This study examined the meteorological characteristics of precipitation systems that produced 38 “sprite-class” negative cloud-to-ground (CG) strokes (i.e., peak currents in excess of 100 kA and charge moment changes in excess of 800 C km) as well as those that produced three confirmed negative sprites on 23 different days during 2009–2011. Within 15 km of the negative sprite-parent/class stroke, the median characteristics for these systems were to produce negative CGs as 69.2% of all CGs, and for the 30 dBZ radar reflectivity contour to reach on average 14.2 km above mean sea level (MSL), during a 25 min period encompassing the occurrence of the stroke. The median contiguous area of 30 dBZ composite radar echo (i.e., maximum value in the vertical column) for these systems was 6.73 × 103 km2. All but three of the discharges occurred in intense multicellular convection, with 30 dBZ exceeding 10 km MSL in altitude, while the others occurred in the stratiform regions of mesoscale convective systems. All but six of the systems produced greater than 50% negative CG lightning, though flash rates tended to be low near the stroke (1–2 min−1 on average). The results suggest that negative sprite-parent/class lightning typically occurs in precipitation systems of similar size and intensity as those that produce positive sprites, but not necessarily the same systems, and the negative lightning normally strikes ground in the convection rather than the stratiform precipitation. However, upper-level positive charge in the convection may play an important role in sprite-class/parent lightning of either polarity.
 While the meteorology of positive sprites (i.e., sprites associated with positive cloud-to-ground lightning) is relatively well known, at least in a gross sense [Lyons, 1996, 2006; Williams and Yair, 2006; Lyons et al., 2009], the meteorology of negative sprites (i.e., sprites associated with negative cloud-to-ground lightning) is largely unexplored. This is because there are so few observations of negative sprites. To date, there have been only a small number of peer-reviewed studies documenting the occurrence of approximately 10 confirmed negative sprites [Barrington-Leigh et al., 1999, 2001; Taylor et al., 2008; Soula et al., 2009; Lu et al., 2012; Li et al., 2012]. By contrast, a single precipitation system can produce more than 10 times that number of observed positive sprites in just a few hours [e.g., Lang et al., 2010].
 The scientifically accepted theory of sprite production is polarity agnostic [Wilson, 1925], so the relative rarity of negative sprites in relation to positive sprites has been a topic of interest in the atmospheric electricity community. Research has progressed along two main lines of inquiry. One has been demonstrating that while negative sprites may be rare, other transient luminous events (TLEs) like negative halos are not and indeed may be as common or more common than their positive counterparts [Bering et al., 2004; Frey et al., 2007; Williams et al., 2007, 2012; Newsome and Inan, 2010]. The other line of inquiry has demonstrated that negative cloud-to-ground (CG) lightning with large charge moment changes (CMCs) is much rarer than its positive-CG counterpart. Given the primacy of CMC in determining the likelihood of sprite production (typically hundreds of C km are needed to initiate sprites) [Huang et al., 1999; Hu et al., 2002; Cummer and Lyons, 2005; Lyons et al., 2009; Qin et al., 2012], it then follows that negative sprites should be much rarer since only a small number of negative CGs have the requisite characteristics for producing sprites [Cummer and Lyons, 2005; Williams et al., 2007; Cummer et al., 2013]. Moreover, the minimum CMC threshold for initiating negative sprites may be ~50% greater than that for positive sprites [Qin et al., 2012].
 Positive sprites are normally associated with positive CGs in the stratiform regions of mesoscale convective systems (MCSs) [Boccippio et al., 1995; Lyons, 1996, 2006; Lyons et al., 2003; Williams and Yair, 2006]. For example, Lyons [1996, 2006] has found that positive sprites tend to occur after a mesoscale system reaches 20,000–25,000 km2 in total area and contains both active convection as well as a large region of stratiform precipitation. The sprite-parent discharges either initiate within the convective region and propagate out into the stratiform region before coming to ground, or initiate in situ within the stratiform region before coming to ground [Lang et al., 2010]. The large CMCs produced by these CGs are primarily the result of discharging the laterally extensive regions of positive charge that exist within the stratiform regions of MCSs [e.g., Williams and Yair, 2006], although the altitude of the in-cloud flash component may play a role as well [Lyons et al., 2003; Lang et al., 2010, 2011]. Note that a few exceptions to the occurrence of positive sprites over the stratiform region of MCSs exist [e.g., Lyons et al., 2008]. However, even in these cases, a common motif is the transition of active convection toward a more stratiform precipitation structure, providing increased lateral areas of space charge.
