We have succeeded in observing sprites for winter lightning in the Hokuriku area (Japan Sea side) of Japan in the winter of 2001/2002. The optical results on 3 days are compared with the corresponding characteristics of parent (causative) lightning with particular attention to the significant differences between Hokuriku winter lightning and the more widely studied continental lightning. Despite significant differences with Hokuriku winter lightning, we have found nearly the same sprite properties as already observed in the U.S. continent with a significant difference (simpler shape for Hokuriku winter sprite). Then, we have also discussed the criteria for sprite occurrence. Specifically, two similar criteria are found: (1) cloud-to-ground discharges of positive polarity and (2) the presence of a certain threshold in vertical charge moment (200–300 C km) (roughly consistent with that for the U.S. continent). Mesoscale convective systems are not necessary to store the charge necessary for sprites, but the parent Hokuriku winter clouds are substantially smaller than the minimum scale for sprite occurrence in the continental lightning; however, it is larger in area than ordinary summer thunderclouds. However, there may exit another condition such as clustering or self-organizing effect of thunderclouds for sprite production.
 Sprites are a newly discovered optical emission in the mesosphere over large thunderstorms. Since their discovery [Franz et al., 1990] many papers have appeared on their occurrence, their morphological features, their association with causative lightning, and their associated phenomena (including ELF sferics and ionospheric perturbations) (see the reviews, for example, by Rodger  and Williams ). An additional optical phenomenon named elves has been found by Fukunishi et al. , and it has been hypothesized that it is due to coupling of the electromagnetic pulse of the lightning discharge with the lower ionosphere [Inan et al., 1991; Nickolaenko and Hayakawa, 1995]. Several generation mechanisms have been proposed for sprites by different authors (see the reviews by Rodger  and Fukunishi ), including the QE (quasi-electrostatic) electric field model by Pasko et al. [1995, 1997] in which a strong secondary electric field above the cloud is imposed after the positive ground discharge, leading to the generation of sprites. The QE model is appropriate in explaining some general macroscopic characteristics of sprites, but other aspects like the fine structure of sprites, which are becoming essential subjects, need further theoretical study, though streams, which are the cause of the sprite fine structure, might be a part of the QE model.
 Recently, sprites have been observed in different geographical locations all over the world like North and South America, Australia, Europe, and Japan [Boeck et al., 1995; Hardman et al., 2000; Fukunishi et al., 1999; Neubert et al., 2001; Su et al., 2002]. However, the significance of observations in different geographical situations has not been previously addressed. The important question arises as to whether the characteristics and morphology of sprites in the other geographical locations are the same as those already found in the United States and what are the essential criteria for inducing sprites when the lighting characteristics seem to be different from those in the U.S. continent. In this sense, the observation of sprites for the lightning discharges in the Hokuriku area (Japan Sea side) is of essential importance because the lightning characteristics here [Brook et al., 1982; Takeuti and Nakano, 1983; Takeuti, 1987] are significantly different from those in large continents like the U.S. where large mesoscale convective systems (MCSs) are known to be prevalent. This paper presents our own optical observation of sprites for winter lightning discharges in the Hokuriku area. We study the general features of Hokuriku sprites and then explore essential criteria for inducing sprites in response to different Hokuriku lightning characteristics in comparison to continental lightning.
2. Optical and Elf Observations
 The 2001/2002 sprite winter campaign was performed between December 2001 and February 2002. We have installed our optical measurement system on the top of the building of Tokai University at Shimizu, Shizuoka (geographic coordinates: 34.96°N. 138.52°E). The camera system used in this campaign consisted of a CCD imager and an image intensifier (Hamamatsu Photonics CCD camera C3077, Hamamatsu Photonics Nightviewer C3100, lens of Nikon Ai AF Nikkor) and a video cassette recorder (SVO-9600, S-VHS). This camera system has a viewing area of 36 degrees (height) × 26 degrees (viewing azimuth). The optical field site Shimizu is far from our university, so we visited the site only when there seemed to be a high probability of having sprites in Hokuriku on the basis of the forecast information from the Hokuriku Power Electric Company. Once we were there, we fixed the look direction from Shimizu in such a way that the center of our viewing azimuth was coincident with the direction of the most probable lightning area in Hokuriku, and we carried out a 10-hour continuous observation. We will show the data for 3 days (14 December 2001 and 23 January and 28 January 2002).
