the channel axis. If the tip potential is U ,
the energy dissipated per unit length of a new portion of the system including
the channel and the space being charged is C1U 2 / 2 ,provided the spark devel-
ops steadily. Since the capacitance per unit length of a new system portion is
independent of its radius R or of any other geometrical dimension, we are
free to assume any nature, size, and volume of this charged space. (The
specific capacitance of the central portion of a long system does vary with
total length and radius but only logarithmically, as is clear from formula
(2.8)) The dissipated energy includes all expenditures for the creation of a
new channel portion and space charge. Attribution of this energy to the
various expenditures is a special problem which requires details of the
process to be specified. But we can estimate the upper limit of air mass
that can be heated to the necessary temperature, say, to T = 5000K. For
this, let us assume that all energy has been used heating an air column of
initial radius ro. This will be an estimate of the upper radius limit. This
temperature will lead to considerable thermal gas expansion, because a hot
channel, as will be shown later, develops much more slowly than a cold
streamer channel. Current must have enough time to heat the gas, because
it is eventually the released Joule heat of current that does the heating. If
the heating rate is not high enough, pressure in the gas space is equalized,
so that the gas of a thin channel becomes less dense. The air heat capacity
does not remain constant within a wide temperature range, so energy calcu-
lation should be made in terms of specific enthalpy h( T , p ) .Therefore, the
expression to a maximum radius rOmaxof a cold air column that can be
heated to temperature T is
7rrim,,p0h(T) x T & ~ U ~ / ~ . (2.34)
Here, h( T ) is specific enthalpy for air at p = 1 atm and po is its density at
p = 1 atm and To = 300K.
With tip potential U, = 1 MV and T = 5000 K, when h ( 5 ) = 12 kJ/g, an
air column that can be heated must have an initial radius less than
yomax = 0.054cm. The maximum radius due to thermal expansion will be

less than rmax = ro [ p o / p (5 ) ]' I 2 M 0.26 cm, where p( 5) is the air density at
T = 5000 K and pressure 1 atm. A channel of this thickness has been observed
in laboratory spark leaders. At U, M 100 MV, characteristic of very powerful
lightning, the radius estimated from formula (2.34) must be two orders of
-
magnitude larger. A lightning leader, however, has a temperature higher
than 5000K and h T 2 approximately (h(10) = 48kJ/g), so that the
radius does not grow as much as U, and remains as small as several centi-
metres. It may seem surprising, but a leader channel is thinner than a streamer
channel at the same tip potential (their radii are established due to different
reasons: a leader radius follows the heating conditions, while a streamer
radius is such that the lateral field is too low for intensive ionization).
2.3.2 The necessity of a streamer accompaniment

The existence of a long spark and lightning are due to two main mechanisms,
even in the presence of a very high voltage source. One is the mechanism of
current contraction in a thin channel which can practically be heated. The
other is the attenuation of a very strong radial field that arises at the lateral sur-
face of a very thin conducting channel under a very high potential relative to the
earth. We shall begin with the second mechanism, because it opens the way
for the first one. In reality, the tremendous value of U / r M 10-100 MV/cm is
not the field scale near the channel tip of radius r . Nor does the value of
U/[rln (Ilr)] l-lOMV/cm, which is somewhat less, determine field E, at
the lateral surface of a channel of length I behind the tip, as could be
suggested from formula (2.9) and the Gaussian theorem, E, = 7/(27rq,r).
This would be valid only for such a simple structureless channel as a
streamer, but its lateral field cannot maintain a high strength for a long
time. Lateral ionization expansion would immediately increase the channel
radius. On the other hand, a channel cannot be heated to the necessary
high temperature unless its radius is small. This is the reason why a single
simple channel cannot be heated.
A long-living spark requiring megavolt voltages will inevitably have a
complex structure. The reader has, no doubt, guessed that we mean the
streamer zone in front of a leader tip and its production - the leader cover
representing a thick charged envelope around the channel (figure 2.2). The
space charge of a streamer zone and leader cover, having the same sign as
that of the channel potential, greatly reduces the field at the channel surface.
Roughly, owing to the field redistribution by space charge, the huge potential
U now drops across a much longer length R of the streamer zone and the
charge cover radius, rather than across a length nearly as short as the channel
radius r . In this case, the field scale is a moderate magnitude U / R but not
U / r , because even a laboratory spark has R of about a metre long.
Indeed, the radius of a streamer zone and, hence, of a leader cover is
defined by the maximum distance streamers may cover when they travel

away from the leader tip. We already know (sections 2.2.6 and 2.2.7) that the
average field necessary for streamer development in atmospheric air must be at
least E,, x 5 kV/cm. Since streamers stop at the end of the streamer zone, the
voltage drop along the zone length R is about AU, x E,,R (cf. formula (2.32)).
About as high voltage drops outside the streamer zone, because the field there,
i.e., in a zero-charge space, drops from about E,, to zero, as for a solitary
sphere of radius R. Hence, we have U x 2AUs and R = U/2Ec,. At
U x 1 MV, the streamer zone radius is R x 1 m, in agreement with laboratory
measurements.
It follows from both calculations and measurements that the current,
field, electron density, and conductivity of a heated leader channel are
generally comparable with respective parameters of a fast streamer. If they
are somewhat larger, the difference is not orders of magnitude. So the heating
time to achieve a much higher gas temperature must be much longer. This
explains why a leader propagates much more slowly than a fast ionization
wave.
The capacitance per unit length of a leader system (the channel plus a
charged cover) will be described by the same formula (2.8) if I is substituted
by leader length L and the conducting channel radius Y by cover radius R , the
actual radius of a charged volume. This follows directly from electrostatics.
Similarly, the current iL at the leader channel front is related to the tip
potential U and leader velocity VL by the same expression (2.1 l)t
(2.35)
For a laboratory leader of length L x 10m and R M 1 m, the logarithmic

values are several times smaller, while the linear capacitance is larger than
in a streamer with Y x lo-’ cm.
The linear capacitance of a conventionally semispherical streamer zone
is C1 M 27reo, like the capacitance of a streamer tip. The tip current flowing
into the streamer zone
it = 27r&OU v, (2.36)
is by a factor of In ( L I R )higher than iL,again like in a streamer. But since the
leader logarithm is closer to unity, currents iL and it do not differ as much as
for a streamer. If the current along a leader channel does not vary much, as in a
fairly short leader at constant voltage, the tip current will not differ much from
experimental current i in the external circuit. A typical laboratory leader
has i x iL x it M 1 A, U x 1 MV, and from (2.36) VL M 2x lo4m/s whch is
close to numerous measurements, in which VL x (1-2.2) x lo4mjs [27,28].
Formula (2.36) or (2.35) permits the estimation of any of the three parameters
t Here and below, the external field potential Uo(x)is omitted for brevity. It is indeed small in
laboratory leaders normally observed in a sharply non-uniform field.

- i, U , or VL - from the other two. This is especially useful in studying
lightning leaders when actual data are very scarce.

Thus, a key condition for a long-term spark development is the formation
of a thick space-charge cover around it, having the same sign as the channel
potential. The charge reduces the field on the channel surface, depriving the
channel of its ability to expand due to ionization. It is only a channel with a
small cross section that can preserve the ability to be heated. A charge cover
also contributes somewhat to the linear leader capacitance, because it is
now determined by the much larger cover radius R rather than by the small
channel radius r. An increase in linear capacitance is accompanied by an
increase in the energy input into the channel.
If Nature were a living being and decided to make a spark or lightning
travel as large a distance as possible, it would do this by organizing the
streamer zone and charge cover. In actual reality, everything happens
automatically: the huge voltages that create long sparks produce numerous
streamers at the front end (figure 2.10).
This reminds us of a high voltage electrode creating, under suitable
conditions, a multiplicity of streamer corona elements. This kind of corona
can be registered in laboratory experiments.
Currents of all streamers starting from a leader tip are summed up,
heating the spark channel. This total current charges the region in front of
the tip, neutralizing the charge of the old tip, and when a new tip is
formed, the spark elongates by a length of about the tip length, as in a
Figure 2.10. Photograph of a positive leader in a rod-plane gap of 9 m length at 2 MV;

the electronic shutter was closed at the moment of contact of the streamer zone and
the plane.

single streamer. Part of the streamer zone appears to be behind the tip,
transforming to a new cover for the newly born leader portion. But this
does not decrease the streamer zone length, because meanwhile the zone
has moved forwards together with the tip. Note only that if there are
many streamers they are very close to one another, and they travel in a
self-consistent field close to the critical field (sections 2.2.6 and 2.2.7). Such
streamers are slow and have a low current [4], so that the leader current is,
indeed, a sum of numerous low streamer currents.
2.3.3 Channel contraction mechanism

The mechanism of current contraction in the front region of a leader channel
is not quite clear, especially quantitatively. One may assume the existence of
ionization-thermal instability. This effect looks like the one leading to glow
discharge contraction [26], but it has its own specificity [4]. The instability is
associated with the dependence of electron impact ionization frequency on
field and molecular number density: v,(E,N ) = N f ( E / N ) , where f ( E / N )
is a rapidly rising function at small E / N (figure 2.4). This is the ionization
component of the instability. Its thermal component is due to the fact that
the gas pressure p rapidly equalizes in small volumes at a moderate heating
rate. With p N T = const, a more heated site proves less dense, and the
N
reduced field E / N , determining the ionization frequency, increases there.

As was mentioned above, numerous streamers start from the front end
of a developing leader. The frequency of streamer emission has been shown
experimentally to exceed lo9 sC1 at a typical laboratory spark current of 1 A
[29]. Younger streamers have not lost their conductivity yet. The streamers at
the leader tip are so close to each other that they form a continuous conduct-
ing channel of radius rSum.Current it flows along t h s and the initial leader
channel. It is external current relative to the tip, because it is created by the
whole combination of charges exposed and displaced by the streamer zone
bulk. This current is practically independent of the tip conductivity. In
terms of electric circuit theory, the streamer zone acts as a current source
(an electric power generator with an inner resistance R + m) relative to the
leader tip. Its actual value is very large: R M AU,/i, M U/2it, where A U , is
the voltage drop across the streamer zone. At U M 1 MV and i x 1 A, the
value is R ZZ 0.5MR. No matter what happens to the leader tip or its short
front portion, the current there does not change. What changes is the electric
field, because it depends on the conductivity and radius of the region loaded by
current (in a glow discharge, the field is fixed and the current can vary during
the instability development).
Suppose the current density, whose average cross section value is
j = it(7rr;,J1, has increased, for some reason or other, in a thin current
column of radius ro << r,,,. Then the released energy density j E and gas
temperature T will also increase. The gas density N will become smaller

and E I N larger due to thermal expansion. As the ionization frequency is a

steep function of reduced field, it will grow much faster than E / N . So, the
electron density ne and conductivity a n,/N will rise. As a result of this
N
long chain of cause-effect relationships, the current density j = aE in the

fluctuation region will become still larger, etc. The process may begin with
any link in the chain. In any case, the current density in a particular fluctua-
tion region will be rising without limit until all current it accumulates there.
At the initial stage of instability development, the perturbed current density
does not exceed much an average value. But as current concentrates within a
small cross section mi, the gas heating rate there rises sharply. The instability
now develops very quickly, acquiring an explosion-like character. The
acceleration effect is manifested better in a thinner column with high density
current.
A perturbation region, however, cannot be infinitely thin, and this sets
a limit to the rate of instability development. The matter is that non-
uniformities of electron density ne are dispersed by diffusion, which is
ambipolar at very high density values. The characteristic time for perturba-
tion dispersion is Tamb = ri/4Da, where D , = p+Te is an ambipolar diffusion
coefficient (p+ is ion mobility and T, is electron temperature in volts). In
addition to charge diffusion, non-uniformity dispersion is due to heat
conduction with a characteristic time Tth = r i / 4 ~where
, K is thermal diffusiv-
ity. The former mechanism appears to be more effective in initial air plasma,
since D, x 4 cm2/s (p+ x 2 cm2/s V, T, GZ 2 eV) is an order of magnitude
e
larger than K x 0.3 cm2/s. If a non-uniformity takes less time for dispersion
than for development, i.e., if Tamb is smaller than the instability lifetime qns,
the latter is suppressed at its origin.
The scale for qns is the characteristic time of, say, gas temperature
doubling in a perturbed plasma column, as compared with initial tempera-
ture To. This time is pocpTo/jE,where j E is the power of Joule heat release
and cp is specific heat at constant pressure with the account of thermal
expansion. But this is not all. The higher the instability development rate
-
is, the greater is the steepness of the ionization frequency dependence on
reduced field E / N E T , i.e., on gas temperature. For instance, if a 10%
increase in T raises the ionization rate by 20%, the instability will, generally,
double its rate, as compared with a 10% increase in the ionization rate. This
circumstance brings the factor Ci d In vild In ( E / N ) into the theoretical
formula for qns[26], which characterizes the q ( E / N ) function steepness.
This yields the following expression to be used for estimations:
(2.37)
For calculations, we shall take laboratory leader current i x 1 A and conductiv-

ity a x lop2 (R cm)-’ corresponding to the electron density ne GZ 1014cmp3 of
air ionization by a streamer zone; Ci = 2.5. Suppose the current density in a

perturbation region is j x 40A/cm2, i.e., somewhat higher than the average

value of 30A/cm2 along a channel with the initial radius taken to be
rsum= 0.1 cm. We shall obtain T , , ~x 1O--6 s. From the condition Ta,b 2 -qTins,
under which the instability has a chance to develop further, we find that the
initial radius of a column with accumulated leader current must exceed
ro min M 3 x lO-3 cm. Taking into account the upper limit ro max x
5 x 1O-2 cm derived from energy considerations, we conclude that a probable
-
leader radius prior to thermal expansion is about ro lop2cm. For details, the
reader is referred to [4],but reservation should be made concerning the result
accuracy, which cannot be too high in the present state of the art.
2.3.4 Leader velocity

Streamers generated at the leader channel front cover a distance of several
metres and stop. As was mentioned in section 2.3.2, such streamers are
weak and their propagation is slow; their velocity is close to its low limit
of V, M 105m/s, which means that their lifetime is R/Vs x lO-5 s. This
time is so long that the streamer plasma decays considerably. Only young
streamers, whose lifetime is about the electron attachment time ra x lO-’s
(section 2.2.5), can preserve good conductivity. A young streamer length is
It M Vs7a M 1 cm. A dense fan of such plasma conductors starts from the
channel front. It is this young streamer fan that seems to be registered in
-
photographs as a bright spot with a radius of r, 1 cm in order of magnitude
(figure 2.11) and is generally considered as a leader tip.
Figure 2.11. An instantaneous photograph (0.1 ps exposure) of the tip region of a

leader.

The streamer zone and cover 67
This suggestion is supported by the fact that the radius of a leader

travelling through air pre-heated to 900K is r, RZ lOcm [20]. Indeed, the
plasma decay slows down and the values of r, and It become higher.
Thus, a necessary condition for leader propagation is the tip region
contraction to a very small radius. This results from instability development,
taking a time of about qns. Over this time, all short young streamers
supplying the leader with current transform to the leader channel. Therefore,
over the time of the process providing a steady propagation of the leader tip,
the latter must cover a distance of about its size, i.e., a young streamer length.
Only in this case can a new front region be formed to replace the old one.
Hence, the leader velocity can be evaluated from the respective parameters as
In the absence of attachment or if its rate is low, the role of 7, is performed by

the time of another plasma decay process - recombination. The evaluation
with (2.36) gives a correct order of magnitude for the velocity of laboratory
N -
leaders: at rins lop6 s and It 1 cm, we obtain V , 104m/s.We should
N
like to note that these qualitative and, probably, questionable considerations

have not yet been substantiated by a more rigorous treatment.
Some of the above problems of the leader process will be discussed in
more detail in the subsequent sections of this chapter and further. Here,
our aim was only to give a general idea of the propagation of a long spark
and, presumably, of the first lightning component. A reader interested
exclusively in lightning hazards may find this information sufficient.
2.4 The streamer zone and cover
We have shown above that a streamer zone plays the key role in a leader
process. It is here that a space charge cover is formed which stabilizes the
leader channel, preventing its ionization expansion which would otherwise
exclude plasma heating. A streamer zone is the site of current generation
for heating the leader, providing its long life. In this section, we shall deal,
in some detail, with processes occurring in the streamer zone and leader
cover, defining the priorities in the causative relationships among leader
parameters. We shall show how the process of streamer generation from a
leader tip becomes automatic.
2.4.1 Charge and field in a streamer zone

The tip of a long leader possesses a very high potential: U, 1 MV forN
laboratory sparks and -10 MV or, probably, more for lightning. Streamers
are continuously produced in a leader tip, which means that the field at its
surface Et exceeds the ionization threshold Eiz 30 kV/cm (under normal

conditions). This excess cannot be very large, otherwise a streamer flux would
become too intensive. The excessive charge of the same sign as U , introduced
into the space would create a much stronger reverse field which would reduce
Et to a level close to E,. Therefore, the field E, does not exceed much E, and
has the same order of magnitude. An automatic field stabilization is inherent
in any continuous threshold process of charge generation by an electrode, for
example, in a steady-state corona. Measurements have shown that the field
near a corona-forming electrode is stabilized with high accuracy and does
not respond to voltage rise across the gap; what changes is the corona
intensity, i.e., its current. A leader tip, too, is the site of corona formation,
with an intensity high enough to support a quasi-stationary state in the tip
and streamer zone, corresponding to potential U,. The field at the corona
electrode E,, is shown by stationary corona experiments to be by a factor
of 1.5 higher than E,, if the electrode radius is about the leader tip radius
r - lcm.
At E, = E,, = 50 kV/cm and Y, = 1 cm, the leader tip charge q, =
4mO~:Et % 5 x lO-'C is capable of creating only a small portion of
E,r, = 50kV of an actually megavolt potential U,. The main potential
source is, therefore, the space charge of the streamer zone and cover
surrounding the tip. But the value of U, is primarily determined by character-
istics external relative to the tip. This is the electrode (anode) potential minus
the voltage drop across the leader channel. Consequently, the charge Q, and
the size R of a streamer zone, as well as respective cover parameters, are
established such that they correspond to the proper potential U,. The
mechanism by which a leader 'chooses' the values of Q, and R are directly
related to streamer properties. There are many streamers present in the
zone at every moment of time. They are emitted by the tip at a high frequency
(see below), have different lengths at any given moment and are at different
stages of evolution, with their charges filling up the zone space. Every single
streamer moves in a self-consistent field created by the whole combination of
streamers. The contribution of the leader tip itself (or of its channel) to the
total field has just been shown to be small. One exception is the region
around the tip with a size of its radius.
There are experimental and theoretical grounds to believe that the field
strength in the streamer zone, except for the tip vicinity, is more or less
constant and close to the minimum at which streamers can grow. This is
indicated by measurements of streamer velocity, which does not vary along
the streamer zone. (Attempts to measure a single streamer in the tip
region, where the streamer density is high, have so far failed.) Experiments
show that until the streamer zone of a laboratory leader touches the opposite
electrode, streamers move slowly, at a nearly limit velocity of about lo5 mjs.
This is possible only in a uniform field close to Eo mn (section 2.2.7).Streamers
can travel for such a distance R,at whch the field Eo mn still exists, but they
stop on entering the region with E < E,, ml,.

Suppose, for simplicity, that the streamer zone is a hemisphere with the
centre in the leader tip. The hemisphere changes to a cylindrical cover of the
same radius R with the same order of the space-charge density. A thin con-
ducting leader channel goes along the cylinder axis as far as the hemisphere
centre. When evaluating the zone parameters, one should take into account
the cover charge at the leader end, which also affects the zone field. We can
do this simply by connecting the hemisphere, simulating a streamer zone, to
another hemisphere by mentally cutting it out of the cover space. Let us
assume that there is a uniform radial field E = Eomi, in the sphere. As was
mentioned in section 2.2.7, the theoretical limit of Eo is close to the experi-
mental critical average field in the streamer channel, below which a streamer
cannot propagate. For air, therefore, we have E x E,, x 5 kV/cm. A uni-
form field in sphere geometry corresponds to the space charge density
p = 2 ~ & / r . If the leader tip is far from the earth and grounded electrodes,
its potential is
(2.39)
The sphere charge Q and its surface potential UR are
(2.40)
UR = U, - E R = Ut/2.
For example, for Ut = 1.5 MV, we have R = l S m , the charge of a hemi-

spherical streamer zone equal to Q, = Q / 2 = 6.2 x c. The leader tip
-
charge qt = Q ( E , / E ) ( r , / R ) 2 lOP3Q is indeed negligible, as compared
with the zone charge. Its physical role, however, is very important: the
high field it creates near the tip, Et > Ei >> E,,, is capable of generating
streamers.
As the streamer zone approaches the grounded plane, its length
increases because its boundary potential UR decreases under the action of
charge of opposite polarity induced in the earth. Now it is most of the voltage
U,, rather than its half, which drops across the streamer space. At the
moment of streamer contact with the 'earth', potential UR = 0 and the
zone length L, = U,/E is doubled relative to the value of R from formula
(2.39). This is clearly seen in streak pictures of a laboratory spark (figure
2.12). It is at the moment of streamer contact with a grounded plane that
the critical field E,, x 5 kV/cm was registered experimentally. The measure-
ments make sense only for short (compared with the interelectrode distance)
leaders, when the voltage drop across the channel could be neglected. By
equating the potentials of the anode U , and of the tip, we can write:
E,, = U,/L, U,/L,.

Figure 2.12. Streak photograph of the positive leader channel bottom in an air rod-
plane gap of 12 m in length. The streamer zone is seen to elongate when approaching
the plane cathode.
2.4.2 Streamer frequency and number

The number of streamers present in a streamer zone at every moment of time
is N , = Q,/q,, where q, is the average charge of a streamer. Both charges
were measured experimentally [29,30], the first from the integral of con-
duction current through the anode for short leaders with as yet small cover
charge and the second from the integral of current through a cathode
measurement cell with such a small radius that only one streamer could
touch its surface (with good luck). After a successful contact, the charge of
a conductive streamer section flew into the cathode and through an integrat-
ing circuit. The charge averaged over many registrations was found to be
q, = 5 x lo-’’ C. For the illustration mentioned in the previous section of
this chapter, we find that the number of streamers in a streamer zone of
length 1.5m and charge Q, = 6.2 x lO-5 C is N , M 1.2 x lo5. Similar data
for qs can be derived from the calculations presented in figure 2.8 or from
a simple theoretical treatment, we shall just perform.
A streamer produced by a leader tip crosses the streamer zone over the
time t, = R/V, min M lop5s, which is by two orders of magnitude larger than
the attachment time 7, M lO-’s. Therefore, all electrons are lost from older
streamer portions comparable in length with R . Conductivity is preserved

only along the length I, - -

V , m i n ~ a 1 cm behind the streamer tip. During
the motion of the streamer tip, the charge of the older streamer portions
flows into the new ones located closer to the tip before conductivity turns
to zero. Under steady-state conditions, when the tip goes far away from
the start, the charged portion of length I, moves together with the tip,
supporting by its charge the minimum excess of tip potential over external
potential, (AU,),, x 5 kV, necessary for the streamer propagation (section
2.2.7). The conductor of length I, and radius r, M 0.1 cm carries the charge
(2.41)
As for the charge accumulated in the streamer tip

qst z 2 r ~ O ( A U ! ) ~ Mi ~mO(AU,)iin/Em
r~ M 5 x lo-" C,
which was not taken into account in the calculation or (2.39), it is by one
order less than that distributed along the channel. At N , M lo5, the average
interstreamer space is R/N;I3,i.e., about several centimetres. With this large
separation of streamer tips, the streamers can really be considered solitary
and propagating in an average self-consistent field.
When a streamer reaches the end of the streamer zone, it stops because it
enters a field lower than Eo ,in. Since the streamer zone approaches this field
fairly slowly, at leader velocity VL an order of magnitude lower than
streamer velocity V , (for laboratory streamers), the streamer loses its
conductivity entirely. The ions of its space charge are gradually repelled
(and diffuse), reducing the field near the charge trace, so that the streamer
trace becomes lifeless 'forever'. The streamer zone still passes by for a time
tL = R / V , N lop4s, after which the immobile charged trace, which is now
behind the leader tip, becomes a cover component. Viable streamers fly
-
across the streamer zone over time t , = R / V s m i n lop5s, an order of
magnitude shorter. Therefore, if the frequency of streamer production is
-
v,, the number of viable streamers in a streamer zone is N I v,t,, while
--
the number of non-viable traces, practically coinciding with the total
- -
number of charged streamer portions, is N v,tL. Hence, the streamer
generation frequency is v, N / t L N V L / R lo9 s-'.
2.4.3 Leader tip current

The streamer production frequency is directly related to the leader tip current:
it = qp,. With 9, = lop9C and v, = lo9 s-l, we get it x 1 A, a value typical
for laboratory leaders in the initial stage while the streamer zone has not yet
reached the cathode. T h s current value has been registered in many experi-
ments [27-291, and the relation v, = i L / q s has been confirmed by direct
measurements. The streamers were counted by piece from current impulses
in small cathode measurement cells after the streamer zone boundary had

72 The streames-leader process in a long spark
touched its surface [29]. The counts were integrated over the area. Measure-
ments made at different currents showed that the relation
iL/v, = qs RZ 5 x 10-”C remained constant. It is consistent with measure-
ments and the estimations of average streamer charge presented here.
The formula for leader tip current can be given in a conventional form of
the type i = 7rr2 eneve = T ~ Vwhen
~ , current is expressed as the number of
charge carriers (electrons) per unit current column length T, and electron
velocity V,. This formula can also be changed to the phenomenological
expression (2.36) describing the result of the current process without
indicating the nature of carriers. Although the current carriers in this case
are electrons, we can also speak of ‘macroscopic’ carrier-moving charged
streamer sections. With what we mentioned at the end of section 2.3.2 and
formula (2.36), we have
(2.42)
where T, = Q J R RZ m O U t is the linear charge in a streamer zone and

T~ = q , N 1 / R is the linear charge of ‘macroscopic’ carriers. The three
expressions for current are equivalent to one another but reflect different
aspects of the current process. The second expression in the chain of
equalities (2.42) indicates the current origin whle the latter shows the actual
process of charge transport; the penultimate expression is phenomenological
and describes the result of travel of the streamer zone as a whole.
The mechanism of leader current production just described is valid until
a streamer zone touches the electrode of opposite polarity (the earth or a
grounded object in the case of lightning). Then the situation changes
radically. In the final jump, the charges of all streamers ‘hitting’ the electrode
leave the gap through its surface. Nonviable streamers are no longer
produced, and they possess an ever-decreasing portion of total streamer
zone charge. The last expression in (2.42) is valid in this case, too,
it = T~V , mln, but the value of T~ is no longer equal to the portion VL/ V , mln
of the total streamer zone charge Q,. In the limit, when the leader tip reaches
the electrode, all zone charge will be provided by moving streamers, so the
current will be it = T,V,. The linear charge of the streamer zone in the final
jump remains the same in the order of magnitude as in the initial stage. Of
course, the streamer zone has now a different shape - it is elongated, looking
more like a cylinder than a hemisphere. But still, its longitudinal L, and
transverse R , dimensions are comparable, and the value of ln(L,/R,)
which appears in the denominator of the expression for linear capacitance
in cylindrical geometry (2.8) and (2.35) is about unity. Therefore, leader cur-
rents after the transition to the final jump and before it are related as V s /VL,
the ratio being above 10.
Moreover, the streamer velocity rapidly increases as the streamer zone is
reduced. The potential difference between the leader tip and the grounded

electrode remains the same, U,, whereas the length L, becomes shorter. The
average field in the streamer zone E = U,/L, rises, together with streamer
velocity (section 2.2.7) and leader current it V,. The final jump current
N
was shown by laboratory leader experiments to rise by a factor of tens or

hundreds, from about 1 A to 102-103 A. This fast current rise lasting for
several microseconds is a prelude to a still higher current of the return
stroke. The latter begins when the leader channel reaches the electrode.
The current rise of the final jump is stimulated by the fact that fast streamers
cross a shorter streamer zone much faster than before, so that the zone
plasma is unable to decay as much, preserving the streamer conductivity.
At the end of the final jump, the leader channel appears to be linked to the
opposite electrode by numerous streamer filaments with current (for details,
see [4]).
2.4.4 Ionization processes in the cover

A leader cover contains a large number of non-viable charged traces of
earlier streamers. They were produced and developed when a streamer
zone was passing through this site. The axial field in the cover is very low,
much lower than in the streamer zone. No ionization can occur in it, so it
is of no interest to us. What is important is the radial field created by the
leader surface charge and all cover charges, as in a streamer zone, the only
difference being in geometry. In contrast to a channel cover, however, a
leader tip with a streamer zone is formed as a self-consistent system from
the very beginning. The tip pumps into the zone as much charge as necessary
to maintain the tip field at the level Ei, providing the production of the
necessary number of streamers. A leader channel ‘inherits’ a ready-made
cover. The charge amount and distribution in the former quasi-spherical
zone is unsuitable for the cylindrical geometry and channel potentials U ( x )
different from U,. But the inherited charge of dead streamer traces is
invariable, which means that there must be a mechanism to make the
channel-cover system self-consistent and controllable, since the potential
distribution U ( x ) and linear leader capacitance vary in time.
We mentioned in section 2.3.2 that the ‘intrinsic’ charge of a conducting
channel of length L and radius rL x 0.1 cm would create at its surface a huge
radial field Er, = U / [ r LIn ( L / r L ) ]= 1 MV/cm at channel potential
U x 1 MV. This critical situation is unfeasible owing to the presence of a
cover. The cover charge induces in the conductor an opposite charge which
is to be subtracted from the intrinsic surface charge. As a result, the field E,
created by the resultant charge has a moderate value. It is hard to imagine,
however, that the inherited charge will be as large as is necessary for confining
the resultant surface field in the narrow range -Eln < E, < Ein with
Ei, = 50 kV/cm << El, x 1 MV/cm. No doubt, the cover charge will turn out
to be either too large or too small. In the first case, the channel will be charged

I4 The streamer-leader process in a long spark
negatively (if the leader is positive), the field around it will become negative, its
module exceeding Ei,. A negative corona will be excited and introduce into the
cover as much charge as is necessary to reduce IE,I to Ein.This situation
becomes feasible when the gap voltage is constant or decreases, since then
the channel ‘enters’ the cover with a too-large charge. Indeed, suppose the
charge of a streamer hemisphere of radius R is distributed as p = 2 ~ ~ E / r
(section 2.4.1), creating potential U, = 2ER (2.39) in its centre. The cover
inherits a charge of the same radial density distribution p = 2 ~ ~ E=/ r
~ E ~ U ,and/ R amount
~ r’ = 27reoUt per unit length. At point x far enough
from the channel ends, L >> R , this charge will create potential
U’(x)
=-
1
YTEO
5”J’ 2 ~ r dr
p dz
o o [r2+ ( z - x) ]
1,2 M
2
r’
-In-
~
Le
= U, In-
~R 0
Le
R
(2.43)
where e is the natural logarithmic base. The potential U’ is by the logarithmic

factor larger than the actual value of U,. The excess cover charge which has
created excessive potential is U’ - U, and must be compensated by intro-
ducing a charge of opposite sign.
The other situation, when the inherited charge is too small, is usually
feasible if the gap voltage rises appreciably during the leader evolution.
The channel potential U ( x ) at a given point increases, and the cover must
be charged up.
It is this mechanism of direct or reverse corona display by a ‘wire’, such
as a leader channel, which leads to a self-consistent channel-cover system.
The system is controlled and corrected automatically but not very quickly.
It is sensitive to the slightest variation in potential distribution along the
channel due to the field Ei, being too small compared with El., of the ‘intrin-
sic’ channel charge. A slight effect on the cover is sufficient to change the field
value, and even its direction, at the channel surface. The cover of a develop-
ing leader with a reverse corona acquires a double-layer structure: outside is
the charge inherited from the streamer zone and inside is the new charge of
opposite sign, introduced by the corona. For example, at U, = 1.5 MV,
R = 1.6m, and L = 10m, the linear cover capacitance from (2.43) is
C1 = r’/U’ = 2 x lo-” Fjm. It is defined by the same formula (2.8) but
with the effective radius Reffvarying with the radial charge distribution; for
p M 1/r, re^ = R / e and for p ( r ) = const, re^ = R / e l i 2 , etc. With U = U,,
the corrected steady state charge in the cover is rL = C1U = 3 x lo-’ Cjm.
Since the values of C1 and rL gradually decrease with the leader length,
the surface field must support a negative corona; hence, E, M -50 kV/cm.
The actual channel charge (intrinsic charge minus induced charge) is found
to be rLc= ~ T E ~=E-8.2 , x IO-’ Cjm << rL.Therefore, it is easy to control
the value of rLc and even to reverse its sign.
We have deliberately considered the mechanism of cover-leader self-
regulation in so much detail, because a reverse corona neutralizes the
cover charge in a laboratory spark and lightning during the return stroke,

A long leader channel 75
when the channel potential becomes equal to the earth’s zero potential (see
section 4.4).
2.5 A long leader channel
All ionization processes responsible for the leader development are localized in
the streamer zone, leader tip and a short channel section behind it. In the latter,
gas heating is completed and a quasi-stationary state characteristic of a long
spark is established. In t h s sense, the rest of the channel plays a minor role,
simply connecting the operating part of the leader to a high-voltage source.
High potential and current vital to the ionization and energy supply are
transmitted through the channel. But how much voltage reaches the leader
tip depends on the channel conductivity which, in turn, is determined by the
channel state. For this reason, what is going on in a developing channel is
as important to the leader process as the mechanisms described above.
2.5.1 Field and the plasma state

There are no direct experimental data on the state of a lightning leader
channel. Therefore, of special value is the information derived from laboratory
spark experiments, since it can serve as a starting point in lightning treatments.
Here we present some values derived from experimental data [27] with a
minimum number of assumptions. Streak photographs were taken continu-
ously of a leader propagating from a rod anode to a grounded plane. Pulses
of voltage U, with the microsecond risetime were applied to gaps of various
length d. By measuring the streamer zone length L, in the photographs at
the moment the zone touched the grounded electrode and assuming the aver-
age zone field to be E,, = 4.65 kV/cm (section 2.4), one can find the leader tip
potential U, = E,,L, and evaluate the average field in the leader channel as
EL = ( U o - U , ) / L , where L = d - L, is the channel length (table 2.1).
The accuracy of EL evaluation, however, is not high, because one
calculates a small difference between large values. Besides, measurements
of the streamer zone length at the moment of contact with the cathode
contain errors as large as those of the channel length. When determining
the latter from streak pictures, one can hardly take into account all channel
Table 2.1. Leader parameter derived from experimental data.
5 1.3 2.3 2.7 1.1 750

10 1.9 3.2 6.8 1.5 590
15 2.2 3.6 11.4 1.7 440

bending, which may increase the length by 20-30%. Finally, the accuracy of
E,, is not as high as is necessary for such a delicate operation. Experimental
researchers know that the streamer zone field varies with air pressure,
humidity, and temperature but they do not know the respective corrections.
Nevertheless, the data in table 2.1 demonstrate a decrease in the average field
with increasing channel length. This is also evident from experiments with
superlong sparks. A voltage of 3-5 MV is sufficient to create a spark 100 m
long or longer. The tip potential necessary for the development of a streamer
zone of several metres in length is 1-2 MV, as is clear from table 2.1, there-
fore the average field in such a long channel will be as low as 200-250 V/cm.
These values are more applicable to older, remote channel sections which
have acquired a quasi-stationary state, but the fields close to the tip are much
higher. This follows from many experiments indicating a regular increase in
average field with decreasing leader length. More explicitly this was shown
by supershort spark experiments, when the channel length was only a few
dozens of centimetres [31]. The field in a supershort leader and, therefore,
the field at the respective distance from the tip of a long spark may be 2-
4 kV/cm. At the site of ionization-thermal instability, where current is
accumulated within a thin column, the field was found (section 2.3) to be
20 kV/cm [4]. But far from the tip, it is nearly two orders of magnitude lower.
At a typical experimental leader current of i x 1 A, its velocity is
V , M 1.5-2cm/p, and the lifetime of a leader section at a distance of 3 m
from the tip is at least 150 ps. This time is long enough for the relaxation
processes in the channel to be nearly completed and for a nearly steady-
state to be established. The thermal expansion of the channel, very fast at
the beginning, is also completed by that time. Measurements made in a
10m gap between a cone anode and a grounded plane [28] (voltage 1.6-
1.8 MV, average current about 1 A, and average leader velocity 2 cm/ps)
showed that the channel expansion rate was l00m/s at first but loops
later the rate dropped to 2m/s. Measurements made by the shadow tech-
nique showed the average thermal expansion radius to be rL = 0.1 cm.
According to spectroscopic measurements, the temperature of a channel
which has reached the gap middle is 5000-6000 K. Some other experimental
data on laboratory leaders can be found in [4, 27, 28, 31, 321.
The air ionization mechanism changes radically at temperatures T M 3000-
6000K and relatively low reduced fields (for example, at E = 450V/cm,
T = 5000K, and p = 1 atm, we have E / N = 3 x V/cm2).t In cold air,
t We should like to warn against the commonly used postulate that the field in a leader channel
has a constant value of E / N 8 x 10-'6VcmZ. Some authors use it for the calculation of
breakdown voltages in air gaps, including long ones. At T = SO00 K, we have E = 1.15kV/cm,
which disagrees with experimental data for more or less long leaders (table 2.1) and contradicts
the physics. The underlying implicit suggestion is that the only consequence of air heating is a
change in its density. We shall show that this is not the case.

A long leader channel 71
it is ionization of O2 molecules by electrons gaining energy in a strong field, but

at the above value of E / N , the ionization rate of unexcited oxygen and nitro-
gen molecules and atoms by electron impact is negligible. Electrons are mostly
produced in the associative ionization reaction
N + 0 + 2.8eV + NO' + e. (2.44)
Due to a low ionization potential of NO (9.3eV), the reaction requires a
small activation energy and occurs at a large rate constant. Recent data
[33] give
k,,, = 2.59 x 10-'7T'.43exp (-31 140/T) cm3 s-'. T [K] (2.45)
Direct NO ionization by electron impact may compete with associative
ionization (2.44) but at an electron temperature higher than T, M lo4 K.
Both estimations and kinetic calculations [34] show that thermodynami-
cally equilibrium concentrations of N, 0, NO and electrons at T M 4000-
6000 K (table 2.2) are established for 20-50 ps.
During this time, a leader elongates only by 20-100cm, i.e., the process
of establishing a thermodynamic equilibrium in the channel seems to occur
concurrently with the transitional process of channel formation and heating
to a quasi-stationary state. Although the electron density does not practically
differ from the density due to streamer generation, ne M 10'4cm-3, the
ionization degree at T = 5000, n,/N M 3.3 x lo-', is an order higher than
that in the streamers, n,/N M 4 x lop6. Therefore, intensive ionization
occurs during the evolution of ionization-thermal instability and subsequent
heating to the final temperature.
Thus, a long laboratory leader channel can be subdivided into two
unequal parts. First, there is a relatively short (about 1 m) transitional por-
tion just behind the tip where the gas is gradually heated and additionally
ionized. This is accompanied by a change in the plasma density and con-
ductivity. Second, there is the rest of the channel heated to 5000-6000K,
which has reached a quasi-stationary state. The suggestion of an equilibrium
electron density in this part of the channel generally leads to a correct value
of its radius, Y x 0.13 cm, close to the measured thermal radius [28]. It can be
Table 2.2. Equilibrium air composition at p = 1 atm.
T, K 4000 4500 5000 5500 6000

N, 10'~cm-~ 1.79 1.60 1.48 1.35 1.27
ne, 1013cm-3 0.63 1.70 4.90 11.2 21.4
N ~10" , cm-3 4.70 4.90 4.60 4.35 3.81
N ~1 0, ' ~cm-3 0.25 1.15 3.61 9.92 20.6
N ~1 0 ~ ' ~cm-3
, 1.62 4.54 2.73 1.67 1.03
n e / N , lop5 0.35 1.06 3.31 8.30 16.8

derived from the relation for current i = 7rr2en,p,E if we take electron

mobility to be pe w 1.5 x 1022N-’ cm2(V.s)-‘, i w 1 A, E M 250V/cm and
if we use the values of ne and N from table 2.2, corresponding to T = 5000 K.
In reality, with the T and E / N values characteristic of remote channel
portions, the electron temperature T, may differ considerably from the gas
temperature: T, may be as high as l0000K at T = 5000K. This slightly
shifts the quasi-stationary values of ne relative to the thermodynamic equili-
brium values corresponding to T (as in table 2.2). Stationary n, corresponds
to the equality of forward and reverse reaction rates in (2.44). The forward
reaction rate is independent of T,, while the reverse reaction rate at
T, x 104K is proportional to TL3’*. Hence, the stationary value of II,will
be larger than that in table 2.2 by a factor of ( T e / T ) 3 / 4 2.
As for the less heated, recent channel sections, the difference between
the electron and gas temperatures is greater. The reduced field E”, and
Te must be higher in the unheated channel to provide impact ionization,
since there is no other source of electron production. At temperatures
T < 2500K, this is O2 ionization by electron impact. As the channel is
heated further, NO ionization requiring lower T, and E”, begins to domi-
nate, and only at T > 4000-4500K requiring a still lower field does the
reaction of (2.44) become important. Clearly, the channel field cannot
follow the condition E”, = const because of the change of ionization
reactions with different energy thresholds. Calculations show [34] that the
value of E / N drops from 55 to 1.5Td with heating from 1000 to 6000K
(figure 2.13).
1o4
0 1 2 3 4 . 5 6
Temperature, kK
Figure 2.13. Parameters of the initial leader channel right behind the tip as a function
.
of the gas temperature (model calculations of [34]; 1 Td = lo-’’ V cm2).

A long leader channel I9
2.5.2 Energy balance and similarity to an arc

The older leader portions are similar to an arc in atmospheric air. Current
1 A and temperature 5000 K correspond approximately to the minimum
arc values and the field equal to 200-250V/cm is only by a factor of two
or three stronger than that in a low current arc. So it is natural to look at
the leader channel as an arc analogue.
All characteristics of a long stationary arc at atmospheric pressure when
the plasma is usually quasi-stationary (maximum temperature T , along the
axis, longitudinal field E , current channel radius ro) are defined only by one
‘external’ parameter, normally, by current i. Joule heat released in the current
channel is carried out by heat conduction. Radiation is essential only for very
intensive arcs when the channel temperature exceeds 11 000- 12 000 K. The
further fate of the energy depends on the arc cooling providing its steady-
state. Heat can be removed via heat conduction through the cooled walls of
the tube containing the arc. It can be carried away by the cooling gas flow
or due to a natural convection if the arc burns in a free atmosphere. Definite
relations among E , T , and ro with current i are obtained if the heat release
mechanism is known. None of the above mechanisms (convection does not
seem to have enough time to develop) are operative in a leader channel. One
may suggest that heat is carried farther away from the channel via heat con-
duction, gradually heating an ever increasing air volume. Strictly, this is not
a steady state process, so it is not a simple matter to find all relations.
However, the state of the channel itself is close to a stationary one. This is
due to a small and definite temperature variation in the current channel owing
to an exceptionally strong dependence of equilibrium plasma conductivity on
its temperature. So one can find the relation between the leader channel
temperature T, and the power PI= iE released per unit length. Using the
available experimental values for T and i, one can find E to see that a fairly
low field is sufficient to support plasma in a well developed leader channel.
The electron density in an equilibrium plasma is ne E exp (-Zeff/2kT),
where Zeff is an effective ‘ionization potential’ of the gas. The relation with
an actual ionization potential of atoms is strictly valid for a homogeneous
gas (the Saha equation). For the temperature range T E 4000-6000K, we
have, in accordance with table 2.2, Zeff = 8.1 eV and Ien/K = 94000K,
which is close to INo= 9.3 eV. Since Ieff/2kTE 10, the conductivity 0 ne N
is strongly temperature dependent and decreases with radius much more

than the temperature. Therefore, we can use the concept of a current channel
with a more or less fixed boundary - the radius ro (figure 2.14). By denoting
the channel boundary temperature as Ti and bearing in mind that the
temperature variation in the channel is A T = T , - Ti << T,, we can write
an approximate expression for the channel energy balance:
PI = -27rroX, (g) 10
AT
M 47rr0X, -= 47rX,AT
YO
(2.46)

.T
Figure 2.14. Schematic arc channel with nearly the same distributions of T and o at
the axis of the ‘old’ portion of a leader channel.
where A, is heat conductivity at T = T,. The refined factor 4 appears instead

of 2 if the heat conduction equation for a uniform distribution of heat
sources is integrated over the range 0 < r < yo.
An arbitrary channel boundary should be set such as to allow an adequate
current through the channel. The current should not be too low, because
another channel will appear ‘outside’; it should not be too high either, because
a current-free periphery will arise inside the current channel. Assume, for the
sake of definiteness, that the channel boundary conductivity cr( Ti) is by a
factor of e less than the axial value cr( T,). With the exponential dependence
of cr(T), this approximately yields A T x 2 k T i / I e f . By substituting this
expression into (2.46), we find the desired relation:
kT i 112
PI= iE 87rAm-, T, = (*PI) . (2.47)
Ieff 87rA,k
Expression (2.47) does not contain the radius ro, its account requires a
consideration of the channel environment [26]. The channel temperature
T, grows more slowly with power than Pi12because the air heat conductivity
rises rapidly with temperature in the range of interest. At T, = 5000K,
A, = 0.02 W/cm K; so we have PIx 130 W/cm. For current i x 1, for
which this temperature seems to be characteristic, E x 130 V/cm. The
values of E and PI are only by a factor of two smaller than those found
from experimental evaluations of the field in older leader channel portions.
It is possible that the channel instability and the necessity to heat an increas-
ing air volume require a higher power and a higher field at the given current.
This problem remains unsolved and deserves close study. When applied to
arcs, formula (2.47) gives a fairly good agreement with experiments.
These considerations of the thermal balance of a leader channel permit
establishing the ‘current-voltage’ characteristic (CVC) E ( i ) that we shall
need in the next section. Heat flow from the channel grows with temperature

Voltage for a long spark 81
but not very rapidly. But the conductivity 0 and current i = .irr2uE would
grow very quickly if the field remained constant. As fast would be the
growth of energy release P I = iE,which would set the system out of heat
balance. The balance is maintained because the field drops with rising
current, while the power and temperature do not change much. In an ideal
case with T, = const, which is close to the low temperature conditions for
a leader and low current arc, we have E x i-' from (2.47). The arc CVC is
indeed a descending curve, though its goes down somewhat more slowly
because T, and P1 slightly rise with current [26].
A similarity between the states and CVC curves for quasi-stationary
leaders and arcs was established in model experiments [4]. Sparks 7cm long
were generated in air between rod electrodes. The circuit parameters were
chosen such that the channel current was stabilized at a level characteristic
of a long laboratory spark at the moment of gap bridging. The stabilizing
mode lasted several milliseconds. During this time, a quasi-stationary state
was established under energy supply conditions close to those of the leader
process. The CVC thus measured is approximated by the expression
E = 32 + 52/i V/cm, i [A]. (2.48)
The obtained field appears to be lower than in a leader (84 V/cm at 1 A) and
closer to the arc field.
2.6 Voltage for a long spark
The problem of minimum voltage, at which a spark can develop to a certain

length is of primary importance for high-voltage technology. This quantity
characterizes the electric strength of an air gap, since its bridging by a
leader results in a breakdown. This problem also applies to lightning, because
it is interesting to know the minimum cloud potential at which a lightning
discharge is possible.
Experiments show that the leader process has a threshold character. An
initial leader cannot survive in normal air at gap voltages less than 300-
400 kV. A leader can only be formed at low voltages in a short gap when it
develops as a final jump from the very beginning. Then streamers immedi-
ately reach the opposite electrode, and the energy supply mode differs
from that of the initial stage, with the streamer zone isolated from the
grounded electrode. The reason for a threshold is easy to understand in
terms of the discussion in section 2.2.3 and formula (2.34). A leader channel
has a minimum possible radius. The radius of a cold air column, in which
current can accumulate, is ro > lOW2cm.A thinner current channel is
immediately enlarged by ambipolar diffusion. To heat a column of such
initial radius to 5000K, the leader tip potential from (2.34) must be at
least 200 kV. If we consider the inevitable energy expenditure for ionization

and gas excitation in the streamer zone, this value will increase by, at least, a
factor of 1.5 [4].Therefore, the tip potential in the initial leader stage will be
several hundreds of kilovolts even under favourable conditions.
The voltage U. applied to the gap drops across the leader channel and is
partly transported to the tip. The general formula is
U0 = EL + U, (2.49)
where E is the average field in a leader channel of length L. We showed in
section 2.5 that in a long channel, most of which is in a quasi-stationary
state, E is a more or less definite value varying with current i. The channel
field decreases with increasing current. But current growth requires that
the tip potential determining the leader velocity and current i = C1U, VL
should be raised. At a fixed length L , the Uo(i, L ) function has a minimum,
since it is the sum of a falling component and a component rising with i.
Minimum voltage U 0 ~ * ( Lcorresponds
) to current iOpt(L)optimal for a
leader of length L. It is hardly possible, in the present state of the art, to
find the Uomin(L) function theoretically. We shall try to define its character
using semi-empirical data.
Many experimental physicists have measured the leader velocity varia-
tion with applied voltage U,. Much work has been done on short leaders
because one can neglect the voltage drop across the channel, assuming
U, = U,. With the account of this approximate equality, Bazelyan and
Razhansky [35] suggested an empirical formula: VL z uU:’~, where
a E 1.5 x lo3V-1/2 cm s - l . Physically, the velocity increase with voltage
looks quite natural (though this variation is not strong). We also know
that the tip current is defined by (2.36). This gives the relation
U, = Ai2l3(VL i 1 / 3 )with A = ( C ~ U ) =
N - ~( /2~7 r ~ ~ a ) -Let
~ / ~us. use the
analogy between a well developed leader and an arc and take the CVC
E = b / i typical for a low current arc. Let us put b = 300 V A/cm for numerical
calculations and ignore the difference between the tip and channel currents.
+
We shall then get U, = Lb/i AiZi3and after differentiation
iopt= (3Lb/2A)3i5, = $ U, opt
U. mln = A3/5(3bL/2)215 (2.50)
If a leader develops under optimal conditions, the applied voltage is
shared by the tip and the channel in comparable proportions. The mode
with a low tip potential close to the limit admissible from the energy criteria,
is unprofitable for a long leader, because it corresponds to low current lead-
ing to a considerable voltage drop across the channel. The long spark param-
eters in table 2.3 found from (2.50)with semi-empirical constants are quite
reasonable: these orders of magnitude for current, voltage, and velocity
meet the requirements on the optimal experimental conditions for long
leader development. Besides, the experiment requires a nonlinear, slow
dependence of minimum breakdown voltage on the gap length. It is generally
known that increasing the length of a multi-metre gap is not a particularly

A negative leader 83
Table 2.3. Long spark parameters.
50 3.3 2.0 1.1 2.1 260

100 4.3 2.6 1.3 2.4 170
3000 17 10 17 4.1 22
effective way of raising its electrical strength. This is a key challenge to those
working in high-voltage technology.
The results of extrapolation of formula (2.50) to a lightning leader
( L = 3 km) also lie within reasonable limits.
What is the rate of gap voltage rise necessary for the optimal mode of
spark development? Clearly, the gap voltage must be raised as the spark
length becomes longer according to (2.50), where L is an instantaneous
leader length. The existence of an optimal mode of spark development has
been confirmed experimentally [36-381. It has been shown that for a
breakdown to occur at minimum voltage, the pulse risetime t f must increase
with the gap length d. The authors of [39] recommend the following empirical
formula for the evaluation of an optimal risetime:
lfopt 50d [PSI, d [ml (2.51)
Generally, optimal voltage impulses have a fairly slow risetime. Their values
vary between 100 and 250ps in modern power transmission lines with the
insulator string length of 2-5m. We shall return to this issue in chapter 3,
when considering the diversity of time parameters of lightning current
impulses. The minimum electric strength of an air gap with a sharply non-
uniform field can be found from the formulas [4]
3400
= -[kV], d < 15m
u50% min
1 + 8/d (2.52)
U,,,, min = 1440 + 55d [kV], 15 < d < 30 m
2.7 A negative leader
Most lightnings carry a negative charge to the earth because they are ‘anode-
directed’ discharges. It is always more difficult to break down a medium-
length gap between a negative electrode and a grounded plane. A negative
leader requires a higher voltage. The difference between leaders of different
polarities is due to the streamer zone structure, while their channels and
voltage drop across them are quite similar. Indeed, a gap of about lOOm
long, in which an appreciable part of voltage drops across the channel, is
bridged by leaders of both signs at about the same voltages [2,3].

The streamer zone formation in a negative leader requires a higher tip

potential for the same reason as a single anode-directed streamer needs a
higher voltage for its development. Fast streamers with the velocity V,
much higher than the electron drift velocity V, do not exhibit much difference
associated with polarity. But streamers in a leader streamer zone are slow:
V, x V,. It is of great importance whether the components of electron
velocity relative to the streamer tip are summed, V, + V,, as in a cathode-
directed streamer, or subtracted, V, - V,, as in an anode-directed one. In
the former case, electrons produced in front of the tip move towards it,
and the ionization occurs in a strong field near the tip. In the latter, electrons
tend to ‘run ahead’ of the moving tip and spend most of their time in a lower
field, so that the ionization occurs under less unfavourable conditions.
The fact that negative streamers generally require a higher field and
voltage has been supported by many experiments. They show that the
average critical field, which defines the maximum streamer length in formula
(2.32), is twice as high for an anode-directed streamer as for a cathode-
directed one: E,, x 10 kV/cm against E,, x 5 kV/cm. We shall illustrate
this with figure 2.1 5 for a gap of length d = 3 m between a sphere of radius
ro = 50cm and a grounded plane. The streamers stopped, having covered
the distance I,,, at negative sphere potential U, = 1.5 MV (the unperturbed
potential at the stop with the account of the sphere charge reflection in the
plane is Uo(Zmax)x 0.25 U,). Under these conditions, cathode-directed
streamers practically cross the whole gap.
The propagation mechanism and streamer zone structure of a negative
leader are much more complicated than those of a positive leader and are still
Figure 2.15. Anode-directed streamers from a spherical cathode of 50 cm radius at a

negative voltage impulse of 1.8 MV and a 50 ps front duration.

poorly understood. In the 1930s, when Schonland started his famous studies
of lightning [41], a negative leader was found to have a discrete character of
elongation, so it was termed stepwise. Streak photographs exhibit a series of
flashes, indicating that the leader propagates in a stepwise manner. Later, a
similar process was found in a long negative leader produced in laboratory
conditions 142,431. With every step, a negative leader elongates by dozens
of centimetres, or by several metres in superlong gaps [3]; steps of a hundred
metres have been registered in negative lightning discharges. Every step of a
laboratory leader is accompanied by a detectable current overshoot which
quickly vanishes during the time between two steps.
Without going into theoretical explanations of this mechanism, based
on an unverified hypothesis, let us see what information can be derived
from streak photographs of the process, made during laboratory experiments
[44]. These are naturally more informative than streak photographs of a step-
wise lightning leader. It is seen from figures 2.16 and 2.17 that in the intervals
between the steps, the tip of a negative leader slowly and continuously moves
on together with its streamer zone made up of anode-directed streamers. The
main events occur near the external boundary of the negative streamer zone.
It seems that a plasma body elongated along the field arises there and is
polarized by the field (compare with the discussion in section 2.2.7). The
positive plasma dipole end directed towards the main leader tip serves as a
starting point for cathode-directed streamers. They move towards the tip,
thus elongating the conducting portion of the channel and enhancing the
negative field at its end directed to the anode. Almost at the same time, the
plasma body generates an anode-directed streamer. T h s nearly mystic
picture of streamer production in the gap space is clearly seen in a streak
photograph in figure 2.18. Nothing like this has ever been observed with a
positive leader.
Figure 2.16. A schematic streak picture of a negative stepped laboratory leader: (1,2)
secondary cathode- and anode-directed streamers from the gap interior; (3) secondary
volume leader channel; (4)main negative leader channel; (5) its tip; (6) plasma body;
(7), (8) tip of secondary positive and negative leader (9) leader flash concluding step
development.

Figure 2.17. A streak photograph of the initial stage in a negative laboratory leader.
Marking numbers correspond to figure 2.16.
The polarized plasma section becomes the starting point not only of
streamers but also of secondary leaders which follow them. They are
known as volume leaders. A positive cathode-directed volume leader grows
intensively. Normally, its streamer zone almost immediately reaches the
main negative leader, so it looks as if the secondary positive leader develops
Figure 2.18. The origin of anode-directed (1) and cathode-directed (2) streamers from
the gap interior; (3) initial flash of a negative corona (static photograph) which trigger
a streak photograph regime; (4)arisen negative leader.

in the final jump mode, i.e., very quickly. The negative volume leader moves
towards the anode somewhat more slowly. When the tips of the main
negative and of the positive volume leaders come into contact, they form a
common conducting channel, giving rise to the process of partial charge
neutralization and redistribution. As a result, the former volume leader
acquires a potential close that of the main negative leader tip. This process
looks like a miniature return stroke of lightning, accompanied by a rapidly
rising and just as rapidly falling current impulse in the channel and external
circuit. The intensity of the channel emission increases for a short time. It is
hard to say what exactly stimulates this increase - the short temperature rise
or the ionization in the channel cover: which changes the cover charge,
thereby getting ready for a potential redistribution along the channel (see
section 2.4.4). The negative portion of the plasma dipole turns to a new
negative tip of the main leader. This is the mechanism of step formation
and stepwise elongation of the main channel. Then the story is repeated.
The picture just described gives no ground to draw the conclusion about
a stepwise character of negative leader development. The motion of a nega-
tive leader is continuous, but secondary positive volume leaders, also contin-
uous, produce a stepwise effect. Discrete is the final result of their ‘secret
activity’, but only if the observer is equipped with imperfect optical instru-
ments. In other words, what is generally known as a step is an instant
result of a long continuous leader process. As for gap bridging by a main
negative leader, one should bear in mind that most of the channel is created
by auxiliary agents - by a succession of positive volume leaders.
This picture has been reconstructed from streak photographs. But we
still do not know how polarized plasma dipoles are formed far ahead of
the main leader tip. Their appearance is hardly a result of our imagination.
Steps can be produced deliberately by making a volume leader start from a
desired site in the gap. For this, it suffices to place there a metallic rod several
centimetres long (figure 2.19). A series of rods placed in different sites of a
gap will create a regular sequence of volume leaders. The work [45] describes
an experiment with a negative leader 200 m long. Its perfectly straight trajec-
tory was predetermined by seed rods suspended by insulation threads at a
distance of 2-3m from each other. A volume leader started from a rod
when it was approached by the negative streamer zone boundary of the
main leader. Clearly, the rods are polarized by the streamer zone field to
serve as seed dipoles instead of natural (hypothetical) plasma dipoles.
There are many hypotheses concerning the stepwise leader mechanism,
but they are so imperfect, lacking strength, and, sometimes, even absurd that
we shall not discuss them here. We are not ready today to suggest an alter-
native model either. Additional special-purpose experiments could
certainly stimulate the theory of this complicated and challenging phenom-
enon. It would be desirable to take shot-by-shot pictures of a negative
leader tip region with a short exposure. A sequence of such pictures would

Figure 2.19. An artificially induced step: (1) initial flash of a negative corona from a
spherical cathode; (2, 3) cathode- and anode-directed leaders from a metallic rod
2.5 cm long, placed in the gap interior; (4) leader flash concluding the step develop-
ment; (5) new streamer corona flash from the tip of the elongating channel.
form a film more accessible to unambiguous interpretation than continuous

streak photographs with confusing overlaps of many details.
References
[l] Lupeiko A V, Miroshnizenko V P et a1 1984 Proc. II All-Union Conf Phys. of

Electrical Breakdown of Gases (Tartu: TGU) p 254 (in Russian)
[2] Baikov A P, Bogdanov 0 V, Gayvoronsky A S et a1 1998 Elektrichestvo 10 60
[3] Gayvoronsky A S and Ovsyannikov A G 1992 Proc. 9th Intern. Conf on Atmosph.
Electricity 3 (St Peterburg: A.I. Voeikov Main Geophys. Observ.) p 792
[4] Bazelyan E M and Raizer Yu P 1997 Spark Discharge (Boca Raton, New York:
CRC Press) p 294
[5] Loeb L B 1965 Science (Washington D.C.) 148 1417
[6] D’aykonov M I and Kachorovsky V Yu 1988 Zh. Eksp. Teor. Fiz. 94 32
[7] D’aykonov M I and Kachorovsky V Yu 1989 Zh. Eksp. Teor. Fiz. 95 1850
[8] Shveigert V A 1990 Teplofz. Vys. Temperatur 28 1056
[9] Bazelyan E M and Raizer Yu P 1997 Teplofiz. Vys. Temperatur 35 181 (Engl.
transl.: 1997 High Temperature 35)
[lo] Raizer Yu P and Simakov A N 1996 Piz. Plazmy 22 668 (Engl. transl.: 1996
Plasma Phys. Rep. 22 603)
[ l l ] Dutton J A 1975 J. Phys. Chem. Rex Data 4 577
[12] Cravith A M and Loeb L B 1935 Physics ( N . Y . ) 6 125
[13] Raizer Yu P and Simakov A N 1998 Piz. Plazmy 24 700 (Engl. transl.: 1996

References 89
[14] Vitello P A, Penetrante B M and Bardsley J N 1994 Phys. Rev. E 49 5574

[15] Babaeva N Yu and Naidis G V 1996 J . Phys. D: Appl. Phys. 29 2423
[16] Kulikovsy A A 1997 J . Phys. D: Appl. Phys. 30 441
[17] Aleksandrov N L, Bazelyan E M, Dyatko N A and Kochetov I V 1998 Fiz.
Plazmy 24 587 (Engl. transl. 1998 Plasma Phys. Rep. 24 541)
[18] Bazelyan E N and Goryunov A Yu 1986 Elektrichestvo 11 27
[19] Aleksandrov N L and Bazelyan E M 1998 J . Phys. D: Appl. Phys. 29 2873
[20] Aleksandrov D S, Bazelyan E M and Bekzhanov B I 1984 Izv. Akad. Nauk
SSSR. Energetika i transport 2 120
[21] Bazelyan E M, Goryunov A Yu and Goncharov V A 1985 Izv. Akad. Nauk
SSSR. Energetika i transport 2 154
[22] Aleksandrov N L and Bazelyan E M 1999 J. Phys. D: Appl. Phys. 32 2636
[23] Gayvoronsky A S and Razhansky I M 1986 Zh. Tekh. Fiz. 56 1110
[24] Kolechizky E C 1983 Electric Field Calculation for High- Voltage Equipment
(Moscow: Energoatomizdat) p 167 (in Russian)
[25] Raizer Yu P, Milikh G M, Shneider M N a n d Novakovsky S.V. 1998 J . Phys. D:
Appl. Phys. 31 3255
[26] Raizer Yu P 1991 Gas Discharge Physics (Berlin: Springer) p449
[27] Gorin B N and Schkilev A V 1974 Elektrichestvo 2 29
[28] ‘Positive Discharges in Air gaps at Las Renardieres - 1975’ 1977 Electra 53 31
[29] Bazelyan E M 1982 Izv. Akad. Nauk SSSR. Energetika i transport 3 82
[30] Bazelyan E M 1966 Zh. Tekh. Fiz. 36 365
[31] Bazelyan E M, Levitov V I and Ponizovsky A Z 1979 Proc. 111 Inter. Symp. on
High Voltage Engin. (Milan) Rep. 51.09 p 1
[32] Meek J M and Craggs J D (eds) 1978 Electrical Breakdown of Gases (New York:
Wiley)
[33] Makarov V N 1996 Zh. Prikl. Mekh. Tekhn. Fiz. 37 69
[34] Aleksandrov N L, Bazelyan E M, Dyatko N A and Kochetov I.V. 1997 J. Phys.
D: Appl. Phys. 30 1616
[35] Bazelyan E M and Razhansky I M 1988 Air Spark Discharge (Novosibirsk:
Nauka) p 164 (in Russian)
[36] Stekolnikov I S, Brago E N a n d Bazelyan E M 1960 Dokl. Akad. Nauk SSSR 133
550
[37] Stekolnikov I S, Brago E N and Bazelyan E M 1962 Con$ Gas Discharges and the
Electricity Supply Industry (Leatherhead, England) p 139
[38] Bazelyan E M, Brago E N and Stekolnikov I S 1962 Zh. Tekh. Fiz. 32 993
[39] Barnes H and Winters D 1981 IEEE Trans. Pas-90 1579
[40] Gallet G and Leroy J 1973 IEEE Conf. Paper C73-408-2
1411 Schonland B 1956 The Lightning Discharge. Handbuch der Physik 22 (Berlin:
Springer) p 576
1421 Stekolnikov I S and Shkilev A B 1962 Dokl. Akad. Nauk SSSR 145 182
[43] Stekolnikov I S and Shkilev A B 1963 Dokl. Akad. Nauk S S S R 145 1085; 1962
Intern. Con$ (Montreux) p 466
1441 Gorin B N and Shkilev A V 1976 Elektrichestvo 6 31
[45] Anisimov E I, Bogdanov 0 P, Gayvoronsky A S et a1 1988 Elektrichestvo 11 55

Chapter 3
Available lightning data
Scientific observations of lightning were started over a century ago. Much

factual information has accumulated about this natural phenomenon since
that time. Most of it, however, has been obtained by remote observational
techniques which can reveal only external manifestations of lightning. This
is not the researchers’ fault. Even a long laboratory spark keeps the
experimenter at a respectful distance: there have been single and mostly
unsuccessful attempts to study the leader interior and the ionization region
in front of its tip. No attempts of this kind have yet been made with lightning.
Nevertheless, the accumulated material is being analysed and systematized,
so that our knowledge about atmospheric electricity is gradually expanding.
A number of carefully written books has made the results of field studies
of lightning accessible to specialists. Among them, of great interest is the
recent book by Uman [l] and the co-authored work edited by Golde [2].
The reader will find there nearly all available data on lightning, so there is
no need to discuss them in this book. We have set ourselves a different
task - to select the few data available on the lightning discharge mechanism
and to try to build its theory. In addition, we shall make a detailed analysis of
lightning characteristics important from the practical point of view. The
nature of hazardous effects of atmospheric electricity on industrial objects
will be considered in much detail and lightning protection principles will
be offered.
This task cannot be solved completely, because many lightning param-
eters have never been measured or, more often, even estimated in order of
magnitude. One hope is a method similar to the identical text analysis used
in cryptography to read a text written in a dead language. If there is at
least part of the text written in an accessible, better, related, language, the
task is not considered hopeless. With patience and ingenuity, the researcher
has a chance if he compares these texts carefully. In this respect, we expect
much from long spark studies. Clearly, a spark and lightning are phenomena
90

Atmospheric field during a lightning discharge 91
of different scales, but it is also clear that both have a common nature. For
this reason, we shall often compare the parameters of lightning with those of
a long spark. We should like to emphasize that this will be a comparison
rather than a direct extrapolation, because there is no complete analogy
between the two phenomena.
3.1 Atmospheric field during a lightning discharge
There is no strict answer to this physically ambiguous question. It is neces-

sary to specify what part of the space between the cloud and the earth is
meant. One thing is clear - the electric field at the lightning start must be
high enough to increase the electron density by impact ionization. This
value is Ei= 30 kV/cm for normal density air and about 20 kV/cm at an
altitude of 3 km (the average altitude for lightning generation in Europe).
Such a strong field has never been measured in a storm cloud. The maximum
values were recorded by rocket probing of clouds (10kV/cm, Winn et al,
1974 [7]) and during the flight of a specially equipped aeroplane laboratory
(12kV/cm). The value obtained by Gunn [4] in 1948 during his flight on a
plane around a storm cloud was about 3.5 kV/cm. The values between 1.4
and 8 kV/cm were obtained from some similar measurements [3-91. It is
hard to judge about the accuracy of these measurements, especially those
made in strong fields, because parts of the field detector or the carrier-
plane parts close to it can produce a corona discharge. In any case, the
corona space charge will not allow the strength in the region being measured
to go beyond a threshold value (for details, see [20]). There are reasons to
believe, however, that a corona on hydrometeorites (water droplets, snow
flakes, ice crystals) keeps the field at a level below Ei in the whole of the
cloud, If this is indeed so, a field can be enhanced above Ei only in a small
volume for a short time, say, as a result of eddy concentration of charged
hydrometeors. This enhancement will be reduced to zero by a corona for
less than a second. The experimenter has no chance to guess where the
field may be locally enhanced to be able to introduce a probe detector there.
Theoretically, it is also important to know the average gap field capable
of supporting a lightning leader. The field decreases in the charge-free space
from the cloud towards the earth. At the earth, the storm field was found to
be 10-200 V/cm. Such a low field did not prevent the lightning development.
Lightnings were deliberately produced in numerous experiments described
by Uman [l, 10-161. A rocket was launched from the earth, pulling behind
it a thin grounded wire. A lightning leader was excited at 200-300m above
the earth’s surface. The near-surface field during a successful launching
was usually 60- 100 V/cm.
Strictly, measurements made at two points, at the earth and in the cloud,
are insufficient for an accurate evaluation of an average electric field. The

92 Available lightning data
km
6-
4-
2-
0 I
Figure 3.1. The ‘dipole’ model of the charge distribution in a storm cloud.
space between the cloud and the earth should be scanned, and this must be
done for some fractions of a second, just before a lightning discharge, in
the vicinity of its anticipated trajectory. Unfortunately, such attempts have
not been quite successful. More successful were measurements made at
points on the earth’s surface separated at a distance of hundreds and
thousands of metres [17-191. These have been used to reconstruct the
charge distribution within a storm cloud, invoking the results of direct
cloud probing. The reconstruction procedure and its possible errors are
discussed in [20]. Generally, with simultaneous field measurements made at
n points, one can write a closed set of equations for the same number of
parameters of charged regions. Its solution provides the parameters, for
example, the average space charge densities in pre-delineated regions.
Most often, the number of points is too small, so the results obtained only
permit the construction of simplified models with point charges. Very common
is the dipole model with a negative charge at 3-5 km above the earth with the
same value of the positive charge raised at a double altitude. Sometimes, a
small positive point charge is added to them, which is placed at a distance
by 1-2 km closer to the earth than the negative charge. All point charges are
assumed to be located along the same vertical line (figure 3.1).
The concept of a cloud filled by charged layers of different signs is
based on probe measurements of charge polarity in hydrometeors. In this
respect, this model raises no doubt. But as for the field distribution, the
measurement error is too large, especially for the space in the cloud between
two point charges. Luckily, the descending lightning trajectory lies mostly
outside of the cloud, in the air free from charged particles. For this part of
the trajectory, the average field evaluation in terms of a simple model
makes sense.