 Based on the observations of positive sprite meteorology, one might expect that negative sprites, though rarer, would feature the same basic characteristics. Indeed, the initial observations of confirmed negative sprites suggested their occurrence within the stratiform regions of large MCSs, though the meteorology of these sprites was not particularly well documented [Barrington-Leigh et al., 1999, 2001; Taylor et al., 2008].
 However, there are reasons to doubt that this would be the most common phenomenology. Lu et al.  showed lightning mapping and other data for 85 negative CGs with high (> 200 C km) impulse charge moment change values (iCMC; i.e., the CMC produced within the first 2 ms of the return stroke). They also presented observations of four strokes that produced confirmed negative sprites. These sprites were observed over active convective regions. Lightning mapping data, which were available for one of the sprites, suggested that the parent discharge consisted of a hybrid intracloud (IC)-CG discharge, which also involved the upper positive charge layer in the normal polarity thunderstorm tripole, as opposed to just occurring between the midlevel negative and lower positive charge like typical negative CGs [Lu et al., 2012]. This hybrid IC/CG activity also was commonly seen in negative CGs that produced the largest iCMCs (~200 C km or more) in the Lu et al.  data set. Such hybrid IC/CGs would be more likely to produce sprites than typical negative CGs because of the greater charge moment changes. Note that because negative CGs do not typically feature long continuing currents, their iCMC values are very similar to their total (impulse plus continuing current) charge moment changes [Lu et al., 2012].
 Thus, a reasonable hypothesis for the meteorology of negative sprites—or of negative CGs with extremely large charge moment changes that very likely should produce sprites even if there is no camera to observe them—is that they mainly occur over convective regions of large mesoscale precipitation systems and that these convective regions contain normal polarity tripoles and thus produce predominantly negative CG lightning. The purpose of this paper is to test this hypothesis and thus shed additional light on the nature of precipitation systems that produce negative sprites, or powerful negative CGs that likely created a sprite, even though camera observations were not available.
2 Data and Methodology
 Radar and lightning data for all cases were analyzed using locally developed software programmed in the Interactive Data Language. In this section, discussion of each data source is divided along the different platforms providing those data.
2.1 Charge Moment Change Network
 The Charge Moment Change Network (CMCN) consists of two extremely low frequency sensors: one near Durham, NC, USA and one at Yucca Ridge near Fort Collins, CO, USA [Cummer et al., 2013]. These sensors provide real-time estimates of iCMC values for CG lightning strokes that occur within most of the contiguous United States. The National Lightning Detection Network (NLDN) is used to geolocate strokes with reported iCMC values and also provides peak current information, following the methodology of Cummer et al. .
 Since photographic or video confirmation of negative sprites is rare, it is attractive to exploit the known relationship between CMC and sprite occurrence in order to develop a larger data set. However, it is difficult to establish a fixed criterion for this purpose. Charge moment changes for past observations of negative sprite parents range from 450 to 1550 C km [Barrington-Leigh et al., 1999, 2001; Taylor et al., 2008; Soula et al., 2009; Lu et al., 2012; Li et al., 2012]. However, recent theoretical work has established that negative CGs with CMCs as low as 300 C km could produce sprites [Qin et al., 2012]. One would expect that the probability of sprite occurrence would increase as CMC increases, although mesospheric and ionospheric conditions should be important as well [Huang et al., 1999; Hu et al., 2002; Cummer and Lyons, 2005; Li et al., 2008; Lyons et al., 2009; Qin et al., 2012]. Thus, to ensure the highest probability of a negative sprite in the CMCN data, the CMC threshold should be set as large as possible. On the other hand, the value of utilizing the CMCN data set is to drastically increase the available number of cases, so setting the threshold too high will prevent that.