 The details of our fully calibrated observation (frequency <1 kHz) in Moshiri, Hokkaido (geographic coordinates; 44°22′ N, 142°16′E) have already been described by Hobara et al. [2000, 2001]. Two induction coils are used as magnetic sensors, and magnetic field calibration was carried out by inserting each coil antenna in a long solenoid. Also, the vertical antenna was calibrated with the use of a parallel-plate calibration antenna as is shown in the work of Nickolaenko and Hayakawa . The dynamic triggering level of recording an ELF transient waveform is set in such a way that the total magnetic field (B = ; where Bx and By are two horizontal magnetic field components) exceeds 20σ (where σ is the standard deviation of the field fluctuation).
3. Optical Results for Hokuriku Sprites
Figure 1 shows two examples of Hokuriku sprites; Figure 1a shows a group of column sprites (event 5 in Table 1) and Figure 1b shows a carrot-type sprite (event 5 in Table 1). Table 1 is the summary of sprite observations, in which we detect sprites on only two nights (14 December 2001 and 28 January 2002) (no sprites were observed on 23 January 2002). Table 1 indicates that 12 sprite events were observed. The first column indicates the event number. The second and third columns indicate the date and time (in UT) of sprite occurrence. The fourth column indicates the estimated height range of sprites. The cases where there are several height ranges for one particular time means that there are several observed sprites in a group. Now we describe how to estimate the height range of sprites. We first identify the causative (or parent) lightning of sprites. The sampling of optical images is 30 ms, so that the accuracy in timing of images is considered to be ±15 ms. The accuracy in timing of lightning occurrence by the JLDN (Japan Lightning Detection Network) is 1 ms and also the GPS time in the camera is given every 10 ms. We take into full account the accuracy in these timings, and also we consider the previous observational fact that almost 50% of sprites are observed to be initiated in the first 20 ms after the causative lightning discharge, while many other sprite are associated with much longer lightning-to-sprite delay (up to several hundred milliseconds) (see a review by Rodger ). So we tried to find the lightning by the JLDN data in a range from −1 s to +20 ms of the sprite time, and the corresponding results are given in the fifth column. When a single value is indicated, this means that there is only one possible lightning discharge responsible for those sprites, and its peak current is also given in the sixth column. Once we know the location of the causative lightning for each sprite from the JLDN data, we assume that a sprite is taking place over the lightning position as based on previous finding that a sprite is occurring mainly with a radius of ∼25 km above the causative lightning [Lyons, 1994; Sentman et al., 1995; Winckler et al., 1996]. By taking into account this uncertainty, the distance between the optical site and possible lightning position, and the elevation angle, the maximum error in estimating the sprite height would be ∼10 km. We then estimate the height range of the sprite by using the elevation angle measured in the optical image. Figure 2 illustrates the location of parent lightning discharges associated with the sprites in the table. The numbers in Figure 2 correspond to the numbers in Table 1. Twelve cases in Table 1 are described one by one. For the first event there are observed two column sprites successively, which are associated with a single positive ground flash (CG). The second case is again a column sprite, but there are two possibilities of parent lightning discharges as seen in Figure 2. However, we take the lightning whose position is closer to the viewing direction of those sprites from Shimizu so that the second possibility is given in a bracket. For the next two events, 3 and 4, it is found that there is no corresponding lightning within our timing limit in the JLDN data, so that the peak current is not given and also the sprite height cannot be estimated. For the fifth event, there are indicated two values of peak current (+142 kA and −122 kA), which means that there were two lightning discharges with their temporal separation of 1 ms but nearly at the same place. It seems uncertain which lightning discharge is the parent of the sprite, but positive-polarity lightning is confirmed because we have found the positive polarity for Qds. The sixth event is the same as the cases of events 3 and 4. In events 7 and 8 there was only one lightning discharge which is responsible for several sprites. For event 7 we find two height ranges for one sprite; the higher height corresponds to the sprite main body, while the lower height corresponds to its so-called tendril (as shown by Williams ). Event 8 is the weakest sprite in our observations. Events 9, 10, and 11 are the same as events 3 and 4 in the sense that no corresponding lightning satisfying our timing limit is detected by JLDN. In event 12 we observed two lightning discharges with the same positive polarity, but we have chosen the lightning whose azimuthal direction is closer to our camera viewing direction, as the first promising one. As in the case of event 7, there are two indications for the only one carrot sprite for event 5, and the lower height corresponds to the tendril.