Atmospheric field during a lightning discharge 93
We shall illustrate the procedure of deriving information from such

measurements. Suppose we have at our disposal the values of field El at
the earth's surface just under an anticipated charged centre of a storm
cloud, as well as the values for field E2 at the lower cloud boundary (also
under the charged centre) measured during a plane flight around the
cloud. The altitude of the lower boundary, h, is also known. Assume the
centre of the main lower charge q to be above the lower cloud boundary
at an unknown distance r. If we ignore the effects of the remote upper
charge and of the additional charge lying under the main one, we can
write
2q 4
El = E2 = - q + (3.1)
47r&o(h+ Y)2 ' 47reor2 47r&o(2h+ r)2 '
The factor 2 in El and the second term in E2 are due to the action of charge
induced in the earth's conducting plane (mirror reflection). Since in reality
El << E2, it is natural to suggest that r << h. Then, with the second term in
E2 ignored, we find
q = 27r~o(h+ r)2E1, r M ah, a = (E1/2E2)1'2. (3.2)

Substituting the values above, i.e., El = lOOV/cm, E2 = 3000V/cm, and
h = 3 km, we shall find q = 6.3 C, a = 0.13, and r M 390m. The average
field in the region between the lower cloud boundary and the earth,
equal to the lower boundary potential p2 M q / 4 7 r ~ ~divided
r by the cloud
distance from the earth, h, Ea" ( E l E 2 / 2 ) 1 1 2M 390V/cm. Allowance for
the second term in E2 (the effect of mirror charge reflection by the earth)
can hardly be justified, because our model did not take into account the
effect of the upper charge of opposite sign. This charge is closer to the
cloud edge than that reflected by the earth and has, therefore, a greater
effect on E2, Its consideration, however, simple though it may seem,
would require field measurement at another point of space and another
equation for finding a new unknown - the altitude of the upper charge
centre of the dipole.
A larger scale correction would, probably, be necessary to account for
the effect, on the near-earth field, of the space charge induced by coronas
from pointed grounded objects (tree branches, high grass, various buildings,
etc.) [21]. Estimations made just at the earth's surface show that this charge
reduces the actual field of a storm cloud by half. So one cannot say that
lightning moves in an unusually low electric field. These are just the values
at which superlong sparks are excited in laboratory conditions (see
chapter 2). Therefore, there is no need to invent a special propagation
mechanism for lightning, different from that of a long laboratory spark, if
we deal with average electric fields capable of breaking down the cloud-
earth gap.

3.2 The leader of the first lightning component
The leader of the first component of lightning develops in unperturbed air, so

only this leader behaviour should be compared directly with laboratory data
on long spark leaders. The comparison can be carried out along two lines. We
can first compare the leader structure in lightning and a spark and, second,
their quantitative parameters, primarily velocities.
3.2.1 Positive leaders

Streak pictures of a positive leader are easy to interpret. So we shall begin
with positive lightnings, though their occurrence is not frequent. Many
books and papers refer to the successful streak photographs of a descending
positive leader taken by Berger and Fogelsanger in 1966 [22].Its schematic
diagram is reproduced in figure 3.2. The leader became accessible to photo-
graphy at 1900m above the earth's surface. It moved down in a continuous
mode, without an appreciable intensity variation. The average leader velocity
over the registration time was 1.9 x lo6 m/s, increasing somewhat as the
leader tip approached the earth. Such a leader is much faster than a long
laboratory spark, whose velocity is 50-100 times lower at minimum break-
down voltage before the streamer zone contacts a grounded electrode.
In the streak picture, the leader tip looks much brighter than its channel,
but no signs of a streamer zone can be identified. We cannot say how the
original negative looks, but in the published photograph the resolution
threshold is hardly less than 50m. It is a very large value for a streamer
leadp channel
L- return
E stroke
b- 1.5 ms -A
,
Figure 3.2. Schematic streak picture of a positive descending lightning leader regis-
tered on the San Salvatore Mount in Switzerland [22].

The leader of the first lightning component 95
Figure 3.3. Schematic streak picture [22] of a positive ascending leader from a 70-m
tower on the San Salvatore Mount.
zone. With the data of section 2.4.1, we can calculate the leader tip potential
U,, at which the streamer zone will exceed at least twice the length of about
loom, a threshold value for the measuring equipment. For the average
streamer zone field E, = 5 kVjcm, the potential derived from (2.39) is
U, % 100 MV. Such a high value cannot be typical of lightning with average
parameters. There is another circumstance preventing streamer zone registra-
tions - different radiation wavelengths of a hot leader channel and a cold
streamer zone. Violet and ultraviolet radiation from streamers is dissipated
by water vapour and rain droplets in the air much more than long wavelength
radiation characteristic of a mature channel. At a distance of about a kilo-
metre between the lightning and the registration site (closer distances are
practically unfeasible), a streamer zone may become quite invisible to the
observer's equipment. Note that this is totally true of optical registrations
of an ascending positive leader.
A schematic streak picture of an ascending positive leader, based on 18
successful registrations [22],is shown in figure 3.3. All lightnings started from
a 70-m tower on the San Salvatore Mount near the Lake Lugano. The leader
does not exhibit specific features that would distinguish it from a long labora-
tory spark. On the whole, it developed in a continuous mode with irregular
short-term enhancements of the channel intensity. Normally, they did not
accelerate the leader development. Something like this has been observed
in a long laboratory spark. The streak picture in figure 3.4 demonstrates
this with reference to a positive leader in a sphere-plane gap 9 m long. But
this phenomenon has nothing to do with a stepwise elongation of a negative
leader channel.
The velocity of an ascending positive leader near the starting point is
close to that of a laboratory spark, about 2 x 104m/s. From some data
[22], it was in the range of (4-8) x 104m/s for a channel length of 40-
100m; but when the leader tip was at a height of 500-1 150m, it increased
by nearly an order of magnitude, to 105-106 mjs.

Figure 3.4. A streak photograph of the initial stage of a positive leader in a 9-m rod-
plane gap, displaying short flashes of the channel.
3.2.2 Negative leaders

Negative lightnings occur more frequently than positive, and their registra-
tion is more common. The main distinguishing feature of the negative
leader of the first lightning component is its stepwise character. The leader
tip leaves a discontinuous trace in streak pictures which look like a movie
film (figure 3.5). One can sometimes find such pictures in a sports magazine
illustrating the successive steps in a sportsman’s performance. The bright
flash of the tip and the channel right behind it are followed by a dead zone
with practically zero intensity. This is followed by another flash showing
that the tip has moved on for several dozens of metres. Such negative
leader behaviour was observed by Schonland and his group as far back as
the 1930s [24,25]. According to their registrations, the average pause between
the steps was close to 60ps, with a spread from 30 to loops, and the step
Figure 3.5. Schematic streak picture of a descending negative leader.

The leader of the first lightning component 91
I I I I I 1
b
0 5 10 15
20 25 Yp
Figure 3.6. Typical integral velocity distributions for descending leaders of the CY- and
P-types (from [24, 251).
length varied between 10 and 200 m with the average value being 30 m. The
duration of a step is likely to be within several microseconds. The available
streak photographs are not good enough to identify the details of a step.
In any case, it is hard to decide whether it is similar to a step of the long
negative spark described in section 2.7.
A stepwise negative leader approaches the earth at an average velocity
of 105-106m/s. Two descending leader types can be identified in terms
of their velocity: slow a-leaders and fast P-leaders. The former travel
at their step-averaged velocity; it varies with the discharge in the range of
(1-8) x105m/s with the average value of 3 x 105m/s. The respective P-
leader parameters are 3-4 times higher. This can be seen in figure 3.6 showing
the integral velocity distributions described in [24,25]. Usually, P-leaders are
more branched and their steps are longer. They abruptly slow down when
they approach to the earth, after which they behave as a-leaders.
An ascending negative leader also has characteristic steps. Most of the
13 registered leaders ascending from a 70-m tower on the San Salvatore
Mount [22] were identified as a-leaders. They have relatively short steps
(5-18 m) and a velocity of (1.1-4.5) x lo5mjs. Two of the discharges were
referred to @-leaders because their velocity was (0.8-2.2) x lo6 m/s and the
step length up to 130m. On the whole, ascending and descending stepwise
leaders do not show significant differences.
The registrations of ascending discharges from the San Salvatore Mount
provide direct evidence for the existence of a streamer zone in a lightning
leader. Registrations made at a sufficiently close distance, which became

Figure 3.7. Schematic diagram of the initial development of a leader ascending from
a 70-m tower on the San Salvatore Mount, as viewed from close distance streak
photographs.
possible due to the tower top being the only starting point, show streamer
flashes arising at the moment a new step begins. Streamers were initiated
not only from the tip of the main channel but also from its branches
(figure 3.7).
3.3 The leaders of subsequent lightning components
Leaders of lightning components following the first one are known as dart
leaders because of the absence of branches. The streak photograph in
figure 3.8 shows the trace of only one bright tip looking like a sketch of an
arrow or dart. A dart leader follows the channel of the previous lightning
component with a velocity up to 4 x lo7mjs. Averaging over many registra-
tions gives the value (1-2) x 107m/s, with the minimum values being an
order of magnitude less than the maximum one [23,25]. The dart leader
velocity does not vary much on the way from the cloud to the earth.
Figure 3.8. A schematic streak picture of a dart leader.

The leaders of subsequent lightning components 99
Figure 3.9. A streak photograph of a well-branched lightning striking the Ostankino

Television Tower; the components along the branches A and B are formed at different
moments of time.
Similar results have recently been obtained from 23 streak photographs

of fairly good quality showing dart leaders taken in Florida by a camera with
a time resolution 0.5 ps [26]. The average velocity of a dart leader varied from
5 x lo6 to 2.5 x lo7 m/s in some registrations; it was (1.6-1.8) x lo7 m/s for
three typical pictures presented in the publication.
It is clear that a dart leader somehow makes use of the previous channel
with a different temperature, gas density and composition. There are several
indications to this. First, there is a tendency for the dart leader velocity to
decrease with increasing duration of the interleader pause. This is because
the gas in the trace channel is gradually cooled to return to the original
condition. If a pause lasts longer, the subsequent component may take its
own way. Figure 3.9 shows a lightning discharge which struck the Ostankino
Television Tower in Moscow. Some of its initial components followed a
common channel but then the discharge trajectory changed. Naturally,
there is nothing like a dart leader in unperturbed air - the leader of each
next component develops in a step-wise manner.
Second, it has been found in triggered lightning investigations [ 15,271
that a dart leader requires a current-free pause for its development. The
long-term current, which supports the channel conductivity in the period
between two subsequent components, must entirely cease to allow the
channel to partly lose its conductivity. Only then will the channel be ready
to serve as a duct for a dart leader. But if a new charged cloud cell is involved
and raises the potential of the channel with current, an M-component is
produced instead of a dart leader. Its distinctive feature is a higher intensity
of the existing channel lacking a well defined tip (figure 3.10). The absence of

Figure 3.10. A streak photograph of a well-branched lightning with M-components.
a clear luminosity front is a serious obstacle to the measurement of its

velocity. Investigators often point to a nearly simultaneous increase of the
light intensity in the whole channel. This suggests either an almost sub-
light velocity of the leader or an exceptionally smeared boundary of its
front. In contrast to a dart leader, an M-component is never followed by a
distinct return stroke with a high (10-100 kA) rapidly rising current impulse.
Both the external view and photometric data obtained from streak
photographs reveal clearly a dart leader tip. Some authors [26] made an
attempt to measure the time variation of the leader light intensity. Although
the measurements were performed near the time resolution limit, they indi-
cate that the light pulse front at the registration point rises for 0.5-1 ps
and then is stabilized for 2-6 ps. Therefore, with the dart leader velocity of
1.5 x lo7m/s, the extension of the front rise is 7.5-15m with the full pulse
front length of 35-105m. It can be mentioned, for comparison, that the
M-component has a pulse front, if any, of a kilometre length.
It is important for the theory of dart leaders that they always move from
a cloud down to the earth. This means that the voltage source that excites
them is 'connected' to the trace channel of the previous component right in
the cloud. The direction of the previous leader does not matter much because
their channels are equally suitable for the development of a dart leader.
3.4 Lightning leader current
We can only guess about the values of descending leader currents or estimate
them from indirect data. We shall make such estimations in section 3.5, using
leader charge data, also obtained indirectly. Ascending leader currents are

Lightning leader current 101
Figure 3.11. A schematic oscillogram of the leader current in an ascending lightning.
not difficult to measure, and there have been many measurements of this
kind. Normally, a current detector is mounted on top of a tower dominating
the locality [28-301. The current impulse of an ascending leader, registered on
an oscillogram, lasts for about 0.1 s, corresponding to the time of ascending
leader development. The current nearly always rises in time (figure 3.11). The
current supplies an elongating leader with charges. Physically, these charges
are induced by the electric field of a cloud. When a leader approaches a cloud,
going through an increasingly higher field, the linear density of induced
charge T increases. Besides, the leader goes up with an increasing velocity
V,, reducing the time for the charge supply. A combination of these factors
raises the current i = 7VL. At the moment an ascending leader starts its
travel, its current is lower than 10A, whereas at the end of the travel, it
may rise to 200-600 A, with an average value of about 100 A. Sometimes,
just before the leader begins its continuous elongation, impulses with an
amplitude of several amperes may arise against the background of a milli-
ampere corona current.
Current oscillograms of an ascending leader triggered from a thin wire
elevated by a small rocket to 100-300m [13,31] give a similar picture.
They show the same slowly rising impulse with an amplitude of 100-200A
and duration 50-looms. It has no overshoots at the front, even if the
leader goes up in a stepwise mode.
There are no reasons to suggest any principal difference between average
currents of ascending and descending leaders. In both cases, the leader is
supplied by charges induced by the electric field of a storm cloud, and the
leader lifetimes are comparable because they move at approximately the
same velocity.
Qualitatively, the current variation of the first component leader is
similar to that of a laboratory spark. When the gap voltage is raised
slowly, one can observe initial leader flashes at the high-voltage electrode,
followed by distinct current impulses [32]. As for long spark steps, they
practically do not change the current at the leader base. It has been shown

[20] that this is to be expected if the charge perturbation region is separated

from the registration point by an extended channel section with high
resistivity and distributed capacitance (section 4.4). The perturbation wave
travelling along the channel towards the detector is attenuated. Of course,
the current of a laboratory spark rarely exceeds a few amperes, but such a
difference is predictable. it follows from the expression for the current
i = T V , cited above. A lightning leader has an order higher velocity
V , and, at least, an order larger linear charge T (due to the voltage being
10-20 times higher). All in all, this increases the current to within the
anticipated two orders of magnitude.
Now one can judge about the current of a dart leader. There are no
direct registrations of this current. One exception was an attempt at its
measurement in a triggered lightning just before its contact with the earth.
This is principally possible since the point of contact is known exactly -
this is the point of wire fixation to the earth. The wire evaporates completely,
having passed the current of the first lightning component. Using the still hot
trace channel, a dart leader follows the path of the wire. A current detector
can be placed at the wire grounding site.
It is much more difficult to interpret the recorded oscillograms, because
it is unclear at what moment of time the development of a dart leader stops
and the return stroke with the high current begins. Nevertheless, the
published current measurements vary from 0.1 to 6kA with the average
value of 1.7 kA [33]. The lower limit of the range is more typical for the
first component (this may be the next component, too, but after a long
current-free pause, when the previous trace channel has nearly totally
decayed). The value of several kiloamperes seems reasonable, since the
velocity of a dart leader is 30-50 times higher than that of the first
component.
3.5 Field variation at the leader stage
The subdivision of experimental data between this and the previous section is
somewhat arbitrary. Electric field measurements provide information about
leader charge, while charge and current are related by leader velocity. On the
whole, this is a general problem. If the observations were arranged properly
and the data analysis was made carefully, relatively simple field measure-
ments can add much to our knowledge of electrical parameters of lightning.
The knowledge of the field itself is rarely of importance, probably, except in
some applied problems of lightning protection of low voltage circuits. For
this reason, it is not the measurements but, rather, methodological
approaches to their treatment which are significant. So we shall begin with
these approaches and the general principles underlying a treatment of most
lightning stages.

Field variation at the leader stage 103
-
Figure 3.12. Measurement of fast variations in the electric field during the lightning
development.
Suppose the electric field at an observation point near the earth’s surface
is Eo(0) at the moment of the lightning start. When the discharge is com-
pleted, the field takes a new value Eo(tm).The field has changed by the
value AEo = Eo(tm)- Eo(0)over the time t,. When the measurement time
is relatively short, it is more convenient to register the field change rather
than to measure its values. This is usually done with electrostatic antennas,
i.e., metallic conductors (normally, flat) grounded through a reservoir capa-
citor C (figure 3.12). If the capacitor and the measurement circuits connected
to it have an infinitely high leakage resistance RI,the capacitor voltage, at
any moment of time, is
(3.3)
where Sa is the area of a flat antenna and qc is the charge induced on it. If RI
is finite (which is always the case due to the input resistance of the circuit
taking voltage readings from the capacitor), the use of (3.3 ) requires the con-
dition RIC >> t,, which can be easily met for the lightning duration ~ 1 0 - s*
but becomes problematic for a time interval of several minutes between two
flashes.
An accurate measurement of the field change AEo(t,) requires the
necessary time constant of the measurement circuit RIC and the account of
effects of external field variation in the atmosphere, Eo, by making allowance
for local effects. (The antenna may be raised above the earth, say, mounted
on a building roof, so that the field there will be higher than on the earth. On
the other hand, a nearby high construction may reduce the field, acting as an
electrostatic screen.) The field value obtained is not particularly informative.
In order to get information about the lightning discharge, we have to make
certain assumptions concerning the distribution of charges which have
changed the field.
Let us begin with a simple illustration. Suppose a lightning leader
passing from a spherical volume has changed the charge of only one sign

in a storm cloud cell. If there are other changes in the sphere charge during
the leader travel, they are assumed to have been completely neutralized later,
at the return stroke stage. If this assumption is correct, the measured value of
A E o ( t m )can give an idea about the quantity of charge transported by the
leader from the cloud to the earth:
+
2 7 r ~ o ( H ~R2)3/2AEo(tm)
AQM = H (3.4)
Here, H is the altitude of the charged storm centre and R is its radial dis-
placement relative to the registration point. Both parameters should be
measured by an independent method or simultaneous field registrations at
two more points should be made at given distances from the first one. This
will provide additional equations for the unknown values of H and R.
Such an unambiguous treatment results from the simple model we have
chosen, which contains no geometrical parameter except for the distance to
the charge. However, a slightly more complicated, dipole model deprives the
measurement treatment of this advantage. Still, electric field measurements
have always been attractive to lightning researchers owing to their
simplicity. Interest in such measurements increased with the application of
lightning triggering by small rockets raising a grounded wire to 150-300 m
above the earth’s surface (triggered lightning). The first component of such
lightning is genuinely artificial, but then the first trace channel is used by
practically natural dart leaders travelling to the earth. Their point of contact
with the earth is predetermined, so field detectors can be placed at any dis-
tance from the leader. This registration system is quite sensitive and capable
of responding to the linear charge density not far from the leader tip when it
approaches the earth.
To illustrate our analysis, we shall use the field measurements described
in [34,35]. The authors of this work kindly made them available to us after
their discussion at the IXth International Conference on Atmospheric Elec-
tricity, held in St. Petersburg in 1992. The files contained detailed records of
electric fields, taken during the flight of dart leaders, and of their return
stroke currents. Detectors were placed at the distance of R = 500m and
30m from the contact point. Regretfully, the recordings at these distances
were not simultaneous but made in different years. Their comparison is
still possible because the fields were recorded at the same time as the
return stroke currents. By sorting out identical current oscillograms, one
can select lightning discharges with about the same leader tip potentials.
This provides close values of leader velocity and linear charge density in
the charge cover not too far from the leader tip. Some representative oscillo-
grams of AE(t)/AEmax normalized by their amplitudes are shown in
figure 3.13. They correspond to discharges with really close currents in the
return strokes (IM = 6 kA at point R = 500 m and 7 kA at point 30 m). The
amplitude values of field variation AE,,, over the time of the dart leader

Field variation at the leader stage 105
0 100 200 300 400 500
1.0-
I
0.8 -
0.6 -
w
0.4-
10 20 30 40
Time, ps
Figure 3.13. Oscillograms taken in Florida, USA [35], from the vertical field com-
ponent during the development of the dart leader in the subsequent component of
a triggered lightning. The detectors were positioned at 30 and 500 m from the point
of strike; the pulses are related to their maximum amplitudes.
travel were 6.9 V/cm and 120 V/cm, respectively. Note that the measure-
ments in [35] result in AEm,,/IM M const at every point. There is no
geometrical similarity between the pulses AE( t)/AEmax at the different
points. On the contrary, there is a sharp difference in the rates of strength
rise, as a dart leader was approaching the earth. The field increase in the
range (0.5-1.0)AEm,, took Atl12 = 76ps for point R = 500m and only
5 ps for point R = 30 m.
These data will be treated in terms of a simple model, in which a dart
leader is represented as a uniformly charged axis with linear charge density
rL. Naturally, the real cover radius R, can be ignored in the field calculation
at a distance R.We shall show below that field calculations can only take into
account the charge distribution along a relatively short length behind the tip,
comparable with R. This will justify the assumption of rL being constant,
because it actually refers to a short length of about R near the tip. Therefore,
the field change due to the leader charge at point R at the earth, with the
allowance for its mirror reflection by the earth, is described as

where h is the height of the leader tip from the earth at the moment of regis-
tration and H is the height of the leader base. The field change is maximum
when the tip contacts the earth, and for R << H this gives
It indeed follows from (3.6) that only the charge distribution along a short
length comparable with R is important for field evaluation. (For example,
at H > 5R, the error of the model with rL= const will be less than 20%
for any charge distribution, unless TL grows rapidly from the tip toward
the base; but there is no reason for this, because the channel field E, is
weak and the cloud potential does not vary much.)
Formula (3.6) allows charge density evaluation with a good accuracy,
since the leader is strictly vertical at the earth - it reproduces the path of
the rocket taking up the wire which has evaporated. The value calculated
from the measurements at point R = 30m appears to be unexpectedly
small: rLx 2 x lop5Cjm. Nearly as much charge is transported by long
laboratory sparks (section 2.4). The potential of a lightning leader tip,
U,, does not seem to be much larger than that of a laboratory spark.
According to (2.8) and (2.35), the linear leader capacitance is
C1 FZ 2mo/ln ( H / R L )x (2-10) x F/m, even with indefinite leader
radius R L . From this, we have U, x rL/C1x 2-10MV.
The velocity of a dart leader proves to be very high. For its evaluation, we
shall use the measured value of A t l p ,which is 5 ps for R = 30 m. Formula (3.5)
gives A E = AE,,,/2 at h = J?;R. Hence, the average velocity along a path of
length h FZ 50m at the earth’s surface is VL FZ f i R / A t 1 / 2= lo7 m/s, quite
consistent with direct measurements. It should be emphasized that this velocity
refers to the perfectly vertical path at the earth’s surface, so it is the true velocity.
Similar evaluations can be made with the measurements at the far point
R = 500 m but with a lower reliability, since the parameter averaging is to be
made over a leader length of about lo3m with an unknown path. Neverthe-
less, the values of T~ = 2.3 x Cjm and VL = 1.15 x lo7 mjs are found to
be close to those above. It will be shown in the next section that an indefinite
trajectory may produce an error much larger than the obtained difference in
the values of TL and VL. So the dart leader of triggered lightning with the
definite path at the earth is a lucky exception.
Another illustration of A E ( t ) ,cited in [35], characterizes a more power-
ful dart leader. The current amplitude in the return stroke was as high as
40 kA. The maximum field change was found to be AE,,, = SlOVjcm,
i.e., a little more than a value proportional to current, while the characteristic
time of the process, A t l p , decreased to 1.8 ps. Calculations similar to those
described above give T~ x 1.35 x 10-4C/m, U, =20-30MV, and
VL = 2.9 x 107m/s, thereby supporting the hypothesis of a direct, though
not very strong, dependence of the leader velocity on the tip potential. For

Perspectives of remote measurements 107
the calculated values of linear charge and velocity at the earth, the leader cur-
rent is found to be iL = T~ VL = 3.9 kA, only an order of magnitude lower
than the current amplitude in the return stroke.
3.6 Perspectives of remote measurements
What we described in the previous section is a very favourable situation, in

which the point of leader contact with the earth is fixed and its final path
is strictly vertical, at least, at a length of 150-300m above the earth. One
should not expect such favourable conditions for natural lightnings, espe-
cially for their first components. Still, one should take quietly and with
some scepticism the idea of indirect remote measurements of lightning
parameters. The experimeter resorts to them because, otherwise, his life
would turn out too short to bring his experiment to a conclusion. Recon-
struction of an electromagnetic field source from strength measurements
made at definite points is an incorrect solution to a fairly common problem
of electrodynamics in various areas of science and technology. Lightning is
not an exception to the rule. We shall consider critically the treatments of
results obtained from solutions to such problems and discuss inverse electro-
static problems, as applied to the lightning leader.
Generally, the density of space charge p ( x , y . z ) between some boundary
surfaces can be found if the electric field in the whole confined volume is
known. Experimentally, this means simultaneous field measurements at an
infinitely large number of points, which is practically unfeasible. A well
organized service for field lightning observation has, at best, several synchro-
nized field detectors. A theoretical treatment of the field records always
suggests an a priori construction of a simplified field source model. The
inverse problem can be solved if the number of unknown parameters in
this model does not exceed the number of registration points. What follows
is quite obvious. One writes down a set of equations with the measurements
on the right and the expression for field at a given point (derived from the
model with yet unknown charge parameters) on the left. The solution defines
the parameters as rigorously as the measurements permit. One should always
remember, however, what has been found from the equations, since these are
parameters of a speculative model rather than a real phenomenon. How
much they coincide is not a matter of accuracy of measurements or calcula-
tions but that of the model adequacy ‘to the phenomenon under study. Most
often, it is here that possible errors originate.
3.6.1 Effect of the leader shape

Without claiming a general analysis, we shall consider a special but fre-
quently used model of near-earth field variation at a large distance from a

<” 0.00-
w
U
z
WO
e
-
-0.02
N
I I
-0.04-
-0.06-
.
Figure 3.14. Electric field variation at the earth, evaluated for a large distance from
the vertical channel of a descending leader.
descending lightning leader. In this model, the leader is represented by a thin

vertical uniformly charged thread with a linear charge density rL,and the
storm cell, from which the leader started, is taken to be so small that it is
replaced by a point charge Q at height H . The value of Q may be unknown,
because the analysis uses the time variation of the field rather than its
absolute value [l].
The field at point R near the earth (figure 3.14(a)) varies in time for two
reasons. The absolute field decreases because of the charge reduction in the
storm cell by the charge AQ = TLL carried away by the leader on its way
to the earth.t The second component is due to the charge accumulation on
the leader of length L ; as the leader moves on, this charge goes down,
enhancing the field at the earth. As a result, for the change of the vertical
field component at moment t with L = VLt, where VL is average leader
velocity, we have
AE(L)= -
AQH
+
27rq,(H2 R2)312 2%
+‘“I (H-x)dx
o [ ( H- x)’ + R2I3/’’ (3.7)
This expression takes into account the doubled field associated with the
earth-induced charge. The evaluation of the integral gives the known
expression
AE(L)= -
[ ( H- L)’
1 1
+ R2I1/’ - (H’ + R’)”’
-
(H‘
LH
+ R2)3/2
}. (3.8)
t This component may be ignored in field measurements at a small distance from the leader, as
described in section 3.5.

When analysing this expression, an experimenter will find it difficult to avoid

a temptation. In the range of R / H < 1.4, the function AE(L) has an
extremum (figure 3.14(b)) which can be easily recorded by an oscillogram.
What remains to be done is to find the height H a t which the lightning started
(it may be taken to be equal to the average height of the storm front), to
measure the distance between the observation point and the leader (e.g.,
from the thunder peal delay time, since the sound velocity is known and
the point of sound wave excitation is near the earth's surface), and to find
the moment of time when the leader tip descended to the height H - L,
by calculation, from (3.8), the leader length L, corresponding to the
extrema1 point. At least one more moment of time is registered exactly by
the oscillogram - the moment of the leader contact with the earth, giving
rise to the return stroke. The oscillogram indicates this moment by a
field strength overshoot. The time interval At in figure 3.14(b) defines the
leader average velocity along the path length H - L, at the earth:
VL M ( H - L,)/At. Note that one of the boundaries of the measured
length might also be found from the moment of sign reversal of the field
being registered. It follows from the analysis of (3.8) that the curve A E ( L )
intercepts the abscissa if the registration point lies at a distance R z (0.8-
1.4)H from the vertical path axis. Technically, the reference point is easier
to find than the extremum.
It is known that the appetite comes with eating. If one substitutes the
geometrical parameters used and the measured values of A E ( L ) into (3.8),
one can find the average linear charge density T ~Together
. with the velocity,
this provides the average current in the leader for the final period of time
At: iL M T ~ F ' ~The
. calculation of charge density was replaced in [36] by
graphical differentiation of the oscillogram E( t ) at the point corresponding
to the moment of leader contact with the earth; this, however, gave a
low accuracy. Therefore, the electric field registration only at one point on
the earth's surface seemed to be sufficient to evaluate one of the least
accessible parameters - the leader current in a descending lightning
discharge.
Let us now try to assess this situation without considering the measure-
ment errors. Obviously, the main error is associated with finding the starting
point of a lightning spark and the distance to it. A common 10% error in
measurements gives much larger errors in evaluations of leader velocity
and current. This always happens when one deals with the difference of
two comparable parameters. We shall focus on errors of the model itself.
The problem of leader branching effects will be ignored. After all, one can
always consider a dart leader which has no branches. The representation
of a real leader as a vertical axis is quite another matter. Any photograph
shows numerous bendings of a lightning trajectory, so a straight vertical
leader is nothing more than a speculative mathematical concept. To assess
its implications, let us make another step and consider a tilted straight

0.054
L/H
Figure 3.15. Electric field variation at the earth for a leader deviating from its vertical
path; R I H = 0.7.
leader. Suppose a leader moves in a vertical plane passing through a field

detector and is tilted towards it by the angle Q relative to the vertical line
(figure 3.15). Then, instead of (3.8), we have
where a = R - H tan a. The result of numerical integration of (3.9) is

presented in figure 3.15 for R = 0.7H. Even a slight tilt from the vertical
line ( a zz 20") entails a nearly three-fold (two-fold for R / H = 0.7) change
in the pulse amplitude A E ( L ) and in the parameters usually derived from
field measurements at the earth's surface. The value of L corresponding to
the extremum A E ( L ) depends only slightly on the tilt, but the signal ampli-
tude variation is sufficient to make one treat the derived leader parameters
only as estimations of orders of magnitude. It is quite another matter if the
leader shape is recorded simultaneously with electric field registration at
two points. The leader trajectory can then be reconstructed in space quite
accurately, and a computer processing of records can completely eliminate
this type of error.

Perspectiws of remote measurements 111
3.6.2 Effect of linear charge distribution

The model of a leader with a uniform charge distribution and the concept of a
storm cell as an electrode with a capacitor battery supplying conduction
current to the leader are extremely far from reality. A storm cloud does
not look like a giant capacitor plate, to which a lightning leader is connected
galvanically during its motion to the ‘plate’ of opposite sign, i.e., to the earth.
In actual reality, the cloud charge is concentrated on hydrometers which do
not contact one another and their assemblage does not possess the properties
of a metallic electrode. A better analogy would be that of an electrode-free
spark, rather than of a spark starting from a high voltage electrode of a
laboratory generator.
To illustrate this, consider a small metallic rod suspended along the
field vector in an inter-electrode gap, where the field is supported by a
high-voltage generator. The rod has no contact with the generator poles.
Two sparks of opposite sign are excited simultaneously at the rod ends,
i.e., in the region of enhanced local field (figure 3.16). The charges appearing
on the sparks must be regarded as polarization charges. This is the way a
Figure 3.16. Streak photographs of a simultaneous development of a positive and a

negative leader from the ends of a metallic rod of 50 cm in length in a uniform electric
field: (1) rod; (2,3) channel and streamer zone of a positive leader; (4,5)the same for a
negative leader.

metallic conductor is polarized when it is introduced into an electric field.

Similarly, a charged storm cloud possessing no conductivity is only a field
source in the space extending to the earth. A plasma conductor arising in
this way or other is polarized in the field and grows, being supplied by polar-
ization current. This system is definitely not a perpetuum mobile. Its energy
source is the electric energy of the cloud field. As the leader develops, this
energy decreases, in accordance with the conservation law. If the external
field and the conductor are homogeneous, the linear density of polarization
charge is equal to zero exactly at the conductor centre and its absolute value
rises towards the ends of different polarities. As long as the conductor has no
contact with the high-voltage generator terminals, its total charge, naturally,
remains equal to zero. The latter is also valid when the field and conductor
are inhomogeneous. Using numerical methods, one can find the polarization
charge distribution for any electric field. The distributions presented in
figure 3.17 have been found by the equivalent charge method [37]. This
method is simple and convenient for long conductors, like those used to
simulate lightning leaders.
Numerical computations show that a uniform field in a perfectly con-
ducting rod creates a polarization charge 7 rising almost strictly linearly
from the rod centre towards its ends. The ends are an exception, because
Figure 3.17. Polarization charge distribution along a straight conductor (a leader

system) in the cloud dipole field, with allowance for a dipole reflection in the conduct-
ing earth.

the charge density here (along a length of the rod radius) rises rapidly. The
charge distribution at the rod base can be approximated as T ( X ) = ax,
where the coordinate origin x is at the rod centre. The reader will soon see
that the contribution of the end charges fq is small at larger distances, if
the rod length 2d is much greater than its radius r.
For simplicity, let us assume that the charge is concentrated on the outer
rod surface, as is the case when it has a conductivity, and that the potential
will be calculated along the longitudinal axis. The rod centre will be taken as
the zero point of the external field Eo potential. The potential p at the point x
is a sum of potentials created by the external field, -Eox, the end charges, pq,
and the charges distributed along the rod, pr:
- + + r2]1/2 + x In
[(d x12
(d - x )
-(d
+ [(a- x12 + r2]'12
+ x ) + [(d + + r 2 p 2
X)* 1
(3.10)
Here, the last approximate expression refers to the rod sites lying far from its
ends, Id f X I >> r. Here, the term pqcan be neglected, and we shall approxi-
mately have pT - Eox = 0 . With the actual charge distribution ~ ( xand ) the
end charges providing pq, the rigorous equality p ( x ) = 0 should be valid
along the whole rod length. By relating the approximate equality to the
centres of the semi-axes x = f d / 2 , we find
(3.11)
The potential at the rod ends must be calculated with the account of
their higher charges. Assuming this charge to be concentrated along the
end circumference, the potential at the centre of the end plane (at the
points x = f d on the axis) can be described as
(3.12)
The potential pr must now be calculated from the unsimplified formula

(3.10). It follows from (3.12) that the end charge is approximately equal to
q x 27r~orEodand by a factor of
ad2 d
K=-NN (3.13)
2q 2r In (&/er)

smaller than the charge distributed over each half (it is an order of magnitude
smaller for d / r z 100). Therefore, the account of localized tip charge may be
necessary only for the calculation of electric field in the region close to the tip,
at a distance less than 10r from it, In a remote region, where measurements
are usually made, such a subtlety is unnecessary - it is sufficient to consider
only the charge distribution along the leader channel. Clearly, this is not a
uniform distribution, taken for granted by some researchers.
It is time to look at the shape of a field strength pulse at the earth, deter-
mined by the charge of a linearly polarized vertical axis with charge
=ra
~ ( x=) f a x per unit length. It is defined by the algebraic sum of terms
from the positively and negatively charged semi-axes and is equal to
AE(L) x ( H - X) dx
-L 2 r E o [ ( H- x)’ + R2I3/’
where L are the lengths of leader sections which have moved away from the
starting point to the earth and upwards. Integration with (3.1 1) gives
L L
AE(L) =
In (fiL/er)
Eo [ [ ( H - L)’ + +
R2]‘/’- [ ( H L)’ + R2I1/’
- In
+
[H (H’ R2)”’]2 +
{H -L + +
[ ( H- L)* R2I1/’}{H L [ ( H L)’+ + + + R2]’/’J
(3.14)
Here, H is the height of the leader start, r is its radius, and R is the
distance between the leader axis and the observation point. It can be
shown that the function A E ( L ) of (3.14) rises smoothly with L and has
no extrema.
The linear charge distribution assumed in the above illustration is, of
course, another speculation (section 4.3). Moreover, a leader goes up and
down non-uniformly, and the field in the earth-cloud gap is far from
being uniform: its strength decreases towards the earth. This limits the
linear charge growth from the start downward. The finite channel conductiv-
ity exhibits similar behaviour, reducing the tip potential. So it is impossible to
find the actual charge distribution exactly without knowing these param-
eters. Thus, a processing of field oscillograms can give nothing more than
what they actually show. The field at a point is an integral effect of the
whole combination of charges created or transported by a given moment
of time. It is probably worth speculating about registrations but one
should assess the results soberly, considering all possible variants and
insuring oneself whenever possible. The best insurance is, of course, to
increase the number of registration points and parameters determined by
independent methods.

Lightning return stroke 115
3.7 Lightning return stroke
All lightning hazards are associated with the return stroke, and this accounts
for the great effort of investigators to learn as much as possible about this
discharge stage. It has been established that the contact of a descending
lightning leader with the earth or a grounding electrode produces a return
wave of current and voltage. It travels up along the leader channel, partially
neutralizing and redistributing the charge accumulated during the leader
development (figure 3.18). The travel is accompanied by an increased light
intensity of the channel, especially at the wave front. At the earth, the
wave front intensity acquires its maximum over 3-4ps [31]. As the wave
goes up to the cloud, the wave intensity steepness and amplitude decrease
many-fold, indicating a considerable decay. Judging by streak pictures, the
region of a high light intensity at the wave front extends to 25-1 10m. The
whole wave travel takes 30-5Ops. This time is especially convenient for
electron-optical methods of streak photography. However, available
attempts to use such methods can hardly be considered successful. A serious
obstacle is the exact synchronization of a streak camera and lightning contact
with the earth. Although there are many synchronization methods, they have
no simple technical solutions and are seldom used in lightning experiments.
Continuous (e.g. sinusoidal) electron streak photography has not justified
hopes. Basic results on return stroke velocities have been obtained using
cameras with a mechanical image processing, which do not need synchroni-
zation (Boyce camera).
Figure 3.18. Scheme of the return stroke propagation after the contact of a descend-
ing leader with the earth (at moment t = 0). A leader brings potential U < 0; ZM is
return stroke current.

3.7.1 Neutralization wave velocity

The measurements made half a century ago [25,39] and those performed
recently [40] indicate a high velocity of a return current-voltage wave. The
minimum measured values are close to (1.5-2) x lo7 mjs and the maximum
ones are an order of magnitude higher, reaching 0.5-0.8 of light speed c. A
velocity comparable with light speed does not mean that we deal with
relativistic particles or purely electromagnetic perturbations. The wave
velocity is the phase velocity of the process.
There are not so many successful optical registrations of the return stroke,
the number of really good ones being about 100. Most of the available data
concern subsequent lightning components. This is natural because every
successfully registered discharge includes the return strokes of several compo-
nents. The wave velocities of subsequent components are somewhat higher
than those of the first ones. According to [40], the first component has an aver-
age velocity V, x 9.6 x lo7mjs while the subsequent ones are a factor of 1.25
higher. Similar data are cited by other authors for subsequent components of
lightning discharges triggered from a grounded wire elevated by a rocket.
To illustrate the statistical velocity spread in individual measurements
and in those made by different researchers, figure 3.19 shows integral
distribution curves for the data of [25] and [40]. The first and subsequent
1.0 -
0.8 -
-3
.-O 0.6 -
cd
D
& 0.4 -
0.2 -
0.01 - . * . I 1
0.05 0.1 0.2 0.5 1

v,ic
Figure 3.19. Velocity distribution of the lightning return stroke: (1) averaged over the
visible channel length [25];(2) averaged over 1.3 km above the earth [40].

components were not separated. Within a 50% probability, there is a 2-fold

difference between the velocities. Earlier measurements generally give lower
return stroke velocities. The point is that most measurements performed
during the 1980s were two-dimensional, usually providing higher velocities,
whereas the earlier data had allowed conclusions only about the vertical
component of velocity. Moreover, the application of improved optics and
photographic materials, as well as higher relative motion rates of the
image and film, improved the time resolution of streak photographs. As a
result, the velocity value obtained in the 1980s was more accurate and
higher because the measurements were averaged over the initial stroke
length of about 1 km at the earth’s surface, where the wave moves 1.5-2
times faster, rather than over the whole stroke length.
All measurements show that the return stroke velocity gradually
decreases and that the velocity V, drops abruptly when the wave front
passes through the point of leader branching. The latter fact suggest a certain
relation between the stroke velocity and the current transported by the wave:
at the branching point, the current is divided among the branches, so the
velocity becomes lower. The knowledge of this relation could improve the
calculation accuracy of overvoltages in electrical circuits during lightning dis-
charges. Unfortunately, the available data are insufficient to allow finding
this relation reliably. Simultaneous registrations of current and velocity
have been made only for return strokes of subsequent components of
triggered lightnings but they cannot provide a representative statistics.
With reference to [12,41], there is note in [l] about a satisfactory agreement
between these registrations and Lundholm’s semi-empirical formula
V,/c = (1 + 40/1M)-1’2, where lMis a return stroke current amplitude
expressed in kA (see section 3.7.2). The lack of factual data is sometimes
compensated by a superposition of distribution statistics. It is assumed
that the values of current and velocity characterized by an equal probability
correspond to each other. There are no serious arguments in favour of this
operation but it is used for the lack of a better method.
3.7.2 Current amplitude

The current amplitude is an important lightning parameter. Most hazards of
lightning are associated, directly or indirectly, with stroke currents, whose
registration has taken much time and effort. Very few of them were made
by direct methods, using a shunt and a Rogovski belt [28,29,42-461. Still
fewer direct measurements were made by equipment with a wide dynamic
range, which can register both powerful impulses with an amplitude to
200 kA and low currents of a few hundreds of amperes, which are equally
important for the understanding of the lightning physics.
A large number of measurements have been made by magnetic detec-
tors. Such a detector represents a rod several centimetres in length, made

from magnetically hard steel. Preliminarily demagnetized rod detectors were

placed at a fixed distance from a conductor aimed at leading lightning current
to the earth. This could be a grounding lead of a lightning conductor or a
metallic tower of a power transmission line. With the appearance of lightning
current, the detector proves to be within the range of its magnetic action and
becomes magnetized. One measures the residual steel magnetization and
calculates the current by solving the inverse problem. The advantages of
this method are its simplicity and low cost. Usually, magnetic detectors are
installed by the thousand to obtain the necessary statistics. However, they
can yield nothing else but a current impulse amplitude. Of course, by mark-
ing the ends of the detector, one can also determine the direction of current
and attribute it to the lightning type (positive or negative). The accuracy of
current measurements is very low for several reasons.
First, there are few objects with a simple system of current spread over
metallic constructions. A single conductor would be ideal in this respect,
because it excludes current branching. In reality, lightning current is distribu-
ted among many conductors, the distribution pattern being unpredictable
since it varies with temporal parameters of the impulse. We shall illustrate
this situation with reference to a simple system consisting of two parallel induc-
tively connected branches with their own inductances L1 and L2, mutual
inductance M , and resistances R1 and R2. Suppose a rectangular current
impulse I with a short risetime is applied to the system. The current distribu-
tion between the two branches is described as
Initial currents ilo and i20 at the stage when current I ( t ) is stabilized to Z
are generated over a very short time equal to the I risetime. The branch
currents, therefore, rise from zero very quickly. The reactive components of
voltage drop -di/dt produced by them are much larger than the ohmic
ones -i that can be neglected for the time being. Hence, we have
i1o/i20 = (L2- M ) / ( L 1- M ) , and the initial current, say, in the first branch
is ilo = Z(L2- M ) / ( L 1+ L2 - 2 M ) . When the transitional process, whose
duration is defined by the time constant At = (Li + L2 - 2 M ) / ( R I+ R 2 ) ,
is over, currents ilx = IR2/(R1+ R2) and i2x = ilxR1/R2 are established
in the circuits. The durations of lightning currents are usually comparable
with the time constant At. Therefore, a magnetic detector placed in one of
the branches will register a current intermediate between the initial and estab-
lished values having a maximum amplitude, since the residual magnetization
of the rod contains information only about the maximum magnetic field of
current. For this reason, one can calibrate a magnetodetector for deriving a
full current amplitude only if the impulse shape is known. This cannot be
done in a real experiment, so one has to resort to a rough estimation of current
distribution over metallic constructions and use it in data processing.

Second, the operating range of the rod magnetization curve is not large,
and the transition from a linear to saturation region may produce additional
errors in data processing. To avoid saturation, the magnetodetector is placed
far from the conductor, which creates difficulties in data processing of light-
nings with low current and low magnetic field. Besides, when the distance
between a conductor and a detector is large, the magnetic field effects of
other metallic elements with current are hard to take into account. So a
100% error does not seem too high for magnetodetectors, even when several
detectors are placed at different distances from a current conductor. Their
records provide sufficient material for engineering estimations or for a
qualitative comparison of storm intensity in different regions, but they are
insufficient for theory. Organization of direct registrations takes much time
and effort. There are no more than a hundred successful registrations
made over a decade. Let us see what information can be derived from them.
Current impulse amplitudes vary widely, from 2-3 to 200-250 kA.
Some magnetodetector measurements give even 300-400 kA, but these
amplitudes seem doubtful. According to [42,46], the integral amplitude dis-
tributions for the first and subsequent lightning components obey the so-
called lognormal law, in which it is current logarithms, rather than currents
themselves, that meet the normal distribution criterion. The probability of
lightning with a current larger than ZM, is defined as
(3.16)
where (lg I)av is an average decimal logarithm of the currents measured and
olg is the mean square deviation of their logarithms. This approximation
cannot be considered accurate. The relative deviation of the value of (3.16)
from the real one may be several tens percent; it may be even more for prac-
tically important current ranges. Nevertheless, lognormal distributions allow
measurement comparison and serve as a guide to engineering estimations.
For example, about 200 current oscillograms for lightnings that struck the
70m tower on the San Salvatore Mount in Switzerland [42] satisfactorily
obey the lognormal law with (lgZ)av= 1.475 and olg= 0.265 for the first
component currents of a negative lightning discharge. This means that the
50% current value is estimated to be 30 kA; 95% of lightnings must have
currents exceeding 4kA and 5% of lightnings 80kA. The probability of
higher currents rapidly decreases: 100 kA is expected in 2% of cases and
200kA in less than 0.1% of cases (figure 3.20). It should be emphasized
again that the distribution boundaries must be treated with caution. The
curve shape in the low current range strongly depends on the sensitivity of
the measuring instruments used (its left-hand limit is usually taken to be
1-3 kA in distribution plots). There are few measurements in the high current

0.01
I ' I . I . I . 1
50 100 150 200 250

Lightning current, kA
Figure 3.20. Lognormal distributions of return stroke currents: A, the first com-
ponent of a negative lightning with (lg I)," = 1.475 and clg= 0.265; B, subsequent
components with (lgZ),v = 1.1 and olg= 0.3; C, positive lightnings with
(lgZ)," = 1.54 and clg= 0.7.
range: it is considered as good luck if they provide a reliable order of magni-

tude. Note that negative lightning currents above 200 kA have never been
registered reliably.
The approximation of data on subsequent lightning components in [42]
gives a much lower integral probability for high currents. A lognormal distri-
bution can be satisfactorily described by (lg& = 1.1 and clg= 0.3. The
calculated 50% current is 12.5kA, 5% current is only 39kA, and the
chance for a subsequent component to exceed lOOkA is close to 0.1%
(figure 3.20).
The statistics for positive lightnings, whose number is about 10% of the
total registrations, is less representative. All descending positive lightnings
are one-component. The integral current distribution for them has a large
spread. The probabilities of both low and high currents are larger without
an essential change of the 50% value. The 50% value is close to 35 kA,
i.e., it is nearly the same as for the first component of negative lightning.
An approximate description of the lognormal distribution of positive cur-
rents in [42] can be made with (lgI)av= 1.54 and clg= 0.7 (figure 3.20).
Positive high current lightnings are more frequent than negative ones. A
5% probability corresponds to 250 kA, and 100 kA can be expected with a

20% probability. Among 26 successful registrations of positive lightnings in

[43], only one showed 300 kA current. It seems likely that many positive light-
nings were, under the observation conditions [43], ascending ones, which
may account for the large spread. Such lightnings have practically no
return stroke, and the equipment seems to have registered the relatively
low leader current of the final development stage. These data were used to
derive the integral distribution extending to the low current region.
The great importance of lightning current statistics to applied lightning
protection necessitated a unification of theoretical distribution curves. Other-
wise, engineers would have been unable to compare the frequency of harmful
lightning effects and protection efficiency. This work is being done within the
frame of the CIGRE (Conference Internationale des Grands Reseaux
Electrique a haute tension) - an operating international conference on
high-voltage networks. Data on current from all over the globe are collected
and analysed. However, there is no unified approach to these data: different
data are discarded for different reasons, so that the distributions obtained
differ markedly. For example, a report submitted to [47] compares two log-
normal laws with (lg&" = 1.4 and 1.477al, = 0.39 and 0.32. The latter is
preferable for power transmission lines, since the measurements for objects
higher than 60 m were excluded from this derivation (power transmission
lines are usually lower). In his book, Uman [l] gives a table of lightning
currents mostly based on the measurements of [42].
Attention to details is inevitable, since slight corrections in parameter
distributions may cause manifold changes in the calculated probabilities of
currents above 100 kA, especially important in lightning protection of
important objects. Both theory and applications suffer from a lack of light-
ning current measurements. We shall list here some key issues to be discussed
in more detail below.
We have mentioned the importance of an object's height. It has been
known since Benjamin Franklin's experiments that h g h constructions
attract more lightnings. It seems likely that the process of attraction depends
on the potential of a descending leader. If this is so, the statistics of descend-
ing leader currents for objects of various height may prove different: there
will be a kind of lightning separation, In this case, a comparison of reliable
current statistics for various objects could help resolve the much debated
problem of lightning-object interaction mechanism.
Of interest in this connection is the following fact. In the case of a very
high construction, many first components are of the ascending type having
no return stroke. But the first component is followed by subsequent descend-
ing components, whose average stroke currents are lower than in subsequent
components affecting low buildings. This suggests an involvement of high
ground constructions in the formation of storm clouds. It appears that an
ascending leader starts from a high construction before the cloud has
matured. Its charge and potential are, therefore, lower. This accounts for

the fact that subsequent components discharging an immature cloud will

transport lower stroke currents than a mature cloud.
Finally, an important issue is the effect of grounding resistance of objects
on lightning current. This may provide information on the resistance of
lightning itself. This resistance is to be introduced in equivalent circuits,
when calculating overvoltages affecting various electrical circuits. This
problem is still much debated: some investigators suggest the substitution of
a lightning channel by a current source with an ‘infinite’ resistance, others
ascribe to the channel the wave resistance of a common wire (about 300 0).
It would not be hard to solve this problem if we had at our disposal reliable cur-
rent statistics for objects of various height but different grounding resistances.
No such statistics exist yet. To speed up the work in this area and to reduce its
cost, various remote registration techniques are being employed. They register
electromagneticfields and coordinates of points where the lightning strikes (ide-
ally, the lightning trajectory), followed by the solution of the inverse problem
for the field source, i.e., lightning current (see section 3.7.4).
There is also an increasing number of direct current registrations from
lightnings triggered from a wire lifted by a rocket to the height of 150-
250m. The first component of a triggered lightning (ascending leader) has
no return stroke; therefore, one deals only with subsequent components. A
comparison of such registrations with natural lightning currents was made
in Alabama, USA [15]. The statistics were not particularly representative
(45 measurements), so no principal differences were revealed. The lognormal
distribution of currents corresponded to the parameters (lg& = 1.08 and
olg = 0.28, nearly the same as those obtained in Switzerland for subsequent
components of natural lightnings [42]. We should like to warn the reader
against a possible overestimation of this coincidence. The comparison
involved measurements from geographical points separated by large dis-
tances, whereas the global variation of lightning parameters still remains
unclear. A more important thing is that the lightnings studied in [42]
cannot be regarded as totally natural. They struck a 70-m tower on a
mountain elevated at 600m above the earth’s surface close to a lake. The
conditions here are more similar to those of lightning triggering than to its
natural development in a flat country. Lightning parameters are known to
differ with altitude: currents registered by magnetodetectors at an altitude
of 1-2km were two times lower than in a flat country, for less than 50%
probabilities [48].
3.7.3 Current impulse shape and time characteristics

Records of lightning current impulses look more like abstractionists’
pictures - they are so diverse and fanciful. The conventional approximation
of a impulse by two exponents Z ( t ) = Io[exp(-at) - exp ,I)@-(
which is
suggested in various guides to equipment testing lightning resistance, is

Figure 3.21. A schematic oscillogram of a current impulse in the first component of a

negative lightning.
intended for currents of laboratory sources simulating lightning, rather than

for natural lightning. Let us try to identify the main features of the time
variation of current, essential for the understanding of the return stroke
mechanism and applications.
The most reliable data have been obtained for first component currents
of a negative lightning. This current is easy to register, since the impulse front
takes several microseconds and an oscillographic record reproduces it in
detail. A sketch of a current impulse averaged over many oscillograms is
shown in figure 3.21 in two time scales. Note the concave shape of the
front. An expression of the type 1 - exp (-pt) looks least suitable for its
description. The first current peak is often followed by a higher one, and
evaluation of the impulse risetime tf is associated with some reservations.
For example, [42] measured the time of current rise from 2kA, a value
close to the resolution threshold, to the first maximum .,Z In this case,
about 50% of negative lightnings had the risetime of the first component
over 5.5 ps, 5% exceeded 18 ys, and another 5% less than 1.8 1s. The knowl-
edge of the risetime allows calculation of the average impulse slope
AI = Z,/tf, However, the calculation of electromagnetic fields of lightning
and the evaluation of possible hazards require a maximum slope
AI,,, = (dZ/dt),,,, rather than an average one. The error in evaluations
of this parameter from current oscillograms may be very large, because
one has to replace the tangent to the Z ( t ) curve by a secant. Nevertheless,
this operation has a sense for a fairly long impulse of the first component.
The integral distribution of the values, like the current itself, is described
by the lognormal law with the parameters (lgZ)," = 1.1 and glg= 0.255, if
the slope is expressed in kA/ps. It results in 12 kA/ps for 50% current, and
the slope exceeds 33 kA/ys with a 5% probability.
To describe the electromagnetic effect of lightning current, let us find
the induced emf U , in a frame of area S = 1 m2 placed at distance D = 1 m
from the channel or a grounding conductor, when the first component
current flows through it (the frame is in a plane normal to the current
magnetic field). Even for a moderate steepness AI = 33 kA/ps, we have
U , = p0AImaxS(2~D)-' = 6.6 kV, where p0 = 47r x lo-' Hjm is vacuum

magnetic permeability. The role of a frame can be performed by any metallic

structure within the construction affected by lightning-wires,wall fittings, rails,
metallic stripes touching each other, etc. At the site of a poor contact, induced
emf will produce a spark, much more effective than that in an electric lighter.
This is dangerous because the spark may come in contact with an explosive gas
mixture.
Sometimes, lightning current behaves in a dual way, creating the induc-
tion emf and voltage at the resistance of the grounding electrode U , = IR. It
is important, therefore, to have knowledge about the relation between the
current amplitude and maximum slope. Although both parameters obey
the same lognormal law, no correlation has been found between them.
This is bad for engineering applications, for one has to calculate the
probabilities of each current with a whole set of possible slopes.
There have been attempts at a more detailed description of the current
impulse front. They were initiated by the CIGRE mentioned above to
handle hazardous effects of lightning on power transmission lines. A set of
additional parameters has been suggested to reduce errors in current oscillo-
gram processing and some regions of the impulse front have been described
quantitatively. This is illustrated in figure 3.22 and requires no comment.
Some of the results were cited in 1471. The processing technique used did
not lead to considerable data refinement, since the 50% maximum slope
AI,,, = 12kA/ps is only two times larger than the 50% average slope
A~o% ~ 30/5.5 = 5.5 kA/ps. But the factor of 2 is essential to
= Z 5 o y O / t f 5 o=
the electrical strength of ultrahigh voltage insulation.
-I Eo -
Figure 3.22. The distribution of current impulse parameters in the return stroke,
based on oscillograms.