 A peak current (Ipk) threshold is difficult to set as well, though all past observations of confirmed negative sprite parents with the exception of one (Ipk ~ 50 kA) [Soula et al., 2009] featured Ipk values of 90 kA or greater [Barrington-Leigh et al., 1999, 2001; Taylor et al., 2008; Lu et al., 2012; Li et al., 2012]. In fact, Ipk is likely not a primary control on sprite occurrence [Lyons, 2006], although a low Ipk for a corresponding large iCMC value (which should be driven by a high Ipk since iCMC by definition does not include much continuing current) could be indicative of a potential data quality problem [Cummer et al., 2013]. Thus, a high Ipk threshold would be preferable to none at all, simply to minimize the chance of data problems.
 Thus, for the purposes of this study, negative CGs with both peak currents of 100 kA or greater and iCMC values of 800 C km or greater (<< 0.01% of all negative CGs with a computed iCMC in the data set) were considered “sprite-class” lightning, that is, lightning powerful enough to have likely produced a negative sprite. The 800 C km value is much larger than the theoretical minimum CMC [Qin et al., 2012] and is larger than that for many confirmed negative sprite parents [Taylor et al., 2008; Lu et al., 2012; Li et al., 2012], but is close to the median of past studies of negative sprites (−819 C km). Thus, to the extent that the limited past observations allow, one can be confident that negative CGs of this magnitude are very likely to produce sprites. For example, Lyons et al.  found that positive CGs with iCMC values greater than 300 C km were 75–80% likely to produce a sprite. This study's criterion, though for negative CGs, is nearly 3 times that number.
 The Ipk criterion (−100 kA) also matches the pattern of setting a threshold that is near the median observed in the literature for sprite parents (−126 kA) and that is high enough to avoid data quality issues, but not so high as to create a data set that is too small. Overall, the two criteria provided a reasonable number of cases for an initial analysis of negative sprite meteorology (38 sprite-class events which also featured available radar data).
 It is certainly possible that these fixed thresholds—though consistent with the literature—could exclude potential sprite-producing CGs or for that matter include powerful discharges that did not actually produce a sprite. Thus, to ensure that any meteorological inferences based on this data set are not grossly at odds with those for confirmed negative sprites, analysis of the meteorology of three negative sprite parents has been performed. The lightning data for these sprite parents were analyzed by Li et al. , and their iCMC values were computed in postprocessing—following the methodology of Cummer and Inan —that is more robust than the real-time CMCN computations. For these particular events, the postprocessing reduced iCMC estimates by approximately one third on average, compared to the real-time values. This is in line with the typical factor of 1.5 of random error that exists in real-time iCMC estimates [Cummer et al., 2013].
2.2 NMQ National Radar Mosaics
 NOAA has developed the National Mosaic and Multi-Sensor Quantitative Precipitation Estimation (NMQ) system in order to improve precipitation estimates and achieve other related goals [Zhang et al., 2011]. A major component of this system is the development of three-dimensional radar mosaics covering the entire contiguous United States. These mosaics, available starting in 2009, are developed from the nationwide network of S-band Doppler weather radars, following a methodology described by Zhang et al. . Radar reflectivity is provided every 5 min on a 0.01° latitude/longitude grid, with a vertical coordinate that ranges from 0.25 km MSL to 18 km (vertical spacing is 0.25 km near the surface gradually stretching to 2 km aloft). Due to file size limitations, the national mosaics are broken into eight different tiles. When necessary (e.g., a storm of interest straddled the border between two tiles), these tiles were stitched together in the analysis software before examination was performed.
 The radar characteristics of storms producing negative sprites or negative sprite-class lightning were examined. The time interval of the radar volume including the time of the sprite-class lightning, as well as the two leading and two following volumes (i.e., a 25 min period inclusive of the discharge in question), was included in this analysis. The vertical behavior of the 10 and 30 dBZ echo contours was examined within 15 km of the discharge in question in all volumes. These are common reflectivity thresholds used in a variety of studies of precipitation systems, and their behavior will be sufficient to diagnose gross characteristics of the storms of interest. In addition, two-dimensional composites of maximum reflectivity in the column were developed for the volume containing the sprite-class lightning, and the contiguous 30 dBZ areas of the parent storms were calculated from these. Except in limited circumstances, the 10 dBZ area was not examined because some of the precipitation systems were so large that it would have required stitching more than two NMQ tiles together to properly compute 10 dBZ statistics, a process that consumed excessive resources for comparatively little additional information. In those cases, the 30 dBZ contour was more than adequate for indicating the massive size of the precipitation systems.
