Sprite observations over France in relation to their parent thunderstorm system

Lars Knutsson

OBERON : Through the house give gathering light, By the dead and drowsy fire : Every elf and fairy sprite Hop as light as bird from brier ; And this ditty, after me, Sing, and dance it trippingly.

(W illiam Shakespeare, A midsummer night‘s dream)

Abstract

As a part of the European research program CAL, sprite observations were carried out from the OMP observatory in the French Pyrenees during the summer 2003. Images of the sprites were taken by two remotely controlled CCD cameras. The 23 July was considered particularly interesting because we then had access to data concerning both cloud-to-ground and intracloud lightning activity. This day was therefore chosen as the object of the present study.

A large thunderstorm with two convective cores, one to the north and the other to the south, developed over the South of France during the late afternoon, and about two hours after sunset, the first sprite was detected. During a little more than three hours, 13 sprites were observed, 7 over the northern system and 6 over the southern system. The images enabled us to determine the azimuth angle of each sprite from the OMP observatory. 12 of the 13 sprites could be associated to positive cloud-to-ground flashes, and by putting together the sprite directions and the locations of the associated flashes on the radar images, we managed to get a rough idea of the position of the sprites in the storm system, and also to estimate their vertical and horizontal extent. Satellite images were included at this point of the study, and it appeared clear that sprites tend to occur over the stratiform region of the storm system in the area with the coldest (highest) cloud tops. The associated positive flashes were also within or close to this portion of the storm.

The sprite occurrences were studied in relation to the cloud-to-ground and to the intracloud activity. W e found that sprites seem to occur in a late stage of each storm system, when the rate of negative cloud-to-ground flashes has considerably decreased, and when the ratio of positive cloud-to-ground flashes is much higher then during the most active phase of the storm. Globally, the intracloud activity is also low during the sprite-producing periods, but sudden —bursts“ of intracloud lightning could frequently be observed at the moment of the sprite. The peak current of the positive flashes was found to be rather weakly correlated to their sprite-generating capacity. The available Schumann resonance measurements seem to indicate that the charge moment is a much more adequate parameter in this respect.

The areal coverage of the radar echo was calculated. The result supports the idea that sprite events tend to appear almost exclusively over large thunderstorm systems.

Contents

1 Introduction… … … … … … … … … … … … … … … … … … … … … … … … … . 1 2 Background and theory… … … … … … … … … … … … … … … … … … … … .. 2 2.1 The history of sprites… … … … … … … … … … … … … … … … … … … … ... 2 2.2 A worldwide phenomenon… … … … … … … … … … … … … … … … … … .. 3 2.3 The TLE family… … … … … … … … … … … … … … … … … … … … … … .. 5 2.3.1 Sprites… … … … … … … … … … … … … … … … … … … … … … … 5 2.3.2 Blue jets… … … … … … … … … … … … … … … … … … … … … … . 7 2.3.3 Blue starters… … … … … … … … … … … … … … … … … … … … … 8 2.3.4 Elves… … … … … … … … … … … … … … … … … … … … … … … ... 9 2.3.5 Sprite haloes… … … … … … … … … … … … … … … … … … … … ... 11 2.3.6 Trolls… … … … … … … … … … … … … … … … … … … … … ...... 11 2.3.7 Gnomes… … … … … … … … … … … … … … … … … … … … … … .. 11 2.3.8 Pixies… … … … … … … … … … … … … … … … … … … … … … … .. 11 2.3.9 And what else ?… … … … … … … … … … … … … … … … … … … .. 11 2.4 Theory… … … … … … … … … … … … … … … … … … … … … … … … … … . 12 2.4.1 Lightning… … … … … … … … … … … … … … … … … … … … … … 12 2.4.2 The global atmospheric electric circuit… … … … … … … … … … .. 13 2.4.3 W hat causes TLEs ?… … … … … … … … … … … … … … … … … ... 15 3 The CAL project… … … … … … … … … … … … … … … … … … … … … … … .. 21 3.1 W P5… … … … … … … … … … … … … … … … … … … … … … … … … … … . 21 4 Instrumentation and available data… … … … … … … … … … … … … … … ... 22 4.1 Images from the Pic du Midi… … … … … … … … … … … … … … … … … ... 22 4.2 ARAMIS… … … … … … … … … … … … … … … … … … … … … … … … … . 23 4.3 Météorage… … … … … … … … … … … … … … … … … … … … … … … … … 24 4.4 SAFIR… … … … … … … … … … … … … … … … … … … … … … … … … … .. 25 4.5 Meteosat… … … … … … … … … … … … … … … … … … … … … … … … … .. 27 4.6 Schumann resonance measurements… … … … … … … … … … … … … … ... 27 5 Case study… … … … … … … … … … … … … … … … … … … … … … … … … … 29 5.1 The meteorological situation… … … … … … … … … … … … … … … … … .. 29 5.2 Observed sprites… … … … … … … … … … … … … … … … … … … … … … . 31 5.3 The sprites in relation to the cloud-to-ground lightning… … … … … … … 34 5.3.1 The northern storm system… … … … … … … … … … … … … … … . 38 5.3.2 The southern storm system… … … … … … … … … … … … … … … . 42 5.4 Intracloud activity… … … … … … … … … … … … … … … … … … … … … ... 46 6 Conclusion… … … … … … … … … … … … … … … … … … … … … … … … … … 53

Acknowledgements… … … … … … … … … … … … … … … … … … … … … … … … . 55

References… … … … … … … … … … … … … … … … … … … … … … … … … … … ... 56

Appendix 1, Estimated vertical and horizontal extent of the sprites… … … … … .. 61 Appendix 2, Charge moments for some of the associated +CGs… … … … … … … 62 Appendix 3, Soundings from the airport of Lyon… … … … … … … … … … … … ... 63 Appendix 4, Intracloud activity 1 second before and 1 second after each sprite… 64

1 Introduction

Sprites are transient luminous glows at altitudes of approximately 40-90 km above large thunderstorms. They are the most dramatic visible evidence of electrodynamic coupling between thunderstorm systems and the overlying mesosphere and lower ionosphere. Although they are much more rare than —ordinary“ lightning, they may have long-term effects on the atmosphere such as chemical changes, persistent heating of ionospheric electrons, and increased production of mesospheric and stratospheric nitrogen oxides (Reising, 1998). All these effects can influence global climate. The understanding of the physical processes behind sprites, the meteorological conditions that generate them, and their global occurrence rate, are therefore of interest not only for a small group of scientists, but for humanity as a whole.

An important issue is to find out whether sprites significantly influence the global electrical circuit. In the conventional picture, the main components of Earth‘s electrical circuit include thunderstorms, the conducting ionosphere, the downward fair-weather currents and the conducting Earth. The thunderstorms are the generator that drives currents upward from cloud tops and maintains an electric potential of hundreds of kilovolts between the ionosphere and the ground. It is possible that sprites play an important role in this circuit. This is a topic of future research.

For easily understandable reasons, sprites are very difficult to observe from the ground. Occurring at high altitude over the massive cloud cover of large thunderstorms, they can be seen only under certain conditions from high mountains or from aircrafts. This is of course why they have remained unknown to science for so long. Scientific study of sprites began only a little more than ten years ago. Since then, there has been an explosion of research in optical and radio measurements and theoretical modeling, not only of sprites, but of all the parent mesospheric phenomena that have been discovered recently : blue jets, elves, trolls…

If sprite exploration is a very young research area, sprite exploration in Europe is even younger. The first sprites over the European continent were observed in year 2000 (Neubert et al., 2001), and the following summer campaigns have shown that these light emissions are a regularly occurring phenomenon in this part of the world.

The summer 2003 produced several large thunderstorms over France and was therefore a —good“ summer for the European sprite research. A great number of observations were made and the results are waiting to be analyzed. In this study, we concentrate on the 23 July 2003. This was one of the few days for which we had access to data concerning the intracloud activity.

1 2 Background and theory

2.1 The history of sprites

The 6 July 1989 is known to be the day when the phenomenon now called sprites was discovered. A group of scientists from the University of Minnesota, lead by Professor John R. W inckler, were then testing a low-light-level TV camera intended for a sounding rocket flight. W hile directing their camera to the northern horizon, they recorded a twin flash that seemed to propagate from some distant cloud tops up in the stratosphere. The flash lasted for less than 30 milliseconds and was associated with a very active storm center located on the northwestern side of Lake Superior. Its vertical extent was about 20 km and thus much greater than that of cloud-to-ground flashes. The publishing of these results (Franz et al., 1990) gave birth to an intense research in the electrodynamics of the middle atmosphere.

But the phenomenon did not suddenly appear on that particular day. For some time already, rumors had been telling about —upward flashes“ and airplane pilots claimed to have seen —strange things“. In fact, for more than a century there had been accounts from credible witnesses, but the science community had generally ignored them, perhaps because they had never been captured on film. By an odd coincidence, the same year as W inckler documented the event on video, Vaughan and Vonnegut (1989) had collected 15 reports of cloud-to-clear air lightning flashes by airline pilots.

Very soon, NASA got interested. Their first concern was to find out whether this newly discovered form of lightning might pose a potential threat to space shuttle missions, especially during launch or recovery phases1. Scientists began searching the archives of images from the space shuttle low-light TV camera system and found 17 cases of luminous events in the stratosphere and in the mesosphere (Boeck et al., 1995 ; 1998). The images seemed to indicate that there was some kind of relationship between the stratospheric phenomenon and the electric activity in the underlying cloud. The exact nature of this relationship, however, was unclear.

One now started to chase —upward lightning“ from the ground, in particular from high mountains, and within a relatively short time the inventory of recorded events grew considerably. In 1993, observations from the Yucca Ridge Field Station (YRFS) in Colorado (a perfect platform for studying the thunderstorms on the plains) documented more than 600 events on 11 nights (Lyons, 1994). And during a campaign in 1994, using two jet aircrafts over the Midwest US, the first color imagery and unambiguously triangulated physical dimensions and heights of these optical emissions were obtained. This campaign popularized the name —sprites“ for these events, —after their elusive nature“ (Sentman et al., 19952), and showed that, in fact, there were also another type of emissions : narrow beams of blue light that propagated upwards from the tops of thunderstorms. This new discovery was called —blue jets“ (Wescott et al., 1995).

A decade of research has significantly enlarged our knowledge of these phenomena. Since 1989, more than 10 000 images have been obtained by various research teams (Lyons et al., 2003b). New types with even more peculiar names have been observed : elves, haloes, gnomes, trolls… and there may be additional categories to be uncovered. Collectively they

1 After the loss of the space shuttle Columbia in February 2003, wild rumors were circulating about the possibility that the spacecraft had been felled by interaction with a sprite (Lyons and Armstrong, 2004). 2 Sentman got the name from the mysterious fairies populating some of Shakespeare‘s plays.

2 are frequently termed Transient Luminous Events (TLEs). Final report (2000) even uses the term TRansient ElectroMagnetic Events (TREMEs), implying the presence of energetic processes across a broad range of the electromagnetic spectrum.

But to a large extent, these stratospheric emissions remain somewhat mysterious. W hy do certain storms produce hundreds of sprites, while others, with apparently similar lightning activity, fail to do so ? W hat are the exact physics behind sprites ? These and many other questions remain to be answered.

2.2 A worldwide phenomenon

The space shuttle videotapes demonstrated that sprites are a widespread phenomenon, which occurs over land and sea, for example over Australia, Africa and south Pacific (Boeck et al., 1998). The following years of hunt for sprites revealed similar mesospheric events above Japan, near China, Brazil (Lyons et al., 2003a), South America, Europe, the Balkans (Neubert et al., 2001), Madagascar, Mexico, over the Hurricane George (Final report, 2000) and of course, on numerous occasions, over the midwestern high plains of North America (Lyons et al., 1999). The U.S. High Plains may represent one of the highest TLE-producing regions in the world. But of course, one should take into account the rather ideal observation conditions at the YRFS, the Colorado station being located on the locally highest terrain with unobstructed horizon, little light sources and clean air.

Figure 1. Global distribution of confirmed observations of TLEs. Triangles indicate ground and aircraft observations, squares space shuttle observations (Lyons et al., 1999).

Figure 1 shows the global distribution of sprites confirmed by ground, aircraft and space shuttle observations (to which should be added the recent sprites detected over Europe). If we compare this map with the global distribution of lightning events (figure 2), there appears to be a rough correspondence between the occurrence of sprites and the regions of high lightning

3 frequency. However, as noted by Lyons et al. (1999), given the regional variability of characteristic storm types, sizes and frequencies of positive cloud-to-ground flashes (+CGs), a more detailed TLE density map may very well not correspond closely to the global lightning pattern. Many tropical convective systems have very high lightning rates, but they are not necessarily sprite producers, because of smaller areal coverage and lower number of +CGs than mid-latitude mesoscale convective systems (MCSs).

Figure 2. Global distribution of lightning as detected from the Optical Transient Detector (OTD) launched from a Pegasus rocket in 1995. Map produced for the year of 1999 (source : NASA).

Considering the local variations, which are believed to be considerable, it is very difficult and hazardous to estimate the global TLE rate of the Earth. But Final report (2000) tries to determine an —upper bound“. Statistics from a number of U.S. storms show a rate of one TLE per every 200 CGs, a value that is thought to be the highest possible that could be globally applied. Since there are roughly 4 intracloud (IC) discharges for every CG, for U.S. storms there are about one TLE per 1000 flashes. Satellite estimates of global total flash rates suggest a mean value on the order of 50 flashes per second. Thus, if the U.S. statistics were valid for the entire planet, this would mean a rate of about one TLE every 20 seconds. In a more recent article, Sato and Fukunishi (2003) estimated the global occurrence rate of sprites to about 720 events/day on average (or one sprite every 2 minutes).

4 2.3 The TLE family

2.3.1 Sprites

In the beginning of the nineties, the name —sprite“ had not yet been adopted. One continued to talk about —upward lightning“, —cloud-to-clear-air lightning“ or even —cloud-to-space lightning“. But the eyewitness descriptions and the increasing number of photographic documents made clear that these terms were somewhat improper. The stratospheric emissions were perceived as highly atypical of conventional lightning, even though they seemed to be in some way associated with lightning discharges, and therefore, it appeared preferable to find another designation. Franz et al. (1990), followed by Lyons (1994), called the phenomenon —cloud-to-stratosphere“ (CS) events, but soon —sprites“, the name chosen by Sentman (Sentman et al., 1995), imposed itself as the generally accepted term.

Sprites are brief, optical emissions that occur in the stratosphere above thunderstorms. Like snowflakes they are all different ; their spatial structure range from small single or multiple vertically elongated spots to vast groupings. Various terms have been used to describe the diversity of the sprite forms. The most commonly reported feature is the —carrot“ (figure 3), with a strong, luminous center tapering towards lower altitudes. The brightest region, the —head“, is predominantly red and lies in the altitude range 66-74 km, above which there is often a faint red glow with a kind of wispy structure (—hair“), separated from the head by a dark band (—hairline“), and extending to about 90-95 km and sometimes more (Sentman et al.,1995). Below the bright red region, tendril- like filamentary forms extend downward to about 40 km or lower. Pasko et al. (2001) claim that these tendrils in certain cases can reach the cloud tops, but to our knowledge this has not been confirmed. Just below the head, the tendrils are red, but they fade to blue at lower altitudes.

Figure 3. Carrot sprites (Tohoku University, Japan).

Another type is called —columniform“ or —C“-sprite (figure 4). It consists of long vertical columns about 10 km long, less than 1 km in diameter, and shows virtually no variation in brightness along their length. Some show faint diffuse —hair“ or tendrils extending above and below the column. On some evenings, C-sprites are the dominant form of sprite above thunderstorms, but on other nights they may not be observed at all (Wescott et al., 1998b).

5

Figure 4. A cluster of C-sprites observed from YRFS (Final report, 2000).

Generally, the sprite starts at an altitude between 70 and 75 km. The initial growth is often downward, very much in the manner of corona streamers1. Downward propagation speed of the tip of the streamer at the end of the developing tendril is around 1-3×107 m/s. As the streamers move downward to an altitude of about 40-50 km, they begin to weaken after 1-2 ms. Then upward growth begins, often at a lower altitude (∼65 km) than the original starting point. The upward branching streamers move at a similar velocity, but the elements tend to be more persistent by approximately a factor of five (Final report, 2000).

The red color is a result of the excitation of the N2 first positive bands (N21P) without any discernable contribution from other emissions (Mende et al., 1995). But it is curious to note that, although the dominant colour is red, sprites are far from always perceived as red. In the historical reports of sprites, there is on the contrary a large variance in the colors. Lyons (1996) reports that, during one night of sprite observations, eyewitnesses described 37.5% of the sprites as green, 25% as orange, and only 37.5% as red or salmon pink. This is probably due to the poor colour response of the human eye at low light intensities. By the way, even under ideal conditions from a high mountain, only some sprites are visible to a dark-adapted eye.