Current impulses of subsequent components have a shorter risetime. In

the work cited above [42], 9 < 1.1 ps for a 50% probability and 0.2 ps for a
5% probability. The latter should be treated with caution because this value
is close to the resolution limit of the measuring equipment. The impulse front
in subsequent components is likely to rise faster. It is mentioned in [l] with
reference to other publications that in many digital registrations the
current could rise to a maximum during the first detector reading (for 0.2 ps).
The maximum slope of a impulse front in subsequent components
obeys, in the first approximation, the lognormal law: (lg AI,,),,, = 1.6
and glg= 0.35. The value of AI exceeds 40 kA/ps with a 50% probability
and is larger than 120 kA/ps with a 5% probability. When affected by such a
steep impulse, the amplitude of induced voltage would exceed 25 kV in the
above example of a frame.
Current of positive lightnings rises slowly. In 5 % of cases, the front
duration was 9 > 200ps. With these impulses, the electric strength of air
gaps of several metres in length is close to a minimum (section 2.6, formula
(2.51)). The voltage with tr 200ps is much more dangerous than a
‘common’ lightning overvoltage impulse with a risetime of several micro-
seconds. Minimum breakdown voltage in air gaps with a sharply non-
uniform field (see formula (2.52)) is about 1.5 times lower than in a standard
lightning overvoltage impulse of 1.2/50 ps (in accordance with the con-
ventional way of presenting time characteristics of a impulse, 1.2 is the
risetime and 50 is the impulse duration at 0.5 amplitude, all in ps).
The duration of a current impulse is as important for lightning protec-
tion practice as the risetime. Impulse duration is usually characterized as a
time span between its beginning and the moment its amplitude decreases
by half, Since current is related to the neutralization wave travelling along
the channel, the impulse duration t, is comparable with the time of the
wave travel. If its velocity is V, E 108m/s and the average channel length
is 3km, the value of tp will be several tens of microseconds. A similar
value is derived from experimental data. The impulse duration in the first
component of a negative lightning is above 30, 75 and 200 ps for the prob-
abilities 95, 50 and 5%, respectively. For subsequent components, the
impulse is much shorter: 6, 32 and 140 ps for the same probabilities. Positive
lightnings must be longer because most of the positive charge of a storm
cloud is located 2-3km higher than the negative charge. Indeed, tp is
above 230ps with a 50% probability. The shortest durations for positive
lightnings are the same as for the first component of a negative one. ‘Anom-
alously’ long impulses stand out against this background - about 5% of posi-
tive currents decreased to half the amplitude for 2000 ps.
Today, we know nothing about the nature of superlong positive
impulses. One thing is clear: they are unrelated to the wave processes in
the lightning channel. One may suggest that hydrometeor charge is accumu-
lated and descends to the earth due to an ionization process in the positively

charged region of a cloud. But we can only speculate about the nature of this
process producing final current of 100 kA and ask why it is manifested only in
positive lightnings.
3.7.4 Electromagnetic field

Electromagnetic field of lightning is familiar to those leaving a TV or radio
set on during a thunderstorm. Sound and video noises inform about a
storm long before it actually begins. Lightning was the first natural radio
station used by the founders of radio engineering for testing their receivers.
The lightning detector designed by A S Popov in 1885 is still Russia’s
national pride. For many years meteorologists surveyed approaching
storm fronts by registering so-called atmospherics - pulses of electro-
magnetic radiation from lightning discharges occurring hundreds of
kilometres away. In the late 1950s, much interest in atmospherics was due
to the nuclear weapon race: suspiciously similar to radiation pulses from
nuclear explosions, they interfered with the diagnostics of the latter.
It is clear from the foregoing that in a return stroke the charge accumu-
lated by a leader cover varies and is redistributed rapidly along the channel,
producing variation of the static component of the electric field. Charge
variation occurs simultaneously with the propagation of a current wave
along the channel, inducing a magnetic field. The induction emf varying in
time gives rise to an induction component of the electric field. Finally,
variation in the current dipole moment (a leader channel can be regarded
as a dipole, with the account of its mirror reflection by the earth) gives rise
to an electromagnetic wave producing a radiation component of the electric
field with a concurrent magnetic radiation component. There is another mag-
netic component - a magnetostatic one proportional directly to current.
It is common practice to distinguish between the near and far regions of
electromagnetic radiation. In the near region, static field components may be
dominant: the electric component, damped in proportion to the cubic
distance Y to the dipole centre, and the magnetic component, varying with
distance as F2.These can be neglected for the far region, because they are
much smaller than the radiation components E , H cz Y - ’ . Now, after these
preliminary remarks, we shall turn to experimental data showing how
much the shape of a registered pulse varies with distance between a lightning
discharge and a field detector.
The shapes of return stroke radiation pulses are shown schematically in
figure 3.23 for the near and far regions. At large distances, where the static
components of magnetic and electric fields are nearly completely damped,
the pulses E( t ) and H ( t ) become geometrically similar. Both are bipolar
and have a high front slope, a well defined initial maximum and several
smaller ones along the slowly falling pulse slope, producing the effect of
damping oscillations. Note that the oscillation period is smaller than the

Figure 3.23. Schematic oscillograms of electromagnetic pulses of lightning in the near

(top) and far (bottom) zones at the distances 2 km (top) and 100 km (bottom).
double time of the wave run along the channel. After passing the zero point,
the pulse part opposite in sign rises and then decreases with nearly the same
rate; its amplitude is 2-3 times smaller that the first ‘half period’.
The inverse proportionality of radiation components to the distance
from the radiation source was the reason why measurements are presented
in the above form: they are normalized to the basic distance Ybas = 100 km
as E:,, = 10-5Em,,r with r in metres. For the first lightning component,
the average values of the initial pulse peak of the vertical component,
EA,,, lie within 5-10 V/cm [49-541 (compare: radio receivers detect well
signals of lmV/m in the medium bandrange). The electric component of
subsequent lightning components is 1.5-2 times smaller. The spread of
measurements is as large as that of lightning currents. The standard deviation
oE is in the range 35-70% for the first lightning component and 30-80% for

subsequent ones. The horizontal component of magnetic field strength

= (po/~O)-'/2E~ax varies respectively. Magnetic induction
B,,, = poHmax is about lo-* T at a distance of 100 km from the lightning.
The radiation pulse of the first lightning component rises to the initial
peak with an increasing rate. In oscillogram processing, the risetime is
arbitrarily subdivided into two components: the initial slow one of 3-5 ps
duration and the final fast one taking 1-0.1 ps. The standard deviation is
also large here: 30-40% of the average value for the slow front and
about 50% for the fast front. In the final stage, the signal rises for about
0.5-1.0E~,,. With some reservations, a subdivision into a slow and fast
component can be also made for radiation pulse of the return stroke of
subsequent lightning components. But it would be more correct to consider
that the rise to the initial peak occurs quickly there, for 0.15-0.6 ps. Note that
the risetimes for the first and subsequent components are close to those of
their current impulses in a return stroke.
The moment of sign reversal for radiation pulses of the first components
is delayed, relative to the onset of a return stroke, by 50ps in temperate
latitudes [54] and by 90 ps in the tropics [52]. The sign reversal for subsequent
components occurs by a factor of 1.3- 1.5 earlier. The time for maximum field
to be established after the sign reversal is of the same order of magnitude as
that prior to the reversal.
The radiation components E and H are, naturally, present in the near
region, too, but they are much smaller than the static component. One
exception is the initial moments of time. The initial peaks in oscillograms
E ( t ) and H ( t ) should be attributed to radiation, since the static field
components did not have enough time to reveal themselves. The monotonic
rise of electric field over 20-50ps, the time long enough for the radiation
component to be damped, is nearly totally due to electrostatic effect. The
induced electrostatic field is quite powerful, because the charge accumulated
by the stepwise leader of the first component or by the dart leader of
subsequent components is neutralized during the return stroke. For
example, the electrostatic field changes by several kV/m at the distance of
1 km from the channel lightning during the first 50ps (for the subsequent
component, the signal is 2-3 times lower than for the first one ); a slower
field rise may continue for about 100 ps. All in all, the field of the first light-
ning component is an order of magnitude higher than the initial radiation
rise. With increasing distance r to 15-20km, the radiation component
becomes dominant over the others, and the initial radiation peak becomes
an absolute maximum of the registered signal.
The magnetostatic component in the near region is not so important.
Still, at a distance of 1 km, it contributes as much to the signal as the
radiation component (figure 3.23). The magnetic induction here is as high
as lop5T. The absolute magnetic field maximum is achieved later than the
stroke current peak registered at the earth's surface. This is clear because

Total lightning flash duration and processes in the intercomponent pauses 129
the magnetostatic component is proportional not only to the current but to

the conductor length. The length increases as a neutralization wave travels
from the earth up to the cloud. For the same reason, the times for the first
and subsequent components do not differ much. The duration of pulse
B(t) in the near region is comparable with that of current inducing a
magnetic field.
3.8 Total lightning flash duration and processes in the

intercomponent pauses
A descending negative lightning flash has on average two or three compo-

nents, each terminated by a more or less powerful current impulse of the
return stroke. The average number of components in an ascending lightning
is four. The maximum number of components in a lightning flash may be as
large as 30. The pauses between the components At,,, vary from several
milliseconds to hundreds of milliseconds. With a 50% probability, their
duration exceeds 33 ms; the integral distribution curve is described by the
lognormal law with the parameters (lg At,,,)av = 1.52 and olg= 0.4, at
At,,, [ms]. The total flash duration varies with the number of components.
Negative one-component flashes are the shortest ones, since their current
often ceases right after the return stroke, for less than a millisecond. An
ascending one-component positive flash can pass current for a longer time,
0.5 s, in spite of the absence of a return stroke. Of course, this is a low current,
less than 1 kA. The average flash duration is close to 0.1-0.2s and the
maximum is 1.5s. These large times are discernible by the naked eye, so
lightning flickering is not a physiological by-product of vision but a physical
reality.
Intercomponent pauses take most of the flash time. They cannot be said
to be current-free. A lightning leader is supplied by current nearly all the
time, and this current is high enough to support plasma in a state close to
that of a steady-state arc. Current of an intercomponent pause is referred
to as continuous current, which is a fairly ambiguous term. Average
continuous current varies between 100 and 200A. Nearly as high current
supplies an arc in a conventional welding set used for cutting metal sheets
or for welding thick pipes. Most thermal effects of lightning are associated
with its continuous current, rather than with return stroke impulses which
are more powerful but shorter. The hghest continuous current measured
[55] was 580A. Continuous current usually slowly decreases with time. In a
one-component ascending lightning having no return stroke, the contact of
the leader with the cloud is terminated by charge overflow from the cloud to
the earth as a decreasing continuous current of about the same value.
Cloud discharging by continuous current can be easily registered by an elec-
tric field detector. Field varies monotonically, as long as current flows through

the channel. These are appreciable changes, since current of lOOA extracts,
from a cloud, charge A Q x 1OC over the time 0.1 s. The field on the earth
right under a cloud changes by the value A E = AQ/(27qH2) M 200V/cm if
the height of the charged cell centre is H = 3 km; at distance r = 10 km from
the lightning axis, A E = A Q H / [ 2 q , ( H 2+ Y~)~'~] zz 5V/cm. Similar values
were registered during observations.
Continuous current flow is accompanied by slowly rising and as slowly
decreasing current impulses with an amplitude up to 1 kA. These are M-
components of lightning. The risetime of a typical M-component is about
0.5ms, an average impulse duration (on the level 0.5) is twice as much, an
average amplitude is 100-200 A, although M-components with current up
to 750 A have also been registered [56,57]. Pulsed current rise is always
accompanied by an increase in light emission intensity of the whole channel,
from the cloud down to the earth. Streak photographs (even taken slowly) do
not show the propagation of a well defined emission wave front similar, say,
to the tip of a dart leader. It seems as if most of the channel flares up
simultaneously, although excitation, no doubt, propagates down from a
cloud with a high velocity, (2.7-4)x lo7mjs (from measurements of [58]).
Two M-components were identified in [58] as ascending ones. In later
measurements, the existence of ascending processes were questioned, because
there were no clear physical reasons for the appearance of an inducing
perturbation at the earth's surface.
Variations in current and electric field of M-components were registered
in triggered lightning flashes at a short distance from the channel (r = 30 m)
[57]. The field variation of a vertical component at the earth is shown in figure
3.24. The pulse A E rises to its maximum 70ps earlier than the current
impulse. The field rises and decreases at nearly the same rate. The pulse
component of field perturbation is nearly completely damped while the
current still has a high amplitude.
I- --
2?0 I 400 600
b
m t, PJ-
,e-
0.5
loO1
1.5
E, kV/m
Figure 3.24. Superimposed schematic oscillograms of M-component electric field and
current at the earth [ 5 7 ] .

Flash charge and normalized energy 131
The number of M-components in a flash may even be larger than that of

subsequent components, but they are of little interest to lightning protection
practice - their charge and current are too low. Theoretically, however, these
components are of great interest, because they seem to contain information
on unobservable processes occurring in storm clouds. It is quite likely that
these processes give rise to a dart leader with a return stroke or to a
stroke-free M-component. Some authors [27] believe that an M-component
is always formed against the background of continuous current, whereas a
necessary prerequisite for a dart leader is a current-free pause, during
which the grounded lightning channel partly loses its conductivity. This is
a very important detail shedding light on processes occurring in a storm
cloud after a grounded plasma channel of the first lightning component
has penetrated it. The transport of the earth’s zero potential to a cloud by
a conducting channel, resulting in a rapid increase in the cloud electric
field in the vicinity of the channel top, is a powerful stimulus for gas discharge
processes there (for details, see sections 4.7 and 4.8).
3.9 Flash charge and normalized energy
During intercomponent pauses, charge is transported from a cloud to the

earth by both powerful return stroke impulses and continuous current, the
latter being much lower but longer-living. The contributions of these currents
to the total charge effect are comparable. With a 50% probability, the stroke
charge transported by the first component of a negative flash is over 4.5C,
while 5% of flashes transport over 20C and another 5% less than 1.1C
[42]. The lognormal law described above is suitable for an approximate repre-
sentation of the integral distribution curve with the values (lg = 0.653
and olg= 0.4. The return strokes of subsequent components have, for the
same probabilities, five times smaller charges due to their shorter duration
and lower currents. The largest spread of charge measurements is character-
istic of positive lightning, in agreement with the diversity of their shape and
duration. Positive pulse charges exceed 16C with a 50% probability, 150C
with a 5% probability, and are less than 2C with a 5% probability. These
seem to be positive lightning with no return stroke. For the description of
integral charge distribution for positive pulses, the lognormal parameters
may be taken to be (lg Q)av = 1.2 and rlg= 0.6.
We have already mentioned that the charge of a lightning flash is always
larger than the sum of charges transported by the return strokes of the first
and subsequent components, since a substantial contribution to the total
charge is made by continuous current. The total negative flash charge exceeds
7.5C with a 50% probability, 40C with a 5% probability, and is nearly the
same as the first negative pulse charge in the least powerful flashes. The
total positive charge is appreciably larger - with 95%, 50% and 5%

probabilities, it exceeds, respectively, 20, 80 and 350C. One cannot say that
the charge transported by a flash is very large. For comparison, even a very
large lightning charge of 350C flows through the arc of a conventional
welding unit for 3-5 s.
Charge transport is accompanied by energy release. An average negative
flash with a charge Q = 1OC and gap voltage 50 MV dissipates about QU =
5 x 10sJ, which is equal to the energy released by a 100 kg trinitrotoluene
explosion. While most energy is released within the lightning trace, the
problem of energy release and heating of metal constructions is of much inter-
est. Normally, the resistance of metallic conductors and that of a grounding
electrode are much less than the equivalent resistance of a lightning channel
RI = U / I M (IM is the impulse amplitude of a return stroke); RI = 1 kR if
U x 50 MV and ZM = 50 kA. Therefore, lightning can be regarded as a current
source, assuming that current IM is independent of the object’s resistance. Any
conductor with lightning current flow releases the energy
K =R 1; i2 dt (K/R)R. KIR = 1; i2 dt
proportional to the conductor resistance R. For practical calculations, data

on ‘normalized’ energies K / R characterizing lightning only are published.
According to [42], 95%, 50% and 5% probabilities correspond to the
measured values exceeding 2.5 x lo4, 6.5 x IO5 and 1.5 x 107A2s for
positive flashes and 6.0 x lo3, 5.5 x IO4 and 5.5 x 105A2s for negative
flashes, respectively. For subsequent components of negative flashes, the
respective values are an order of magnitude smaller and do not contribute
much to the total energy release. To get an idea about thermal potency of
lightning, evaluate the heat of a steel conductor with a cross section of
S = 1 cm’. With resistivity p = lop5R cm, the energy density released by a
powerful positive flash ( K / R = 1.5 x lo7A’s) is ( K / R ) ( p / S 2=
) 150 J/cm3,
with the conductor temperature increasing by 40°C. Owing to Joule heat, a
lightning flash is capable of burning down only a very thin conductor with a
cross section less than 0.1 cm’. In many cases, however, heating just by several
hundred degrees may become hazardous.
3.10 Lightning temperature and radius
Plasma temperature is usually measured by spectroscopic methods. Light-

ning spectroscopy is a hundred years old, and it was used even before
photography and field-current measurements. Reviews of spectroscopic
results can be found in Uman’s books [l, 591 together with extensive
references. However, direct data on lightning plasma are still very scarce.
Lightning spectra, naturally, contain lines of molecular and atomic oxygen
and nitrogen, as well as singly charged ions N2, argon, cyane and some

Lightning temperature and radius 133
other impurities. No doubly charged ions have been detected, indicating that
the temperature does not exceed 30 000 K. Measurements of time resolved
NI1 (N') line intensities show that the return stroke temperature reaches
30 000 K for the first 10 ps [59,62] and drops to 20 000 K in 20 ps. Average
temperatures are estimated to be about 25 000 K. These results are obtained
assuming that a plasma channel is optically transparent and that the
excitation of atoms in the plasma is equilibrium (of the Boltzmann type).
The estimations justify this assumption.
Electron densities found from the Stark broadening of the Ha lines
are 1 0 ' * ~ m -for
~ the first 5ps of the stroke life. Under thermodynamic
equilibrium conditions at T = 30 000 K, this value of ne corresponds to
the pressure of 8atm [63]. About lops later, ne decreases to l O ' ' ~ m - ~ ,
corresponding to the pressure drop down to the atmospheric pressure.
Then the value of ne remains unchanged over the time of the N I I line
registration. This does not seem strange. Equilibrium electron density in
air at p = const = 1 atm changes only slightly in a wide temperature
range 15000-30 000 K, remaining about 1017~ m - As ~ . the channel cools
down, the ionization degree x = n e / N certainly decreases, but when the
pressure reaches the atmospheric value, the gas density N rises simul-
taneously. For this reason, ne = x N does not change much. High intensity
radiation is observed for about loops (from 40 to lOOOps). The first
peak is often followed by another one several hundreds of microseconds
later.
Spectroscopic measurements were mostly made during a return stroke,
but some authors [64] managed to register the spectrum of a 2-m portion of a
stepwise leader. The leader tip temperature calculated from the N 11 lines lies
within 20 000-35 000 K. The diameter of the radiation region is less than
35 cm. More accurate evaluations are unavailable. It seems unlikely that
this temperature is characteristic of the whole leader channel. Rather, the
experiment registered a short temperature rise during a powerful step
which was akin to a miniature return stroke (section 2.7). The step-induced
perturbation involving part of the channel region is most likely to be
damped rapidly along the leader length.
It is not only the plasma dynamics but the channel radius, too, which
still remains enigmatic. In making evaluations of the radius, one usually
relies on photographs. But in this case, it is very important to agree on the
kind of radius being evaluated. This may be the radius of the channel,
through which current flows during the leader and stroke stages. Clearly,
such a radius will include the best conducting and, hence, the hottest core
of the plasma channel. Or, one can follow another approach. When solving
the problem of electric field variation during the lightning development, one
has to deal with the radius of the leader cover where most of the space charge
is concentrated. This is the charge radius of lightning. Therefore, each time
we speak of radius, we must define exactly what we mean.

Here, we shall use the concept of channel radius as applied to the region
where the lightning current is accumulated and the concept of cover radius to
the region where most of the space charge is concentrated. The former can.be
determined, to some extent, by using optical methods, although this is a com-
plicated task. With reference to the optical measurements [65], one usually
deals with radii of several centimetres. This resolution is accessible to
modern cameras at a distance of about a kilometre, but the cameras must
have the highest resolution possible. Anyway, we have never heard about
the application of such perfect optical equipment in lightning research.
In addition to using special-purpose optics, the experimentalist must
match perfectly the sensitivity of photographic materials and exposures.
A longer exposure produces a halo, increasing the actual radius. Unless
special measures are taken, the error may be very large, especially for
flashes with a high light intensity. For some reasons, the optical radius of
a lightning channel may exceed manifold the thermal radius. Such an
effect was observed in studies of spark leaders in laboratory conditions
[66]. Registration of the thermal radius appears problematic even for trig-
gered lightning, with a fixed point of contact with the earth. For natural
lightning, this task is much more complicated. As for the cover radius,
there is no reliable technique for its registration at all. So lightning radius
measurements cannot provide unquestionable data, and the researcher is
to rely on theoretical evaluations only.
3.1 1 What can one gain from lightning measurements?
It was not our task to review all experimental studies on lightning: this has
been well done in [l, 591. We believe that the latest experimental data will
be presented in a new Uman book now in preparation. But the basic facts
have been discussed here, and we can now ask ourselves whether the
available data are sufficient to build lightning theory and to check it by
experiment.
The situation with lightning is somewhat similar to that for a long
laboratory spark, i.e., experiments give mainly external parameters of a dis-
charge. In the laboratory, these are velocities of the major structural elements
(streamers and leaders), their initiating voltages, currents, transported
charges, and, possibly, some other characteristics Sometimes, we have
some information on channel radii, or on the time variation of radii, or
scarce data on plasma parameters. But that is all.
The arsenal of lightning researchers is much smaller. First, they have no
information about the voltage in the cloud-earth gap at the lightning start,
and there are no data on the initial distribution of electric field. Both literally
and figuratively, the bulk of a storm cloud, where a descending leader
originates, is obscure. Measurements made at the earth’s surface cannot

What can one gain from lightning measurements? 135
help much, because the number of registration points is too small, so it is

impossible to reconstruct the initial field distribution along the whole
lightning path.
The fine structure of a lightning flash is not clear either. Observations
give no information about the size of the streamer zone in a lightning
leader, and even the existence of such a zone is largely speculative. Nor do
we know the origin and structure of volume leaders, which are responsible
for the stepwise pattern of a negative leader, at least, observable in laboratory
conditions. There is no information on the gas state in the track of a
preceding component, when a dart leader travels along it. The only dart
leader parameter that has been measured is its velocity. What has just been
listed refers primarily to the return stroke. It appears that space charge
neutralization - the basic process occurring in it - is related to the fast
radial propagation of streamers away from the channel. This is the way
the cover charge is supposed to change. But there are no experimental data
on this process, nor can we hope to obtain any in the near future.
Most available findings concern lightning currents and transported
charges. As in a laboratory spark, lightning currents are usually registered
at the earth’s surface, so we have data on leader currents for ascending dis-
charges only. There are no direct measurements of currents for descending or
dart leaders, the latter fact being especially disappointing. There are more or
less detailed descriptions of currents for return strokes, but the measurements
made at one point (that of contact with the earth) restrict the possibilities of
both a theoretical physicist and a practical engineer. Data on the current
wave damping along the leader are important for the former because then
he may try to reconstruct the plasma conductivity variation. The latter
needs them to be able to calculate the lightning electric field at the earth
and in the troposphere, because it is hazardous to both ground objects and
aircraft.
Lightning current statistics deserves special attention. Normally, they
are used in calculations of the occurrence probability of lightning with
hazardous parameters, e.g., a critically fast rise of the impulse front and/or
amplitude. The practical requirements on the calculation reliability are
extremely high. Indeed, it is impossible to provide the necessary accuracy,
using lognormal parameter distributions. Any approximation of an actual
distribution lognormally would be approximate, especially in the range of
large values important for lightning protection. The error may be as high
as 100%. One should keep this in mind when comparing calculations of
hazardous lightning effects and the available experience in object protection.
This is the reality not to be ignored either by a theorist attempting to
create a lightning model or by an engineer working on lightning protection.
No matter how ingenious a theorist may be, he will not be able to check his
model, filling the gaps by laboratory spark data or by general physical
considerations. As for practical lightning protection, one usually gained

the unfortunate experience from analyses of emergencies that resulted from

the lack of knowledge of atmospheric electricity.
References
[l] Uman M 1987 The Lightning Discharge (New York: Academic Press) p 377
[2] Golde R H (ed) 1977 Lightning (London, New York: Academic Press) vols 1, 2
[3] Imyanitov I M 1970 Aircraft Electrization in Clouds and Precipitation (Lenin-
grad: Gigrometeoizdat) p 210
[4] Gunn R 1948 J . Appl. Phys. 19 481
[5] Gunn R 1965 J . Atmos. Sci 22 498
[6] Evans W H 1969 J . Geophys. Res. 74 939
[7] Winn W P, Schwede G W and Moore C B 1974 J . Geophys. Res. 79 1761
[8] Winn W P, Moore C B and Holmes C R 1981 J. Geophys. Res. 86 1187
[9] Kazemir H W and Perkins F 1978 Final Report, Kennedy Space Center Contract
CC 69694A
[lo] Newman M M, Stahmann J R, Robb J D et all967 J . Geophys. Res. 72 4761
[ l l ] Kito Y, Horii K, Higashiyama Y and Nakamura K 1985 J . Geophys. Res. 90
6147
[12] Hubert P and Mouget G 1981 J . Geophys. Res. 86 5253
[13] Hubert P, Laroche P, Eybert-Berard A and Barret L 1984 J . Geophys. Res. 89
251 1
[14] Idone V P and Orville R E 1984 J . Geophys. Res. 89 7311
[15] Fisher R G, Schnetzer G H, Thottappillil R et a1 1993 J . Geophys. Res. 98 22887
[16] Wang D, Rakov V A, Uman M A et a1 1999 J . Geophys. Res. 104 4213
[17] Malan D J and Schonland F G 1951 Proc. R. Soc. London Ser. A 209 158
[18] Malan D J 1963 Physics of Lightning (London: English Univ. Press) p 176
[19] Malan D J 1963 J . Franklin Inst. 283 526
p 294
[21] Chalmers J A 1967 Atmospheric Electricity (2nd edn) (Oxford: Pergamon) p 418
[22] Berger K and Fogelsanger E 1966 Bull. S E V 57 13 1
[23] Schonland B 1956 The Lightning Discharge. Handbuch der Physik 22 (Berlin:
Springer) p 576
[24] Schonland B, Malan D and Collens H 1938 Proc. Roy. Soc. London Ser. A 168
455
[25] Schonland B, Malan D and Collens H 1935 Proc. Roy. Soc. London Ser. A 152
595
[26] Jordan D M, Rakov V A, Beasley W H and Uman M A 1997 J . Geophys. Res.
102 22.025
[27] Fisher R G, Schnetzer G H, Thottappillil R et a1 1992 Proc. 9th Intern. Conf. on
Atmosph. Electricity 3 (St Peterburg: A I Voeikov Main Geophys. Observ.) p 873
[28] McCann G 1944 Trans. A I E E 63 1157
[29] Berger K and Vogrlsanger E 1965 Bull S E V 56 No 1 2
[30] Bazelyan E M, Gorin B N and Levitov V I 1978 Physical and Engineering Funda-

References 137
[31] Saint Privat d’Allier Research Group 1982 Extrait de la Revue Generale de
I’Electricite, Paris, September
[32] Gorin B N and Shkilev A V 1974 Elektrichestvo 2 29
[33] Idone V P, Orville R E 1985 J . Geophys. Res. 90 6159
[34] Rubinstein M, Uman M A and Thomson P 1992 Proc. 9th Intern. Conf. on
Atmosph. Electricity 1 (St Peterburg: A I Voeikov Main Geophys. Observ.) p 276
[35] Rubinstein M, Rachidi F, Uman M A et a1 1995 J. Geophys. Res. 100 8863
[36] Thomson E M 1985 J . Geophys. Res. 90 8125
[37] Kolechizky E C 1983 Electric Field Calculation for High-Voltage Equipment
(Moscow: Energoatomizdat) p 167 (in Russian)
[38] Jordan D M and Uman M A 1983 J . Geophys. Res. 88 6555
[39] Schonland B and Collens H 1934 Proc. Roy. Soc. London Ser. A 143 654
[40] Idone V P and Orville R E 1982 J . Geophys. Res. 87 9703
[41] Idone V.P, Orville R E, Hubert P et a1 1984 J . Geophys. Res. 89 1385
[42] Berger K, Anderson R B and Kroninger H 1975 Electra 41 23
[43] Berger K 1972 Bull. Schweiz. Elekrtotech. Ver. 63 1403
[44] Gorin B N and Shkilev A V 1979 in Lightning Physics andLightning Protection
(Moscow: Krzhizhanovsky Power Engineering Inst.) p 9
[46] Eriksson A J 1978 Trans. South Afr. ZEE 69 (Pt 8) 238
[47] Anderson R B and Eriksson A J 1980 Electra 69 65
[48] Alizade A A, Muslimov M M et a1 1974 in Lightning Physics and Lightning
Protection (Moscow: Krzhizhanovsky Power Engineering Inst.) p 10
[49] Master M J, Uman M A, Beasley W H and Darveniza M 1984 ZEEE Trans. P A S
Pas-103 2519
[50] Krider E P and Guo C 1983 J . Geophys. Res. 88 8471
[51] Cooray V and Lundquist S 1982 J . Geophys. Res. 87 11203
[52] Cooray V and Lundquist S 1985 J . Geophys. Res. 90 6099
[53] McDonald T B, Uman M A, Tiller J A and Beasley W H 1979 J . Geophys. Res.
84 1727
[54] Lin Y T, Uman M A et a1 1979 J . Geophys. Res. 84 6307
[55] Krehbiel P R, Brook M and McCrogy R A 1979 J. Geophys. Res. 84 2432
[56] Thottappillil R, Goldberg J D, Rakov V A, Uman M A et a1 1995 J . Geophys.
Res. 100 25711
[57] Rakov V A, Thottappillil R, Uman M A and Barker P P 1995 J . Geophys. Res.
100 25701
[58] Malan D J and Collens H 1937 Proc. R. Soc. London A 162 175
[59] Uman M A 1969 Lightning (New York: McGraw-Hill)
[60] Orvill R E 1968 J . Atmos. Sci. 25 827
[61] Orvill R E 1968 J. Atmos. Sci. 25 839
[62] Orvill R E 1968 J . Atmos. Sci. 25 852
[63] Kuznetsov N M 1965 Thermodynamic Functions and Shock Adiabata for High
Temperature Air (Moscow: Mashinostroenie) (in Russian)
[64] Orvill R E 1968 J. Geophys. Res. 73 6999
[65] Orvill R E 1977 in Lightning, vol 1, R Golde (ed) (New York: Academic Press)
p 281
[66] Positive Discharges in Air Gaps at Les Renardieres - 197s 1977 Electra 53 31

Chapter 4
Physical processes in a
Iight ning discharge
Here we shall discuss the basic phenomena occurring in a lightning discharge:

a descending negative leader, an ascending positive leader, the return strokes
of the first and subsequent components, a dart leader, and some others.
Lightning may travel not only from a cloud towards the earth, or from a
grounded object towards a cloud, but it may also start from a body isolated
from the earth - a plane, a rocket, etc. About 90% of all descending
discharges are negative and about as many ascending discharges are positive.
For this reason, an ascending leader is said to be positive. Available experi-
mental data on lightning as such are of little use in our attempts to explain the
mechanisms underlying the above processes. There are very few observations
that might shed light on their physical nature. So, one has to resort to spec-
ulations, invoking both theory and experimental data on a long laboratory
spark, which relate primarily to a positive leader. Since this process is most
simple (to the extent a lightning process may be considered simple), we
shall begin with the discussion of an ascending positive leader.
4.1 An ascending positive leader

4.1.1 The origin
The lightnings people observe most frequently are descending discharges,
which originate among storm clouds and strike the earth or objects located
on its surface. However, constructions over 200m high and those built in
mountainous regions suffer mostly from ascending lightnings. These are of
nearly as much interest to the physicist as the seemingly common, descending
discharges. An ascending leader is initiated by a charge induced by the elec-
tric field of a storm cloud in a conducting vertically extending grounded
object. If a metal conductor of height h with a characteristic radius of the
rounded top r << h is fixed on the earth and then affected by a vertical
138

A n ascending positive leader 139
external field Eo, a field El x Eoh/r >> Eo is created by the induced charge at
the conductor top (see section 2.2.7). This field rapidly decreases in air (for a
distance of several r values), creating a potential difference between the
conductor end and the adjacent space, AU x Eoh. When the cloud bottom
is charged negatively and the vector Eo is directed from the earth up to
the cloud, the grounded conductor becomes positively charged, since the
field makes some of the negative charges leave the metal to go down to
the earth.
No stringent conditions are necessary for the field E , to initiate the air
ionization (at sea level El x E, x 30 kV/cm) or for a corona discharge to
arise at the pointed parts of a high structure (it is necessary to have
El x40-31kV/cm for r = 1-10cm). The conditions for a leader to be
initiated in the streamer corona stem are much more rigorous. The energy
estimations made in section 2.6 show that there is no chance for a leader to
arise if the leader tip potential U,, or, more exactly, its excess over the external
potential at the tip, A U = U, - U,, is less than AVrm,,x 300-400 kV. This
estimate is supported by experiments with leaders, whose streamer zones
have no contact with the electrode of opposite sign at the initial moment of
time. Therefore, for the desired potential difference A Ut,, to be produced,
the structure must have, at least, h x AUrm,,/EO x 20-30m if the average
field of the storm cloud at the site of the grounded object is -150 V/cm.
On the other hand, even if a leader is produced at such a low potential,
AVfm,,,it can hardly travel for a large distance. The leader current will be too
low to heat the channel to a sufficiently high temperature. As a result, the
channel resistance will be too high so that a very strong field will be required
to support the current in the channel. The channel field E, is, however,
limited by the external field Eo. Indeed, a grounded body of height h, from
which a positive leader has started, possesses zero potential. Having covered
the distance L, the leader tip acquires the potential U , = -E,L. Here, the
+
potential of the unperturbed external fields is U. = -Eo(L h), and we have
AUt = Ut - U0 = AU, + (Eo - E,)L. AU, = Eoh. (4.1)
For a leader to develop from the initial threshold conditions, the potential
difference A U , should not decrease relative to the initial value of AU,. For
this, the average channel field E, must be lower than the external field Eo.
However, a mature channel possesses a falling current-voltage characteristic
E, (i). A decrease in E, to 100 V/cm requires a channel current higher than
N
1A. We discussed this issue in sections 2.5.2 and 2.6. With the approximation
accepted there (E, x b / i and b =300 VAjcm), the leader current is to exceed
,,i = b/Eo x 2 A at Eo x 150 V/cm.
Let us see how large the potential difference A u t should be to make the
current exceed i,. In chapter 2, we derived formula (2.35) relating the
channel current behind the leader tip to the tip potential U, and the leader
velocity vL. That formula was applicable to the laboratory conditions

140 Physical processes in a lightning discharge
considered in that chapter, when a leader travelled through the rapidly

decreasing field of a high-voltage electrode. Having covered a distance of
only a few radii of the electrode curvature, usually very small, the leader
tip found itself in a space with a nearly zero potential, U, << U,. The neglect
of the external field was justifiable in that case. An ascending leader is quite
another matter: the potential difference A U , = U , - U, continuously
increases with the leader velocity, as its tip approaches a charged cloud,
since the tip enters a region of an ever increasing external field. Hence, we
have I Uti << I Uoi and the value of A U , largely determined by U,. Therefore,
the approximate formula of (2.35) must be rewritten in its general form, with
U , replaced by AU,:
. ~TEOAU~VL
’
= ln(L/R) ’
aut = U, - U,
where R is the effective radius of the leader charge cover.
The available leader theory fails to provide a clear and convincing
physical expression to describe the relationship between vL and AU,. So
we shall further use the empirical relation suggested in section 2.6:
wL = a(AU,)”2, a = 15m/~V’/~. (4.3)
This relation was derived from experimental data on rather short gaps, in
which the tip potential could be taken to be identical to that of a high voltage
e1ectrode.t In accordance with (4.2), expression (4.3) corresponds to the
relation vL = ill3 also supported by some laboratory experiments. Now,
using the value of imin,we shall find AUi which provides the leader viability:
Assuming L = 10m and R = 1 m for a still-short initial leader at

Eo = 150V/cm, we obtain AVjmln= 3.1 MV and hmin = 210m. The result
of this simple estimation agrees with that of lightning observations. In a
flat country, ascending lightnings make up an appreciable fraction of the
total number of strikes affecting grounded objects of about that height.
The continuous ascending leader of a triggered lightning (initiated from a
grounded wire raised by a rocket above the earth) is also excited at about
200m. Note that the value of Eo used in the calculations is somewhat
larger than those measured at the earth surface. The storm cloud field
near the earth is always attenuated by the space charge introduced in the
air by corona discharges from thin conductors of small height, such as
tree branches, shrubs, grass, constructions, etc. Some measurements show
that the field at the earth is half that at a height of 10-20m.
t The streamer theory has been advanced further. Note, for comparison, that the streamer
velocity is V, RZ AU, from formulae (2.6) and (2.8) with AU, instead of U, and E, = const.

An ascending positive leader 141
4.1.2 Leader development and current

Two main leader parameters are accessible to measurement: its velocity and
the current through the channel base contacting a grounded object. The
current is due to the charge pumped by an electric field into the growing
leader. If the field in the leader channel is lower than the external field
(otherwise the leader is non-viable), the difference between the tip potential
and the unperturbed potential at its site becomes larger as the leader becomes
longer (see expression (4.1)). According to (4.2) and (4.3), the leader current,
proportional to i AU:”, also rises. The current rise becomes more rapid as
N
the leader becomes longer, especially when the leader reaches the region of a
very high cloud field. The rising current heats the channel more, so that its
linear resistance and field drop. With time, the channel becomes a nearly
perfect conductor. Grounded at its base, the channel possesses the same
potential U ( t ,x) everywhere along its length, including the tip, which is
low relative to the absolute external potential IUo(x)I. In this case, the
value of A U = U ( t , x )- Vo(x)z -Uo(x)varies only slightly with time at
every point x along the channel.
The linear leader capacitance C1, given by formula (2.8) with length L
and cover radius R instead of 1 and Y, also varies very little. Indeed, the
cover radius behind the tip is about the same as the streamer zone radius
which, according to (2.39), is R = AUt/2E,, where E, M 5 kV/cm is the
streamer zone field under normal conditions. The height of the charged
region centre in the cloud, H PZ 3 km, is much greater than that of the
leader starting point, h PZ 200m. Suppose the leader length is greater than
h but smaller than H by such a value that the cloud field non-uniformity
along the channel can be neglected. We then have AU, M IUo(L)I M IEoLI,
and the value of LIR M 2E,/Eo under the logarithm in C1 of (4.2) and
(4.4) is independent of time, If this relation does change, which happens
when a leader rises so high that it enters the region of a rapidly increasing
external field, the logarithm changes much more slowly. Thus, the linear
charge ~ ( x M) C , A U remains nearly constant in time at every leader point.
But if there is no charge redistribution along the channel, the current in it,
i(t, x), does not change along its length but changes only in time. Entering
the channel through its grounded base, the current supplies charge only to
the front leader portion. The current in the base is the same as in the channel
right behind the tip. It is defined by formula (4.2) with Aut M I Uo(L)I,close
to the unperturbed potential of the cloud charge at the tip site. Similarly, the
leader velocity can be found from (4.3).
Therefore, the velocity and current of a fairly long leader (long relative
to the start height), which develops in the average field Eo, are described as
1‘ = 2TEoa(AUt)3’2
WL = a(ALJ,)1/2, AU, = EoL, e = 2.72.. . (4.5)
ln(2&E,/eEo) ’

The numerical factor f i / e in the second expression of (4.5) has resulted

from a more rigorous calculation. This result is obtained if the linear charge
T(X) is calculated directly from external field Eo with r(x) = const x, as in
section 3.6.2, rather than from average linear capacitance C1 and A U ( x ) ,
as was done in the derivation of (4.2). The current is found from
i = dQ/dt, where Q is the net charge of the conductor (the integral of ~ ( x )
in x).
It follows from (4.5) that the current rises rapidly with time as the
leader develops, whereas the velocity increases much more slowly:
U, = dL/dt x L1I2 M t and i M L3I2M t3 at Eo = const. In stronger external
fields, the leader current also rises with EO much faster than the velocity.
Numerically, a leader with L = 500 m has i = 4.5 A and = 4 x lo4 cmjs
in an average field Eo = 150V/cm, i.e., about the same values as for an
extremely long laboratory spark. An increase of L to 2000m and Eo to
300V/cm gives the typical lightning parameters: i = 120A and
U, = 12 x 105cm/s. A rapid rise of the leader current and a much slower
increase of its velocity were inevitably registered in observations of both
natural and triggered lightnings [ 1,2]. These estimations reasonably agree
with measurements.
In contrast to (4.2) and (4.3), expression (4.5) ignores the voltage drop
across the leader channel because U, << lUo(L)l.It is easy to see the validity
of this assumption in the next approximation using the derived formulae.
With the voltage-current characteristic E M [ lip', the voltage drop across
the channel decreases as U , E EcL M L'I2 M t-' with the leader develop-
ment. At the tip site, on the contrary, IUo(L) = lEoiL grows even faster
than L t2 if one takes into account the increase of the average external
field along the channel during its travel up to the cloud. Note that E, and
U, do not drop to zero in reality but only decrease to a certain limit, because
the field Ec(i) in a very heated channel with high current is stabilized due to
the greater plasma energy loss for radiation (the current-voltage character-
+
istic should be expressed as E, = c b / i rather than as E, = b/i). This issue,
however, is of no importance to an ascending leader, since its current
becomes very high only when the tip reaches the region with 1 U o ( L ) />> U,.
If one desires to refine these simple results by taking account of the
voltage drop, charge redistribution, and current variation along the channel,
one should regard it as a long line, as was done with the streamer in section
2.2.3. The distributions of potential U ( t . x ) ,charge per unit length r(x,t ) ,
and current i ( x , t ) along the line can be described by equations similar to
(2.13) and (2.14):
dr di dU
-
at +-
dx = 0, --
dX = E,(i), i(L) = r(L)vL (4.6)
where E, is the longitudinal channel field expressed through current i ( x , t)
from the current-voltage characteristic (the field in (2.13) was expressed

through current and linear resistance, E, = iR1). The leader velocity uL is

given, for example, by formula (4.3): dL/dt = uL.
Equations (4.6) must involve an electrostatic relation between charges
and potentials. In a simple approximation, expressions (2.13) and (4.2)
) C1[ U ( x )- U o ( x ) ] through
were allowed to contain a local relation, ~ ( x = ,
linear capacitance C1.This approximation was shown by many calculations
to be quite acceptable to the case of a uniform or weakly non-uniform
external field, but it appears insufficiently rigorous for a strongly non-
uniform field, which a leader crosses on entering a storm cloud. In fact,
the potential at every point along the channel length is also created by
charges located at adjacent channel sites. To simplify the non-local relation,
the leader charge can be assumed to be concentrated on a cylindrical surface
with an effective cover radius R; then the desired relation takes the form
r ( z ,t) dz
A U ( x , t) = U ( X ,t) - U ~ ( X=) (4.7)
[(z - x ) +~ R2I1l2
*
The boundary conditions for the set of integral differential equations (4.6)
and (4.7) are described by the third equality in (4.6) and U ( 0 ,t) = 0, since
the leader base is grounded. Practically, it is convenient to subdivide the
channel into N fragments and consider the charge density in each fragment
to be dependent only on time, thus replacing the integral equation of (4.7) by
a set of linear algebraic equations. Each of them will relate the potential
U(xk) at the middle point xk of the kth fragment to the intrinsic and all
other linear charges. After integrating (4.7), one can easily see that radius
R enters logarithmically the factors of the set of equations (compare with
(4.2)), thereby justifying the use of linear leader charge r instead of its
cover space charge. The set of algebraic equations for U(xk) and '(Xk) is
solved in time at each step, and the progress is made by using equations
(4.6). We are presenting the result of this solution.
As the leader tip approaches the cloud, the external field at the tip site
becomes stronger and the ever increasing portion of the channel finds itself
in a strongly non-uniform field. Since the velocity and current are largely
defined by the potential Uo(L) at the tip site, formulae (4.5), in which Eo is
an average field, remain valid. In a simple model of a cloud with a spherical
unipolar charged region, the potential distribution in the space free from
charges is the same as for a point charge. If H is the height of the spherical
charge centre, Q,, the potential at height x at the point displaced from the
vertical charge axis for distance r (with the account of the mirror reflection
by the earth's plane) is
"{ 1
- }.
(4.8)
= 4T&o [ ( H - x ) 2 + r2I1l2 [ ( H+ + y2I1l2
Figure 4.1 presents the parameter calculations for an ascending leader
propagating in such a non-uniform field. The calculations were made from

300
- 1.4
3
- 1.2 e
E . E
- 1.0 *z
Y
~ 200
s
8 -8 3 .
-0.8 'g6
5
U 3
d
?
100 2 - 0.6 L
1 3
- 0.4 2
- 0.2
Figure 4.1. The propagation of an ascending leader from a grounded object in a

negative cloud field. Indicate current iL calculated from (4.5); Q, = 5C,
H = 3km, r = 0.5km.
the set of equations (4.6) and (4.7), as described above. The current at the
channel base is defined by the total charge Q and the velocity by expression
(4.5):
For comparison, the current was also calculated from (4.5). The results show
the good accuracy of this simple formula, so the use of average linear
capacitance C1 can be considered justifiable in the calculation of
T ( L )= CIAUo(L)and in the case of a sharply non-uniform field.
4.1.3 Penetration into the cloud and halt

There are two questions to be answered here: how high the maximum leader
current is and where the leader halts. To answer the first question, one should
keep in mind that the cloud charge is concentrated in a certain volume but
not at a point. Suppose it is a sphere of radius R, with the centre at height
H in (4.8). Measurements made during flights through storm clouds indicate
that R, is most likely to be by an order of magnitude smaller than H . The
maximum potential at the centre of a uniformly charged sphere is by a
factor of 1.5 higher than on its surface and equals U,,,, = 3Qc/8mORc.
Penetrating into the charged region, an ascending leader acquires a
considerable velocity and a very high current. To illustrate, calculated

from (4.8) for H = 3 km, the field near the earth under the charge centre,
EO= -Qc/27qH2 = 150V/cm, is created by charge Q, = -7.4C and
~Uo,,,I = 340MV for R, = 300m. From (43,a leader that has reached
the charged region centre acquires the velocity wL = 2.8 x 105m/s and the
current i,, = 5 kA (the field at the sphere boundary is EOmax = 7.5 kV/cm,
decreasing to zero towards the centre; in the current estimation, the
logarithm was taken to be 1). The lifetime of this high current is short,
about Rc/w~,,, lO-'s, with the total duration of the leader ascent of
N
about 3 x s (these estimations ignore the effect of air density, which is

1.5 times lower than normal at a height of 3 km). Maximum currents of
the kiloampere scale were registered during observations of ascending
lightnings.
On its way up through the charged region, the leader enters an area of
reciprocal external field at height x > H. The potential difference AU, is,
at first, positive but decreases as the leader elongates. Its velocity and current
now decrease with time, but this process has its limits. There is a region of
positive charge of nearly the same value high above the negative charge
region. Representing it as a sphere with the centre at height H D and +
taking the mirror reflection effect into account, as in (4.8), we can find the
potential of the dipole thus formed:
-
+
1
+
+
[(H D - x ) ~ r2I1l2 [(H+ D
1
+ x ) +~ r2I1l2
}. (4.10)
Without allowance for the voltage drop across the channel (U, 0) the
leader tip will reach the point x,, where the absolute potential U. drops to
AU, E 400 kV, remaining negative as before. Since AUlmnis small relative
to huge potentials of charged regions (I Uolm,, N 100 MV), a positive ascend-
ing leader halts at a slightly lower height than the zero equipotential surface
of the external field. Because of the effect of charges reflected by the earth, the
zero potential line lies somewhat lower than the dipole centre. For example,
at D = H , which corresponds, more or less, to reality, we have x, = 1.4868
exactly on the vertical axis (Y = 0) instead of 1.5H, as would be the case with
a solitary dipole. With greater radial displacement Y, the zero equipotential
line comes closer to the earth, slowly at first but then more rapidly at
r > H (figure 4.2). This is the reason why ascending leaders taking different
vertical paths halt at different heights.
It has just been mentioned that the equipotential line U o ( x ,Y) = 0
corresponds to the maximum height attainable by a single ascending
leader. With allowance for the voltage drop across the channel, which
may appear appreciable in some situations, AU, drops to the threshold
value AUc, below the maximum height. This is supported by numerical

0.00
0.0 0.5 1.o 1.5 2.0
r/H
Figure 4.2. The zero potential line of a cloud dipole with the allowance for charges
reflected by the earth for D = H.
calculations made from the set of equations (4.6) and (4.7) and illustrated in
figure 4.3. They also indicate the leader retardation rate. As the leader
velocity decreases, the channel current becomes lower, causing the field E,
to rise. The tip potential decreases respectively, together with the potential
difference AU,, which limits the current still more, and so on.
2.0 200 1000
800
Q 1.5
E
150 s
-, +
0
d 1.0 $ 600
a
.s
-
0
0
loo z 400 $
el
8
U
8 0.5 50
el 200
0.0 0 0
Leader length, km
Figure 4.3. Numerical simulation of an ascending leader propagating in a cloud
dipole field (Qc = 12 C, H = D = 3 km, r = 0.5 km), with allowance for the voltage
drop across the channel.

When a leader goes beyond the lower cloud charge region, the external
field changes its direction along the channel: below the negative charge
centre, x = H , its vertical component is directed upwards but above the
centre it is directed downwards. Correspondingly, the external potential U,
is non-monotonic and has an extremum at height H (absolute maximum).
The leader continues to develop beyond the maximum point, as long as
the relation
Aut ut(L) - Uo(L)= IUo(L)I- Iut(L)/> AUt,,,,,
is valid. But now, the leader velocity decreases continuously, because I Uo(L)I
drops with the leader elongation and I U,(L)I rises due to the rising channel
field. It is clear that a leader can develop successfully in any other direction,
since it is capable of propagating in the direction strictly opposite to the
external field. The calculations show the leader path along the equipotential
line in a zero external field. Here, AU,, i and wL decrease slowly, only due to
the greater voltage drop across the channel; otherwise, the leader would
travel for an infinitely long time.
We have focused on this circumstance because it is here that the princi-
pal features of a leader process manifest themselves clearly. The external field
at the tip site is usually low and cannot affect the instantaneous leader velo-
city, current and direction of motion. The direction may vary randomly, a
fact well known to those making lightning observations. What is important
is the voltage U, created by this field along the leader path, rather than the
field strength. The propagation of a positive leader is provided by the trans-
port of a fairly high positive charge to its streamer zone. The current of many
streamers taking the charge out accumulates in the channel, heating it and
providing its viability. But for many streamers to be excited off from the
leader tip, the latter must possess a high potential relative to the unperturbed
potential AU, = U,(L) - U o ( L )x IUo(L)I.This is indicated by the absence
of appreciable discrepancies between the current calculations made straight-
forwardly from the linear density of induced charge T(X) in a strongly non-
uniform external field and from formula (4.5)containing only AU,.
Therefore, it is not surprising that the lightning paths exhibit the diversity
illustrated in figure 4.4.No random change of the leader path can disturb its
viability. A leader can follow any direction: it can move along the external
field or in the opposite direction, along the equipotential line, etc. - all ways
are open as long as the condition AU, > AUtmnis valid. But the leader
acceleration does depend, of course, on its direction of motion. Moving
along the field, the leader is accelerated, because the voltage drop is compen-
sated excessively by the increase in I U, 1, When the leader moves in the opposite
direction, it is decelerated. The maximum acceleration is achieved in the
direction of the maximum gradient AU,, and this seems to be the reason for
the fact that the main leader branch darts in the direction of the rising field,
i.e., towards a charged cloud, a high object, etc. (for details see section 5.6).

Figure 4.4. A photograph of a well-branched lightning with the path bendings.
4.1.4 Leader branching and sign reversal

Leader branches are nearly always visible in photographs of ascending
lightnings. Branching may start almost from the channel base or after the
leader has covered many hundreds of metres (figure 4.4). The currents of
branches are summed up at the branching points, so it is higher at the channel
base than in any branch. It is very unlikely that branches would start
simultaneously and that the potential differences AU, at their tips would
be the same at any moment of time. Rather, the values of AU, are distributed
randomly. An abrupt decrease or even an entire cut-off of current in one of
the branches does not at all mean that a similar thing has happened in
another branch or in the base. Therefore, at least one of the ‘main’
branches will have a relatively high current and, hence, a greater probability
to go up very high and even to reach the maximum leader height x, than a
single leader does. This event is stimulated by the decreasing voltage drop
along the branching leader ‘stem’, where the total branch current has accu-
mulated and where the field is low (in accordance with the current-voltage
characteristic), especially if the stem is long and branching occurs at different
heights.
Branching can send the leader up above the zero equipotential surface,
where its sign reversal occurs. Imagine the situation, in which a branch,

A n ascending positive leader 149
running up far from the charged cloud region, has reached the maximum
height x , ( r ) and stopped. The channel plasma cannot decay immediately
but persists for some time. During this time, another, luckier branch has
approached the negative cloud bottom and even partly penetrated it. In
this region, the cloud has potential U,,,,, so that the branch portion that
has entered it acquires the positive charge
which may be as large as S x 10% of the negative cloud charge. Due to the
partial compensation of the lower cloud charge, with the upper charge
being constant, the zero equipotential surface will become lower by the
length Ax, which is about the same percentage of x, - H , so that
Ax x S(x, - H ) . As a result, the upper portion of the first halted branch
(from the tip down to the new zero equipotential surface) will be in a field
directed downwards. The new external potential at the tip site,
U; x IdUo/dxlxsAx, will become positive and the potential difference
AU, = U,(L)- U ; ( L )will be negative. For the example given in the previous
section with x, - H x 1.5 km at S x 0.1, we have Ax x 150 m; from formula
(4.10) with r << H , we have IdUo/dx~xs x 600V/cm, so that eventually
U;(L) x 9MV. Even if the branch penetrating the cloud charge misses its
centre to enter a region with a potential several times lower than U,,, (as
a result, UA(L) will be reduced as much), this will still be sufficient to
revive the first leader branch.
Therefore, the halted leader has a chance to revive and move on up to
the upper positive cloud charge but as a negative leader this time. The
leader position at the point of the first stop is unstable. Even a slight
perturbation, such as a decrease in the lower cloud charge (in the example
presented, due to the penetration of another branch) may stimulate its
further growth with the opposite sign. As the leader develops, it will
penetrate into an increasingly higher field of the upper charge and
become accelerated. Having passed the upper charge centre, H D, it +
will be retarded and stop, for good this time, at a height H,,, > H D, +
where the potential of (4.10) will drop to a relatively low value of AUtmin.
The height H,,, may be 10-20km or higher if one accounts for the air
density decrease. The currents flow in different directions in different
portions of this leader. Above the equipotential surface, the current flows
downwards, as in a negative leader. In the lower leader portion which
serves as a stem for many positive branches, the current remains directed
upwards. The observer, who registers the current at the earth, may not
suspect the sign reversal occurring up in the clouds. The channel field is
established in accordance with the current. It reverses in the upper channel
portion, thereby reducing the total voltage drop across the branch that
went far up and stimulating its further development.

4.2 Lightning excited by an isolated object
Like a high grounded body, a large object isolated from the earth can become
a source of lightning in a high electric field of a storm cloud. Discharges can
be induced by fields extending not only between the earth and a charged
cloud but also between oppositely charged clouds. A lightning discharge
can be excited by a large aircraft, rocket or spacecraft when it travels through
the troposphere, and this is a serious hazard to its flight. Therefore, this
phenomenon is of primary practical importance.
4.2.1 A binary leader

In contrast to an ascending leader starting from a high grounded body
(section 4.1), an isolated body produces two leaders, one going along the
external field vector and the other in the opposite direction. The physical
reason for the excitation of two leaders is the same. The external field induces
charge in the conductor, so that a large difference between its potential and
the external potential arises at the conductor end. If the body is extended
along the field, the electrical strength at its end increases abruptly. In contrast
to the situation with a grounded conductor, the opposite charge does not
flow down to the earth but accumulates at the other end, polarizing the
isolated body. A grounded conductor in an external field possesses the earth’s
potential, while an isolated conductor acquires a potential corresponding to
an average external potential along its length. Large differences between the
body’s and external potentials (of opposite signs) now arise at the ends of the
body, and both ends are capable of exciting leaders of the respective signs. A
long conductor absolutely symmetrical relative to its average cross section
transversal to the uniform field acquires potential U equal exactly to the
external field at the body’s centre. The distribution of unlike charges in
each of its halves is identical to the charge distribution in a grounded
conductor of the same size and shape as the isolated conductor half.
The process described here can be easily reproduced in laboratory
conditions. Figure 4.5(a) shows streak pictures of leaders which have started
from a rod of 50 cm in length, suspended by thin plastic threads in a 3-m gap
in a uniform field. One can see all characteristic features of a positive leader
propagating continuously to the upper negative plane and those of a stepwise
negative leader travelling down towards a plane anode. Generally, leaders
arise at different moments of time because of the threshold field difference
for the excitation of positive and negative initial streamer flashes or due to
the difference in the curvature radii of the rod ends. The leaders may have
different velocities because the same voltage drop A U, creates streamer
zones of different sizes at the positive and the negative ends. The instanta-
neous currents at the growing channel ends may also differ. But on average,
every leader transports the same charge, since the net charge remains to be

Lightning excited by an isolated object 151
Figure 4.5. Streak photographs of leaders from the ends of a metallic rod placed in a
uniform electric field: (a) general view; (b) fast streak photograph demonstrating the
relationship between the positive and the negative leaders; (1) rod, (2), (3) tip and
streamer zone of positive leader, (4),(5) tip and streamer zone of negative leader,
(6), (7) negative and positive leader flashes.
zero in a system isolated from the voltage source. The discharges appear to be
interrelated. Any fluctuation - say, a flash - of one leader appreciably acti-
vates the other: the space charge (e.g. positive) incorporated in front of the
rod stimulates the accumulation of negative charge across the conductor,
thereby enhancing the field at its negative end. The high-speed streak pictures
in figure 4.5(b),resolving individual streamer flashes, show an activation of
the positive leader channel following a negative leader flash.
The conditions for the start of leaders from a long isolated conducting
body are the same as from a grounded conductor, and they are also defined
by expression (4.4). But now, when estimating the threshold field Eo from the
value of AVimin = Eod, one should keep in mind that d is a half length of an
isolated leader. For a field capable of exciting a discharge from a conductor
of length 2d, we find
E o = - [1 bln(L/R) ] 2/5 .
(4.11)
d3/5 27qa
As in the illustration in section 4.1, we take the ratio of the channel length of
a young leader to the equivalent charge cover radius to be L/R M 10. Then
we have Eo = 440 V/cm for an aircraft of length 2d = 70 m. This estimate
describes the external field component along the aircraft axis. But an aircraft
often flies at an angle to the field vector, so that the threshold external field
may be several times higher. Fortunately, the lightning excitation threshold is

not very low, otherwise airline companies would suffer tremendous losses
from lightning damage. On the other hand, fields of this scale are not very
rare: much higher fields were registered during airborne cloud surveys. For
this reason, the problem of lightning protection in aviation is regarded as
being very serious.
Having started from an isolated body, each leader develops as long as
the external field permits. This process is basically the same as that discussed
in section 4.1 for an ascending leader. Below, we shall consider the specific
behaviour of two differently charged leaders developing simultaneously.
This specificity becomes especially clear in a non-uniform field typical of a
storm cloud.
4.2.2 Binary leader development

The principal features and quantitative characteristics of a binary leader can
be understood from a simple model. The x-coordinate will be taken along the
leaders. The leader paths should not necessarily be straight lines but they may
have various bends, as is the case in reality. Denote the external field poten-
tial along the leader lines as Uo(x) and their tip coordinates as x1 and x2. In
figure 4.6 Uo(x) corresponds to the field of a negatively charged cloud. The
leaders were excited by a conducting body somewhere half way between
the cloud and the earth. The x-axis is directed upwards, the leader with the
subscript 1 travels downwards and the one with the subscript 2 upwards.
Let us neglect the voltage drop across the leader channels, ascribing the
same potential U to the channels and the initiating body. The whole system
now represents a single conductor. In the satisfactory approximation above,
in which the capacitance per unit length C1 at every moment of time was
Figure 4.6. A schematic diagram of a binary leader channel in a cloud dipole field.
x1: descending leader tip coordinate; x2: ascending leader tip coordinate; xo:position
of zero charge point.

assumed to be the same along the conductor length, the general condition for
an uncharged conductor is
1 x*
C 1 [ U - Uo(x)]dx=O; U=-/ U ~ ( Xdx.
) (4.12)
x2 - X I XI
The condition of (4.12) defines the conductor potential U , which generally

varies in time during the leader development (it is constant only if the electric
field and the binary leader are symmetrical relative to the centre of the
initiating body, which is also symmetrical). The conductor potential is
equal to the average external potential along its length. The leader velocity
can be calculated from (4.2). It was pointed out above that the available
theory cannot provide a clear physical expression for the velocity of even a
relatively simple, continuous positive leader, let alone a negative stepwise
one. It is this circumstance which makes one resort to the empirical formula
(4.2) derived from results of laboratory experiments with currents up to
lOOA and justifiable, to some extent, for positive leaders. No similar
measurements for negative leaders are available, and this is especially true
of natural lightning observations. So, one has to rely on close experimental
data on breakdown voltages in superlong gaps at the sign reversal of the
high voltage electrode, as well as on the moderate velocity differences
between positive and negative lightning leaders. The deviations of their
measured values usually overlap. These facts provide good grounds for
extending expression (4.2), as a first approximation, to negative leaders. In
the latter case, we mean the average velocity neglecting the instantaneous
effects of stepwise development. This approximation is the more so justifiable
that the direct dependence of the leader velocity on the potential difference at
the tip, AU,, raise no doubt and that the variation of the factor a or of the
power index in (4.2) cannot change the picture qualitatively.
Thus, with the account of the x-axis directions and velocities, as well as
the signs of AU at the leader tips, the equations for the leader development
can be written as
= -a[U - Uo(x1)]lI2
dx2
-=a[Uo(x2) - U ]112 . (4.13)
dt ’ dt
Together with (4.12), expressions (4.13) describe the evolution of the two
leaders starting from the body ends, whose coordinates xl0 and x20 are
given as the initial conditions for equations (4.12) and (4.13). The sign
reversal point of the conductor, xo(t), defined by the equation
U(t) = UO(x0) is displaced, during the leader propagation, in accordance
with the nature and degree of field non-uniformity along the channels.
Having solved the equations, one can find the currents at the leader tips
from (4.3) with L = x2 - x l . Generally, they differ quantitatively from one
another and from the current in other channel cross sections, including the
sign reversal at point xo,through which the total charge flows during the

polarization. The current i(xo)= io is defined as
l'o = - = - - Ql = C1r [ U - Uo(x)]dx. (4.14)

dt dt ' XI
This current is used for changing the charge of the old leader portions,
increasing or decreasing them as A U ( x , t), and for supplying charge to its
new portions. This leads to the current variation along the channels, which
can be found by solving the problem.
For some simple distributions of Uo(x),the division of equations (4.13)
by one another
(4.15)
allows the functional relationship between x2 and x1 to be found from

squaring, after which finding the final result x1( t ) , x2(t) reduces to squaring,
-
too. This becomes possible if the cloud field is approximated by the point
charge field Uo(x) IxI-', and if a new variable z = x2/x1 is introduced.
The resultant formulas allow an analytical treatment of some characteristic
relationships. To avoid cumbersome derivations, we invite the reader to do
this independently, while we, instead, shall present some numerical calcula-
tions for several variants.
The calculations prove to be quite simple in integrating the set of
equations (4.13) and (4.14) as well as in the case of a more rigorous approach
to the problem, when the charge distribution along the conductor length is
found from an equation similar to (4.7). Figure 4.7 demonstrates the propa-
gation of vertical leaders in the field of a cloud dipole (with the allowance
for the earth's effect). The calculation was made using an equation similar
to (4.7). The initiating vertical body is located between the lower negative
charge of the dipole and the earth, being displaced horizontally by
Y = 500m from the charge line. As the ascending leader moves up, its tip
approaches the bottom charge centre and enters a region of an ever increas-
ing field. The descending leader moves more slowly towards a weaker field.
The external field potential approaches zero at the earth but increases rapidly
near the charged cloud. As a result, the negative potential of the conductor
made up of the leader channels, U , rises with time, with the sign reversal
point xo going up closer to the cloud. At the initial moment of time, the
potential is U = -27 MV and the point is at an altitude xo = 1603m.
When the ascending leader reaches the charged centre 17 ms later, we have
the altitude xo = 2040 m and U = -64 MV. The absolute potential rise
stimulates the descending leader, increasing its velocity by a factor of three
during this time in spite of its propagation through an ever decreasing
external field. The calculations made with (4.13) and (4.14) have yielded
similar results.

"."I . I . , . I . I ,
0 5 10 15 20
Time, mc
Figure 4.7. The propagation of leaders from a metallic body located between the
cloud and the earth (Q, = -1OC, H = D = 3 km, Y = 0.5 km).
Figure 4.8 illustrates the propagation of one of the leaders in a zero

external field and refers to the situation when the descending leader has
suddenly changed its direction for some reason at a certain height to
follow the equipotential surface, i.e. along the zero field. The calculation
was made with (4.13) and (4.14). Similar to the first variant, this situation
exhibits a remarkable property of a binary leader. The leader developing
along a rising field sustains the other leader, which has travelled in less
favourable conditions, allowing it to move with a certain acceleration even
in a zero field.