 The performance and data characteristics of the NLDN are well known to the atmospheric electricity community [Cummins et al., 1998; Biagi et al., 2007; Cummins and Murphy, 2009] and will not be reviewed here. Flash-level data were analyzed to examine the CG characteristics of storms producing negative sprites or negative sprite-class lightning. CGs were examined within 15 km of the sprite-class lightning for the 25 min period inclusive of the discharge. Positive CGs were not considered if their peak currents did not exceed 10 kA [Cummins et al., 1998]. NLDN-detected IC discharges of either polarity were considered CGs if their peak currents exceeded 25 kA (R. Holle, 2009, http://www.srh.noaa.gov/media/abq/sswhm/Holle_sw_hydromet_09.pdf). IC lightning flashes were not examined in this study due to their low and potentially regionally dependent detection efficiency by NLDN [Cummins and Murphy, 2009].
 The North Alabama Lightning Mapping Array (NALMA) [Goodman et al., 2005] provides three-dimensional mapping of very high frequency (VHF) radiation from lightning. These data were used to examine the structure of one sprite-class discharge (on 26 March 2011) that occurred within range of the NALMA. All of the other sprite-class lightning examined in this study occurred too far from any of the various LMA networks that populate the United States, or the LMA data were already examined in other studies [Lu et al., 2012].
3.1 General Observations
 CMCN data were examined for all negative CG strokes that met the 800 C km and 100 kA sprite-class threshold over the period from August 2007 to July 2011. The results are plotted in Figure 1. A total of 92 discharges meeting these criteria were detected and geolocated. This number was less than 5% of the number of positive CGs (1875) meeting the same criteria. This supports the findings of other studies that large-CMC negative discharges are much rarer than their positive counterparts [Cummer and Lyons, 2005; Williams et al., 2007; Cummer et al., 2013]. Most of the discharges were loosely clustered around the lower Mississippi River valley and Gulf of Mexico, as well as the Gulf Stream off the eastern coast of the United States. These regions are roughly consistent with those featured in the stroke density maps for weaker (−200 and −600 C km thresholds) negative discharges shown by Cummer et al. , which bolsters confidence that the present study's data set can be considered an approximately representative sample of a negative CG data set with a more relaxed iCMC criterion, at least from the perspective of regional climatology. For every 100 C km that the iCMC threshold is relaxed, the available sprite-class data set for this same time period approximately doubles in size (203 strokes for 700 C km threshold, 519 strokes for 600 C km, 1271 strokes for 500 C km, 3254 strokes for 400 C km, and 8755 strokes for 300 C km).
 An interesting question to ask is whether the 92 sprite-class negative CGs commonly occurred with other large-iCMC strokes, positive or negative. Fifty-six of these sprite-class negatives occurred within 15 min and 100 km of at least one other powerful negative CG (threshold: iCMC < −300 C km, Ipk < −50 kA). The median number of companion powerful negatives was 1, and the maximum was 12, for each of the 92 sprite-class events. However, it was rare for powerful positive CGs to occur in concert with the sprite-class negatives. Only one sprite-class negative CG occurred within 15 min and 100 km of a positive CG with the same characteristics (iCMC > 800 C km, Ipk > 100 kA), and only 9 of 92 occurred within the same distance and time of positive CGs with relaxed criteria (iCMC > 300 C km, Ipk > 50 kA).
 Most (79) of the 92 sprite-class negative CGs occurred during the 00–12 UTC time period, which in the United States corresponds to evening, nighttime, and early morning hours. The mode of the diurnal distribution occurred at 07 UTC (roughly 02 local time in the central United States). The 92 strokes were distributed roughly evenly throughout March through November, averaging ~10 strokes per month during that period, but none occurred during December through February.
