Variations in Cloud-to-Ground Lightning Characteristics among Three Adjacent Tornadic Supercell Storms over the Tennessee Valley Region

The magazine publisher is the copyright holder of this article and it is reproduced with permission. Further reproduction of this article in violation of the copyright is prohibited. To contact the publisher: http://www.ametsoc.org/AMS/

ABSTRACT
The contrasting behavior of cloud-to-ground (CG) lightning associated with three adjacent supercell thunderstorms observed on 18 May 1995 is examined. Thunderstorm characteristics and anvil interactions are related to north-south variations in CG lightning properties. While tornadic activity was not consistently related to variations in CG properties, radar reflectivity factor area greater than 65 dBZ was generally inversely related to CG frequency. It is hypothesized that suppression of CG activity was produced by reduction of large number concentrations of precipitation-sized particles (i.e., presence of large hail) in the particle interaction mixed phase region. It is further hypothesized that seeding from upstream storm anvil ice was associated with nearly coincident enhancement of CG activity in downstream storms. Likewise, it is hypothesized that the reduction in 65 dBZ echo area (inferred suppression of hail) is related to this inferred anvil seeding process.

1. INTRODUCTION
    A large number of published studies have investigated the relationship among cloud-to-ground (CG) lightning, supercell storm type(FN1)--low precipitation (LP), classic (CL), and high precipitation (HP)--tornadoes, and large hail. A subset of these have explored relations between CG frequency variations (reductions) and tornado development (Doswell and Burgess 1993; MacGorman 1993; Perez et al. 1997; Bluestein and MacGorman 1998). Considerable attention has been devoted to positive CG (CG+) behavior in tornadic supercell storms (MacGorman and Nielsen 1991; Branick and Doswell 1992; Curran and Rust 1992; MacGorman and Burgess 1994; Stolzenburg 1994) and rapid changes in dominant CG polarity during supercell LP-CL and CL-HP transitions (Branick and Doswell 1992; Seimon 1993). These investigations have indicated appreciable variability in tornado-CG relationships (MacGorman and Burgess 1994; Bluestein and MacGorman 1998), which is not surprising since lightning precursor charge separation processes remain enigmatic, and depend on a combination of storm dynamical and microphysical processes (e.g., Bringi et al. 1997). Intense thunderstorms, including tornadic supercell storms, may exhibit remarkably variable CG rates ranging from very high (e.g., MacGorman 1993; Perez et al. 1997) to extremely low (e.g., Lang et al. 2000; McCaul et al. 2002). Moreover, thunderstorms that exhibit high average rates display considerable temporal variability (MacGorman 1993; Perez et al. 1997; Bluestein and MacGorman 1998). Such variability is perplexing and has placed limits on using CG as a diagnostic tool for storm intensity, or as a predictor for tornado occurrence.
    This paper describes an analysis of radar reflectivity and cloud-to-ground lightning behavior of three tornadic supercell storms, arranged as a sequence of storms oriented NE to SW, approximately parallel to mid- to upper-level flow. Each storm exhibited similar radar reflectivity structures, and produced one F3-F4 tornado in addition to weak tornadoes of F0-F1 intensity. While the storms were separated by distances (60-100 km) large enough to generally be considered isolated, they appeared to be influenced by anvils of storms located toward their upstream direction.
    The primary objective of this paper is to document systematic variations in CG behavior among these three supercell storms. Despite similarities in storm echo structures and lifetimes (7-9 h), the northernmost storm produced the most lightning (3894 flashes) and exhibited the largest fraction (43%) of positive CG (CG+) flashes. The southernmost storm generated significantly lower CG flashes (1259), and a much lower fraction (20%) of CG+. The storm located between the northern and southern storms exhibited intermediate flash totals (2732) and CG+ fraction (34%). For short time periods, the northern and intermediate storms were dominated by CG+, but the southern storm did not exhibit this behavior.
    One interesting finding of this study--and one not apparently reported previously in the literature--is an association between the increased CG rates and the merger of supercell convective cores with upstream anvils. Such an increase was observed for all three storms. We explore the hypothesis that seeding from upwind anvils may alter the microphysical structure of supercell storm updraft cores by increasing the number concentrations of graupel embryos, leading to increases in CG (and total) lightning. A secondary objective is to examine the relationship between tornadoes and CG lightning within these storms and to compare with studies in the Midwest and Great Plains regions.