Very early, it was noticed that sprites were almost always associated with tropospheric +CG flashes with time lags of less than one to over 100 ms (Sentman et al., 1995 ; Winckler et al., 1996 ; Lyons, 1996). Certain +CGs seem to trigger the sprites in some way, but high +CG rates do not automatically lead to sprite production, and the exact characteristics of the sprite producing +CGs remain to be discovered.

Sprites often appear in clusters of two, three or more. Some of the very large events seem to be tightly packed clusters of many individual sprites, but they can also consist of multiple spatially distinct sprites, which may be laterally distributed across distances of 40 km or more.

1 A corona streamer is characterized by electron impact ionization within a high electric field region at the streamer tip with rapid recombination of electrons behind the tip (Stanley et al., 1999).

6 The lateral dimension of —unit“ sprites is typically 5-10 km, corresponding to volumes that may exceed 1000 km3. Compound sprites and sprite clusters may occupy volumes greater than 10,000 km3 (Sentman et al., 1995).

High speed photometer measurements indicate that the brightest emissions only last on the order of 1-3 ms. But it is an often observed fact that the time duration of sprite images recorded on video often is much longer, several tenths or sometimes even several hundreds of ms (Lyons, 1996). Winckler et al. (1996) suggest that this puzzling fact may be partly explained by the different spectral response of the two instruments, the photometer being blue sensitive and the video photocathodes red sensitive, although the origin of the longer-lived red emission is not yet resolved.

However, Winckler et al. (1996) add that the long duration of some sprites is not at all connected with this process, but with a long-lasting cloud discharge following the +CG stroke that triggered the event. As a result of this discharge, the sprite —metamorphizes“, in the sense that some of the vertical striations fade and are replaced by others. The new elements are often laterally displaced from the old, which sometimes gives the impression that the sprite —dances“ across the sky (—dancing sprites“). The associated +CG moves large amounts of charge to ground by a continuing current (Reising et al., 1996), and in some cases, a series of +CGs may occur sequentially over a large horizontal distance within a fraction of a second (—spider lightning“), suggesting the existence of an expansive travelling network of intracloud currents (Barrington-Leigh et al., 2002).

Detailed analysis of the apparent brightness of several separate unit sprites yield a maximum brightness of roughly 600 kilo Rayleighs (kR) within the sprite1. W hen averaged over the dimmer outer portions of the sprite, the mean brightness is about 50 kR. Assuming the optical emission is centred at 600 nm and spatially uniformly distributed throughout a mean estimated emission volume of 2000 km3, this translates into a total optical energy of 1-5 kJ per event. If the duration of the sprite is 3 ms, the instantaneous optical power is in the range 0.3-1.5 MW (Sentman et al., 1995). Large sprite clusters may of course greatly exceed these figures.

Sprites are most likely to occur above large mesoscale convective systems (MCS) exhibiting horizontally extensive stratiform precipitation regions. Linearly extensive squall lines also produce sprites. But intense, compact superscell storms, in spite of their extremely high intracloud flash rates and numerous +CGs, rarely generate many sprites, with the exception of brief bursts of sprites during their dissipation phase when a sufficiently large stratiform precipitation area develops (Final report, 2000 ; Lyons et al., 1999 ; 2000a ; 2000b)

2.3.2 Blue jets

The 1994 jet aircraft observations revealed the existence of another TLE event, rapidly named —blue jet“ (Wescott et al., 1995). Although this new variant was the first to be discovered after the sprites, later research has shown that this phenomenon is the very rarest of all known TLEs. In fact, even today, the greatest fraction of the blue jet observations comes from this 1994 campaign, when as many as 56 examples were recorded during a 22 minute interval in a storm over Arkansas.

1 A Rayleigh is the number of photons per second per cm2 column integrated and normalized to 106 photons.

7 Blue jets are narrowly collimated cones of blue light that appear to propagate upwards from the top of the anvil in electrically active storms and reach altitudes of 40-50 km. The jets ascend at a 10-20 degrees angle, travel at speeds on the order of 100 km/s and flare out as they reach maximum altitude such that they resemble a trumpet. In most cases the jet seems to fade away all along the cone simultaneously, about 200 ms after it began (Wescott et al., 1995). At the time of the discovery, it was obvious that blue jets were fundamentally different from sprites in at least two ways : the relatively low speed made it easy to follow the upward motion at video rates, and the emissions were blue.

Blue jets do not seem to be temporally associated with specific CG discharges of either polarity, although they do occur in the same general area as negative CG flashes. All of the blue jets observed in the 1994 campaign appeared in two very active thundercells with winds on the ground of 100-120 km/h and baseball sized hail (up to 7 cm diameter). This suggests that very strong updrafts may be required to produce the large electric fields necessary to produce blue jets. This may also explain why blue jets are so rarely observed, even though it is quite possible that they are more frequent than we are aware of ; their blue emissions (primarily below 480 nm) are severely scattered by the transmission through the atmosphere, which makes them difficult to detect, especially from the ground. There are indications that these emissions come from the N2 second positive bands. Blue jets are very bright, at least of the order of 1000 kR (Wescott et al., 1998a). Figure 5. Blue jet observed from La Réunion Island (Final report, 2000).

The cumulative distribution of the œCGs during the interval 5 seconds before and 5 seconds after the blue jet shows some strange behaviour. The jets seem to occur at the end of a more- frequent-than-average series of œCGs, but after each jet, there is a significant lull in activity for about 2 seconds and within a radius of 15 km, as if the jet extracted electrical energy from the storm cell (Wescott et al., 1998a).

2.3.3 Blue starters

During the first pass by the Arkansas storm in 1994, Wescott et al. (1995) recorded nearly two dozen other, smaller events that they interpreted as —upward lightning“. They were brief, very bright and blue in colour, and extended upward from the cloud tops (17-18 km) for much shorter distances than the blue jets (maximum 25.5 km). Later, this kind of event was considered as a distinct phenomenon (Wescott et al., 1996), even if it seemed probable that it was related to the initial phases of blue jets.

8 Like the blue jets, the blue starters do not appear to coincide with either negative or positive CG strokes. Another resemblance is that, at the time of the event, they are followed by an abrupt decrease in the number of CG flashes in the surrounding area. However, there is one significant difference in the cumulative CG flash distribution between the jets and the starters: the increase in the rate of œCGs observed before the blue jets do not seem to occur before the blue starters. This suggests that more charge transfer to the ground precedes jets than starters (Wescott et al., 1998a).

The velocities of the blue starters vary considerably. Wescott et al. (1996) found values between 27 and 153 km/s. Furthermore, the velocity was not always constant ; in most cases starters slowed down with time, but one of them was observed to increase its speed abruptly (Wescott et al., 1996).

2.3.4 Elves

Of the 17 luminous events found on the space shuttle video, 15 were identified as sprites, one as a blue jet, and one could at first not be classified at all. It was —an enhanced airglow luminosity“ (Boeck et al., 1992) observed at an altitude of about 95 km in coincidence with a lightning flash in a tropical oceanic thunderstorm near French Guyana. The two phenomena were separated by a clear air layer of about 80 km and there was no visible discharge connecting them.

Brief high altitude flashes were noted at YFRS in 1994 (Final report, 2000), but it was not until the SPRITES‘95 campaign that one was able to confirm the existence of diffuse optical flashes with a duration of less than 1 ms at altitude 75-105 km in the lower ionosphere just after the onset of cloud-to-ground lightning discharges (Fukunishi et al., 1996). This new member of the TLE family was called ELVES, the acronym of Emission of Light and Very low frequency (VLF) perturbations from Electromagnetic pulse (ELP) Sources.

Figure 6. Elve captured during the night of the Leonids storm 1998 (Space Dynamics Lab., Utah State University).

9 The discovery was not entirely a surprise ; Inan (1990) had predicted that VLF signals produced by lightning may substantially heat the lower ionosphere, and Taranenko et al. (1993) had brought up the idea that electromagnetic radiation originating in lightning discharges may excite optical emissions with substantial intensities in the night time lower ionosphere.

Had it not been for their extremely short lifetime, eyewitnesses would certainly have observed elves a long time ago. The fact is that they would fill the entire night sky for any observer within a 100 km radius from the causative lightning flash if the human eye was able to see them (Boeck et al., 1998). Elves are very broad (up to 400 km diameter) expanding, disk shaped structures on the edge of the ionosphere. They were observed first in association with positive CGs, but Barrington-Leigh and Inan (1999) have shown that they can be associated with negative CGs as well. This is consistent with the theory of EMP heating, a process which is independent of the polarity of the field. And since œCGs are much more common than +CGs, it is likely that the occurrence rate of elves exceeds that of sprites. In fact, the authors claim that nearly all discharges with an EMP intensity above a certain threshold may trigger elves. In other words, it is quite possible that only a very small subset of elves are detectable with current instruments. Not surprisingly, elves detected using a high speed photometric array (named the —Fly‘s eye“) were associated with œCGs and +CGs with peak currents primarily exceeding 75 kA. However, elves do not require the large cloud-to-ground charge transfer that apparently occur in sprite-producing lightning discharges (Reising, 1998).

Though very brief, the luminous intensity of elves is high : >1 MR (Inan et al., 1997). The onset time of elves is on the order of ∼300 µs (Final report, 2000), which corresponds to the round-trip traversal time (with the speed of light) of the radiation from the ground discharge to the ionosphere and then to the observer. Theoretical models (Inan et al., 1996) predict that elves should in fact be a rapidly expanding —doughnut- shaped“ emission, but only a few images actually exhibit this Figure 7. Pictorial view of the principal TLE phenomena. morphology.

10 2.3.5 Sprite haloes

Some sprites are preceded by a diffuse disk-shaped glow that superficially resembles elves but at a lower altitude (70-85 km). Initially, they were interpreted as elves followed by sprites (—sprelves“), but are now being termed —sprite haloes“. They are usually less than 100 km wide and propagate downward. Their duration is considerably longer than for elves, about 1-3 ms. Columnar sprite elements sometimes emerge from the lower portion of the sprite haloe‘s concave disk (Lyons et al., 2000a).

2.3.6 Trolls

During the SPRITES‘99 campaign at the YRFS in Colorado, video confirmation was obtained of another TLE : the troll (Transient Red Optical Luminous Lineaments). These resemble blue jets, but are clearly dominated by red emissions. They occur after an especially vigorous sprite in which tendrils have extended downward to near cloud tops. Their brightness is comparable to that of weak sprites. The trolls exhibit a luminous head leading a faint trail and move upwards initially around 150 km/s, gradually decelerating and disappearing by 50 km. They are not directly associated with any apparent CG (Lyons et al., 2000a ; Final report, 2000).

2.3.7 Gnomes

The STEPS (Severe Thunderstorm Electrification and Precipitation Study) program during the summer of 2000, revealed a series of unusual luminous events atop the overshooting convective dome of a supercell storm. They were brief (33-136 ms duration) upward propagating lightning-like channels. Estimated to be less than 200 m wide, they did not grow more than 1 km above the cloud top. They appeared to be brighter, much more compact in shape, and more optically uniform than blue starters (Lyons et al., 2003b). To the already existing fairy tale menagerie was added the —gnomes“.

2.3.8 Pixies

During the same supercell storm, very small but intense pinpoints of light appeared, also scattered about upon the surface of the convective dome. They were estimated to be on the order of 100 m in size and lasted less than 16 ms. Provisionally they were referred to as —pixies“ (Lyons et al., 2003b).

2.3.9 And what else ?

Although the TLE family has not ceased to grow during the last decade, we may not yet know all the members of it. Lyons et al. (2003b) observe that numerous eyewitness reports of lights in the high atmosphere do simply not fit the descriptions of sprites, elves, blue jets, etc. Not even all the 15 observations by airline pilots published by Vaughan and Vonnegut (1989) can be classified in the known TLE groups. So there may very well be new kinds of emissions to be uncovered. Another possibility is that the already discovered groups express themselves over a wider variety of shapes, morphologies and optical intensities than we know of today.

11 2.4 Theory

2.4.1 Lightning

Lightning is a transient, high-current discharge whose path length is measured in kilometers. Although well over half of all flashes take place within the cloud, so called intracloud (IC) discharges, cloud-to-ground (CG) lightning has been studied far more thoroughly because of its potential for damage to property and living things, but also because of its relative accessibility for study. The polarity of a CG flash is by convention classified as —positive“ or —negative“ based on the polarity of its net effect on the charge of the thundercloud. If the net effect of the discharge is to move negative charge (electrons) from the cloud to the ground, it is called a negative CG. If it has the net effect of transferring positive charge from the cloud to the ground (electrons moving upward), then it is called a positive CG (Reising, 1998).

In the classic model for the charge structure, the thundercloud forms an electric dipole, as shown in figure 8. This means that a primary positive charge region is found above a primary negative charge region. At the base of the cloud, there is often also a small localized zone of positive charge.

(b) (a)

(c) (d)

Figure 8. Thundercloud charge distribution and categorization of the four types of lightning between cloud and ground (BOLT Lightning protection).

This simple structure has a general validity, but in any given cloud the charge distribution can be much more complex. The charges can for example be stratified into several layers (as many as 6 and maybe even more) alternating in polarity in the vertical (Stolzenburg et al., 1998).

12 The mechanism behind the electrification of the clouds is highly complex and not fully understood. Several processes are thought to be involved in the charge separation. W ater droplets freeze from the outside, and when doing so, positively charged ice fragments may splinter off. They are then carried upward into the anvil of the storm. Other processes are graupel-ice crystal collisions, selective ion capturing, inductive charge separation and phase transitions of water. The fact is that there is a multitude of mechanisms by which clouds can become electrified, and in spite of considerable efforts to explain the electrical properties of thunderclouds, many questions are still unanswered.

A CG flash consists of a series of leaders, typically two or three, each followed by return strokes, which occur each time there is a completion of the electrical connection between the cloud‘s charge reservoir and the ground. The first return stroke is initiated by the —stepped leader“, the visible channel following a high conductivity path formed by preliminary breakdown preceding the flash. If the flash ends when the first return stroke ceases, it is called a single stroke flash. But if additional charge is available in the cloud, a —dart leader“ may propagate down the residual channel and initiate a subsequent stroke. As many as 15 more return strokes may occur in the same flash, with typical delays of 30-100 ms between strokes (Reising, 1998). Lightning often appears to —flicker“ because the human eye can just resolve the individual pulses of luminosity that are produced in each stroke.

Four different types of CG lightning have been identified. Flashes initiated by downward- moving negatively charged leaders (figure 8a) account for more than 90% of the CG discharges worldwide, while less than 10% are initiated by a downward-moving positive leaders (figure 8c). CG discharges can also be initiated by leaders of either polarity that move upward from the Earth (figure 8b and 8d), but they are rare and usually occur from mountain peaks and tall man-made structures.

Negative and positive CGs differ in their properties and occurrence rates. Positive flashes are generally composed of a single stroke, sometimes followed by a period of continuing current. Positive flashes are often initiated by the upper positive charge in thunderclouds, where this charge has been separated horizontally from the lower negative charge by wind shear, but this may not always be a necessary condition. There is a widely spread notion that large peak current CGs are generally of positive polarity. This is not true ; actually, for all classes of peak currents >75 kA, the negative flashes outnumber the positive ones (Final report, 2000).

Lightning is preferentially produced over land because of the strong updrafts in continental clouds as opposed to oceanic clouds. As we can see on the map of the global distribution of lightning (figure 2 in 2.2), the major part of the lightning activity takes place in the intertropical convergence zone (ITCZ). There is a pronounced concentration of flashes in the Amazon Basin of South America, the Congo Basin of Africa, Southeastern Asia and Indonesia. The satellite measurements made by the Optical Transient Detector indicate a rate of ∼ 40 flashes per second worldwide (Reising, 1998).

2.4.2 The global atmospheric electric circuit

The ionosphere is the region in the upper atmosphere where there are enough electrons and ions to make the atmosphere a reasonable good conductor. Charged particles are created when solar radiation at wavelengths shorter than 102.7 nm is absorbed by atmospheric molecules and atoms. Energy is transferred to an electron in the molecule, which then escapes to become

13 a free electron, leaving the molecule positively charged. This process is called photoionization. In the ionosphere overall, neutral atoms and molecules greatly outnumber electrons and ions, but there are still enough charged particles to create a discontinuity between the ionosphere and the lower atmosphere. The conductivity of the ionosphere is due primarily to free electrons, because the mobility of free electrons is much greater than the mobility of ions. Therefore, ionization of the ionosphere is often described in terms of the number density of electrons, Ne (MacGorman and Rust, 1998).

