We present the results of space-based observations of sprites obtained during the Mediterranean Israeli Dust Experiment (MEIDEX) sprite campaign conducted on board the space shuttle Columbia during its STS-107 mission in January 2003. A total of ∼6 hours of useful data were saved from 21 different orbits, of which 1/5 contained lightning. We imaged sprites from an altitude of 280 km using a calibrated multispectral camera above thunderstorms in various geographical locations, mainly in central Africa, northern Australia, and South America, and also over the Pacific and Indian Oceans. In this paper we report on sprites observed from ranges 1600–2000 km from the shuttle, at altitudes of 40–90 km above the ground. Their brightness was in the range of 0.3–1.7 mega-Rayleighs (MR) in the 665 nm filter and 1.44–1.7 MR in the 860 nm filter. On the basis of the frequency of observed events and the number of tropical thunderstorms, we estimate the sprite rate in the tropics to be of the order of several per minute.
 Transient luminous events (TLEs) is the collective name given to a wide variety of optical emissions which occur in the upper atmosphere above active thunderstorms. These very brief colorful phenomena were discovered in 1989 [Franz et al., 1990] and have been studied since from the ground [Lyons, 1994a, 1996], aircraft [Sentman and Wescott, 1993], balloons, the space shuttle [Boeck et al., 1994, 1998], and the International Space Station [Blanc et al., 2004]. Distinct classes and names were given for the various forms of TLE, all of which allude to their fleeting, unpredictable nature: jets, sprites, elves, and haloes, to name but a few. Sprites seem to play an important role in mesosphere-troposphere coupling [Pasko et al., 2001] that bears on the global electrical circuit [Rycroft et al., 2002]. Indeed, recent observations suggest that TLEs connect the top of thunderstorms to the ionosphere [Pasko et al., 2002; Su et al., 2003].There is a growing body of literature which covers the phenomenology and theory of TLE generation, and we refer the interested reader to recently published reviews [Lyons et al., 2000, 2003; Rodger, 1999].
 Early images of TLEs from space were obtained in conjunction with the mesoscale lightning experiment that was conducted in 1989–1991 [Boeck et al., 1994, 1998]. The analysis of hundreds of hours of video identified 17 events of vertical flashes that appear to connect cloud top and the ionosphere. These events were geolocated by using stars and ground lights and were found to occur over Africa, South America, USA, Australia, Borneo, and the Pacific Ocean. The oblique view of the illumination inside the cloud (caused by strong lightning flashes) from the space shuttle provided the first unambiguous optical link between the parent stroke and the subsequent TLE. Recently, the Lightning and Sprite Observations experiment (LSO), which consisted of automatic nadir-view observations of thunderstorms conducted from the International Space Station, succeeded in separating the weak sprite signal from the bright lightning light preceding it by using a camera with a very narrowband filter in the 756–766 nm range. This filter allowed only the sprite light from the molecular nitrogen band N2 1 PG (3-1) at 762.7 nm to enter the camera. In total, ten events were detected in several hours of automatic undirected observations [Blanc et al., 2004].
 In this paper we report results of space-based observations of sprites, obtained during the Mediterranean Israeli Dust Experiment (MEIDEX), that was conducted on board the space shuttle Columbia during its STS-107 mission in January 2003. The mission lasted 16 days and was performed in a 39° inclination at an altitude of 280 km (150 nautical miles (NMs)), passing over the major thunderstorm producing regions on our planet. Nocturnal observations of the mesosphere above these storms were conducted as a secondary objective of the MEIDEX.
 A detailed description of the MEIDEX science payload and of its technical specification had already been described [Yair et al., 2003]. Spectral data from earlier studies [Hampton et al., 1996] showed that at least five out of the six wavelengths chosen for the MEIDEX were adequate for TLE observations. We used an image-intensified Xybion IMC-201 camera, with a rectangular field of view (FOV) measuring 10.76° vertical and 14.04° horizontal (diagonal 17.86°), with a 486 × 704 pixels CCD, where each pixel corresponds to 1.365 × 10−7 steradian. The camera was calibrated before flight at the Laboratory for Atmospheres at NASA Goddard Space Flight Center and during the mission using the Moon as a calibration source, enabling us to obtain calibrated images of the observed phenomena. The camera was mounted on a single-axis gimbal and allowed a 44° scan of the limb, which in terms of potential coverage gave us a 1600 km arc across the horizon.