Table 1. List of Sprite Events Observed in the Hokuriku Area and Characteristics of ELF Transients on 2 Days
Height range, km
Peak current, kA
Transient (Y or N)
Qds, C km
Figure 3 is the summary on the height range occupied by sprites. We compare this result with the previous American result: Lyons  came to the conclusion on the basis of 36 events that the upper-end height is, on the average, 77 km, with the maximum value being 88 km. We estimate the average height of upper end of our winter sprites to be ∼78 km, with the maximum value being 104 km. Even thought we think of the maximum possible error, the upper end height is nearly the same for both sprites in the Hokuriku and in the United States.
 As for the associated ELF transients, the existence (or absence) of the associated ELF transients at Moshiri is given in the seventh column (Y, meaning Yes, for existence and N, meaning No, for absence) and the eighth column gives you the information on the characteristics of ELF transients observed. Then, the last, ninth column provides the estimation of the vertical charge-transfer (charge-moment) when we observed an ELF transient. Based on the observed frequency characteristics of ELF waves, we estimate the charge transfer using the impulsive approximation [Huang et al., 1999; Hobara et al., 2002]. Figure 2 illustrates the position of sprite-inducing lightning discharges observed in the Hokuriku area on the 2 days, together with our optical site (Shimizu).
 We summarize the features of Hokuriku sprites as follows.
 1. Hokuriku sprites are observed in the UT time interval form 1400 to 1800 (or LT = 2300 to 0300; midnight phenomena), which is consistent with the diurnal variation of winter lightning occurrence in the Hokuriku area [Takeuti and Nakano, 1983; Takeuti, 1987].
 2. Most of the sprites observed are of columnar shape and we have observed only one carrot shape, which suggests much simpler shapes for Hokuriku sprites than for sprites over large continents.
 3. Height distribution of sprites is found to be in a range from 50 km to 90 km, with the most probable occurrence at 70–75 km. The height of upper end of sprites is not found to be very different from that for continental sprites. The lowest end height of the sprites seems to be higher in Hokuriku than in the United States.
 4. The maximum horizontal spread in groups of sprites is about 70 km (with the dimension of one column sprite 1–2 km).
 The second conclusion on the simpler shape based on 12 events in this paper is supported by our previous paper by Hobara et al.  by using the optical measurement by the Tohoku University group. Such a structure difference including the shape and also its fine structure may be related with the charge moments of the parent lightning, which will be studied in future. As the conclusion we can say that the general features of Hokuriku sprites are found to be consistent with or to support previous results observed for the U.S. campaign [Sentman and Wescott, 1995; Lyons, 1996; Williams, 2001] in the sense that morphological characteristics of Hokuriku sprites are not exceptional as compared with those of continental ones. However, there seem to exits one significant difference like the reduced variety of sprite structure as compared with other observations (as in the work of Williams ).
4. Criteria for Sprite Generation
 Sprite observation over large continents like Europe, Africa, Australia, etc. [Boeck et al., 1995; Hardman et al., 2000; Neubert et al., 2001; Su et al., 2002] might be important to evaluate or elaborate the previous results extensively obtained already in the U.S. continent because those continental summer lightning discharges are nearly the same as those in the U.S. continent. The lightning characteristics of Hokuriku winter lightning in Japan are known to be significantly different from those for the summer lightning over large continents [Brook et al., 1982; Takeuti and Nakano, 1983; Michimoto, 1993; Adachi et al., 2002], therefore it is useful to study the effect of difference in lightning characteristics (between the summer lightning in the big continent and winter lightning in the Hokuriku area in Japan) on the occurrence and characteristics of sprites. In this sense we will pay close attention to the two important publications by Huang et al.  and Hobara et al.  in order to try to find the essential condition for sprite generation. Sprites are known to occur almost exclusively with positive ground flashes (though there is a report by Barrington-Leigh et al.  that suggested the existence of negative-lightning-produced sprites), but a majority of positive flashes are not linked with sprites which suggests the presence of any key factor.