1.4-
1.2-
2 1.0-.
0 .
60.8-
.e
7B 0.6-. leader moving along

equipotential path
S
8
0.4-
0.2 -
.
. \
0 . 0 - I
10 15 20 25
Time, ms
4.0 1.6
1.4
3.5
E.
l3.0
J
.s7
U
2.5
7
upward leader
/rd downward
leader
1.2
1.0 2
-
0.8 3
5
0
*
0.6
$
.I
0.4
2.0
0.2
1.5 0.0
I 1'5 ' 20 25
Time, ms
Figure 4.8. The development of a leader pair from a metallic body at 1.5 km above the
earth in a cloud dipole field ( H = 3 km, D = 3 km, Q, = - 10 C ) .At the moment of time
N 10 ms and 1 km altitude, the descending leader turned to follow an equipotential path:
(top) leader velocities; (bottom) position of the zero charge point (T = 0), the altitude of
the ascending leader tip and the length of the portion along the equipotential path.
Where an isolated conducting body may initiate a lightning discharge

depends, to some extent, on a mere chance. A leader may start under a
storm cloud, as in the illustrations just described, or inside a cloud at the
height of the lower or upper charges or somewhere between them. These
variants differ considerably in the polarization charge distribution along
the conductor and, hence, in the leader propagation conditions. A situation
may arise when the positive leader penetrates into the field of the negative
lower cloud, thereby transporting a positive charge to the earth, which

Lightning excited b y an isolated object 157
must apparently be ‘attracted’ by the negative cloud. (In the two situations
above, it was the negative leader that travelled to the earth, ‘naturally’
extracting a negative charge from the cloud). This exotic situation arises
when two leaders are excited from a body located somewhat above the
negative charge centre. The negative leader then goes up to a positive
cloud. The strong field created by the cloud dipole induces a large negative
charge in the ascending leader, displacing the positive charge down. The
average external potential between the two leader tips, U < 0, appears to
be ‘more positive’ than the external potential of the lower tip, Uo(xl) < 0,
so that AU,, = U - Uo(xl) > 0. This is the reason why the descending
leader is positive. With time, when the ascending tip comes closer to the
positive charge, the potential U of the binary system does become slightly
positive (3-3.5 MV), making up several percent of 1 Uol,,,. Then it persists
as such, sustaining the descending leader travel to the earth. By the
moment of contact with the earth, the positive charge is distributed along
the channel in about the same way as in a grounded conductor in a negative
cloud field, being mainly concentrated at the height of this charge. For this
reason, the return stroke current, which only slightly contributes to the
charge after the contact, is weak. The return stroke can be said to make no
contribution to the charge redistribution, since the channel potential
should be corrected only slightly (as compared with I Uolm,, x 100 MV), by
reducing it from 3 MV to zero. This reduction enhances, though only slightly,
the ascending leader, which travels on until it stops high above the positive
charge of the storm cloud.
When an isolated body initiates two oppositely directed leaders, it does
not always happen that the descending positive leader reaches the earth. For
the contact with the earth to take place, the potential U of the conductor
made up of the two leaders must become positive at a certain moment. Other-
wise, the descending leader will stop at the point xls, where the negative
leader tip potential U,, will be by a small value of AVtm,, higher than the
negative potential of the external field (assuming AVtm,,= 0, when U is
equal to U o ( x l s ) )The
. condition for the average conductor potential to be
positive at the moment of contact with the earth is described by the inequality
JZ dx > o
u0(x) (4.16)
which follows from (4.12). Here, Uo(x) is the cloud dipole potential given by
(4.10) with the allowance for its reflection from the earth, and x2 is the
altitude the ascending leader tip has reached by that moment. In principle,
there are no reasons for this inequality to be violated, since the integral of
(4.16) in the limit x2 = CO is necessarily positive (and equal to
-21n(l + D/H)Q,/4mo at r << H ) for the negative lower cloud charge
(Q, < 0). This means that the descending positive leader has a chance to
reach the earth - this only requires that the ascending leader should reach

a sufficient altitude. If the horizontal channel displacement from the vertical

line crossing the centres of the cloud charges, Y << H , we obtain x2 > 2 . 4 7 8
from (4.16) and (4.10) at D = H. The ascending leader must cover a distance
of about 0.5H above the upper positive charge centre.
4.3 The descending leader of the first lightning component

4.3.1 The origin in the clouds
Although lightning observers are familiar with the propagation of a descend-
ing (negative) stepwise leader, the conditions and the mechanisms of its
origin are literally foggy. No one has ever observed the lightning start or
its development in the clouds. Its origin cannot be totally reproduced in
laboratory conditions, although negative stepwise leaders have been pro-
duced experimentally (section 2.7). But the conditions for their initiation
by a high-voltage metallic electrode connected to the condenser of a impulse
generator have little in common with what actually occurs in the clouds - a
cloud is not a condenser winding and, of course, not a conductor. The
negative cloud charge is scattered throughout the dielectric gas on small
hydrometeors. It is very hard to perceive how the charges, fixed to particles
with low mobility and dispersed in a huge volume, can come together to form
a plasma channel in a matter of a few milliseconds.
In our terrestrial practice, we encounter events somewhat similar to the
spark initiation in the clouds. Investigation of what has caused an explosion
or a fire in industrial premises containing an abundance of electrostatic dust
particles or droplets can provide evidence for a spark discharge arising in a
medium with a dispersed charge. Lately, there have been reports of studies
with gas jet generators ejecting into the atmosphere miniature electrically
charged clouds [3,4]. Sometimes, extended bright structures of about 10 cm
in size were observed along a charged spray boundary; on some occasions,
they were observed to form spark channels of about 1m in length. Unfortu-
nately, no measurements could be made of the field at the discharge start, so
the fact of discharge excitation was only stated. Therefore, one can do
nothing more than just make conjectures about the excitation mechanisms
of lightning in the clouds and of sparks in laboratory sprays.
Speculations concerning these mechanisms (the only type of conclusion
we can draw today) have to be arrived at via the process of elimination. A
cloud medium cannot be considered as being conductive when we speak of
current supply to the leader channel. Common charges are not transported
directly to the leader, nor do they leave the cloud by themselves during the
fast leader process. Therefore, the cloud charges play a different role -
they are the source of electric field which ionizes the air molecules, producing
the initial plasma, and then sustains the leader process. To fulfil the first task,
the field somewhere in the charged region is to exceed the ionization

The descending leader of theJirst lightning component 159
threshold (Ei M 20-25 kV/cm at the height of the cloud charge) or the cloud
is to contain inclusions enhancing the field locally via the polarization charge.
It seems that neither mechanism should be discarded entirely, although cloud
probing rarely registered fields exceeding several kilovolts per centimetre.
These results do not testify to the absence of higher fields, because most of
the measurements concerned fields averaged over lengths of several dozens
of metres. No measurements were made at the moment of lightning initia-
tion, because the probability of a detector registering a field at the right
place at the right moment is extremely low. On the other hand, the conditions
necessary for the excitation of a leader process in a cloud are quite rare;
otherwise, the number of lightning strikes per square kilometre of the earth’s
surface would greatly exceed 2-5 per storm season.
Let us estimate the volume to be occupied by a cloud charge capable of
creating an ionization field. It was mentioned above that the field Eo at the
earth was often found to be lOOV/cm during thunderstorms. This value
should not be considered to be the cloud dipole field, since the near-earth
charge provided by microcoronas from various pointed objects attenuates
the cloud field at the earth. A similar value is obtained from a small positive
charge supposed to lie under the principal negative charge [5]. Taking, for
estimations, the intrinsic dipole fields Eo to be 200V/cm and the heights of
the lower (negative) and the upper (positive) charges to be x = H = 3 km
and H + D = 6 km, respectively, we find, from (4.10), the dipole charges
Q, = 13.3C. These values will serve as guidelines in further numerical calcu-
lations. The charge Q, can create field Ei M 25 kV/cm at its boundary if it is
distributed throughout a sphere of radius R, = 220 m. Measurements show
that the charged region is, in reality, 2-3 times larger, but one should not
discard the possibility of a short accidental charge concentration in a smaller
volume due to the action of some flows in the clouds.
More probable is the situation when a macroscopically averaged maxi-
mum field of cloud charge is several times lower than Ei and local fields,
enhanced to Ei x 25 kV/cm, arise near polarized macroparticles. Note that
the maximum field near a metallic ball polarized in an external field E is
E,,, = 3E. Similarly enhanced is the external field of a spherical water
droplet, since water possesses a very high dielectric permittivity E = 80 and
E,, = 3 E ~ / ( + 2 E ) . Therefore, if charge Q, is concentrated in a sphere of
& times larger radius, R, = 380 m, the field three-fold enhanced by polari-
zation can achieve the ionization threshold. Following the ionization onset,
streamers may be produced around large droplets, giving rise to a possible
leader, because streamers may be branched and extended in an average
field of -10 kV/cm.
Leaving aside the mechanisms of ionizing fields and leader origin,
because they are still poorly understood, we shall take for granted only the
mere fact that a leader does occur. At its start, a descending leader is
devoid of the possibility of taking the charge it needs away from the cloud.

Observations show that this charge is quite large: an average negative leader
transports to the earth a charge QL RZ -5 C and, sometimes, it is as large as
-20 C [l], a value close to the evaluations of Q, for the storm cloud. But if the
cloud charge remains ‘intact’, the only thing that can provide the charge
balance is the ascending leader of opposite sign, which is to develop simulta-
neously with the descending leader. This idea was suggested in [6],which
presented a qualitative distribution of the charge induced along a vertical
conductor made up of two leaders prior to and following its contact with
the earth. What happens is principally the same as in the excitation of two
leaders by a conducting body isolated from the earth and is affected by an
external field (section 4.2). This process is independent of the descending
leader sign; therefore, one should not think that a negative cloud can produce
only a negative leader while a positive cloud always produces a positive one.
In any case, two oppositely charged leaders are produced simultaneously,
and which of them will travel to the earth depends on the charge position
in the cloud and on the leader starting point.
A binary leader is most likely to be initiated near the external boundary
of the charged region, because the field there is highest. The field at the
centre of an isolated charged sphere is zero. In the case of a uniform
charge distribution throughout its volume, the field rises along the radius
as E r but decreases from the outside as e rP2 with the maximum
N N
E,, = Qc/47r~& at the boundary. For a dipole configuration of real

charges, the field does not practically vary across the boundary surface of
the charged region. For the above values of D,H and R,, the field at the
upper point of the lower sphere is about 5% higher than at the lower
point. The probability of a binary leader being initiated at either point is
nearly the same. However, the final result of the binary leader development
will differ radically, and this circumstance was essentially demonstrated in
section 4.2. If both leaders are initiated at the bottom edge of the lower
negative charge, the negative leader will go down and the positive one will
go up. The negative leader has a real chance to reach the earth with a high
negative potential equal to that of cloud charges averaged over the whole
conductor length. The conductor is mostly in the region of high negative
potential, nowhere entering the positive potential domain. The closing of
this highly charged channel to the earth leads to the wave processes of
charging and charge exchange (the return stroke) involving high current.
The latter represents a real hazard. This is what happens in the case of a
negative lightning. If a binary leader is initiated at the upper boundary of
the lower charge, the positive leader goes down to the earth and the negative
one goes up. A positive descending leader can never reach the earth unless it
acquires a positive potential. For this, its ascending partner must necessarily
go beyond the zero potential point, closer to the upper positive charge of the
dipole. Owing to the compensation of positive and negative charges at
various sites along the path, the average potential transported down to the

The descending leader of the first lightning component 161
earth is quite low. This actually cancels the return stroke current. Positive
lightnings with very low currents of about 1 kA present no danger and are
quite frequent. The number of their registrations by the observer increases
with increasing sensitivity of the detectors used.
4.3.2 Negative leader development and potential transport

The stepwise propagation pattern has been believed by many to be the
principal problem for a theoretical description of a negative leader [7].
However, it is of little importance to the leader evolution whether it develops
continuously or by relatively short steps. The leader current and velocity are
averaged over many steps. Averaged also is the channel energy balance,
although the energy release at a distance of several step lengths from the
tip has a well defined periodic pulse character. We shall discuss the stepwise
effect in section 4.6, following the consideration of the return stroke, since
this process is involved in every step as the main component.
The evolution of the descending channel of a binary leader is intimately
related to that of its ‘twin brother’ - the ascending leader. (In t h s sense, the
term ‘Siamese twins’ would be more appropriate.) A characteristic feature of
the twins is the break-off of their potential, which varies but little along
their highly conductive channels, from the external potential at the start. In
this respect, a lightning leader differs considerably from a laboratory leader
starting from an electrode connected to a high-voltage source. Being ‘tied
up’ to the electrode, a laboratory leader with a well conducting channel carries
the electrode potential, which may be close to the source emf. Generally, it is
lower than the emf by the value of the voltage drop across the external circuit
impedance when a discharge current is flowing through it. The underestima-
tion of the principal difference between a laboratory spark initiated from a
high-voltage electrode and a natural electrodeless lightning leads to erroneous
attempts to derive from observations the voltage drop value across the leader
channel. The reasoning is usually as follows. The potential U , transported by a
lightning leader to the earth can be estimated from the return stroke current
and the characteristic channel impedance (section 4.4). The cloud potential
U,, can also be estimated (see formula (4.17) below). The leader channel
base has the same potential - as if the cloud were an electrode. Therefore,
the voltage drop across the leader length, from the cloud to the earth, is
AU, = lUoRl - lull, and the average field in the channel is expressed as
E, = A U J L , where L is the leader length ( L M H , or 30-50% greater with
the allowance for the path bendings). Such estimations lead to incredibly
large values of A U, 100 MV and E, M 1 kV/cm. A mature leader channel
with i ~ 1 0 0 A current cannot have such high fields. Its state is very much
like that of the quasi-equilibrium hot plasma in an arc, which has a field 1-2
orders of magnitude lower. This follows from theory and from evaluations of
fields in superlong laboratory sparks.

The attempts to solve the ‘inverse’ problem using the expression

1 U,,l = AU, + 1 U 11 to calculate the storm cloud potential have also failed.
When one includes in this expression the generally correct values of arc
field E,, one gets unjustifiably low cloud potentials U,, inconsistent with
atmospheric probing measurements and other calculations.
The methodological error of both approaches is due to the rigid relation
of the base potential of a descending leader to the external field potential at
the leader start. In actual reality, the channel potential undergoes a consider-
able time evolution, being determined by the polarization charge distribution
along the binary leader length. When the descending leader approaches the
earth, its base potential may differ significantly from the potential created
by the cloud charge at the start site at the moment of start.
A simple calculation of the leader development can be made from
equations (4.13) and (4.12), but a more rigorous solution can be obtained
from equations (4.13) and (4.7), which were used for that purpose in section
4.2. One should also bear in mind that if both leaders start from the
boundary of a charged cloud region, at least one of them will enter the
charged volume and may even pass through its centre. Then, we have to
discard the point model of a cloud dipole and make the next approximation
by assuming that charge Q, is distributed uniformly with the density
3QC/47rRf in a sphere of radius R,. Inside the sphere, the potential of its
intrinsic charge is radially symmetrical and is equal at point r to
3Qc r d R,. (4.17)
The potentials from the upper dipole charge and from charges reflected by
the earth can be found as from point charges. They do not contribute
much to U,$. For example, for the centre of a negative sphere with
Q, = -13.3C and R, = 500m, we have U,, = -360MV and at the
5
boundary U,, = U,, = -240 MV. With all other charges taken into
account, we get U,, = - 196 MV for the bottom edge of the lower sphere
at H = D = 3km.
Figure 4.9 presents the results of this calculation including those for the
charge distribution along the conductor length from an equation similar to
(4.7). We have evaluated the development of both leaders along the dipole
axis, following the start from the bottom edge of the lower negative
Figure 4.9. (Opposite) The model of a descending leader from the lower boundary
of the negative dipole charge (Qc = -12.5C, H = 3 km, D = 3 km, R, = 0.5 km).
Vertical channels have no branches: (top) tip positions of the negative descending
leader, x l , and its positive ascending partner, x2,with the points of zero potential
differences, xo; (centre) charge distribution along the leader channel; (bottom)
potential and velocity of the descending leader.

2-
1-
--.
E
U 0
E 4
I-" -1 -
-2 -
-3
Time, ms

sphere. The dipole potential from (4.10) was used as Uo(x), except for the
length in the charged region, where expression (4.17) was employed with
r = jx - HI. The descending negative leader of a binary system is accelerated
quickly after the start. Having covered about 500 m, it travels farther to the
earth with a slightly decreasing velocity wL x (1.6-1.7) x 105m,’s, a value
close to observations. The leader strikes the earth in 16ms. By that time,
the ascending leader has reached the height x2 = 3.6 km. This is far even
from the zero potential point located at x x 4.5 km, let alone from the
upper positive charge located at an altitude of 6km. The descending
leader, which started from a site with the local potential - 185 MV, trans-
ports to the earth nearly half of this value, U1 % - 105 MV, in spite of the
initial assumption of the zero voltage drop across the channel assumed to
be a perfect conductor.
The reader should not feel discouraged by the large calculated value of
U1, which is more appropriate to record strong lightnings rather than to a
common lightning discharge, especially considering that the cloud param-
eters taken for the calculation were quite moderate. It will be demonstrated
in section 4.3.3 that leader branching, which is a rule rather than an excep-
tion, reduces considerably the potential transported down to the earth. It
is quite likely, however, that lightnings of record intensities are produced
in ordinary clouds rather than in those having a record high charge, but
only if the descending leader does not branch (or does so slightly).
The above calculation for an ideal situation with unbranched leaders is
interesting and useful for two reasons. First, one should understand the
physics of a simple observable phenomenon before one turns to its complex
modifications. The other reason is, probably, more important. Practical
lightning protection requires the knowledge of both typical average lightning
parameters and their record high values. It is the latter that become more
important in designing prospective measures for especially valuable con-
structions and objects. As was pointed out above, the case of an unbranched
leader just discussed is likely to be one of the rare but most hazardous
phenomena.
The potential U1 transported by a lightning leader to the earth is an
important parameter for practical lightning protection. The return stroke
current (section 4.4), the most destructive force of lightning, is proportional
to U1. The nature of U1 becomes clear from the above conception of
descending leader development in a binary leader process. Ideally, potential
U1 is that of a perfect conductor, made up of two leader channels, at the
moment of its contact with the earth. But the ascending and descending
leaders develop differently, because their paths cross regions possessing
different distributions of cloud potentials Uo(x). The descending leader
travels nearly without retardation because the potential difference at its
tip, AU,, = U - Uo(xl), remains almost constant (a decrease in 1 U1 is largely
compensated by 1 Uo(x)Idecreasing towards the earth). The ascending

I”
Figure 4.10. Estimation of the potential transported by a negative leader to the earth.
positive leader moves ‘against’ the field and soon enters a region of a rapidly
rising external potential; as a result, AU,, = U - U0(x2) becomes relatively
low soon after the start. This leads to a lower velocity of the ascending
leader, which now goes up ‘unwillingly’, being affected by its more active
twin which moves faster, pumping its charge into it. For this reason, just
before the descending leader contacts the earth, the total potential of the
system nearly coincides with the external field potential U0(x2) at the site
of the ascending leader tip (AU,, = U - Uo(x2) << 1 Ul).
This circumstance makes it possible to determine the transported poten-
tial U1 = U just from the condition U = U0(x2) at x 1 = 0. The condition has
a clear geometrical interpretation (figure 4.10). The shaded regions between
the external potential curve Uo(x) and the horizontal line intercepting it must
be identical on both sides of the left-hand interception point (the point of the
conductor sign reversal, xo).This results from the net polarization charges of
both signs being identical; they are proportional to the shaded regions (see
formula (4.12)). This approach can be used to find U1 in different charge
distribution models and for different horizontal deviations of the vertical
leader path from the dipole axis. In a simple case when both leaders propa-
gate along the dipole axis, formulae (4.12) and (4.10) with D = H and r = 0,
together with expression (4.17), yield a dimensionless equality for finding
point x2 and then U 1 :
(4.18)
For the variant shown in figure 4.9 with n = i, expressions (4.18) give
e2 = 1.27 and U / U O R= 0.63, in a fairly good agreement with the
calculations of the leader evolution (note that U,, RZ U,, is the external

field potential at the bottom edge of the lower cloud charge that has
triggered both leaders).
A negative descending leader can start from any point on the lower
hemisphere of the bottom negative charge of the cloud. The location of
this point is quite likely to be a matter of chance, since the field across the
surface is nearly uniform. Depending on the location of the starting point,
the ascending twin crosses the charged region along chords of different
lengths, and this, along with the other factors, affects the potential trans-
ported to the earth. The maximum potential U1 at the moment of contact
with the earth is characteristic of a leader that has started from the lowermost
point of the charged sphere, when the paths of both twins cross the regions of
maximum potentials and the ascending leader path in it is longest. The value
of U1max is about 60% of the external potential U,, at the start and 40% of
the maximum U,, value at the centre of the negative cloud charge. But even
if the descending leader is initiated near a lateral point of the hemisphere
located at a maximum distance from the dipole axis, it transports a
considerable potential found from calculations to be 0.65U1max x O.4UoR.
Therefore, an unbranched negative leader transports to the earth a high
potential, (0.6-0.4)UOR, no matter where it has started from the lower
hemisphere.
4.3.3 The branching effect

Measurements of return stroke current show that a descending negative
lightning rarely transports to the earth a potential as high as 100MB
(Z, = U 1 / Z , where 2 is the channel impedance; see section 4.4.2). The
reason for this is not the supposedly lower potentials of most clouds. The
value of -100MV is characteristic of cloud charges moderate in size and
density. The reason is most likely to be the leader branching, since an
unbranched leader is an exception rather than the rule. Numerous downward
branches of a descending leader can be well seen in photographs. Although
ascending leaders are screened by the clouds, their branching can be
registered by radio-engineering instruments [8- lo]. However, the potential
U 1 is affected by the branches of a descending negative leader rather than
of its positive ascending twin brother.
Let us make sure first that the branches of an ascending leader do not
change the situation much. In the limit of a very intensive branching, the
negative cloud bottom, pierced by numerous conductive channels, is electro-
statically identical to a continuous conductive sphere of capacitance
C, = 4mORc,whose charge has been pushed out on to the surface. The net
charge of a system made up of a sphere and a negative leader attached to
it remains equal to the initial charge -Qc. With the neglect of the voltage
drop across the descending channel, the binary system possesses the same
potential U along its length. At the moment of contact with the earth,

when the leader capacitance CL corresponds to the length L x H , this

potential is
U=-=--
+ C, - 1 + L/2R, ln(L/R) (4.19)
Qc cc UOR
cc+ CL Cc
where UoR would be the boundary potential of a charged sphere, were it
isolated (the small capacitance gain due to charges induced in the earth are
ignored). For L/R x 100 and H/Rc = 6, as in the previous numerical
illustration, the potential U1 x O.6UoRis nearly the same as that transported
to the earth in the absence of ascending leader branching. The potential of
the cloud-leader system drops because of the outflow of some of the cloud
charge to the new capacitance of the descending leader just produced. A
similar effect has been observed in long laboratory sparks. The capacitance
of an extremely long spark is often only one order of magnitude smaller
than the output capacitance of a impulse voltage generator, connected
directly to a gap without a large damping resistor. The charge inflow into
the leader is quite appreciable and reveals itself as a voltage drop across
the gap.
A branched descending leader possesses a larger capacitance than an
unbranched one; it takes away a higher charge from the cloud and decreases
the potential more. To estimate this effect, let us represent a branched leader
as a bunch of n identical conductors of radius R and length L, spaced at
distance d (L > d >> R). Supplied by the same power source, they possess
the same potential U and linear charge T . The potential at the centre of
any of these conductors is found by summing the potentials of all charges
of all conductors, including the intrinsic potential. Integration with the
neglect of the small effect of the earth yields
The total capacitance of the n conductors, Ctn = nrL/U, is larger than that
of a single isolated conductor, but this gain is less than n-fold:
Ct n - nln(L/R)
ct1 ln[(L/R)(L/d)n-']
The reduction in the potential transported to the earth roughly follows
the distribution of the cloud charge Q, between the capacitances of the
charged cloud cell, C,, and of the leader, Ctn,described by the first equality
of (4.19). For n = 10 branches separated at distances d = L/3 and
L/R x 100 derived from photographs, the capacitance is Ctl0x 3.2Ct1. This
well-branched leader will transport to the earth potential U1 x 0.3Uo~.In
view of the real length of a leader (especially, a well-branched one) which is
about 1.5 times longer than the charge height H , i.e. L M 1.5H, the

168 Phjsical processes in a lightning discharge
potential decreases to O.2UoR,in good agreement with the data on negative

lightning currents. The number of branches and their lengths vary randomly
with the lightning. The potential U1 determining the return stroke current
vary together with them. This variation is likely to produce a wide range
of current amplitudes. The variation of storm cloud charges seems to be
less significant.
There is another source of reduction in the potential transported by a
negative leader to the earth. In more complex models than the vertical
dipole variant, the reduction is due to a low positive charge assumed to be
present at the very bottom of a cloud [5]. Calculations show that if a positive
4 C charge of 0.25 km radius with the centre at 2 km above the earth is added
to the above dipole with Q, = 113.3C, R, = 0.5 km and H = D = 3 km, the
negative leader initiated from the bottom edge of the negative charge will
transport half of the potential to the earth.
4.3.4 Specificity of a descending positive leader

Positive leaders do not occur very frequently. Statistics indicate that in
Europe their number is 10 times smaller than that of negative ones. But it
is quite likely that their actual number is larger than the number of their
registrations. It was pointed out in section 4.3.1 that a descending positive
leader does not carry high potential to the earth and that its return stroke
current is low. For this reason, the electromagnetic field of a positive
lightning discharge can be detected at a much shorter distance than that of
a negative discharge and, probably, not all of them are registered.
If the bottom charge of a cloud dipole is negative, a positive descending
leader may start either from the upper negative hemisphere or from the
bottom hemisphere of the upper positive charge. The leader will reach the
earth, transporting to it a positive potential, provided the condition of
(4.16) is met. With a small deflection of the leader vertical axis from the
dipole axis (Y << H ) , the transported potential found from (4.12) and (4.10)
with x = 0 will be
where H1and H2are the heights of the bottom and top charge centres and x2
is the ascending leader height at the moment the descending leader contacts
the earth. We mentioned at the end of section 4.2.2 that an ascending leader
must go up at least to x2 = 2.47H1 at H2 = 2H1; then we have U = 0.
At x 2 M 4H1, the function U ( x 2 ) crosses the smooth maximum,
U,,, KZ -Qc/207r~OH1M 8MV, if Q, = -13.3C and H1 = 3 km, as in the
previous examples. Even the maximum potential transported to the earth
is small. This means that the return stroke current of a descending positive
leader travelling along the dipole axis will be low. The potential and the

The descending leader of the$rst lightning component 169
current will be still lower, with the real voltage drop U, across a channel of
total length 4H x lOkm taken into account. Even for the channel field
E x 10 V/cm, the value of U, 10 MV is comparable with U,,. This light-
ning is so weak that it has little chance of being registered and included in the
statistics.
Vertical channels demonstrate maximum positive potentials transported
to the earth. They go through more or less identical regions of negative (at
the bottom) and positive (at the top) external potentials, and the respective
contributions to the integral of (4.12) are mutually compensated. Positive
lightnings, however, can possess very high currents. With the foregoing
taken into account, one can suggest at least two reasons for this. One is a
favourable random deviation of the channel path from the vertical line.
Suppose the ascending leader of a binary system, starting from the upper
positive charge point closest to the earth, xo = H2 - R,, moves up vertically,
while the other leader, having descended to the zero potential point between
the charges, turns aside and goes along the zero equipotential line. After it
has deviated for a large distance r from the dipole axis, it turns down
vertically to contact the earth this time. In this case, the descending leader
misses the region of high negative potential, and positive contribution to
the integral of (4.12) remains uncompensated. Calculations with formulae
(4.12), (4.10) and (4.17) made at H I = 3 km, H2 = 6km, R, = 0.5 km, and
r = 1 km show that the descending leader will transport to the earth a poten-
tial 4.3 times greater than that to be transported along the dipole axis.
Another principal possibility is the deviation of the dipole axis itself
from the vertical line, with the vertical leader path preserved. The centres
of the top and bottom charges can be shifted from the same vertical line
because of the difference in the wind forces at different heights. Then the
leader that has started up vertically from the top charged region passes
through the region of high positive potential, while its twin, descending ver-
tically, will appear to be shifted aside relative to the bottom charge and go
through the region of low negative potential. The effect will be the same as
in the first case. Quantitatively, it may even appear to be stronger, since
the length and capacitance of the descending leader are smaller due to the
lack of an extended path along the zero potential line.
4.3.5 A counterleader
The descending lightning leader does not reach the earth or a grounded body,
because it is captured by the ascending leader developing in the electric field
of cloud and earth-reflected charges. This field is enhanced by the charge of
the descending leader approaching the earth. This can also happen in labora-
tory conditions, especially if the descending leader is negative. Then the
counterleader is positive and requires a lower field for its development.
Streak pictures of laboratory sparks clearly show the counterleader start

Figure 4.11. A streak photograph of a long spark with a counterleader coming from a
grounded electrode.
and motion towards the descending leader (figure 4.11). The altitude at which
their encounter occurs depends on the descending leader sign and charge.
The length of the counterleader at the moment of their contact is important
for lightning protection practice, because it defines the number of strikes at
bodies of different heights and, to some extent, the current rise parameters
of the return stroke from the affected body.
Let us estimate the altitude z, which the descending leader tip is to
reach to be able to create a field at the earth high enough to produce a
viable counterleader. The latter does not differ from any other ascending
leader, and its development from a body of height d requires that the
near-terrestrial field should exceed the value of Eo from formula (4.1 1).
For the height d = 30m characteristic of industrial premises, the field
must be Eo x 480V/cm. If the cloud field is -lOOV/cm, the field
AE = 380V/cm must be created by the descending leader with charge.
The main contribution to the near-terrestrial field is made by the charge
concentrated at the leader channel bottom. Therefore, the calculation of
the field AE under a very long vertical conductor should utilize the constant
value of r averaged over this bottom of length -z, rather than the linear
density of the non-uniform charge r(x).With the charge reflected by the
earth, we have
r "dx - r u-uo -U
(4.21)
*E(z) =GIZ T - G - z l n ( L / R ) -zln(H/R)
N

Return stroke 171
where U is the channel potential and L x H is its length, which is about the
cloud height at the tip height z << H. Here, we have used the conventional
expression T = C1( U - U o ) with average linear capacitance and accounted
for the near-terrestrial potential of the cloud, 1 Uoi << 1 UI. For an unbranched
descending leader carrying high potential U x 50 MV, we obtain z x 260 m
at ln(H/R) x 5.
The counterleader arises at the last stage of the descending leader
development, i.e., near the earth. Its velocity is not high and is equal to
wL, x 2 x 104m/s from the first formula of (4.5), because the potential
difference on the leader tip is quite low, A U x Eod x 1.5 MV. The descend-
ing leader has an order of magnitude higher velocity. For this reason, the
counterleader acquires the length L1 x (wL,/vL)z x 25 m by the moment of
encounter. This is a large value, since the length L1 is summed with the
body’s height d, so that the total height of the grounded conductor becomes
nearly doubled. This affects the frequency of the body’s damage by lightning
strikes.
It follows from formulae (4.21) and (4.11) that the height z,to which the
leader descends before it can initiate a counterleader, is greater for higher
premises, from which the counterleader starts, z d3I5,although this depen-
N
dence is not very stringent. It is important that as the altitude of a body and z
become greater, the counterleader has more time for its acceleration and can
acquire a longer length. It is important for applications that it is not only the
length L1 which increases but also the L1 / d ratio.
The simple estimation obtained from (4.21) and (4.11) can be refined by
accounting for the T ( X ) non-uniformity in the integral of (4.21) arising from
the proportionality T U - U 0 ( x ) and by rejecting the approximation of
N
constant linear capacitance C1. In the latter case, T ( X ) should be found

from equation (4.7). Calculations show that the two corrections are rather
small, so the estimations above can be considered to be satisfactory.
4.4 Return stroke

4.4.1 The basic mechanism
A return stroke, or the process of lightning channel discharging, begins at the
moment the cloud-earth gap is closed by a descending leader. After the
contact with the earth or a grounded body, the leader channel (it will be
taken to be negative for definiteness) must acquire zero potential, since the
earth’s capacitance is ‘infinite’. Zero potential is also acquired by the ascend-
ing leader, which is a continuation of its descending twin brother. The
grounding of the leader channel carrying a high potential leads to a dramatic
charge redistribution along its length. The initial channel distribution prior
to the return stroke was T~ = C1[Ui- U 0 ( x ) ]Here
. and below, the potential
transported to the earth, which acts as the initial potential for the return

Figure 4.12. Schematic recharging of a lightning channel after the contact of the
descending leader with the earth. Shaded regions, charge; ( a ) moment of the leader
contact with the earth; (b) the return stroke reaching the upper channel end; (c)
charge change.
stroke, will be denoted as Vi. As before, it will be taken to be constant along

both leader lengths, and the voltage drop across the channel will be ignored
as an insignificant parameter. We shall assume that the channel is character-
ized by linear capacitance C1, which does not vary along its length or in time
during the return stroke process. After the whole channel has acquired zero
potential, U = 0, the linear charge becomes equal to T~ = -C1 Uo(x). The
channel portion belonging to the negative descending leader does not just
lose its negative charge but it acquires a positive charge (Uo < 0, r0 < 0,
T~ > 0). Not only does it become discharged but it is also recharged. The
twin positive channel high in the cloud acquires a larger positive charge
(figure 4.12).The linear charge variation for the return stroke lifetime is
AT = T~ - T~ = -CoUi. At Ui(x) = const, the charge variation is constant
along the channel length and has such a value as if a long conductor (a
long line) pre-charged to the voltage Ui becomes completely discharged (as
if it were r0 = C1Vi to become T~ = 0).
It has been emphasized that the leader charge is concentrated in its
cover. The charge in a non-conducting cover changes due to the charge
incorporation from the conductive channel, owing to the streamer corona
excitation at the channel surface. This is an exceptionally complicated pro-
cess, whose rate can be found only from an adequate theoretical treatment.

Return stroke 173
For this reason, the assumption of capacitance C1 being constant, which

implies a zero-inertia charge variation in the cover with varying channel
potential, is quite problematic. But if we discard this seemingly essential
assumption, nothing will change qualitatively or even quantitatively.
Indeed, suppose the cover charge does not change at all during the time
the whole channel acquires zero potential. This is equivalent to the assump-
tion that the channel capacitance is determined, during the return stroke pro-
cess, by the conductor radius r, rather than by the cover radius R.Because of
the logarithmic dependence of linear capacitance (2.8) on the radius, it
decreases to the value Ci equal to about a half of C1; for example, we
obtain C1 % 10pF/m and Ci RZ 4.4pF/m at I = 4000m, R = 16m and
Y, = 1.5 cm. Then the charge variation during the stroke
A T = T ~- ~ o = [ ( C l- C i ) ( U i - U o ) + C ’ 1 ( 0 - U o ) ] - [ C 1 ( U i - U o ) ]
= -c:ui (4.22)
remains the same in order of magnitude. Consequently, when considering
fundamental stroke mechanisms, one can take C: % C1 and assume the
equivalent line to be charged uniformly.
Measurements made at the earth show that a descending leader is
discharged with a very high current. For negative lightnings, the current
impulse of a return stroke with an amplitude ZM -10-100 kA lasts for 50-
100 ps on the 0.5 level. A short bright tip of the return channel well seen in
streak photographs runs up for approximately the same time. Its velocity
v, M (0.1-0.5)~is only a few times less than light velocity c. It would be
natural to interpret this fact as the propagation of a discharge wave along
the channel; this wave is characterized by a decreasing potential and rising
current. Due to an intensive energy release, the channel portion close to
the wave front, where the potential drops from U, and a high current is
produced, is heated to a high temperature (from 30000 to 35000K, as
shown by measurements). This is why the wave front is so bright. The
channel behind it is cooled due to expansion and radiation losses, becoming
less bright. A return stroke has much in common with the discharge of a
common metallic conductor in the form of a long line. The line discharge
also has a wave nature, and this process was taken to be a model discharge
in shaping the ideas concerning the return lightning stroke.
A lightning channel is discharged much faster than it was charged
during its development with the leader velocity vL % (10-3-10-2)~,. But
the variations in potential and linear charge during the charging and the
discharge are expressed as values of the same order of magnitude: ro AT.
In agreement with the velocity, the channel is discharged with current
-
1, M ATV, by a factor of v,/uL %102-103 higher than the leader current
iL M r0uL ~ ~ 1 0 0 The
A . linear channel resistance R Idecreases approximately
as much during the leader-stroke transition. This decrease is due to the

channel heating by high current. As a result, the plasma conductivity

increases and the channel expands, making the conductor cross section
larger. In this respect, a lightning discharge certainly differs from a discharge
of a common conductive line, whose resistance remains constant (if the skin-
effect is ignored). Since resistance is a plasma characteristic, its decrease can
be found straightforwardly only if the physical processes occurring in the
channel are taken into account. (This situation will be analysed in section
4.4.3.) But this conclusion can be arrived at indirectly from general energy
considerations. Over the return stroke lifetime, t, H/wr, where H is the
N
channel length, the energy dissipated in the channel must be approximately

equal to the initial electrical energy C1U:/2 per unit length:
C1 U:/2 N IhRlt, N IhR1 H / v , N ATHI-WR~. (4.23)
About as much energy was dissipated in the leader when the capacitance
C1 was charged. If the leader develops in the optimal mode (see section 2.6),
to which a natural lightning process is, probably, very close, because Nature
usually takes optimal decisions, the voltage drop across the channel is com-
parable with the excess of the leader tip potential over the external potential.
Therefore, the resistances of the channel and the streamer zone are compar-
able, because the same current flows through them. Therefore, the unit length
of the leader dissipates the same energy C1U,'/2 (in order of magnitude),
expressed by the leader parameters il, vL and R I L similar to (4.23). This
yields R I I M RILiL,i.e., R 1 / R I L 10-2-10-3. It is also found that the
N N
average electric field in the leader channel and behind the discharge wave
in the return stroke, E, M R1IM x RILiL,have the same order of magnitude.
This is consistent with the conclusion to be made from a straightforward
analysis of the established states in both channels. The situation there is
similar to that in a steady state arc. But the channel field E, in a high current
arc does vary but slightly with the current [ 111.
It follows from the foregoing that if a leader has iL M 100A, E, z 10Vi
cm and ROLM 0.1 R/cm, the return stroke must have Ro x 10-3-10-4 R/cm
in the steady state behind the wave front; the total resistance of a channel of
several kilometres in length appears to be lo2 R. This value is comparable
with the wave resistance of a long perfectly conducting line in air, 2,whereas
the total ohmic resistance of a leader of the same length is two orders of
magnitude larger than Z . The ratio of the ohmic resistance of the line portion
behind the wave to the wave resistance indicates the degree of the wave
attenuation during its travel along the line (section 4.4.2). If the channel
resistance were constant and remained on the leader level, the lightning chan-
nel discharge wave would attenuate, being unable to cover a considerable
channel length. The current through the point of the channel closing on
the earth would also attenuate too quickly. Experiments, however, point to
the contrary: the visible bright tip has a well-defined front, and a high current
is registered at the earth during the whole period of the tip elevation. The

Return stroke 175
transformation of the leader channel during the wave travel decreases its
linear resistance considerably, determining the whole return stroke process.
4.4.2 Conclusions from explicit solutions to long line equations

Long line equations with the allowance for the main factor - variation in
linear resistance - can be solved only numerically (section 4.4.4). However,
the nature of the process and its essential physical characteristics can be
understood from the analysis of well-known analytical solutions for simple
situations. Their comparison with lightning observations indicate the impor-
tant points for the formulation and solution of the real problem.
In the absence of transverse charge leakage due to imperfect insulation,
a long line is described by the equations
au -L1-+Rli,
di di
-- =cl--.
dU
(4.24)
dX dt dX at
They generalize equations (2.12) by accounting for inductance. The induc-
tance per unit conductor length, L 1 , as well as its capacitance C1, can be
assumed to be approximately constant. For an isolated conductor of
radius Y, and length H >> rc, it is
Po H H
L1 x -In - = 0.2 In -pH/m (4.25)
27r rc YC
Here, the channel length H is about equal to the height of an ascending

leader tip. For a perfectly conducting line with R1 = 0, equations (4.24)
are re-differentiated to produce a simple wave equation:
a2u 1a2u - 0, 21 = (L1C1)y2. (4.26)
8x2 212 at2
If the line is charged to voltage Ui and short-circuited on the earth by its base
x = 0 at the moment of time t = 0, a rectangular wave of complete voltage
elimination (from Ui to 0) and an unattenuated current wave of the same
shape will propagate with velocity U from the grounding point (figure 4.13):
While the wave propagates along the line, a detector mounted at its
beginning will register direct current. If voltage Ui is low and there is no
charge cover around the conductor, the capacitance and inductance are
characterized by the same radius Y, in the logarithms of (2.8) and (4.26). In
this case, we have

ti
Figure 4.13. Distributions of potential and current during the discharge of a perfectly
conducting line.
where c is light velocity. If one describes the capacitance, in contrast to the

inductance, with the leader cover radius R = 16m, then Y, = 1.5cm,
H = 4km (as above), C1 = 10pF/m, L1 = 2.5pH/m, v = 0.67C, and
Z = 500R. The wave velocity is now lower than light velocity, but not
much. Current of amplitude ,Z = 30 kA, typical of the return stroke of the
first negative lightning component, arises at I Uil = ZZM = 15 MV. The
values of Vi and w are correct in order of magnitude, but the wave velocity
w exceeds several times the observable velocity, and it is impossible to
reduce this discrepancy by varying the reactive line parameters. In a line of
preset length, C1and L1 vary only slightly (logarithmically) with the conduc-
tor radius. What remains to be done is to focus on the only parameter that
has not been accounted for - resistance R1 which is very high in a leader
but reduces by 2-3 orders during a return stroke.
Let us discuss the exact solution of equations (4.24) describing the line
discharge at R1 = const:
i ( x ,t ) = -
Z
(4.29)
where Zo(z) is the Bessel function of a purely imaginary argumentjz:

ZO(Z) 1 + (z/2)2 at z << 1
(4.30)
+
z0(z) M eZ(2~z)-1/2[1~ ( z - ' ) ] at z >> 1.
The wave has the same velocity as an ideal line, without losses, but the
current at the wave front falls exponentially, as it propagates:
(4.31)

Return stroke 111
The attenuation if is described by the ratio of the ohmic resistance Rlxf of

the line behind the wave to the wave resistance. The line base current,
which is just the current registered in lightning observations, arises instanta-
neously (with instantaneous short-circuiting of the line on the earth) and, at
the first moment, is determined exclusively by the wave resistance, indepen-
dent of the value of R I : i(0,O) = -Ui/Z. As the wave moves on towards the
cloud, the ohmic resistance the current has to overcome becomes increasingly
higher, so the base current decreases. At at >> 1, or at Rlxf/2Z >> 1, the
current through the base is
(4.32)
This current decreases much more slowly than at the wave front, because in
spite of the negligible front current, the line far behind the wave front is
discharged all the same, and all the charge that flows down from it goes
through the base.
The wave front propagates at a rate of electromagnetic perturbation. It
is independent of the line ohmic resistance but is determined exclusively by
its reactive parameters and is close to light velocity. This is a 'precursor'
which exists under any conditions, no matter whether the line has a
resistance or whether it changes behind the wave front. The precursor
carries information about the changes in the line, in our case about the
line grounding. If the resistance is zero or, more exactly, has no effect yet
because it is much less than the wave resistance (Rlxf << Z), the line is
discharged in a resistance-free way, and its initial potential and charge
practically vanish right behind the front of the primary electromagnetic
signal, the precursor. When the resistance becomes much higher (practically
several times higher) than the wave resistance, the charge and potential dis-
appear gradually, and the rate of their reduction decreases as the linear
resistance RI increases. At R1 = 10 n / m corresponding to the leader chan-
nel resistance, the time constant is a = 2ps-l and the precursor current
decreases, in accordance with (4.3 l), by an order of magnitude as compared
with the initial value of i(0, 0) over the period of time t 1 ps, for which the
N
precursor covers only 200m (U = 0.67C). Half way up to the cloud

(x = 1500m), the front current decreases by a factor of 3 x lo6. According
to (4.32), the line base current at that moment will be 3 x lo5 times higher
than at the front. Therefore, the current somewhere behind the precursor
will inevitably rise to a much larger value. Let us see where this happens
and what will be the velocity of the high current region carrying the
charge away from the line.
For the analysis of the relatively late stage in the discharge process with
at >> 1, we shall employ the second, asymptotic formula of (4.30) for the
Bessel function. For the region x << xf = ut located fairly far from the
weak precursor, the root in the argument of Io can be expanded. Using

formulae (4.29) for a and w, we get
This expression is the explicit solution to equations (4.22) without the

inductance term but with the same boundary and initial conditions. The
potential is
vi-
s/(4xr)’:2
U ( x ,t ) exp( -t2)d[ = U, erf
f ?i [0
Expressions (4.33) and (4.34) have a clear physical sense, demonstrating the
nature of a non-ideal line discharge.
When the perturbation front (the precursor due to the action of induc-
tance) goes far away, the current decreases slowly from the line base to the
front. It also varies slowly in time at every point, except for the region
close to the front. This is the reason why the inductance effects in the main
discharge region are very weak. With the neglect of the inductance term,
equations (4.24) transform to equations similar to those for heat conduction
or diffusion:
(4.35)
To use an analogy, the potential acts as temperature, current as heat flow,

and x as thermal conductivity (heat diffusion). We did not take x = const
out of the derivative deliberately to be able to come back to this equation,
also valid at RI# const.
The process of line discharge is similar to the cooling of a uniformly
heated medium, when a low (zero) temperature is maintained at its bound-
ary, beginning with the moment of time t = 0. Formulas (4.34) and (4.33)
describe the diffusion of the earth potential along the channel (figure
4.14(a)). The current-potential wave, smeared in contrast to the precursor,
Figure 4.14. The potential wave ( a )in linear diffusion with x = const and (b) in non-
linear diffusion with rising x.

Return stroke 179
propagates in such a way that its characteristic point x l , say, where the
potential is reduced by half relative to the initial value of Vi, x 1 / ( 4 ~ t ) ~=/ *
0.477, obeys the diffusion law x1 x ( x t ) l l 2 with a decreasing velocity
v1 x ( ~ / t ) M” ~x / 2 x 1 .From expressions (4.33), the current at point x1 is
20% lower than that at the channel base x = 0. Substituting the leader
resistance R1 x 10R/cm and C1 = 10pF/m into the formulae, we get
x = 1010 m2/ S . Over the time t = 10 ps (at at = 20), during which a weak
precursor will cover a distance of 2000 m, the half-potential point character-
izing the propagation of the line discharge wave will diffuse for 3 15 m only
and will be moving at velocity v1 x 0 . 0 5 ~ By . that time, the base current
i(0, t ) will have dropped by a factor of 11 relative to the initial current
i(0,O) (formula (4.32)).
The calculated values of x l , wl, and i(0,t ) can be brought closer to
measurements at a certain stage of the lightning discharge. Instead of the
leader resistivity, one should then deal with a lower resistivity averaged
over the perturbed region. This makes sense in some evaluations. But the
illusion of a satisfactory numerical agreement with measurements in a
short stage of the process is destroyed, as soon as we recall one of the
important qualitative observations. At the return stroke stage, a bright
and well-defined wave front - the channel tip, which becomes smeared
only slightly with time - is moving up to the cloud. This indicates that the
energy release and, hence, the current rise occur faster than in the solution
to (4.33). Clearly we deal with a wave possessing a steep front, at least for
powerful lightnings, rather than with diffuse current profiles. This contra-
diction can be resolved by rejecting the approximation R1 = const and by
including, in the theoretical treatment, the time evolution of the leader
channel and its transformation to a return stroke channel.
Note that the simple and attractive model of an immediate transforma-
tion of the leader channel at the wave front to an ideal conductor cannot
rectify the situation. This model would take us back to equalities (4.26)
and (4.27) describing the wave of immediate voltage removal and sustained
current, which propagates with the velocity of an electromagnetic signal
close to light velocity. But this possibility was already refused above. It
was mentioned in section 4.4.1 that the key to the phenomenon of return
stroke should be the analysis of the channel transformation dynamics.
The effect of the gradual resistance reduction during the Joule heat
release can be understood from equations (4.35) and (4.24) without the
inductance term. It would be justifiable to replace U by the potential varia-
tion AU = U - Ui, since Ui = const in our approximation, so we get
- - aX - d A U
dAU dAU 1
-- i = - C 1 X X ’ x = - (4.36)
at ax ax RlCl .
The resistance decreases while x increases, as the amount of charge flowing
through the particular channel site becomes larger, or with the increase in

AU. Consequently, the rear sites of the diffusion wave, where A U and
diffusion coefficient x are already higher, propagate faster than the front
sites, where A U and x are still low. To supply current to the region close
to the discharge wave front (a weak precursor is out of the question now),
the potential gradient there must be large because of the small diffusion
coefficient. Both circumstances indicate that the wave acquires a sharp
front, its profile becomes steeper and convex. In contrast to the gradual
asymptotic approximation at x = const, the curves U ( x ) and i ( x ) for a
given moment of time look as if they stick into the abscissa (figure 4.14;
the same will be seen from the numerical simulation illustrated in figure
4.17). The effect described here is well known [12]; this is a non-linear heat
wave driven, for example, by radiative heat conduction, whose coefficient
drops with decreasing temperature T approximately as x = T3.t
The variant with R I = const, for which the solution to (4.33) and (4.34)
is valid, probably corresponds to low current lightnings, when the energy
release is too small to provide an essential reduction in the former channel
resistance. In any case, there are streak pictures of return strokes with unclear
wave fronts or those becoming smeared after the propagation for a few
hundreds of metres [13,14]. To obtain conclusive evidence, stroke streak
pictures should be analysed at different currents. Regretfully, no simulta-
neous recordings of currents and stroke waves are available.
One can draw another conclusion from the solution to the set of
equations (4.24) at R I = const # 0, which is important for the analysis of
observations and for the formulation of boundary conditions necessary for
finding a numerical solution. According to (4.29), when the line closes on
the earth instantaneously the discharge current through the closed end also
reaches its maximum instantaneously. As mentioned above, the maximum
is independent of R I , being determined exclusively by the wave resistance.
Clearly, the same will also be true for any time-variable resistivity, and the
only question is how fast the current will decrease after the maximum.
However, the current in a real return stroke rises for several microseconds,
sometimes for several dozens of microseconds, and this time may become
even comparable with the total impulse time. Such a slow current rise may
t Equations (4.36) with A U ( x . 0 ) = 0, AU(0. t ) = -L'i and no inductance terms allow self-
similar solutions. The simplest of them are (4.33) and (4.34) for x = const. The process is self-
similar in a more complex approximation for x = b( 1 U,l)ntv,which corresponds qualitatively
-
to the R I x-' evolution during the channel transformation. Constants 6 , n, and U can be
chosen from the analysis of R I behaviour (section 4.4.3):n x 1-2; U 0.5-1. The wave front
follows the relations
x/ = E[b(lC:l)"]'!?t'"-'~:?
Vf = i ( V + l)E[b(lC:l)"]':2t-('-V'".
where E of about 1 is to be found by solving an ordinary differential equation [12].

Return stroke 181
be only due to the properties of the commutator, whose role is played by the
streamer zones of the descending leader and the counterleader. Their contact
actually gives rise to the return stroke. The streamer zone field rises, as the
streamer zones are reduced and the leader tips come close to each other or
as the descending tip approaches the earth with no counterleader formed.
The streamers are accelerated to a velocity lo7m/s, transporting kiloampere
currents even in laboratory conditions [ 15,161. Thus, the rate of current rise
and the impulse front duration at the earth, 9, are determined by processes
occurring in the vanishing streamer zone rather than in the former leader
channel. Measurements provide indirect evidence for this, showing that the
impulse rise time tf in positive lightnings possessing a longer streamer zone
than negative ones, at the same voltage, is several times longer.
4.4.3 Channel transformation in the return stroke

It has been shown above that electromagnetic perturbation propagates along
a line with a velocity equal to or somewhat lower than light velocity, indepen-
dent of the initial resistivity. When the resistance is high, as in the leader
channel, the current and the potential variation induced by the perturbation
attenuate rapidly. But the precursor is followed by a stronger perturbation
propagating at a lower velocity, which reduces the potential considerably,
to zero with time. The potential of a negative lightning drops to zero at
this channel site due to positive charge pumping; this compensates the initial
negative potential there. This process is accompanied by Joule heat release
with a linear power i2R1, which is high at first since the impulse front of
the ‘genuine’ (not the precursor) current is quite short and the initial
(leader) resistance R I is relatively high. The processes that follow - the
channel heating, its radial gas-dynamic expansion, the shock wave propaga-
tion, and the resistivity reduction - have much in common with those in
powerful spark discharges in short laboratory gaps. The latter have been
extensively studied experimentally, theoretically, and numerically [ 17-24].
Also, calculations have been made with the initial parameters characteristic
of a lightning return stroke, accounting for radiative heat exchange which is
especially important in this large-scale phenomenon [22-241. The stroke
channel gas is heated up to 35000K. Most of the Joule heat is radiated by
the highly heated gas in the ultraviolet spectrum. The emission from this
spectral region is absorbed by the adjacent colder air, adding the newly
heated gas to the conductive channel.
Such a treatment of the process would take us far from the point of
interest, so we shall restrict ourselves to a description of two numerical results
for atmospheric air, obtained with a rigorous allowance for radiative heat
exchange [23,24]. In both calculations, the shape and parameters of the
current impulse were preset, as is usually done in lightning calculations. Of
course, the current behaviour here depends on what happens in the whole

10 15 20 25 30 35
T, kK
Figure 4.15. The conductivity of thermodynamically equilibrium air at atmospheric
pressure.
of the perturbed region. But the formulation of a self-consistent problem

requires a combined solution of a set of equations for a long line discharge
and extremely cumbersome equations describing the physical evolution of
each section along the line. A problem of this complexity has not been
approached yet.?
Both calculations for one-dimensional cylindrical geometry were made
with the current impulse
i(t)= ZMt/tf at t < tf, i ( t ) = ZMexp[-(t - tf)/tp] at t > 9
possessing a linearly rising front and exponentially decreasing tail. The
calculation in [23] was made with moderate parameters tf = 5ps,
ZM = 20 kA, and tp = 50 ps corresponding to a moderate power lightning.
The other calculation [24] was for tf = 5 ps, ZM = 100 kA, and tp = 100 ps
of a very powerful lightning. It is generally believed that the air conductivity
0 corresponds to its thermodynamic equilibrium and is determined by
temperature (figure 4.15).
Figure 4.16 shows the evolution of pressure, gas density, temperature,
and radial velocity distributions behind the shock front for a powerful
current impulse [24]. The curves for moderate current impulses are qualita-
tively similar.
tThe problem of a short laboratory spark is much simpler. The set should include a simple
discharge equation for a capacitor bank as a high-voltage source for a spark gap with the desired
resistance and allowance for the circuit inductance. Note that this kind of LRC circuit usually
registers damped oscillations unobservable in lightning.

Return stroke 183
25- t=5p
20 -
a
$15
20
10 -
50
5-
" ' " "

OO 5 IO 15 20 0 5 10 15 20
r, cm r. cm
4
sa
2
0
0
r, cm r, cm
Figure 4.16. The radial distributions of pressure p , density p, temperature T , and

the velocity v behind the cylindrical shock wave of a return stroke: ,Z = 100 kA,
tf = 5 ps, t p = 100 ps; po = 1 atm, pa and co are the initial presure, air density and
sound velocity; To = 300 K.
The point of primary interest in a return stroke treatment is the beha-

viour of the integral channel parameter - its linear resistance:
RI = [ 1; 27rra(r) dr] -'. (4.37)
Table 4.1 presents, among other parameters, the linear resistance values
obtained from T(r) data borrowed from [23,24]. One can see that the
resistivity drops at first for 1 ps but then falls rather slowly. This decrease
ceases closer to the pulse tail, and the resistance begins to rise gradually.
The dramatic initial drop in R I is due to the primary heating of a very thin
initial channel by high density current.? As T increases to about 20000K,
the conductivity a rises but remains nearly constant with further temperature
t The gas is assumed to be in thermodynamic equilibrium at every moment of time. This assump-
tion is justified by a fast energy exchange (for lO-*-lO-'s) between electrons and ions, resulting
in a small difference between the gas and electron temperatures. The ionization is of thermal
nature: a Maxwellian distribution is established in the electron gas, and the amount of ionizing
electrons is defined directly by the electron temperature, rather than by the field. The electron
temperature, in turn, is determined by the Joule heat release and energy balance of the gas.
Equilibrium ionization is also established rapidly (for details, see [ll]).

0.074 8 5 2 >24 0.1 - 1600

3.7 28 18 120 5.5 0.3 8 3.1
11 24 15 170 4 0.7 26 0.38
39 17 12 110 2 1.2 38 0.20
91 14 10 60 1 1.4 46 0.27
Current impulse 100 kA with tf = 5 p,tp = 100 ps, Q = 1OC
5 35 25 180 16 0.8 - 0.28
20 22 20 140 5.7 2 - 0.057
50 18 15 90 2.6 3 - 0.039
100 14 12 70 1.8 4 - 0.028
200 12 11 40 1.0 5 -150 0.032
300 10 10 20 1.0 5 - 0.064
Note. T,,, is the temperature along the channel axis, Teffis the average temperature in the
conductive channel, oeffis an average channel conductivity, p is channel pressure, reffis the effec-
tive radius of the conductive channel, W is the total energy released (no data for the second
variant; the given values was estimated as W x i ~ a x R 1 t p )and
r Q is the charge transported
during the current impulse.
-
rise. In a strongly ionized plasma, with ions of constant charge o T3I2,but
doubly charged ions appear with increasing T . Since U ZC2,where Zi is the
-
ion charge multiplicity, the two effects compensate each other. The resistance
of a highly heated channel decreases with time due to its expansion only.
Some time later, however, the pressure at the channel centre drops to
atmospheric pressure, and the expansion ceases. The conductive channel
cross section is reduced gradually because of the gas cooling caused by
thermal radiation. The channel resistance begins to rise slowly because of
decreasing reff and Teff.
The expansion time of the channel becomes longer and its minimal
resistance decreases for the stronger current impulses. Physically, the linear
resistance is affected by the energy released per unit channel length, W ,
rather than by the current. This value is not described unambiguously by
- - -
the current amplitude; what is more important is the amount of the
transported charge Q: W1 i 2 R l t QiRl QE and the field does not
vary much. The calculations, however, deal with the current impulse but
not with W1. Semi-quantitatively, the time dependence of resistance can be
understood using the relations for the shock wave of a powerful cylindrical
explosion. The explosion can be considered to be strong as long as energy
is released in a thin channel and the pressure of the explosion wave does
not fall close to the atmospheric pressure. In this case, the flow is self-similar.

Return stroke 185
The shock front radius rs and pressure p in the affected region depend, within
the accuracy of numerical factors, on W 1 and t as r, ( W 1 / p o ) 1 / 4 t and
N 1/2
p N Wl/rz ( Wlpo)'/2t-',where po is cold air density. The channel
N
expansion is completed when the pressure drops to a certain value close to

atmospheric pressure. This sets the limit to the validity of formulae for self-
similar motion. This means that they are still applicable, and the duration of
the resistance reduction then is t ( W l p o ) 1 / It
N 2 . can be shown [12] that for
a self-similar cylindrical explosion in the central region with the pressure
equalized along the radius (figure 4.16), the internal specific energy depends
on r, t and W1 as E ( Wy/2t2-Tr-2)1/(T-
N 'I, where y is the adiabatic exponent.
A point with fixed temperature, e.g., T M 10000 K, can be regarded as the
conductive channel boundary, since the plasma conductivity below this
point is relatively low. The radius of a point with fixed T and E ( T ) varies
with time as r W:'4t1-y/2,reaching a value proportional to r,,,
N N W;I2
by the moment the channel stops expanding, t W:12.Therefore, the linear
N
channel resistance drops to a value proportional to Rl,, N rmax W-' , and

-2
N
this occurs for the time t W'/'. These relationships are qualitatively
N
consistent with the calculations for the two variants described in table 4.1.
4.4.4 Return stroke as a channel transformation wave

The first substantiated attempt to make a numerical simulation of the light-
ning return stroke with allowance for the resistance variation was undertaken
as far back as the 1970s [25,26]. The most important features of the process,
which are due to an abrupt conductivity rise at the site of intensive Joule heat
release, became evident at once. The simulation showed that a weak initial
perturbation (precursor) propagating up along the channel at an electro-
magnetic signal velocity close to light velocity does not change the plasma
state and cannot be treated as the return stroke wave front visible in streak
photographs. The main wave of current and decreasing potential travels
several times slower; its velocity is defined by the transformation of the
low conductivity leader to the low resistivity stroke channel. This conclusion
was formulated explicitly in [25-281; it reflects the nature of the lightning
return stroke.
Turning to numerical simulation today, we should like to formulate this
problem in a simple and clear physical language and to try to outline
problems to be solved within this model. An obviously essential aspect of
the theory still is the resistivity dynamics of the lightning channel. An
exhaustive formulation of this problem would involve a simultaneous
solution of equations describing the propagation of a current-voltage
wave and the channel dynamics at every point along its length, affected by
the ever varying energy release. So we shall restrict the discussion to a
simple model, having accepted a probable law for the linear conductivity
rise, G = R:', and focusing on the qualitative results of the solution.

Let us describe G in the simplest way reflecting the main qualitative

features of the channel evolution. It will be assumed that the linear conduc-
tivity increases with current. This partly reflects the fact that resistance
decreases with increasing charge through a particular channel site. But the
resistance is stabilized with time, even though the current continues to
flow. In principle, the stable state of a lightning channel hardly differs
from that of an arc. The field E in a h g h current arc only slightly varies
with current; in other words, the linear conductivity of the channel is
G,, = i / E i. Assume E to be equal to the field EL in the lightning leader,
N
whose current is not low on the arc scale; then the conductivity is
G,, = i / E L . In a mature gas-dynamic process when the shock wave is still
strong, the resistance will drop with time. As the shock wave becomes
weaker, the decrease in R1 and the increase in G become slower. These
tendencies are described by the relaxation-type formula
dG
- - i/EL - G(t) -
- Gst(i) - G(t) (4.38)
dt Tg Tg
where Tgis the characteristic time of linear conductivity variation (relaxation
time). In a simple case with i = const, Tg = const, and G(0) = 0, we have
G = GSt[1- exp(-t/T,)].
Equations (4.24) are solved with the initial conditions U ( x ,0) = U, and
i ( x ,0) = 0, RI(x30) = RI,-;the reactive parameters are taken to be constant:
C1 = 10 pF/m and L1 = 2 pH,”. The channel does not close on the earth
in an instant but does so through the time-decreasing resistance of the
commutator (similarly to the real lightning length decreasing through the
streamer zone). The accepted values of R,,, = R(0)exp(-cut), R(0) = 10 R
and cu = 1 ps-’ provide a typical duration of the negative current impulse
front tf FZ 5 p . The boundary condition at the grounded end of the line
raises no doubt: U ( 0 .t ) = i(0,t)R,,,. The problem of the far end up in the
clouds, x = H , is much more complex. Conventionally, it is considered as
being open, assuming i ( H ,t) = 0. In reality, the situation is far from being
self-evident. When the line gets discharged and its end in the clouds takes
zero potential, a high electric field must arise near it due to the voltage
difference A U = - U o ( H ) . This gives impetus to very intensive ionization
processes, probably involving high current. This situation will be partly
discussed below. Now, we shall assume the upper end to have no current.
The results to be presented were obtained for a vertical unbranched channel
with the total length H = 4 km. This is the height the ascending leader tip
reaches when the descending leader, which has started from the point closest
to the earth in the bottom negative sphere with the centre 3 km high, contacts
the earth. It is normal practice to use the following averaged parameters
of the leader prior to the contact: EL = 10V/cm, iL = lOOA, and
RIL = EL/iL= 10 O/m. For a realistic description of the resistivity dynamics
(section 4.4.3), the relaxation time should be taken to be Tg = 40 ps, when the

Return stroke 187
current at this channel site rises, and Tg = 2 0 0 p , when the current

decreases. The model calculation reproduces the distributions of current
i ( x , t ) and potential U ( x ,t ) along the channel; the linear charge is ~ ( xt ), =
C1[ U ( x ,t ) - U o ( x ) ]Generally,
. the external field potential Uo(x) can also
vary in time, because we should not discard a possible partial
neutralization of the charge in one of the regions of the cloud dipole. The
latter point will not be discussed for the time being.
The calculations are presented in figures 4.17-4.2 1. The precursor tra-
velling with velocity 0 . 6 4 ~is damped so fast that this is not shown in the
plots after a noticeable break from the principal wave re-charging the chan-
nel (we shall term it a discharge wave for simplicity). The wave in figure 4.17
travels along the channel with velocity U, x 0.4c, i.e., 1.6 times slower than
the precursor. This velocity somewhat decreases as the wave moves up. Its
variation can be conveniently followed from the change in the well-defined
maximum linear power of the Joule losses i2Rl (figure 4.17 (centre)). The
wave front power rises abruptly along a 100-200 m length, then it decreases
towards the earth, making the channel tip with intensive energy release stand
out clearly. It seems that it is this region which is clearly discernible in streak
photographs. The linear power proportional to the squared current drops
remarkably on the way up the cloud, and the maximum becomes smeared.
This is also consistent with observations of radiation intensity [14,29]. A
photometric study has shown that the radiation from the wave front is
attenuated and the front loses its clear boundary. The current wave is not
attenuated so rapidly (figure 4.17 (top)). For the time of its earth-cloud
travel lasting for 3 4 p , the current at the channel base drops from the
maximum of 35 kA to 24 kA. This agrees with observations indicating that
an average current impulse duration in a negative lightning is close to
75ps on the 0.5 level. The wave front deformation depends on the initial
potential Ui delivered by the leader to the earth. The higher the value I Uil,
the higher the discharge current. The rate of resistivity decrease at the
wave front grows respectively, so the front steepness increases. This is evident
from a comparison of figures 4.18 (top) and 4.18 (bottom). At 1 Uii = 50 MV,
the current wave goes along the channel practically without elongating the
front? while at lUil = 10MV it has a lower velocity and a smooth front.
Unfortunately, there have been no registrations of current and streak
photographs of the return stroke taken simultaneously. A comparison of
the relationships between current and wave velocity could provide a good
test for the return stroke theory.
As the current impulse amplitude rises, the linear resistance falls more
quickly and to a lower level, so the wave is damped more slowly during its
t The motion of a high current wave with attenuation but without noticeable distortions
facilitates the electromagnetic field calculation necessary in many applied problems of lightning
protection and in substantiation of remote current registration methods.

0 1 3 4
X,2km
t=8p
0 1 2 3 4
x,km

Return stroke 189
I,,,= = 67 kA
U
3
\
.3 0.4 -
0.2 -
0.0 I . , . , . , . ,
0 1 2 3 4
x, lan
1- = 8,15 kA
0.8
1
0.6
.I
0.4
0.2
0.0
0 1 2 3 4
Figure 4.18. Deformation of the current wave front at leader potential (top)
Ui = -50MV and (bottom) -10 MV; for the other parameters, see figure 4.17.
propagation along the channel. There is no damping at a very high current and
the impulse front becomes steeper, as was discussed in section 4.4.2 (figure
4.18). Non-linearity is also observed in the current amplitude dependence
on the initial potential U, at the earth. If the commutator were perfect
(R,,, = 0), the current at the earth at the moment of contact would instantly
Figure 4.17. (Opposite) Numerical simulation of the return stroke excited by a

descending leader with potential -30 MV: (top) current and (centre) voltage dis-
tributions; (bottom) the power of Joule losses. The initial leader resistance, 10 n/m.
Steady state field in the channel behind the wave, 10 V/cm.