 The geometry of observation is shown in Figure 1. For a limb distance of 1900 km the camera field of view covered the altitude range 0–150 km, where all known TLEs occur. During sprite observations, the shuttle was pointed such that the center of the FOV was pointing 50 km above the limb (17° below local horizontal of the shuttle).The camera was operated in the “locked” mode, each time on a specific filter with a fixed integration time of 33 ms. Filter choice and camera gain settings evolved throughout the mission to enhance the probability of registering TLEs.
3. Operational Methodology
 Owing to the MEIDEX orbital constraints for its primary objective, i.e., dust measurements which required daytime dust over-flight in the Atlantic and the Mediterranean regions, most of the nighttime TLE observations were conducted in the SE Pacific (Australia and Fiji), Africa, the southern Indian Ocean, and South America. The necessary shuttle attitude maneuvers and camera gimbal changes were deduced 24 hours in advance on the basis of a forecast method [Ziv et al., 2004] that evaluated the probability of lightning activity in the regions of interest (ROI) on the basis of significant weather maps (SIG maps) used for aviation.
 The necessary shuttle attitude maneuvers and camera gimbal changes were deduced based on almost real-time IR satellite images that were available on the Web (http://www.bom.gov.au/weather/satellite/) and on VLF lightning locations from the Tropical Ocean-Global Atmosphere (TOGA) network (http://ritz.otago.ac.nz/~sferix/TOGA_network.html) operated by the University of Otago, New Zealand. These were used for short-term corrections of the 24 hour forecast. This “now-casting” method allowed us to request of mission control an adjustment of the shuttle attitude, which was generally granted, provided that it was calculated no later than 4.5 hours ahead of an observation (i.e., the time of three revolutions).
 Most of the observations were conducted with filter 5 (665 ± 50 nm), a spectral range in which considerable radiation from sprites is expected from the N2(1 PG) system [Heavner et al., 2000]. The transmittance of this filter was between 80 and 95%. Additional measurements were conducted with filter 6 (860 ± 40 nm), which was considered to be prospective for near-infrared (NIR) emissions [Clodman and Yair, 2003].
 The payload was commanded alternatively from the crew cabin and from the ground according to a predetermined schedule. Time stamping on the Xybion image was inserted from the ground as part of the camera setting and was corrected by the crew if the lag was greater than 2 s. Thus the accuracy of event timing may be considered to be ±2 s. January 2003 was rich in Intertropical Convergence Zone (ITCZ) lightning activity, with major storm centers found in the summer hemisphere around northern Australia, Indonesia, Fiji, and south of the equator over the Pacific Ocean. Intense storms were also observed over Argentina, the Amazon Basin, and the Congo Basin in central Africa. Most observations were conducted before local midnight, when convective activity had not yet subsided. The observations relied on the astronauts' visual observations and real-time adjusting of the camera pointing angle, which were based on initial storm location forecasts transmitted beforehand.
 TLE observations during the MEIDEX were performed in 24 dedicated observation windows, each approximately 20 min long. A total of 583 min were recorded on board, out of which 458 min were transmitted to the ground. In some orbits the data were downlinked (live) during the actual observation and recorded simultaneously on board and on the ground, and in other orbits it was recorded on board and re-played to the ground at a later time. Nonoverlapping data from 21 orbits constitute our 357 min database, with a possible addition of 11 min in tapes recovered on the ground by search teams after the Columbia accident. A summary of the MEIDEX-sprite observations is presented in Table 1. Mission elapsed time (MET) denotes the observation windows (note that on the images presented in the next sections the format of the time stamp is 01/xx/03, where “xx” is the mission day, and the time is the MET). “Start” and “stop” refer to ingress and egress into and out of the predetermined sprite ROI and do not always match the actual recording times. On some occasions the recording continued well outside of the ROI, as long as it did not violate the mission flight rules or interfered with other shuttle activities.