 From these 92 discharges, 38 that occurred during 2009–2011 (when NMQ radar mosaics were available) were selected for further analysis, based on their geographic proximity to radar coverage. This naturally excluded some discharges from this time period that occurred over the Atlantic Ocean, Mexico, the Caribbean Sea, and the Gulf of Mexico. This limits the applicability of this study's results to precipitation systems occurring mainly over land (although, as will be seen, some storms occurred in coastal regions), despite the fact that powerful negative CGs are disproportionately favored over salt water [Lyons et al., 1998]. An additional three discharges were added to this database, for a grand total of 41 negative CGs. The three additions were negative CGs that occurred in 2010 and 2011 and did not meet the iCMC criteria (values ranged from 450 to 710 C km), but did produce confirmed negative sprites as reported by Lu et al.  and Li et al. .
 The radar data for the storms producing these 41 discharges, which occurred over 23 different days, were examined in detail. A representative example of a negative sprite-class storm is shown in Figure 2. This storm was an intense, multicellular system that occurred over northeast Texas on 3 May 2009. The sprite-class discharge occurred within the convection, which itself was producing mostly negative CGs (62.5% of 104 CGs during the 25 min analysis period). The 30 dBZ contour extended to an average maximum height of 16 km MSL, and the areal coverage of contiguous 30 dBZ echo in this storm was nearly 5500 km2.
 Overall statistics for the 41 cases are shown in Table 1. The results suggest that most sprite-class/parent negative CGs occurred in or near large (at least MCS-scale), intense, multicellular convection that was mainly producing negative CGs. Indeed, all but three events occurred within 15 km of a convective core that featured 30 dBZ reaching to at least 10 km MSL. The remainder occurred within the stratiform regions of MCSs. Trends in 30 dBZ echo heights and volumes during this period (not shown) were mixed, with no clear trend toward increasing or decreasing intensity at the time of sprite-class stroke production. Given the behavior, the most general statement that can be made is that the precipitating structures near the sprite-class CG were mature. All but six events produced greater than 50% negative CG lightning during their respective 25 min analysis periods, although overall CG flash rates tended to be low (1–2 min −1 on average).
Table 1. Characteristics of the 41 Negative Sprite-Producing or Sprite-Class Precipitation Systems Observed in This Study
CG flash rate (min−1) within 15 km
% Negative CGs within 15 km
10 dBZ Max height (km MSL) within 15 km
30 dBZ Max height (km MSL) within 15 km
30 dBZ Contiguous area (km2)
6.73 × 103
1.99 × 105
Maximum reflectivity within 15 km (dBZ)
Maximum reflectivity in feature (dBZ)
 Figure 3 shows a distribution of contiguous 30 dBZ echo areas for all 41 events. Only seven were smaller than 2000 km2 (the smallest observed system was 616 km2), while a full third of the data set occurred in massive systems totaling 50,000 km2 or more in area. For context, the areas of those large systems were comparable to the commonly accepted definition of mesoscale convective complexes [Maddox, 1980]. However, the present study used a much more restrictive criterion, namely, contiguous 30 dBZ radar echo as opposed to satellite cold cloud structure. As another point of comparison, the prolific positive sprite-producing MCS on 20 June 2007, studied by Lang et al. , featured peak radar-defined convective areas near 50,000 km2. Also, recall that the typical precipitation system does not produce positive sprites until it reaches 20,000–25,000 km2 in total radar echo coverage (not just 30 dBZ, as in this study) [Lyons, 1996, 2006]. A common motif in the data set was the presence of large, quasi-steady (i.e., not rapidly evolving) precipitation systems aligned along fronts and other boundaries.