200 1 r 0.6
Voltage, MV
Figure 4.19. Calculated dependencies of the current amplitude and average wave
velocity in the return stroke on the leader potential Vi.
take the maximum value,Z = I Ui l/Z,independent of the actual channel resis-

tance, and would be ZM NUi. With the finite time of R,,, decrease to zero, the
current wave is able to cover some distance along the channel and to include in
the circuit the ohmic resistance of this channel portion. For this reason, the
current amplitude appears to be lower than Ui/Z and rises somewhat faster
than potential Ui (figure 4.19). It is important that the lightning current
amplitude ZM is found to be appreciably smaller than its theoretical limit
Ui/Z: e.g., ,Z M 0: 6Ui/Z at Vi = 30 MV. This is another source of errors
in evaluations using the equality Vi = ZZ ,, in particular, in the calculation
of cloud potential from lightning current data.
4.4.5 Arising problems and approaches to their solution

The current at the earth is independent of the boundary condition at the
upper channel end, until the reflected wave comes back to the earth with
the information about the processes occurring there. Before that moment,
the positive charge is pumped into the line from the earth. In virtue of the
boundary condition - zero current at the upper end - the current wave is
reflected there with the sign reversal. As a result, the current behind the
reflected wave, i.e., between its front and the channel end (figure 4.20),
decreases (it would drop to zero in the absence of damping). The incoming
positive charge now re-charges the line making it positive (an ideal line
would be re-charged to -Ui). The reflected wave moves faster and is
damped more slowly, because the linear resistance in most of the channel
has dropped by an order of magnitude or more due to the action of the
forward current wave.

Return stroke 191
0 1 2 3 4
x, km
-8a& 15--
10-
5-
0-
-5 -
Figure 4.20. Current and potential distributions during the propagation of waves
reflected by the cloud end of the channel.
When the reflected wave reaches the earth, delivering a positive potential
to it, a new discharge cycle begins. The newly acquired positive charge flow-
ing into the earth is equivalent to the extracted negative charge. The current
sign at the grounded end is reversed (figure 4.21). In the absence of
dissipation in a distributed system such as a long line, undamped oscillations
with a period T = 4H/w, would arise similar to those in an LC circuit.
Nothing of the kind is observed in lightning registrations, nor is there a
single change in the current direction. This means that the discharge wave
is either not reflected by the upper end of the line or the reflected wave
becomes so damped on the way back to the earth that it is unable to manifest

.-
-0.4 j U
Figure 4.21. Calculated current impulse through the grounded channel end.
itself against the background of other variations in the current. By changing

the parameter Tg or the quasi-stationary channel field EL, one can reduce
or even cancel part of the current impulse of opposite sign, but it is
impossible to attain a portrait likelihood between the calculated and
observable currents. The suppression of the reflected wave by raising the
instantaneous values of the channel resistance R I (x, t ) inevitably results in
an excessive reduction of the impulse duration at the grounded end of
the line. There seems to be no way of avoiding this even by changing the
resistivity reduction law.
The first thing that seems to be suitable for rectifying this situation is to
question the boundary condition at the upper end. This idea appears reason-
able because it generally agrees with lightning current registrations at the
earth for the double path time t FZ 2H/w, while the model solution for
i(0,t ) remains independent of the boundary condition. It is obvious that
the open circuit condition is an excessively rough idealization. It was
mentioned in section 4.3.3 that if the negative cloud bottom is filled with a
large number of branches which stem from the ascending leader, this
region becomes similar to a metallic sphere. Assuming such a 'metallization'
of the cloud, it would be more reasonable to consider the upper end to be
connected to a lumped capacitance C, = 4 7 r ~ ~ Rdefined ~, by the cloud
charge radius R,, instead of being open. The boundary condition at x = H
would have the form i ( H ,t ) = C, dU/dt. When the current wave reaches
the line end, the delivered current also discharges the negative 'metallized'
cloud region. This, however, does not prevent the appearance of the reflected
wave. At the first moment of time, the capacitance still preserves its charge and
is similar, in accordance with the reflection condition, to a short-circuited

Return stroke 193
end of the line, which generates a reflected current wave of the same sign and
amplitude as the incident one. As the capacitance becomes discharged, the
reflected wave amplitude decreases and then the sign is reversed. The com-
pletely discharged capacitance, incapable of supporting current, eventually
becomes equivalent to an open line end. It is clear even without a numerical
calculation how much the current changes at the earth after the arrival of a
reflected wave of such complexity. It should be emphasized again that
nothing of the kind has ever been observed in real lightning.
One can also try to rectify the situation by complicating the boundary
condition with the allowance for the final resistivity of the ‘metallized’
cloud region. The streamer and leader branches filling the cloud possess a
resistance at the moment of their generation. A leader branch can hardly
be heated as much as a single descending leader. The resistance of the
‘metallized’ cloud, R,,,is quite likely to be high during the whole return
stroke stage. If this is so, the boundary condition should be formally repre-
sented as i ( H ,t) = C, dU/dt - R,,dildt. Strictly, it is not only the boundary
condition that changes in this case, like in the case of ideal metallization, but
also the set of equations. The cloud potential U. can no more be considered
as being constant in time. The second equation of (4.24) should be re-written
as
having taken into account the change in U, due to the change in the cloud
charge Q,. Then the function Uo(Q,) must allow for the delay because of
the finite rate of the electromagnetic field propagation. The problem becomes
extremely complicated. Although radar registrations do indicate the develop-
ment of a wide network of branches in clouds, there has been no investigation
of cloud ‘metallization’. The reason for this, no doubt, is the lack of initial
data.
One should not discard two other factors unaccounted for by the numer-
ical model. First, a leader channel can practically never be single. Owing to
the numerous branches of different lengths developing at different heights,
numerous reflected waves will arrive at the earth at different moments of
time, creating a sort of ‘white noise’ with a nearly zero total signal. This
will deprive the current of its characteristic bending which is usually created
by a single reflected wave at the earth. Second, constant linear capacitance
only approximately describes the real re-charging of a lightning leader. We
have mentioned above that the cover charge around a leader channel is
changed by numerous streamers starting from it. Their velocity decreases
rapidly when the streamer tips go away from the channel surface with its
high radial field. So, when the voltage at the wave front changes, the
charge near the channel changes almost immediately, while its change at
the external cover boundary occurs with a delay. In other words, a lightning

discharge can proceed for a fairly long time. The quasi-stationary current
from the discharge of the cover periphery, having the same direction as the
current in the forward wave, can compensate for the reverse current induced
by the wave reflected by the earth.
To conclude, the model of a single long line with varying linear resis-
tance allows elucidation of many aspects of the return stroke but cannot
claim to give reliable quantitative description of all characteristics of this
phenomenon. The much more simplified models of return stroke are usually
used calculating electromagnetic field for technical application. A review of
this model is given [30].
4.4.6 The return stroke of a positive lightning

Two kinds of current impulse can be distinguished in oscillograms taken at
the earth after the arrival of a positive leader. Common impulses are similar
to those registered in negative lightnings, although they have a slightly longer
duration t p and less steep fronts. Such impulses can be naturally interpreted
as return stroke currents corresponding to the wave discharge of the leader
channel, as described above. Sometimes, however, quite different impulses
are registered with an order longer duration and an amplitude as large as
200 kA. A closer examination shows that impulses with an ‘anomalous’
duration cannot be interpreted as resulting from a grounded leader dis-
charge. They appear to result from another process, and we shall offer
some suggestions concerning their nature in section 4.5. Here, only
common impulses will be discussed.
It was shown in section 4.4.2 that the stroke current front is unrelated to
the wave discharge process in the channel but, rather, is associated with an
imperfect commutator closing the channel on the earth. The leader streamer
zone acting as a commutator possesses a finite resistance and is reduced
during a finite period of time. The front steepness is determined by the rate
of resistance reduction in this transient link between the channel and the
earth. But the streamer zone length of a positive leader at the same voltage
is about twice as large as that of a negative leader and takes more time to
be reduced. It is quite likely that this is the main reason why, with the
50% probability, the current front duration in positive lightnings,
tr = 22 ps, is four times longer than in negative leaders [l]. Approximately
the same proportion is characteristic of the maximum pulse steepness.
The duration of the pulse itself, t p , is primarily determined by the stroke
channel length. It was shown in section 4.3 that it is only positive leaders
starting from the top positive region of a storm cloud which have a real
chance to reach the earth. This region is twice as high as the negative
cloud bottom. Hence, the channel length of a positive descending leader is,
at least, twice as long. But the vertical positive channel transverses the
negative cloud region, delivering a very low potential U, to the earth, so it

Anomalously large current impulses of positive lightnings 195
is incapable of producing a return stroke with an appreciably high current

(section 4.3.6). Only those lightnings, whose positive descending leaders
bypass the negative cloud region along a very curved path, can actually be
identified in the registrations. The statistics shows that the total length of
such a leader, including the path bendings, is 1.3-1.7 times greater than
that of a straight leader. Therefore, a positive channel length and its stroke
pulse duration appear on average to be three times greater than in a negative
leader. As for other characteristics, common positive pulses are the same as
the negative ones described above.
4.5 Anomalously large current impulses of positive

lightnings
Anomalous impulses of a positive lightning have the duration t, M 1000 ps of

the 0.5 amplitude level and the rise time q x 100 ps. The current in some of
them is as high as 100 kA or more [l]. Although such lightnings are rare, their
effects on industrial objects are so hazardous that they should not be
underestimated. A current impulse delivers to the earth a charge
Q M 1OOC; therefore, as large a charge must be located in the cloud cell
where the lightning originated. The potential at the boundary of a charged
cloud region of radius, say, R, M 1 km is U,, M 1000 MV, with 1500 MV
at its centre. Any attempt to treat a long current impulse as return stroke
current inevitably leads to contradictions. Indeed, in order to reduce the
near-earth current by half of its maximum value for lOOOps, it would be
necessary to assume in (4.29) at - 0.7 and a = R1/2L1 = 700s-’; hence,
p. -
the average linear resistance behind the wave front of the return stroke
would be Rl M 3 . 5 ~ a / m . The total resistance of a channel of length
H = 4000m would be R1H M 14R, i.e., 40 times less than the wave
resistance. The line would seem to be discharged as an ideal line, i.e., for
20 ps instead of 1000 ps, with the velocity of an electromagnetic signal.
Excessively smooth impulses are sometimes observed in ascending
leaders. A positive impulse IM M 28 kA with t, = 800 ps was registered
during the propagation of a negative ascending leader from a 70-m tower
on the San Salvatore Mount in Switzerland [31]. This fact in itself is of
interest, but its analysis may offer an explanation of ‘anomalous’ currents
of descending positive lightnings. Note, at first, the unusual situation at
the start. Since the negative charge of the dipole is located at the cloud
bottom, the ascending leader is to be positive rather than negative. Therefore,
the dipole axis has either deviated from the normal or the bottom negative
charge was neutralized earlier by, say, an intercloud discharge. This situation
occurs rarely but it is possible.
We mentioned in section 4.1 that ascending lightnings have no return
strokes because their channels are grounded from the very beginning.

However, when the ascending leader penetrates the charged cloud region
(positive, in this case), a large potential difference arises between the front
end of its grounded channel and the space around it, so the leader current
has been found from many registrations to rise to several kiloamperes.
This event seems to be triggered by the same mechanism, but its effect is
greatly enhanced by the leader hitting the very centre of a large cloud
charge of, say, Q, x 30C and radius R, M 500m, where the potential is as
high as U, M 500-800MV. At such voltages, the streamer zone and cover
appear much extended. Negative streamers develop until the average field
in their streamers drops below E, M 10 kV/cm under normal conditions (or
1.5 times less at a 5-6km height [16]). Streamers elongate very quickly
when the field is higher. Therefore, a very powerful streamer corona con-
sisting of numerous branched streamers (they are likely to originate not
only from the stem but from its numerous branches, too) will fill up a
space of size R M Uo/EsM R,. The negative charge of the streamer zone
will partly neutralize the positive charge of the cloud cell. If the streamers
have velocity U, x 106m/s, they will fill the charged cloud region for
t x Rc/vU,M s. Since the capacitance of the leader portion inside the
cloud, CL,is comparable with that of the charged cloud region, Ccl,
a charge of opposite sign, comparable with the cloud intrinsic charge,

penetrates the cloud. The resultant effect is such that most of the cloud
charge would seem to run down to the earth with current i x Q o / t M 30 kA
for t M lop3s. Microscopically, the cloud medium remains non-conductive,
as before. Charges do not recombine but neutralize one another on average.
The process of current organization reduces to the neutralization of the
cloud rather than leader charge.
Returning to long current impulses after the positive leader arrival at the
earth, let us imagine that the leader has been developing along a vertical line
somewhat away from the axis of a powerful cloud dipole with Q, x lOOC or
more, R, x 1 km, and U,, M 1000 MV. Suppose the leader cover has no
contact with the cloud charge boundary but is close to it. All the same, a
huge, actually induced charge comparable with Q, arises in the vicinity of
the cloud charges. Note that the arrival of a vertical positive leader does
not reveal itself in any way, since its potential is close to zero because of a
nearly complete symmetry of charges induced in the lightning channel.
Suppose now that while this leader still preserves conductivity (this
period of time is measured in dozens of milliseconds because of the current
supply of -100 A), an intercloud discharge occurs, connecting the lower
negative charge of the dipole to another positive charge. Intercloud dis-
charges have been observed to be a much more frequent phenomenon than
cloud-earth discharges. So our suggestion is not at all improbable. The

Stepwise behaviour of a negative leader 197
charges of opposite signs connected by intercloud leaders will gradually

neutralize each other via the same mechanism as the one underlying an
ascending 1eader.t As the neutralization goes on, the earlier induced but
now liberated positive charge of the grounded leader will flow down to the
ground. This will occur at a lower velocity than the return stroke velocity,
in accordance with the neutralization rate of the negative cloud charge.
This is a likely explanation for long powerful current impulses in positive
lightnings.
4.6 Stepwise behaviour of a negative leader
When discussing the negative leader in section 4.3.2, we put off the considera-
tion of its stepwise behaviour until the reader has become familiar with the
return stroke, since a similar phenomenon is the principal event occurring
in each step. It would be reasonable, at this point, to turn to the nature
and effects of the stepwise leader behaviour. But we should like to warn
the reader that there is no clear answer to the question why a negative
leader has a stepwise structure while a positive one has not. Nonetheless,
some observations of stepwise positive lightning leaders were presented in
[32]. This phenomenon has never been observed in laboratory conditions.
4.6.1 The step formation and parameters

The only thing one can rely on today in discussing the nature of leader steps is
the results of laboratory experiments with long negative sparks (section 2.7).
Natural lightning observations are not informative, except for the step
lengths Ax, x 5-100m [13,32-371 and the registrations of leader channel
flashes occurring at the step frequency. Streak photographs indicate that
only the front channel end of 1-2 steps in length shows bright flashes. But
weak flashes may appear even along a kilometre length (the vision field of
a photocamera does not always cover the whole channel).
Laboratory streak pictures show that a step originates from two second-
ary twin leaders at the front end of the streamer zone in the main negative
leader (in the Russian literature, these are termed bulk leaders). The positive
leader moves towards the main leader tip while the negative one develops
along the latter (figure 4.22). During the pause between two steps, the
secondary negative and the main leaders do not have a high velocity, but
the positive leader moves faster for two reasons. First, as the distance to
the main leader tip becomes shorter, the difference between the positive tip
t The fact that intercloud discharges do neutralize charged regions is supported by electric field
measurements, and high currents that flow through lightning channels are indicated by peals of
thunder.

Figure 4.22. The potential distribution for various stages of a step formation. The main
leader potential U ( x )is counted from the external potential U,:(top) secondary leaders
1 and 2 are formed at point 3 at the streamer zone end; (centre) the tip of a positive sec-
ondary leader has reached tip 4 of the main leader, and a discharge wave has started its
travel along the secondary leader channel (dashed line); (bottom) the main leader tip
after the step formation has taken a new position, and the process is repeated.
potential U 1 and the external (for the tip) potential U ( x l )at the tip site x1
increases (figure 4.22 (top)). Second, the streamer zone field of the main
leader, E, % 10 kV/cm, which must support the generated negative streamers,
is higher than the field E, x 5 kV/cm required for the development of positive
leaders. For this reason, the streamers generated by the secondary positive
leader tip develop in a fairly strong field, become accelerated and all reach
the main leader tip. Since the long channel of the main leader has a
capacitance greatly exceeding that of the short secondary leader, it absorbs
completely all charges carried by the positive streamers. In other words,
the secondary positive leader develops in the final jump mode. We know
from section 2.4.3 that this leads to its acceleration. The secondary negative
leader, on the contrary, develops in a decreasing field beyond the streamer
zone of the main leader, whose streamers stop in space, so it moves much
more slowly, similarly to the main leader.
When the tip of the secondary positive leader comes in contact with the
main channel, the positive leader experiences the transition to the return
stroke. Charge variation waves run along both channels, as described in
section 4.4, and their potentials tend to become equalized (figure 4.22
(centre)). But the capacitance of the kilometre length channel is much
higher than that of the shorter secondary channel, so their fusion results in

Stepwise behaviour of a negative leader 199
establishing a potential only slightly differing from the initial potential of the
main leader tip, U,.
The moment at which the potential U1 is taken off the secondary leader
channel and the latter joins the main leader, manifests the end of the step.
The main leader tip ‘jumps’ over to a new place, the one occupied previously
by the tip of the secondary negative leader, delivering to it its potential U ,
(figure 4.22 (bottom)). The tremendous potential difference that arises in the
vicinity of the newly formed tip at this moment produces a flash of a powerful
negative streamer corona, which transforms to the novel streamer zone of the
main leader. Then the sequence of events is repeated. The combination of the
charge utilized for the short recharging of the secondary positive leader and
for charging the secondary negative one, plus the charge incorporated into
the new streamer zone, create the step current impulse. (Recall that there is
always a local current peak at the streamer tip or in the leader streamer
zone, related to the displacement of the charge in this region; see sections
2.2.3 and 2.3.2.) Part of the step impulse creates a current impulse in the
main channel and the other part is spent for the formation of a new cover
portion. The charge Q, pumped into the main channel can be evaluated in
terms of the mean velocity of the stepwise leader, wL M 3 x 10m/s, the
length of a step Ax, M 30m, and the current iL x lOOA averaged over the
whole duration of the process. Since the time between two steps is
At, x Ax,/wL M lop4s, the charge is Q, M iLAt, x lop2C.
4.6.2 Energy effects in the leader channel

The energy pumped by the charge pulse Q, into the channel can be evaluated
if the effect is assumed to be similar to that observed when the small capaci-
tance (of the secondary leader) is added parallel to the large capacitance (of
the main leader). While a common potential is being established, there is a
dissipation of energy nearly equal to the electrical energy stored by the
small capacitance at a voltage equal to the difference between the initial
capacitance voltages U, - U1, where U1 is the potential of secondary leaders.
It was pointed out in section 2.4.1 that the leader tip potential is shared
nearly equally between the streamer zone and the space in front of it.
Secondary leaders are produced at the streamer zone edge, so we have
4
U1 - U, M ( U , - U,); hence, U , - U1 = f ( U , - U,,). With the accepted
average values of current i x C1( U , - Uo)vL and of velocity wL and assuming
C1 x 10pF/m, we find U, - U , = 30MV and U, - U1 = 15MV. Thus, the
step energy is
W M Q,( U, - U1)/2 x 7.5 x lo4J. (4.39)
Of course, not all of this energy is released in the main channel. During the dis-
sipation of charge Q,, the channel potential rises appreciably. This leads to the
radial field enhancement and to the excitation of a streamer corona which

pumps some of the charge into the leader cover. This process, the cover
ionization in particular, requires much energy. But even if the energy released
in the channel is assumed to be W M lo4 J, t h s is still a very large energy.
The power required to support an average current of lOOA in remote
channel portions, where the effects of current impulses are averaged and
smeared, should be P1 M lo5Wjm. It is removed from the channel by heat
conduction and, partly, by radiation. These parameters correspond to the
maximum channel temperature T M 10 000 K, field E x 10 Vjcm, and resis-
tance R1 M 10R/m taken for the above estimations (section 2.5.2). At this
power, the energy released between two flashes per unit channel length will
be Wl,, M 10J/m for the time At, M lOP4s. Therefore, the single pulse
energy W would be sufficient to support a channel 1 km long. In reality, a
step pulse is damped at a much shorter length. At a distance of about
1 km, the step effects are smeared and the energy released in the channel
becomes totally averaged. But at a short distance from the tip, the energy
effect of the step is very strong, as is indicated by the intensive flash. The
temperature registered in some measurements was as high as 30 000 K [35],
i.e., the same as at the wave front in a return stroke.
Let us evaluate the distance at which the energy effect of an individual
step is still essential. When a short step joins a long channel, charge Q, is
pumped into the channel for a short time. Since we are interested in distance
and time much larger than the real length and duration of a charge source, let
us assume the source to be instantaneous and point-like, as is usually done in
physics: a point charge Q, is introduced at the initial point of the line, x = 0,
at the initial moment of time t = 0. The resistance of not very short channel
fragments, Rlx, is higher than the wave resistance, so the inductance effect
will be neglected. At an average resistance of 10R/m, this distance is just
about the step length Ax,. At shorter distances, the instantaneous point
source model is invalid, since it implies an infinite initial voltage and
energy. They drop to realistic values only if the charge affects a length
exceeding Ax,, at which the source was actually placed. Therefore, with
the neglect of inductance (and the precursor), the line charging to potential
U,(x, t ) above the background potential is described by equations (4.36).
On the assumption of R1 = const ,t they have an exact solution correspond-
ing to heat flow from an instantaneous lumped source:
I.The value of resistance R Ito be taken for evaluations may be smaller than that in the leader,
having in mind a transformation of the channel due to the step current.

Stepwise behaviour of a negative leader 20 1
The power released by the current step per unit channel length is
described as
(4.42)
At the point x, the power reaches a maximum at moment t, = x2/6x, and
(4.43)
For the time of the pulse action, the energy released per unit length at point x
is
W, M 1; Q2
iiR, dt = 2x - -
W Ax,
d 1 x 2 Ax, (
x
, x > Ax, (4.44)
where W is the total energy injected into the channel by the pulse, with its
upper limit given by formula (4.39). The effective duration of energy release
from a single step at point x is expressed as
X2
At, Wl
- 2.2t - -. (4:45)
- 2.7%
N
PImax
Consequently, the contribution of charge injection to the energy release at
a given channel site decreases in the direction of perturbation propagation as
Wl x x - ~and is independent of R 1 . The latter fact justifies the use of
R1 = const without reservations concerning the resistance variation during
the current impulse passage. The energy pulses released at point x owing to
the two subsequent steps superimpose at x > (2.7xAts)ll2; t h s critical
distance follows from the condition At, M At,. For example, at the average
resistance R I = 10 R/m, with x = 1010 m2 /s and step frequency At, = lop4s,
this happens at a distance x x 1.6 km in t , M AtJ2.2 M 45 ps, after the pulse
arrival here. Thus, the effects of energy release from individual steps are
detectable even along an extended lightning path, and this is the cause of
observable flashes of almost the whole channel. For a flash to arise, there is
no need for a strong energy effect. A short temperature rise of, say, above
l000K over l0000K would be sufficient for a flash to be detected by
modern photographic equipment.
The channel energy is affected by the temperature rise above the average
background, rather than by the time separation of the energy pulses between
two subsequent steps. In this respect, the impulse effect on the channel during
the wave propagation is damped at distances close to the tip. The plasma
temperature modulation determining the flash intensity at large distances is
due to the imbalance between the energy release and heat removal from
the channel during the pauses between the steps. There is no imbalance
at a large distance from the tip after the channel development has been

established. At T x 10 000 K, the losses for air plasma radiation are not par-
ticularly great but become appreciable at T M 12- 14 000 K. The Joule heat of
current is eliminated from the channel primarily by heat conduction. This
process occurs at constant (atmospheric) pressure when the energy release
is moderate, as is the case for distances of hundreds of metres from the tip.
At T M l0000K, the air heat conductivity is X x 1 . 5 ~ WjcmK and
the thermal conductivity at pressure p = 1 atm is XT = X/pcp x 180cm2/s,
where p is air density and cp is heat capacity. The average conductivity in
the channel corresponds to a temperature lower than the maximum tem-
perature. To illustrate, at c x 10 (Cl cm)-’ corresponding to T = 8000 K,
the effective radius of a conductive channel is r M (mrR1)-”2M 0.6cm
for R1 = 10R/m. The heat is removed from the channel for time
t r 2 /2xT
N N s, an order of magnitude longer than the pause between
the steps. The repeated energy pulses dissipate rather slowly, and the
temperature modulation relative to its average value T M 10000K is not
large at long distances x.Indeed, the energy released in the remote channel
portions during a pause is Wla, x PlavAt,x 10 Jim at an average power
PlavM lo5 Wjm. Even if we assume that all energy of a step is released in
the channel and W/Ax, x 2500 Jim in (4.45), the excess of the pulse release
over the average heat removal, which is equal to the average energy release
Wla,, will be small at x > Ax,( Wl/Wlav)1’2x lOAx, zz 500m. With allow-
ance for other energy expenditures, this reduction in the pulse effect will be
evident even at shorter distances. This circumstance makes the use of average
parameters reasonable in the consideration of the evolution of long stepwise
lightning leaders, ignoring the stepwise behaviour effects. In any case, labora-
tory experiments show that there is no appreciable difference between a
positive continuous and a negative stepwise spark discharge as for the
velocity, average leader current or breakdown voltage in superlong gaps.
However, even a small excess of the average temperature over its aver-
age value may be sufficient for a flash to be registered optically. As for chan-
nel portions located at a distance of one or two steps from the tip, the energy
pulses and outbursts of temperature and brightness are found to be very
strong there. A gas-dynamic expansion of the channel is also possible, as
happens in the return stroke (section 4.4), although it occurs on a smaller
scale. No doubt, a flash is also produced by a powerful impulse corona
giving rise to a new streamer zone of the elongated leader. Photographs
show that the transverse dimension of a step flash is about 10m [38].
4.7 The subsequent components. The M-component
The processes in the lightning channel following the first component are
known as subsequent components. Of interest among these are so-called
M-components and dart leaders. In the first case, the current impulse

The subsequent components. The M-component 203
registered at the earth has a very smooth front (0.1-1 ms), a similar duration
and an amplitude of several hundreds of amperes, sometimes of 1-2 kA. The
channel radiation intensity increases abruptly during the impulse, but one
can hardly identify in the photographs a structure similar to the impulse
front. The current impulse of an M-component is always registered against
the background of about 100A continuous current of the interpulse
pause. For a dart leader to arise, this current must necessarily be cut off
[39,40]. A few microseconds after the cut-off, a short high-intensity region
- a dart leader tip - runs down to the earth along the previous channel
with a velocity of -107m/s. The contact of the dart leader with the earth
produces a return stroke with its typical characteristics but having a much
shorter impulse front than in the first component (less than 1 ps or even
0.1 ps in some impulses). It is hard to say anything definite about the lower
limit of the front duration: it is likely to lie beyond the time resolution of
the measuring equipment.
The papers published almost simultaneously [41, 421 interpret the sub-
sequent component qualitatively as representing the discharge, into the
earth, of an intercloud leader after its contact with the upper end of the preced-
ing grounded but still conductive channel. Here we describe the evolution of an
M-component in terms of a numerical simulation.
The model underlying the simulation is as follows (figure 4.23). Initially,
there is a grounded plasma channel of length HI with zero potential, which
was left behind by the preceding lightning component. At time t = 0, a leader
channel of length H 2 and potential Ui joins it in the clouds (the voltage drop
from the leader current and from the intercomponent current is neglected).
The short process of the channel commutation through the streamer zone
Figure 4.23. The formation of a subsequent component: (a) the grounded channel of
the previous component and an intercloud leader; (b) channel charging-discharging
waves.

of the intercloud leader is ignored. At the moment of closure, the leader

channel possesses a typical resistance RIL % 10O/m. The resistance RI, of
the previous channel depends on the duration of the intercomponent
pause. After the return stroke current impulse of the previous component
25001
2000 f
d
g 1 500
C
-
E
61000-
500-
0-
Figure 4.24. Simulation of the M-component on closing an intercloud leader 2 km in

length and lOMV potential on a 4-km grounded channel. The initial linear resistances
of the channels R I = 10 n/m and the steady field EL = 10 Vjcm. The waves of potential
(this page, top), current (this page, bottom), the power of Joule losses (opposite, top)
and the current impulse at the grounded end of the channel (opposite, bottom); for
comparison, the latter is also given for R I= 20 G/m and EL = 20 V/cm (curve B).

T h e subsequent components. The M-component 205
is damped, the channel resistance increases gradually due to the gas cooling.
But if the intercomponent current is comparable with the leader current, as
is usually the case by the moment the M-component arises, the increased
resistance of the grounded channel may be suggested to be limited by the
value of R I , M r l L . The reactive parameters of both lines, which are not
very sensitive to the channel plasma state, can be taken to be identical to
those of the leader: C1 M 10pF/m and L1 M 2.7 pH/m.
During the intercloud leader discharge into the earth via the preceding
channel path, the channel resistances change, as in the return stroke
14
12
.
E
3 10
28
g6
b
7-
& 4
0
0 i 2 3 4
x, km
800 -
<
600-
m
a,
a,
c,
4-3
400-
E
t,
200-
0 100 200 300 400 500 600 700

Time, ps
Figure 4.24. Continued.

(section 4.4.4).Suppose that these changes follow the relaxation law expressed
by formula (4.38). This formula describes adequately the qualitative tenden-
cies; there are no quantitative results to compare them with.
This process is described by the long line equations of (4.24) with the
following initial and boundary conditions:
U ( x , O ) = 0 at 0 < x < H I ,
U ( x , O )= U, at H I < x < H I + H2, (4.46)
i ( x ,0) = 0; U ( 0 ,t ) = 0, i(H1 + Hz, r ) = 0.

At the site of contact of the two lines, their potentials are identical at r > 0.
After the contact of the intercloud leader with the grounded channel, current
and voltage waves start running along both lines away from the point of
contact. As a result, the grounded channel becomes charged while the
leader channel is discharged. If the initial parameters of the lines are
identical, the initial current at the point of contact is i = Ui/2Z, where Z
is the wave resistance of the lines. Rapidly attenuated precursors run in
both directions at velocities of electromagnetic signal, while the main current
and voltage waves propagate via the diffusion mechanism (figure 4.24). These
waves spread much stronger than in the return stroke, since the current and
voltage are lower here (the more so that the initial voltage amplitude is
half the value of Ui). The channel resistance decreases more slowly and the
wave fronts become smooth instead of becoming steeper. The initial voltage
Ui in the subsequent components seems to be lower on average than in
the stepwise leader of the first component because this process involves the
increasingly less mature cloud cells with lower charges. If we ignore the
weak displacement current induced by the changing charges of the recharged
channels, the current impulse at the earth can be registered only after the
diffusion wave front has reached the earth. By that moment, the wave has
become very diffuse, so the current impulse front appears to be very
smooth (figure 4.24 (bottom, page 205)).t The more or less uniform power
distribution along the channel is to look as a uniform enhancement of its
radiation intensity. The calculations of this distribution and such current
impulse characteristics as the front steepness, duration, and amplitude are
similar to their observations in M-components. A still better agreement
with the measurements can be attained by varying the parameters in the cal-
culations, primarily R L Land quasistationary field EL in fully transformed
channels. These arguments favour the above suggestions concerning the
nature of lightning M-components.
t The current impulse of a return stroke is of a different form. The current amplitude is registered
right after the short-term commutation process when the leader contacts the earth via its
reducing streamer zone, which takes a few microseconds.

Subsequent components. The problem of a dart leader 207
4.8 Subsequent components. The problem of a dart leader
There is still no clear understanding of the nature of a dart leader, so we shall

discuss the scarce experimental data available and suggest a hypothesis based
on them. Then we shall consider some possible consequences of this hypoth-
esis and the difficulties that may arise. The dart leader problem remains
unsolved but it cannot be put aside because of the importance of the dart
leader process.
4.8.1 A streamer in a ‘waveguide’?

There are no grounds to believe that the mechanism of dart leader initiation
in the clouds is essentially different from that of a M-component. Rather,
both processes result from the closure of an intercloud leader on a grounded
channel remaining after the passage of the return stroke in the previous light-
ning component. But the potential wave running along this track to the
earth, known as a dart leader, differs radically from that of an M-component.
It has a well-defined front identifiable by the intense radiation of dart leader
tip travelling to the earth with velocity ?&L N lo7 m/s, which is at least by an
order of magnitude hgher than the typical velocities of the first stepwise
leader. The ability of a dart leader to travel so fast is especially remarkable
because its potential is most likely to be lower than that of a stepwise
leader. This is indicated by the return stroke currents, which are on average
2-2.5 times lower ( I , M U , / Z ) . The potential drop from the dart leader tip
in the previous, still untransformed channel towards the earth must occur
very quickly. This is indicated by a very fast front rise of the return current
impulse, tf. To gain the full current ,Z M U / Z , where U is the potential
carried by the dart leader, the return wave must run along a leader section
with a rising potential Sx and reach the totally charged portion of the
channel. Therefore, we have Ax ‘U,?,and if the return wave velocity is
N
w, M 10’ m/s and M 0.1 ps, the length of the region with an abrupt poten-
tial drop in the dart leader front is Sx M 10m. It is quite possible that this
value is actually smaller because the return wave cannot at first gain the
full return stroke velocity U, M lo’,/,. On the other hand, the potential
drop region should not be shorter than Ax M vdL?f x 1 m, since the cross
section of a channel with total potential U approaches the earth at velocity
‘udL. Such a steep front of 1-10m is unattainable not only by a diffusion
wave with its potential varying along many hundreds of metres (figure
4.24) but even by an ‘ordinary’ leader of the first lightning component, in
which Ax is determined by the streamer zone length. At the moment of
contact with the earth, the latter is measured in dozens of metres at the tip
potential of 20-30MV. This is the reason why the time necessary for the
return wave current to grow to its maximum value is dozens of times
longer than that for the dart leader.

It follows from the foregoing and the fact that a dart leader travels as
fast as a very fast streamer that the former has no streamer zone which
would serve as the primary prerequisite for a leader mechanism. It appears
that the dart leader, contrary to its name, is essentially not a leader, although
it has a charge cover, which it has to acquire under somewhat different
circumstances (see below). Nor does it look like a diffusion wave of the M-
type. The latter would have an order of magnitude higher velocity and a
very diffuse front.
A dart leader looks more like the oldest of the known types of propa-
gating plasma channel - a streamer, whose head represents an ionization
wave. The velocity of a dart leader is close to that of a high voltage streamer.
The principal reason for a streamer channel being non-viable in air - a rapid
loss of conductivity by the cold plasma - is very weak in this case. A dart
leader follows the track heated by the preceding component, so that the
still-hot track serves as a kind of waveguide to the leader. The high gas
temperature greatly retards electron losses. Therefore, the possibility for the
region behind the tip to be heated to arc temperatures increases considerably.
This provides a stable highly conductive state inherent in a ‘classical’ leader.
The preheated air pipe serves another, probably more important, func-
tion. Its hot and rarefied air is surrounded laterally by cold dense air, Since
the rate of ionization due to the field is described by the E / N ratio, the
radial expansion of the channel region behind the streamer tip is abruptly
retarded as compared with the forward motion of the ionization wave. So
the air mass to be heated by the current is reduced, permitting the channel
gas to be heated to a higher temperature. The cold air restricts the channel
expansion because it acts as a charge cover produced by the streamer zone
of the leader.
One should not think that the channel does not expand through the
ionization mechanism at all. This process is just much slower than the
forward motion of an ionization wave towards the earth, so most of the
Joule heat is released into the yet unexpanded channel having a smaller
radius. The radial field leads to the channel expansion only at the beginning,
as is the case with common streamers (section 2.2.2). When the radial field is
somewhat reduced, the channel becomes the source of a radial streamer
corona which does not require a high field. Radial streamers rapidly lose
their conductivity in cold air, and their immobile charges form a cover of
the type that surrounds a common leader channel. Now, though with some
delay, the mechanism of radial field attenuation and hot channel stabilization
comes into action. Thus, a dart leader, being essentially a streamer (i.e., an
ionization wave having no streamer zone in front of the tip), must possess
a charge cover, as a leader. Unlike the case with a common leader, the
cover is not inherited from a streamer zone but is formed entirely behind
the tip, which is the seat of the principal processes driving the dart leader.
(In a classical leader, the cover formation partly continues behind the tip,

as described in section 2.2.4.)

The streamer mechanism of the dart leader development due to impact
ionization of the gas in the strong field of the tip can manifest itself only if the
conductivity in the channel of the previous component has dropped below a
critical value by the time the next lightning component is to arise. This is
unambiguously supported by the following observations. The M-component
is produced against the background of a continuous current of the inter-
pause, whereas the dart leader arises some time after this current is cut off.
As long as the medium preserves a high conductivity, the diffuse penetration
of the field and current prevents the ionization wave propagation. The diffu-
sion wave has practically no ionization due to a direct action of the low field.
The medium pre-ionization does not stimulate but rather hampers the propa-
gation of the ionization wave. The latter requires a strong field, but the high
conductivity of the medium in front of the wave leads to the field dissipation.
In order to focus the potential drop to a narrow region, one must stop the
charge flux (electric current) by concentrating charge in a narrow region to
produce a strong field. The charge flux can be ‘locked in’ only by creating
resistance to it if one places a poor conductor in front of the well-conducting
portion of the channel.
4.8.2 The non-linear diffusion wave front

At this point, we have to make a short digression to discuss the structure of
the near-front region of a diffusion potential wave. One will see later that this
is directly related to the ionization wave problem. The diffusion wave velocity
w is determined by the propagation process along the whole wave length.
Its variation along the path from the cloud to the earth is illustrated in
figure 4.24. By order of magnitude, the velocity of a non-linear wave is
U FZ x,,/-xf, where -xf is its total length from the source to the initial front
point and xavis an averaged diffusion coefficient in the transformed channel
behind the front, which better fits the final linear resistance of the channel
than to its initial resistance. If constant potential Ui is applied to the initial
channel end, the value of xavdoes not change much. The velocity changes
appreciably over the time, during which the wave covers a distance compar-
able with that between the cloud and the earth. But its change is relatively
small over the time the wave covers a distance of the order of its front
width where the potential U(x) rises steeply. This means that we have
U(x, t ) M U(x - wt)in the wave front, and the distributions of all parameters
along the x-axis are quasistationary in the coordinate system related to the
moving front (as in a non-linear heat wave [12]). With this circumstance in
mind, we can rewrite the potential diffusion equation (4.35) as
(4.47)

Taking into account E = -dU/dx = 0 and U = 0 in front of the wave at

x -+ m, the integral of this equation is
(4.48)
The familiar relation i = rv is valid at every point of the quasistationary

wave portion but not only at the site of the current cut-off in front of the
streamer or leader tip.
The energy W1 per unit length of the quasistationary channel is
described as
(4.49)
and is expressed directly through the local potential. Indeed, reducing the
rank of the set of equations (4.48) and (4.49) by dividing them by one
another, we find
w, - w,o = c1u2/2 (4.50)
where Wlo is the initial energy in the channel far out the wave front. Thus the
statement repeatedly used in evaluations that the energy dissipated in the
channel is of the same order as that stored in its capacitance is valid exactly
in the stationary case.t
We shall consider moderate waves, when the gas is heated at constant
pressure, and the Joule heat is released at constant mass m = r r 2 p = mopo2
per unit channel length (yo and p are the initial radius and gas density in
front of the wave). Then we have W1 = mh, where h is the specific gas
enthalpy. Assume for simplicity that thermodynamically equilibrium ioniza-
tion is established at every point of the wave, so that conductivity 0 and
x = ma(pC1)-’ are the functions of temperature T or h ( T ) . Then x is
unambiguously related to U through formula (4.50).$ With (4.48)-(4.50),
finding the distributions along the wave reduces to the quadrature
1- = -2vx. U=[
2m(h - h,)
cl ] 1’2
. ho=-.w
ml o (4.51)
Let us calculate the integral by approximating the relationship x E o / p

by the power function x = Ahn in the temperature range typical of the wave.
The coordinate origin x = 0 is taken at an arbitrary point of the wave front
t This is quite natural because under the problem conditions the channel is not created anew but
exists from the very beginning with its linear capacitance C , . Then every channel portion is
charged as lumped capacitance (cf. the comment o n formula (2.17) in section 2.2.4).
$In sections 4.7 and 4.4, the quantity x was related to electrical parameters through relation
(4.38) which refers to strong waves with a high energy release. If desired, one can use this relation
after substituting aG/ar by -U dC/dx and doing the above operations.

Subsequent components. The problem of a dart leader 21 1
start, in front of which (x > 0) the channel transforms very slightly, so that x
increases slightly, followed by (x < 0) where it changes noticeably. The
parameters of the initial front point will be marked by the subindex 1, assum-
ing for definiteness x1 = 2x0, where xo corresponds to the initial channel
conductivity. Then we have h l - ho = Sho, where S = 2l/" - 1. An exponen-
tially damping tail of the electric field and current extends forward along the
wave where the diffusion is 'linear':
Within the front, where h exceeds ho considerably or, asymptotically at

x + -CO, we have
By matching the approximate solutions asymptotically valid at x + +xand

x -+ -CO, the parameters of the front start and the matching point coordi-
nate x1 can be found as
U1 [($)r5(9)1'n]1'2,U1 ' El =-
2nxl
Ax
x1 = -.
2n
(4.54)
This point is closer to the a priori position of the front start x = 0 than Ax,
which justifies the approximations.
Let us illustrate this situation numerically with reference to the conditions
typical of the M-component (figure 4.25). Suppose the diffusion wave has
velocity v = 10' mjs running along a channel with the initial radius ro = 1 cm,
temperature To = 5900 K (h, = 14.8 kJ/g) and po = 5 x lop5g/cm3 which
is by a factor of 25 less than the normal; m = 1.54 x 10-4g/cm;
the initial linear resistance R1 = 10R/m, xo = 10 m /s
10 2
ne 1 . 8 10'4cm-3,
~
and C1 = 10pF/m. For the temperature range T x 6-10000K in air at
1 atm, we have a l p = 17h3 (where .[(a cm)-'], p[g/cm3], and h[kJ/g]).
a
Hence, A = 2.7 x lo6 (m*/~)(kJ/g)-~ and S = 0.25. From formulas (4.51)-

(4.53), we find for the initial front point U1 = 3.5 MV, El = 2.2 kV/cm, and
x, 0 A x
Figure 4.25. Schematic diagram of the nonlinear diffusion wave front.

the effective field length before the wave front Ax = 100m. The point behind
the wave with U = lOMV (h M 50 kJ/g, T M 100OOK) lies at a distance
x = 500m from the front. There, x M 3x10" m2/s, the resistance is by a
factor of 30 lower than before the front, and the field drops to 33 V/cm. The
field maximum lies near the initial front point. The qualitative picture presented
in figure 4.25 agrees with the numerical results of figure 4.24(a).
4.8.3 The possibility of diffusion-to-ionizationwave transformation

Let us define the conditions, under which a diffusion wave can transform to
an ionization wave which is supposed to be a dart leader. Consider a simple
situation. It is suggested that potential U, is applied to the upper end of a
grounded conductive channel of the previous lightning component. It
begins to diffuse into the channel. It is assumed that there is no transfor-
mation and the initial conductivity corresponding to the diffusion coefficient
xo is preserved. The diffusion is 'linear' in this case. The potential and field
vary as
(4.55)
E = iR1 =
(..xot) lI2
exp (- A).
t4x0
At every point x, the field first rises with time but then falls after the
maximum E,, = 2(7re)-li2U,/x at moment t = x2/2x0. The point E,,
moves at velocity wg = xo/x, and the potential at this point is U , = 0.33U1.
An ionization wave can be formed if the maximum field is sufficiently
high and exceeds a certain critical value E,. The ionization wave is assumed
to propagate at velocity w, supposed to be equal to that of a dart leader. Since
E,, Nx-l tC1I2, the ionization wave could principally arise at earlier
N
times when E,, > E,, but if its velocity is U, < wg,is immediately overcome
by a diffusion wave. This will not happen if wg drops below U, while E,, is
still higher than E,, i.e., if the conditions vs 3 wg, E,, 2 E, are fulfilled
together. For this to happen, the diffusion coefficient must be smaller and
the linear channel resistance larger:
(4.56)
For this, the gas temperature in the initial channel should not be high. On
the other hand, for the 'waveguide' properties to manifest themselves, the
temperature must be as high as possible to make the air rarefied. Because
of the very sharp temperature dependence of conductivity (when it is
low), these conditions are met only in a very short temperature range,
T M 3000-4000K, where the air density is by a factor of 10-15 lower than

normal. For the estimations, we take Ei = 3 kV/cm corresponding to a field

characteristic of initial air ionization, 30-40 kV/cm under normal conditions.
Suppose w, = 107m/s, Ui = 5MV (such potential usually provides current
IM M lOkA for the next component at 2 = 500R during the return stroke),
and C1 = 10pFjm. We find xOcrx l.lx10sm2/s and Rlcr = 880R/m. The
resistance is two orders of magnitude higher than that supposed to
precede the M-component. For this reason, a dart leader can appear only
after the current cut-off during the interpause and a partial cooling of the
channel.
In reality, the channel undergoes transformation due to the diffusion
wave, its conductivity rises, and the field dissipates faster than what is
expected from the second formula of (4.55). To provide for the critical con-
ditions, the initial conductivity may seem to be lower than the estimated
value. So we should consider the other extremal case when the diffusion
wave heats a limited amount of air and an equilibrium ionization is estab-
lished. The diffusion wave is now non-linear. Its maximum field is near the
initial point of the wave front, and one should use, instead of (4.56), the last
relation of (4.52) similar to it with El = Ei.One should keep in mind that of
interest are the temperatures 3000-6000K, at which 0 varies with T much
more strongly: a l p M 1.8 x 10-6h9 (the dimensionalities are the same as in
the illustration of section 4.8.2). Now we have n = 9; at U, = 107m/s and
channel radius yo = 1 cm, we have U 1 = 1.2MV, xo = 4x lo7m2/s, and
R I , M 2500 a / m . The field extends before the wave front only for Ax = 4m.
The electron density in the initial channel under critical conditions is
ne x 6x 10l2~ m - corresponding
~, to its temperature 4000 K.
4.8.4 The ionization wave in a conductive medium

The values obtained in section 4.8.3 on two extremal assumptions do not
differ much and seem to be reasonable. The problem of the conditions
necessary for a dart leader to arise may seem to have been solved. This
optimism will, however, disappear as soon as one evaluates the parameters
of an ionization wave when it propagates through a medium with critical
conductivity.
Consider a wave in the front-related coordinate system, as was done in
section 4.8.2. The equation for field (4.48) will be supplemented by an ioniza-
tion kinetics equation written directly for x because x CT ne:
N N
-U-
dX = uix, vj =Nf(E/N). (4.57)
dx
This equation describes a new law for the channel transformation. Owing to
(4.48), the ionization frequency 4 turns to the potential function. Then, by
dividing (4.48) and (4.57), the problem is reduced to the equation for x( U )
and the quadrature, as in section 4.8.2.

To advance further, one should choose the functionf(E/N) in a way

suitable for integration.1 But difficulties and doubts arise immediately
here. The approximation of vi by the power function vi = bEk, as in the
streamer theory when this approximation with k = 2.5 provided fairly
good results, does not work in this case. The ionization wave propagating
through a conductive medium appears absolutely diffuse, as in any other
channel transformation law: (4.38) or on the assumption of equilibrium
ionization (section 4.8.2). Let us impart a threshold nature to the function
vi(E) in a simple way - 4 = 0 at E < E* and 4 = const at E > E*.This is
what was done by the authors of [43] when solving a similar problem for
the laboratory ionization wave in a tube. The integration of (4.48) and
(4.57) by the above method yields the following result. The change in the
electron density and x in the ionization wave is defined by the ratio of
potentials U2 and U1 at the points of ionization outset and onset, where
E = E*. A potential 'tongue' of effective length Ax = xo/uextends in front
of the ionization wave, as in other diffusion modes. The potential at the
front is U 1 = E*Ax. The parameter ratios at the wave boundaries are
(4.58)
This relation can be regarded as the dependence of the wave velocity on an

'external potential' U2 applied to its back. On the other hand, the velocity
is expressed by a formula similar to (2.2) for the streamer:
where Axi is the extension of the ionization region from the initial to the final
point of the wave. For a wave to survive, its parameters must meet the last
inequality of (4.59). Otherwise, the field within the wave will be unable to
exceed E*, so no ionization will occur.
The capabilities of an ionization wave are limited, and this limit increases
with increasing initial conductivity of the medium. For example, if the initial
parameters ne0 M lOl2cmP3 and x0 x 107m2/s were even lower than the
critical values found in section 4.8.3 and if the threshold field was 3 kV/cm,
it would be necessary to have the ionization frequency 4 = 2.1 x lo6 s-'
and potential U2 = 300MV in order to increase ne and x by three orders of
magnitude (Ax = 1 m, U 1 = 0.3 MV) and U2 = 30 MV by two orders. The
wave width in t h s case is Axi E 22m, i.e., it is very extended. Only when
t Sometimes, it seems better to describe vi the function of E and, on the contrary, to remove U
from (4.48)and (4.57). Instead of (4.48), we then get

the initial conductivity is still an order of magnitude lower (nCo= 10" cmP3,
xo = lo6m2/s, and R = lo5fl/m), the wave width begins to approach what
would be desired for a dart leader. In the same medium at the same U
and E*, the parameters necessary for the ratio x 2 / x 0= lo3 would be
~ sP1, U2 = 30 MV, Ax = 10 cm, Axi = 5 m, and U , = 30 kV.
vi = 1 . 4 lo7
A still narrower region of the potential rise would be obtained at a still
lower initial conductivity. But then we approach the applicability limits of
the basic concepts of the theory of perturbation propagation in a conductive
medium and of the long line theory, and we are probably coming closer to the
understanding of criteria for the dart leader production.
4.8.5 The dart leader as a streamer in a 'nonconductive waveguide'

The diffusion mechanism of field evolution in a channel, or in a long line, is
incompatible with abrupt potential changes and, hence, with strong fields. If
abrupt changes do arise, they are rapidly smeared by diffusion. We believe
for this reason that neither a narrow ionization wave nor a dart leader can
be formed in a well-conducting channel. To find the conditions, in which a
very strong field can be induced, we should remind ourselves of the prerequi-
sites for the long line equations.
The electrostatics equation for cylindrical geometry has the form:
dEx 1 d
-+--rE -- P (4.60)
r-
dx r dr EO
where p is space charge density. By integrating, in the cross section, a conduc-

tor of radius ro and neglecting the dependence of the longitudinal field E, on
r, we obtain
2 8Ex
m-0 -
dX
+ 27rroEr, = -
EO
7-
, 1;
r = 27rrpdr (4.61)
where Er0is the radial field on the surface of a conductor of length I >> yo:
(4.62)
If the longitudinal field varies along the channel so slowly that the axial diver-
gence can be neglected (the characteristic length for the variation of E, is
Ax >> ro), we arrive at one of the basic conceptions of the long line theory,
~ ( x=) C1U(x), whose implication is the potential diffusion mechanism. It
is suggested implicitly that the resistance varies very slowly along the
channel, so this variation cannot be an obstacle to a charge flux, making
the flux velocity decrease abruptly and create a space charge due to its
local accumulation (a long line has no 'jams').
However, space charge does accumulate at a sharp boundary between a
poorly- and a well-conducting channel portion. A charged tip is formed at
the end of an ideal (or non-ideal) conductor, the potential in front of it

drops abruptly, at distances about equal to ro, inducing there a strong field
capable of sustaining an ionization wave. This is what happens in a
common streamer in a non-conductive medium. It is then clear what is
necessary to support a sharp potential drop at the ionization wave front for
a long time. The conductivity along the perspective trajectory must drop to
a value low enough for the diffusion field tongue to be unable to smear the
sharp potential drop. Therefore, the tongue length must become comparable
with the channel radius Ax x xo/v N yo. Because of the strong temperature
dependence of the degree of equilibrium ionization in air at low temperatures,
a drop to T x 3000K would be sufficient. The equilibrium electron density
established for the long zero-current pause will be neo 10'o-lO1l cmP3;
- -
hence, Rlo 105-1060jm, and xo 106-105m2/s. But the air density in
-
the cooled channel of the previous component at T x 3000K is by an order
of magnitude lower than that of cold air, so that the conductivity drop will
not interfere with the 'waveguide' properties of the track.
The velocity of a dart leader as an ionization wave is defined, in order
-
of magnitude, by the same formula (2.2) as the streamer velocity. But the
'pre-ionization' in this case (nee 10'o-lO'l cmP3) is considerable, and a
much smaller number of electron generations (1n(ne2/neo)x 5 ) is to be
produced in the wave. With the account of the similarity law for vi at
vi - --
an order of magnitude lower gas density, the ionization frequency is
10" sC1 and ro 1 cm; then we obtain a correct order of the velocity
U = qro/ In(ne2/ne0) io7 mjs.
One cannot say that all the details of the dart leader behaviour have been
clarified by the above considerations. For the dart leader channel to be well
conductive, the electron density in it must be at least 5-6 orders of magnitude
higher than the initial value for the track. But the capabilities of the ioniza-
tion wave to produce more electrons are limited. The maximum conductivity
of an ionization wave propagating through a non-conductive medium is
defined, in order of magnitude, by the relation ~ F ~ ~vi (section~ / -
E 2.2.2),
because the space charge of the streamer tip, providing a strong ionization
~
field is dissipated with the Maxwellian time TM = Eo/cF.t After the wave
t It also determines the rate at which the linear charge T = C1U is established, if it is, in the
channel. Let us integrate the relation for charge conservation in the conductor cross section.
Neglecting, for simplicity, the variation in 0 along the channel length, we obtain
Using (4.61) and (4.62), we arrive at a refined equation for the relation between T and U :
The postulate of the long line theory, T = C , U, is valid if the changes in the system, which also
define ~ ( t )occur
, slower than with T , =~ co/u. When applied to the wave front moving at
velocity U in a line with conductivity uo,this happens at 00 >> u i o / r o and xo >> vr0.

Experimental checkup of subsequent component theory 217
has passed, the channel still needs to be heated and ionized, but both pro-
cesses are to occur in a moderate electric field, as in a classical leader channel.
Besides, this must take place before a strong radial field makes the channel
expand beyond the hot gas tube, or if it has already become enveloped by
a stabilizing charge cover (section 4.8.1).
There are still many questions about the processes in a dart leader that
remain to be answered; the development of its quantitative theory is also a
task of further research.
To conclude, it is worth noting some specific features of a current
impulse in the return stroke of subsequent components. Generally, the
impulse duration is related to the time it takes the return stroke to run
along the whole channel. For the subsequent components, this time must
be longer than for the first component due to the attached intercloud
leader. But the impulse duration in the subsequent components is about
twice as short, although the return wave velocities are generally the same.
The reason for this difference is likely to be the absence of branches in a
dart leader. It is quite possible that the relatively slow process of their re-
charging elongates the current impulse tail of the first component. The
impulses of the subsequent components do not reverse the sign, similarly
to those of the first one. In the absence of branches, the action of the reflected
wave can no longer be screened by the randomly reflected waves of the
numerous branches (section 4.4.5). The hypothesis of ‘white noise’ should,
probably, be discarded as being inadequate. This problem, like the others
above, awaits its solution.
4.9 Experimental checkup of subsequent component theory
The theoretical treatment of processes occurring in the channel of the

previous component has been reduced to the various wave propagation
modes - the diffusion mode in the M-component and the ionization wave
mode in the dart leader. The former has a strongly elongated front with a
slowly varying potential, and the latter must possess a tip with a concentrated
charge, producing an abrupt potential change. Indirect evidence for the
significant difference in the field distribution is the registrations of current
impulses at the earth. The impulse front durations are found to differ by
2-4 orders of magnitude between an M-component and a dart leader.
There is a possibility for a direct experimental evaluation of the potential
distribution in a wave approaching the earth. This can be done by measuring
the electric field gain at the earth during the wave motion. If the potential
slowly rises along the whole wave length, as in an M-component (figure
4.24(a)), the distributions of the potential and linear charge from the initial
front point, located at height h, to the cloud can be considered to be
linear, ~ ( x=) a,(x - h ) (x is counted from the earth and x 2 h). For the

field at distance r from a vertical channel. we find
AE(r)= -
aq [ +
lnH ( H 2 +r 2 y 2
- H-h ] (4.63)
27rEO +
h (h2 + r2)1'2 +
( H 2 r2)1/2
where H is the height of the grounded channel. If H is, at least, several times
larger than r, the dependence of field A E on the distance between the
registration point and the channel line will be only logarithmic. The same
is true of the front duration of a field pulse.
The situation must be quite different for a dart leader with the abrupt
potential drop at the wave front, since the first approximation in the field
calculation may assume a uniform potential along the channel and
~ ( x= ) const at x > h. This gives formulae (3.6) and (3.7), which yield the
maximum value AE,,,(r) N r - ' . Such a large difference in the field variation
is easily detectable experimentally, especially if we remember that it concerns
not only the field pulse amplitude but also its front rise time. To see that this
is so, it is sufficient to introduce into (4.63) and (3.6) the h-coordinate for the
wave front, expressed through the respective velocities: h = H - ut.
Triggered lightning is a perfect source for such measurements. A triggered
lightning is initiated by launching a small rocket raising a very thin wire which
evaporates during the development of the first component. The point of
contact of the lightning with the earth is strictly defined, so it is easy to
position current detectors at the necessary distances. Besides, the channel at
the earth follows the wire track and is strictly vertical, as is implied in
the numerical formulae. Such measurements have been partly made [44-451.
In section 3.5, we discussed the measurements of field A E at distances
r1 = 30m and r2 = 500m from the channel during the dart leader develop-
ment. These measurements were not synchronized. However, the ratio
AE(30)/AE(500) = 17.4 for approximately equal currents is nearly the
same as r2/r1 = 16.7.
The field measurements for M-components have been reported only for
r = 30m [42]. The oscillogram of A E ( t ) is accompanied by a simultaneous
registration of a current impulse with the amplitude of 800A and the front
rise time -100 ps. The duration of the impulse front A E is approximately
the same, but the field reaches its maximum of 1350 V/m earlier, when the
current has reached half of its maximum amplitude (until the potential
wave arrives, the current at the earth is zero, whereas the field begins to
rise since its start). Figure 4.26 shows the calculated functions i(t) and
A E ( t ) at the observation points with r = 30m and 500m. The long line
model described in section 4.7.1 was used with the same C1 = 10pF/m,
L1 = 2.7 pH/m, and R I (0) = 10 njm. The length of the grounded channel
was 4000m and that of the intercloud leader contacting it was 2000m. The
experimentally observed current of 800 A was reproduced in the calculation
at the leader potential U, = 9.7MV. Under these conditions, the field

References 219
Figure 4.26. Calculated variations of the electric field at the earth’s surface due to the
M-component under the conditions of figure 4.24. The dashed curve shows the
current impulse I .
amplitude of 1500V/m at the point r = 30 m is close to the measured value. It

follows from figure 4.26 that the temporal parameters of the current impulse
are also consistent with the measurements. At the point r = 500m, the
calculated field amplitude is a factor of three smaller and the time for the
maximum amplitude is nearly the same as for r = 30m. Both parameters
would differ by an order of magnitude in a dart leader with this increase in
r . Therefore, the diffusion model of the M-component reproduces fairly
well the available observations. It would, certainly, be most desirable to
make simultaneous field registrations at different distances from a grounded
lightning channel.
References
[l] Berger K, Anderson R B and Kroninger H 1975 Electra 41 23

[2] Idone V P and Orville R E 1985 J. Geophys. Res. 90 6159
[3] Antsurov K V, Vereschagin I P, Makalsky L M et a1 1992 Proc. 9th Intern. Con$
on Atmosph. Electricity 1 (St Peterburg: A I Voeikov Main Geophys. Observ.)
360
[4] Vereschagin I P, Koshelev M A, Makalsky L M and Sysoev V S 1989 Izvestiya.
Akad. Nauk S S S R , Energetika i transport 4 100
[5] Simpson G C and Robinson G D 1941 Proc. R . Soc. London A 117 281
[6] Kasemir H W 1960 J . Geophys. Res. 65 1873


[8] Proctor D A 1971 J. Geophys. Res. 76 1078
[9] Mazur V, Gerlach J C and Rust W D 1984 Geophys. Res. Lett. 11 61
[lo] Mazur V, Rust W D and Gerlach J C 1986 J. Geophys. Res. 91 8690
[ll] Raizer Yu P 1991 Gas Discharge Physics (Berlin: Springer) p 449
[12] Zel’dovich Ya B and Raizer Yu P 1968 Physics of Shock Waves and High-
Temperature Hydrodynamic Phenomena (New York: Academic Press) p 916
[13] Schonland B 1956 The Lightning Discharge. Handbuch der Physik 22 (Berlin:
Springer) 576
[14] Orvill R E 1999 J. Geophys. Res. 104
p 294
[17] Abramson I S, Gegechkori N M, Drabkina S I and Mandel’shtam S L 1947 Zh.
Eksper. i Teor. Fiz. 17 862
[18] Drabkina S I 1951 Zh. Eksper. i Teor. Fiz. 21 473
[19] Dolgov G G and Mandel’shtam S L 1953 Zh. Eksper. i Teor. Fiz. 24 691
[20] Braginsky S N 1958 Soviet Phys. JETP 7 (Eng. Trans.) 1068
[21] Zhivyuk Yu N and Mandel’shtam S L 1961 Soviet Phys. JETP 13 (Eng. Trans.)
338
[22] Plooster M N 1971 Phys. Fluids 14 2111 and 2124
[23] Paxton A N, Gardner R L and Baker L 1986 Phys. Fluids 29 2736
[24] Sneider M N 1997 Unpublished report
[25] Gorin B N and Markin V I 1975 in Research of Lightning and High-Voltage
Discharge (Moscow: Krzhizhanovsky Power Engineering Inst.) p 114 (in
Russian)
[26] Bazelyan E M, Gorin B N and Levitov V I 1978 Physical and Engineering
Fundamentals of Lightning Protection (Leningrad: Gidrometeoizdat) p 223 (in
Russian)
[27] Gorin B N 1985 Elektrichestvo 4 10
[28] Gorin B N 1992 Proc. 9th Intern. Conf. on Atmosph. Electricity 1 (St Peterburg:
A I Voeikov Main Geophys. Observ.) 206
[29] Jordan D M and Uman M A 1983 J. Geophys. Res. 88 6555
[30] Rakov V A and Uman M A 1998 IEEE Trans. on E M Compatibility 40 403
[31] Berger K 1977 in Lightning, vol. 1, Physics Lightning (R Golde (ed) New York:
Academic Press) p 119
[32] Berger K and Vogrlsanger E 1966 Bull. S E V 57(13) 1
[33] Schonland B, Malan D and Collens H 1935 Proc. Roy. Soc. London Ser A 152
595
[34] Schonland B, Malan D and Collens H 1938 Proc. Roy. Soc. London Ser A 168
455
[35] Orvill R E 1968 J . Geophys. Res. 73 6999
[36] Orvill R E and Idone V P 1982 J . Geophys. Res. 87 11177
[37] Krider E P 1974 J. Geophys. Res. 79 4542
[38] Uman M A 1969 Lightning (New York: McGraw Book Company) p 300
[39] Fisher R G, Rakov V A, Uman M A et a1 1993 J . Geophys. Res. 98 22887
[40] Fisher R G, Rakov V A, Uman M A et a1 1992 Proc. 9th Intern. Con$ on Atmosph.
Electricity 3 (St Petersburg: A I Voeikov Main Geophys. Observ.) p 873

References 22 1
[41] Bazelyan E M 1995 Fiz. Plazmy 21 497 (Engl. transl.: 1995 Plasma Phys. Rep. 21
470)
[42] Rakov V A, Thottappillil R and Uman M A 1995 J . Geophys. Res. 100 25701
[43] Sinkevich 0 A and Gerasimov D N 1999 Fiz. Plazmy 25 376 (Engl. transl.: 1999
[44] Rubinstein M, Rachidi F, Uman M A et a1 1995 J . Geophys. Res. 100 8863
[45] Rakov V A, Uman M A, Rambo K J et a1 1998 J. Geophys. Res. 103 14117

Chapter 5
Lightning attraction by objects
In this chapter, we shall describe the way a lightning channel chooses a point
to strike (a terrestrial or a flying body). This is the principal issue for lightning
protection technology. In any case, a direct stroke is more hazardous than a
remote lightning effect via the electromagnetic field or shock wave in the air.
Historically, direct lightning strokes were observed earlier than indirect ones,
and the first research into lightning protection problems was associated with
direct strokes.
Everyday experience and scientific observations, including those made
as far back as the 18th century, indicate that lightning most often strikes
individual structures elevated above the earth. These may be towers,
churches, houses on high open hills, and just high trees. Today, this list is
much longer and includes power transmission lines, transmitting and
receiving antennas, and the like. The experience in maintaining such
structures indicates that the frequency of strokes increases with the
object’s height. This observation was used as a basis for the most common
lightning protection techniques. A grounded rod higher than the object to
be protected - a lightning rod - put up in the vicinity of the object is
supposed to attract most strokes, thus protecting the object. The underlying
principle of this approach has not changed since the first lightning rod was
constructed two and a half centuries ago. What has changed is the require-
ment for the protection reliability, which have become extremely stringent.
For this reason, the specialists have to deal with exceptions rather than the
rules, focusing on the rare cases of lightning breakthroughs to the object
being protected, because they lead to emergencies and sometimes to
catastrophes.
The study of lightning attraction mechanisms is extremely time-
consuming and expensive. Even a simple measurement of the number of
lightning strokes at objects of various heights is very hard to arrange.
Most apartment houses and industrial premises in Europe are less than
222

The equidistance principle 223
50m high. On the average, a lightning strokes a 50m building once in five
years. Every kilometre of a power transmission line 30m high attracts
approximately one lightning discharge per year. Long-term observations of
a large number of buildings and multi-kilometre transmission lines are
necessary to accumulate a representative statistics. The difficulties increase
many-fold when one needs to extract information on the protection reliabil-
ity from the observational statistics. To illustrate, 10-20 years of continuous
observations of a 50 m building would be required to obtain information on
the lightning discharge frequency, and at least 1000 years would be necessary
to check whether its lightning rod can really provide a ‘99% protection’
promised by the rod producers.
In a situation like this, one has to resort to theoretical evaluations,
and this is one reason why lightning attraction theory has been the focal
point of research for many lightning specialists. Here, as in many other
lightning problems, there is an acute lack of factual data. The available
evidence obtained from laboratory investigations on long sparks does not
always provide an unambiguous interpretation, and this makes one treat
with caution many, even generally accepted, concepts. We shall focus on
the most advanced approaches, discussing, where necessary, alternative
hypotheses.
5.1 The equidistance principle
This approach is oldest and clearly correct in its theoretical formulation.