The “useful data” column refers to the video in minutes transmitted by the shuttle, either by the analog or digital downlink systems, which was recorded in the Payload Operations Control Center (POCC) at NASA GSFC and/or at NASA Johnson (overlapping times excluded). “Status” denotes if the data was saved on the ground or lost in the accident. The “analyzed” column marks the orbits whose data has been analyzed thus far.
 Only part of the data set covered stormy regions where TLEs are generated. Here we report the results of the analysis of 2/3 of our data set, concentrating on the detection of sprites. A separate report on our elves observations is presented in the work of Israelevich et al. . The total energy of each sprite event was calculated from the average of the radiance-exposure product, obtained in a region that matches the shape of the sprite by the use of a closely fitted polygon. The total area of this polygon in pixels was determined, and the average of its radiance-exposure product was calculated in mJ m−2 sr−1. Since an atmospheric background emission exists, it was necessary to subtract its radiance-exposure product from that of the sprite. We assumed that for sprites detected in filter 5, the main source of the emissions was from red first positive group (1 PG) bands of neutral molecular nitrogen at 662.4 nm (N2 1 PG) [Heavner et al., 2000]. An estimation of the range of each event from the shuttle, based on its location within our FOV, enabled us to accurately calculate the energy and the brightness of the sprite in each of the given spectral bands of the MEIDEX filters 5 and 6. The procedure is detailed in Appendix A.
4.1. Detection of Sprite Emission in the NIR Orbit 66: 20 January 2003, 1824:32 UT, Filter 6 (860 nm)
 The shuttle flew across the southern Indian Ocean and eastern Australia crossing the northeastern coast. At that time, two main lightning activity centers were located within a mesoscale convective system (MCS) on the northwestern Australian coastline to the west of our planned observation area. At 1824:32 UT the shuttle was located at 36.07°S 158.12°E, with the payload bay pointed directly backward (no bias; Figure 2a). The recorded image (Figure 2b) shows a very large sprite above the lightning illuminated cloud top. On the basis of an estimated range of 1800 km from the shuttle to the flash, we calculated the sprite to occupy the altitude range 45–90 km, with a lateral dimension of 30 km. This detection of sprite emission in the NIR spectral range complements the EXL98 observations reported by Siefring et al.  and Bernhardt et al. . The detected emissions in filter 6 confirm spectral analysis work by Bucsela et al.  and arise from the N2 1 PG. The calculated brightness of this event was 0.96 ± 0.1 MR (Figure 2c). Interestingly, the airglow layer (Meinel OH emission bands 6-2 and 7-3) [Chamberlain, 1961] is clearly visible in this image as a diffuse glow parallel to Earth's surface, and the sprite seems to pierce through it. Bakans  evaluated the emission intensity of these airglow spectral bands to be 680 ± 40 and 880 ± 50 R. Less than a minute later (at 1825:25.26 UT), we observed two additional sprites located above the horizon (Figure 2d). No visible lightning-induced cloud illumination was associated with these sprites; presumably, the causative ground flash was behind the geometrical horizon at a range exceeding 2000 km. It is worth noting that no previous sprite event was ever recorded at such a large range. The calculated emission from these two sprites was 0.78 ± 0.08 MR. These events were separated by more than 500 km from the first one and originated from a different part of the storm system.
4.2. A Sequence of Meteors and Sprites Orbit 87: 22 January 2003, 0153:14 UT, Filter 5 (665 nm)
 In this orbit we started sprite observations over Argentina, crossing the Atlantic Ocean, continuing south of the equator over Africa, and terminating in the Indian Ocean. When flying eastward into continental Africa over Namibia, a strong storm was observed over the Congo Basin. In the span of less than 2 min, two meteors and several bright TLEs (sprites and elves) were observed (Table 2). Although the space shuttle traversed some 900 km northeastward during that period of time, the fact that the astronaut kept tilting the camera toward the center of lightning activity ensured that we approximately imaged the same storm system. Figure 3a presents the overlapping footprints (on the ground) of 40 s of the orbit on the basis of the fact that the shuttle was tilted with respect to the velocity vector, enabling an almost cross-track observation. The first meteor (marked M1 in Figure 3a) was observed at 0152:47.23 UT and penetrated the upper atmosphere in what appears in the image as a steep angle (Figure 3b). It is hard to establish the exact trajectory of this meteor from the visible trail because it can be either coming toward or receding with respect to our line of sight. On the basis of a comparison with visible background stars, we estimate this meteor to peak at magnitude +1.5, and the estimated range for the termination of the visible track was 1200 km. The second meteor (marked M2 in Figure 3a) was brighter (+1.0) and occurred at 0153:05.32 UT. Figure 3c shows the trajectory of the second meteor, consisting of 14 superimposed frames spanning a total duration of 0.462 s. The trajectory shows that the meteor entered the atmosphere between the shuttle and the limb, as Earth is clearly visible in the image due to the illumination by the Moon (phase 68.5%) and was seen in an oblique angle from the northeast. The computed range for the termination point is 800 km from the shuttle.