3.2 Systems That Produced Several Sprite-Class Negative CGs
 The typical pattern was for a storm system to produce only one to three negative sprite-class discharges during its lifetime. As noted before, these generally were produced near mature, deep, and intense convective cores. However, there were a couple of notable storms that produced more than three sprite-class strokes. On 5 October 2010, a relatively small linear MCS along the Gulf coast produced seven sprite-class negative CGs over a ~6.5 h period (Figure 4). The 30 dBZ composite areas of the convective elements producing these sprite-class strokes ranged between 1500 and 3000 km2. However, a less restrictive threshold, such as a 10 dBZ contiguous area, found total storm areas ranging from ~20,000 km2 to ~35,000 km2, in line with Lyons [1996, 2006]. The MCS was slow moving, remaining mostly over east Texas and Louisiana for the majority of this time, and was aligned roughly parallel to the coast, likely along a sea-breeze front (Figure 4). Though it struck ground within a weak echo, the sprite-class negative in Figure 4a occurred within 15 km of a deep convective core and thus is not considered a stratiform-produced stroke.
 The other example was a large mesoscale system that occurred on 9 March 2011 and produced seven sprite-class negative CGs over a ~16 h period (Figure 5). When it first started producing sprite-class negatives, the system consisted of a large number of distinct multicellular convective clusters (Figures 5a–5c). The sprite-class negatives were mostly split between separate, mature clusters at a few different times during this day. As the system moved eastward over the Gulf coast states—ahead of an eastward moving cold front—it gradually organized into a classic leading-line, trailing-stratiform MCS. Distinct from the other events on this day, the last sprite-class negative CG occurred in the stratiform region of the MCS, unfortunately too far from the NALMA for useful analysis (Figure 5d). This stratiform region was intense, however, with 30 dBZ reaching a maximum altitude of 4.5 km on average, and the maximum reflectivity within 15 km of the stroke was 47.5 dBZ.
3.3 Lightning Mapping Analysis
 One negative sprite-class storm, on 26 March 2011, was close enough to the NALMA for useful analysis. The discharge in question (peak current −145 kA, iCMC −974 C km) occurred near a large cluster of convective cells around 1043 UTC (contiguous 30 dBZ area of the sprite-class storm was 15,330 km2), and the LMA showed initiation and termination of the discharge near the convection (Figure 6). Similar to what Lu et al.  found for large-iCMC negative CGs in general, the flash appeared to be a hybrid IC/CG discharge, which started between the midlevel negative and upper-level positive charge regions and then came to ground late in the flash's lifetime. This behavior (upper-level intracloud activity preceding an eventual negative leader to ground) is qualitatively similar to what has been noted for so-called “bolt-from-the-blue” (BFB) flashes [Krehbiel et al., 2008; Lu et al., 2012], suggesting that BFBs may disproportionately feature large CMCs and therefore greater halo and/or sprite potential than normal negative CGs. Note that this flash occurred ~120 km from the NALMA network center, so three-dimensional mapping and source detection was not optimal, preventing good resolution of the midlevel negative and low-level positive charge regions. This storm produced 68.3% negative CG lightning flashes (out of 41 total CGs), and its 30 dBZ echo reached 13 km MSL on average during the 25 min analysis period encompassing the sprite-class discharge. The NALMA indicated a decreasing trend in total lightning flash rate in the cell that produced the sprite-class stroke, with a flash rate of 10 min−1 at 1039 UTC declining to only 3 min−1 (including the sprite-class one) during 1043 UTC.
3.4 Confirmed Negative Sprite-Producing Systems
 Three of the events in this study were observed to produce confirmed negative sprites. These occurred on 9 September 2010, 29 July 2011, and 25 August 2011. The detailed analysis of the sprite-parent flashes can be found in Lu et al.  and Li et al.  and will not be reviewed here. Two of the events (9 September 2010 and 25 August 2011) occurred within the convective regions of multicellular systems (Figure 7), similar to the overall results with other sprite-class lightning in this study, while the third (29 July 2011) occurred over the stratiform region of an irregularly shaped MCS (Figure 8). The 9 September 2010 storm included the remnants of Tropical Storm Hermine and 40–45 min later produced two gigantic jets [Meyer et al., 2013]. The contiguous 30 dBZ areas for the convective negative sprite events were 52,181 km2 and 3072 km2, respectively. For the 25 min analysis periods encompassing the sprite parents, the first storm (9 September 2010) featured 30 dBZ reaching 12.8 km MSL on average and produced 67% negative CG lightning (out of 24 CGs total), while the second (25 August 2011) featured 30 dBZ reaching 14.6 km MSL on average and produced 100% negative CG lightning (19 CGs). By contrast, the stratiform event had 30 dBZ reaching 6.7 km MSL on average, and there were only two CGs total (100% negative). Nevertheless, the 30 dBZ contiguous area (which included both stratiform and convective echo) was just under 20,000 km2, and the maximum reflectivity within 15 km of the sprite-parent stroke was 45.5 dBZ.