Between the ionosphere and the Earth, both highly conducting, there is a voltage difference of approximately 250 kV, the ionosphere being positively charged and the Earth negatively. This generates an electric field directed downward (in fair weather) with a strength of about 100- 300 V/m at the surface. There are diurnal, seasonal and other time variations in this field that are caused by many factors (Rycroft et al., 2000).

The conductivity of the atmosphere σ below the ionosphere is weak, but far from negligible. It is maintained primarily by cosmic ray ionization and it increases with altitude. Near the surface, it is on the order of 10-14 mho/m, and it increases nearly exponentially with altitude up to 60 km due to the energy spectrum of the cosmic rays and the charged particles precipitating from the magnetosphere (Rycroft et al., 2000). In fact, a current with a density of -12 2 about Ja ≈ 10 A/m flows vertically through the atmosphere. It carries some 1800 A to ground and it would be sufficient to eliminate the potential difference between the ionosphere and the surface of the Earth within a few minutes (Stozhkov, 2003).

Clearly, this does not happen, and the reason for this is the permanent lightning activity around the world. The electric field under a thundercloud is reversed with respect to the fair- weather field, and far more intense (∼5 kV/m). CG lightning and the powerful precipitations transfer negative charge to the ground. Each thunderstorm cell drives a vertical electrical current upward of ∼1 A. And since there are permanently about 1000-2000 active thunderstorms occurring over the world, this compensates the fair-weather current in the opposite direction and closes the circuit. Thus, thunderclouds are the generators of the global electric circuit, as illustrated in figure 9.

Figure 9. Diagram of the global electric circuit. (Rycroft et al., 2000)

An upward current is driven through a —charging resistor“ of 105-106 Ω (i.e. the thunderstorms occurring over less than 1% of the Earth‘s surface), and a downward current flows through a —discharging resistor“ of about 102 Ω, representing the parallel resistance of all fair-weather air columns (>99% of the Earth‘s surface). In a simplified picture, one can compare this

14 structure with an giant spherical capacitor, in which the ionosphere and the Earth are the two charged plates separated by the concentric shell of the atmosphere. The thunderstorms then act as batteries charging the capacitor. The capacitance is given by

C = 4πε0RE×/H ≈ 0.7 F (1) where RE is the radius of the Earth, ε0 the permittivity of free space and H the scale height of the atmosphere (∼7 km). Inserting V=250 kV, this means that the energy associated with the global electric circuit is enormous :

W = CV2/2 ≈ 2×1010 J (Rycroft et al., 2000) (2)

The upper plate of the capacitor is not confined to a single level, but rather is distributed through the atmosphere, reflecting changes in atmospheric conductivity and the fair-weather electric field, so that the density of the leakage current remains roughly constant with altitude (Pasko, 2003). This current density can be written as

J = σE (3)

But since the conductivity of the atmosphere σ increases with height, the electric field must then decrease, which it does quite rapidly ; at 20 km, the fair-weather electrical field is ∼1 V/m, and at 50 km, it is only ∼10-2 V/m (Rycroft et al., 2000).

2.4.3 What causes TLEs ?

The thunderstorm generator hypothesis was proposed by the British physicist C.T.R. W ilson in 1920 (Rycroft et al., 2000), and it was also him who outlined the first theoretical explanation of the phenomenon that we now know as sprites. As early as 1925, he predicted that electric fields could cause ionization at great heights and give rise to discharges between clouds and the upper atmosphere, and in 1956, he wrote : —It is quite possible that a discharge between the top of the cloud and the ionosphere is a normal accompaniment of a lightning discharge to Earth“ (quoted by Lyons et al., 2000b).

The coupling between the TLEs and the electrical activity in the underlying thunderstorm appeared obvious from the very beginning. But still today, there is considerable uncertainty as to the exact mechanisms of this coupling. Basically, there are three sources of field energy that can power these upper atmosphere events :

- the electrostatic field from the initial, relatively slow separation of charges within the cloud prior to the discharge ; - the electromagnetic pulse (EMP) from the propagation of the return stroke ; - the quasi-electrostatic (QE) field that temporally exist at high altitudes following the sudden removal (by a lightning discharge) of thundercloud charge.

To connect this energy to the neutral atmosphere and generate sprites, two hypothesis have been proposed : heating of ambient electrons by the QE thundercloud fields, and runaway electron processes driven by the same QE fields (Rowland, 1998).

15 The first of these hypothesis relies on conventional dielectric breakdown, largely developed by Victor P. Pasko at Stanford University (Pasko, 1996 ; Pasko et al., 1995 ; 1996 ; 1997). As the thundercloud charges slowly build up before a lightning discharge, high-altitude regions are shielded from the QE fields of the thundercloud charges by the redistribution of free charges in the mesosphere and the lower ionosphere, creating an electric field approximately equal and opposite to the —applied“ electric field due to the thundercloud charges (figure 10a). W hen one of the thundercloud charges is quickly removed by a lightning discharge (e.g., the positive one as in figure 10 b), the remaining charges of opposite sign in and above the thundercloud produce a large QE field that appears at all altitudes above the thundercloud, and endures for a time equal to approximately the local relaxation time1 at each altitude :

τr = ε0 / σ (4) where ε0 is the permittivity of free space, and σ the local conductivity. Beyond the time τr, the conductive upper atmosphere nullifies the imposed electric field.

(a) (b) (c)

Figure 10. Electron heating, breakdown ionisation and excitation of optical emissions by penetration of large quasi-electrostatic thundercloud fields to mesospheric altitudes (Pasko et al., 1997).

This temporarily existing QE field accelerates low energy electrons to energies where they can collisionally excite or ionize the neutrals and in this way lead to optical emissions (Pasko et al., 1997). The QE field falls off with altitude z as ∼z3, according to :

1 2M ∆E = [V / m] (5) πε 3 4 0 z where M is the charge moment, defined as the product of the charge lowered from the cloud and the height from which it is lowered (the unit is therefore Coulomb meter, C m, or more frequently used, C km), (Huang et al., 1999). But the dielectric strength of the atmosphere

1 Defined as the time the electric current takes to adjust to 1/e of its final value after an electric field is suddenly applied, assuming that the conductivity remains constant (Rycroft et al., 2000).

16 decreases more rapidly, since it is fundamentally proportional to the air density (rather than pressure), which declines nearly exponentially with altitude. At some critical altitude the electric field will therefore exceed the local breakdown strength and breakdown will commence, unless the pulse length is too short or the ambient conductivity is too high. And as is well known, the emission of light from a gas increases drastically when the electric field exceeds the breakdown strength of the gas. The occurrence of breakdown thus depends on the altitude, the ambient conductivity, and the amplitude and the duration of the QE field, which relaxes with a time constant of ∼1 ms at 80 km and ∼3 ms at 70 km altitudes, consistent with the observed few ms duration of sprites (Fernsler and Rowland, 1996 ; Pasko et al., 1995 ; 1997).

Since removal of a positive charge from the thundercloud by a +CG discharge is equivalent to the deposition of a negative charge of the same magnitude at the same location, the resultant fields at ionospheric altitudes do not depend on the complexity of the initial charge configuration in the thundercloud, but only on the effective charge moment of the removed charge. Thus, a 100 C charge removed from 10 km altitude produces approximately the same effects in the lower ionosphere as a 50 C charge removed from 20 km altitude (Pasko et al., 1996 ; 1997). And that is why, in recent years, the charge moment has emerged as the single most important parameter that determines whether or not a +CG produces a sprite. The discharge duration plays a secondary role, as long as the charge removal time is shorter than the local relaxation time at ∼70-90 km altitude (Reising, 1998). The relaxation time of the electric field depends on the conductivity of the atmosphere ; for low ambient conductivities of the upper atmosphere, the lightning discharge duration can be significantly longer than 1 ms and still produce similar levels of optical emissions (Pasko et al., 1997).

Pasko et al. (1996) also suggested that localized conductivity changes which occur due to external sources (i.e., electron precipitation from the magnetosphere, meteors, runaway electrons, perturbations left from a previous sprite, etc.) may significantly modify the electric field and explain how one CG can create clusters of sprites. Moreover, the model of Pasko et st al. (1996 ; 1997) showed that above 50 km, the 1 positive band of the neutral N2, excited (but not ionized) by colliding free electrons, should be the dominating optical emission in the altitude range around ∼70 km, which accounts for the observed red colour of sprites, in excellent agreement with spectroscopic observations of sprites (Mende et al., 1995). As for the bluish color of + the tendrils, it is caused by collisional excitation of N2 ions by free electrons. This can be illustrated by an electrically excited glow discharge tube as in figure 11. This cathode tube is filled with air at a temperature and a pressure that roughly match conditions in sprites. A long uniform region of red emission can be seen near the anode end of the tube. As is the case with field observations of sprites, no spectral evidence for Figure 11. Sprite light in the ionization is found in this region of the tube. The blue atmosphere (left) and in a light (hard to distinguish on the image) near the cathode + laboratory glow discharge tube is caused by the impact of electrons on N2 ions (right) (source : Williams, 2001). (Williams, 2001).

All this looks quite convincing. Nevertheless, some have expressed doubts as to whether the thunderstorm lightning can produce sufficient charge moments for the conventional breakdown (Roussel-Dupré and Gurevich, 1996 ; Huang et al., 1999). The plot of the electric

17

field versus altitude in figure 12 gives us an idea of the kind of charge moments required for this process to take place. The diagram also shows the behavior of dielectric strength versus altitude. Since its vertical decline is close to exponential, it appears as nearly a straight line in the logarithmic scale. High altitude breakdown is possible, but at lower altitudes, it seems hard to obtain a sufficient charge moment. To have a sprite initiated by conventional air breakdown at 60 km, a charge moment of about 3500 C km would be required. Even if charge moments well over 1000 C km have been detected (Cummer and Stanley, 1999 ; Huang et al., 1999), they seem to be rather rare. One important Figure 12. Breakdown electric field versus height and charge moment (source : Huang et factor in this respect is the height of the al., 1999). positive charge reservoir, since the charge moment is the product of the removed charge and the altitude. The conventional view in atmospheric electricity has been that this reservoir is located at altitudes at and above 10 km. This picture may not be accurate. In recent years, observations indicate that sprites are produced by laterally extensive mesocale convective systems (MCS) in which the positive charge reservoir predominates in the 4-6 km ranges of altitude, close to the 0°C isotherm (Boccipppio et al., 1995 ; Williams, 1998 ; Marshall et al., 2001). This would seriously limit the possibility of very large charge moments. Furthermore, large charge moments do not always seem to be needed to produce sprites ; Hu et al. (2002) found sprites initiated by charge moments as high as 3070 C km - but also as low as 120 C km.

Figure 12 also exhibits the altitude dependence for another kind of breakdown, often referred to as the runaway electron breakdown, and this is the other hypothesis that has been put forward to explain the sprite emissions. This curve is given by

ρ(z) E(z) = 218 [kV / m] (6) ρ(0) where ρ (z) is the altitude dependant density of air (Huang et al., 1999). As we can see on figure 12, the threshold electric field needed to initiate this breakdown is considerably lower (about a factor of 10) than for the conventional breakdown, which makes this process attractive, since the breakdown can be obtained with a lower charge moment.

In the first hypothesis, we were dealing with low energy electrons (a few eV). The runaway process involves high energy (∼MeV) electrons provided by the cosmic rays. These rays produce ∼10-5 > 1 MeV electrons per cm3 and per second at 10 km altitude (Rowland, 1998). The electrons are turned around and accelerated by the QE field. To illustrate this mechanism, we can plot the frictional force experienced by a high-energy electron as it propagates through

18

Figure 13. Rate of electron energy loss in air. The frictional drag dE/ds experienced by an electron in air as a function of the electron energy (Roussel-Dupré and Gurevich, 1996) air as a function of the electron energy (figure 13). W e see that a minimum exists at approximately 1 MeV. If an electric force whose magnitude exceeds the minimum is applied to the medium, then electrons with energies greater than the critical value εc, at which the electric force equals the frictional force, will be maintained and accelerated (runaway) to higher energies. Through collisions, these electrons produce ions and new electrons. Most of these new electrons thermalize due to collisions, but those whose energies exceed the critical value εc become part of the runaway population and contribute to further ionization that also populates the runaway regime. The net result is an avalanche in which the electron population grows exponentially. Collimation of these relativistic electrons by the electric field leads to the formation of an electron beam propagating through the medium as long as the electric field exceeds the threshold. The basic properties of this electron beam depend on the parameter δ0 = E/Et, where Et is the magnitude of the threshold electric field for runaway breakdown defined by equation (6), (Roussel-Dupré and Gurevich, 1996). δ0 determines the percentage of low energy secondary electrons produced by the energetic runaway electrons which becomes runaway electrons themselves after being accelerated in the QE field. Of course, the exponential growth cannot go on forever ; in fact, the increased conductivity reduces the E field and stops the runaway process after about 1 ms (Bell et al., 1995).

In several regards, the runaway theory seems to fit the observations rather well. Bell et al. (1995) have shown that, with the runaway model, the optical emissions should peak in the 55- 70 altitude range, and substantial intensity should be found at altitudes as low as 45 km. As already mentioned, Sentman et al. (1995) found that the brightest region of the sprites was located between 66 and 74 km altitude, and he also observed that some sprites exhibit significant intensity at altitudes as low as 40-50 km.

A strong argument is also the unexpected satellite measurement of γ ray flashes in the vicinity of thunderstorms by the Burst And Transient Signal Experiment (BATSE) aboard the Compton Gamma Ray Observatory (CGRO) (Fishman et al., 1994). Indeed, one of the unique signatures of runaway breakdown is the strong γ ray flux generated when the electron beam collides with air molecules, causing the electrons to slow down and to emit bremsstrahlung radiation. Gamma rays cannot be produced by a thermal breakdown model (Roussel-Dupré and Gurevich, 1996).

19 Furthermore, even in the few cases for which the charge moment is sufficiently large for conventional breakdown, one can wonder (as does Huang et al., 1999) why runaway breakdown would not occur first, since the breakdown strength is much lower for this process ?

Still, experimental validation of the runaway process is, to our knowledge, not yet available and evidence for a runaway mechanism in sprites is scant (Williams, 2001). Moreover, problems arise when the geomagnetic field is accounted for. Firstly, the field threshold conditions for the creation of the electron beam is substantially changed when the magnetic field is incorporated in the equations. Secondly, the magnetic field significantly deflects the runaway beam from the vertical, which means that the frequently observed vertical structure of the sprites appears to contradict the runaway theory. And at the geomagnetic equator, the horizontal geomagnetic field is perpendicular to the vertical thunderstorm electric field and can prevent the development of relativistic electron avalanche at altitudes ≥ 40 km (Lehtinen et al., 1999).

Of course, one cannot exclude that thermal breakdown and runaway breakdown could occur at the same time. Rowland (1998) points out that the thermal sprite will move down in altitude as Q increases. The runaway sprite, when it is triggered, moves up at nearly the speed of light. W hich one triggers first will increase the plasma density. This could short the fields out fast enough to prevent the other from triggering.

The only TLE mechanism about which there seems to be a relative consensus is the origin of elves. The return stroke produces an electromagnetic pulse (EMP) that is inherently short- lived (∼0.1 ms) and the conductivity of the atmosphere is not large enough to affect the fields until they reach the lower ionosphere, where a thermal breakdown takes place. According to Fernsler and Rowland (1996), breakdown should occur whenever the return-stroke current exceeds ∼50 kA. The accompanying enhanced airglow (elve) are centered at ∼90 km altitude. The briefness and the diffuse character of the elves make them very difficult to detect optically.

20 3 The CAL project

Coupling of Atmospheric Layers (CAL) is a Research Training Network1, funded by the European Commission‘s Improving Human Potential (IHP) program. It began 1 November 2002, has a duration of four years, and involves scientists from the Danish Space Research Institute, the Belgian Institute for Space Aeronomy, l‘Université Paul Sabatier , le Commissariat à l‘Energie Atomique (CEA), the University of Oulu, Johann W olfgang Goethe Universität, the University of Leicester, the De Montfort University, the Finnish Meteorological Institute, the University of Crete, and the Danish Meteorological Institute.

The research concerns thunderstorm electrical and space radiation effects in the stratosphere, mesosphere and lower thermosphere. The network studies unanswered questions relating to Transient Luminous Events (TLEs) in the stratosphere and mesosphere, their relation to various aspects of the atmospheric system, and the overall dynamic response of the atmospheric layers to forcing of the mesosphere and lower thermosphere regions by thunderstorm and solar activity.