Suppose that an object of small area and height h is located on a flat earth’s
surface (it is a rod electrode in laboratory simulations). Let us assume further
that a lightning channel is shifted from it horizontally at a distance r, and the
channel tip is at an altitude Ho (figure 5.1). In order to predict whether
the lightning will strike the object or the earth, we shall take into account
the breakdown voltage measurements of long air gaps with a sharply non-
uniform electric field. They show that the longer the gap, the higher the
average voltage required for its breakdown and the longer the time necessary
for the discharge formation. This means that the shortest gap has the best
chance to experience a breakdown, provided that the same voltage is applied
simultaneously to several gaps. Let us keep in mind that the distance from the
lightning tip to the object, [(Ho- h)’ + r2I1/’, is shorter than that to the
earth’s surface Ho at
The distance Re, is known as the equivalent attraction radius for an object of
height h. It indicates the surface area, from which lightning discharges that

224 Lightning attraction by objects
Figure 5.1. Estimating the equivalent attraction radius.
have descended to the altitude Ho are attracted by the object. For a compact
object of small cross section, this is a circle of area SeqM T R ; ~for; an
extended object of length L >> h and width b << h (e.g., a power transmission
line), this is a stripe of area Seq x 2ReqL.The average number of strokes per
storm season is evaluated from Seqas
NI = nlSeq (5.4
where nl is the year density of lightning discharges into the earth at the
object’s site. Global and regional maps of storm intensity are made from
meteorological survey data [l, 21. The n1 data are usually given per 1 km2
per year. Quite often, the maps indicate the number of storm days or
hours, together with empirical formulae to relate this parameter to nl.
The equidistance principle, simple and clear as it may seem, is of little
use, because one can employ formulas (5.1) and (5.2) to advantage only if
one knows the altitude Ho (the attraction altitude), at which a descending
lightning leader begins to show its preference and selects the point to
strike. The condition of the earth’s surface and the objects located on it
cannot influence the lightning behaviour high up in the clouds. A lightning
develops by changing its path randomly. As it approaches the earth, the
field perturbation by charges induced by terrestrial objects become
increasingly comparable with random field fluctuations. Eventually, the
perturbation begins to play the dominant role, determining the channel
path more or less rigorously. The average altitude Ho at which this happens
is known as the attraction altitude.
It is unlikely that the altitude Ho should be determined only by the
terrestrial object’s height h. It must also depend on the leader field varying
statistically with the lightning due to the variations in the storm cloud

The equidistance principle 225
charge, the starting point of the descending leader, its path, number of
branches, etc. This diversity of lightning conditions is uncontrollable. The
only parameter that can, to some extent, depend on observations is the
attraction altitude averaged over all descending discharges. It deserves
attention because the averaging will require only the statistics of descending
lightnings which have struck objects of various heights. These statistics
cannot be said to be reliable but it provides some factual information
important for lightning protection practice.
Before we use the stroke statistics in a theoretical treatment, we think it
worthwhile defining the range of object heights. Unfortunately, one has to
discard the stroke data concerning high constructions. Ascending discharges
become dominant at heights h > 150m. Data on such strokes cannot be
included in the statistics without reservations, even though they were
obtained from well-arranged observations, in which every discharge was
identified unambiguously. The point is that ascending lightnings partly
discharge the clouds, reducing the number of descending discharges. This
interference into the storm cloud activity is so appreciable that a further
increase of h above 200 m does not practically change the stroke frequency
of an object by descending lightnings. Of little use are the data on low
structures (10-15m). The number of strokes in this case is greatly affected
by the nearest neighbours and the local topography. Account should be
taken of the statistics for low buildings, but such observations are scarce.
The overall data have a too large spread.
The authors of [3] selected the most reliable data and, by averaging
many observations, derived the relationship between the number of descend-
ing strokes and the terrestrial object height. Figure 5.2 shows individual
representative values to demonstrate the data spread. All of the results are
normalized to the intensity of the storm cloud activity, which is 25 storm
days per year. In the range h 9 150 m considered, we can admit, with some
reservations, the existence of a quadratic height dependence of the number
of lightning strokes for concentrated objects and a linear dependence for
extended ones. Both dependencies mean Re,/h const.
Figure 5.2 shows that the expression Re,= 3h, sometimes used for
rough estimations of the expected number of strokes, agrees fairly well
with the averaged values of Re, derived from observations. The substitution
of Re, = 3h into (5.1) yields for the average attraction altitude for descending
leaders
Ho = 5h. (5.3)
This does not seem to be a large height. A lightning is insensitive to the
earth’s surface along most of its path and it is only its last 50-500m which
are predetermined. Below, we shall discuss the mechanism of the more or
less rigid determination of the leader behaviour by a particular site on the
earth (section 5.6).

0.
Figure 5.2. The average number of strokes per year for compact (top) and extended
(bottom) objects of height h. The dashed curves bound spread zones in observation
data. Solid curves are plotted using the equivalent attraction radius.
5.2 The electrogeometric method
Popular among some lightning specialists, this method of calculating the

number of lightning discharges into a grounded structure [4-81 should be
considered as a modification of the equidistance principle. The main

The electrogeometric method 227
Figure 5.3. Lightning capture regions.
calculation parameter in this method is the striking distance r,. Surfaces

located at a distance r, from the upper points of a structure (roof), the
adjacent buildings, and from the earth’s surface define by means of their
interception lines the lightning capture regions (figure 5.3).
The further path of a lightning channel which has reached the capture
region is considered unambiguously predetermined. The leader will move
to the object (or to the earth), whose capture surface it has intercepted.
With these initial assumptions, the calculation of the number of strokes NI
reduces to geometrical constructions, since the lightning density nl at an
+
altitude z > h r, is considered to be uniform, and the value of NI can be
calculated if one knows the area S, of the capture surface projection on to
the earth’s plane, NI = ytlSs.
The long history of the electrogeometric method has witnessed only one
improvement - that of the selection principles concerning the striking
distance r, [l, 31. Discarding inessential details, the quantity Y, is found from
an average electric field E, between the object’s top (or the earth’s surface)
and the lightning leader tip which has reached the capture region. Usually,
the values of E, were taken to be equal to the average breakdown strengths
of the longest laboratory gaps. As the laboratory study of increasingly longer
sparks progressed, the values of E, introduced into the calculation method
decreased from 6 to 2 kV/cm, entailing larger striking distances. In this
approach, the parameter r, is independent of the grounded object’s height
but is sensitive to the leader tip potential U, (Y, zz U,/&). However for applica-
tions, attempts are made to find the relation to the current amplitude Z,w of the
return stroke, rather than U,, using simulation models of the kind discussed in
section 4.4. If the function Y, =f(ZM) and, hence, S,(ZAw) are known, the
number of lightning strokes at an object with current 1 , > Zlwo is found as

Figure 5.4. The dependence of striking distance on lightning currents. The lower
curve is plotted using [4] data, the upper using [5] data. The spread region is hatched.
where p(Z) is the probability density of current of amplitude Z found from

natural measurements. To find the total number of strokes, the lower limit
of the integral in (5.4) should be taken to be zero.
The generally correct idea of differentiating distances r, in current
amplitude actually fails to refine the calculation of N I . There is no factual
information to determine the function rs = f ( Z M ) experimentally, while
theoretical evaluations suffer from an unacceptably large spread, so that
the values obtained by different authors differ several times (figure 5.4).
The stroke statistics for objects of various heights could, to some extent,
be used for fitting the calculated total number of strokes N I , but it proves
unsuitable for finding the function r, =f(Z,w).
The calculation procedures in the electrogeometric approach do not
involve a strong dependence of stroke frequency on an object's height.
Indeed, for a single construction, like a tower, the capture region projection
on to the earth's plane is a circle of radius
R = (2r,h - h2)'I2 at rs 2 h
(5.5)
R = r, at r, < h
and for an extended object, like a transmission line, it is a stripe of width 2R.
Therefore, the number of strokes of low power lightnings with a small stroke
distance (rs < h ) will be entirely independent of the object's height, while the
frequency of powerful discharges with r, > h must increase with height
slower than h for compact objects and as for extended ones. Actually
both of these dependencies are steeper.
5.3 The probability approach to finding the stroke point
A predetermined choice of the discharge path through an air gap contradicts

the experience gained from long spark investigations. Neither a spark nor a

The probability approach to Jinding the stroke point 229
lightning travel along the shortest path. When voltage is simultaneously

applied in parallel to several air gaps of various lengths, it is the longest
gap that is sometimes closed by a spark. This is supported by the large
spread of breakdown voltages: the standard deviation u for multi-metre
gaps with a sharply non-uniform field is 5-10% of the average breakdown
voltage.
If two gaps, tested individually, possess the distributions of breakdown
voltage of the probability densities cpl ( U ) and p2(U ) ,then, provided that the
voltage of a common source is applied simultaneously, the breakdown prob-
ability for one of the gaps, say, the first one, is described as
where a2is the integral distribution defining the probability of the gap break-
down at a voltage less than U . If the distributions cpl and p2 are described by
the normal law with the standard deviations u1 and u2 and by the average
values of Uavl and Uav2, the breakdown voltage difference AU = U1 - U,
also obeys this law, with AU,, = Uavl - Uav2 and a = (a: +
This
allows us to rewrite (5.6) using the tabulated probability integral:
P1 = A2 [I - g / : e x p ( - x 2 / 2 ) d x 1 , A= Uavl - u a v 2
(0: + a;,li2.
The expressions of (5.7) are valid for Uavl 2 Uav2.Otherwise, one should find
(5.7)
the breakdown probability P2 for the second gap, writing for the first one
P1 = 1 - P2.
The formal relations of the probability theory (5.6) and (5.7) are valid if
the discharge processes in the gaps do not affect one another and if every
individual breakdown can be considered as an independent event. Multi-
electrode systems of this kind can be termed uncoupled. A classical example
of an uncoupled multi-electrode system is an insulator string of a power
transmission line. The distance between the adjacent towers is so large that
there is no electrical or electromagnetic effect of discharges occurring in
one string on those of its neighbours. The earth’s surface and an object
located on it can also be regarded as an uncoupled system, with a descending
lightning leader acting as a common high voltage electrode. Such systems
have been studied in laboratory conditions [9],in which the distribution of
breakdown voltage was used as the indicator of an uncoupled nature of
the system. If the individual gaps comprising a system are tested individually
and have the integral distributions Q1( U ) and a2(U ) with the probability
densities cpl(U) and cp2(U),the system will have the following distribution
of the breakdown voltages:
(5.8)

Figure 5.5. The breakdown voltage probability for the uncoupled multielectrode
system involving the high-voltage and two grounded electrodes. x : measured
QsYs(U),0 : measured @ ( U )for the single gap. The dashed curve is evaluated for
the system using @ ( U ) .
In the particular case of equal gap lengths with @I ( U ) = (a,( U ) = @( U ) ,we

have
@&q = 1 - [l - qU)]”*. (5.9)
Therefore, the uncoupled character of a system can be tested experimentally
by comparing the measured distribution of its breakdown voltages with those
calculated from formulas (5.8) and (5.9) and the distributions in the
individual gaps. Experiments show that if the distance between grounded
electrodes is comparable with their height, the leader processes in each gap
develop independently, so that the system they comprise can, indeed, be
regarded as an uncoupled one (figure 5.5).
Suppose now that the attraction of a descending leader begins when its
tip reaches the altitude Ho. The problem reduces to finding the breakdown
path in an uncoupled system with a common high voltage electrode - the light-
ning leader - and two grounded electrodes, namely, the earth’s surface and an
object of height h located on it. The probability of lightning attraction towards
the object from the point with the x- and y-coordinates in the attraction plane
is equal to that of the gap bridging between the leader tip and the object’s top,
P,(x,y). This probability is defined by the integral of (5.7). When the relative
standard deviations are identical, al/UaVl= a2/Uav2= a,, at identical
average breakdown voltages, the upper probability limit is expressed through
the shortest distance from the leader tip to the earth, d , ( x , y ) , and to the

The probability approach to jnding the stroke point 23 1
object, d o ( x , y ) :
(5.10)
The expected total number of lightning strokes at the object, NI, is found by
integrating Pa over the attraction plane. If the earth's surface is flat and the
lightning discharge density nl is constant, then we have for a compact object
of height h and for an extended object of average height h and length L ,
respectively:
NI = 2nnl :1 P,(r)rdr, NI = 27rnlL[r P,(y) dy,

(5.11)
[z2+ (Ho - h)2]'12- Ho
A, = z = r,y.
+
ua[z2 (Ho - h)2 + H i ]
!
The relations obtained from the equidistance principle are identica, to (5.11)
at cr, = 0.
In virtue of the approximate symmetry of the function Pa(r)relative to
the point with Pa = 0.5 ( r / h E 3; see figure 5.6), the calculations of NI slightly
depend on the standard attraction deviation 0,. When cra varies from zero
to 10% (there are practically no greater deviations in pure air), the value
of NI increases only by 15% for a compact object and by less than 5% for
an extended one. It would be unreasonable to discard the simple and clear
equidistance principle for the sake of this small correction, but for the greatly
inclined (almost horizontal) paths of lightnings attracted by objects.
rh
Figure 5.6. Evaluated attraction probability for the attraction altitude Ho = 5h ( h is
the object height, Y is the object-lightning stroke distance).

lightning
rod f
hr
I / , , a ,,,//,/,,/;
h0
I
Ir I-
Figure 5.7. Why a lightning rod is less effective when an inclined lightning approaches
from a side of the protected object.
It is clear from the foregoing that the larger the distance between the
lightning and the object (compared with that from the lightning to the rod)
the greater the protective effectiveness of a lightning rod. For a lightning
travelling in the attraction plane strictly above the lightning rod, the
difference between the two paths (figure 5.7) is largest:
Ad = [(Ho- ho)’ + - Ho + h,
A d z h, - ho at a << [2(h,- ho)(Ho- h0)]1/2.
With increasing side shift of the lightning in the attraction plane r , the value
of A d decreases, and this decrease is especially noticeable when the lightning
approaches from the side the object being protected, as is shown in figure 5.7:
Ad = [(Ho- h0)’ + ( r - a)2]1/2- [(Ho- A,)’ + ~’1”’.
In the limit r + x,the distance to the object is smaller than to the lightning
rod ( A d x -a), and it is quite ineffective for lateral strongly inclined lightnings
coming from the object side. In the equidistance approach, there should be no
events like ths, since lightnings are not to strike an object at a distance r > Req.
In reality, the proportion of lateral strokes is found to be fairly large. The fact
that the probability method considers this circumstance correctly (figure 5.6) is
very important for the evaluation of the lightning rod effectiveness.
5.4 Laboratory study of lightning attraction
Laboratory investigations of lightning attraction were initiated in the 1940s

by simulating a descending lightning by a long spark and placing small model

Laboratory study of lightning attraction 233
rods and objects to be protected on the grounded floor [lo, 111. At that
time, experimental researchers expected to derive information necessary for
a numerical evaluation of lightning rod effectiveness. The naive optimism
has long vanished. The measurements showed that the attraction process
did not obey similarity laws. Essentially different results were obtained
from gaps of different lengths and different time characteristics of the voltage
pulses applied [12-141. But the interest in laboratory investigations of
lightning has survived, and they are currently performed in an attempt to
understand the attraction mechanism of long leaders.
The primary question is when the attraction begins. Clearly, the
condition of the earth’s surface does not affect the leader propagation
while its tip is far from the earth. Here, the spark paths become distributed
randomly. If one projects a multiplicity of paths on a sheet of paper and
finds the mean deviation Ax from the normal passing through a high-voltage
electrode with the account of the sign (e.g., plus on the right and minus on the
left), one obtains Ax = 0 for altitudes z > Ho. The mean path in a gap
perfectly symmetrical relative to the normal proves strictly vertical down
to the altitude Ho. The attraction onset is indicated by the mean path
deviation towards the electrode simulating a terrestrial object (figure 5.8).
Data treatment for determining the attraction altitude was made in [14] for
spark discharges of up to 12 m in length. The path statistics involved different
time characteristics of the voltage pulse. In the case of a steep pulse front
( 6 p ) , a leader was attracted from the moment of its origin; for a smooth
front (250p), it had enough time to cover an appreciable gap length
before the deviation towards a grounded electrode became noticeable
(figure 5.9).
I I
I- r-I
Figure 5.8. Determination of the attraction altitude Ho by the bend point of a mean
path deviation onset.

0.0 0.2 0.4 0.6 0.8 1.0

Lld
Figure 5.9. The average leader deviation from the high-voltage electrode axis toward
the grounded electrode, Ar, depending on the leader length L. The results are given
for configuration of figure 5.8 and two voltage fronts, t f ; d = 3m. In fact each
curve presents a leader trajectory in cylindrical z - r coordinates, averaged over
many experiments.
The reason for this is as follows. When the voltage rises rapidly, the
streamers of the initial corona flash reach the grounded cathode, and the leader
develops in the jump mode from the very beginning. But when the voltage rises
slowly, the leader channel covers about one third of a 3 m gap before the transi-
tion to the jump mode. This suggests that the streamer zone imposes a definite
direction on the leader as soon as it touches the grounded electrode. If t h s is
the case, the attraction of a laboratory leader must begin later in a long gap
than in a short one. The experimental data presented in figure 5.10 show that
the attraction delay time does increase with the length of the discharge gap d:
Ho = d at d = 0.5 m but Ho x (0.4-0.5)d at d = 10 m. The ratio of the streamer
zone length to d decreases nearly as much by the moment of the final jump.
Another independent method for the study of spark attraction is to use a
blocking electrode. Suppose we are able to set up instantaneously a metallic
electrode on a grounded plane in the right place at the right moment of time.
Let us do this many times for different lengths of a developing leader channel
L and plot the probability of the electrode striking, Bo, as a function of L .
The possible curves are presented in figure 5.1 1. The first version corresponds
to the ‘instantaneous’ choice of the striking point at an altitude Ho < d (at
the critical leader length L,, = d - Ho).A probability of the electrode striking
falls sharply if the electrode is set up with delay (at L > Lcr)since the leader has
already chosen some other point for stroke at the moment of the leader start
(Ho = d, L,, = 0) while in the third version the leader chooses a striking
point gradually also but beginning from the altitude Ho < d when L,, > 0.

Laboratory study of lightning attraction 235
1.0
0.8
0.6
0.4
0.2
Figure 5.10. The deviation Ar at various gap length d for tf = 250ps under the
conditions of figure 5.9.
Figure 5.11. ‘Blocking electrode’ experiment. A supposed qualitative probability @ of

a stroke to the electrode when: ( A )a leader is instantly chooses the striking point when
its length reaches the critical length L,,, ( B ) a leader is gradually attracted from the
very beginning, (C) a leader is gradually attracted reaching Lc,. (D)measurements
for d = 3 m, tf = 6 ps (curve 1) and 250 ps (curve 2).

This can be done experimentally if the electrode displacement is replaced

by its screening by an electric field. The electrode should be insulated from
the earth and a high voltage of the same sign as that of the leader should
be applied to it. This will create a counter-propagating field which will
block the electrode from the leader. The electrode will become accessible
only after the blocking voltage is cut off. The electric circuit provides a
precise control of the voltage cut-off [ 131. The experimental relationships
in figure 5.11 show again that the attraction begins since the leader origin,
if it develops in the final jump mode from the very beginning. In long gaps
with a smooth voltage pulse, when the initial leader phase is well defined,
the attraction is delayed as much as the transition to the final jump (curve
2 in figure 5.1 1).
Experiments on negative leaders have yielded similar qualitative results
[15], but the attracting effect of a grounded electrode on a negative leader
proves to be stronger.
5.5 Extrapolation to lightning
The scale of laboratory experiments, 1 : 100 or 1 : 1000, is too small to resolve

the details or to make long-term predictions. Laboratory studies have so far
failed to clarify an important point: Does the attraction onset really coincide
with the moment of the leader transition to the final jump, or is this process
controlled by a counterleader starting from the grounded electrode? The
interest in the counterleader and its relation to lightning attraction arose
long ago [4]. The counterleader seems to increase the altitude of the grounded
electrode. The difference between the tip potential and the external potential,
A U , increases, so the counterleader goes up with acceleration (section 4.1.2)
to meet the descending leader.
One of the difficulties is that the moments of the descending leader tran-
sition to the final jump and of the counterleader origin in a laboratory are
hardly discernible. The experiment accuracy is insufficient to separate them
reliably in conventional laboratory gaps of about 10 m long. For lightning,
these moments may be considerably separated, but a direct measurement is
practically unfeasible. So, one has, as usual, to rely on numerical evaluations.
Let us first evaluate the field perturbation in the atmosphere by the
charge of the grounded electrode of height h and radius ro before a counter-
leader starts from it. The external threshold field necessary for a counterlea-
der to arise and develop is defined by formula (4.1 l), in which d must be
equalized to h. The field Eo for an industrial building of height h = 50m
was found to be 350 V/cm at the parameters used in section 4.1.1. Note
that this field results not so much from cloud charges as from the charge
of the descending leader approaching the earth. The field induces a charge
on the grounded rod, whose density per unit length can be considered to

Extrapolation to lightning 237
depend linearly on height: ~ ( z=) aqz (section 3.6.2). The value of a, is

defined by (3.11) where d = h and r = r0. For Eo = 350V/cm, h = 50m
and ro = 0.1 m, we have aq M 3 x lop7C/m2.
The field gain AEo at the altitude zo above the rod, associated with the
rod charge and its reflection by the earth, is
AE, = 4 = &( p 2zoh - l nzo e- h) . (5.12)
At the attraction altitude zo = Ho M 5h = 250m and the found value of aq,

we get AEo M 30V/m. This is about IOp3 of the unperturbed atmospheric
field at the altitude Ho and 3 x lo4 times lower than the field in the streamer
zone of a negative leader. It is hard to imagine a lightning leader which would
respond to such weak perturbations. In any case, laboratory experiments
have failed to reveal changes in the breakdown probability of a gap for
such a small relative increase in the voltage. Consequently, the attraction
process cannot begin before the counterleader is excited.
Let us follow the excitation of a counterleader by the field of a descend-
ing leader, relating the tip altitude of the latter to the grounded rod height.
For this, expression (4.1 1) should be supplemented by the dependence of
the average near-earth field on the descending leader charge. Consider a
simple situation. Suppose a descending leader starts at altitude H 1 and
moves together with its partner, a positive ascending leader, vertically
without branching right above the grounded rod of height h. At the
moment the descending channel acquires the length L , with its tip having
descended to the altitude Ho = H 1 - L, the potential of the leader charge
at the rod top, together with the charges reflected by the earth, is
(p, a, [ ( H l - h ) In 2H1 - Ho - h - ( H l
=-
4T&o Ho - h
+ h ) In 2H1Ho- Ho
+ h-t '1. (5.13)
The field average in the rod height is E,, = p q / h + Eo (with the account of
the cloud field Eo). By equating Eay to the threshold field necessary for the
excitation of a viable counterleader (formula (4.1 l)), we find the attraction
altitude Ho from (5.13), assuming the attraction to begin at the moment of
the counterleader start.
We shall not be interested in the quantity Ho linearly dependent on the
poorly known parameter aq to be averaged over all descending lightnings.
Rather, we shall focus on the tendency in the variation of the Ho/h ratio
with varying h in the range 10-150m. Buildings lower than 10m are rarely
affected by lightning, while the picture for high structures is greatly distorted
by ascending lightnings, as pointed out above. Suppose that the attraction
altitude for an object of average height, say, h = 50 m, is found from (4.11)
and (5.13) to be really close to the experimental value Ho = 5h. This yields
the estimate for a, (which is aq/4neo E 1.5 kV/m at Eo = lOOV/cm and

H I = 3 x 103m), permitting the calculation of Ho/h for constructions of

other heights. The calculations are
h.m 10 20 30 50 100 150
Holh 9.3 7.0 6.0 5.0 4.0 3.6
Of course, this is not the linear dependence Ho FZ h obtained from a prelimin-
ary treatment of observational data. The attraction altitude definitely rises
with the object’s height, as Ho x ho65 according to the calculation. A
better agreement could hardly be expected since the observational data are
limited and have a very large spread.
Another result would be obtained if one related the attraction onset to
the moment of the leader transition to the final jump. The attraction altitude
+
would then be determined by the streamer zone length L,, Ho = h L,. The
length L, only slightly depends on the grounded rod height. At its zero
height, the streamer zone is totally created by anode-directed streamers of
the negative descending leader, which require an average field of 10 kV/cm
(there is no counter-discharge). If the rod has a large height, the active vol-
tage is shared equally between the streamer zones of the descending leader
and the positive counterleader. Cathode-directed streamers of the latter
can develop in a 5 kV/cm field, thereby decreasing the average field in the
common streamer zone, at most, by a factor of 1.5-2. This would set a
limit to the possible variation in the attraction altitude. It may seem that
the result obtained is quite promising. The attraction altitude at the leader
potential U x 100MV is also found to be close to lOOm for low objects
and about 300m for structures 100-150m high. But one should keep in
mind that only unique unbranched leaders are capable of delivering to the
earth such a large cloud potential (section 4.3.2). Such leaders occur rarely
in nature; the potential of a normally branched lightning is several times
lower. As smaller will be the streamer zone length proportional to U . The
attraction altitude would then become equal to the object’s height, provided
it is not too low.
In other words, the attempt to relate the attraction process to the final
jump unambiguously relates the quantity Ho to the potential of a descending
leader, making it strongly dependent on the factors discussed in section 4.3.2,
which change this potential (e.g., branching). If one relates the attraction to
the excitation and development of a counterleader, the dominant factor will
be the total charge delivered by all the components of a descending leader to
the earth.
The idea that the attraction onset is associated with the excitation of
a counterleader leaves little hope for an unambiguous relationship between
Ho and the return stroke current I M , as was implied, for example, by
the electrogeometric method. Indeed, the current IM is determined by the
potential U, delivered by the leader to the earth (section 4.4. l), whereas the
field at the grounded rod is due to the total charge of the descending

On the attraction mechanism of externaljeld 239
leader. The branching and path bending typical of a descending lightning

greatly affect the value of Ui and, hence, 1, (section 4.3.3), but they do
not much change the total leader charge.
5.6 On the attraction mechanism of external field
There is no doubt that lightning attraction is due to the electric field which is
related to the object. It is difficult to imagine another remote way to affect a
leader. As for the field source, the evaluations made in section 5.5 show that
the field created by the charge induced in the object itself proves very weak.
When the distance between the descending leader tip and the object is
sufficient for an attraction effect to reveal itself, the object charge field at
the leader tip is by a factor of 102-103 lower than the cloud charge field.
There are no reasons why such a slight perturbation should make the
leader change its path, which is subject to various random bendings even
without the influence of any terrestrial objects. No doubt, a counterleader
excited by the object serves as a mediator between the object and the
descending lightning. It looks as if it elongates the object, thereby increasing
the charge acting on the descending leader. The counterleader travelling
towards its tip attracts it to itself, and this eventually results in the lightning
stroke at the object. The mutual attraction of the two leaders becomes
especially pronounced when the fields they excite at the tip are comparable
with or, better, exceed the differently directed cloud field. It is only then
that the descending leader changes its path to go to the object, and the
counterleader is attracted by the descending leader rather than by the
cloud charge centre, as is usually the case. It is the excess of the perturbation
field over the cloud charge field which imparts a quasi-threshold character to
the attraction process.
This unquestionable and fairly trivial reasoning is certainly useful for
lightning protection practice. Physically, however, it remains quite meaning-
less until the mechanism of the external field effect on the leader is known.
This is equally true of the cloud field which is also involved in the attraction
of the leader, generally directing it to the earth. It is not clear at first sight
what exactly is affected by the external field, which may be very weak. The
fact is that the leader moves along the field even at Eo M lOOV/cm. Fields
of this scale cannot affect directly the leader development - we have
emphasized this several times above. The leader propagation, which occurs
via turning the air into the streamer and leader channel plasmas, requires
much stronger fields. These are present in the leader tip, in the tips of numer-
ous streamers, as well as in the streamer zone where the strength (the lowest
of the three) exceeds 10 kV,km in a negative leader and 5 kV/cm in a positive
one. High driving fields are created by the charges of the tips, streamer zones
and, partly, by the nearest portions of the channel and leader cover. They

cannot be created by the cloud or any other remote objects. The

instantaneous leader velocity is entirely independent of the low external
strength Eo but is determined by the potential difference between the
leader tip U, and the external field U. at the tip site. This great difference,
A U = I U, - Uoi x 10-100 MV, along the relatively short length of the
streamer zone creates in it the field E, x 5-10 kV/cm >> Eo necessary for
the streamer and, eventually, leader development. What is then the instanta-
neous effect of the negligible field Eo and its weak perturbations produced by
the remote counterleader on the motion of the descending leader?
Apparently, this effect is that the external field accelerates the leader. We
mentioned this at the end of section 4.1.3 and shall now discuss it at length.
The underlying mechanism is as follows. Voltage determines the leader
velocity, while the voltage gradient determines its acceleration. Velocity is
a function of the absolute potential change at the tip, V L = f ( A U ) , with
U. in the expression for A U being a function of the space coordinates or
of the tip vector radius r . A particular form of the functionf(AU) in this
case does not matter; what is important is that VL grows with AU. So, retain-
ing the generality, we can use the empirical approximation of (4.3),
4).
V L jAUl’ (y = The algebraic value of the leader acceleration is
N
dVL
dt :i (
=&- dUt f -duo
--
dt
- d r ) =*y-
dr dt
(: dUt
dt
E0VL) (5.14)
where plus refers to a negative leader and minus to a positive one. The first
term in the sum of (5.14) does not depend on the direction of the external
field. One of the reasons for the variation of U, with time was discussed in
section 4.3.2. Another reason is the increasing voltage drop across the
channel with its elongation. Normally, a variation in U, has a retarding
effect on descending leaders of both signs.
The second term in (5.14) leads to acceleration if the negative leader
moves in the direction opposite to the field vector, with the positive leader
moving along the field. The accelerating effect of the external field increases
as the field becomes higher and the angle between the field and velocity
vectors becomes smaller. Both terms have been estimated to have the same
order of magnitude (10’ m/s2); the second term may sometimes be even
larger. For this reason, the attractive action of the external field proves
essential.
We can now make clear the attraction mechanism. The actual mechan-
ism, by which a leader chooses its propagation direction, has a statistical
nature. This is indicated by numerous random path bendings and branching.
Clearly, there is a high probability that the leader moves towards a site where
it can acquire the greatest acceleration or the least retardation. It will be able
to develop a maximum velocity in this direction, bypassing other competitors
on its way. Large-scale leader photographs taken with a very short exposition
nearly always show several leader tips on short, variously oriented branches

How lightning chooses the point of stroke 24 1
Figure 5.12. A still photograph of the leader channel front with exposure of 0.3 ps.
(figure 5.12). Among these, only one tip has a real chance of survival - for a
positive leader, it is the one which belongs to the branch oriented along the
external field; for a negative leader, the respective branch must be oriented
against the field vector. The other tips usually die.
The mutual attraction of the descending leader and the counterleader,
mediated by the electric fields created by their charges, is a self-accelerating
process. This is due to a positive feedback arising between them. An
enhanced field of one leader accelerates the other leader towards the first
one. Because the distance between the leaders becomes shorter, the field of
each leader rises at the site of the other leader tip, and the mutual acceleration
proceeds at an increasing rate. This goes on until the streamer zones of the
leaders come in contact and their channels unite. As a result, the common
channel appears to be tied up to the object, from which the counterleader
started.
5.7 How lightning chooses the point of stroke

Suppose the descending lightning leader has deviated from the vertical line to
go to some high terrestrial structures. The highest structure is a lightning rod,
or several lightning rods. If the objects to be protected are much lower, the

Figure 5.13. Increasing the grounded electrode effective height by a counterleader.
lightning usually bypasses them to strike one of the rods. This can be predicted
from the equidistance principle. But when designing lightning protection
devices, one usually focuses on exceptions rather than the rules. So the question
arises of how large is the probability that the leader will miss the rod and strike
the object, having taken a longer path. It seems justifiable to apply the concepts
of a multi-electrode system to this problem. The lightning which has become
oriented towards a group of grounded ‘electrodes’ has to choose among
them. Let us make an estimation from formulas (5.10) and (5.1l), substituting
the distance from the leader tip to the earth, de,by the distance to the rod top, d,
(do is, as before, the distance to the object’s top). Lightning protection
experience shows that there is no need to make a lightning rod much higher
than typical terrestrial constructions (ha < 50m). Arranging them close to
each other, one can provide a reliable protection of the 0.99 level (of 100 light-
nings, 99 are attracted by a protection rod) if the rod height h, is only 15-20%
larger than ho. For an ‘average’ lightning, displaced at a distance equal to the
attraction radius Re, x 3ho relative to the grounded system, we have
Ad = do - d, RZ (0.12-0.15)hr at Ho = 5h, (figure 5.13(a)). The substitution
of these values into (5.10) with oa RZ 10% gives A , x 0.2. After taking the
integral of (5.7,) one gets the probability of the lightning stroke at the object
Po x 0.4 instead of the experimental value 0.01.
The complete failure of the theory was predictable. A system with a close
arrangement of grounded electrodes cannot be considered to be discon-
nected, Its counterleaders affect one another. The first leader that has started
from one of the electrodes decreases the electric field behind it, via its cover
space charge, preventing the upward development of counterleaders from the
other electrodes. Appearing with a delay, if they do, these counterleaders
cannot retard their faster competitor, because the field is enhanced in the
direction of the first leader propagation (figure 5.13(b)). This makes all of

How lighttiitig choo.scs 11w poiti t r!/'.strokc '43
Figure 5.14. The oscillogram shows how the counterleader started from the 'active'
grounded electrode of 1.1 m height screens an electric field on similar 'passive'
electrode located at a distance of 10cm. The gap length is 3 m. E the field at the
passive electrode tip. Q - the counterleader charge. I . the gap voltage.
the counterleaders interconnected: therefore. one now deals with a connected

multielectrode system.
Turn to laboratory experiments [9]. The oscillograms in figure 5.14
illustrate the field variation on the grounded. 'passive' electrode when a
counterleader develops from the nearby 'active' electrode. To simulate this pro-
cess for a sufficiently long time, a planeeplane gap 3 m long was used with two
rod electrodes on the grounded plane. A high negative voltage pulse was applied
to the other plane. A possible discharge from the passive electrode was excluded
by placing a thin dielectric screen totally covering the rod top. Before discharge
processes came into action. the passive electrode field rose in a way similar to the
voltage pulse. After a leader had started from the active electrode, the field rise
on the passive electrode became slower. and the shorter the distance between
the electrodes. the greater the rate of slowdown. At a very short inter-electrode
distance, the passive electrode field stopped rising with voltage and even
decreased somewhat. This obvious result indicates that the degree of mutual
effects of discharge processes and grounded electrodes becomes greater with
decreasing distance between them. Eventually. the role of passive electrodes
becomes negligible the grounded electrode system behaves as if it is replaced
by one active electrode which attracts nearly all descending leaders.
Owing to the feedback mechanism considered. the choice of the
stroke point made by a lightning become more definite. Even the slightest

x
.&
3
c,
’g
0.8
0.6-
s
E!
o.21.r/
a 0.4-
0.0 , , , . , ,
0.90 0.95 1.00 1.05 1.10
U”%
Figure 5.15. The breakdown voltage distributionfor the system of figure 5.5, but with a
small distance 10 cm between the grounded electrodes,which makes the system coupled.
advantages in the conditions in which a counterleader arises acquire an

additional significance, being enhanced by the weakening electric field in the
vicinity of the passive electrode, below the leader channel. It seems as if the
passive electrode entirely disappears from the system. The breakdown voltage
distribution in it nearly exactly coincides with that characteristic of a solitary
active electrode (cf. figures 5.15 and 5.5). Formally, this can be accounted for
by introducing a smaller relative standard deviation for the distribution of the
breakdown voltage difference a, in expressions (5.10) and (5.11). We shall term
it a choice standard. The upper limit of the probability integral
A, =
do - dr (5.15)
+
a,(di &)‘I2
defines, as in (5.7), the probability of choosing the stroke point on grounded
electrodes:
(5.16)
Formula (5.16) describes the probability of a lightning striking a body more

remote from its leader, and expression (5.15) is valid as long as do > dr.
Otherwise, instead of finding the probability of a lightning stroke at an
object (P,,), one should find this probability for a lightning rod (Per), then
defining P,, as 1 - PCr.t If the height of a lightning rod is h,, that of an
? A t A , >> 1 and, hence, P, << 1, one can use the approximate expression P, Y
(27r-’/*A;’ exp(-Af/2), which is valid and more convenient for estimations.

How lightning chooses the point of stroke 245
object is ho and the distance between their top projections on to the earth’s
surface is A r , we have
A, =
+
[(Ho- h ~ )( r -~ A r ) 2 ] 1 ’ 2- [ ( H ,- +r2]1/2 (5.17)
+
o,[(Ho- h0)2 ( r - Ar)’ + ( H o- h,)’ + r2I1l2 ’
Here, as before, o, is given in relative units, r is the horizontal distance from

the descending leader tip to the lightning rod axis, and Ho is the attraction
altitude. With increasing r , A , and the probability integral values become
smaller. As a consequence, the probability of a lightning stroke at the
object increases. Therefore, remote lightnings make protection measures
complicated, especially when their paths are greater deflected from a vertical
line (section 5.3).
It would be useless today to try to define the choice standard from
theoretical considerations. One should also bear in mind that the final
result of the integration of (5.16) in area for finding the number of lightning
strokes at an object strongly depends on o,,in contrast to (5.11). The quan-
tity U, can no longer be taken to be constant, since it must decrease as the
distance between the rod and object tops is made shorter. It is only the prac-
tical experience gained with various lightning protection systems which can
give some hope. The choice of objects to be observed is strictly limited.
Bulk registrations of stroke locations are made only for power transmission
lines of high and ultrahigh voltages. Sometimes, registration equipment is
mounted on unique constructions such as skyscrapers or very high television
towers [3,16]. In order to derive the values of choice standard from observa-
tions, it is necessary to calculate the expected number of lightning strokes at
the object of interest at various oc values, trying to get the best possible
agreement with the observations. As a first approximation, one may consider
the lightning attraction by a system of grounded electrodes and the choice of
the stroke point within the system to be independent events described by the
probabilities Pa and P,. Then, by analogy with (5.1 l), the expected number
of breakthroughs to a compact and an extended object (of length L ) will be
1
described as
x
Nb = 27rnl
sol Pa(r)Pc(r)rdr, Nb = 27rnl
0
dy.
Pa(y)PC(y) (5.18)
The observational data processing made in [3,17] revealed the dependence

of the choice standard U, on the distance between the object and the
lightning-rod tops D. For a relative choice standard, the following formula
is recommended:
a, = 7 x + 8 x ~ o - ~ D . D [m]. (5.19)
Its use in the calculations of (5.16)-(5.18) provides reasonable agreement
with observations of 0.9-0.999% reliability rods. There are no data on
rods with a higher reliability.

Note the following important circumstance concerning the protective

action of a lightning rod. Even common sense indicates that the rod height
must be increased with increasing distance between the object and the rod.
Let us discuss the opposite situation when a rod is mounted directly on an
object of small area. How large must be the excess Ah = h, - ho to provide
a given protection reliability? Essentially, we deal here with the frequency
of lightning strokes below the lightning rod top. This question is justified
by observations of such a high construction as the Ostankino Television
Tower in Moscow (540m). During the 18 years of observations, descending
lightnings have struck it at various distances below the top, down to 200 m
(figure 1.10). The rod has been unable to protect itself. This sounds ridicu-
lous, but this is the reality.
The results of a numerical integration of the first expression in (5.18),
using (5.10), (5.11), (5.16), and (5.17) at Ar = 0, are presented in figure
5.16. The stroke probability @b = Nb/Nl shows the fraction of lightnings,
attracted by the whole system of grounded electrodes, N I , which have
missed the lightning rod to strike the object. The calculations were made
with the attraction standard ra= 0.1 for objects of height ho = 30- 150 m.
The actual protective effect is achieved only if the height of the lightning
rod considerably exceeds that of the object. For short constructions with,
say, ho = 30m, a 99% protection reliability ( a b = lo-*) requires the light-
ning rod height excess of Ah x 0.2ho, which is quite feasible technically
because it is equal only to 6 m above the object. An object 150m high will
require a lightning rod 50m higher than the object (Ah x 0.3ho), which
Figure 5.16. The evaluated probability of a lightning breakthrough to an object of

height ho, protected by an adjacent lightning rod of height h, > ho.

W h y are several lightning rods more effective than one? 247
will be more expensive and complicated. In technical applications, the ten-

dency of the Qb(Ah)curves to saturation is very important. This tendency
becomes greater with increasing construction height, which means that a
single rod will be ineffective for a high protection reliability. It is hard to
protect an object with ho > lOOm with a reliability above 0.999%
(ab= lop3), an object with ho > 150m above 0.99%, etc. The higher the
construction, the more complicated is the problem, and this is the reason
why the Ostankino Tower is unable to protect itself. Nine lightning strokes
were registered photographically along its length of 200 m from the top [18].
The protection efficiency decreases as the distance between the top of a
high lightning rod and that of an object of similar height increases, reducing
the mutual effect of counterleaders. Formally, this manifests itself as a larger
choice standard U,, in accordance with (5.19). Sooner or later, its effect begins
to dominate over that of lengths in formula (5.17), so the upper limit of the
probability integral A , stops rising.
5.8 Why are several lightning rods more effective than one?
The answer to this question can be found geometrically. Let us consider two
lightnings which travel in the same vertical plane going through an object
and its lightning rod in opposite directions. Suppose both leader tips are at
an attraction altitude Ho at the same distance from the rod. They have, there-
fore, an equal chance to be attracted by the object-rod system. The only
difference is that one leader will approach it on the lightning rod side
(version 1) and the other on the side of the object to be protected (version 2).
Assume, for definiteness, that the displacement of the lightnings relative to the
rod axis is equal to the attraction radius Re, = 3h, (i.e., an average displace-
ment), Ho = 5h,, and the horizontal distance between the rod and the object
is AY= h, - ho << h,. From (5.17), the upper limit of the probability integral
for version 2 is nearly seven times less than for version 1:
7Ar Ar
4 1 = u,25fih, > 4 2 = u,25&hr ’
Consequently, the Probability integral from (5.16) for version 2 also

decreases, increasing sharply the probability of striking the object. To
illustrate, for A r = 0.2hr and uc = 0.01, the parameter A , takes the values
of 4 and 0.57, respectively. When the lightning approaches on the lightning
rod side, the probability of striking the object is, according to (5.16),
nearly zero, but on the object side it is 0.28. Therefore, a single lightning
rod can protect an object reliably only from the ‘back’, while its protection
efficiency from the ‘front’ is much lower. This situation can be rectified if
the object to be protected is placed half-way between two rods; it is still
better if there are three rods and so on - this becomes only a matter of

cost. No rod palisades are known from the protection practice; nevertheless,
it is tempting to surround the object of interest with a protecting wire,
especially if it is not very high but occupies a large area.
As an illustration, let us consider a case simple for the calculations. This
will allow us to get numerical results and demonstrate the calculation
procedures. Suppose a circle of radius Ro = lOOm is densely filled by
constructions of height ho = 10m. All of them must be protected with a
0.99% reliability, i.e., the probability of a lightning stroke should not exceed
@bmu = Let us now place a circular grounded wire at a distance of
10m from the external perimeter of the premises. This distance is necessary
for technical considerations. For example, we must prevent a sparkover
between the grounded wire and the communications systems and other struc-
tures, whether it occurs across the earth’s surface or through the air due to high
current pulses of the lightning discharge. Therefore, the circular grounded wire
will have a radius R, = 110 m. Let us find the wire height h,, whch will provide
the necessary value of Bb,,,. For the radial symmetry, the probability of the
lightning breakthrough is found from formulae (5.11) and (5.18) as
(5.20)
The probabilities of attraction P,(r) and point choice P,(r) for a lightning,
whose tip (in the horizontal plane at the attraction altitude Ho) is at the
instantaneous distance r from the area being protected, are defined by similar
expressions (5.7) and (5.16). These differ only in the values of the upper limit
of the probability integral. For the attraction probability, the limit A , ,
according to (5.10), is described by the difference between the minimal
distances from the leader tip at the attraction altitude Ho to the system of
grounded electrodes and to the earth, Ad, = d, - de. In the case being
considered, A , is defined by the smaller of the values (at Y < Ro):
Ad,, = [(R,- r ) 2 + (Ho - hr)2]1’2 - Ho, Ad,, = ho.
At r > Ro,we have Ad, = Ad,, . In the calculation of the choice probability,
the upper integral limit is given by formula (5.15). When calculating the
difference between the minimal distances to the object and the protector,
A d = do - d,, one has to keep in mind that we have domi,= H - ho at
+
r < R and do,,, = [ ( r - R0)2 ( H o- hO)2]1’2at r > Ro.
The calculation procedure reduces to finding, for every value of r , the
upper limits A , and Act in the integrals of (5.7) and (5.16) to calculate
(extract from tables) these integrals, which give P,(Y)and P , ( r ) , and to
calculate the integrals of (5.20). Practically, it is sufficient to make the
f The value of the choice standard oc necessary for the calculation of A , is found from formula
(5.19). taking into account the distance D between the protector top and the point on the object’s
surface nearest to the lightning with the instantaneous coordinate r; ua 0.1.

Some technical parameters of lightning protection 249
I .1
10” 1O‘* 10’‘

Lightning breakthrough probability 0
Figure 5.17. The object of 10 m height and of 100 m radius is protected by a bounding
circular wire. In the graph is presented the evaluated wire height h necessary to
decrease the probability of a lightning breakthrough to the object up to the value
of CJshown on the abscissa axis.
calculations with the step Ar M (0.1 - 0.2)hr and finish them when P,(r)
drops to 10-6-10-7 with growing r . If the probability integral is given reason-
ably (by an empirical formula or by borrowing it from a table, e.g., using a
spline), the volume of calculations proves so small that they can be made with
a programmed calculator. With a modern computer, the time necessary for
numerical computations is only that for the data input.
The calculations made for the above example are shown in figure 5.17.
The probability of a lightning breakthrough to the object decreases to the
given value of lo-* when the protective wire is suspended at a reasonable
height h, 34m. Note, for comparison, that a single lightning rod placed
at the centre of a similar area provides the same protection reliability only
if its height is h, > 150m. Even if one builds such a rod, the result may
prove disappointing. Quite often, it is impossible to provide a safe delivery
to the earth of a high lightning current impulse, when conductors with
current pass close to structures being protected. Electromagnetic induction,
sparking capable of setting a fire, etc. may also be dangerous.
5.9 Some technical parameters of lightning protection

5.9.1 The protection zone
It follows from the foregoing that a lightning-rod has a better chance of
intercepting descending lightnings if it has a greater height above the

object and is closer to it. Practically, it is important to identify a certain area

around a protector, which would be reliably protected. This is the protection
zone. Any object located within this zone must be considered to be protected
with a reliability equal to or higher than that used for the calculation of the
zone boundary. There is no doubt that this idea is technically constructive.
When the configuration of the protection zone is known, the determination
of the grounded rod or wire height reduces to a simple calculation or geo-
metrical construction - this was an important factor in the recent age of
‘manual’ protection designing. At that time, the general tendency was to
simplify the zone configuration as much as possible. In Russia, for instance,
a single lightning rod zone was usually a circular cone, whose vertex
coincided with the rod top [lo]. When lightning protection engineers realized
that the height of the rod was to exceed that of the object to be protected
(section 5.7), the cone vertex was placed on the rod axis under its top [19].
The greater the protection reliability required, the more pointed and lower
was the zone cone. For a grounded wire, the protection zone had a double
pitch symmetry; when intersected transversally by a plane, it produced an
isosceles triangle with nearly the same dimensions as those of a vertical
cross section made through the rod cone half. Lightning protection manuals
give a set of empirical formulas to design protection zones for simple types of
lightning protector [2,4].
The long-term practice has somewhat screened the principal ambiguity
of the notion of protection zone. Indeed, having only one parameter - the
admissible probability of a lightning stroke abmaX - one is unable to determine
exactly the zone boundary. So one has to resort to some additional consid-
erations of one’s own choice. In particular, there is nothing behind the
concept of a conic zone except for the consideration of an axial symmetry
and the desire to make the geometry simple. The value of BbmaX corresponds
to a wide range of zone configurations, so the chosen configuration may
appear to be far short of optimum. A protection zone is rarely filled up.
When an object occupies a small fraction of this area, which is frequently
the case in practice, the lightning rod height proves excessive. For high
objects and still higher lightning rods, this results in unjustifiably large
costs, which increase when high reliability is required. When the engineer
places an object within a protection zone, he has no idea about its actual
protection. But by decreasing the distance from the zone boundary inward,
the probability of a lightning stroke may decrease by several orders of
magnitude. To specify its value, one has to make numerical calculations
similar to those illustrated in section 5.8.
Finally, the most important thing is that protection zones can be built
with sufficient validity only for two types of lightning-rods - rods and
wires. Even an attempt to combine them causes much difficulty. The same
is true of multirod protectors, non-parallel two-wire protectors, and sets of
rods of different height. All of them find application, especially when natural

Some technical parameters of lightning protection 25 1
‘protectors’ are used, such as neighbouring well-grounded metallic structures

or high trees. The analysis of protection practice shows that preference is
often given to easily-calculated designs rather than to effective designs. How-
ever, the statistical techniques used for the calculation have no limitations on
the protector type, their number, or the geometry of the objects to be pro-
tected. Some problems may arise only in finding the shortest distance from
the lightning leader tip to the lightning-rod and to the object. But they are
surmountable with the use of modern computers. One should also bear in
mind that the calculation provides the engineer not only with the break-
through probability but with the number of expected breakthroughs over
the time a particular object is in use. The latter parameter is more definite
and cost-significant.
5.9.2 The protection angle of a grounded wire

The concept of protection angle cy is used in designing wire protectors for
power transmission lines (figure 5.18). The protection angle is considered
positive when the power wires are suspended farther from the axis than the
grounded wires, so they are open, to some extent, to descending lightnings.
The value of Icy1 decreases with the grounded wire suspension height and
with decreasing horizontal displacement of the power wire relative to the
grounded one. The protection reliability is lower when the positive angle is
larger. The angle was introduced as a parameter necessary for the generaliza-
tion of observations of lightning strokes at transmission lines of various
designs. It turned out that the angle cy could not serve as an unambiguous
characteristic of the protective quality of a grounded wire. A transmission
line must also be described in terms of the grounded wire height above
power line wires, Ah, and of the grounded wire height above the earth, h,.
Figure 5.18. Positive and negative protection angles. A : grounding wire; B power
wire; C: insulator string.

This determines the distance between the grounded wire and the power wire
at fixed a, which defines, through the choice standard a,, the degree of the
system connectivity. Of lines with an identical protection angle, the best pro-
tected line is the one with the largest value of Ah and the lowest value of h,.
Empirical formulas, which relate the lightning breakthrough probability
to wires with a and h,, have found wide practical application. Their accuracy,
however, is not very high because they do not include Ah. For example, there
are expressions identical in composition [21,22]:
ah;'=
lg@b =--- 4, lg@b =-- 3.95, h, [m], a [degree]. (5.21)
90 75
They give the probability of a lightning stroke with a 300% error related to a
value supported by practical observations. These formulae should be treated
with caution when the line supports are higher than 50m at small positive
and, especially, at negative protection angles. This is because most main-
tenance data refer to lines of up to 40m high with positive protection
angles of 20-30". Besides, very few of the data used for deriving empirical
formulas represent direct measurements. Usually, the data are derived
from registrations of storm cut-offs minus the calculated return sparkovers
(section 1.6.1). The latter calculations often give a large error. Still, expres-
sions (5.21) demonstrate that negative protection angles are quite attractive.
The action of protection wires placed farther from the tower axis than line
wires (cy < 0) is similar to that of a closed grounded wire surrounding a
region being protected (section 5.8). This type of protector could provide
an exceptionally low probability of a lightning stoke at line wires, but the
implementation of negative angle protection requires larger towers and,
hence, a higher cost. This approach is, for this reason, unpopular.
5.1 0 Protection efficiency versus the object function
No doubt, there is a close relationship between the protector efficiency and a

particular function (purpose) of a protected construction, especially when it
is under high potential relative to the earth (e.g., ultrahigh voltage wires) or
ejects a highly heated gas into the atmosphere. By raising the object potential
to values comparable with the absolute potential of a descending leader, one
can either increase the field at its tip, making the leader move towards the
object, or lower it, suppressing the lightning attraction. It is a matter of
the quantitative effect produced rather than its principal feasibility. High
object potential U,, may affect both the process of lightning attraction
and the choice of the stroke point. The latter is more sensitive to external
effects owing to the positive feedback in the connected system. Expressions
(5.10) and (5.15) are suitable for estimations. They include the standards
of attraction, ca,and of choice, ac. It is easier to control the process when

Protection efjciency versus the object function 253
their values are lower. For objects of regular height (-30m), we have
ua lo-' and uc x This enables us to focus on the choice only. The
effects of the high potential action of the object, U&, will noticeable at
U& u,U,, where U , is the tip potential at the moment the leader has
descended to the attraction altitude. An 'average' lightning has
U , = 50MV. Therefore, in order to get an effect on the process of choice,
one must apply Uob M 500 kV to the object (or to the protecting wire). In
order to affect the attraction process, the applied voltage must be -5 MV.
The latter value is, certainly, not feasible for the present power industry,
but available operation voltage of an ultrahigh voltage (UHV) line is high
enough to affect the lightning preference to a protecting wire or to a line
wire [22,23].
Most UHV lines operate at alternative voltage of frequency f = 50 Hz
(60Hz in the USA). Over the time H o / V Lx s along the flight path
Ho, during which the lightning chooses a point to strike, the wire potential
changes but little, and its values U & ( [ )= Ufmax sinwt (U = 2759 can be
taken to be equally probable. By the initial moment of attraction, u o b ( t ) ,
may have the same or opposite sign relative to the lightning. If the sign is
the same, the development of a counterleader from the wire will be delayed,
so the probability of the lightning striking the wire will be reduced. In the
other situation, the effect will be opposite. To get a total result over a
long-term observation of the line operation (or a short-term observation of
a very long line), one should average the operating voltage effects over an
oscillation period. For this, expression (5.15) for the parameter A , must be
extended to the case in question. Expression (5.15) was based on the differ-
ence in the average fields along the lengths from the leader tip at the attrac-
tion altitude to the protector and to the object. Now, this difference can be
calculated with the potential U& to get, instead of (5.15),
where U, is the descending leader tip potential at altitude Ho.

The qualitative result of the calculations to be given below is predictable.
We are interested in the effect of alternative voltage on the preferential choice
of the stroke point between a protection wire and a power wire, since the
operating line voltage is too low to affect appreciably the lightning attraction.
In the half period when Uob(t) and U, have the same sign (suppose it is a
negative descending leader and negative voltage), the lightning is 'repelled'
by the power wire; in the positive half period, it is attracted by it. Owing
to the protecting wire, the probability of a lightning stroke at the wire in
the off-voltage mode is low, 10-2-10-3. Therefore, the favourable effect of
all negative half periods is small. Even if no lightning strikes the power wire
during this time, the number of strokes at it will, for a long time, be reduced
only by half relative to the no-load mode, because negative half periods take

Figure 5.19. Effect of the AC transmission line operation voltage on the lightning
breakthrough probability. The lower and upper curves correspond to probabilities
of and lo-', respectively, without voltage.
only half of the on-voltage time. The unfavourable effect of positive half
periods may be much stronger. In principle, the potential difference U,, > 0
and U, < 0 at a high alternative voltage amplitude may produce such a
strong 'attracting' field that all lightnings going to the power line will strike
its wires. The probability of a strike at the line wire during the positive half
periods may rise by 2-3 orders of magnitude (even as much as unity) against
its two-fold reduction during the negative half periods. As a result, the
stroke probability averaged over a long time for the line wire grows. The
operating voltage effect on power lines reduces the reliability of lightning
protection.
The numerical calculations of this effect are illustrated in figure 5.19.
The probability of lightning breakthrough to AC lines increases by an
order relative to the probability x lop2 for the off-voltage mode at
y = Uobmax/ccUtx 3.75; at ab x lop3, this effect is produced at 1.5 times
lower voltage. For the typical size of modern power line towers with
oc x 0.008, the stroke probability for the power wire at U , x 30MV rises
from lop3 to lop2 at phase voltage amplitude Uobmaxx 625kV. Such are
the line voltages (750 kV) in some countries. Only the next generation of
power lines with 1150 kV can be expected to produce as strong effect on light-
ning at U , x 50MV. An experimental line of this kind has been in use in
Russia for a short time.
Direct current line has a more pronounced effect on lightning. Lightning
separation is possible in D C lines: a positive line wire more strongly attracts

Lightning attraction by aircraft 255
negative lightnings and a negative wire more strongly attracts positive ones.
Since the frequency of positive descending lightnings is an order of magni-
tude smaller, a positive UHV DC line wire will attract a larger fraction of
strokes. This effect may become well pronounced at the wire potential of
f 5 0 0 kV and higher.
The treatment of a hot air flow from an object to be protected is
generally similar to the above analysis. The density and electrical strength
of hot air are lower, and the strength is proportional to the density in the
first approximation [25-261. Formally, this is equivalent to the reduction
of distance do from the lightning to the object in expression (5.15), as if the
object height were increased. As a result, the lightning protector has a
lower efficiency. Consider, as an illustration, a chimney lOOm high with a
10m lightning rod fixed on its top. With the practically zero horizontal
distance between the rod and the object and in the absence of hot smoke
gases, the rod will intercept about 90% of all lightnings attracted by the
chimney (figure 5.16). But if the chimney ejects a hot gas flow with the
temperature of 100°C along the length of 30m, the probability that
the lightning will miss the rod to strike the chimney will rise from 10 to
50%. Actually, the lightning rod becomes ineffective.The question is whether
it is worth constructing this purely decorative device on the chimney top.
5.1 1 Lightning attraction by aircraft
Protection of aircraft and spacecraft has always been a complex and demand-
ing problem - poor protection may have serious repercussions. It has been
mentioned that an aircraft can be damaged by an ascending lightning starting
from its surface or by an attracted descending discharge in the atmosphere, as
happens with a terrestrial construction. Naturally, the concept of attraction
refers only to descending lightnings. There are no observational data on the
interaction between aircraft and descending lightnings, and one has to resort
again to laboratory experiments. Figure 5.20 gives a set of static photographs
taken from the screen of an electron optical converter. Of many pictures, we
have selected the most typical ones. The electronic shutter was shut at differ-
ent moments of time, so the result is not exactly a movie film but something
close to it. One can see that a vertical rod insulated from the earth has
attracted one of the leader branches together with its streamer zone,
having first excited a streamer flash and then a counterleader. Its contact
with the descending leader has produced a short luminosity enhancement
of their, now common, channel, like a step of the negative leader with its
miniature return stroke (sections 2.7 and 4.6). As a result, the channel and
the rod have become the extension of a high voltage electrode. The leader
has started off towards the earth from the lower end of the rod which now
seems to be part of the leader channel.