Table 2. Timing and Sequence of Events During the Observation in Orbit 87 Over a Thunderstorm in Central Africa
Type of TLE
 The sequence shown in Figure 3d illustrates the raw data obtained from the Xybion camera. The triple columniform sprite was recorded at 0153:15.89 UT, less than 30 s after the first meteor penetrated that same atmospheric volume. This event (marked S1 in Figure 3a) was located near the limb ∼1500 km away from the shuttle. Another small sprite was detected at 0154:17.96 UT. A subsequent image from 0156:01.60 UT (Figure 3e) shows a carrot-shaped sprite with a distinct bright body and a dimmer set of branches extending upwards toward the ionosphere. Some tendrils also appear to protrude from below the main sprite body. These tendrils are known to occur no later than 10 ms after sprite initiation with speeds of the order of 106–107 m s−1 [Moudry et al., 2003], but our camera was unable to resolve this downward evolution. The bright spot to the right of the sprite is the star B-Tauri, and Saturn appears on the left-hand margin of the image. This area of Earth was already predicted to be a major producer of TLE [Fullekrug and Price, 2002], and so there is little surprise in the discovery of sprites over Africa.
4.3. Sprites Over an Oceanic Storm Orbit 48: 19 January 2003, 1513:50 UT, Filter 5 (665 nm)
 The shuttle passed to the east of the Australian southeastern coast observing a massive storm centered at 38°S, 138°E near Tasmania. At 1513:50 UT the shuttle was located at 36.07°S, 158.12°E, with nose down the payload bay pointed in azimuth 268.7 (meaning that Earth is seen on top in the video images). An IR satellite image shows an extended cold front emanating from Antarctica toward SE Australia, with cloud bands over the ocean (Figure 4a). The crew gimbaled the camera by 7.48° to observe a lightning flash located near the limb. The recorded image shows Earth occupying the upper part of the frame and the limb at the bottom third, a brightly illuminated cloud top, and two distinct sprites at a height of 80 km above the ground. The main body of the sprites and diffuse elongated branches are clearly seen (Figure 4b). The bright line at the middle left-hand side of the image is light coming out of cities on the southeastern coast of Australia. Lightning activity around Australia for this UT day is shown in Figure 4c. These lightning observations come from the experimental World Wide Lightning Location network [Rodger et al., 2004], which uses linked VLF receivers to locate discharges from the electromagnetic pulses they radiate. The compilation of lightning locations from the network clearly shows the lightning activity organized along the cold front. A search for correlation with the lightning data showed that the only flash detected within the 2 s accuracy of the image time stamp was at 1513:51.26 UT and was located at 33.19°S, 132.89°E, well outside the camera FOV. Thus it cannot be the parent lightning seen in the image. Figure 4d presents the radiance-exposure product for the two sprites in mJ m−2 sr−1. Assuming an average duration of 10 ms for the sprites (observed in a single video frame), we computed a total (surface) brightness of 1.14 ± 0.1 (left element) and 0.79 ± 0.08 (right) MR. The present observation of sprites above the ocean is a significant addition to the limited number of oceanic sprites reported thus far.