 We now illustrate the characteristics of winter Hokuriku lightning on the 3 days mentioned before. Three factors of lightning are mainly discussed: (1) spatial scale, cloud height etc., (2) peak current (plus polarity), (3) charge-transfer (charge moment), and finally (4) height range of sprites. The first point is studied by means of the radar image data observed at the Komatsu base, and the study of the second point is based on the information from the JLDN data. The third point is investigated on the basis of the ELF observation in Moshiri, Hokkaido [Hobara et al., 2002], and the estimation of charge moment (Qds) has already been explained before [Nickolaenko and Hayakawa, 2002]. Here, we need the information on the distance and azimuth of the source (lightning) from the observing station. So, when we know the position of the parent lightning, we use this information to determine the source-receiver distance in the ELF waveform analysis and the results of Qds are given in Table 1 (last column). However, you can find the value of Qds even though there is no lightning detected by the JLDN, for example, in events 4, 6, 9, and 10. For event 4, the sprite direction is very close to that for event 5, so that we assume that the causative lightning is taking place at the same place as in case 5 (i.e., Komatsu in Figure 2) and we estimate the Qds (the result is indicated in the last column in Table 1, with the notation of ∼ meaning “about”). The same procedure was adopted for two cases, events 6 and 9, in order to estimate Qds. The ELF transients in case 4 are multiple being composed of a few transients, so that the accuracy in Qds estimation is worse. The ELF transient in case 10 is a noisy nature, and hence it seems difficult to estimate Qds. The accuracy of these estimations with the notation of ∼ is about 10% by changing the source-receiver distance by ±100 km.
4.1. Is a Mesospheric Convection System (MCS) Essential?
 From the observation in the U.S. continent it is initially found that sprites tend to occur for a large mesoscale convective system (MCS) whose spatial scale of radar coverage is 500 km × 500 km [e.g., Boccipio et al., 1995], and further studies by Williams  and Huang et al.  have indicated that MCS is a necessary condition for sprite occurrence. However, a paper has recently been published that states that sprites have occasionally been observed above smaller storms covering 2500–7500 km2 [Stanley, 2000], but such occurrences appear to be much more the exception than the rule. Is the extremely large scale essential for sprite occurrence in the Hokuriku area? Our study of radar images during the most developed thundercloud phase for Hokuriku winter lightning on our three days has indicated that the horizontal scale of radar reflectivity is not so large as that for MCSs (as defined to be 20,000km2 by Mohr and Zipser ) in the U.S. continent, and Figure 4a shows one example of a radar CAPPI (Constant Altitude Plan Position Indicator) at a height of 5 km with the strongest dBZ when we observed a carrot sprite in Table 1 (14 December, 1640–1782 UT). The observing station, Komatsu is located in the center, and the image intensity is indicated in color. Also, the position of the channel to ground of the lightning inducing that sprite is indicated by plus sign. A multiplicity of embedded cells is evident, each with a horizontal dimension of the order 20–30 km. Even at the place when we observed the carrot-shaped sprite, the cloud structure is of the same spatial scale. The sprite cloud in our study is found to be considerably smaller than the MCS as defined by Mohr and Zipser . Also, our scale is smaller than the minimum value for sprite generation by Stanley , but it is still an order of magnitude larger than that of a typical isolated thunderstorm [MacGorman and Rust, 1998; Williams, 1998]. This statement is confirmed to be valid for all sprite events.
 Next we discuss the cloud height problem. The cloud top of MCSs in the U.S. continents is known to extend up to ∼15 km [MacGorman and Rust, 1998]. In this sense, the Hokuriku winter lightning can be concluded to be located at a much lower height by at least a factor of two. Figure 4b shows one example of vertical cross section of radar image for the same event in Figure 4a but along the line in EW direction crossing the sprite lightning. Figure 4b illustrates that the thundercloud is shaped as a flat-topped stratiform, extending extensively horizontally, but extending in height only from 2 to ∼6 km. The two important distinctions with U.S. continental lightning are (1) that Hokuriku winter thundercloud is not a large MCS and (2) its different cloud structure, but we observe sprites in Hokuriku despite these differences. Here we have to refer to the latest study by W. A. Lyons et al. (Characteristics of sprite-producing positive cloud-to-ground lightning during the 19 July 2002 STEPS mesoscale convective systems, submitted to Journal of Geophysical Research, 2003). They have found on the basis of lightning mapping array for two MCS studies in the U.S.A. that +CG did not produces sprites until the centroid of the maximum density of lightning radiation emissions dropped from the upper part of the storm (7–11.5 km altitude) to much lower altitudes (2–5 km). This suggests a possible linkage between sprite-parent CGs and melting layer/bright band charge production mechanism in MCS stratiform precipitation regions. A comparison of this new result for summertime American sprites with our corresponding result for winter lightning in the Hokuriku area would be a future subject to study.