The key topics investigated in CAL are :

1. Physics of the TLEs 2. The characteristics of the thunderstorms generating TLEs 3. Chemical effects on the atmosphere 4. Global distribution of TLEs 5. Modifications to the global atmospheric electrical circuit 6. D- and E-region dynamics 7. Dynamic response of the atmosphere

3.1 W P5

The CAL project is divided into ten different W ork Programs (W Ps). W P5 is called —Cloud Electrification and Meteorology“ and is orchestrated by the Atmospheric Electricity Group at the Laboratory of Aerology of the University of Toulouse in France. Its objective is to characterize the physics of the thundercloud which produces the TLEs. Aspects of the meteorological environments, cloud system structure, microphysics, dynamics, and electrical activity are considered. The main questions to be answered are : (i) W hat is the intra-cloud activity during the TLE production ? (ii) W hat are the characteristics of the CG flash associated with the TLE ? (iii) W here is located the TLE with respect to the associated CG flash ? (iv) In which area of the cloud (convective of stratiform) is produced the TLE ? (v) At which evolution phase of the thunderstorm are the TLEs produced ?

The present work is a small piece of the W P5 research.

1 A Research Training Network (RTN) consists of a consortium of teams (e.g. universities, research organizations, industrial firms, international organizations, etc), located in different countries, that propose a common research project to serve as a vehicle for providing training and, where necessary, transfer-of- knowledge (Marie Curie Research Training Networks Handbook, European Commission, 2nd ed. 2003).

21 4 Instrumentation and available data

4.1 Images from the Pic du M idi

The sprite2003 campaign was conducted during the summer from Observatoire Midi- Pyrénées (OMP) located at the Pic du Midi de Bigorre (42°56′12.0′′N ; 0°0′34.16′′E), in the French Pyrenees (figure 14). At an altitude of 2877 m, the observatory enables a full 360° view of the horizon and very clear skies when local weather conditions are favorable. W ithin a range of about 1000 km, storms can be observed over most of France, the Alps, northern Italy, the western Mediterranean Sea, the Iberian peninsula and the Biscay.

Figure 14. The OMP observatory and its location in southern France (photo : Observatoire Midi-Pyrénées).

An optical video camera system was deployed by the team of the Danish Space Research Institute (figure 15). During the campaign, it was remotely controlled via the internet from Denmark. The system included trigger software for automated sprite detection. The station consisted of two JAI CV- S3200 low-light CCD cameras, based on the Sony ICX248AL 1/2′′ monochrome Ex View HAD sensor. A CCD camera uses a small, rectangular piece of silicon rather than a piece of film to receive the incoming light. The silicon element is called a charge-coupled device (CCD). The CCD is segmented into an array of individual light-sensitive cells called —photosites“. Each photosite is one element (—pixel“) of the whole picture that is formed. They transmit their light- measurements directly to the computer without involving any film or chemicals. The cameras were equipped with two

Figure 15. The two low-light level CCD video identical 16 mm F 1.4 lenses (FOV of 20 cameras (photo : Observatoire Midi-Pyrénées). degrees), and provided both composite and S-video outputs. Also, the camera settings, such as the gain, shutter and integration period could be adjusted over serial RS232 links. The cameras were mounted in a weatherproof housing on top of a QuickSet 20 motorized pan/tilt unit. The pan/tilt unit featured a two-axis control interface, which was accessible through a serial RS232 link to a control PC one floor below. The unit allowed for the pointing of the camera within 360° of azimuth, and from œ35° to +35° of elevation.

22 The two video feeds from each camera were connected to video interface cards in the control PC. This PC was configured with : - One Intel Pentium IV 2.5 GHz processor - 1024 MB RAM - 120 GB internal hard disk capacity - 260 GB external hard disk capacity - Network interface - Two Pinnacle PCTV Rave Video Acquisition cards, with the Conexant BT878 chipset, one S-video and one composite video input on each - One Sensoray 616 MPEG-II encoder with one composite and one S-video input - One dual-port RS232 PCI interface card.

The data were : - Pan images taken every minute from the composite input of the first PCTV card - Event images from the S-video input from the second PCTV card. The event images were uploaded to the campaign site every 5 minutes during observations - MPEG Video sequences, stored at the external hard disk. The videos were recorded using the Sensoray card. The video files consistent with events were uploaded at 5.30 (UTC) every morning.

All images and video files were time stamped using the PC system time, that was synchronized to UTC time through the Network Time Protocol (NTP). The PC clock was synchronized every 10 minutes.

In principle, the system time was supposed to be correct within 0.006 seconds, with an almost constant drift of 0.012 seconds per 10 minutes. However, after the campaign, Tomas Allin, of the Technical University of Denmark, pointed out that there appeared to be a constant +40 ms offset in the event timing, due to a misinterpretation of video frame time stamps by the trigger software. This offset has been taken into account in this present work.

4.2 ARAMIS

The French territory is covered by the meteorological radar network ARAMIS (Application Radar la Météorologie Infra-Synoptique), operated by Le Centre de Météorologie Radar (CMR) within Météo-France. It consists of 18 conventional radars, of which 10 are C-band radars (λ = 5 cm) and 8 are S-band radars (λ = 10 cm). Each radar has a range of approximately 250 km and produces images every 5 minutes. In order to compose one Plan Position Indicator image, the radar beam makes two complete scans with an elevation of respectively 1.4° (for low distance) and 0.8° (for large distance). Thus, with account taken for the rotundity of the Earth, the cloud is, at the maximum distance from the radar, scanned by the radar at a height of about 8 km. The parameter provided by the radar is the reflectivity factor directly related to the precipitation rate. The computer system Castor commands the radar, supervises it and processes the data. The software Sycomore then concentrates the data to Météo-France in Toulouse and generates composite images with a resolution of 1.5 × 1.5 km. These images are then enriched with radar images from other European countries thanks to the program OPERA (Operational Program for the Exchange of weather RAdar information) within the Eumetnet1.

1 Network grouping 18 European national meteorological services.

23 4.3. Météorage

The French detection network for cloud-to-ground lightning, Météorage, covers the entire country as well as parts of the adjacent territories. It comprises 17 regularly spaced (200-250 km) sensors that work in conjunction with sensors in neighbouring countries. The Météorage stations use a technology called IMPACT (IMProved Accuracy from Combined Technology) that combines the advantages of a magnetic direction finding system (LLP) and a time of arrival system (LPATS), (Cummins et al., 1998).

The oldest and most widely used location system is the LLP, named after its original manufacturer, Lightning Location and Protection Inc. Each station is equipped with a direction finder (DF) and a flat plate antenna that determines the field polarity. The DF consists of two crossed magnetic loop antennas which detect the north-south and west-east component of the magnetic field radiated by the return stroke of the CG flash, the most energetic phase in terms of low Figure 16. Météorage detector (Météo-France). frequency electromagnetic fields (1- 300 kHz). By goniometry, each station provides the direction of the electromagnetic wave generated by the source, i.e. the direction from the station to the ground-strike point. W hen the same signal is detected by several stations, the location of the ground-strike point can be estimated by triangulation. A shape recognition procedure identifies the signal waveform generated by the return stroke in order to distinguish between CG discharges and other signals like intra- and intercloud lightning. (Soula and Chauzy, 2001). The quality of detection of the magnetic signal is sensitive to distortions of the magnetic field by obstacles and metallic objects. Hence, the installation site must be carefully chosen. The detection range for lightning signals is about 200 km or more (Finke and Kreyer, 2002).

LPATS (Lightning Position and Tracking System) measures the time of arrival (TOA) of the leading edge of the lightning impulse. Using bandwidths extending to a few megahertz, the leading edge can be determined to within 1 µs or better. The exact time of the signal‘s arrival at the antenna stations is compared for the different locations. The position of the signal source is given by the intersection of the hyperbolas for equal time differences. To determine the 3 unknown parameters (time, longitude, latitude) of the signal source, at least 3 independent measurements are necessary. Due to the ambiguity of the intersection of 2 hyperbolas, in the general case 4 measurements of arrival time are needed. The uncertainty in the determination of the time is in the order of the time uncertainty of the single sensor (∆ti). The location uncertainty inside the network is in the order of ∆ti ⋅ c. Hence for location uncertainty lower than 1 km, a time resolution higher than 3 µs is necessary (Finke and Kreyer, 2002). To avoid reception of the skywave (the propagation of radio waves by reflection against the inner surface of the ionosphere), receiver sites must be less than a few hundred kilometers apart.

24

Figure 17. Lightning location principles : TOA - Time Of Arrival (left), DF œ Direction Finding (right) (from MacGorman and Rust, 1998).

Since 1997, the Météorage uses both these techniques of detection. The IMPACT system combines magnetic and electric field measurements in the frequency range 0.4 kHz - 400 kHz. The basic principle is the same as before, but IMPACT is synchronized by GPS signals, which keeps the absolute timing error between sensors smaller than 300 ns. The system provides information of the location in 2-D, the time of occurrence, the polarity, the multiplicity, and the peak current of each individual return stroke. The detection efficiency is around 90% (Morel and Sénési, 2000), and about 70% of the locations are determined with an accuracy of < 4km (Seity, 2003).

4.4 SAFIR

The SAFIR (Surveillance et Alerte Foudre par Interférométrie Radioélectrique) lightning location system was originally developed by a research group of the French National Aerospace Research Agency (ONERA) in a research and development program about lightning prevention and storm forecasting for aeronautics, defence and aerospace applications.

SAFIR operates at a detection frequency in VHF in the worldwide protected frequency band at 110-118 MHz. This high frequency allows to detect signals from the single step of the leader process and from recoil streamers of cloud discharges. These processes emit presumably in the high frequency range. Another consequence of the VHF range is the line-of-sight propagation of the lightning signals. Thus, the Earth curvature limits the detection range of the antennas. A ray arriving at a detection station from 200 km distance was emitted from an altitude of at least 3 km. The antenna represents an array of 5 electric dipoles comprising an interferometer. Additionally to the VHF registration, a second antenna at LF (300 Hz œ 3 MHz) is used for the detection of return stroke signals. The combination of both antennas enables the discrimination between intracloud and cloud-to-ground lightning (Finke and Kreyer, 2002). Figure 18. SAFIR detector (Météo-France).

25 For total coverage, a minimum of three stations is required. Each station detects, by interferometric technique (based on the phase difference between waves received by several antennas), the direction of the electromagnetic wave generated by the leader phases of any cloud discharges. In the 2D configuration used for the permanent networks installed in France, the detection of the same signal by several stations leads to the location in the horizontal plane of the radiation sources. The time resolution is 100 µs (Soula and Chauzy, 2001).

The mean distance of stations in SAFIR networks is at about 80-100 km. W ith the angular resolution of 0.25°, this yields a location accuracy better than 1 km. The detection efficiency is around 90% for ICs and CGs for distances smaller than about 150 km. The detection capacity is limited currently to 100 events per second. Since SAFIR detects many signals from a single lightning channel, this limit can be reached easily during intense storms. This low value was chosen to meet the slow data transmission lines (Finke and Kreyler, 2002).

All detected VHF emissions (called —point sources“) are classified into different categories according to certain spatial and temporal criteria. For the IC activity, we have four groups :

0 = Isolated IC point 1 = Start of IC lightning 2 = Intermediate point 3 = End of IC lightning

The spatial criterion has been fixed to 4 km. This means that if the distance between two point sources is greater than 4 km, they are not considered as being parts of the same lightning discharge. The temporal criterion corresponds to the maximal duration of a lightning discharge. In SAFIR, this limit has been fixed to 500 ms. An isolated point, is a point source that cannot be associated to other point sources according to the distance-time criteria. The groups 1, 2 and 3 are supposed to illustrate the progression of a lightning discharge. However, the category n° 2 (intermediate points) can vary only between 0 and 98, since the storage capacity is limited to 100 points per second (Defoy, 1998).

Initially, there were three SAFIR systems working over the French territory, one over the Paris region, one on the Atlantic Coast, and one in the South-East. Today, only two of them are left, because the system in Paris is no longer active. For the thunderstorm of the 23 July 2003, we used the data produced by the south- eastern system. Unfortunately, one of the three stations was not working. This had the consequence that the coverage was incomplete and that a

—blind spot“ appeared in the Figure 19. Coverage of the south-eastern SAFIR system the summer 2003 with the —blind spot“ (Météo-France).

26 center of the area. However, except maybe for the case of one sprite, we do not believe that this partial loss of data significantly affected our results (cf. 5.4).

4.5 Meteosat

Eight Meteosat satellites have been launched to this day, the first in November 1997 and the most recent in August 2002. The project was developed as the European Space Agency‘s contribution to the Global Atmospheric Research Program (GARP) to provide global data sets used in improving weather forecasting. The satellites are placed in a geostationary orbit at 36 000 km altitude, in the so called —Clarke belt“1. Presently, there are four Meteosat satellites operating : n° 5, at longitude 63° E over the Indian Ocean, and n° 6-8, located over Europe, close to the Greenwich Meridian.

The radiometer is the principal payload of the satellite. It provides the basic data of the Meteosat system and operates in three bands of the electromagnetic spectrum : - 0.4 to 1.0 µm visible band (VIS) - 5.7 to 7.1 µm water vapour absorption band (W V) - 10.5 to 12.5 µm thermal infrared band (IR)

The communication package aboard these satellites consists of a transponder and its antenna subsystem. It transmits digital image data every half hour to the ground facilities located in the European Satellite Operations Center (ESOC) in Darmstadt, Germany, where the image data is processed.

4.6 Schumann resonance measurements

In order to calculate the charge moments of the sprite-associated CGs, the CAL project includes measurements of electromagnetic radiation in the lower ELF (Extremely Low Frequency) band. These are carried out by the Geodetic and Geophysical Research Institute at Sopron in Hungary, using a detection station located in Nagycenk (47°38′N ; 16°43′ E).

ELF (3 Hz œ 3 kHz) and VLF (Very Low Frequency : 3 kHz œ 30 kHz) transient signals are generated by various natural phenomena (volcanic eruptions, dust storms, tornadoes… ), but the far most significant source of —noise“ in these frequencies is the lightning activity. In fact, lightning discharges radiate the bulk of their electromagnetic energy in these low frequencies. The discharge currents generate transient radio pulses termed —atmospherics“ or —sferics“. In these low frequencies, the attenuation suffered by globally propagating electromagnetic waves is extraordinarily small. It amounts to only 0.3 dB/1000 km at 10 Hz, increasing with frequency to about 1 dB/1000 km at 60 Hz, which allows the waves to propagate a few times around the globe before dissipating (Barr et al., 2000).

Between the conducting terrestrial surface and the lower boundary of the D-region of the ionosphere, a resonating cavity is formed, a spherical parallel-plate waveguide, commonly referred to as the Earth-ionosphere waveguide. The sferics are —trapped“ in this waveguide by multiple reflections from the ground and the lower ionosphere, and at certain frequencies,

1 After the writer Arthur C. Clarke who, in 1945, worked out that an object in this position in space would appear motionless from the Earth‘s surface.

27 constructive interference occurs, which generates quasi standing electromagnetic waves, called Schumann Resonances (SR).

In very simple terms, the circumference of the Earth being about 40 000 km, and the speed of electromagnetic radiation 300 000 km/s, the fundamental frequency should be roughly f = c/λ = 7.5 Hz. In practice, SR depends on many factors, and the actually observed fundamental frequency is approximately 7.8 Hz, with the following higher order resonances occurring at 14.2, 19.6, 25.9 and 32 Hz. The low attenuation rate of frequencies in the ELF range allows the SR to be prominent in sferic noise spectra and enables the radiation from intense individual lightning flashes to be detected at global ranges. Powerful lightning discharges produce a large signature in the lower ELF band that appears as a transient pulse (called —Q- burst“) in a waveform record (Barr et al., 2000).

The ELF measurements can then be used to derive the charge moment of the lightning discharge that caused the Q-burst. This is not easy, and we will not here go into details. It involves computing the wave impedance, which is independent of the complexities of the lightning source characteristics and thus contains only information of the propagation characteristics (the attenuation constant and the phase velocity) and the source-observer distance. The direction of the flash can be determined from the time evolution of the horizontal magnetic field components, the 180° ambiguity being resolved by calculating the Poynting vector, which points away from the source. Given the direction and the distance, the source can be uniquely located on the Earth. The electric and magnetic spectra for the recorded event can then be compared with the theoretical spectra to estimate the source current moment and eventually the charge moment1.

1 This is far more complicated than this brief summary suggests. A more detailed description can be found in : Huang et al. (1999).