 Overall, the characteristics of the confirmed negative sprite-producing precipitation systems were similar to those observed for the sprite-class storms. The parent strokes occurred either within intense convection contained within large mesoscale systems, or within a well-developed stratiform region, and these storm areas produced low rates (~1 min−1 or less) of mainly negative CG lightning.
4 Discussion and Conclusions
 This study featured the use of several numerical thresholds, such as 10 and 30 dBZ echo heights and/or areas, as well as 15 km ranges. These thresholds, though arbitrary, were informed by common thresholds for convective studies used by the scientific community. For example, as reviewed by Lang and Rutledge , the 30 dBZ contour is commonly used in radar studies of storm electrification. Moreover, the results of Lu et al.  suggest that 15 km is an approximate minimum horizontal distance that the in-cloud components of high-iCMC negative flashes can travel between initiation and ground strike. Regardless, the focus of this study was on gaining a qualitative sense of the meteorology of negative sprite-class lightning, and sensitivity studies (not shown) demonstrated that adjusting these thresholds did not affect its main conclusions.
 These conclusions are the following:
 Negative sprite-parent and negative sprite-class lightning mainly occurred in or near intense, deep convection.
 The systems producing these discharges were frequently large—at least MCS scale, thousands if not tens of thousands of square kilometers in size. This is in basic agreement with past studies of negative sprite-producing storms [Barrington-Leigh et al., 1999, 2001; Taylor et al., 2008; Soula et al., 2009]. Large, long-lived, and quasi-steady systems organized by fronts and other boundaries could produce multiple sprite-class strokes over several hours.
 The systems producing these discharges also produced mainly negative as opposed to positive CG lightning near the locations of the sprite-class discharges, although overall CG flash rates were low (1–2 min −1, on average).
 This latter conclusion suggests that negative sprites would tend to occur over normal polarity thunderstorms, albeit ones that are very large in area to be able to support the large observed charge moment changes. There were notable exceptions to all these observations, for example, the three systems that produced stratiform sprite-class or sprite-parent lightning. In these cases, the stratiform regions also were mature and intense, with maximum reflectivities near the sprite-class/parent strokes ranging from 39.5 to 47.5 dBZ and maximum heights of the 30 dBZ contour ranging from 4.5 to 6.7 km MSL.
 The general pattern that was observed for negative sprite parents or sprite-class lightning suggests an interesting contrast to the meteorology of positive sprite-parent lightning. Positive sprite parents normally occur within the stratiform regions of MCSs [Boccippio et al., 1995; Lyons, 1996, 2006; Lyons et al., 2003; Williams and Yair, 2006; Lyons et al., 2009], with some notable exceptions [e.g., Lyons et al., 2008]. Thus, while the sizes of precipitation systems producing negative or positive sprite-parent/class lightning appear to be similar, there is a distinct difference in where the opposite polarity discharges typically occur, one that mirrors the typical pattern of negative and positive CG occurrence in MCSs [e.g., Rutledge and MacGorman, 1988; Rutledge et al., 1990].
 However, this does not necessarily mean that the same precipitation system would produce negative and positive sprites concurrently, due to the observed lack of high-iCMC positive and negative CG strokes occurring close in space and time to one another. This is consistent with the observed regional offset in the United States between the occurrence of powerful positive strokes (favored mainly over the northern Great Plains) and powerful negative ones (favored mainly over the southeast and nearby oceanic regions) [Lyons et al., 1998; Williams et al., 2005; Cummer et al., 2013]. Why this offset exists, and to what extent it impacts individual storms, is a topic of future research. Of particular interest is the fact that negative and positive sprites have been observed in close proximity to one another in other regions of the world [Barrington-Leigh et al., 1999; Taylor et al., 2008; Soula et al., 2009].