Figure 5.20. Attraction of the spark leader by the isolated metal rod suspended in the
gap middle.
It appears that the attraction of a descending lightning by an insulated

conductor, as well as by a grounded one, is stimulated by the excitation of a
counterleader. The similarity in their mechanisms accounts for the similarity
in the basic parameters of attraction. Below, we present some laboratory
measurements of equivalent radii Re, for spark attraction by a vertical metal-
lic rod of length 1 = 0.5m, suspended at height H above a grounded plane.
The spark was produced by a positive voltage pulse with a loops front in
a rod-plane gap of 3 m. The front provides a more or less reliable field rise
time for a real object during the development of a descending lightning
leader. The measured values of Req(H) are normalized to the value of
Re,(0) for a rod that has descended to a plane to become grounded:
HI1 0 12.8 3.4
R,,(H)/R,,(O) 1.0 0.9 0.9 0.8
The response to the conductor rise above the earth is fairly weak. A 10-20%
decrease in Re, seems to be regular, although it lies within the experimental
error range. To extrapolate this result to lightning, one should assume that
the number of descending lightning strokes for aircraft with the maximum

Lightning attraction by aircraft 257
size I is not larger than that for a grounded object of the height h = 1. This
limit for the number of strokes does not follow only from the experimental
fact of a certain decrease in Re, with H . Of greater importance are the
possible variations in the aircraft position relative to the external field
vector, Eo, during the flight. The field enhancement at the ends of its fuselage
of length I is defined by field projection on to the aircraft axis, rather than by
the value of EO.High terrestrial constructions are always aligned with the
field since it is vertical at the earth.
Let us now estimate the possible number of descending lightning strokes
at an aircraft of length I = 70m, using the concept of attraction radius
Re, M 31. We shall have Nd x n17rR&kh, where n1 is an average annual
frequency of lightning strokes at the earth and kh is the ratio of the total
flight hours per year to the total number of hours in a year. For kh = f
and nl x 3 kmP2 per year, we get Nd M 0.1 per year. This is at least an
order of magnitude less than what follows from official statistics. One
should not think that the discrepancy is due to the neglect of intercloud
discharges, whose number is 2-3 times larger than that of lightnings striking
the earth. In order to be attacked by intercloud lightnings, aircraft must
penetrate through the storm front, but this is absolutely forbidden and
may happen only as an accident. Rather, the result was overestimated
because any pilot tries to stay as far away from a storm as possible.
Therefore, descending lightnings are responsible for fewer than 10% of
strokes at aircraft. The other 90% or more are due to ascending lightnings
excited by aircraft and spacecraft themselves (section 4.2). However, the
interest in descending lightnings remains active because of the poor predict-
ability of the stroke points on the aircraft surface. A similar situation but for
high terrestrial constructions was discussed in section 5.7. The probability of
a lightning striking much below the top is rather high. This situation can be
readily simulated in the laboratory for a long positive spark excited by a
voltage pulse with a smooth front, tf M loops and higher. The photograph
in figure 5.21 illustrates a spark stroke almost at the rod centre, together
with the integral distribution of the stroke points along its length. The
wide, if not random, spread of stroke points over the aircraft surface creates
additional problems. The aircraft has many vulnerable areas. In addition to
the cockpit and fuel tanks, these are hundreds of antennas and external
detectors providing a safe flight. It would be desirable to hide them from
descending lightnings but the chances for this are quite limited. One con-
solation is that most lightnings affecting aircraft are of the ascending type
starting mostly from the ends of the fuselage and wings, where the external
electric field is greatly enhanced.
The excitation of ascending lightnings by aircraft was considered in
section 4.2. Formula (4.11) allows estimation of the hazardous field Eo for
an aircraft of length I = 2d. The field Eo decreases with growing d, some
slower than d-315.Note that the parameter 2d is not necessarily the fuselage

Figure 5.21. The stroke probability at various points of an isolated rod for two
voltage front durations. The photograph shows how the spark has struck at the
rod centre.
length; this may be the wing length if it is larger. In general, the experience
indicates a direct relationship between the aircraft size and the frequency
of lightning strokes. There are exceptions, of course. The statistics of flight
accidents shows that aircraft of identical size may differ considerably in the
capacity to excite lightnings. In one design, the engines are mounted on
the wing pylons, and the ejected hot gas jet passes near the metallic fuselage,
where the low fields cannot excite a leader. In another design characteristic of
rockets also, the engine nozzle is placed in the tail, so that the hot jet serves as
the fuselage extension. This is a perfect site for a counterleader to be excited
since the leader development needs a lower field in a low density gas.
In the estimation, we shall assume the jet length to be half the fuselage
length, lj = d , and its average temperature to be twice as high as the ambient
air temperature. Suppose that the jet radius is large enough for the streamer
zone to be entirely within it and that the leader develops in a gas of relative
density S = 0.5. When the gas density becomes lower due to the heating, the
field providing the streamer propagation decreases at a rate 6 [25,26].The
rate of the electric strength decrease in long gaps is approximately the same
for mountainous regions, although the density variation range in these
experiments was narrower, 6 M 0.7 [25]. We shall assume from these data
that a leader developing within a hot jet requires a potential drop 5-' times
smaller than that given by formula (2.49), i.e., AU = 36A3/5(3hd/2)2/5
(here, the leader length L has been replaced by the jet length d ) . The total
length of a conductor consisting of a fuselage of 2d long and a leader of

Are attraction processes controllable? 259
length d is equal to 3d. Hence, we have AU = 3Eod/2, and the estimated

external field providing the leader formation in the jet is
The field is found to be Eo(6)x 165V/cm at the values of A = ( 2 7 ~ ~ o a ) - ~ ' ~ ,

a = 1.5 x 103V-'/* cm/s, b = 300VA/cm, used in chapters 2 and 4, and
d = 35m. At this field, there will be a breakdown of the jet, increasing the
aircraft size by the jet length value. This will create favourable conditions
for an ascending lightning to develop from the fuselage in a low field. To
+
estimate the field, d should be replaced by Le, = (2d lj)/2 in formula
(4.11); for the present example, it should be 1.5d. The 25% decrease of the
threshold field which will follow may greatly change the total number of
lightning strokes at the aircraft.
5.12 Are attraction processes controllable?
We gave an affirmative answer to this question, when discussing the effects of

operating voltage in ultrahigh voltage lines and hot gas flows. The further
consideration of this problem should be concerned with quantitative aspects
and particular methods of lightning control. Lightning control has two aims:
to raise the reliability of lightning protection of nearby objects and to expand
the area being protected by using conventional techniques. These may only
seem to be two sides of the same effect. For example, increasing the lightning
rod height increases both the protection reliability for a particular object and
the maximum radius of the protected area. This, in principle, is the case, but
quantitatively the two results differ considerably.
Turn to the estimations above. It follows from the calculations in figure
5.15 that the increase in the lightning rod height h, by only 4 m (from 36 to
40 m) reduces the probability of a stroke at an adjacent 30 m object from lo-*
to or by an order of magnitude. The effect is significant. As for the
expansion of the protected area on the earth, its radius Ar does not grow
faster than h,. This can be demonstrated by putting ho = 0, r = 0, and
HO= 5h, in formula (5.17). Then we shall have A r h, for a given proba-
N
bility of choosing a stroke point, i.e., at fixed A , and D~ = const. In actual

reality, the standard oC grows with h,, due to which A r rises still slower. In
our example, A r increases by less than lo%, and this insignificant effect is
of no interest to us.
Lightning control eventually reduces to a change either in the electrical
strength of the discharge gaps between the descending leader tip and the
protector and the earth or in the gap voltage. For this reason, the particular
conclusion that follows from the above example can be extended to any
control measures - their effectiveness falls with distance between the object

top and the lightning-rod, since the mutual effects of the components in a
multi-electrode system become weaker. Formally, this weaker effect mani-
fests itself in increasing standard cc.It appears that the lightning control is
easier for objects of low height and area, when conventional protectors are
sufficiently effective. It is much more difficult to deviate a lightning from
an object without mounting a metallic rod on top of it. The application of
destructive technologies to storm clouds and their charge neutralization
are not discussed in this book, because this is a special problem having no
direct relation to lightning processes.
The physics of the effect of a voltage pulse rise on a descending lightning
leader is clearer than that of other effects. The effect can be expected to be
favourable when the potential applied to the lightning rod is of opposite
sign to that of the lightning, or the potential applied to the object is of the
same sign. In the former case, the conditions for a counterleader to start
from the rod are quite favourable. To initiate a preventive start of a counter-
leader from a lightning rod is to deviate the stroke point from the object. But
in order to produce a noticeable effect, the counterleader must have a channel
length comparable with the length difference between the object and the rod,
or between their tops (the latter quantities are comparable). Only then does
the effective rod height really grow and the charge space of the counterleader
considerably limits the field at the object top. Therefore, one deals with
channels of metre lengths, sometimes of tens or even hundreds of metres,
especially if one takes into account the multi-fold increase in the radius of
the area to be protected. This is a fairly complicated task.
A short-term ‘elongation’ of the rod by exciting a plasma channel from its
top is very similar to the counterleader behaviour. A laser spark or a short-
term long plasma jet would be sufficient for ths. Laboratory studies have
shown that a man-made plasma conductor affects a long spark path as a
metallic conductor. The problem is the technological complexity and consider-
able cost of the project rather than the principal possibility of control.
Imagine an ideal pulse generator, whose effectiveness is so high that it
blocks a lightning breakthrough to the object with 100% probability. The
protection reliability will then be determined by the reliability of the genera-
tor itself, primarily by its synchronizing unit. It is a difficult task to design a
reliable synchronizing unit capable of responding to a nearby descending
lightning leader. A leader always chooses a complicated, poorly predictable
path and has many branches. It is necessary either to distinguish a branch
from the main channel or to trigger the control unit repeatedly. The latter
is undesirable not only because this is resource-consuming. A control pulse
can stimulate a branch to become the main channel, which is the first to
reach the grounded electrode, producing a powerful return stroke pulse.
The close vicinity of a strong current may be as hazardous to the object
being protected as a direct stroke. Finally, we should not discard multi-
component lightnings - 50% of subsequent components do not follow the

Are attraction processes controllable? 26 1
channel of the first component [27].So it is necessary to design a control unit

capable of generating a series of pulses with millisecond intervals. Such a pro-
ject would be very costly.
High costs have been the main reason for the decreasing interest in
lightning control among specialists. They think of using nonmetallic rods
or other unconventional measures only in exceptional situations when the
common approaches are incompatible with the technological functions of
the object being protected. Designers have suggested some exotic ways of
lightning protection. Specialized firms advertise lightning rods with radio-
active, piezoelectric and other wonder tops. The performance of radioactive
sources has been tested in a laboratory, and no noticeable effect has been
registered even on the leader start, let alone its propagation along the dis-
charge gap. This should have been expected, because a leader arises from a
pulse corona flash resulting from a long application of an electric field (as
happens during a storm). Every pulse flash represents a streamer branch
with a channel electron density of 10'2-10'4cm-3 [28].A radioactive top
can hardly add anything to this density, unless its power is so high that it
kills everything alive around it.
It appears that leader suppression may be more promising than its
excitation. Laboratory experiments have long been known [29], in which
an ultracorona was successfully used to suppress the leader start. The
corona arises as a thin uniform cover on the anode or the cathode made of
a thin wire (-0.1-1 mm). A slow voltage rise does not change the corona
structure or the ionization region thickness. The electric field strength on
the electrode is stabilized by the space charge of ions drifting slowly on the
corona periphery. The field stabilization prevents the formation of an ioni-
zation wave, or a streamer flash, which would otherwise produce a leader.
With no consequences, the average field in a gap of several dozen centimetres
long could be raised to 20-22 kV/cm, whereas 5 kV/cm was commonly
sufficient to produce a breakdown in the absence of an ultracorona.
It would be tempting to extend the laboratory effect to lightning protection
practice to suppress the counterleader start from the object being protected. An
obstacle here is the rate of external field variation at the top of a grounded
electrode of height h, rather than the much greater gap length. The electrode
possesses zero potential, U = 0. The potential of the external field EO at the
top is U, = Eoh,so that the air at the electrode top is affected by the potential
difference equal to U - U, = - U,. The linear charge of a leader descending
directly on to the object creates field AEo at the earth, given by formula
(3.5). As the leader approaches the earth, potential U, rises at the rate
(5.23)
where z is the altitude of the descending leader tip and VL = -dz/dt is its
velocity.

262 Lightning attraction b y objects
Let us find the maximum rate of the field rise at which the ultracorona
can still survive. Assume, for simplicity, that a corona (positive, for definite-
ness) arises at a sphere of radius yo, attached to the electrode top. TO
prevent the corona transformation to an ionization wave capable of
initiating a streamer flash and then a leader, the field on the sphere
should not rise in time with AEo. The surface with maximum field should
not detach from the sphere to move into the gap interior. In an ultracorona,
the field on the sphere is stabilized by space charge on the level of E,
depending on radius yo. The sphere concentrates a constant charge
Q, = 4neor&. A short time A t after the corona ignition, the voltage
increases by the value A U = A,At, which is supposed to increase the posi-
tive sphere charge by AQ1 = CAU = 4mOrOAuAt. To avoid this, the sphere
charge AQ1 must be compensated. The compensation occurs owing to the
gas ionization in the thin surface layer. Positive ions transport the charge
AQ for the distance Ar = plE,At (where p, is the ion mobility), so that
+
the negative charge induced in the sphere AQ, = - A Q r o / ( r o Ar) is able
to neutralize AQ1. The charge actually induced in the sphere is transported
into it by electrons produced in the near-surface layer, whose number is
excessively large since lAQll = AQ, < AQ. ‘Excessive’ electrons leave for
the external circuit and then to the ‘opposite’ electrode - the earth. The
+
field on the radius r = yo Ar now becomes equal to E ( r ) = (Q, AQ)/ +
+
[4mO(r0 AY)^] and should not exceed E,. To the small value of about
A r / r o , this requirement is met at A, d 2p,E; M 3.6 k V / p (E, M 30 kV/cm,
p1 x 2 cm2/VSI.
We have analysed the other extrema1 situation when the corona exists so
long that charge Q >> Q, is incorporated into space and the ion cloud radius
becomes r1 >> YO. A well-developed corona can exist at the sphere for a long
time if the voltage U. does not grow in time faster than U, = Aut. The
maximum admissible growth rate A, coincides, in order of magnitude,
with the above estimate but is slightly lower. At a fast voltage growth,
say, U M t” with n > 1, there necessarily comes the moment when the
ion cloud field becomes higher than E,, stimulating the transition to a
streamer flash. For the typical values of h = 50m, rLx 5 x Cjm, and
VL x 3 x lo5m/s, the voltage growth rate reaches the estimated critical
value when the leader descends to the altitude z x 200m, at which the
attraction process begins. A little later, A , N z-* becomes even more critical,
and the ultracorona dies giving way to a counterleader.
To conclude, lightning can be controlled but this task is costly and very
complicated technologically. So it would be unreasonable to discard
conventional protection technologies where they can solve the problem suc-
cessfully. One should not expect miracles in lightning protection. If particular
circumstances make one turn to unconventional measures, one must be ready
to create complex devices, whose protection reliability will be determined by
their operation, rather than by the interaction with a lightning.

I f the lightning misses the object 263
5.13 If the lightning misses the object
This is likely to happen more often than direct strokes at an object. Some-
times, the object attracts a lightning branch which could hit the object if it
had enough time before the return stroke develops from the main channel.
Such a situation is illustrated in figure 5.22. The counterleader, which has
started from the television tower top, has no time to transform to an ascend-
ing lightning or intercept the descending leader, because the latter has struck
a metallic tower below the tower top. As a result, the counterleader remains
uncompleted. The counterleader channel has, however, become several
dozens of metres longer. This is now a mature channel, whose temperature
is at least 5000-6000K. If it had touched a hot gas jet, it would inevitably
ignite the gas. Practically a leader of any length is suitable for ignition of
inflammable exhausts into the atmosphere. To excite and develop a leader
in air under normal conditions, a voltage of 300-400 kV would be sufficient.
Such a potential difference AU = Eoh can be produced in objects of height
h > 30m even in the absence of lightning because this would require a
storm cloud field of Eo M lOOV/cm. If the object is lower, uncompleted
counter-leaders can be excited even by remote lightnings. From formula
(3.7), a descending leader that has started at an altitude of H = 3 km and
has touched the earth creates a field Eo = lOOV/cm at a distance R = 1 km
from the stroke point if it carries the linear charge T~ M 8 x lop4Cjm. This
charge is characteristic of a descending leader with average parameters.
This is one of the long-range mechanisms of lightning, which should be
Figure 5.22. The long incomplete counterleader (2) started from the top of the
Ostankino Tower while the descending lightning struck lower than the top (1).

taken into account when treating possible emergencies for objects containing
large amounts of inflammable fuels.
References
[l] Uman M A 1987 The Lightning Discharge (New York: Academic Press) p 377
[2] Operating Instruction for Lightning Protection of Buildings and Works RD
31.21.122-87 1989 (Moscow: Energoatomizdat) p 56 (in Russian)
[3] Bazelyan E M, Gorin B N and Levitov V I 1978 Physical and Engineering Funda-
141 Golde R H 1967 J . Franklin Inst. 286 6 451
[5] Linck H and Sargent M 1974 CIGRE, Sec. N 33/09 (Paris) 11
[6] Wagner C F 1963 AZZZ Trans. 83 (Pt 3) 606
[7] Wagner C F 1967 J . Franklin Inst. 283 (Pt 3) 558
[8] Darveniza M. Popolansky F and Whitehead E R 1975 Electra 41 39
[9] Bazelyan E M, Levitov V I and Pulavskya I G 1974 Elektrichestvo 5 44
[lo] Stekolnikov I S 1943 Lightning Physics and Lightning Protection (Moscow,
Leningrad: Izdatelstvo Akademii Nauk SSSR) p 229 (in Russian).
[ l l ] Akopyan A A 1940 Res. All-Union. Electr. Inst (Moscow) 36 94
[12] Bazelyan E M, Sadychova E A and Filippova E B 1968 Elektrichesrvo 1 30
[13] Bazelyan E M and Sadichova E A 1970 Elektrichesrvo 10 63
[14] Aleksandrov G N, Bazelyan E M, Ivanov V L et a1 1973 Elektrichesrvo 3 63
[I51 Bazelyan E M. Burmistrov M V, Volkova 0 V and Levitov V I 1973 Elektri-
chesrvo 7 72
[16] Cann G 1944 Trans. AIEE 63 1157
[17] Gorin B N and Berlina N S 1972 Elektrichesrvo 6 36
[18] Gorin B N, Levitob V I and Shkilev A V 1977 Elektrichesrvo 8 19
[19] Bazelyan E M 1967 Elektrichesrvo 7 64
[20] International Standard Protection Structures against Lightning 1990 IEC 1021
P 48
[21] Burgsdorf V V 1969 Elektrichesrvo 8 31
[22] Kostenko M V, Polovoy I F and Rosenfeld A N 1961 Elektrichesrvo 4 20
[23] Bazelyan E M 1981 Elektrichesrvo 5 24
[24] Larionov V P, Kolechitsky E S and Shulgin V N 1981 Elektrichesrvo 5 19
[25] Bazelyan E N, Valamat-Zade T G and Shkilev A V 1975 Zzvestiya. Akad. Nauk
S S S R , Energetika i transport 6 149
[26] Aleksandrov N L and Bazelyan E M 1996 J. Phys. D: Appl. Phys. 29 2873
[27] Rakov V A, Uman M A and Thottappillil R 1994 J. Franklin Inst. 99 10745
p 294
[29] Uhlig C A 1956 Proc. High Voltage Symp. Nut. Res. Council of Canada

Chapter 6
Dangerous lightning effects on

modern structures
This chapter is concerned with the mechanisms of hazardous lightning effects

on various objects in the atmosphere, having no contact with the earth, on
terrestrial constructions and underground communications lines. The dis-
cussion will be restricted to those effects which are, in this way or other,
produced by the electrical and magnetic fields of lightning. No doubt, a
hot lightning channel can ignite inflammable material but their direct contact
is a rare phenomenon, whereas a remote excitation of sparks in such material
due to electrostatic or magnetic induction is a regular thing. Lightning can
destroy constructions by a purely mechanical action but this does not
happen often. The burn-offs and holes at the site of contact of a hot lightning
channel with metal are hazardous only to thin (one millimetre thick) metallic
coatings. On the other hand, the range of electromagnetic effects is very wide.
They can damage both microelectronic devices and ultrahigh voltage lines.
The test maintenance of a 1150kV transmission line in Russia has shown
that it is not resistant to powerful lightning discharges. Most of the material
presented in this chapter concerns the physical mechanisms of electrical,
magnetic and current effects of lightning. We shall discuss simple and clear
qualitative models illustrating the physics of these processes. We believe
that this is the key issue to lightning protection theory. The process of
equation solution, so important two decades ago, is not so essential today.
If a physical model describes the reality adequately and the respective
equations are available, modern computers are able to overcome almost
any mathematical complexity.
When a lightning strikes a grounded metallic construction, a high return
stroke current I, passes through it. Because of an imperfect grounding
having a resistance R,, the construction potential rises by the value
U = IMR,, for example, by 1 MV at I, = 50 kA and R - 20R. This is
g.-
one of the reasons for the overvoltage due to a direct lightning stroke.
Another reason is the emf of magnetic induction (the intrinsic induction
265

266 Dangerous lightning effects on modern structures
due to an abrupt current change in the construction and the mutual induction
produced by the current wave running through the lightning channel). But
lightning overvoltages may result not only from a direct stroke but from
remote lightning discharges as well. Their effect is associated with electro-
static and electromagnetic inductions. In the former case, an overvoltage
results from the time variation of the electric field strength at the object,
created by the lightning channel charges during the leader and return
stroke stages (sometimes, by the slowly changing charge of the storm
cloud). Another reason for a remote excitation of overvoltage is the varying
magnetic field of the rapidly changing lightning current. Overvoltages
became a very serious hazard at the beginning of the twentieth century
when the first power transmission lines were built, and the engineer still
associates an overvoltage with a powerful effect of tens and hundreds of kilo-
volts. This is true of transmission lines of high and ultrahigh voltages (UHV
lines). However, an overvoltage as small as several hundreds or dozens of
volts may become hazardous to electric circuits with a low operating voltage.
Especially vulnerable in this respect are the circuits of microelectronic
devices.
Historically, the theory of overvoltages has developed with reference to
power transmission lines. Naturally, the mechanisms of ultrahigh voltage
excitation were the first to attract the researchers’ attention. So this theory
is now very detailed [l-41 and the numerical procedures suggested are
capable of solving engineering problems with a desired accuracy. We shall
not describe these approaches here but rather focus on the physical aspects
of the overvoltage problem, because in many practical applications they
are not as self-evident as in a lightning stroke at a power line.
The calculation of overvoltage includes the solution of two equally
important problems. One is to find the electromagnetic field of a lightning
discharge at the site where the object to be protected is located. These calcu-
lations may prove very cumbersome and time-consuming, especially when
one tries to take into consideration such parameters as the real path and
length of a leader channel, the non-uniform charge distribution along the
channel length, and the lightning current spread over the metallic parts of
a particular object and underground service lines. The physical aspects of
this problem, however, are quite clear and the numerical methods are well
known. The other problem is to determine the response of an object and
its electrical circuits to the electromagnetic field of lightning. The physical
aspects of this problem are much more diverse, and the basic mechanisms
of overvoltage excitation are not always obvious. So the latter are the subject
of special interest in this chapter.
An induced overvoltage is normally smaller than an overvoltage pro-
duced by a direct stroke, especially by remote strokes, but it affects the
object more frequently. When one calculates the frequency of emergencies
for a high-voltage circuit with an insulation designed for hundreds of

Induced overvoltage 261
kilovolts, one usually deals with direct strokes, because induced overvoltages
cannot damage the insulation. Objects with metallic shells which can screen
well the internal electric circuits (including low-voltage ones) are designed in
a similar way. However, unscreened low-voltage circuits suffer equally from
overvoltages due to direct strokes and from induced overvoltages. Since the
latter are more numerous, they should not be discarded when choosing the
protective measures.
6.1 Induced overvoltage

6.1.1 ‘Electrostatic’ effects of cloud and lightning charges
The atmospheric electric field varies in time during a storm. The slowest
changes, lasting for several seconds or tens of seconds, are due to the
accumulated charges of the storm cloud cells and their transport by the
wind. Field variations associated with the leader propagation last for several
milliseconds. Changes of microsecond duration arise from the charge re-
distribution during the return stroke. In any field variation, the electrostatic
potential of a perfectly grounded object would remain equal to zero. In
reality, however, the grounding resistance R, is always finite. If the change
in the charge induced on the object surface creates current i, = dqi/dt
through the grounding rod, the object acquires potential U = -i,R, relative
to the earth.
A grounded body of capacitance C possesses a potential difference
AU = U - U, relative to the adjacent space (here, U, is the average potential
of the external field Eo at the object’s site). The charge induced on the body is
q, = CAU; hence, current i, is defined by the equations
-
d i, + A - -A-, . A = -d U,
dt R,C- R, ’ dt ‘
2, = - exp(-t’RgC)
R,
1: A,(t‘) exp(t‘/R,C) dt’
where we assume i g ( 0 )= 0. In a simple case with A , = const, we have

i, = -A,C[l - exp(-t/R,C)], U = A,R,C[l - exp(-t/R,C)]. (6.2)
For the estimation, we put C = 100 pF, corresponding to a sphere of 1 m
radius, and set the overvoltage amplitude below 1 kV. During a storm
without lightning discharges (the field variation A , lo4 Vjs),the desired
N
grounding resistance should be R, < 1OOOMR. But in the presence of a

close descending lightning leader with the field variation A , lo9Vjs, theN
grounding resistance must be reduced to 10 kR. With the account of the

return stroke at A , 10” Vjs, this value must be decreased further to
N
1000. Therefore, a good grounding of an object seems to be an effective

268 Dangerous lightniTzg effects on modern structures
tool for its protection against overvoltages excited by electrostatic induction.

No doubt, faster variations in the external field impose more stringent
requirements on the grounding rod. Resistances exceeding 1000 MO are
hardly realistic because of the leakage across the unclean surface of even a
perfect insulation. For this reason, overvoltages due to a slow variation in
the storm cloud charge present a problem only in exceptional situations
(for example, in providing protection to the explosives industries or to
storages of explosives). It is not difficult to provide a 100Q resistance but
special designs are necessary.
We should like to mention an exotic but fairly realistic situation when
the object capacitance is subject to a change. This happens, for example, in
apparatus with remote wire control. When the apparatus goes away horizon-
tally from the operator and the cable elongates with a constant velocity
v = const, the capacitance grows linearly in time, C ( t ) = Clvt. At a constant
external field, the grounding electrode current and the object voltage relative
to the earth do not change in time and are
During the object motion up to a cloud, the overvoltage will be larger

because of the higher average potential of the conductor, U, x Eowt/2.
Let us calculate the overvoltage due to the return stroke current. Its
specificity results from a high velocity of the recharging wave through the
channel, wr, which is comparable with light velocity c. Strictly, this requires
account to be taken of the delay time of an electromagnetic signal in the
calculation of charges induced on the object. When faced with this complex
task, engineers sometimes feel a mystic horror. In actual fact, the delay
changes little in many situations, especially in the case of a compact object.
To illustrate this, consider the limiting case when a terrestrial compact
object is located right under a vertical, descending leader, more exactly,
when the horizontal distance to the stroke point is r << z , where z is the
height of the return wave front. At the moment of time t, the leader charge
is neutralized, and the channel is recharged along its portion from the
earth to the altitude z = vrt. But the object 'is aware' of the charge change
along a shorter portion only, z, = cv,t/(c + vr). The effect of the delay is
equivalent to a decrease in the return wave velocity by a factor of
(1 + vr/c). The equivalent velocity is v,, = z e / t > vr/2, because U, < c. For
a lightning of medium power with v, x 0.25c, the velocity is wre RZ 0 . 8 ~ A~ .
20% correction is of little importance, particularly as the neglect of the
delay leads to an overestimated overvoltage, thus providing a certain reserve
for the engineering solution. The effect of the delay will be smaller at
comparable values of r and z . Indeed, the distance between the charge
neutralization front and the object, (r2 + z2)lI2,increases more slowly than
z. It remains nearly unchanged at r >> z. Therefore, the time evolution of

the field, Eo(t),at the object's site will not differ from that calculated neglect-
ing the delay. The phase delay which acts for the time A t = r / c does not
affect the overvoltage.
Let us make a direct evaluation of the 'electrostatic' component of
overvoltage during the return stroke, assuming that a rectangular charge
neutralization wave (section 4.4) is moving along a vertical, perfectly
conducting channel towards a cloud. At any point of the channel behind
the wave front z = wrt, the charge changes by the same value r.The electric
field follows the charge variation. Without the account of the delay, its
change AE, at the distance r from the channel is described by an expression
similar to (3.5) (with h = 0, H = z and R = z):
The time constant for real electric circuits, R,C < 0.1 ps, is several orders of
magnitude smaller than the time of the return stroke flight from the earth to
the cloud. Then, according to (6.2), the electric component of the overvoltage
(relative to the earth) for a compact object is defined as
dAE, r R Ch vr2 t
U, R,Ch- -
dt 2 m (vft2 + r2)3'2
where h is an average object height. The short-term action of this overvoltage
load must be endured by all the insulation gaps separating the object from
the adjacent constructions and service lines, whose potentials were not
changed by the lightning or, if they were, to a different extent.
At the moment of time tmax, the pulse Ue(t) reaches its maximum
In the second formula of (6.6), we have substituted I, = TU,. A lightning of

medium current IM = 30 kA, which has contacted the earth at the distance
r = lOOm from the object of medium height h = 10m and capacitance
C = lOOOpF (a wire l00m long), is capable of exciting an overvoltage
pulse with an amplitude Uem,,= 2 kV at R, = 10 R because of the channel
recharging during the return stroke.
Most of the parameters in (6.6), are beyond the engineer's capacity when
he requires a high protection reliability. It is hardly possible to change the
capacitance or average height of the object being protected. It seems more
feasible to reduce the overvoltage to a safe level by decreasing the grounding
rod resistance R,. This is an effective way of overvoltage protection against

electrostatic induction. However, this measure, like any other technological

tool, has its limitations. It is difficult to provide R, < 1 R in a impulse
mode. The obstacles are the relatively low conductivity of the earth and
the inductance of the grounding conductors, which are fairly long when
the grounding mat occupies a large area. After the potentialities of R, reduc-
tion have been exhausted, there remains only one way - increasing the dis-
tance r to the nearest lightning discharge. To do this, one has to protect
from direct strokes not only the object itself but the area around it together
with the other constructions located on it, some of which are higher than the
object to be protected. In that case, all lightning rods must necessarily be
mounted outside this area; otherwise, the protectors will be able to attract
lightnings, bringing their charges close to the object.
In contrast to the amplitude, the duration of the overvoltage pulse front
is practically independent of the object's parameters, being primarily deter-
mined by the distance to the stroke point, r . From the first formula of
(6.6), we have t,, 0 . 7 at
~ ~r = lOOm and TI, x 0 . 3 ~ Overvoltages
. of
microsecond duration are typical of the lightning return stroke. Pulses
with the front duration of 1 - 1 . 2 ~are
~ still used as standards in insulation
tests for resistance to lightning overvoltages, although they do not always
reflect the reality.
6.1.2 Overvoltage due to lightning magnetic field

The problem of overvoltage induced by the magnetic field of a lightning
discharge is the most common one among overvoltage problems. The
lightning current varying in time and space induces the emf in any circuit.
If a circuit is formed by conductors, the emf excites electric current. If the
circuit is disconnected, the voltage equal to the induced emf is applied to
the break. Let us estimate the maximum effect produced by an infinitely
long straight conductor with current i. At the distance r from the conductor,
the magnetic field is H = pOi/27rr. Consider a rectangular frame in a plane
intercepting the conductor (figure 6.1). Suppose the side parallel to the
conductor has a length h and the side normal to it has r2 - rl = d ; the short-
est distance between the frame and the conductor is r l . The magnetic flux
through the frame is defined as
--In-.
r1
The emf induced in the circuit, U , = -dQ/dt, is
At the maximum rate of the current change, Ai z 10" A/s, characteristic of the
return stroke of subsequent lightning components, the emf'induced in a circuit

I Pr
Figure 6.1. Estimating the overvoltage magnetic component.
with the sides h = d = 10 m at the distance rl = 100 m from the conductor with
current is U , = 19 kV. The emf for a smoother current impulse of the first
component of a moderate lightning with Ai = 5 x lo9A/s is U , = 1 kV.
Overvoltages excited electrostatically and electromagnetically are gener-
ally comparable. The former can be coped with using an effective grounding
of the object, but overvoltages due to the electromagnetic mechanism do not
respond to the grounding efficiency. Imagine metallic columns buried deep in
the ground, which support rails for a mobile overhead-track crane mounted
high up at the ceiling of industrial premises. The whole construction has a
perfect grounding owing to the metallic columns which provide a complete
absence of electrostatic overvoltages from close lightning strokes. However,
a pair of columns with a rail and the conducting earth forms a closed circuit
with an area of several hundreds of square metres, in which the time-variable
lightning current excites an emf. The same thing occurs in a circuit formed by
columns, fixed at the opposite sides of the premises, and an overhead crane.
A possible disconnection at any site of the metallic construction cannot be
ignored either. A disconnection may arise due to metal erosion, poor welding
or inadequate contact between the crane wheel and the rail. In that case,
practically all emf of the circuit will appear to be applied to the site of
defect, provoking a spark discharge through the air or a creeping discharge
across the surface to bypass the defective site. A spark-induced emergency is
inevitable if there is an explosive gas mixture in the premise.
The fact that any construction may serve as a circuit capable of inducing
an emf increases the hazard - this may be a metallic ladder on a conductive
floor, a metallic pipe leaning against a wall, etc. Such casual circuits present
an even more serious hazard, because their parts may have only a slight
contact between them, so that the probability of a spark gap is extremely
high. An explosion would, no doubt, destroy the casual circuit, creating a
mystery to the fire brigade in the spirit of Agatha Christie’s stories.
The sequence of procedures for the calculation of overvoltages due to
lightning current is similar to that for lightning charge calculation. One

should first find the magnetic flux through the circuit in question and
calculate the induced emf. The magnetic flux is often replaced by the
vector potential A(t) to simplify the calculations. For current i in a thin
conductor such as a lightning channel, the vector-potential is
where the integral is taken in the conductor length and r is the distance from
the current element id1 to the point, at which A is determined. The emf
induced in the circuit of interest is defined as
(6.10)
where EM is the strength of a vortex electric field excited by the time-variable

magnetic field of the lightning. For a straight conductor with current, the
vector EM is parallel to the current. If the lightning channel is vertical, the
vector EM is also vertical.
Let us represent a lightning return stroke as a rectangular wave of
current ZM propagating at velocity vr along a vertical channel from the
earth up to the cloud. Without accounting for the delay, leading to a certain
overestimation of the result, we have
Factor 2, instead of 4, in the denominator results from the allowance

for the current spread in the earth. The field EM is vertical, so the horizontal
sections of the circuit do not contribute to U,. In the vertical sections, the
values of EM are summed algebraically. For a metallic frame with an air
gap, like the one shown in figure 6.1, the magnetic component of overvoltage
in a small gap of A << h is
u.44 = h P d r 1 ) - E M ( Y 2 ) I (6.12)
where EM are taken to be values averaged over the conductor heights h.
All the results obtained within the model of a rectangular current wave
of the lightning return stroke overestimate the overvoltage; the smoother the
front of the real current wave, the greater is the overestimation.
6.2 Lightning stroke at a screened object

6.2.1 A stroke at the metallic shell of a body
Overvoltages due to a lightning stroke at the metallic shell of a body, such as
a plane or other objects, occur very frequently. To get an idea of what
happens in this case, let us look at the schematic diagram in figure 6.2.

Lightning stroke at a screened object 273
Figure 6.2. Lightning current flows along a pipe with a conductor inside.
Suppose lightning current runs along a closed metallic shell of an object, inside
which there is a conductor connected to the shell at one of its ends, say, at the
lightning current input. The potential at this contact will be taken to be zero. If
R I is the linear resistance of a shell of length 1 and L1 is its linear inductance,
the voltage applied to the shell will be Uf = -(L1 di/dt + Rli)l.The lightning
current does not branch into the inner conductor disconnected at the other end
(the capacitance is neglected). The conductor potential changes only due to
the mutual inductance, U, = Mlldi/dt. Since the magnetic flux of the shell
current is entirely attributed to the inner conductor, the linear mutual induc-
tance M I is equal to the linear inductance of the frame, L1. Then the potential
difference between the shell and the inner conductor at the far end of the latter
is described as
U, = U, - Uf = iR1l. (6.13)
The remarkable property of a cylindrical system with an inner wire to
compensate completely the induction emf is well known to impulse measure-
ment technology. This property is the basis for making shunts for measuring
current impulses with very short fronts (to a few fractions of a nanosecond).
The respective theory, useful for the understanding of the overvoltage
mechanism, is discussed in detail in [ 5 ] . We shall turn to it when evaluating
the skin-effect in a shell. Here, it should be noted that the shape of an
overvoltage pulse, U,(t), in the absence of a skin-effect is similar to that of
a current impulse, i(t). This is valid as long as the time of the electromagnetic
wave propagation along the frame is much shorter than the impulse
duration.
No principal changes will occur when the conductor ends are connected
to the shell via resistances Rkl and Rk2.The voltage U, will then appear to be
operative in the inner closed circuit consisting of the shell, conductor, and
resistors. When the resistances of the conductor and the shell are small, the
current i = U e / ( R k l+ Rk2) arising in the circuit will distribute the over-
voltage U, between the turned-on resistors in reverse proportion to their
values. The same will happen when the conductor is connected to the shell
via spurious capacitances. Of course, if a massive aluminium shell has a
cross section of 100 cm2 and the linear resistance is R1 E 3 x lop6n/m, the

overvoltages may be small, U, RZ 30V, even at the maximum lightning

current Z,w = 200 kA and a long 1 = 50 m. But one should bear in mind
that modern microelectronic units stuffed into an aircraft or spacecraft can
be damaged easily even by lower voltages.
6.2.2 How lightning finds its way to an underground cable

The problem of lightning access to an underground cable deserves special
attention, because the spark propagation under these conditions has a
peculiar physical mechanism. A direct stroke of a descending lightning at an
underground cable is a rare event, since the leader cannot ‘sense’ its presence.
The current flows to the cable along a spark channel creeping along the earth’s
surface. Its path can sometimes be identified easily because of the bulging
loosened soil. Normally, the spark path is as long as several dozens of
metres, or even hundreds of metres in low conductivity soils. It seems unlikely
that a creeping spark should move towards a cable purposefully; rather, this is a
matter of chance. But the local topography can stimulate the spark access to the
cable. Suppose a cable is laid along a forest path, and the current of the light-
ning that has struck a nearby tree flows down to its roots, giving rise to a spark
channel, which propagates across the path until it hits the cable.
A high current spreading through a poor conductor such as soil induces a
fairly high electric field which initiates ionization. This fact has long been
known, so the calculations of grounding resistance take into account the
increasing radius of the metallic conductor owing to the larger ionization
zone around the metal. It has also been suggested that a strong electric field
induced by high current may cause a breakdown of some gaps by a spark fila-
ment in the soil [6]. The soil ionization creates a natural ’grounding electrode’,
when a lightning channel contacts the earth’s surface. The mechanism of
current spread through the soil can become clear from analysis of the simple
case of a spherically symmetrical distribution of current I,. At the distance
Y from the point of lightning stroke, the current is sustained by the electric
field E = p1M/(27rr2), where p is the soil resistivity. The soil represents a
porous medium with the pores filled by air. Experiments show that the soil
air is ionized more readily than the atmospheric air at E > EISE 10 kV/cm
[6,7]. This is due to the local field enhancement around sand grains, etc.
(cf. section 4.3.1). Therefore, the medium in a hemisphere of radius
Y, = (I~p/27rE1g)’i2 is ionized to become a natural well-conducting grounding
electrode. The grounding electrode resistance, i.e., the resistance to the current
spread through a non-ionized soil, is defined as
R --I1 ” Edr=---.
- IM r,
P
~ T Y ,
(6.14)
For example, the grounding resistance is found to be R, = 720 at

ZM = 30 kA, p = lo3 0 . m (sandy soil), and ri = 2.2m.

This situation is very unlikely because the process is unstable. Even a

slight asymmetry, which is always present in nature, say, the asymmetry
created by tree roots at the site of the lightning strike, may produce a creeping
discharge. A plasma channel similar to a leader channel originates at the
strike site. It acts as a long grounding electrode, from which the lightning
current spreads through the soil. The leakage current per unit channel
length, Zl, is proportional to the channel potential U at this site, Zl= G, U .
The linear conductivity GI of the leakage through the channel surface
contacting the soil is defined by an expression similar to (6.14) but with
allowance for the cylindrical (or, rather, semi-cylindrical) geometry. The
radial field at radial distances r smaller than the conductor length I is
E M Z1p(7rr)-', where I , = Z M / I is the leakage current per unit channel
length. When integrating the field over the radius to find the channel poten-
tial U , one should take the upper limit I , x I , because at r > I the field
decreases as l/r2 and the integral converges quickly. Hence, we have
(6.15)
Here, ri is the radius of a well-conducting channel. Because of the logarithmic

dependence of G1 on ri and 11, these values do not affect G1 much.
Laboratory experiments [8] have shown that the principal difference
between a classical leader in air and a spark running along a conducting
surface is the mechanism of current production providing the energy for
the channel heating. In the former case, the current is produced by the
streamer zone in front of the leader tip (section 2.4.3) and in the latter,
owing to the transverse current leakage from the surface of the channel
contact with a conducting medium. A streamer-free leader process was
clearly observed under these conditions in laboratory experiments [5,8].
The streak picture in figure 6.3 does not show even a trace of the streamer
zone, whereas the air gap of the same length is filled by streamers nearly
from the very beginning of the leader process, in the absence of a conducting
surface. The spark process occurring along a conducting surface is very
effective. A creeping leader requires an order of magnitude lower voltage
for its development than an ordinary leader - 135 kV instead of 1300kV -
for bridging a gap of 5 m long. Of primary importance here is the medium
conductivity and the current supplied to the channel. To make a streamer-
free leader move on, the field at its tip must be E > Elgto be able to initiate
the ionization, to supply the initial channel with a current as high as the
ordinary leader current, it > if,,, 1 A, to heat the gas rapidly, and to
N
maintain the channel conductivity (section 2.4.3).

A small portion of the lightning current, it << ZAw, delivered to the leader
tip is sufficient for a creeping leader to develop successfully. The tip current
is, at first, very high. But as the channel grows, more and more lightning
current leaks down to the earth because of the increasing contact area of

Figure 6.3. Streak photographs of a leader creeping along the soil (top) and an air
leader (bottom). 1: channel, 2: tip, 3: streamer zone.
the plasma column with the conducting soil. So, the leader eventually stops.
Let us evaluate the maximum length I of the leader channel. Suppose current
,Z is delivered to the channel through its base at the stroke point. The current
value is determined by the recharging of the lightning leader channel at the
return stroke stage and is independent of the creeping spark length. For
simplicity, we take the delivered current I.w and the longitudinal field E,,
supporting the creeping leader current, to be constant. At high currents
(i > 1 A), the dependence E,(i) is, indeed, not particularly strong. By the
moment the leader has stopped, the tip potential U, and current it are low
relative to U(x) and i(x) at distances .x from the tip, comparable with the
channel length. We then have
di GIEcx2
U ( x )M E,x, _
dx
-- Zl= G,U(.x), 2 .
i(x) = (6.16)
~
With i(Z) = 1
, at the channel base, the maximum channel length is defined,
1
with the account of (6.15), as
1%
( 2zM - [
-
)'/2
N
21,p In ( I / ~ J 'I2
(6.17)
GI E, ..E,

If the longitudinal field E, is lOOV/cm, as in the case with a common

leader channel which is usually close to the arc state, the channel length I
will be 40m (ri x 1 cm) for an average lightning with IM x 30 kA and a
-
well conducting soil with p x 100 R m. The channel length will grow with
rising lightning current and decreasing soil conductivity: its value is
I M 220 m at the maximum current 1, x 200 kA and p x 1000 R/m. These
estimates are consistent with observations. If a creeping leader encounters
a cable, the still available current in it will penetrate to the cable sheath.
6.2.3 Overvoltage on underground cable insulation

If one digs the soil to expose the site of the lightning current input into a
cable, one can observe the cable cores with damaged insulation, which are
in contact with the metallic sheath. The damage may be stimulated by the
presence of a gas-generating dielectric in the cable. The dielectric is decom-
posed, because of the heating by high current, to produce an electrical
hydraulic effect, so that the cable appears literally compressed by the
shock wave. A similar effect can be produced by an explosive evaporation
of soil water. The elimination of the damage at the current input may not
remove the emergency, because there may be several others along a distance
of several hundred metres, on both sides of the strike point. These damages
result from overvoltages arising between the core and the sheath during the
lightning current flow along the cable. The overvoltage mechanism is similar
to that described in section 6.2.1, except that the conductor with a sheath has
a longer length, sometimes of many kilometres. When the lightning has
incorporated its current into the sheath, the cable in a soil of infinite
volume should be regarded as a long line with distributed parameters, or,
more exactly, as two lines. One is the sheath in a conducting soil. The light-
ning current flowing along the sheath gradually leaks into the soil and goes to
‘infinity’, thus raising the sheath potential U,(., t ) relative to an infinitely far
point on the earth. The other line is the core with the sheath. It is affected by
the magnetic field of the sheath current, giving rise to an induction emf and
voltage drop in the conductive sheath due to its finite linear resistance RI,. As
a result, the cable core acquires potential U,(x, t ) relative to infinity, which is
generally different from U , ( x .t ) . The difference U, = U, - U , represents the
overvoltage on the cable insulation capable of damaging it.
A rigorous solution to the problem of U e ( x .t ) follows from a combined
solution of the set of equations describing the lightning current flow along a
cable sheath and the voltage wave propagation (between the core and the
sheath) along the cable core. This would be a correct approach, provided
that the waves in the sheath and inside the cable had approximately the
same velocities. But we shall show that these velocities differ by several
orders of magnitude, which necessitates the subdivision of this problem
into two problems. One will describe the lightning current flow along the

sheath and the other the propagation of waves, excited by this current, inside
the cable.
Let us first follow the fate of lightning current i ( x ,t ) in the cable sheath.
Its variation along the length due to the displacement current associated with
the charging of the sheath linear capacitance C1, to the voltage U,(x,t ) can be
assumed to be negligible, as compared with the large current leakage into the
soil through the linear conduction G1 of the sheath grounding. One can also
neglect the mutual induction emf in the sheath, produced by the core current
i,, because it is small compared to the self-induction emf. Since the total
magnetic flux of the sheath current i involves both the sheath and the core,
the mutual inductance M 1 per unit length of the sheath-core system is
equal to the linear sheath inductance L1. However, the current in the core
is i, << i. Indeed, the current in the core screened from the earth by the
sheath is only due to the charging of the cable capacitance C1, to the voltage
U, acting between the core and the sheath. The value of U, does not exceed
the electrical strength of the cable insulation, U, M 2 kV. Even if the current
wave velocity in the core were close to light velocity, the core current would
be of the order i, M C1,Uec M 10-30A (for a cable of a small cross section,
C1, M 20-50 pF/m), which is much lower than the lightning current
i M 10kA. So, one can ignore the current deviation into the core even
when its insulation is damaged at the lightning current input into the cable
so that the core appears to be connected to the sheath.
Therefore, on the above assumptions, the current flow along the sheath
is defined by the equations
aU, ai ai
(6.18)
ax
-L1-+Rli,
at ax
-- = G1 us
where L 1 is given by formula (4.25) and R 1 is its linear resistance. If the cable
were on the earth's surface, with the lower half of the sheath touching the
earth, formula (6.15) would be valid for G1. When a cable is buried at a
large depth, the current spreads radially from it in all directions uniformly,
so the value of GI is doubled. In intermediate situations, one can use the
empirical formula
2n
G1 = Y Q h K 114 (6.19)
p In ( 12/2hr)'
where h is the cable depth. The boundary condition for (6.18) is expressed by
the equality i(0,t ) = Io, where Z,,is one half of the current delivered by the
lightning to the cable at the input x = 0 (the current flows in both directions
from this point).
The cable sheath possesses a low active resistance and a fairly high
inductance because it is a solitary conductor. The self-induction emf has
a greater effect on the distribution of the rapidly varying lightning current
in the sheath than the active voltage drop. If Rli is neglected in the first

0.0 0.5 1.o 1.5 2.0

Z
Figure 6.4. The function 1 - erf(z).
approximation, the set of equations (6.18) changes to the familiar diffusion

equation, with the only difference that the diffusion coefficient xo is now
defined by LIGl rather than by RIC1.The current varies along the sheath
length in both directions from the input as
X 1
x O
--
- LlGl
x -.
Po
2P (6.20)
The latter expression for xo corresponds to a surface cable; at a large depth, this
would be xo = p / p o : xo x 160-600m2/ps at p x 102-103Rjm ( p x 500 s2jm
for a common sandy soil). The point with a fixed value of illois shfted with
a decreasing velocity ‘U x xo/x x ( ~ , / t ) ” ~in, agreement with the diffusion
law x (4xor)’/*. The current covers a 1 km cable length for t x 2000-
200 ps, decreasing rapidly at the wave front (figure 6.4). The sheath potential
from (6.18) and (6.20) is
The potential at the current input drops with time, from an ‘infinite’value at
t = 0, which results from the neglect of Cls.tThe equivalent sheath resistance
t If C1, is taken into account, there is a weak precursor which propagates with the velocity of an
electromagnetic signal (L, overtaking the diffusion wave described by (6.20) and (6.21)
(cf. section 4.4.2).

also decreases with time. It is defined by the resistance of the soil around the
elongating cylindrical surface, through which the current leaks.
Now turn to the wave process inside the cable. The type of overvoltage
under consideration is dangerous only to communications lines, whose linear
inductance L1, is small because of the narrow gap between the core and the
sheath. The self-induction term is usually small relative to the voltage drop
on the active core resistance RI,. The current leakage through a high quality
insulation can also be neglected, since it is small relative to the displacement
current charging the linear core capacitance C1, (relative to the sheath). With
these assumptions, the core potential U, relative to infinity and the core
current i, are described by the equations
au, - Rlcic+ L1di ai, = Clca(uc - Us)
-- (6.23)
dX at!
ax at
which account for M 1 = L1. The electrical signal induced in the core by the
lightning stroke has a much higher propagation rate than the process of filling
the sheath with current. Indeed, we have U, = 0 and a i l a t = 0 far ahead of
the filled part of the sheath. Equations (6.23)transform to the diffusion equation
for U, and i, with the coefficient xc = (RlcClc)-lx 2.5 x 106-2 x lo5m2/ps
( R I ,x 0.01-0.1 fl em) exceeding xo by several orders of magnitude. This
means that the charging of the cable capacitance occurs very quickly, and a
quasi-stationary mode is established in the cable, in which U, and i, follow a
relatively slow variation of the sheath current.
By subtracting the first equalities of (6.18) and (6.23) from one another
and keeping in mind R I , R I and i, << i, we obtain the equations for over-
N
voltages in the cable:

au, = Rli. U,(x. t ) FZ i ( x , t ) R 1dx + UJO, t ) . (6.24)
dX
If the cable insulation at the lightning current input is intact, V,(CQ,t ) = 0
and U 2 ( m .t ) = 0 , the overvoltage value is maximal at the input point and
is defined as
U 2 ( 0 .t ) = -
sox i ( x . t)R1 dx KZ -ZoRlxl,
where x1 is the equivalent sheath length with the lightning current at the
= ( 4 ~ ~ t ) ” ~(6.25)
moment of time t. Overvoltages rise in time as long as the lightning

current is high; more exactly, as long as its decrease is compensated by the
elongation x1 (the situation for a realistic current impulse will be discussed
below).
If the cable insulation is damaged at the current input ‘instantaneously’
and the core contacts the sheath, then we have U,(O, t ) = 0 and the over-
voltage grows with distance from the stroke point up to the maximum
value of (6.25) at x > x1 = ( 4 ~ ~ t ) provided
’/~, of course that the lightning

Lightning stroke at a screened object 28 1
current is still high at the moment t.t For example, at Io = ZM/2 = 10kA and
R1 = 3.5 x lop4sljm (the aluminium sheath is 1 mm thick and 30“ in
diameter), we have U, M ZoRlxl x 2 kV at a distance x1 M 600m from the
current input. This happens at the moment of time t M x:/4x0 x 6 0 p (at
xo x 1000m2/ps, if the lightning current is still high relative to its amplitude.
This is the duration of comparatively short current impulses of negative
lightnings. For anomalously long impulses (- 1000ps) of positive lightnings,
the length of the ‘active’ cable portion where the overvoltage arises can
increase to 1-10 km, with the overvoltage amplitude becoming appreciably
larger. It is clear now why the repair of the damaged insulation at the
lightning input is insufficient and other damaged sites must be found and
removed along several kilometres of the cable length. In regions with
poorly conducting soils (rocks, permafrost), a damaged line may extend to
dozens of kilometres.
So far, we have evaluated the overvoltage for a rectangular current
impulse. To calculate it for a real lightning pulse, we should first find a
more rigorous solution for the current input into the cable with an intact
insulation. This will provide the maximum value of U,. We shall apply the
operator approach to equation (6.18), omitting the term R1i, as before. As
a result, we get the expression
A = (P/xo)1/2= (PPo/2P)1/2
in which the last term corresponds to a cable on the earth’s surface. If unit
current i(0,t ) = Io = 1 flows into the sheath, the integration constant is
A = 1. The operator form of the overvoltage is
The inverse transform of (6.27) is the function
(6.28)
which coincides, within the accuracy of the numerical coefficient of the order
of unity, with (6.25) at Io = 1. Expression (6.28) for unit current Z o ( t ) = 1
represents the unit step function of y ( t ) providing the solution for arbitrary
lightning current i( t ) by taking the Duhamel-Carson integral. In particular,
we get the following expression for a bi-exponential current impulse
tThe equivalent core resistance Rlcx2, with x2 x ( 4 ~ ~ t ) grows

’ / ~ , in time, in contrast to the
decreasing input resistance of the sheath. This is another argument in favour of the current enter-
ing primarily the sheath rather than the core, even if they come in contact at the input.

0.01 . 1
0.01 0.1 1 10 100
Z
Figure 6.5. The function h(z).
i = 210 = I,w[exp(-at) - exp(-Pt)], allowing for its spread in both direc-

tions from the stroke point:
U,(O. t ) = IMR1 [h(4 h(9r)]
(7rp0/2p)”2 a1J2 fill2
(6.29)
h ( z ) = exp(-z) exp(y2)dy.
The function h ( z ) has a maximum h,, M 0.54 at z M 1 and goes down to

0.5hm,, at z E 5 (figure 6.5). Therefore, the overvoltage pulse front Ue(O,t )
is close to the duration of a lightning current impulse and U, decrease several
times more slowly than the current. The duration of the current pulse front
affects the overvoltage value only slightly. The estimation from (6.29) for a
current impulse of duration t - 100 ps ( a = 0.007 ps-’, P = 0.6 ps-’) gives
p .-.
Uem,,/ZMR1M 145m at resistivity p = l000R .m. For IM = 30kA and
R1 = 3.5 x R/m (an aluminium sheath of 1 mm thickness and 30”
diameter), the maximum overvoltage is 1.5 kV.
The diffusion equation for sheath current can be solved numerically for
any shape of the current impulse in a cable of finite length, when the core
contacts the sheath at the lightning stroke point and when the insulation
is intact. Figure 6.6 illustrates the results for the former situation. At
xo = 2p/p0 = 1.6 x lo9 m2/s ( p = 1000 R m), the bi-exponential current
impulse i(t) = Z M [exp(-of) - exp(-Pt)] with a 5 ps front and duration of
tp = loops (on the 0.5 level) excites an overvoltage pulse with the reduced
amplitude Uem,,/RIIO= llOm along the cable length of 500m. The over-
voltage maximum occurs at the moment of time t, = 60 ps; the overvoltage

I .
5b
" I
0 100 140 . 260 ' 240 300

Time, pi
Figure 6.6. Evaluated overvoltage pulses in the cable of a length x with a core con-
tacting a sheath at the stroke point. The bi-exponential lightning impulse current
with cy = 0.007 ps, p = 0.6 ps; p = 1000 njm.
drops by half in 230 ps. At a distance of 1000 m from the lightning current
input, the overvoltage pulse is somewhat higher, smoother and longer. For
current IM = 30 kA, its amplitude rises to 1.1 kV in an aluminium sheath
with R1 = 3.5 x lop4O/m and to 7.5 kV for a cable with a lead sheath of
the same cross section. All of the calculated parameters are quite comparable
with those estimated from (6.29).
6.2.4 The action of the skin-effect

One of the manifestations of the skin-effect is that the current turned on at a
certain moment takes some time to penetrate into the conductor bulk. The
characteristic time for a conductor of thickness d and conductivity 0 to be
filled by current is T, = p o d 2 ;for example, T, M 6 ps at d M 1 mm and
0 M lo7(a. m)-'. One can neglect the skin-effect when treating overvoltages
in underground cables with about the same sheath thickness but with an
order of magnitude longer time of lightning current flow along the cable.
One can assume the current to flow through the whole sheath thickness, as
was suggested above in the treatment of linear sheath resistance R I .But
when one considers short sheaths, especially those of terrestrial objects, in
which current runs very rapidly (at light velocity), it is often impossible to
ignore the finite time of current penetration in the transverse direction, i.e.,
through the sheath thickness.
The electric field and current diffuse from the conductor surface into its
bulk with the diffusion coefficient xs = (pOo)-' (hence, d2 x s T s ) .Due to
N

this fact, the effective resistance of the conductor is higher than in the case of
direct current. The formal use of this fact in (6.13) would result in an increase
in the overvoltage which is proportional to R I . But an opposite effect is
observed in reality. Owing to the skin effect, the overvoltage pulse front
becomes smoother than the current pulse front, reducing the overvoltage
at a finite pulse duration.
The reason for this paradox is that the last equality of (6.13), which is
strictly valid only for an infinitely thin sheath or for direct current, should
not be used in any situation. If the current varies in time and the sheath
has a finite thickness, its voltage can also be represented as a sum of the
resistance U R ( t )= f ' j l ( j is the current density) and the induction
Ui(t) M d@/dt components (@ is the magnetic flux). But with the same sum
+
U , = UR Vi, the value of each component varies with the point r of the
sheath cross section, for which the calculation is being made, since the
proportion between the current density j ( r ) and the magnetic flux @ ( r )
varies when the total current over the cross section changes. Calculations
of overvoltages between the conductor and the sheath, U, = U, - U,, are
generally indifferent to which r the value of U, is being found, because the
potential does not vary with the thickness. For simplicity, however, it is
reasonable to make calculations for the inner sheath surface: this surface
and the internal conductor are the only elements of the system affected by
equal magnetic fluxes, mutually excluding the induction components of
overvoltage on the conductor and the sheath. Consequently, formula
(6.13) can be replaced, without any restrictions, by the expression
U,(?) =j&)cr-ll = Ein(?)l (6.30)
where j,, and Einare the current density and longitudinal electric field, respec-
tively, on the inner surface of the object's sheath.
The current penetration into a thin sheath is described by the equation
for one-dimensional plane diffusion. It has been mentioned that the diffusion
coefficient is expressed by the quantity xs = (pea)-'. For a rectangular
current impulse of infinite duration, the longitudinal field strength on the
inner surface of a sheath of thickness d can be written as
The exponential series at t > y-' converges very rapidly, so one can restrict
oneself to the first term only. Therefore, the field Ei, rises with the time
constant T : = y-l = p0ad2/7r2; its value is 6 ps at cr FZ 5 x lo7 (Cl. m)-'
and d M 1 mm. This permits the neglect of the skin-effect action on over-
voltages in long underground cables, in which the current diffusion along
the sheath and, hence, the time of the overvoltage rise to the maximum
take 1 or 2 orders of magnitude longer than T : . However, the skin-effect
in objects located on the earth's surface and having relatively short sheaths,

0.0 ' .
50 100
I
150
I
200
Time, p
Figure 6.7. Overvoltage pulse deformation in a cable sheath due to skin-effect with
the time constant T, = 10 p. An exponential current impulse is duration of 100 ps
(dashed curve).
in which lightning current propagates over the time t << T i , decreases the
overvoltage with a greater efficiency in the case of a shorter current impulse.
For an exponential current impulse i(t) = ZMexp(-at), we have from for-
mula (6.31) with the first series term only and the Duhamel integral
exp(-at) -
27 exp(-yt)
-
7-a 1 t > y-'.
The results of the calculations presented in figure 6.7 show that the skin effect
(6.32)
elongates the overvoltage pulse front to T i and the amplitude decreases by

several dozens of percent.
6.2.5 The effect of cross section geometry

We have assumed so far that the sheath has the shape of a circular cylinder.
In that case, the current is distributed uniformly along the cross section
perimeter and there is no magnetic field inside. But this model is inapplicable
to many real objects, for example, the fuselage or wing of an aircraft, having
very complex cross section profiles with different curvatures. The current
flowing through a non-circular sheath is distributed non-uniformly along
its perimeter and the magnetic field is present inside. These factors affect
the mechanism of overvoltage excitation by lightning current.
Let us consider a two-dimensional sheath shaped as a cylinder of a non-
circular cross section and a considerable length I , when the end effects are
weak and the current and field distributions are plane-parallel. The sheath
is considered to have a uniform resistivity and thickness. Let us subdivide

286 Dangerous lightning tlffects on modern structures
the sheath into a set of N parallel conductors of a short length Ark along the
cross section perimeter, such that the current J k per perimeter unit length in
the kth conductor could be regarded as varying only with time (the total
current in the kth conductor is ik = JkArk). In a steady-state mode when
the current becomes direct, all J k values are the same, since they are
determined by equal ohmic voltage drops in all of the conductors. A mere
summing of the magnetic fields of the conductors will indicate that a
magnetic field may be present inside a sheath of an arbitrary cross section
geometry.
When lightning current is introduced into the sheath very quickly, the
magnetic induction emf in the conductors greatly exceeds the ohmic voltage
drop. But in this approximation, all of the conductors will indeed form an
integral ‘perfectly conducting’ sheath, namely, they will be connected in
parallel. This means that all of them will have equal potentials. Hence, the
magnetic Aux coupling @ for each conductor is the same. This provides a
set of equations for finding the currents ik at the initial stage of the process:
N N
Lkik(0) + Mkmim(0) m # k, im(0)= I M (6.33)
m= 1 k= 1
where Lk is inductance, Mkm is the mutual inductance of the conductors k

and m, and IM is the lightning input current. Now, in contrast to the
steady-state mode, the currents will be different even in identical conductors
if they are located at different sites of the sheath. We shall illustrate this with
reference to three parallel conductors of length I and radius r located in the
same plane so that the distances between the adjacent conductors are
identical and equal to D.If il is the current in the central conductor and iz
is that in the end conductors, with M 1 2as the mutual inductance of the
adjacent conductors and M23 of remote ones, we shall have
Lil + 2M12iz = Liz + M12il+ M23i2. il t 2i2 = IAM
The current in the central conductor is lower than in the end conductors
because of M12> M23.
It is easy to solve a set of equations of the type (6.33) even for a large
number of conductors simulating a sheath. Only the calculation of inter-
conductor distances is somewhat cumbersome, requiring knowledge of the
cross section profile coordinates. We shall leave this problem to the reader
and illustrate, instead, the analytical solution for the current distribution
in a long cylindrical sheath with an elliptical cross section [9]. This solution

is useful for the evaluation of many real profiles and for testing computation
programmes:
J(x) = (6.34)
2 4 a 2 - x2(1 - b2/a2)]’I2‘
Here, a is the large and b the small semiaxis of the ellipse and x is the distance
between the ellipse centre and the calculation point projection on the large
axis. The ratio of the minimum linear current density (on the plane part of
the ellipse) to the maximum one (on its tip) is Jmax/Jmin = a / b . The current
non-uniformity may be great in real structures, such as the aircraft wing,
a / b > 100.
There is no magnetic field in the sheath at the moment of time t = 0. This
is the result of the initial current distribution among the conductors owing to
the magnetic induction emf. With the redistribution of the currents under the
action of ohmic resistance, a magnetic field will gradually arise in a non-
circular sheath. The field becomes the source of overvoltages in the inner
circuits of the object. By integrating numerically the set of equations
where U (t) is also the unknown voltage drop along the length of the sheath
‘made up’ of conductors, one can find the variation in the current distribu-
tion along the sheath perimeter. The initial condition for the integration is
the solution to (6.33). The calculation accuracy increases with the number
N of simulating conductors. But the limiting case of N = 1 is also suitable
for the evaluation of the time constant of a transient process: TI, = L 1 / R 1 ,
where L1 and RI are the linear sheath inductance and resistance. The current
is redistributed slowly, T,, 0.1 s, in the sheaths of large objects with radius
N
Ypv 1 m and thickness d 1 mm. During the action of a common lightning

N
current impulse with t, 100 ps, the current distribution along the sheath
N
perimeter differs but little from the initial distribution profile. The results
of a computer simulation support this conclusion. The computation for
a sheath of complex geometry (figure 6.8) with L 1 = 0.57 pH/m and
RI = 1.05 x lop5Rjm (Ttz= 54ms) has shown that the linear current
density at all characteristic points of the sheath takes the steady-state value
for a time about 20 ms. During the first 200 ps typical of lightning current,
the density cannot change appreciably.
Let us consider overvoltages across the insulation between an inner
conductor and the sheath. Suppose the conductor is placed very close to
the inner sheath surface. The contour area between the conductor and the
wall will be very small, and the internal magnetic flux will be unable to
create an appreciable induction emf. The voltage between the conductor
and the sheath will be equal to the integral of the ohmic component of the

O’ fo 20 30 4-0 t,ms
Figure 6.8. Evaluated evolution of a linear current density at indicated points of the
wing-like sheath shown. J , = J ( t + XI).
longitudinal electric field &(x) at the site of the conductor location x. But
now, the evaluation of Einshould not be based on the average current density
in the sheath, using the total current and linear resistance R I .For a sheath of
thickness d , we have
=J(x)~/d. (6.36)
The nearer the current line with the maximum linear density, the higher the
overvoltage across the conductor insulation relative to the object’s shell. One
practical conclusion is quite evident. To reduce manifold the overvoltage in
the electrical line inside an aircraft wing and along its thinnest back end,
where the current density is maximal, it is sufficient to shift the wire closer
to the upper wing plane or, better, to the lower one, where the current density
is minimal due to the wing curvature (figure 6.8). Laboratory measurements
have confirmed this suggestion [lo].
Note the seemingly ambiguous character of the evaluations. The sheath
cross section is practically equipotential, so the inner conductor must be
under the same voltage with respect to any point of the sheath in a particular
cross section. However, the ohmic overvoltage component for a conductor
inside an elliptical cylinder (figure 6.9) with respect to points 1 and 2 of the
large and small semiaxes differ by a factor of Jmax/Jmin = a/b, in agreement
with (6.34). This contradiction is superficial. In the presence of a magnetic
field, there is the magnetic component, in addition to the electrical one,
U = U, + U,. The distance between the conductor and current line 1 is
practically zero, and the magnetic flux induces nothing in such a narrow
circuit, U , = 0. On the contrary, a wide circuit, made up of a conductor
and remote current line 2, is affected by most of the internal magnetic flux.