5.1. Detection and Occurrence Rates
 The MEIDEX sprite campaign was limited in duration and was considered a secondary science objective for the primary dust experiment. In order to achieve the maximum yield out of the very limited set of observations, we developed a forecast method and a pointing capability that helped the shuttle crew conduct targeted observations toward areas with a high probability of TLE occurrence. In addition, the real-time pointing by camera maneuvers was a significant success factor that greatly enhanced the detection rate.
 The total number of TLEs (elves and sprites) we have discovered so far in the data is 17 out of 254 min of observing time analyzed. The summary of detected sprites is presented in Table 3. In three 5 min samples (orbits 48, 66, and 87) of thunderstorms in different locations, we have located seven sprite events, which occurred above different cells separated by hundreds of kilometers. If we consider the number of sprite elements in each event, the total is 11. Although it is known that sprites tend to appear in places preceded by other sprites [Stenbaek-Nielsen et al., 2000; Moudry et al., 2003], we presume that these events were separate and independent.
Table 3. Summary of MEIDEX Confirmed Sprite Events, Found After Analysis of 2/3 of the Dataa
MET Time, UT
Number of Sprites
There were 20 additional suspected events not listed here.
 Only approximately 1/5 of the data we analyzed (254 min) was actually over stormy weather and contained lightning. This means that the average detection rate of TLEs during the MEIDEX was 0.33 events per minute. This is a much higher rate than reported after the MLE [Boeck et al., 1998] or in the International Space Station (ISS) observations [Blanc et al., 2004]. If we only consider sprite events (and not elements), the detection rate is 0.13 sprites per minute. We assume that our camera had detected only the high-energy tail of the sprite brightness distribution, being biased toward bright events that were located near the limb when the shuttle was pointed in that direction. Thus the reported value sets a lower limit on the occurrence rate.
 To date, there is no reliable quantitative assessment of the prevalence of sprites on a global scale. Reports by Lyons [1994b, 1996] during summer observation campaigns from Yucca Ridge Field Station (YRFS) in Ft. Collins, Colorado, stated a number of the order of 1000 sprites for a 6-week campaign. This yields an average of 25 events per night, which translates to 4 per hour. However, this value probably represents only summer conditions above that region of the United States with a bias toward large mesoscale convective systems (MCSs). Fullekrug and Price  estimated a sprite rate of 60–70 per night (or ∼10 per hour) over the African continent. Heavner et al.  suggested that the upper limit for the global rate is 1 sprite per second on the basis of the assumption that the global sprite distribution corresponds to the lightning distribution that exhibits a ground flash rate of 10–14 s−1.
 If we consider that there are approximately 1000 storms globally at any given time [Rycroft et al., 2002], with a total global flash rate of 44 ± 5 s−1 [Christian et al., 2003], and if we assume that only a quarter of these are ground flashes (CGs), we get 750 CGs per minute worldwide. Since sprites are exclusively produced by strong +CGs (though some observations report otherwise) [Sao-Sabbas et al., 1999], we can estimate that only 1% of those are +CGs that have large enough charge moments to generate sprites, so that the statistical overall global rate is of the order of 7.5 sprites per minute.
 The average global sprite rate cannot be accurately calculated from the limited MEIDEX set of observations. However, the wide geographical distribution of sprites found during the mission and the relatively high detection rate can assist in estimating a lower limit. On the basis of our forecast method products and analysis of satellite images during the mission [Ziv et al., 2004], we conclude that the storms we observed were not special in any meteorological aspect. Thus we can assume that they represent the ordinary, ambient condition of tropical lightning activity for the Southern Hemisphere summer. Multiplying the detection rate of sprites (0.12) by a conservative estimate of simultaneous tropical thunderstorms (100), we get a rate of 12 sprites per minute in the tropics. This value is of the same order of magnitude as the one reported by Heavner et al.  and corresponds well with the statistical analysis above. Obviously, only long-term observations from orbiting space platforms can retrieve a reliable estimate of the global sprite rate.