28 5 Case study

5.1 The meteorological situation

On the 23 July, France had already experienced stormy weather conditions for several days. This was due to a talweg that had been moving from the Atlantic Ocean in over the Iberian Peninsula heading towards the Pyrenees (figure 21). The talweg generated a southwesterly wind flow over the whole Central Europe. The scorching heat of the summer 2003 that is now so tragically famous for having killed so many people, had not yet got its grip on the country, but the temperatures had risen to well over 30° C during the afternoons in many regions in the South of France and approached even 40° C on W ednesday 23 July near the city of Avignon (figure 20). The hot air produced instability and lead to a stormy development over the Pyrenees in the afternoon, and later over the Central Massif and over the Alps.

Figure 20. Maximum temperatures the 23 July 2003 (in °C).

Figure 21. 500 hPa geopotential height chart from the 23 July 2003.

29

Figure 22. Sea level analysis from the 23 July 2003.

At 2000 UTC on the evening of the 23 July, as can be seen on the ARAMIS radar image (figure 23), the perturbation was roughly located over the French region Languedoc- Roussillon. During the night it then moved in over South-Eastern France.

Figure 23. ARAMIS radar image, South of France, 2000 UTC, the 23 July. 2003. The color scale to the right indicates the ranges of reflectivity expressed in dBZ.

30 During the evening, the most active region of the storm split into two parts, one in the north, the other in the south of the radar reflectivity area (figure 29). The former tended to fade out progressively after 2130 (UTC), while the latter started weakening after 2300 (UTC), although a small part of the convective system stayed surprisingly active till just after midnight.

5.2 Observed sprites

Between 2111 and 0033 UTC, a total of 13 sprites were recorded. Most of them had carrot- shaped forms (figure 24) and appeared sometimes in clusters. Others were larger, more diffuse and seemed to consist of groups of vertically aligned striations (figure 25).

Figure 24 Sprite at 2111 (UTC) Figure 25 Sprite at 2312 (UTC)

A peak in the sprite production was reached around 2200 UTC when 5 sprites occurred within 17 minutes. The last three sprites were observed within ten minutes about half an hour after midnight, at a very late stage of the storm when the electric activity was extremely low. The total repartition of the sprites is seen in figure 26.

21:00 21:20 21:40 22:00 22:20 22:40 23:00 23:20 23:40 00:00 00:20 00:40 (time UTC)

Figure 26. Temporal repartition of the 13 sprites.

Even if the total number of events is rather low, it is tempting to see some kind of tendency for periodicity in the appearance of the sprites. As reported in other documents (Final report, 2000), once the sprites start, they tend to occur at quasi-periodic intervals. It is also worth noticing that the sprites began about two hours after sunset (local sunset took place at 1916 UTC). According to previous studies, there seem to be a general tendency for sprites not to occur until at least one hour after darkness, perhaps in response to some changes in the ionosphere after the passage of the terminator (Lyons et al., 1999).

31 Since we had only one spot of observation, it was of course not possible to locate the sprites by triangulation. W e could only determine the direction of each sprite seen from the Pic du Midi. However, for all but one of the sprites (as we shall see, cf. 5.3), we did have the location of the parent positive flash(es), which seemed to offer a fairly good approximation, at least for the estimation of the vertical and horizontal extent of the sprites. In fact, this is the method suggested by Lyons (1996) and Final report (2000) to obtain estimates of the vertical extent of sprites from single-image sprite detections. It is based on the assumption that the sprite is generally centered within 50 km of the parent flash. W hen we had multiple parent flashes, we took the mean distance. For sprite n° 2 we did not have any detected parent flash. But it occurred in almost exactly the same angle as sprite n° 1, and since the storm pattern was roughly the same, we found it reasonable to use the same distance as for the previous sprite event.

Our results are listed in appendix 1 and illustrated in figure 27. The mean vertical extent was 27 km (10.1 standard deviation). The mean top was located at 91 km altitude (8.6 standard deviation) with a range from 73 to 104 km. The mean detectable lower limit (often hard to distinguish on the images) was at 64 km altitude (standard deviation 9.2) with a range from 45 to 76 km. Most of the top values fit rather well with those of Sentman et al. (1995), 88±5 km, but are a bit higher than in Lyons (1996). The highest values around 100 km may be a bit surprising, since we have found no other reports of sprites exceeding 100 km. These high values can of course be due to the uncertainty of the measurements, but Final report (2000) points out that different sprite altitudes might be a consequence of seasonal variations of the ionospheric height. Thus, we cannot exclude that they are real.

120

100

80 ) m k (

e

d 60 u t i t l A

40

20

0

2 8 8 1 8 3 5 5 5 6 6 2 4 1 8 7 3 :3 :5 :5 :1 :2 :0 :3 :3 :3 :0 :0 :0 :3 :4 :2 :2 :0 1 4 4 2 6 9 5 5 5 9 9 1 2 1 3 8 3 :1 :3 :3 :5 :5 :5 :0 :0 :0 :0 :0 :5 :1 :2 :2 :2 :3 1 1 1 1 1 1 2 2 2 2 2 2 3 3 0 0 0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 0 0 0 Time (UTC)

Figure 27. Vertical extent of the sprites. Sprites appearing with identical time indications correspond to multiple sprites.

The sprite aspect ratio (horizontal extent divided by vertical extent) varied from 0.12 to 1.47 with a mean of 0.50 (standard deviation 0.41), which means that the typical sprite in this storm was twice as high as it was wide. But this seems to be very variable ; Lyons (1996) finds a mean ratio of 1.84, the lowest being 0.28 and the highest 5.0.

32 One often raised question, because of its fundamental importance for understanding the role of sprites in the global atmospheric electric system, is if the tendrils can reach the underlying cloud tops (Lyons, 1996 ; Neubert et al., 2001 ; Final report, 2000). Pasko et al. (2001) report that this can sometimes be the case. Unfortunately, it is not possible from our images to tell whether this is the case or not. The reason could be, as noted by Sentman et al., (1995), that the red color of the tendrils just below the collar is gradually merging into blue with distance downward, a color to which the camera is not very sensitive.

The brightest portion of the sprites is believed to be the origin of the electrostatic breakdown causing the phenomenon. For the diffuse sprites, it is not always easy to identify this part, but the carrot shaped sprites are much easier to deal with. W e found their brightest portion to be located between 63 and 81 km with a mean altitude of 74 km, which is very consistent with the altitude (75 km) observed by Stanley et al. (1999) and the range (66-74±4 km) found by Sentman et al. (1995).

The sprites had a clear tendency to concentrate in a relatively small segment of the storm, or at least we suppose they did, because they occurred in more or less the same direction seen from the Pic du Midi. Almost 85 % of the sprites were within 5° azimuth of its predecessor, and all of the 7 first sprites appeared within 3.9° azimuth. Between 2209 and 2251 (UTC), the azimuth increased more than 10° because the sprites began to appear over the southern storm system, and ceased to appear in the North.

Most of the sprites ranged in one single video field, which means that the duration was less or equal to 20 ms. However, some exceptions exist. The sprites n° 5 (2159 UTC) and n° 7 (2209 UTC) are seen on two video fields, and the sprite n° 10 (2321 UTC) on as many as 6 video fields, which should mean that its duration was somewhere between 80 and 120 ms. This is not at all exceptional ; Lyons (1996) found sprite events lasting as long as 16 video fields and adds that the longer events were clusters of sprite elements forming and dissipating, possibly in response to a long lasting intracloud electrical discharge. W e observe that the sprite n° 10 does have the character of a cluster (figure 28) and that, furthermore, it had 3 associated +CGs and coincided with a very intense intracloud activity (cf 5.4).

Figure 28. Sprite n° 10, 2321 (UTC).

33 5.3 The sprites in relation to the cloud-to-ground lightning

Between 2111 and 0034 UTC, 1475 negative cloud-to-ground (-CG) and 209 positive cloud- to-ground (+CG) were detected in a quadratic area with a side length of 400 km and the center located in the south of the Department of Ardèche, at 44.5° N and 4.5° E. (Hereinafter, we will use this area as a reference frame when considering the development of the storm.) This means that in average we had one sprite every 113 œCGs and every 16 +CGs.

As we can se on the radar image just before the beginning of the sprite producing period (figure 29), the œCGs were largely concentrated in two high-reflectivity cores separated from each other by some 100 km.

200

150

100

50

0 -200 -150 -100 -50 0 50 100 150 200

-50

-100

-150

-200

Figure 29. Radar image at 2100 (UTC) with œCGs and +CGs detected ±5 minutes of radar image time. Distances in km. The color scale to the right indicates the ranges of reflectivity expressed in dBZ.

Out of 13 sprites, 12 (or 92%) were temporally associated with Météorage detected +CGs. The criterion used was the time interval 500 ms before and after each sprite. 4 (or 33%) of these 12 sprites were associated with multiple +CG events (double or triple), which is very close to the percentage (35%) observed over the U.S. High Plains by Lyons (1996), who found a discernible tendency for the sprite +CGs to have multiple events. If we define multiple events as CGs occurring within one second, 16% of all the multiple +CGs produced sprites, while 9.1% of all the +CGs produced sprites, which seems to confirm the observation of Lyons. Only one of the sprite +CGs had a second return stroke (during the sprite-producing period, 11.1% of all +CGs had a second return stroke).

As shown in table 1, all single sprite +CGs occurred before the sprite. In some of the multiple events (sprites n° 7, 9 and 10), one or even two of the +CGs were detected after the sprite. If we count only the +CGs registered before the sprite, we find an average lag time of 121 ms between the positive flash and the sprite.

34 Table 1. Detected sprites and associated cloud-to-ground flashes.

35 W e should mention that among the associated +CGs, we ruled out one flash that was temporally close to the sprite n° 12 (0028 UTC). It was detected 322 ms after the sprite at latitude 44.2950 and longitude 9.7129, which means, given our detected direction of the sprite from the Pic du Midi, that it was located at least 300 km from the event. This seemed quite unrealistic.

According to Final report (2000), the brightest sprite events tend to have a shorter lag time from the parent +CG. This is not confirmed by our observations. It is true that the sprites n° 1, 6 and 10, which to us appeared to be the brightest (sprite n° 2 was also very bright, but it was not associated to any detected +CG), had the lag times 59, 62 and 38 ms, which is clearly lower than the average value of 121 ms. But the last three sprites, n° 11-13, were all of a weak luminosity and had among the smallest lag times : 28, 33 and 47 ms.

Regarding the œCGs, only three flashes corresponded to the chosen criterion, and all three of them occurred more than 200 ms after the sprite, which might indicate that they had nothing to do with the sprite. Anyhow, we believe that this supports the hypothesis that the vast majority of sprites are associated with +CG events (Lyons, 1996 ; Winckler et al., 1996 ; Reising et al., 1996 ; Final report, 2000). However, as demonstrated by Barrington-Leigh et al. (1999), -CGs can sometimes trigger sprite events, and we find no reason to automatically assume that sprites always are connected with +CGs. Lyons et al. (2003a) point out that the polarity is not decisive in itself, but that the triggering of sprites requires transport of large amounts of charge to Earth in unusually large and long-lasting continuing currents, and that +CGs are fare more likely than œCGs to perform this transport (cf. 2.4.3. ).

Sprite n° 2, at 2211 (UTC), had no detected parent CG at all (of any polarity). Of course, this does not necessarily mean that there was no such flash, since the reliability of the lightning detection system is not 100% (cf. 4.3). However, as we shall see later (cf. 5.4), it is interesting that this sprite was temporally associated with a rather intense intracloud activity.

Previous studies (Lyons, 1996 ; Winckler et al., 1996 ; Final report, 2000) have found a certain relationship between the sprite production and the peak current of the +CGs. In our case, the average peak current of sprite +CGs was 51.3 kA, which is clearly higher than the average peak current for all +CGs during the sprite producing period (40.6 kA). So, yes, most likely there is some kind of connection between them. But there was a very wide range in values for the sprite +CGs, the lowest value being 22.5 kA and the highest 147,5 kA with a standard deviation of 29.2 (Final report, 2000, even find sub-10 kA +CGs triggering sprites). In fact, as much as 55% of the sprite-related +CGs had peak currents under 50 kA. As remarked by Lyons (1996), this suggests that unusually large +CG peak currents may not be a necessary condition for sprite formation. On the other hand, large amplitude +CGs does not seem to be a sufficient condition for sprites either ; 9 +CGs with peak currents over 100 kA (one of them reached 217 kA) occurred without producing any detected sprite.

The average peak current for the œCGs during the sprite producing period was œ16.6 kA, the smallest being œ2.6 kA, and the most powerful œ96.8 kA. The 5 minutes average value stayed relatively stable with some rare bursts without apparent relation with the sprites.

Our average peak current for sprite +CGs is considerably lower than the value reported by Lyons (1996) from the U.S. High Plains (81 kA). But this difference might not be real. Neubert et al. (2001) have brought the attention on the different methods used in U.S. and in Europe for deriving the peak currents. In Europe, the currents are calculated assuming infinite

36 ground conductivity (since the ground conductivity varies so much with location anyway), but in the U.S., the National Lightning Detection Network (NLDN) adopts a finite conductivity value. As a result, European peak currents are typically a factor 1.6 lower than the ones in the U.S., which, by the way, almost exactly corresponds to the difference between our value and the U.S. value.

W e had measurements of the charge moments regarding 9 of the sprite-associated +CGs (cf. appendix 2). In most cases, they were calculated in two different manners : one from the electric field component Ez and the other from the magnetic field component Hϕ. For some +CGs, only one of the two methods was possible. W ith some exceptions, the differences between the two calculations are relatively small (∼30-50 C km). However, for the +CG associated to sprite n° 12 at 0028 (UTC), the E-field measurements yielded 501 C km, and the H-field measurement 1011 C km. The difference between them is so large that none of the values seems reliable. If we exclude this +CG, we find, for the E-field measurements, an average value of 513 C km (standard deviation 216 C km), and for the H-field measurement, 520 C km (standard deviation 256 C km). This is consistent with earlier estimations of sprite- producing flashes ; Cummer and Stanley (1999) measured charge moments between 150-1100 C km for the associated +CGs, and Huang et al. (1999) 200-2000 C km.

Time Area (km²) Several previous studies have found a relation between the 20:00 38713 sprite production and the area of the radar echo (Final 20:15 39159 report, 2000 ; Lyons, 1996 ; Lyons et al., 1999 ; 2003). 20:30 39402 Sprites seem to appear almost exclusively over storm 20:45 36726 systems with large radar areal coverage. A threshold value 21:00 35975 of around 7500 km2 has been estimated (Final report, 21:15 31981 2000). Above this value, the number of TLEs is modestly 21:30 33131 correlated with the echo size. Under this value, according to 21:45 35988 previous observations, sprites occur only exceptionally. 22:00 36159 22:15 37152 22:30 38043 W e measured the total area of the radar echo ≥ 16 dBZ 22:45 37773 between 2000 and 0045 (UTC) of the two storm systems 23:00 33637 together (table 2) As we can see, the surface remains 23:15 30890 between 30000-40000 km2 until about 2315 (UTC), when it 23:30 26797 commences to decrease rather rapidly. At 0030 (UTC), we 23:45 22502 fall below the 7500 km2 which corresponds fairly well to 00:00 12795 the prediction, since we had our last sprite event at 0033 00:15 11400 (UTC). Thus, our observations are in good accordance with 00:30 6808 2 00:45 3757 the approximate threshold value of 7500 km .

Table 2. Radar areal coverage.

W hat is there to say about the frequency of the CGs ? Is it in some way connected to the sprites ? At first sight, the situation appears a bit discouraging. The œCG frequency presents a pattern with several maxima but with an overall decreasing trend throughout the sprite period. Unfortunately, no relation whatsoever with the sprites seems to be in sight. But if we separate the two high-reflectivity cores and examine the flash rates in each one of them separately, things look a bit different.

37 5.3.1 The northern storm system

The images from the Pic du Midi indicate that the sprites n° 1-7 occur in the northern part of the storm system, and from the œCG rate we can see that these TLEs clearly coincide with a late phase of the electric activity in this area (figure 30). The number of negative flashes is falling rapidly and somehow the sprite production seems to get started at this particular moment. This is in agreement with Boeck et al. (1995), who, in their study of the space shuttle videos, pointed out that the typical TLEs are observed in a thunderstorm cell that exhibits a low to moderate flash rate. And on numerous occasions, it has been noticed that sprites tend to occur during the mature-to-late stages of the storms, often after +CGs have been occurring for several hours (Lyons, 1994 ; Lyons et al., 2003a ; Boccippio et al., 1995). Since the associated +CGs tend to have a rather high peak current, this suggests that the sprite may appear after an especially energetic discharge within a storm cell otherwise exhibiting low flash rates (Final report, 2000).