 The limited LMA analysis in this study, coupled with the detailed analysis of negative high-iCMC flashes and a negative sprite parent by Lu et al. , suggests that in-cloud components of the negative lightning remain mostly near convection and consist of a hybrid-IC/negative-CG discharge that involves the upper positive charge region during the flash—characteristics that are qualitatively similar to BFB discharges. This is an interesting point of comparison between high-iCMC negative lightning and positive sprite parents. A typical positive parent often initiates in convection and travels into the stratiform region along a downward sloping upper pathway (which can reach as low as the melting level far into the stratiform region) before coming to ground [Lang et al., 2010, 2011], while negative high-iCMC strokes may involve the upper positive charge only within the convection proper. In this hypothesized model, lower positive charge in the normal tripole structure of the convective region helps to eventually guide the upper-level intracloud activity to ground as a negative CG. However, if the upper-level lightning travels into the stratiform region, then only a sunken positive dipole may exist, and thus, the preferred path to ground is as a positive CG.
 The common involvement of the upper positive charge in sprite-parent or sprite-class lightning, regardless of the polarity of the final CG (and sprite), is a notable result coming from a synthesis of this study with the results of others [Lang et al., 2010, 2011; Lu et al., 2012]. If large-iCMC positive and negative strokes (e.g., sprite parents) compete for available upper positive charge in convection, then that may help explain their observed lack of coincident and collocated occurrence in the United States.
 It is notable that negative CG flash rates were typically low near the sprite-class negative strokes and that for the one analyzed LMA case (26 March 2011), total flash rates in the parent cell were low and declining. This is consistent with the relatively low CG rates observed during the time of the negative sprite reported by Soula et al. . These low flash rates may allow the buildup of large amounts of negative charge that can be subsequently neutralized by a single stroke, leading to a large peak current and iCMC (T. Chronis et al., New evidence on the diurnal variation of peak current in global CG lightning, submitted to Bulletin of the American Meteorological Society, 2013). The large contiguous regions of deep convection in many of the observed sprite-class systems also should assist with copious charge production over a broad area.
 A limitation of the present work is that it lacks confirmation of sprite occurrence for most of the examined cases. It also could not focus on the radar structure of sprite-class storms over the Gulf Stream and other oceanic regions. Moreover, it is well known that negative halos are much more common than negative sprites [Bering et al., 2004; Frey et al., 2007; Williams et al., 2007, 2012; Newsome and Inan, 2010]. This study was not focused on these particular TLEs, but they are related to sprites and can occur over negative CGs with similar CMCs to the thresholds used in this study [Williams et al., 2012].
 In addition to addressing the above concerns, future work should focus on continued searching for negative sprites with high-speed video camera observations as well as adding more cases of negative sprite-class lightning from upcoming and past convective seasons. The results of the present study suggest that video observations over the convective regions of large, quasi-steady precipitation systems—particularly over the southern and southeastern United States—may be fruitful in this regard. Over the long term, a detailed database of the characteristics of storms producing this lightning can be developed, and statistically based analysis of these systems can be performed. The development of the national radar mosaics, along with the proliferation of optical and radio-frequency observing networks for sprites and their parent lightning, will greatly facilitate future research into sprite meteorology.
 Vaisala supplied the NLDN flash-level data analyzed in this study and also enables the CMCN by providing geolocation of high-iCMC strokes via a real-time NLDN stroke-level data feed. These data were absolutely critical to this study, and the authors offer their sincerest thanks to Vaisala for providing them. Katherine Willingham provided the NMQ radar mosaic data on behalf of NOAA. Bill McCaul of Universities Space Research Association provided the NALMA data. Paul Hein of Colorado State University assisted with the data analysis. The authors are extremely grateful to all of these people and their agencies for their gracious help in facilitating this research. The authors also thank the journal editor and reviewers for their assistance with publishing this study. This research was funded by the Defense Advanced Research Projects Agency under the Nimbus program, as well as the National Science Foundation under grant AGS-1010G6S7.