4.2. Positive Cloud-to-Ground Discharge?
Figure 5 illustrates the distribution of maximum current of the Hokuriku lightning on the three days for both polarities (positive or negative) based on the lightning data by the JLDN, together with the corresponding information when we observed sprites. Hokuriku lightning has been found to exhibit an interesting feature of having both polarities in nearly equal number (e.g., the percentage of +CG among the whole is 50% on 14 December 2001 and 52% on 23 January 2002 and 43% on 28 January 2002), supporting the early result by Brook et al.  and Takeuti and Nakano . We can definitely conclude from Table 1 and Figure 5 that sprites are exclusively associated with positive cloud-to-ground discharges, which lends further support to our previous same condition for other winter lightning by Hobara et al. . It is already agreed that +CG maximum current does not appear a robust predictor of sprite potential [Huang et al., 1999].
Figure 6 shows the temporal evolution during 10 hours of the lightning characteristics (maximum current (Figure 6a) and the charge moment (Qds) (Figure 6b)) on a particular day, 14 December 2001. We have to comment on the Qds value determined for the lightning for sprites. As is seen from Figure 6 and Table 1, the Qds value is found to lie in a range from ∼160 C km to >1000 C km. However, Table 1 indicates that the Qds values exceeding 400 C km for three cases are definite and reliable, while there may be included some uncertainty (like multiple ELF in the eighth column in Table 1) in estimating Qds for other cases. Combining our present estimation and the first corresponding results for the Japanese winter campaign (on the basis of about 20 sprites) in the winter of 1998/1999 by Hobara et al. , the Qds value is found to range from ∼200 C km to ≥500 C. When we assume ds = 5 km for the Hokuriku lightning as in Figure 4b and we assume an upper positive charge reservoir of the titled dipole model [Brook et al., 1982; Michimoto, 1993], the charge (Q) must be rather large, Q = 40 C to ≥150 C. We now compare our Qds value for sprites with those observed for the summertime continental lightning by Huang et al.  and Hu et al. , and we have found that the lowest threshold of Qds for inducing sprites in the continental situation is 200–300 C km. We then conclude that the lowest threshold for Qds does not exhibit any significant difference between the Hokuriku winter lightning and summer continental one. Based on the mechanism for sprite production proposed by Wilson , Qds is considered to be a good indicator in discussing sprite occurrence. It seems to us that there must be a certain threshold of Qds for sprites caused by Hokuriku lightning of the order of 200–300 C km. When we look at Figure 6b, there are a few lightning discharges whose Qds is exceeding the observed threshold for sprite occurrence. However, one lightning discharge just before 1200 UT is found to be located just outside our camera viewing range, and another one with a tall black bar at 1530 UT is also found to be completely outside our viewing area. So we cannot verify that these two discharges have induced sprites or not. The corresponding figure on 23 January (though not shown) indicates low lightning occurrence (one third as compared with the previous two days), low peak currents (−190 ∼ +390 kA) and low Qds (less than 230 C km), which suggests an extremely low probability of occurrence of sprites. This is consistent with no observation of sprites on this day.
4.3. Altitude Range of Sprites
 The height of sprite initiation is one possible additional parameter to compare between continental U.S. and Hokuriku winter lightning. Williams  refined the theoretical determination of sprite threshold by (1) using a more accurate altitude-dependence of air density and (2) including the multiple image effects of the ionosphere. Unfortunately, our poor time resolution has not enabled us to estimate the sprite initiation height. The altitude distribution of all the sprites in Table 1 is already summarized in Figure 3. Also, we have found that the upper-end height for Hokuriku sprites seems to be nearly the same as that for the U.S. sprites event though we take into account the error of our estimation. The optical estimation of the lowest-end height seems to be strongly dependent on the S/N at the observing site and the optical environment at Shimizu is not so good. Even if we take into account this measurement sensitivity problem, the lowest-end height for Hokuriku sprites is higher than (of the order of ∼10 km) that for summertime continental ones (the usual lowest height ∼50 km with heights of tendrils at ∼30–40 km [Lyons, 1996; Sentman and Wescott, 1995]. When we observe tendrils for a few cases in Table 1, their altitude is found to be 54 ∼ 60 km. It is not certain whether the observed higher altitude of the lowest-end of our sprites is simply due to the observation uncertainty or it is of physical importance. In the latter case, this difference in the height of lowest-end of sprites has some relations to the charge moment or to the local air density profile.