120

100

80

-CGs 60 sprites

40

20

0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 0 0 0 0 0 0 1 1 1 1 1 1 2 2 2 2 2 2 3 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Time (UTC)

Figure 30. Number of œCGs every 5 minutes in the northern part of the storm and the sprites occurring in this region.

Now, if we examine the +CGs in the northern part of the storm, we observe that they do not decrease like the -CGs. On the contrary, their number even increases slightly after 2100 (UTC), most of them occurring in the large stratiform area. And since the number of œCGs is decreasing, the percentage of positive flashes is rapidly growing (figure 31). And we notice that this is the moment when the sprites begin to appear, as if there was some kind of connection between them and the ratio of positive flashes in the storm.

38 100,00

90,00

80,00

70,00

60,00

% +CGs 50,00 sprites

40,00

30,00

20,00

10,00

0,00

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 0 0 0 0 0 0 1 1 1 1 1 1 2 2 2 2 2 2 3 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Time (UTC)

Figure 31. Percentage of +CGs every 5 minutes in the northern part of the storm and the sprites occurring in this region.

Indeed, this seems to confirm 200 that there is some kind of Angle sprites n° 3-7 connection between the sprites (or at least the 150 majority of them) and the +CGs. But as was the case 100 for the peak currents, the Sprite producing +CGs relationship is by no way 50 simple. W hy doesn‘t the +CG rate between 2225 and 0 2250 (UTC) generate any -200 -150 -100 -50 0 50 100 150 200 sprites, while the +CG rate between about 2100-2210 -50 (UTC) appears to do just that? At this point, it might -100 be helpful to examine the exact location of the sprite -150 producing +CGs in the thunderstorm. Let us take a look at the sprite maximum -200 around 2200 UTC (figure 32), when 5 sprites occurred Figure 32. Radar image at 2200 (UTC) with œCGs and +CGs ±10 within 17 minutes and within minutes. The direction of the 5 sprites, n° 3-7, that occurred 2152- a narrow angle (3.9°) from 2209 (UTC) are all between the two dotted lines. The black, encircled crosses indicate the +CGs associated to the sprites. the Pic du Midi. Distances in km.

39 As we can see, the sprite producing +CGs are concentrated in a rather small portion of the stratiform region, all of them to the south of the sprite. This has often been observed (Final report, 2000 ; Lyons, 1996) and most likely means that parent +CGs have other important characteristics than high peak current, characteristics that may depend on specific cloud dynamical and microphysical mechanisms. According to Lyons (1996), this could indicate the presence of some sort of —sprite generating cell“ within in the stratiform region.

W e also notice that the sprites seemed to occur somewhere between the parent +CGs and the dissipating high-reflectivity core in the northern part of the storm. The previous made assumption (cf. 5.2) that the sprite is centered within 50 km of the parent flash appears to be valid, or is at least not contradicted by the sprite directions from the Pic du Midi. The lateral distance of the +CG location to the sprite is generally believed to show that —dendritic“ or —spider“ lightning is radiating outward away from the storm‘s interior (away from the convective core), a phenomenon frequently observed above the U.S High Plains (Lyons, 1996). Spider discharges are horizontally stratified lightning channels propagating over large distances near the cloud base (Mazur et al., 1998). They owe their name to their highly branched nature and their propensity to —crawl“ along the cloud base at a visually detectable speed. They are part of intracloud flashes and +CGs occurring prior to and during the inverted (fair weather polarity) phase in the decaying stage of a storm. If this kind of lightning plays a key role in the generation of sprites, it could explain why sprites tend to appear in the mature or late phase of the storm. W e will come back to the matter of spider lightning in 5.4.

W e will now return to figure 200 31 and take a closer look on +CGs not producing sprites the second maximum in the +CG percentage that 150 occurred between 2225 and 2250 (UTC) and that 100 apparently failed to produce any sprites. W e can see them in figure 33. One cannot 50 avoid being struck by the narrow clustering of these 0 +CGs in a small area, very -200 -150 -100 -50 0 50 100 150 200 resembling to the sprite producing group of +CGs -50 that we saw in figure 32. Some of them were powerful, the maximum peak -100 current being 162.2 kA and the average 62.1 kA -150 (incidentally, the average peak current of the sprite producing group of +CGs in -200 figure 32 was only 41.8 kA). But unlike the other group, Figure 33. Radar image at 2240 (UTC) with œCGs and +CGs this one did not produce a ±10 minutes. The group of +CGs mentioned in the text is single sprite. encircled. Distances in km.

40 One may suspect that this had something to do with its location in the storm system. The sprite +CGs occurred in the large stratiform anvil region, while the others were much closer to the dissipating center in the northern part of the storm. This is particularly obvious if we look at the infrared satellite images (figure 34).

(a) 200 (b) 200

150 150

100 100

50 50

0 0 -200 -150 -100 -50 0 50 100 150 200 -200 -150 -100 -50 0 50 100 150 200

-50 -50

-100 -100

-150 -150

-200 -200

Figure 34. Meteosat infrared images. The color scale to the right indicates the range of temperature expressed in °C. (a) 2200 (UTC) with œCGs and +CGs ±15 minutes. The sprite producing +CGs are encircled and colored in black. The 5 sprites, n° 3-7, 2152-2209 (UTC) occurred between the two dotted lines. (b) 2230 (UTC) with œCGs and +CGs ±15 minutes. The non sprite-producing +CG-group is encircled. Distances in km.

W e can see that the sprite-producing group were located in the area with the coldest, and hence the highest, cloud tops (between œ56° and -48° C), while the other group occurred in a zone where the cloud tops were significantly lower (temperature range between œ40° and -30° C, or even close to the range between œ30° and œ20°C). Furthermore, at 2230 (UTC), the high-cloud region in the northern system had almost disappeared. Our closest soundings are from the airport of Lyon, Lyon-St-Exupéry, at 1200 (UTC) the 23 July and 0000 (UTC) the 24 July (appendix 3), and they indicate that the coldest cloud tops reached the tropopause, located at about 11 000 meters altitude. This was not the case in the area where the non sprite- producing +CGs appeared.

As for the 5 sprites between 2152-2209 (UTC), we strongly suspect, given the location of the sprite +CGs with respect to the sprite-observation angles, that they also occurred in the high- cloud region. This characteristic is confirmed by Final report (2000), according to which it is apparent that the sprite +CGs occur in the part of the storm which has the coldest cloud tops. In the storm studied in this report, from August 1999 over the U.S. High Plains, it was found that no sprites appeared with cloud tops warmer than œ60° C.

The satellite images from 2100 and 2130 (UTC) suggest that the sprites at 2111 and 2134 also took place over the coldest part of the cloud tops. At 2100 (UTC), we even had a small cloud area with a temperature range between œ64° and œ56° C, very close to the sprite direction observed from the Pic du Midi.

41 After 2245 (UTC), there were almost no flash activity in the northern part of the storm and the radar reflectivity surface is clearly decreasing.

5.3.2 The southern storm system

At first glance, the situation in the southern part of the storm is different because the -CG activity is not really fading out before the appearance of the sprites. On the contrary, it shows a local maximum just before midnight, after the sprites n° 8-10. But if we look closer on the geographic repartition of the œCGs, we find that this late, local maximum is almost entirely due to a very small but highly intensive cell of negative discharges located in a very peripheral region of the system. In fact, it is somewhat separated from the southern part of the storm (figure 35).

200

150

100

50

0 -200 -150 -100 -50 0 50 100 150 200 Intense œCG cell

-50

-100

-150

-200

Figure 35. Radar image at 2345 (UTC) with œCGs and +CGs ±10 minutes and the small, intense œCG cell. Distances in km.

If we exclude these distant flashes from the data, the temporal relation between the sprite events and the œCG activity is quite similar to the one in the northern part of the storm ; the sprites tend to occur during the late, decaying stage of the storm (figure 36). W e first have a smaller, temporary decrease in the number of -CGs around 2130 (UTC), but this do not generate any sprites. However, when the —real“ final phase of the electric activity starts, the sprites make their appearance. They are temporally separated into two groups : n° 8-10 occur between 2251 and 2321 (UTC), and n° 11-13 between 0023 and 0033 (UTC).

42 90

80

70

60

50 -CGs sprites 40

30

20

10

0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 0 0 0 0 0 0 1 1 1 1 1 1 2 2 2 2 2 2 3 3 3 3 3 3 0 0 0 0 0 0 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0 0 0 Time (UTC) Figure 36. Number of œCGs every 5 minutes in the southern part of the storm (without the small, peripheral cell mentioned in the text) and the sprites occurring in this region.

The first three sprites (figure 37, a-c) have in common that they are all associated with multiple +CG flashes (n° 8 double, n° 9 and 10 triple). All but one of the +CGs are, as before, located to the south of the sprite. W e notice that the range of the sprite azimuths from the Pic du Midi is a bit larger than for the northern system (7.8°).

200 200 200 (a) (b) (c) 150 150 150

100 100 100

50 50 50

0 0 0 -200 -150 -100 -50 0 50 100 150 200-200 -150 -100 -50 0 50 100 150 200-200 -150 -100 -50 0 50 100 150 200

-50 -50 -50

-100 -100 -100

-150 -150 -150

-200 -200 -200

Figure 37. (a) Radar image at 2250 with sprite n°8 at 2251 (UTC). (b) Radar image at 2310 with sprite n° 9 at 2312 (UTC). (c) Radar image at 2320 with sprite n°10 at 2321 (UTC). Black crosses indicate the sprite- associated +CGs. Other œCGs and +CGs ±5 minutes of radar image time are marked in red color.

As in the northern part of the storm, the sprites seem to have a certain tendency to occur somewhere between the associated +CGs and the disappearing high-reflectivity core. The multiplicity of the +CGs and the relatively large distance between the +CGs themselves, but also between the +CGs and the sprites, certainly sustain the hypothesis of large and complex horizontal discharges within the stratiform region. It is interesting to note that the triple +CGs

43 appear to —line up“ quite nicely. In these cases, it would be particularly valuable to have the exact location of the sprite with respect to the associated flashes.

If we once again take a look 200 at the corresponding infrared satellite image (figure 38), 150 we see that the sprites n° 8 at 2251 and n° 9 at 2312 (UTC) 100 Sprite at 2312 (UTC) also seem to be located in the coldest (highest) cloud tops. 50 Sprite at 2251 (UTC) One may have the

impression that they are 0 located on the border of the -200 -150 -100 -50 0 50 100 150 200 Sprite +CGs at 2312 (UTC) coldest area, but we should -50 remember that the satellite Sprite +CGs at 2251 (UTC) image is taken at 2300 -100 (UTC) and that the southern

system was moving in the -150 northeastern direction, so at the moments of the sprites, at -200 2251 and 2310 (UTC), they were certainly much closer Figure 38. Meteosat infrared image at 2300 (UTC) with œCGs and +CGs ± 15 minutes, sprite directions (dotted lines) at 2251 and 2312 to the center of the coldest (UTC), and associated +CGs in black color. region.

The last three sprites took place within ten minutes, at a moment when the electrical activity was extraordinarily low. During these ten minutes, only 13 CGs were detected in our quadratic area. Of course, we are reaching the edge of it ; it is true that the activity was a bit more intense beyond the eastern limit. But as we can see from figure 39, the associated +CGs were well within the area, which makes us believe that the sprites were also in this region. And like in the northern part of the storm, the sprite +CGs are all located at almost the same spot, to the south of the sprites, close to a number of other +CGs, forming what looks like some sort of —sprite generating unit“, whatever may be the signification of this.

200 200 200

(a) 150 (b) 150 (c) 150

100 100 100

50 50 50

0 0 0 -200 -150 -100 -50 0 50 100 150 200-200 -150 -100 -50 0 50 100 150 200-200 -150 -100 -50 0 50 100 150 200

-50 -50 -50

-100 -100 -100

-150 -150 -150

-200 -200 -200

Figure 39. (a) Radar image at 0025 with sprite at 0023 (UTC). (b) Radar ima ge at 0030 with sprite at 0028 (UTC). (c) Radar image at 0035 with sprite at 0033 (UTC). W hite crosses indicate the sprite-associated +CGs. Other œCGs and +CGs ±5 minutes of radar image time are marked in red color.

44 Furthermore, the activity may have been low, but a great deal of the flashes were positive, with a rather high average peak current (62.6 kA). If we make a diagram over the percentage of +CGs, similar to figure 31 for the northern part, we once again find what appears to be a relation with the sprites (figure 40).

1 0 0 ,0 0

9 0 ,0 0

8 0 ,0 0

7 0 ,0 0

6 0 ,0 0

% + C G s 5 0 ,0 0 s p r ite s

4 0 ,0 0

3 0 ,0 0

2 0 ,0 0

1 0 ,0 0

0 ,0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 0 0 0 0 0 0 1 1 1 1 1 1 2 2 2 2 2 2 3 3 3 3 3 3 0 0 0 0 0 0 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0 0 0 T i m e (U T C ) Figure 40. Percentage of +CGs every 5 minutes in the southern part of the storm and the sprites occurring in this region.

W ith a mean 15 minutes value, the correlation between the % +CGs and the sprite rate has an R× value of 0.59. The corresponding value for the northern part was ony R×=0.31, because of the concentration of +CGs +CGs between 2230 and 2250 (UTC) that did not produce any sprites. If we exclude this group of flashes, the correlation is much higher : R×=0.56.

200 The infrared satellite image exhibits a similar pattern as before (figure 41). The three last sprites occurred in almost the 150 same direction, and once again, the sprite direction from the Pic du Midi cuts 100 the coldest cloud area, even if it is not possible to conclude from the location of 50 the associated +CGs that the sprites 0 really occurred within this area, which is -200 -150 -100 -50 0 50 100 150 200 now quite small. Still, the seemingly constant tendency of the sprite direction -50 to coincide with the coldest clouds strongly suggests that there is a -100 relationship between the appearance of sprites and the altitude of the cloud tops. -150

-200 Figure 41. Meteosat infrared image at 0030 (UTC) with œCGs and +CGs ±15 minutes, approximate direction (dotted line) of the sprites at 0023, 0028 and 0033 (UTC), and associated +CGs in black color.

45 5.4 Intracloud activity

Our data of the intracloud (IC) flashes are somewhat altered by the —blind spot“ in the SAFIR lightning localization system (cf. 4.4). This spot was located slightly to the south east with respect to the center of our quadratic area, which could partially affect our ability to analyze some of the sprite-related processes in the southern system of the storm. However, it is our belief that the overall picture of the IC activity should not be significantly distorted by this shortfall. Furthermore, as we shall see, the temporal correlation that we find between the sprites and the IC lightning seems to indicate that no essential part of the collected data, except maybe for sprite n° 8, has been lost.

If we look at the whole area (figure 42, solid lines), we see that the intracloud activity reaches a pronounced maximum between 2045 and 2230 (UTC). During this period, the old thumb rule, according to which there are roughly 4-5 IC discharges for every CG, appears to be approximately valid, especially if we take into account the possible loss of some IC data in the —blind spot“. After this maximum and to the end of the storm, the CG and IC frequencies are more or less of the same magnitude (although it should be repeated that the IC data could be slightly underestimated).

500

450

400

350

300

C G s (w hole area) 250 IC s (w hole area)

C G s (nothern part) 200 IC s (nothern part)

sprites (northern part) 150 sprites (southern part)

100

50 Isolated œCG cell

0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 :1 :2 :3 :4 :5 :0 0 0 0 0 0 0 1 1 1 1 1 1 2 2 2 2 2 2 3 3 3 3 3 3 0 0 0 0 0 0 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0 0 0 T im e (U T C )

Figure 42. Frequency of CG and IC flashes.

But this description is not very accurate if we examine the northern and the southern part of the storm separately. As we can see in figure 42, the northern region (dotted lines) plays only a minor part in the high intracloud activity between 2045 and 2230 (UTC). Moreover, in this area there is no substantial difference between the number of CGs and ICs ; the intracloud lightning frequency is sometimes a bit lower, sometimes a bit higher than the CG frequency, but they both stay in the same ballpark. Thus, the southern region is to a large extent responsible for the high IC maximum that we see in the global picture. It is remarkable that, for an approximately equal number of CGs, the southern part is capable to produce about 5 times as much ICs as the northern part. It is also interesting to note that the small, peripheral œ CG cell that appeared around midnight UTC (figure 35), and that we have not excluded from figure 42 (encircled part), corresponds to a very low IC activity.

46 Concerning the sprites, we observe once again that they occur at a late stage of the storm, when the activity is slowing down considerably. Another characteristic could perhaps be the apparent similarity in magnitude between the number of œCGs and the number of +ICs during the sprite productive periods, although it would be hazardous to draw any conclusions in this direction after having studied only one particular thunderstorm.