Figure 6.9. The conductors inside an elliptic cylinder.
The emf induced by the flux adds the ohmic voltage to the necessary value U .
The evaluation of the magnetic flux direction will show that the signs of U,
and U , coincide if the circuit, in which U , is induced, is composed by the
current line with a linear density less than the average value; otherwise, U,
and U, have opposite directions. Therefore, the values of U, and U, vary
with the design circuit chosen, but the sum remains the same.
We shall make use of this circumstance to find the time variation of the
magnetic field inside the sheath. It has been pointed out above that lightning
current i ( t ) acts for such a short time that it cannot be redistributed radically
along the sheath perimeter; therefore, we have J ( t ) w i ( t ) at any point. Hence,
we get Ein(t) i ( t ) for a t h n sheath where the slun effect is inessential. Choos-
N
ing a design circuit with U, = 0, we find U (t ) = U, ( t ) E,, ( t ) i( t ). But in

N N
the general case, this is U ( t ) = U e ( t )+ U M ( t ) ,with the values of U, and U ,

being comparable; hence, U , ( t ) i( t ).
N
Thus, the induction component of overvoltages in a sheath is pro-

portional to lightning current rather than to its derivative! Therefore, the
magnetic flux penetrating into the sheath varies in time as the integral of
current i( t ) . This remarkable result has been confirmed by experiments.
The oscillograms in figure 6.10 illustrate a test current impulse, similar in
shape to a lightning current impulse, and a magnetic pulse H ( t ) inside a
i
-OV
0.2 0.'4 0.'6 ' t, ms
Figure 6.10. Oscillograms of the test current and magnetic field inside the wing-like
sheath.

sheath simulating an aircraft wing [l 11. The response time of the magnetic
field detector did not exceed 0.5ps, so that the H ( t ) pulse front close to
300ps and an order of magnitude higher than the current impulse front
causes no doubt.
6.2.6 Overvoltage in a double wire circuit

Although the use of a metallic sheath as a reverse wire saves on metal, most
internal circuits of objects to be protected consist of two wires, because they
are better screened from noises. When the magnetic field inside an object is
zero, as is the case with a perfectly circular sheath, the lightning current
raises the potential of each wire relative to the shell but no overvoltage
arises between the wires. This is important because the electromagnetic
induction can damage the insulation and produce noises in information
transmission systems. The consequences of an information line disorder
are often as hazardous as a failure in an electronic unit.
It follows from the previous section that the magnetic induction emf
inside the sheath strongly depends on the wire location. The emf value is
maximal when one of the wires goes near the inner sheath surface along
the current line of maximum linear density and the other wire is immediately
adjacent to the line with J,,,. The overvoltages U1 and U, of the two wires
relative to the shell are determined only by the ohmic components, since
the wires immediately adjacent to the inner sheath produce with it zero
area circuits: U1 = J,,,pl/d and U2 = J,,,pl/d. The voltage between the
wires AU,,, = U1 - U 2 = (J,,, - J,,,)pl/d is due to the internal magnetic
field, so the variation rate of the magnetic flux penetrating through the circuit
composed of the wires is dQ,,/dt = (J,,, - Jmln)pl/d.We can again con-
clude that the magnetic field pulse in the sheath is not similar to the lightning
current but to its time integral. During a current impulse of a negative light-
ning tp = 100 ys (on the 0.5 level) with J,, = const, the magnetic field within
a non-circular sheath rises as H ( t ) t (for a circular sheath, Jmax
N = J,,, and
H = 0). At this lightning current, the higher the conductive sheath resistivity
and the greater the non-uniformity of the initial current distribution along
the sheath perimeter, the higher is the internal magnetic field.
To illustrate, an estimation will be made for an elliptical cylinder of
length I = 100 m. The following parameters will be used: a = 1 m, b = 1 cm,
the aluminium sheath thickness d = 1mm, and p = 3 x fl-m. The
lightning current amplitude will be taken to be 1, = 200kA, a value used
in aircraft tests for lightning resistance. Using formula (6.34), we obtain
J,, x 3200 kA/m, J,,, NN 32 kA/m, and AU,, 9.5 kV. In a real construc-
tion, such a great overvoltage could have resulted from a poor design of the
internal electrical network. Wires running to the same electronic unit must
not be separated so much from each other, nor should they be placed on the
inner side of a metallic shell at places differing much in the surface curvature

Lightning stroke at a screened object 29 1
and, hence, in the linear current density. A compact paclung of cable assem-
blies at sites of minimum surface curvature is a good and nearly free means
of limiting overvoltages in internal circuits of objects with metallic shells.
Overvoltages rise considerably if the shell is made from a plastic and if
its electric circuits are located in a special outer metallic jacket extending
from the head to the tail. The linear resistance of the jacket may be 1-2
orders higher than that of the totally metallic fuselage. The ohmic component
of overvoltage will increase respectively. To eliminate the magnetic compo-
nent, associated with the penetration of the magnetic field into the jacket,
it is very desirable to make it as a pipe with a circular cross section.
6.2.7 Laboratory tests of objects with metallic sheaths

The lightning protection practice involves a great many technical problems
associated with the formation of test current and the measurement of all
parameters of interest. Here, we shall be concerned with the physical aspects
of testing, which could allow prediction of the object's response to lightning
current in a real situation, generally different from laboratory conditions.
Let us begin with a laboratory current simulating lightning current. The
best thing to do would be to make a laboratory generator reproduce light-
ning current exactly. The high requirements on the protection reliability
make one apply maximum currents with an amplitude up to 200kA,
especially for testing aircraft. Tests on the 1 : 1 scale are attractive because
they do not require overvoltage measurements. It is sufficient to examine
the object's equipment after the tests to see that there is no damage. However,
the generation of a high current creates problems when the object has a long
length or when the current impulse front to be reproduced must be short. For
example, the maximum steepness for the impulse i( t ) = Zo [ 1 - exp(-pt)]
with an exponential front is (dildt),,, = pZo.To generate such impulses,
the source must develop the voltage U,,, =DIOL,where L is the circuit
inductance close to that of the test object; L z L1l for an object of length
1. The maximum voltage is U,, z 12MV for L1 x 1 pH/m, I x loom,
Io = 200 kA, and p x 0.6 ps-', corresponding to the front t f % 5 ps average
for the current of the first negative lightning component. A generator with
such parameters would have an enormous size and great cost.
The intuitive desire to elongate the current impulse front rather than to
reduce its amplitude in the testing of objects with a solid metallic sheath has a
reasonable physical basis. Due to the longer front duration tf, the ohmic
overvoltage in the internal circuits of the object to be designed could
change only when there is an appreciable current redistribution along its
cross section perimeter during the time t x tf. This would require the time
tf > 100 ,us. Therefore, the application of impulse fronts with a duration of
dozens of microseconds, instead of typical lightning impulses, cannot affect
the test results. The same is true of overvoltages in a double wire circuit,

induced by an internal magnetic flux. Consequently, the increase of the

impulse front duration by one order of magnitude will be unable to affect
appreciably the overvoltage in internal electrical circuits. This considerably
reduces the requirements on the laboratory source of impulse current,
because its operating voltage decreases in proportion with the increase in
tf. The decrease in U by an order of magnitude reduces the costs, because
the costs of high-voltage technologies rise faster than the actual voltage.
Much attention should be given to the simulation of the lightning
impulse duration in laboratory conditions. Anyway, the test impulse
should not be shorter than the real one, for the overvoltage amplitude may
thus be underestimated because of the skin-effect. It would be unreasonable
to reproduce on the test bed the actual amplitude of the lightning current
impulse if there are no non-linear elements in the test object’s circuit and
the overvoltages can be registered by detectors. Since the electrical and mag-
netic components of overvoltage are similar in shape and equally depend on
the applied current amplitude, one can recalculate the measurements in pro-
portion with higher currents and select the test impulse amplitude in terms of
the highest possible accuracy and registration convenience.
Quite another matter is the situation when the test object’s sheath is not
solid but has slits or technological windows. The ‘external’ magnetic field of
the lightning partly penetrates through the sheath; the field is proportional to
the current and the induced overvoltages are proportional to the current
impulse steepness. The total overvoltage now depends on both the current
rise time and amplitude, so the engineer has no chance to select a convenient
test impulse shape. In principle, the recalculation of measured pulses to real
ones is also possible, but this requires a detailed analysis of the overvoltage
mechanism and the responses of the object’s circuits, which does not raise the
testing reliability.
Another problem is to connect the test object to the laboratory
generator. It is obvious that a conductor with the generated current should
be connected to the site of a possible lightning stroke. In the case of terrestrial
objects in natural conditions and on a test bed, the problem of current output
is solved in a simple way - by using a grounding bus. The situation for
aircraft and spacecraft is more complicated. In real conditions, the lightning
current first flows through a metallic sheath (say, the fuselage) and then
enters the ascending leader channel, whose length is much greater than
that of the object. It is difficult to reproduce the real current path in
laboratory conditions - this would require a very high voltage to make the
impulse current run through the long conductor simulating a lightning
channel. Besides, the test object and the numerous detectors would be
under a very high potential relative to the earth.
The return current wire is normally located close to the test object. Its
magnetic field interacts with the object’s metallic sheath, through which
forward current flows. As a result, the current distribution along the sheath

Point number
01
1 2 3 4 1
Figure 6.11. Measured angular distributions of the linear current density along the
circular pipe perimeter at various locations of the reverse current conductor.
Marked points (on the pipe scheme) are presented on the abscissa axis. The curve
A I corresponds to a single reverse wire A for a = 2r, curve A2 is that for a = 4r,
curve B depicts three reverse wires B placed as shown in the scheme. Uniform
distribution C corresponds to the coaxial reverse current cylinder of radius 2r.
perimeter changes, the redistribution being considerable if the return current

wire is close to the object. Inducing the emf of the opposite sign, the reverse
current increases the current load in the nearby parts of the metallic sheath
but decreases it in the remote parts.
For a particular geometry, the current distribution should be found
numerically from the set of equations (6.33) by adding, to each equation, a
term for the magnetic flux from the reverse current wire, -ZMMko, where
Mko is the mutual inductance between the return current wire and the kth
conductor simulating the sheath. Under conditions typical of test beds, the
distortions due to the return current path may be appreciable. The results
presented in figure 6.11 have been obtained from the tests of a sheath
shaped as a circular pipe. In order to avoid the effects of currents induced
in the conductive soil, the sheath was raised above the earth at a height
H = 7 r , where r is the pipe radius. The role of the return current wire was
performed by a thin conductor running parallel to the pipe at the distances
a = 2r and a = 4r from it, three conductors located at 120" at the same
distance, and a coaxial cylinder of radius 2r. The latter design provides a
perfectly uniform current distribution along the perimeter of the sheath
cross section. The return current of a single wire distorts the current distribu-
tion to the greatest extent: its minimum linear density drops to 0.5ja,and the
maximum density rises to 2.3ja, (jav = Z0/27rr). The current distribution
becomes more uniform when the number of reverse conductors is increased.

When a sheath has a complex geometry, it is hard to predict the reverse

current effect on the test results. The current redistribution in the sheath may
lead to both the overestimation and underestimation of overvoltages in the
internal circuits. Much depends on the arrangement of the internal conduc-
tors and return current wire.
6.2.8 Overvoltage in a screened multilayer cable

Overvoltages in screened multilayer cables are due to the skin-effect. As a
result the cable wire screens in the layers are loaded differently by the light-
ning current. Every layer is formed by wires arranged in a circle and having
their own screens (figure 6.12(a)). Depending on the reliability requirements,
a cable may have an outer metallic sheath or a dielectric coating protecting it
from mechanical damage. However, a direct lightning stroke produces a
breakdown of dielectric material, and the lightning current is distributed
among the screens. The adjacent screens in a layer contact each other
along the whole cable length. It can be assumed in a first approximation
that they form a solid sheath of circular cross section with resistance
Rk = R/nk and inductance Lk, where R is the resistance of an individual
wire screen and nk is the number of screened wires in the kth layer
(figure 6.12(b)). For simplicity, we shall consider a double layer cable, mark-
ing the inner layer with k = 1 and the outer layer with k = 2. The adjacent
screens of wires from the adjacent layers are also in contact with one another.
Figure 6.12. Multilayer cable (a) and solid sheath model (b).

For this reason, a set of circular sheaths can be regarded as a solid conductor,
and the current penetration along its radius (from the second layer to the first
one) can be considered as a skin effect. Such a system can also be treated as a
set of discrete circular layers.
In the latter case, the current distribution among the layers at the initial
moment of time t = 0 can be found from the condition of magnetic flux
coupling equality (6.33). Equations (6.35) are valid at t > 0 and have the
following solution for two layers at constant current IM = il i2 = const:+
R2rM IM
[l - exp(-Xt)],
+ R2 [Rl + R2 exp(-Xt)l
i l ( t ) = ____ i2(t) =
R1+ R2 RI
~
(6.37)
where X = (R, + R2)/(L1 - L 2 ) . Equations (6.35) allowed for the mutual
inductance of the layers, M12= L2, as in the treatment of the screen-wire
system in section 6.2. In accordance with the skin-effect law, the lightning
current first loads the outer sheath and then gradually penetrates into the
inner sheath. The current is distributed uniformly between the individual
screens in each circular layer, iSl= il/nl and is2 = i2/n2.The overvoltage
across the insulation between a wire and its own screen (providing that the
skin-effect in an individual screen is neglected) is similar to the current in the
layer, U , ( t ) = Rlil(t) and U 2 ( t )= R2i2(t),but not to the lightning current.
If a double wire circuit uses the cores of one layer, there is no over-
voltage in the instruments connected to it, because the potentials of the
layer cores are identical. If the instruments are connected to the cores of
different layers, the voltage between them is
U12 = U2 - U1 = I M R eXp(-Xt).
~ (6.38)
At I M = 1, expression (6.38) is a unit step function for the set of equations
providing the solution for the lightning current impulse of an arbitrary
shape. In particular, at i(t) = IM[exp(-at) - exp(-Pt)], we have
U12= IMR2[Bexp(-Pt) - A exp(-at) - ( B - A ) exp(-At)]
(6.39)
A = ./(A - a), B = p/(X - p).
Owing to the relatively small value of L1 - L2 x (p0/27r) In (r2/r1) at close
layer radii r2 and r l , the layer current ratio is redistributed rapidly, for
T = A-' FZ l o p . This is the reason for a fast damping of the overvoltage
pulse U12(figure 6.13), which may be remarkably shorter than the current
impulse. It follows from (6.39) that the pulse U12 reverses the sign; its
opposite tail is damped approximately at the rate of lightning current reduc-
tion. The overvoltage amplitude in a double wire cable is close to that in a
wire-shell system, exactly as in a sheath with a sharply non-uniform current
distribution. If the screens are thin and have a high resistance, the hazard of
damaging the connected measuring instruments is fairly great.

-0.2 J
Figure 6.13. Overvoltage pulse on a two-layer cable for the bi-exponential current
impulse with cy = 0.007 ps, ,3 = 0.6 ps and the redistribution time constant T = 50 ps.
The problem for a multilayer cable can be solved in a similar way. The
overvoltages between the cable cores grow with distance between the respec-
tive layers. Other conditions being equal, the overvoltages drop with the layer
depth in the cable. The use of cores of one cable layer reduces considerably
the overvoltage in a double wire system but does not eliminate it entirely.
There are no perfectly circular cables - the cable is pressed under its own
weight and becomes deformed during its winding on a drum. The result is
that the current distribution along the sheath cross section perimeter
becomes non-uniform, producing additional overvoltages between the
cores of the same layer. To minimize these overvoltages, it is desirable to
connect the equipment to the adjacent cores of the same layer. High precision
equipment should be connected to the cores of deeper layers. Overvoltages
arising in a multilayer cable can be evaluated from the same set of
equations (6.35).
6.3 Metallic pipes as a high potential pathway
Modern constructions have an abundance of underground metallic pipes,

and the lightning protection engineer must take them into account as a
possible pathway for currents from remote lightning strokes. This actually
happens when a pipe lies close to a high lightning rod or another object
preferable to lightnings. Spreading through the earth away from the
grounding electrode in a way described in section 6.2.2, some of the current
enters a metallic pipe and runs along its length. A pipe is sometimes

Metallic pipes as a high potential pathway 291
Figure 6.14. Underground pipe as the pathway for a lightning current and the design
circuit for a simple evaluation of the object potential.
connected directly to the object grounding electrode. Figure 6.14 illustrates

the typical situation when a metallic pipe line connects the grounding
electrode (with grounding resistance R g l )of an object, struck by lightning,
to the grounding (with resistance Rg2)of a well-protected object. Although
the lightning is unable to reach the latter directly, some of the current finds
its way to its grounding electrode - the pipe. For applications, it is
important to know the dependence of this current on the line length 1 and
on the soil conductivity.
Section 6.2.3 considered the problem of current distribution for an under-
ground pipe of infinite length. The limited line length in the present case is an
important parameter, especially because it has the grounding resistances at its
ends. Generally, t h s problem can be solved analytically using the Laplace trans-
formation. But the final result is represented as a functional series too complex
for a treatment, so numerical computations are necessary. It is, therefore, more
expedient to solve this problem numerically from the very beginning. Before
presenting the results of a computer simulation, we shall make a simple evalua-
tion. Let us replace an underground pipe by the lumped inductance L = L 1I and
its intrinsic grounding resistance R, = (G1l)-'. The latter will be represented as
two identical resistors R = 2 R , by connecting them to the ends of the line in
parallel to the grounding resistances Rgl and Rg2of the objects it connects
(figure 6.14). This rough approximation makes sense, since we are interested
in the value of current i2 at the far end connected to the grounding mat,
rather than in its distribution along the line. In this approximation, we have
d i2 RR .
+
L-dt Re2i2 = (i - i 2 ) R e l , R . - 2, j = 1 , 2 .
R+Rd
(6.40)
Putting the lightning current to be i = ZM exp(-at) and i2(0) = 0, we find
i2(t)= Re 1AIM [exp(-at) - exp(-At)], X = Re1 + Re2

(Re1 + Re2)(A - a ) L
(6.41)

At the beginning, while the effect of self-induction emf is still noticeable, the
current largely flows through the equivalent resistance Rel at the front end of
the line. After time T = A-', the current gradually penetrates to the far end
of pipe. Some of it, i82 = i 2 R / ( R+ Re]!,finds its way to the grounding elec-
trode of the object of interest, raising its potential to the value U2 = ig2Rg1
relative to a remote point on the earth. For a longer line, the values of ig2
and U2 decrease for two reasons. An increase in L = L1l and G = G 1 l
raises the time constant T , and by the time the current has reached the far
end of the pipe, the initial lightning current is considerably damped. Besides,
a smaller portion of the current i2 that has reached the far end enters the
object's grounding electrode because of the greater pipe leakage. The depen-
dence of ig2 and U2 on 1 proves to be rather strong, especially when the
effective duration of the lightning current, t, x a - ' , is comparable with
T = A-'. Suppose we take t, = 100 ps on the 0.5 level ( a = 0.007 ps-'), the
grounding resistances Rgl = Rg2= 10 R, and L1 = 2.5 pH/m. A metallic
pipe with a lOcm diameter and lOOm in length, lying at the surface of the
soil with p = 200 R/m (G1 = 2.1 x (a/m)-', R = 9.7 R), will deliver
the current igz 0.171ZAvto the ground of the object located at its far end.
The object's potential will be raised to U , x 50kV at ZM = 30kA. At
I = 200m, we have ig2x 0.0861Zjw and, at the same lightning current,
U2 z 25 kV. But even this voltage is quite sufficient for a spark to be ignited
between closely located elements of two metallic structures, provided that
one of them is connected to the grounding electrode and the other is not.
Such a spark can induce an explosion or fire in explosible premises.
In low conductivity soils, current can be transported through metallic
pipes for many kilometres. This refers, to a still greater extent, to external
pipes and rails mounted on a trestle which are grounded only locally, through
the supports separated by dozens of metres. Here, evaluations can also be
made with expression (6.42), putting R = 2Ri/n, where RL is an average
resistance of the support grounding and n is the number of supports.
A comparison of the estimates and computations is shown in figure 6.15
for the above example with I = 200m. The estimates for the current
amplitude at the far end of the pipe and for the moment of maximum
current show a satisfactory agreement with the numerical computations.
The computations will be unnecessary if one finds it possible to ignore the
initial portion of the pulse front and can put up with a 20-25% error.
Let us calculate the potential at the far end of the pipe unconnected to
the grounding electrode at either end. This may happen due to careless design
or poor maintenance of communications lines. The soil will be considered to
have a low conductivity, p = lOOOQ/m; L1 = 2.5 pH,". The curves in
figure 6.16 show the variation in the voltage and current amplitude ratio
UmaX/ZAv for impulses of negative lightnings with tp = 100 ps and for those
of 'anomalous' positive lightnings, which are an order of magnitude
longer. The pipe is capable of delivering a potential of dozens of kilovolts

Metallic pipes as a high potential pathway 299
Time, ~s
Figure 6.15. Portion of a lightning current passed to the object through the communi-
cation pipe of 200 m length. Curve 1: numerical computation, 2: simple evaluation.
for a distance of 1 km to the object even at a moderate lightning current of

30 kA. Damage of the contact between the pipe and the object's grounding
electrode may be fatal if a spark arising in the air gap encounters an
inflammable substance.
20 -
f
E
0 15-
+-
J
.
10-
5-
0 200 400 600 800 1000

1, m
Figure 6.16. Computed maximum overvoltages transferred to an object at the far end
of the underground pipe of 10 cm diameter and of length 1. The pipe is not connected
with the grounding of both an object and a lightning rod. Computations were made for
the usual lightning current impulse of 100 ps duration, and for an 'anomalous' impulse
of 1000 ps, Lightning stroke to the other end of the pipe.

300 Dangerous lightning effects on modern stsuctuses
The delivery of high potential can be controlled in a simple way - all

communications lines must be connected to the same grounding mat. In
that case, the voltage of all mat components will be raised equally by the
brought current of a remote lightning stroke. It should be noted that this
is a reliable means to cope with the overvoltage of kilovolt values. A
simple connection of metallic sheaths to the grounding mat cannot remove
pulse noises of tens or hundreds of volts having a short rise time. Steep cur-
rent impulses spreading across the buses and components of the grounding
mat always create an induction emf, producing abrupt voltage changes
even in conductors of about l m in length. Electrical circuits must be
mounted in such a way as to avoid the appearance of closed contours or
joints of the conductor screens to points remote from each other in the
grounding mat. This sometimes becomes such a delicate matter that the
result depends on the engineer’s intuition rather than on exact knowledge,
6.4 Direct stroke overvoltage
We described the manifestations of overvoltage due to a direct lightning

stroke when discussing the lightning current propagation across a metallic
sheath. The highest current enters the sheath when a lightning discharge
strikes an object directly (section 6.2.1). This happens, for example, when
an aircraft is affected by the return stroke current recharging the descending
leader which has connected the aircraft to the earth. Below, we discuss a
direct lightning stroke at a grounded terrestrial object. Specifically, we
shall be interested in the voltage applied to the insulation of the object
relative to the earth or another construction located nearby. The classical
situation is that a voltage arises between the lightning rod that has
intercepted the lightning and the nearby object being protected. A rough
treatment of this situation was made in section 1.5.1. The fast variation of
a high lightning current i along the metallic parts of a construction raises
+
its potential by U = R,i Ldi/dt relative to a remote point on the earth.
Much depends on what is understood by the grounding resistance R, and
inductance L. These issues are discussed in much detail in the books on
direct stroke overvoltages (e.g., [6]).Here, we outline the most important
physical aspects of the problem.
6.4.1 The behaviour of a grounding electrode at high current impulses

An important parameter of a grounding electrode is the stationary grounding
resistance usually measured during the spread of direct or low frequency
alternative current of several amperes. The value of Rgo found from the
measurements may be several times larger or smaller than R, = Ue/ZM
corresponding to a rapidly varying kiloampere lightning current (here, U , is

Direct stroke overvoltage 301
the potential at the current input into the protector). We have discussed, at sev-
eral points in the book, the two physical mechanisms affecting differently the
ability of a metallic conductor to tap off the lightning current to the earth: the
self-inductance and ionization expansion of the surface contacting the soil.
The voltage drop across the inductance prevents current flow into the
conductor. A long conductor has to be treated as a line with distributed
parameters. The input resistance of the line, Ri, = U ( 0 ,t)/i(O:t ) varies in
time, since the current diffuses along the line, and it takes some time for
the whole conductor to be loaded by current more or less uniformly. As
the limiting case, consider an infinite conductor in a soil with resistivity p .
From formulae (6.21) and (6.22), the voltage at the conductor input is
U e ( t )z U ( 0 ,t ) t - 1 / 2 for the current i(0,t ) = const = lo and t > 0. At
N
Io = 1, formula (6.21) can be treated as a unit step function of the system,

y ( t ) . This allows us to follow the input voltage of a horizontal grounding
conductor at the lightning current i ( t ) with a real impulse front by using
the Duhamel-Carson integral:
U ( 0 ,t ) = y(t)i(O) + s:, y(T)i’(t - 7) d r .
For a impulse with an exponential front i(r ) = Io[ 1 - exp(- p t ) ] we have

(6.42)
U ( 0 ,t ) = 210 (g y 2 1 1 ( 3 t ) (6.43)
where h ( @ ) is a function given by the last integral in (6.29) and figure 6.5. Its
maximum h,, at pt, M 0.9 permits the calculation of the maximum voltage
drop across the grounding electrode:
(6.44)
The effective input resistance of an extended horizontal grounding electrode,
corresponding to U,,,, is expressed as
(6.45)
In contrast to a lumped grounding electrode with R, M p , the input resistance

of an extended one varies much less with the soil resistivity, R,,, pli2. N
Extended grounding electrodes are ineffective, because only a short initial

portion of their length , leffM (RgerG1)-’, is actually operative during the
impulse front time. For example, the effective resistance is R,,, M 13 R and
the effective length of a long grounding pipe with L1 = 2.5 pH/m at the
earth’s surface is leffx 22m in the case of the first component current of a
negative lightning with the rise time tf M 5ps ( p M 0 . 6 ~ ~ - I and )
-
p = 100 R m. In a soil with an order of magnitude lower conductivity, the
respective values are R,, M 42 R and leffM 75 m.

Extending the grounding bus beyond the limit leff,we are still unable to
reduce appreciably the maximum voltage drop across the bus. For this
reason, it is better to introduce current at the centre of a long bus rather
than at its end, such that two current waves would run in opposite directions
along the half-length conductors. Still more effective are three conductors
arranged at an angle of 120”, and so on. When a grounding mat with the
lowest possible value of R,, is desired, it is preferable to load, more or less
uniformly, the whole of the adjacent soil volume. For this aim, a set of
horizontal conductors or a conductor network is combined with vertical
rod electrodes. To avoid the interaction effect of the grounding elements
and to achieve the maximum loading of them by current, the distance
between the elements should be made comparable with their length (or
with the height, for vertical rods). But even in that case, only part of the
grounding mat, within the radius of leEfrom the current input, will operate
effectively at the impulse front.
Thus, the resistance of a grounding electrode for rapidly varying cur-
rents is much higher than for direct current. A grounding mat network
with numerous horizontal buses and vertical rods is able to reduce the effec-
tive resistance to the value of R,,, x 1 0. But when a large number of objects
is being constructed, for example, the towers of a power transmission line,
one has to deal with resistances as high as R,, x 10 R and more.
Laboratory experiments show that the grounding resistance of an
electrode delivering to the earth very high currents is lower than for low
currents. The grounding resistance decreases with the current rise. The
grounding resistance ratio of a high impulsed current and low direct current,
cui = R,/Rpo, is often called the impulse coefficient of a grounding. The
coefficients ai used in the literature are sometimes as small as aix 0.1. To
illustrate, we shall cite the generalized function cui =f(plM) which has
been suggested for a vertical rod of 2.5m in length from the results of
small-scale laboratory experiments [7] (figure 6.17). The grounding resistance
-
is reduced by a factor of four at p = 1000 R m and IM = 30 kA.
In principle, this reduction in resistance might be due to a larger effective
radius of the grounding electrode because of the soil air ionization. In section
6.2.2, we gave formula (6.15) for the linear conductivity of a long rod lying on
the earth with one half of its surface contacting the soil. If the rod is fixed in
the vertical position, the whole of its surface contacts the soil but its leakage
conductivity is lower by a little less than a factor of 2 at the same length (due
to the poorer operation of the upper end of the rod located at the earth’s
surface, because current cannot flow upward into the air). The linear conduc-
tivity GI and the grounding resistance R, of a rod of radius ro, fixed vertically
into the earth for a length 1, are
(6.46)

ai I
1.0’
0.8- 2.5 m
0.6-
0.4.
0.2 -
PI, M V m
0, 1 I
Figure 6.17. Impulse coefficient for the grounding rod of 2.5 m length.
To reduce R, by a factor of 4 at the initial rod radius ro = 1 cm and 1 = 2.5 m,

the radius must be increased to r1 = 105 cm. The field at the ionized volume
boundary must exceed the ionization threshold in the soil, Eigx 10 kV/cm,
and the current density at p = IOOOR/m must be j = Eig p = 1 kA/m2.
For the surface area of the ionized volume S M 27rrll + ,,f
= 24m2, the
total leakage current would be Z = j S = 24 kA, corresponding to the current
of a moderate lightning power.
However, the uniform radial ionization expansion of the initial ground-
ing volume at a rate r l / t f = 2 x 105m/s (this process must be completed
within the rise time of the current impulse, tf = 5ps) can hardly occur in
reality. Anyway, there is no experimental indication for this. More probable
would be the rod ‘elongation’ owing to the leader development into the soil,
because the current density and the field at the rod end are higher than at its
lateral surface. The elongation of a grounding electrode is a more effective
means of reducing the grounding resistance R,,because of R, 1/1, since -
the resistance decreases only logarithmically with increasing radius (but
only at r << 1). However, even this process seems unlikely. There is no
experimental evidence for the existence of long leaders in the soil bulk.
The leakage area of a grounding electrode is likely to increase due to the
elongation of the leader creeping along the soil surface from an element of the
grounding mat. The grounding resistance will then decrease, as 1/L at a long
leader length L. This mechanism, observed in model laboratory experiments,
seems optimal for a natural reduction in R, due to the lightning current. To
change appreciably the grounding resistance of a typical lightning protector,
the leader must grow to L M 10 m in length (the total length of the protector
electrodes) for the time t f M 5 ps of the lightning current rise. For this to
happen, the leader must elongate at a rate of 2 x lo6 mjs. A creeping leader

develops in the air adjacent to the surface of a conductive soil and is shown by
laboratory experiments to be devoid of a streamer zone and a charge sheath
-
(section 6.2.2). In t h s respect, it is similar to a dart leader whch develops a
velocity of about lo7 mls at current i 1 kA. Assuming that the development
of a fast leader along the soil surface requires this current in the leader tip, it, let
us estimate the tip radius, at which this appears possible.
If the grounding resistance is R, and the lightning current is I,, the leader
is supported by the voltage U zz R,IM applied to its base. The tip possesses
approximately the same potential, because the leader channel is a good
conductor and not a large part of the voltage drops across it. The resistance
of the leakage current from a hemispherical tip is equal, from formula
(6.14), to R, % p/27rrt. The tip current is it U / R , . Only part of the lightning
N
current enters the channel, Io. This current mostly leaks into the soil through
the lateral channel surface possessing, according to (6.15), a leakage resistance
RI, = (G,L)-' = pln(L/ro)/rL. Keeping in mind U Rl,Zo. we obtain the
N
formula to be used for the estimation of the tip radius:
(6.47)
The 6cm radius obtained at Io = IAw/2= 15 kA, it = 1 kA, L =10m, and
-
yo zz rt, appears to be quite reasonable. The field at the channel lateral surface
behind the tip, E pZo/*irr,L 80 kV:cm, is high enough for the ionization
expansion of the leader channel to occur there. Radii much larger than
those of a conventional leader in air under similar conditions have been
registered for laboratory leaders creeping along the soil. The photographs
in figure 6.18 illustrate this quite clearly.
To conclude, the spread of high lightning currents reduces the ground-
ing resistance, probably due to the excitation of one or several leaders
creeping along the soil surface, thereby increasing the length of the ground-
ing electrodes. But for a fast leader growth (otherwise, the leakage surface
has no time to become larger for the short lightning rise time), a high
current of about 1 kA must be delivered to the leader tip. This restricts
the process of grounding resistance reduction by the condition under
which the electrodes are arranged in compact groups. A fast reduction is
hardly possible for a modern substation having an extended grounding
network. The reduction is, however, quite possible in the case of a con-
centrated protector consisting of 2-3 horizontal conductors or several
vertical rods.
It is worth saying a few words about the testing of lightning grounding.
The great complexity of a large-scale simulation of lightning current makes
one turn to model tests. in which the surface current density of small
electrodes is preserved while the total current is reduced manifold. The
laboratory studies indicates that the similarity laws are invalid for the
leader process. The questions of how to interpret the small-scale simulation

Figure 6.18. Still photographs of leader creeping along the soil during its develop-
ment ( a ) and at the moment of gap bridging ( h ) .
results and how well they reproduce the real process of lightning current
spread are open to speculation.
6.4.2 Induction emf in an affected object

Let us consider a descending lightning stroke at an object of height h. The
induction emf for the object is proportional to its inductance, L = Llh.
The linear inductance can be estimated in a simple way, assuming that the
current fills up a conductor composed of the object and the lightning channel
of length 1 >> h. If we assign to the conductor a constant radius yo << I and
assume a perfectly conducting soil, we shall obtain L1 M (pO/27r)In ( 2 1 / r 0 ) .
This value will be L1 M 2.3 pH/m at 1 M H = 3 km ( H is the altitude of the
negative cloud charge centre) and ro = 5cm. Actually, the return stroke
wave covers a much smaller distance during the time of the current impulse
rise, when di/dt and the induction emf have maximum values. But owing to
the logarithmic dependence of L , on the geometrical size of a long conductor,
the change in the length of the lightning channel filled by current will affect
but little the value of L1. For example, we obtain 1 = w,tf = 500m and
L1 M 2pH/m for the return stroke velocity w, = IO8 m/s and tf =5 ps,
corresponding to the first component of a negative lightning. At t f = 1 ps
(the rise time of the subsequent component), L1 will be only 20% lower.
The same result is obtained when one uses the vector potential A([) and

vortex electric field EM = -aA/dt for the calculations. Suppose the current
rises linearly with a distance from the current wave front, i ( x ) = b(xf - x),
where xf = v,t and b = const. The current at the point x of the channel
rises linearly with time, Ai = ai/& = bur. Neglecting the delay time, as was
done in (6.9) and (6.1 l), we have
where ro is the average radius of the object affected by lightning. One can see
that the formula for L1 in (6.48) coincides with the one above, provided 1 is
understood as the length of the channel loaded by current by the moment of
time t .
If the finite velocity of an electromagnetic signal is taken into account
and the object is located directly under the lightning channel, which happens
in the case of a direct stroke, the evaluations made in section 6.1.1 give
Once again, we should like to emphasize the small contribution of the delay:
the logarithms in formulae (6.48) and (6.49) differ less than by 3 % at 3 = 0.3,
tf = 5 ps, and yo M 1 m.
Expressions (6.48) and (6.49) define rigorously the vortex electric field
EM at the earth's surface. When the object's height is h << urtf, which is
valid for many practical situations, the variation in EM along the object
can be ignored and the induction emf is U , = EMh.The emf rises linearly
with increasing h. In particular, if we have h = 30m, w, = 0 . 3 ~and the
lightning current rises to the amplitude IM = 100 kA for the time tf = 5 ps,
the maximum value of the induction component of the voltage at yo = 1 m
is U.Mm,,= 780 kV, a value comparable with the electrical component
Uem,, = R,ZM M 1000 kV at R, M 10 R. Of course, the effects of the electrical
component may be more serious because of its longer action. Indeed, in the
first approximation, the pulse Ue(t) is similar in shape to the lightning current
impulse and U,w(t) to its time derivative.
Formula (6.49) can also be used to evaluate the magnetic component
after the current impulse maximum. For this, the real current entering the
lightning channel should be represented as a sum of the two components:
i, = Ait, i2 = -A,(t - tf), and i2 = 0 at t < t f . It is not surprising that U ,
is non-zero behind the impulse front, since the magnetic field continues to
grow, as the lightning channel is filled by current. The total overvoltage
pulse of a direct stroke, U d ( t )= U,(t) + U M ( t ) ,is very different from the

2.0 -
2.01
1.5-
$1.0-
8 .
0.5 -
0 .40 t .
0.0 1 . , q. , . , , 1
I
0 2 4 6 8 10
Time, ks
Figure 6.19. Computed overvoltage of the direct stroke at the transmission line tower
with the grounding resistance R, = 10 s2 for the lightning current with tf = 5 ps and
amplitude of 100kA; U, = 0 . 3 ~ .
lightning current impulse because of an abrupt rise and an equally abrupt fall
of U , with time (figure 6.19).
6.4.3 Voltage between the affected and neighbouring objects

It is important for many applications to know the voltage affecting the
insulation gap between an object of height h, affected by a direct lightning
stroke, and another object of height hl < h, located nearby. For this, one
should find the difference between the evaluated overvoltage of the direct
stroke, Ud,,,, and the maximum overvoltage Uinmax,induced on the
neighbouring object. The latter value strongly depends on the object’s
construction. So, let us analyse two extrema1 situations.
Suppose a lightning strikes a lightning rod located near the mast it
protects (figure 6.20). The magnetic components of the overvoltages are
determined by the vortex field strengths EM from formula (6.49). For the
rod, yo can be taken to be equal to its average radius rl. For the mast, yo
can be assumed to be equal to the distance d between the rod and the
mast. The maximum time-dependent difference between the magnetic
components of the voltage at the height h l , where the distance between the
constructions is minimal, is equal to
(6.50)
Expression (6.50) allows for the value v,t >> d at the moment this maximum
occurs. The magnetic component of the overvoltage across the insulation

Figure 6.20. Estimation the voltage between a lightning rod struck and an object.
gap AUMm,,= UMrOd - U,wob,increases with distance d, because UMrOd is

independent of d and drops as the distance between the object and
the current increases. The return stroke velocity v, has practically no effect
on AUMm,,. Its upper limit is the value of for the affected lightning
rod (AUMm,,M UMm,,/2at d / r l RZ 100).
The situation with the electrical component of overvoltage is less definite.
The overvoltage is also determined by the difference between the two values,
AU, = Uerd - Ueobl,but Uerd = -Rg,IM is a definite quantity and Ueobl
varies with the design of the object’s grounding mat. The latter may be
common with the lightning rod grounding grid and quite compact; in that
case, we have A U , = 0 because the bases of the rod and the object are
interconnected. There may be another extreme situation: the grounding mat
of the object may be so far from that of the lightning rod that it may be
unaffected by the electric field of the lightning current spreading through the
soil. In that case, we shall not have UeOb,= 0 and AU,” = Rg,IM,because
this would be possible only in the absence of current through the object’s
grounding mat. In reality, there is an electric charge induced on the object’s
surface due to the electrostatic induction (section 6.1. l), so a current flows
across the object, creating the electrical component of the overvoltage. Its
value can be found from formula (6.5) and the maximum value from (6.6),
provided that the return stroke is simulated by a rectangular current wave
in a vertical lightning channel. Let us evaluate the possible voltage from
formula (6.6).
To go beyond the zone of the current spread away from the lighting rod
grounding, it is necessary to move away at a distance -20m from it. The
radius of the grounding grid, within which the electrodes are located, is
hardly larger than 5m, so that the distance between the rod and the

object in formula (6.6) can be taken to be r = 25m. Assuming that the height
of a typical object is h = 30m, its linear capacitance C1 z 10pF/m,
C = Clh = 300pF, and R , = l o a 2 ,we shall have UeDb, 30kV at the light-
ning current Z,v = 30 kA. Although Uerodand Ueobihave different signs and
iAU,I > IUerodI,the additional value is not essential because it is an order
of magnitude less than R,ZM in the above example. The situation when a
lightning rod is put up at a distance sufficient for the separation of its own
grounding grid and that of the object is quite realistic. This is done for the
protection of especially important constructions to avoid pulse noises or
sparking due to the induction emf, when some of the current finds its way
to the object’s grounding through the soil.
Another extreme case, in which the electrical component of the object
overvoltage is dominant, is a lightning stroke at a metallic grounded tower
of a power transmission line. The direct stroke overvoltage affects an
insulator string, to which a power wire is suspended. Consider first a
simple and frequent variant (in lines with an operation voltage below
1lOkV) when the line has no protecting wire. In that case, we do not have
to solve the difficult problem of lightning current distribution between the
affected tower and the wire, repeatedly grounded by the adjacent towers.
Nor should we bother about the electromagnetic effect of the protecting
wire on the power wire (section 6.4.4). As in the previous situation, the
insulator string is affected by the overvoltage A U equal to the potential
difference of the tower at the point of the string suspension and the power
wire. The calculation of the tower overvoltage is similar to that for a light-
ning rod, just described. A specific feature of this problem is the existence
of the wire. Being suspended horizontally, it does not respond to the mag-
netic field of the current in the lightning vertical channel. The power wire
is well insulated from the tower grounding by the insulator string. Owing
to its far end being grounded, it would be able to maintain zero potential,
but for the current created by the redistribution of the charge induced on
the wire. The induced charge is very high because some of the wire length
is located close to the lightning channel, and the total capacitance of a
long wire is very large. Naturally, the small distance between the wire and
the lightning channel does not mean the existence of a direct contact between
them, so we can speak only of the effect of electrical induction on the wire.
Even though the wire is connected to the earth at zero resistance, the
induced charge cannot respond immediately to the lightning charge variation
and the wire potential cannot remain at zero. The grounding point is located
far away, at the end of the wire, so the charge liberated by the induction
cannot be delivered to it faster than with light velocity c. For the induced
charge q,n to appear at the point x, a current wave must be excited at this
point, which will eventually transport the charge -q,n out of the wire to
the earth. This wave will propagate at light velocity. During its motion,
the potential at the wave front will rise due the voltage drop on the wave

resistance of the line with distributed parameters, i.e., along a long wire.
Elementary current and potential waves arise at any point on the wire,
where the induced charge is changed by the lightning field. Propagating
with light velocity to the left and to the right of the origin, the currents of
elementary waves are summed, raising the voltage between the wire and
the earth. After the waves are damped, this voltage, naturally, drops to
zero, because the wire is grounded. The response of a long line to the external
field Eo,(x, t ) acting along a horizontally suspended wire is described by the
equations
di di
aue = R l i + L l - - E o , ( x , t ) ,
-- --= C aue
1 T (6.51)
dx at dx
where the potential U e ( x ,t ) is due exclusively to the line response to the
field Eo,(x, t ) . The total potential of the wire relative to the earth, U p e ( xt,) =
U o ( x ,t ) + U e ( x ,t ) , contains another component, Uo(x,t ) , defined by the
charges of the lightning return stroke. Neglecting the ohmic voltage drop
relative to the induction term and taking b’Uo/b’x = -Eox into account, we
arrive at the wave equation with a distributed driving force and containing
no damping term:
The solution to this equation represents a general solution to a homogeneous

equation and a particular solution to an inhomogeneous one, corresponding
zIo
to the two identical waves propagating in opposite directions along the line:
= 1 ‘ d 8O U o ( X l , O ) d O +
Uge(x,t)
(6.53)
XI = x - C ( t - 0 ) . x,= x + c(t - 0 ) .
The integrals give the sum of the above elementary waves moving at light
velocity. The waves are excited by the time variation of the external field
potential U,. For the elementary wave to arrive at the point x at the
moment of time t , the causative variation in U0 must occur at the points
x f Ax earlier, by the time 0 = A x / c . If the time is counted from the
moment of the lightning contact with the line tower, the lower, zero limit
of the integrals of (6.53) should be replaced by the time-of-flight of light
for a minimum distance between the lightning channel and the wire,
In the general case, the difficulties that arise in the calculation of the
integrals depend on how one approximates the lightning current related to
the linear charge in the return stroke wave inducing the field Eo, as well as
on the lightning channel position relative to the wire. Of significance are
the following factors: what object the lightning strikes (the earth or an
element of a power transmission line raised above the ground), the channel

deviation from the normal, as well as its bendings and branching. It is impos-
sible to solve this problem without numerical computations. The question
then arises as to the stage in the study, at which a computer simulation is
most helpful. One should not ignore a numerical integration of initial
equations (6.51), allowing the control of the effect of active resistance R I
which sometimes has a large value. The effective value of R Imay be much
higher than the resistance of the line wire to direct current because of the
skin-effect, the soil resistance, used by the wave as a ‘return wire’, and due
to the energy consumed by the impulse corona. The corona is excited in
the wire by overvoltages and absorbs some of the propagating wave
energy, contributing to its damping. The impulse corona also increases the
effective linear capacitance of the wire, since the electrical charge is localized
not only on the wire surface but in the adjacent air. The charge is delivered
there by streamers of metre lengths. The capacitance CleRdepending on the
local wire overvoltage varies together with the velocity of perturbations in
the wire, U = (C1,,L1)-1/2.This greatly distorts the wave front, since different
sections of the wave front have different velocities. The problem becomes
greatly non-linear and definitely requires a numerical solution.
The calculation formulas given below describe simple situations
neglecting the wave damping in the wire. They have been derived by direct
integration of (6.53) and borrowed from [3]. The lightning channel is
considered to be vertical; the return stroke wave moves towards the cloud
at constant velocity U,. For a rectangular charge wave in the channel of
lightning that has struck the earth (but not a line tower) at a horizontal
distance r from the wire, we obtain for the wire point nearest to the lightning
channel
where p = ur/c, U, is the return stroke velocity, and h is the wire height above
the earth. The time in (6.54) is counted from the moment of the lightning
channel contact with the earth. This formula can be used at t > r / c , i.e.,
after the electromagnetic signal has covered the distance between the channel
and the wire. The overvoltage is still active at a large distance from the stroke
point (x + CQ), where the lightning field effect is negligible, Eox x 0. The
wave reaches that point through the wire, as in the case of a communications
line (this occurs without damping at R I= 0). Such overvoltage waves are
known as wandering waves. For these waves, we have
(6.55)
where the time is counted from the moment of the wave front arrival at the
‘infinitely’ remote point of interest. At t , = r/wr, the function - U g e ( m , t ) has

Figure 6.21. Evolution of an overvoltage at the wire point nearest to the lightning
stroke point (solid curves) and a wandering wave voltage (dashed curve). For a
return stroke, a rectangular current wave model is used.
a maximum:
(6.56)
which is independent of q..At t >> t,, the overvoltage is damped as f-’.

Typically, the amplitude of a wandering wave is somewhat higher than
the voltage amplitude relative to the earth at the site where the wire is closest
to the stroke point (figure 6.21). The reason for this is the opposite signs of U.
+
and U,, causing a reduction in the value of Uge= U, U, in the close
vicinity of the wire, where U. # 0. Far from the stroke point, we have
U , M 0, and the overvoltage is totally defined by the wire response. Although
the overvoltage maximum at the closest point does vary with wr, this
variation is not appreciable. This is good because there are few measurements
of the return stroke velocity and practically no synchronized measurements
of the lightning current.
If the lightning current is supposed to rise at the impulse front as
i(t) = A,t with A , = const, the overvoltage U,,(O. t ) , for the same conditions
and designations as in (6.54), is
( 2+ 1 - 32)”2- 3 K
Uge(0.t ) = -- In (6.57)
27r&oCWu, 1 - 32
+ +
Formula (6.57) has a sense at r / c 6 t 6 t f r / c ( 3 < K < ( t + t f / r ) 3). After
the current impulse maximum, t > tf, the calculation can be made using this
formula and a superpositior. by representing a real current wave as two

waves of different signs shifted in time, as was done in the comments on

formula (6.49). As the time increases within the lightning current rise time,
the value of Ug,(O,t ) rises monotonically. The calculated pulses V,,(O, t)
for a rectangular current wave and for a wave with a linearly rising current
have something in common at r/v, x tf, which is valid for remote lightning
strokes with r 2 100m. Both approximations lead to an increase in the over-
voltage during the current front. But for close strokes, especially for a direct
stroke at a line tower, the discrepancy between the calculated pulses becomes
remarkable. This is the reason for the sceptical attitude to analytical solu-
tions, which we showed at the beginning of the discussion. No doubt, a
linearly rising current is closer to reality than a rectangular impulse, but it
cannot simulate the actual current rise accurately. The same is true of the
impulse amplitude. The discrepancy in the calculations made within the
models considered grows with decreasing distance r .
Nevertheless, another analytical solution [3] may be useful for the estima-
tions. It concerns the case of a direct lightning stroke at a transmission line
tower, when the shortest distance between the lightning channel and the
wire is determined only by the height difference between the tower, h,, and
the wire, h. If a charge wave corresponding to the lightning current front
i(t) = Air moves up along the vertical lightning channel from the tower top
to the cloud with constant velocity w,, we shall have at the point of the wire
suspension
Let us calculate the overvoltage due to the first component of a negative

lightning with the average parameters IM = 30 kA, tf = 5 ps, Ai = 6 kA/ps,
and w, = 0 . 3 = ~ 90m/ps. We shall have U,,(O, tf) = 320kV for ho = 30m
and h = 20m at t = tf. Similar, but of the opposite sign, is the potential
rise on the tower grounding resistance R, x 10 R due to the lightning current,
RgIM =300 kV. This doubles the electrical component of the overvoltage
across the insulator string.
It is worth noting the specific effect of overvoltage on the line insulation.
Overvoltage is not strictly related to any point on the line, as is the case with
the voltage drop across the tower grounding. It has been mentioned that the
charge liberated by electrical induction moves along the wire, creating a
wandering overvoltage wave. With a negligible damping, it can cover a
distance of several kilometres, affecting, on its way, all the insulator strings it
encounters. An insulation breakdown may occur even far from the lightning
stroke, where the line insulation is poor for some reason. Really hazardous is
the encounter of the wandering wave with a hgh-voltage substation, because
the overvoltage wave penetrates to the internal insulation of its transformers
and generators, which is always poorer than the external insulation.

A wandering wave also arises when a lightning strikes a power wire. The
lightning current spreads along the wire from the stroke point in both
directions, producing very strong overvoltage waves, U ( x ,t ) = Z i ( x , t ) / 2 .
Since the wave resistance is 2 x 250-400R (the smaller value is typical of
ultrahigh voltage lines with split wires of a large equivalent radius), the
current ZM = 30 kA would produce an overvoltage with an amplitude of
3.5-6MV. In reality, the overvoltage is limited to the value of breakdown
voltage for the tower insulator string closest to the lightning stroke, where
the flashover does occur, A wave with an amplitude equal to the string break-
down voltage is a wandering wave in this case. Of course, the overvoltage
may rise again, after the string flashover, due to the self-induction emf of
the tower and to the voltage drop across its grounding, to which the lightning
current runs after the string flashover. Wandering waves are damped by the
same processes that determine the resistance RI in (6.51).
It is important for lightning protection practice to compare the over-
voltages due to direct strokes at a line tower and a wire. In the former
case, the voltage drop across the insulator string is the sum of three
components. The voltage drop across the tower grounding and the induced
voltage of the wire are approximately the same quantitatively but have
opposite signs. This totally gives about -2R,ZM over the string. The mag-
netic component L,Ai has a real effect only on the current impulse front,
and its average value is equal to L,Z,w/tf (L, is the tower self-inductance).
The magnetic component for the first leader of a moderate negative lightning
(ZM = 30 kA, t f = 5 ps) and for a tower of standard size (h, M 30 m) does not
exceed 200 kV but 2R,Iy > 600 kV because of R, 2 10 Cl. It appears that
overvoltages due to a direct moderate stroke at a tower can flashover the
insulation only in lines with voltages less than llOkV, which have strings
less than 1 m in length. For a 220 kV transmission line, a hazard may arise
when the currents are twice as high as the average value, but such lightnings
occur only with a 10% frequency. The hazard of a lightning stroke at a tower
is not high for 500-750 kV transmission lines, since they have long strings. A
reverse flashover may arise from a lightning with 100 kA currents and more,
but their number is less than 1YOof the total. If the lightning current strikes
the wire, the current spreads in both directions along it. With the wave resis-
tance Z > 200 52, we get ZZM/2> 3 MV even for a moderate lightning. This
is sufficient to flashover the insulation of any of the currently operating lines.
A lightning stroke at a wire should always be considered to be hazardous.
6.4.4 Lines with overhead ground-wires

When a lightning strikes a tower of a power transmission line protected by a
grounded wire, the current is split between the tower and the grounded wire,
due to which the current load on the tower is reduced. However, the engineer
is then faced with a complex problem of calculating the current distribution.

Another aspect of this problem is the account of the screening effect of a

protecting wire. Since the wire is connected to the tower, it acquires, in a
first approximation, the tower potential, thus creating a voltage wave of
the same sign. Owing to the electromagnetic induction, a similar wave of a
lower amplitude is excited in the power wire. As a result, the voltage in the
insulator string, equal to the potential difference between the tower and
the power wire, drops. These additional problems complicate the calculation
of direct stroke overvoltage for a line with an overhead ground-wire. The
problems that arise here relate to electric circuit theory rather than to
physics, so we shall discuss them only briefly.
Many engineers try to calculate the current distribution between a tower
and an overhead wire within the model of an equivalent circuit with concen-
trated parameters. The lightning channel is regarded as a source of current
i(t). The tower is replaced by its inductance L, and grounding resistance R,,
the two grounding wire branches (on the left and on the right of the stroke
point) are represented by the branch inductances L,/2 and their grounding
resistances in the adjacent towers, R,/2, connected in parallel. One also intro-
duces the mutual-induction emf M , dildt, induced by the lightning current in
the wire-towers-earth circuit (figure 6.22). This circuit can be simplified
further by putting R, = 0, because the principal interest is focused on the
current front of tf = 1-10 ps and because the cable inductance along the
many hundreds of metres of its length is as high as hundreds of microhenries
and the time constant is usually taken to be L J R , > 100 ps. This model circuit
then presents no calculation problems, provided that the mutual inductance of
the vertical lightning channel and the circuit including the ground-wire is
known. As the return stroke wave moves up, the channel is filled by current
so that the value of M , rises in time. The calculations similar to those for
the derivation of formula (6.49) and allowing for the time delay yield [3]
(6.59)
Figure 6.22. The design circuit for a current a tower of the line with the protective
cable.

The considerable simplification of the real process can be justified only at low
grounding resistances of the towers, when the current in the wire circuit is
limited mostly by its inductance, and one can neglect the current branching
off to the grounding resistances of all other towers except the one nearest to
the affected tower.
The value of R, in a real transmission line in areas with low conductivity
soils may be several times higher than the normal value, reaching 100 0. Then
the current distribution problem must also take into account the removal of
some of the lightning current to 2-5 towers away from the stroke point. The
equivalent circuit becomes more complicated (chain-like), representing a
series of link circuits identical to the first one. For a more rigorous solution,
the ground-wire is to be considered as a long line with a wave resistance Z,
and many local non-uniformities produced where the ground-wire contacts a
tower. Each tower is then represented as a chain of L, and R, connected in
series. Figure 6.23 illustrates the variation of the current impulse in the
tower with the design circuit. For a circuit with lumped parameters, the neglect
of the tower grounding resistance R, x 10 R does not affect the result much,
while at R, x 1000, the tower current impulse shape changes radically. A
circuit with distributed parameters permits one to follow the effect of con-
secutive wave reflections at the contacts between the ground-wire and the
towers. The current impulse distortion by the reflected waves is especially
Time, ps
Figure 6.23. Current impulse on the struck tower with (solid curves) and without
(dashed curves) allowance for a grounding resistance of the nearest tower. The
model with a linearly raising current front is used for return stroke.

appreciable for a short impulse front tf characteristic of the subsequent light-

ning components. As the linear resistance of the line, R I ,rises, the effect of
the reflections becomes less pronounced because the reflected waves are
damped more strongly. In any case, the overhead wire takes some of the
lightning current away, thereby unloading the affected tower; this current
fraction cannot be less than 2R,/Z,.
Let us evaluate the screening effect of the protecting wire, which also
reduces the direct stroke overvoltage. Engineers had become aware of this
effect long before overhead wires were used as lightning protectors. Some
even supposed that a wire could reduce the voltage of an insulator string
to a value lower than the flashover voltage. The wire acquires the electrical
potential of the tower, which has increased by the value of the voltage
drop across the grounding resistance. As a result, a high-voltage wave runs
along the wire. The nearby power wire finds itself in its electromagnetic
field inducing a similar wave. If U, is the voltage wave amplitude in the
ground-wire, the voltage produced in the power wire is Ucoup= kcoupUc,
where kcoup= ZCw/Z,is a coupling coefficient and Z,, is the wave resistance
of the grounded wire-power wire system whch can also be regarded as a long
line. We have Z,, = ( L ~ c w / C l ~ )by
1 ’definition.
2 The linear inductance L1,
and the capacitance Clcw between this two wires are calculated in a
conventional way, with the allowance for the earth’s effect. With Llcw
+ - N
ln[(ho h ) / ( h o- h ) ] and Clcw {ln[(ho+ h ) / ( h o- h ) ] } - ’ , the coupling

coefficient is
(6.60)
where Y, is the ground-wire radius. For a rigorous calculation, the geometri-

cal radius in (6.60) should be replaced by an equivalent radius of the space
charge region at the wire, (the space charge is incorporated by a impulse
corona under the action of high voltage). This somewhat increases the
value of kcoup.Measurements give approximately kcoup= 0.25, instead of
the calculated ‘geometrical’ value of kcoupz 0.2. Therefore, the electrical
component of power wire overvoltage is reduced once more, this time by
the value Ucoup= kcoupUc.The total overvoltage reduction owing to the
ground-wire makes up several dozens percent, decreasing the tower-stroke
effect on the transmission line insulation.
We should like to mention a certain relationship between the type of
lightning action and the transmission line cut-off. Even induced overvoltages
are hazardous for low voltage lines (primarily those of 0.4- 10 kV). Induced
overvoltages are much more frequent than direct strokes and are the main
reason for the line cut-offs. A protecting wire is useless in this case, so low
voltage transmission lines do not have it at all. For a line of 35kV or
more, induced overvoltages are practically harmless and direct lightning
strokes are dominant. The favourable effect of an overhead grounded wire

becomes apparent at an operating voltage U 2 110 kV, when the lightning

current leading to the insulation flashover after a stroke at a line tower
exceeds an average value ~ 3 kA.0 About 50% of cut-offs for 110-220 kV
lines equipped with a grounded wire are due to strokes at towers and 50%
of cut-offs occur when lightning breaks through to get to the power wire.
Beginning with 500 kV, an increasing number of cut-offs are due to lightning
breakthroughs to the power wire.
6.5 Concluding remarks
We finish this chapter and the book by describing the lightning effect on
power transmission lines. Scientists are still unable to offer a clear
mathematical description of its complicated mechanism. Modern computer
simulations can infinitely specify and refine a mathematical model of the
lightning effect, with respect to both the electromagnetic field and the object’s
response to a stroke. This is, to some extent, interesting, useful and makes
sense. The process of refining computations has no limit. To illustrate, a
detailed analytical treatment of long line parameters with the account of
the earth’s effect has taken several hundreds of pages in the work by
Sunde [6]. Suppose a superprogramme has been created for the solution of
the lightning protection problem; its application will immediately show
that the great efforts it has required can change but little the existing low
predictability of lightning-induced cut-offs. The key problem today is not a
rigorous mathematical solution of the available equations but an adequate
physical description of the principal physical processes producing a lightning
discharge, its electromagnetic field, and the object’s response to it. For this
reason, we have tried to present simple qualitative models rather than
stringent solutions to the equations. On the other hand, many aspects of
this problem have been omitted, partly because they are not directly related
to lightning as a physical phenomenon and partly due to the lack of space or
to the limited knowledge about the key physical phenomena.
Let us look back at the material presented in this book in order to
emphasize the points of primary importance. After the numerical value of
an overvoltage has been calculated, it is necessary to compare the result
obtained with the flashover voltage of the insulation in order to identify its
possible flashover. Most of the voltage-time characteristics of insulation
strings have been found from tests by standard 1.2/5Ops impulses (here,
the first value is the front duration and the second is the impulse duration
on the 0.5 level). Such a refined impulse has little to do with lightning over-
voltages, and this is clear from figure 6.24. A lightning-induced overvoltage
has necessarily a short-term overshoot arising not only from the current wave
reflection by the grounding of the neighbouring towers but also from the
magnetic induction emf. It is not quite clear how this rapidly damping

Concluding remarks 319
04-
0 5 10 15 20
Time, ps
Figure 6.24. Current impulses in a struck tower ( i s ) and in the few neighbouring
towers (il-i3).The wave problem was solved allowing for wave reflection from the
places where grounded wire is connected with the towers.
overshoot affects the electrical strength of an insulation string. Under certain

conditions, a powerful corona flash saturates the gap with a large space
charge and can ‘lock up’ the leader process, increasing the strength [12]. As
for ‘anomalously’ long overvoltages induced by positive lightnings, the
electrical strength of air may, on the contrary, be several dozens percent
smaller than what standard tests give [5,13]. The question of the real
electrical strength of the UHV transmission line insulation is still to be
answered.
The return stroke models discussed above ignore the lightning channel
branching and bending, whereas an actual discharge channel is far from
being a straight vertical conductor. The channel can deviate from the
normal by dozens of degrees, especially when it approaches high con-
structions. Another complicating point in a return stroke model is the
counterleader. The assumption that the return wave starts directly from
the top of an affected construction, say, from a line tower, is far from the
reality. The length of a counterleader is comparable with the height of the
construction it starts from. Together with the total length of the streamer
zones of the descending and ascending leaders, this will give a value 1.5-3
times greater than the object’s height. Such a high altitude of the return
stroke origin and its propagation in both directions from the point of
contact, not only towards the cloud, may have a considerable effect on
the electromagnetic field of the lightning. The available theoretical models
do not take these facts into account. There are no data on counterleaders
related to the subsequent lightning components. Today, it is even impossible

to confirm, or to disprove, the mere existence of a leader travelling to meet a

dart leader.
Another weak point of the models is the set of statistical data on the
amplitude and time characteristics of the lightning current impulse. There
is some information on medium current lightnings, because these are numer-
ous, whereas lightnings of extremal parameters are poorly understood. The
consequences of this are quite serious. The choice of protection means and
measures depends, to a large extent, on what has actually caused the storm
cut-off of a particular transmission line - a reverse flashover of the insulation
string, when the lightning strikes a tower, or the lightning breakthrough to
the power wire bypassing the overhead protecting wire. Underestimating
or, on the contrary, overestimating the high current probability by ignor-
ance, one may arrive at the wrong conclusion concerning the contribution
of reverse flashover in UHV transmission lines, which may cause great
losses. The determination of extremal lightning parameters is one of the
key problems in natural investigations. We should like to emphasize again
that the exceptions are more important than the rules to lightning protection
practice.
Refe rences
[l] Wagner C F 1956 Trans. AIII 75 (Pt 3) 1233

121 Lundholm R, Finn R B and Price W S 1958 Power Apparatus and Systems 34
1271
[3] Razevig D V 1959 Thunderstorm Overvoltage on Transmission Lines (Moscow:
Gosrenrgoizdat) p 216 (in Russian)
[4] Golde R H (ed) Lightning 1977 vol. 2 (London, New York: Academic Press)
p 294
[6] Sunde E D 1949 Earth Conduction Effects in Transmission Systems (Toronto:
Van Nostrand) p 373
[7] Ryabkova E Ya 1978 Grounding in High-Voltage Installations (Moscow:
Energiya) (in Russian)
[8] Bazelyan E M, Chlapov A V and Shkilev A V 1992 Elektrichesrvo 9 19
191 Kaden H 1934 Archivfur Electrotechnik 12 818
[lo] Babinov M B and Bazelyan E M 1983 Elektrichesrvo 6 44
[ l l ] Babinov M B, Bazelyan E M and Goryunov A Yu 1991 Elektrichesrvo 1 29
[12] Bazelyan E M and Stekolnikov I S 1964 Dokl. Akad. Nauk SSSR 155 784
[13] Burmistrov M V 1982 Elektrot. promyshlennost’; Ser. Appar. wysokogo napryaz-
heniya 1 123

(Eduard M. Bazelyan, Yuri P. Raizer) Lightning

Uploaded by

Copyright:

Available Formats

(Eduard M. Bazelyan, Yuri P. Raizer) Lightning

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Eduard M. Bazelyan, Yuri P. Raizer) Lightning

Uploaded by

Copyright:

Available Formats

Lightning Physics and

Copyright © 2000 IOP Publishing Ltd.

All rights reserved. No part of this publication may be reproduced, stored in a

British Library Cataloguing-in-PublicationData

ISBN 0 7503 0477 4

Library of Congress Cataloging-in-Publication Data are available

Publisher: Nicki Dennis

Published by Institute of Physics Publishing, wholly owned by The Institute of

Typeset in 10/12pt Times by Academic + Technical, Bristol

Copyright © 2000 IOP Publishing Ltd.

2 The streamer-leader process in a long spark 27

Copyright © 2000 IOP Publishing Ltd.

2.2.5 Electron-molecular reactions and plasma decay in

3 Available lightning data 90

Copyright © 2000 IOP Publishing Ltd.

4 Physical processes in a lightning discharge 138

5 Lightning attraction by objects 222

Copyright © 2000 IOP Publishing Ltd.

5.5 Extrapolation to lightning 236

Copyright © 2000 IOP Publishing Ltd.

Copyright © 2000 IOP Publishing Ltd.

the effectiveness of conventional protective measures and to predict technical

Copyright © 2000 IOP Publishing Ltd.

Introduction: lightning, its

1.1 Types of lightning discharge

The above recommendation to remove the camera shutter should be taken

Copyright © 2000 IOP Publishing Ltd.

Figure 1.1. A static photograph of a lightning stroke at the Ostankino Television

Copyright © 2000 IOP Publishing Ltd.

Figure 1.2. A photograph of a descending lightning with numerous branches.

growth. In addition to descending sparks outgrowing from a cloud toward

Copyright © 2000 IOP Publishing Ltd.

Figure 1.3. A photograph of an ascending lightning.

Copyright © 2000 IOP Publishing Ltd.

proportion of positive sparks has been found to be somewhat larger in

1.2 Lightning discharge components

An observer can notice a lightning spark flicker which, sometimes, may

Copyright © 2000 IOP Publishing Ltd.

Figure 1.4. The image of a multicomponent lightning in a slowly moving film.

1.3 Basic stages of a lightning spark

The affinity of lightning to a spark discharge was demonstrated by Benjamin

Copyright © 2000 IOP Publishing Ltd.

Copyright © 2000 IOP Publishing Ltd.

Current impulses of a return stroke accompany all components of a

Copyright © 2000 IOP Publishing Ltd.

1.4 Continuous and stepwise leaders

Figure 1.5. The analysis of an image of a vertically descending light source in a

Copyright © 2000 IOP Publishing Ltd.

An essentially different behaviour is exhibited by the leader shown in

Copyright © 2000 IOP Publishing Ltd.

1.5 Lightning stroke frequency

1.5.2 Human hazard

Copyright © 2000 IOP Publishing Ltd.

1.6 Lightning hazards

Copyright © 2000 IOP Publishing Ltd.

Thermal effects of lightning are most hazardous at the site of contact of

Copyright © 2000 IOP Publishing Ltd.

Thermal damage of conductors, through which lightning current flows,

Copyright © 2000 IOP Publishing Ltd.