5.2. Meteors and Sprite Formation
 Elves and sprites were already observed in conjunction with meteor activity during the Leonid Meteor Shower Airborne Campaign in 1999 [Yano et al., 2001]. There, 11 sprites and 33 elves were observed in 1.06 hours above a European thunderstorm in the Balkans at the time when the Leonid flux was near its peak [Jenniskens et al., 2000, Figure 6]. A meteor was also reported to initiate a blue-jet event in association with the occurrence of a sprite during the SPRITES'98 campaign [Suszcynsky et al., 1999]. Still, a relation between the meteoritic flux and the occurrence of TLEs has not been uniquely established [Wescott et al., 2001]. At least one theory [Symbalisty et al., 2000] suggests a direct link between meteor ablation in the mesosphere and the formation of Columniform (or C-) sprites. According to model calculations, the flux of particles in the meteor trail reduces the ambient atmospheric conductivity so that a strong cloud-to-ground stroke occurring within ∼1 hour of the formation of the meteor trail can trigger a temporally brief column of light known as a C sprite. Zabotin and Wright  have suggested that the presence of small particles of meteoric origin in the mesosphere and stratosphere explains features in sprite formation and fine structure. Presumably, the surfaces of these conducting dust particles contain microspires that amplify the electrostatic field, leading to explosive emission of electrons. Similarly, Belevkina et al.  suggest that cosmic dust particles can act as seeds for the formation of sprite tendrils. Especially, solid iron and magnesium grains can magnify the ambient electric field by a factor of 105–106, making the atmosphere at mesospheric heights more susceptible to electrical breakdown. However, no clear evidence has been obtained linking meteors and sprites. Sao-Sabbas et al.  had suggested that conductivity inhomogeneties in the mesosphere, caused by (among other factors) meteoritic dust particles, may play an important role in the observed lateral displacement of sprites with respect to their parent ground stroke.
 On the basis of the American Meteor Society database, 22 January coincides with the activity period of several meteor showers (Delta Cancrids, Canids, Eta Carinids, Eta Craterids, January Draconids, Rho Geminids, Alpha Hydrids, and Alpha Leonids). Thus it is reasonable to expect that the two meteors we observed represent only a part of the total meteoritic flux and that they were preceded and succeeded by other events not observed by our camera. The termination height of a specific meteor depends, among other factors, on its composition, size, velocity, and trajectory and is different for various showers and sporadic meteors. For example, during the 1999 Leonid storm, Brown et al.  found termination heights between 75 and 110 km (mean 95 ± 0.56 km). Slower meteors end at a lower altitude.
 The termination altitude of the light emission from the first meteor (∼85 km) approximately coincided with the height of the triple sprite observed less than 30 s later. Indeed, it is hard to establish that this specific meteor triggered the observed sprite, being separated laterally by ∼200 km and because the position of the meteor in the image does not correspond to a specific feature in any of the three sprite elements shown in Figure 3d. Presumably, particles deposited along the trail of similar meteors served as the source of mesospheric irregularities and sought to explain sprite appearance [Wescott et al., 2001], but this cannot be verified from our image.
 Thus, even though no direct causative relationship between sprites and TLEs was found during our observations, the proximity of our observed events (e.g., meteors and TLE) in time and space supports existing theoretical studies about the role that meteors play in TLE generation and evolution and may not be a mere coincidence. More observation campaigns of sprites are needed during known peak dates of meteor showers.
6. Summary and Conclusions
 In the video from the 13 orbits analyzed thus far, we have positively identified 17 TLEs in less than ∼51 min of accumulated thunderstorm data (7 sprites, 10 elves together with 20 suspected events), a significantly higher detection rate compared to the MLE and LSO observations. The geographical distribution of these events shows that when there are thunderstorms in the tropics, there is a high probability that some form of accompanying TLE exists. We have obtained images over the Pacific and Indian Oceans, in the central south Atlantic and over Argentina, Brazil, north Australia, Tasmania, Congo, Nigeria, and the Borneo and Fiji peninsula. If we consider the limited observation time and detection efficiency of the MEIDEX payload, the wide geographical distribution of our successful observations provides evidence to the wide-spread nature of TLE occurrence on a global scale.