4.4. Any Other Significant Factor?
 In addition to the above-found necessary conditions of (1) positive cloud-to-ground discharges and (2) the threshold of Qds (irrespective of different characteristics of lightning in the US continent and Hokuriku area), there may be significant factors in generating sprites. A few hints have provided by the foregoing observations.
 The presence of a MCS is found not to be a necessary condition for inducing sprites, but we have found that the parent lightning for winter lightning in the Hokuriku area is still smaller than the minimum size by Stanley  but must be larger-than-average sized. Furthermore, we have found in this paper that the parent lightning whose Qds exceeds the threshold, has always triggered a sprite. Then, the next question is whether the Qds threshold is a sufficient condition. First of all, Figure 6 suggests that sprites tend to appear in a group; indicating some kind of clustering in temporal evolution. Figure 4 has indicated the presence of larger-than-average thundercloud cells responsible for sprite occurrence, but the thundercloud cells are found to form a group or to be clustered spatially (connected with each other) when a sprite is detected. In this sense the Hokuriku lightning is self-organized in such a way that such small elementary cells are connected with each other, forming a very large horizontally extended spider-like structure. Qds is a macroscopic quantity (effective vertical charge moment), which refers only to the macroscopic structure of lightning cloud. However, the fine structure such as the contiguity of charge among the cells would be of essential importance in triggering a sprite, and this is closely related to the macroscopic quantity of Qds. We are now studying this self-organization by means of fractal analysis.
 We have succeeded in observing sprites for winter lightning discharges in the Hokuriku area (Japan Sea side) of Japan by our optical measurement at Shimizu. On 2 days during the period of December 2001 to February 2002, a considerable number of sprites have been observed, and the relationship of those sprite occurrences with the causative lightning characteristics (spatial structure, maximum current, and charge moment (Qds) from ELF observation in Moshiri, Hokkaido) has been discussed. Important finding are summarized as follows.
 1. Sprites observed are found to exhibit morphological characteristics rather similar to those of sprites observed in the U.S. continents, and the most significant difference may be the simpler structure (mainly columnar) for the Hokuriku sprites. These simpler structures might be linked to small charge moment changes, since the charge moment changes of these observed sprites are all below 1000 C km.
 2. Causative (parent) lightning discharges for those Hokuriku sprites are found to satisfy the following two conditions: (1) positive cloud-to-discharge and (2) charge moment (Qds) seeming to exceeding 200–300 C km. The second condition is found to be rather consistent with the value for the summertime continental lightning.
 3. Thundercloud characteristics of Hokuriku winter lightning are found to be essentially different from those of continental lightning (for which sprites are first observed); large Q value, smaller cloud height (ds = a few kilometers), smaller spatial structure (mainly extended horizontally with a thin vertical structure range = 2 ∼ 6 km), etc. Nevertheless, sprites are detailed in Hokuriku. Hence the vertical change moment (Qds) is a good indicator for generating sprites; i.e., Q itself is not an important quantity, but Qds is a universal quantity as theoretical considerations indicate that Qds is the fundamental quantity for predicting dielectric breakdown in the mesosphere over thunderclouds.
 4. Positive polarity and Qds seem to be conditions for sprite occurrences. There may exist any essential factor(s) for sprites such as the self-organization of thundercloud leading to the contiguity of charge among the cells. The change in fine structure might result in an increase in macroscopic Qds value.
 The authors would like to express their science thanks to K. Michimoto for providing them with the radar image data at Komatsu. We are also thankful to T. Nagao of Tokai University for providing us with the optical field site.
 Arthur Richmond thanks H. Fukunishi and another reviewer for their assistance in evaluating this paper.