2. CASE OVERVIEW

A. ENVIRONMENT
    Synoptic conditions at 0000 UTC 19 May 1995 are shown in Fig. 1. A cyclone with modest central surface pressure (1000 hPa) was centered over the Kentucky-Indiana border. A cold front extended from south Illinois into east Texas. The three storms considered in this study were located 250-300 km ahead of this front within a region of relatively strong surface pressure gradient over northeast Mississippi, northwest Alabama, and central Tennessee. Warm air advection was negligible over this region, but differential positive vorticity advection appeared to provide lifting on the synoptic scale. A jet axis (about 40 m s[sup-1] within the 200-300-mb layer) was located over the warm sector over and just to the NW of the study region.
    Two thermodynamic soundings are presented in Fig. 2a. The thin lines represent the 1200 UTC sounding from Redstone Arsenal (RSA, located in northern Alabama near Huntsville; see Fig. 5b for location). This sounding was modified by local deep convection and is thus not representative of the atmosphere 11 h later. The 0000 UTC sounding from Nashville, Tennessee (BNA) represents the sounding closest in space and time to the storms considered in this paper. The boundary layer has been modified with observed surface conditions near the storms. This sounding is also not entirely representative of the storm environment, as the dry upper levels portrayed around the storms in the Geostationary Operational Environmental Satellite (GOES) water vapor image at 2000 UTC (Fig. 3) are not reflected in this sounding. However, the available information indicates virtually no convective inhibition at low levels.
    The wind hodographs in Fig. 2b were derived from the Okolona, Mississippi (OKO) 404-MHz National Oceanic and Atmospheric Administration (NOAA) wind profiler at 2000 and 2200 UTC. (The location of OKO is shown in Fig. 5b.) The southernmost storm was about 50 km north of the profiler at 2100 UTC. Low-level winds exhibit variability over short timescales, and a significant increase in upper-tropospheric wind speed occurred between 2000 and 2200 UTC. Missing winds above 8.5 km at 2000 UTC and 10.5 km at 2200 UTC suggest the presence of very dry air above 6-8 km. This is corroborated by the high brightness temperature, signifying dry air, displayed in the series of 6.7-µm channel water vapor GOES images in Fig. 3. The hodograph exhibits a highly sheared linear structure up to 2 km at 2000 UTC, and a more circular shape up to 4 km at 2200 UTC. Values of CAPE, bulk Richardson number, and surface to 3 km storm-relative helicity, estimated at 2850 J kg[sup-1], 25, and 400 m[sup2] s[sup-2], respectively, indicate an unstable atmosphere favorable for supercell storm development and sustenance (Weisman and Klemp 1984).

B. SEVERE WEATHER EVENTS
    The 18 May 1995 outbreak produced 62 reported tornadoes, 165 damaging wind events, and 120 hail events from Illinois to Alabama (Fig. 4). The three supercell storms considered in this study produced most of the severe weather over northern Alabama and south, central, and east Tennessee. Storm core tracks are defined in the three-panel cloud-to-ground lightning plot in Fig. 5. Hereafter, these storms are identified (from north to south) as A, B, and C. Storm A produced six tornadoes across Tennessee, including an F3, and hail up to 70 mm in diameter. Storm B generated six tornadoes across Tennessee, the strongest an F4, and hail also up to 70 mm in diameter. Storm C produced two tornadoes in north Alabama, one an F4, and hail up to 95 mm in diameter.