As described in 4.4, the SAFIR system discriminates between different types of intracloud electric activity : isolated IC points, start of lightning, intermediate points and end of lightning. Since the number of intermediate points reflects, at least in principle, the length and complexity of the flash, we found it interesting to investigate more closely the distribution of these points at the moment of the sprite events. In order to do this, we chose the interval between one minute before the sprite and one minute after the sprite, we calculated the number of intermediate points and compared it with the number of isolated points and starts. It turns out that, even though the result is not systematic for all the sprites, the vast majority of them are temporally associated with a local —burst“ of intermediate points. Take, for instance, the sprite n° 2 at 2134 (UTC), that had no detected associated CG flash (figure 43).

100

90

80

70

60 Isolated points Starts 50 Intermediate points sprite 40

30

20

10

0

1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 4 4 4 5 5 6 6 6 7 7 8 8 8 9 9 0 0 0 1 1 2 2 2 3 3 4 4 4 5 5 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 Time in seconds (UTC)

Figure 43. Intracloud activity one minute before and one minute after sprite n° 2 at 2134 (UTC).

An IC flash started 59.2 ms before this sprite. It counted no less than 69 intermediate points. Is this to say that the sprite was actually triggered by the IC ? It is impossible, of course, to assert this with any kind of certainty. The Météorage system may simply have failed to detect the associated CG, although it is a fact that the SAFIR system, designed to identity CG flashes as well, did not detect it either. Final report (2000) mentions some cases of sprites apparently without associated CGs. It evokes the possibility that they may have been induced by the charge transfers due to so called —spider“ lightning, but adds that one would then have to accept that sprites must be triggered far more frequently by intracloud discharges, since ICs in severe storms outnumber CGs by 10 to 1 or more. And this does not at all seem to be the case.

47 By the way, are we really dealing with —spider“ lightning here ? If we examine the spatial repartition of the intermediate points (figure 44), we notice that they are concentrated in a rather small area. Its 80 maximum extension is about 12 km, which is 60 somewhat less than typical —spider“ lightning, extending

40 in most cases many tens of kilometers. Furthermore, the speed of propagation of the 20 flash is in the order of 106 m/s, while —spider“ leaders

0 propagate generally at a -80 -60 -40 -20 0 20 40 60 80 lower speed, about 2-4×105 Mazur et al. -20 m/s. ( , 1998). But aside from this, we observe that the IC lightning, in -40 similarity with +CGs associated with many of the

-60 other sprites, appears in a region of relatively low reflectivity and that the sprite -80 occurs somewhere between

this area and the dissipating Figure 44. Radar image at 2135 (UTC). Sprite direction (dashed high reflectivity core. line) at 2134 (UTC). Red crosses are intermediate points.

The IC activity showed a great diversity during the second before and the second after the sprite (appendix 4). Seven of the sprites were preceded by ICs exhibiting more than 10 intermediate points, but the sprites n° 8 at 2251 and n° 13 at 0033 (UTC) had no detected intracloud flashes whatsoever in this particular interval, which could of course indicate a detection problem. It is true that sprite n° 8 took place somewhere in the vicinity of the —blind spot“. On the other hand, this sprite was followed by several detected ICs that were geographically not too far from the two associated +CGs, but rather late, about a second after the sprite, which might signify that they were not connected to the TLE event. It does not seem possible to say anything conclusive about the IC activity at the moment of this sprite.

The correlation between the sprites and the number of intermediate points was particularly obvious when the sprites occurred at an extremely calm moment of the thunderstorm. Sprite n° 11 at 0023 (UTC) was preceded by an IC (-104.5 ms) with 3 intermediate points and followed by an IC (+777.3 ms) with 15 intermediate points, although there was practically no other intracloud activity at all detected the minute before and the minute after the sprite. Sprite n° 12 at 0028 (UTC) was preceded by a small IC (-120.9 ms) and sprite n° 13 at 0033 (UTC) was followed by a small IC (+459.4 ms). For both of them the activity was very close to zero the minute before and the minute after the sprite.

Figure 45 shows some examples of IC activity at the moment of the sprite.

48 100 100

90 (a) 90 (b) 80 80

70 70

60 60 Isolated points Starts

50 50 Intermediate points sprite 40 40

30 30

20 20

10 10

0 0

0 4 8 2 6 0 4 8 2 6 0 4 8 2 6 0 4 8 2 6 0 4 8 2 6 0 4 8 2 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 9 9 9 0 0 1 1 1 2 2 3 3 3 4 4 5 5 5 6 6 7 7 7 8 8 9 9 9 0 0 0 1 1 2 2 2 3 3 4 4 4 5 5 6 6 6 7 7 8 8 8 9 9 0 0 0 1 1 2 6 6 6 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 8 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 Time in seconds (UTC) Time in seconds (UTC) 100 100

90 (c) 90 (d) 80 80

70 70

60 60 Isolated points Starts

50 50 Intermediate points sprite 40 40

30 30

20 20

10 10

0 0

7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 7 1 5 9 3 4 8 2 6 0 4 8 2 6 0 4 8 2 6 0 4 8 2 6 0 4 8 2 6 0 4 8 2 6 0 9 0 0 0 1 1 2 2 2 3 3 4 4 4 5 5 6 6 6 7 7 8 8 8 9 9 0 0 0 1 4 4 5 5 6 6 6 7 7 8 8 8 9 9 0 0 0 1 1 2 2 2 3 3 4 4 4 5 5 6 4 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 6 6 6 6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 Time in seconds (UTC) Time in seconds (UTC)

Figure 45. Examples of intracloud activity ±60 seconds of the sprite. (a) Sprite n° 7 at 2209 (UTC). (b) Sprite n° 8 at 2251 (UTC). (c) Sprite n° 9 at 2312 (UTC). (d) Sprite n° 10 at 2321 (UTC).

50

Sprite direction The IC that occurred 177.7 ms 40 before the sprite n° 10 at 2321 (UTC) was quite remarkable. It 30 had as many as 85 intermediate +CGs associated to the sprite points (figure 45d) and really ) m k

( 20 stands out in an astonishing way Y from the activity before and after

10 the sprite. It lasted 8.6 ms and IC activity was concentrated to an area with

0 a maximum extension of about 7 35 45 55 65 75 85 km. As we can see (figure 46), it

-10 was relatively close to the three X (km) +CGs associated with the sprite and probably, from what we can Figure 46. Location of the IC flash just before sprite n° 10 see of the sprite direction, close at 2321 (UTC) with respect to the associated +CGs and to the to the sprite itself. sprite direction.

49 Sprite n° 6 at 2205 (UTC) is also interesting and a bit different from the others. Firstly, it was our most characteristic cluster with three carrot shaped sprites (figure 47). Secondly, it was followed by a series of ICs (260.1-273.6 ms after the sprite) dispersed over a rather long but narrow —channel“ towards southsouthwest (figure 48). It was also preceded by a group of ICs (386.2-79.3 ms before the sprite) that were spatially close to the associated +CG and that were concentrated to a rather small area, resembling, in this respect, the IC activity Figure 47. Sprite n° 6 (cluster) at 2205 (UTC). just before the previous mentioned sprites n° 2 and 10.

200

150 Sprite directions

100

ICs before sprite 50

0 -200 -150 -100 -50 0 50 100 150 200

-50 ICs after sprite

-100

-150

-200

Figure 48. Radar image at 2205 (UTC). Directions of the three sprites in the cluster (sprite n° 6). Red crosses indicate intracloud activity ±1 second of the sprite. Distances in km.

The maximum extension of the intracloud activity after the sprite was more than 90 km, which corresponds well to the characteristics of —dendritic“ lightning, although the speed, in this case too, seems to be higher than typical —spider“ speed. At least we have here an evidence of a large, horizontal propagation of discharges at the moment of a sprite.

50 It should be added that SAFIR did not account for this horizontal propagation as one single lightning discharge. The detection system described it as a series of nine flashes interrupted by a number of isolated points. But this should not be taken too literally. Neither should, by the way, the 69 intermediate points in sprite n° 2 or the 85 in sprite n° 10 be automatically interpreted as single discharges, even if, in concern of simplicity, we have chosen to do so in this study. Unfortunately, the analysis of the intracloud lightning structure is not yet an exact science. As mentioned in 4.4, SAFIR classifies the data into different categories (isolated points, starts, etc) according to spatial and temporal criteria, and one should keep in mind that this does not necessarily correspond to a physical reality. If we have, for example, a large distance (>4 km) between two detected points, which was frequently the case in the vast electrical activity just after sprite n°6, SAFIR will consider these two points as parts of two different flashes. This classification is of course highly arbitrary and cannot by any means be regarded as a reliable, physical fact. But these imperfections in the detection technique are hopefully not so important in the sprite-related description of the intracloud lightning. W e believe that the simple sum of intermediate points is a good indicator of the intensity and the complexity of the IC activity at a certain moment.

The location of the IC lightning with 200 respect to the sprite-associated +CGs was very various. In figure 46, we saw that the 150

IC was quite close to the sprite +CGs, but 100 sprite n° 7 at 2209 (UTC) is a rather extreme example of the opposite (figure 50 ) m

k 0

49). Here the two sprite +CGs coincide ( spatially almost exactly with the sprite Y -200 -150 -100 -50 0 50 100 150 200 direction (which was quite unusual in our -50 observations), but the IC activity in the -100 second before the sprite is located in a small area about 150 km to the south (an -150 area that belongs to the southern storm -200 system, whereas the sprite and the sprite X (km) +CGs occur in the northern system), and Figure 49. Sprite direction n° 7 (dashed line). The we find the IC activity in the second after two black crosses are the +CGs, the red crosses the intracloud activity before the sprite, and the the sprite some 80 km to the west. These blue crosses the intracloud activity after the sprite. distances are so large that one can seriously doubt that the ICs had any connection with the TLE. Still, if we look at figure 45a, it is hard not to be struck by the temporal correlation between the sprite and the sudden peak in the number of intermediate points. If this is a pure coincidence, it sure is a rather peculiar one.

Other ICs were clearly closely related to the sprite +CGs, like the one that occurred 184.5 ms before sprite n° 3 at 2152 (UTC). All of the 30 intermediate points were gathered within a radius of 2.5 km of the +CG. The following sprite, n° 4 at 2156 (UTC), had two sprite +CGs less than 5 km from each other and was preceded by an IC activity some 15 km to the north of them, between the +CGs and the probable location of the sprite. Sprite n° 9 had two centers of IC activity during the second before the sprite, of which one was located less than 10 km from one of the three sprite +CGs.

51 20 Incidentally, it is interesting that the spatial Intracloud activity location of this latter center with respect to Sprite direction the +CGs (figure 50) resembles very much 10 the location of the ICs preceding sprite n° 0 10 (that we have already seen, figure 46), a 30 40 50 60 70 80 90 100 110

sprite that also had three associated +CGs. ) +CGs associated to the sprite m

k -10 (

Given that the sprites n° 9 and 10 occurred Y within less than 10 minutes and within about the same angle seen from the Pic du -20 Midi, it is tempting to conclude that they both arose from a similar electrical -30 structure in the clouds at that particular moment of the storm. However, the optical -40 X (km) appearance of these two sprites does not Figure 50. Location of one part of the IC activity have much in common, n° 9 showing a just before sprite n° 9 at 2312 (UTC) with respect to large, blurred, columniform aspect, and the associated +CGs and to the sprite direction. n°10 being the bright, carrot shaped type.

Already the study of the space shuttle videotapes revealed the connection between the sprites and the intracloud activity (Boeck, 1995 ; 1998). The TLEs always seemed to occur in association with cloud flashes immediately before and after the event. Most of our observations confirm this temporal pattern in the time interval of one second before and after the sprite, even though we also find some strange absences of data : before the sprites n° 8 and 13 and after the sprite n° 12. Further investigations with a much larger material will be needed in order to analyze the exact characteristics of the sprite-related ICs.

52 6 Conclusion

From the European Sprite2003 campaign, organized within the frame of the CAL project, we have studied sprite observations from the 23 July. These were carried out from the OMP observatory in the French Pyrenees with two low-light CCD cameras. The two thunderstorm systems that developed over the South of France during the afternoon and the evening produced in total 13 detected sprites after sunset. All of these, except one, could be associated to a positive cloud-to-ground flash, which confirms the results of other studies, i.e. the vast majority of sprites seem to be triggered by +CGs.

The temporal repartition of the sprites did not appear to be random. Once the sprite production started, sprites had a tendency to occur fairly regularly with a few minutes interval, as if, at certain precise moments of the storm, favourable conditions for the sprite process were present. In accordance with previous observations, sprites were not detected immediately after sunset, but began almost two hours later.

Since we had only the azimuth angle of each sprite, we could not determine the sprite location with absolute certainty. Making the assumption that the center of the sprite lies within 50 km of the parent +CG, we did however calculate the approximate vertical and horizontal extent of the events. The mean maximum altitude, 91 km, corresponds relatively well to other reports, even if some of the higher altitudes œ close to or over 100 km œ are a bit surprising and may be due to the uncertainty of the measurements. W e were not able to tell whether the sprite tendrils reached the cloud tops.

W e have reason to believe that the sprites occurred in a relatively small portion of each storm system. Firstly, because the sprite directions from the Pic du Midi varied little, especially for the northern storm region. Secondly, because the sprite-associated +CGs tended to form tight clusters in a small area. This suggests that rather special conditions are required for sprite production to take place, and that these conditions exist only in certain thunderstorms (since all thunderstorms do not produce sprites) and in a very limited area of these storms. It is possible that some kind of —sprite-producing cell“ emerges in certain circumstances, even if the nature of this cell remains to be discovered.

It would be a great advance to European sprite research if, in the future, observations could be carried out simultaneously from two different mountain peaks, distant from each other. This would allow us to determine the location of the sprite by triangulation and eliminate the uncertainty of its exact position in the storm system.

W e found a certain relationship between the peak current of the +CGs and the occurrence of sprites. But the relationship was rather weak ; the average peak current of the associated +CGs was higher than the average peak current of all the +CGs, but low peak currents could also generate sprites, and high peak currents were no guarantee for sprite production. Thus, it appears obvious that a high peak current is neither a necessary nor a sufficient condition for sprites.

Our available measurements of the charge moments were rather few and concerned only some of the associated +CGs. Nevertheless, they confirm observations from many other reports : sprite-producing flashes do have considerably higher charge moments than —ordinary“ lightning (∼200-1000 C km). In accordance with the physical processes that are thought to cause sprite events (quasi-electrostatic fields or runaway electron breakdown), the charge

53 moment of the cloud-to-ground flashes therefore appears to be a much more relevant parameter for sprite production than the peak current. In theory, this is not inherent to the polarity of the lightning, but in practice, with some rare exceptions, only positive flashes can reach the sufficient charge moments.

The sprite production started in a late stage of the two storm areas, when the œCG activity as well as the IC activity was rapidly falling, but when the +CG activity remained stable or even increased slightly. W e observed a pattern where the sprite generation seemed to be correlated, not to the number of +CGs itself, but to the ratio of positive CGs to the total number of CGs. Some of the sprites occurred at a stage when the CG activity was close to zero, but almost all of the few CGs that actually were detected at that moment had positive polarity.

The sprite-associated +CGs were located outside the dissipating high-reflectivity core in the stratiform region of the storm system (often several tenths of km from the center of the storm). The sprites had a tendency to occur somewhere between the +CG and center of the storm. Multiple +CG events appeared to produce sprites more easily than single events, and they were dispersed over a rather large area in the stratiform region. In the case of triple events, the +CGs somehow seemed to form a straight line pointing towards the sprite (or at least towards the observed sprite direction). This suggests the presence of a large horizontal electrical activity within the storm system.

The intracloud detecting system SAFIR confirmed in most cases the presence of intense IC activity either in the second before or the second after the sprite, or both. The number of so called intermediate points, indication of complex IC discharges, often showed a sudden —burst“ at the moment of the sprite. This was particularly obvious when the total electrical activity was very low. For several sprites, we had an activity close to zero the minute before and the minute after the sprite, but an abrupt peak at the moment of the sprite. However, with one remarkable exception, the sprite-related IC activity did not have a very large horizontal extension. Mostly, it was concentrated to a rather small area. W e found that the location of the IC lightning was very various, not only with respect to the sprites, but also to the associated +CGs. But even when the distance from the sprite was large, we had a strong temporal correlation to the sprite.

It is interesting to note that one of the most intense —bursts“ of IC activity occurred immediately before the sprite that had no detected associated +CG flash. It is tempting, even though there is of course no proof for it, to imagine that this sprite might have been triggered by the IC flash. IC activity temporally close to sprite events without detected parent CG should be studied thoroughly in the future in order to determine whether IC discharges can have a sprite-triggering function.