 In conclusion, the MEIDEX sprite campaign succeeded in recording various types of TLE in numerous geographical locations and obtained a considerable amount of new observations. Contrary to remote-controlled or automatic robotic observations, the human factor played a significant and indispensable role in the real-time target acquisition, greatly enhancing the probability of capturing TLEs. The mission proved the flexibility and global coverage of sprite observations from space and set a benchmark for future satellite- and ISS-based TLE observations, such as the global survey by the ROCSAT-2 satellite [Chern et al., 2003] and other missions presently in the planning stage, such as the French TARANIS mission [Blanc et al., 2004]. The novel human-based “hunting technique” for TLEs from space proved to be very efficient and may be adapted in future space-based campaigns from satellites and from the ISS.
Appendix A:: Calibration Procedure of Sprite Images
 Here we describe the extraction of radiance data from the MEDIEX calibrated sprite images. During the preflight calibration process at NASA GSFC Laboratory for Atmospheres, each pixel of the camera CCD was filled by the uniform spectral radiance B (W m−2 sr−1 nm−1) emitted from the NIST traceable calibrated integrating sphere. When determining a certain exposure time τ (ms) of the camera to this uniform spectral radiance, a certain video signal (gray level (GL)) was obtained. Since the camera had filters with a finite band pass, the spectral radiance B (W m−2 sr−1 nm−1) was integrated over this band pass and normalized to the filter transmittance yielding the filter radiance N (W m−2 sr−1).
 As a result, GL is proportional to the product of N and τ (ms* W m−2 sr−1) or simply (mJ m−2 sr−1). This proportionality is almost linear. The inverse relation between N and GL is the calibration of the camera. In this calibration method each GL of any pixel is converted to a product N*τ (mJ m−2 sr−1), where the area is related to the object plane (the integrating sphere or the measured target, in this case, the sprite). When the TLE is imaged by the camera, the different N*τ values of all its pixels (area) are summed to yield its total Σ(N*τ) (mJ m−2 sr) value.
 The total energy that is emitted from the TLE to a solid angle (J sr−1) is obtained by multiplying the value of Σ(N*τ) by the area of a single pixel. Note that we multiply by the area of a single pixel and not of all the pixels because the summation on all TLE pixels had already been done (the area of a single pixel (m2) is the product of the angle, the pixel subtends (sr), and the square of the distance (m) to the TLE). The total energy that is emitted from the TLE to a solid angle that is converted to the total number of photons by dividing E (J sr−1) by the quanta of a photon hν (we assume a central emission line in each filter). This yields E (photons sr−1). Assuming that the TLE duration τ (s) is known, we can divide the value of E (photons sr−1) by τ to get the total photon flux intensity from the TLE M (photons s−1 sr−1).
 The measured total photon flux intensity from the TLE M (photons s−1 sr−1) is converted into the luminous flux (photons s−1 m−2 sr−1) at the TLE plane by dividing it by the total area of the TLE (the total number of its pixels multiplied by the pixel area, m2). The result is insensitive to the distance between the observer and the TLE because we multiply and divide by the square of the distance from detector to target.
 The Rayleigh is a unit of luminous flux used to measure the brightness of the airglow and the aurora, first proposed by Hunten et al. . One Rayleigh is 106/4π quanta (photons) per square meter per second per steradian (or 7.96 × 104 photons s−1 m−2 sr−1). We convert the luminous flux (photons s−1 m−2 sr−1) at the TLE plane to Rayleigh units by dividing the result by 7.96 × 104 photons s−1 m−2 sr−1.
 This research was made possible by the devotion and enthusiasm of the Columbia crew: Rick Husband, William McCool, Michael Anderson, David Brown, Laurel Clark, Kalpana Chawla, and Ilan Ramon. The MEIDEX is a joint project of the Israeli Space Agency and NASA. We wish to thank S. Janz and E. Hilsenrath of the Laboratory for Atmospheres at NASA GSFC for their help in the calibrations of the Xybion cameras. Special thanks to the Hitchhiker team at NASA GSFC: T. Dixon, M. Wright, K. Barthelme, S. Applebaum, C. Knapp, K. Harbert, and to A. Lalich and T. Schneider, STS-107 flight planners at NASA JSC, for making this experiment possible. The WWLL network map came courtesy of Craig J. Rodger, University of Otago, New Zealand, supported by Marsden Fund contract 02-UOO-106. Thanks also to Martin Fullekrug, University of Frankfurt, for his help with the Wetterdienst IR satellite images.