3. RADAR AND LIGHTNING ANALYSIS

A. OVERVIEW
    General radar observations were obtained from 2-km-resolution Weather Surveillance Radar-1988 Doppler (WSR-88D) composite reflectivity imagery and single Doppler data from the Huntsville, Alabama (HSV) WSR-74C; and the Knoxville, Tennessee Tri-Cities (MRX) and Birmingham, Alabama (BMX) WSR-88D radars. A time series of low elevation (0.5°-1°) scan reflectivity factor images from the HSV and MRX radars (Fig. 6) reveals the relative locations of three long-lived isolated supercell storms identified in the CG plot of Fig. 5. A summary of general storm chronology is given in Table 1.
    Storm B developed more slowly than A or C, and became well formed after 2000 UTC. Storms A-C traveled from 260° at about 20 m s[sup-1] along parallel paths spaced 60-100 km apart. Thus, these storms remained as separate entities throughout most of their lifetime. Development into right-moving, cyclonic supercell storms was demonstrated by reflectivity structure (Fig. 6), mesocyclones depicted in radial velocity from BMX and MRX, and gradual rightward turns of 15° over a 1-h period. At 0015 UTC 19 May, the northeast (NE) left-moving component of an earlier storm split over central Alabama (annotated in Fig. 6d) clipped the eastern low-level inflow zone of storm C just NE of the Alabama-Tennessee-Georgia border, eventually dissipating it. This storm also intersected the SE side of storm B at 0115 UTC in east-central Tennessee. These collisions resulted in the demise of the supercellular stages of both storms B and C. During this period storm A, located in NE Tennessee, also weakened. The top panels of Figs. 7a-c depict WSR-88D radar parameters of storms A, B, and C, including: (a) echo tops of 12-15 km, (b) vertically integrated liquid (VIL) generally exceeding 60 and approaching 80 kg m[sup-2], (c) reflectivity factor (Z) maxima of about 70 dBZ, (d) a minimum GOES IR temperature of about -70°C, and (e) temporal variability in these parameters. Owing to large distances between the storms and WSR-88D radars, these observations are limited by crude sampling resolution, and hence have some level of uncertainty.

B. SUPERCELL FEATURES
    Large radar-storm distances limited measurements of the 3D structure of storms A, B, and C. The HSV radar was closest to storms B and C, but only 0.5° plan position indicator (PPI) scans were available for analysis. Distinct weak/bounded weak echo regions (WER/BWER), hook echoes, and "V-notch" patterns (Lemon 1980) in Z, all signatures of classic supercell storms, were observed within storms A, B, and C (Fig. 6). All three storms exhibited primarily classic supercell structures according to the radar definitions of Doswell and Burgess (1993) and Rasmussen and Straka (1998). The sequence of Z patterns in Fig. 6 indicates that storm B was likely a CL supercell throughout its lifetime. Values of Z within the hook echo of storm C increased significantly between 2300 and 2310 UTC, becoming comparable to that of the precipitation core (observed with HSV radar data). A distinct transition time from one classification to another is difficult to define since a continuum is observed. Based on Z alone, we surmise that storm C transformed from the CL to HP part of the (continuous) supercell spectrum after about 2300 UTC.
    Storm A appeared to evolve towards the HP end of the supercell spectrum in the HSV radar imagery after about 2200 UTC, but the long distance limits quantification and timing of this transition. As A approached the MRX radar at 0018 UTC, a low-level hook echo of 45-50-dBZ intensity was evident south to southwest of the mesocyclone. Much stronger Z values filled the main precipitation core, indicating a CL supercell. Values of Z within the hook increased to 50-60 dBZ 6 min later, suggestive of a transition towards the HP supercell end of spectrum. In this study, storm A is classified as a CL supercell before 0020 UTC, and an HP storm (but perhaps towards the CL continuum) after this time. Such transformation from CL to HP supercell structures is often observed during the latter mature storm stage (e.g., Bluestein and Woodall 1990; Rasmussen and Straka 1998).