The satellite images showed that sprites seem to occur only in regions with cold (high) cloud tops, cloud tops in the vicinity of the tropopause. The associated +CGs had also a clear tendency to be found within or close to this region. W e believe that this might explain why a tight, powerful cluster of +CGs in a low-cloud area of the northern storm system did not produce any sprites, while a relatively similar cluster in the high-cloud area did just that. However, this needs to be confirmed by future research.

The surface of the radar echo ≥16dBZ was calculated. W e found that the results were in good accordance with the estimated threshold value for sprite production (7500 km2), observed in

54 several other studies. Thus, our observations sustain the hypothesis that sprite production is possible only in large thunderstorm systems.

The object of this study, the sprite observations of the 23 July 2003, is only a small part of the European sprite investigations, a research area that has just begun. A larger study of the observation data from the summer 2003 (and from future campaigns) will undoubtedly yield more statistically reliable results and a deeper understanding of the physical processes behind the sprite phenomenon.

Acknowledgements

This work would not have been possible without the help and hospitality of the Atmospheric Electricity Group at the Laboratory of Aerology of the University of Toulouse in France. I especially would like to thank my supervisor Dr. Serge Soula for all help and encouragement. I am also very grateful to Prof. Serge Chauzy for his valuable comments and good companionship. I thank Thomas Allin of the Technical University of Denmark for his help in interpreting the sprite images. Thanks also to Dr. Gabriella Satori of the Geodetic and Geophysical Research Institute of Hungary for providing the charge moment calculations.

55

References

Barr, R., D. Llanwyn Jones, and C.J. Rodger, ELF and VLF radio waves, J. Atmos. Solar Terr. Physics, 62, 1689-1718, 2000.

Barrington-Leigh, C.P., and U.S. Inan, Elves triggered by positive and negative lightning discharges, Geophys. Res. Lett., 26, 683-686, 1999.

Barrington-Leigh, C.P., U.S. Inan, M. Stanley, and S.A. Cummer, Sprites triggered by negative lightning discharges, Geophys. Res. Lett., 26, 3605-3608, 1999.

Barrington-Leigh, C.P., V.P. Pasko, and U.S. Inan, Exponential relaxation of optical emissions in sprites, J. Geophys. Res., 107, 10.1029/2001JA900117, 2002.

Bell, T.F., V.P. Pasko, and U.S. Inan, Runaway electrons as a source of red sprites in the mesosphere, Geophys. Res. Lett., 22, 2127-2130, 1995.

Boccippio, D.J., E.R. W illiams, S.J. Heckman, W .A. Lyons, I.T. Baker, and R. Boldi, Sprites, ELF transients, and positive ground strokes, Science, 269, 1088-1091, 1995.

Boeck, W .L., O.H. Vaughan, Jr, R. Blakeslee, B. Vonnegut, and M. Brook, Lightning induced brightening in the airglow layer, Geophys. Res. Lett., 19, 99-102, 1992.

Boeck, W .L., O.H. Vaughan, Jr, R.J. Blakeslee, B.Vonnegut, M. Brook, and J. McKune, Observations of lightning in the stratosphere, J. Geophys. Res., 100, 1465-1475, 1995.

Boeck, W . L., O. H. Vaughan, Jr., R. J. Blakeslee, B. Vonnegut, and M. Brook, The role of the space shuttle videotapes in the discovery of sprites, jets and elves, J. Atmos. Solar Terr. Physics, 60, 669-677, 1998.

Cummer, S.A., and M. Stanley, Submillisecond resolution lightning currents and sprite development : observations and implications, Geophys. Res. Lett., 26, 3205-3208, 1999.

Cummins, K.L., M.J. Murphy, E.A. Bardo, W .L. Hiscox, R.B. Pyle, and A.E. Pifer, A combined TOA/MDF technology upgrade of the U.S. national lightning detection network, J. Geophys. Res., 103, 9035-9044, 1998.

Defoy, S., Caractérisation des données SAFIR et approche d‘une étude de corrélation entre précipitation et activité électrique, Rapport de stage d‘approfondissement de l‘Ecole Nationale de la Météorologie, 1998.

Fernsler, R.F., and H.L. Rowland, Models of lightning-produced sprites and elves, J. Geophys. Res., 101, 29653-29662, 1996.

Final report for the 1999 sprites program submitted to the Department of Energy, Mission Research Corporation, 18 April 2000 [available online from http://www.fma-research.com/Sprites99/DOEfinal.html].

56 Finke, U., and O. Kreyer, Detect and locate lightning events from geostationary satellite observations. Report part I. Review of existing lightning location systems, EUMETSAT study, EUM/CO/02/1016/SAT, 2002 [available online from http://www.eumetsat.de/en/area4/pps/ExpertMeetings/Reports/Finke_part1.pdf].

Fishman, G.J., P.N. Bhat, R. Mallozzi, J.M. Horack, T. Koshut, C. Kouveliotou, G.N. Pendleton, C.A. Meegan, R.B. W ilson, W .S. Paciesas, S.J. Goodman, and H.J. Christian, Discovery of intense gamma-ray flashes of atmospheric origin, Science, 264, 1313, 1994.

Fukunishi, H., Y. Takahashi, M. Kubota, K. Sakanoi, U.S. Inan, and W .A. Lyons, Elves: lightning-induced transient luminous events in the lower ionosphere, Geophys. Res. Lett., 23, 2157-2160, 1996.

Franz, R.C., R.J. Nemzek, and J.R. W inckler, Television image of a large upward electrical discharge above a thunderstorm system, Science, 249, 48-51, 1990.

Hu, W ., S.A. Cummer, W .A. Lyons, and T.E. Nelson, Lightning charge moments changes for the initiation of sprites, Geophys. Res. Lett., 29, 10.1029/2001GL014593, 2002.

Huang, E., E. W illiams, R. Boldi, S. Heckman, W . Lyons, M. Taylor, T. Nelson, and C. W ong, Criteria for sprites and elves based on Schumann resonance observations, J. Geophys. Res., 104, 16943-16964, 1999.

Inan, U.S., VLF heating of the lower ionosphere, Geophys. Res. Lett., 17, 729-732, 1990.

Inan, U.S., W .A. Sampson, and Y.N. Taranenko, Space-time structure of optical flashes and ionization changes produced by lightning-EMP, Geophys. Res. Lett., 23, 133-136, 1996.

Inan, U.S., C. Barrington-Leigh, S. Hansen, V.S. Glukhov, T.F. Bell, and R.Rairden, Rapid lateral expansion of optical luminosity in lightning-induced ionospheric flashes referred to as ”elves‘, Geophys. Res. Lett., 24, 583-586, 1997.

Lehtinen, N.G., T.F. Bell, and U.S. Inan, Monte Carlo simulation of runaway MeV electron breakdown with application to red sprites and terrestrial gamma ray flashes, J. Geophys. Res., 104, 24699-24712, 1999.

Lyons, W .A., Characteristics of luminous structures in the stratosphere above thunderstorms as imaged by low-light video, Geophys. Res. Lett., 21, 875-878, 1994.

Lyons, W .A., Sprite observations above the U.S. High Plains in relation to their parent thunderstorm systems, J. Geophys. Res., 101, 29641-29652, 1996.

Lyons, W .A., T.E Nelson, R.A. Armstrong, E.R. W illiams, D.M. Suszcynsky, R. Strabley, M. Taylor, and L. Gardner, Characteristics of thunderstorms and lightning flashes which produce mesospheric transient luminous event, 11th Int. Conf. on Atmospheric Electricity, Guntersville, Alabama, 1999.

Lyons, W .A., T.E. Nelson, and J. Fossum, Results from the SPRITES‘99 and STEPS 2000 field programs, Preprints, 20th conf. on severe local storms, Amer. Meteor. Soc., Orlando, 2000a.

57

Lyons, W .A., R.A. Armstrong, E.A. Bering, and E.R. W illiams, The hundred year hunt for the sprite, EOS, Trans. Amer. Geophys. Union, 81, 373-377, 2000b.

Lyons, W .A., T.E. Nelson, E.R. W illiams, S.A. Cummer, and M.A. Stanley, Characteristics of sprite-producing positive cloud-to-ground lightning during the 19 July 2000 STEPS mesoscale convective systems, Monthly Weather Review, 131, 10, 2417-2427, 2003a.

Lyons, W .A., T.E. Nelson, R.A. Armstrong, V.P. Pasko, and M.A. Stanley, Upward electrical discharges from thunderstorm tops, Bull. Amer. Meteor. Soc., 84, 445-454, 2003b.

Lyons, W .A., and R.A. Armstrong, A review of electrical and turbulence effects of convective storms on the overlying stratosphere and mesosphere, Symposium on space weather, The 84th AMS annual meeting, Seattle, 2004 [available online from http://ams.confex.com/ams/pdfpapers/74073.pdf]

MacGorman, D.R., and W .D. Rust, The electrical nature of storms, Oxford University Press, New York, 1998.

Marshall, T.C., M. Stolzenburg, W .D. Rust, E.R. W illiams, and R. Boldi, Positive charge in the stratiform cloud of a mesoscale convective system, J. Geophys. Res., 106, 1157-1163, 2001.

Mazur, V., X.-M, Shao, and P.R. Krehbiel, —Spider“ lightning in intracloud and positive cloud-to-ground flashes, J. Geophys. Res., 103, 19811-19822, 1998.

Mende, S.B., R.L. Rairden, and G.R. Swenson, Sprite spectra ; N2 1 PG identification, Geophys. Res. Lett., 22, 2633-2636, 1995.

Morel C., and S. Sénési, SAFNW C scientific report on PGE 11 rapid developing thunderstorms, prepared by Météo-France/CNRM, Toulouse, 2000 [available online from http://www.meteorologie.eu.org/RDT/doc/SR_RDT.doc].

Neubert, T., T.H. Allin, H. Stenbaek-Nielsen, and E. Blanc, Sprites over Europe, Geophys. Res. Lett., 28, 3585-3588, 2001.

Pasko, V.P., Dynamic coupling of quasi-electrostatic thundercloud fields to the mesosphere and lower ionsphere : sprites and jets, PhD thesis, Department of Electrical Engineering, Stanford University, 1996.

Pasko, V.P., Electric jets, Nature, 423, 927-929, 2003.

Pasko, V.P., U.S. Inan, Y.N. Taranenko, and T.F. Bell, Heating, ionization and upward discharges in the mesosphere due to intense quasi-electrostatic thundercloud fields, Geophys. Res. Lett., 22, 365-368, 1995.

Pasko, V.P., U.S. Inan, and T.F. Bell, Sprites as luminous columns of ionization produced by quasi-electrostatic thundercloud fields, Geophys. Res. Lett., 23, 649-652, 1996.

58 Pasko, V.P., U.S. Inan, T.F. Bell, and Y.N. Taranenko, Sprites produced by quasi-electrostatic heating and ionization in the lower ionosphere, J. Geophys. Res., 102, 4529-4561, 1997.

Pasko, V.P., U.S. Inan, and T.F. Bell, Mesosphere-troposphere coupling due to sprites, Geophys. Res. Lett., 28, 3821-3824, 2001.

Reising, S.C., Remote sensing of the electrodynamic coupling between thunderstorm systems and the mesosphere/lower ionosphere, PhD Thesis, Department of Electrical Engineering, Stanford University, 1998.

Reising, S.C., U.S. Inan, T.F. Bell, and W .A. Lyons, Evidence for continuing current in sprite-producing cloud-to-ground lightning, Geophys. Res. Lett., 23, 3639-3642, 1996.

Roussel-Dupré, R., and A.V. Gurevich, On runaway breakdown and upward propagating discharges, J. Geophys. Res., 101, 2297-2311, 1996.

Rowland, H.L., Theories and simulations of elves, sprites and blue jets, J. Atmos. Solar Terr. Physics, 60, 831-844, 1998.

Rycroft, M.J., S. Israelsson, and C. Price, The global atmospheric electric circuit, solar activity and climate change, J. Atmos. Solar Terr. Physics, 62, 1563-1576, 2000.

Sato M.., and H. Fukunishi, Global sprite occurrence locations and rates derived from triangulation of transient Schumann resonance events, Geophys. Res. Lett., 30, doi:10.1029/2003GL017291, 2003.

Seity, Y., Relations entre activités d‘éclairs, microphysique, et dynamique au sein d‘orages européens, thèse de doctorat, Université Toulouse III œ Paul Sabatier, 2003.

Sentman, D.D., E.M. W escott, D.L. Osborne, D.L. Hampton, and M.J. Heavner, Preliminary results from the Sprites94 aircraft campaign: 1. Red sprites, Geophys. Res. Lett., 22, 1205- 1208, 1995.

Soula, S., and S. Chauzy, Some aspects of the correlation between lightning and rain activities in severe storms, Atmos. Res., Vol 56/1-4, 355-373, 2001.

Stanley, M., P. Krehbiel, M. Brook, C. Moore, and W . Rison, High speed video of initial sprite development, Geophys. Res. Lett., 26, 3201-3204, 1999.

Stolzenburg M., W .D. Rust, and T.C. Marshall, Electrical structure in thunderstorm convective regions. 3. Synthesis, J. Geophys. Res., 103, 14097-14108, 1998.

Stozhkov, Y.I., The role of cosmic rays in the atmospheric processes, J. Phys. G: Nucl. Part. Phys., 29, 913-923, 2003.

Taranenko, Y.N., U.S. Inan, and T.F. Bell, The interaction with the lower ionosphere of electromagnetic pulses from lightning : excitation of optical emissions, Geophys. Res. Lett., 20, 2675-2678, 1993.

59 Vaughan, Jr., O. H., and B. Vonnegut, Recent observations of lightning discharges from the top of a thundercloud into the clear air above, J. Geophys. Res., 94, 13179-13182, 1989.

W escott, E.M., D.D. Sentman, D.L. Osborne, D.L. Hampton, and M.J. Heavner, Preliminary results from the Sprites94 aircraft campaign: Blue jets, Geophys. Res. Lett., 22, 1209-1212, 1995.

W escott, E.M., D.D. Sentman, M.J. Heavner, D.L. Hampton, D.L. Osborne, and O.H. Vaughan, Jr., Blue starters: Brief upward discharges from an intense Arkansas thunderstorm, Geophys. Res. Lett., 23, 2153-2156, 1996.

W escott, E.M., D.D. Sentman, M.J. Heavner, D.L. Hampton, and O.H. Vaughan, Jr., Blue jets: their relationship to lightning and very large hailfall, and their physical mechanisms for their production, J. Atmos. Solar Terr. Physics, 60, 713-724, 1998a.

W escott, E.M., D.D. Sentman, M.J. Heavner, D.L. Hampton, W .A. Lyons, and T. Nelson, Observations of ”columniform‘ sprites, J. Atmos. Solar Terr. Physics, 60, 733-740, 1998b.

W illiams, E.R., The positive charge reservoir for sprite-producing lightning, J. Atmos. Solar Terr. Physics, 60, 689-692, 1998.

W illiams, E.R., Sprites, elves, and glow discharge tubes, Physics today, 41-47, November, 2001.

W inckler, J.R., W .A. Lyons, T.E. Nelson, and R.J. Nemzek, New high-resolution ground- based studies of sprites, J. Geophys. Res, 101, 6997-7004, 1996.

60 Appendix 1. Estimated vertical and horizontal extent of the 13 sprites. Sprites appearing at identical time indications correspond to multiple sprites.

61 Appendix 2. Charge moments for some of the associated +CGs.

62 Appendix 3. Soundings from the airport of Lyon at 1200 (UTC) the 23 July 2003 and at 0000 (UTC) the 24 July 2003. Source : University of W yoming.

63

Appendix 4. Intracloud activity 1 second before and 1 second after each sprite.

Sprite Sprite -1 second Sprite +1 second event n° hour min sec Isolated points Starts Intermediate points Isolated points Starts Intermediate points 1 21 11 32.356 5 1 4 2 0 0

2 21 34 58.180 28 15 85 0 0 0

3 21 52 11.234 1 3 30 2 4 0

4 21 56 28.482 0 2 12 2 8 10

5 21 59 3.506 2 2 17 3 2 1

6 22 5 35.740 2 5 9 9 9 9

7 22 9 6.826 1 1 44 6 7 13

8 22 51 2.288 0 0 0 0 5 4

9 23 12 34.830 2 2 18 1 0 0

10 23 21 41.147 0 1 85 0 0 0

11 0 23 28.347 0 1 3 4 2 16

12 0 28 27.515 1 1 0 0 0 0

13 0 33 1.585 0 0 0 0 1 1

64