C. LIGHTNING BEHAVIOR

1) CLOUD-TO-GROUND LIGHTNING
    Locations of CG lightning from storms A, B, and C are presented in Fig. 5. The CG strikes are further decomposed into time series of positive (CG+) and negative (CG-) flashes per 5-min time interval in the bottom panels of Figs. 7a-c. CG lightning statistics are summarized in Table 2.
    Significant temporal variations in CG activity were associated with each storm, which is typical of tornadic supercell storms documented in the literature (e.g., MacGorman and Nielsen 1991; MacGorman and Burgess 1994; Bluestein and MacGorman 1998). Storm A exhibited three major CG episodes centered on 2055 [130 strikes(5 min)[sup-1]], 2220 [>100 strikes(5 min)[sup-1]], and 0020 UTC [45 strikes(5 min)[sup-1]]. The first two episodes were dominated by CG+. Storm B exhibited multiple short-lived peaks, the most significant occurring near 2145 [120 strikes(5 min)[sup-1]]. Storm C exhibited one major CG episode late in its life cycle near 2320 [80 strikes(5 min)[sup-1]] after a prolonged period of variably moderate CG- activity. This 2320 event was dominated by CG-, and was closely followed by a significant burst of CG+. Storms A and B were dominated by CG- late in their life cycles after transforming to HP supercells.
    A systematic variation in total CG lightning flashes and polarity from north to south was evident among these storms (Table 2). Storm lightning duration diminished from north to south, with a 130-min difference between storms A and C. In view of similar echo lifetimes, this suggests that the northern (downshear) storms were more efficient lightning producers. Total CG, CG-, and CG+ flash counts, average flash rates (CG+ and CG-), CG+ fraction, and CG+ dominant time fractions all decreased systematically from north (storm A) to south (storm C). Only maximum CG- flash rates did not display this pattern: CG+ rates did decrease from north to south but not as uniformly. Storms A and B were fairly similar in this regard and in CG- activity, but storm C had a greater maximum CG- rate than storms A and B. Essentially, more lightning activity occurred to the north, and the greater fraction was CG+ lightning.
    With regard to lightning activity and supercell storm classification, the supercell storms in our case exhibited a variation in CG+ activity similar to CL supercells that are in some cases CG- dominant and in other cases CG+ dominant (MacGorman and Nielsen 1991; Seimon 1993). Storm A displayed a greater fraction of CG+, while storm C was on the other end of the spectrum with 85% fewer CG+ strikes.
    During their transition period from CL to HP, CG- rates within A and C increased significantly at 0020 and 2300 UTC, respectively. The preponderance of CG- lightning from HP storms is consistent with other documented observations (Branick and Doswell 1992; MacGorman and Burgess 1994). The basis of HP transition used here was comparable Z magnitudes of the hook echo and precipitation core. This behavior is similar to that of the Plainfield, Illinois, tornadic storm, in which precipitation was visually observed to wrap around the mesocyclone as CG+ to CG- domination occurred (MacGorman and Burgess 1994). An increase in CG-, and consistent CG- dominance was also observed in the lightning history of storm B after 0040 UTC, so it is possible that the same CL to HP transition (or progression towards the HP end of the spectrum) occurred for this storm. However, this storm was too far away from both radars this time to resolve low-level hook echo features.

2) TOTAL AND INTRACLOUD LIGHTNING
    The National Aeronautics and Space Administration (NASA) spaceborne Optical Transient Detector (OTD; Boccippio et al. 2000) recorded total lightning (CG and intracloud) for a 4-min period (0106:40-0110:38 UTC) near the end of storms A and B's lifetimes. Storm C was nearly dissipated at this time. Storm A was a weakening HP supercell, 70 min after the last tornado and 50 min before the last recorded CG. In contrast, CG- flash rates from storm B increased during this period. Storm B produced an F0 tornado about 15 min before the overpass, and another F1 tornado developed about 50 min after the OTD observations. The OTD flash-rate estimates were 21 and 63 min[sup-1] (accounting for the 64% detection efficiency determined by Boccippio et al. 2000) for storms A and B. The National Lightning Detection Network (NLDN) detected about 2 min[sup-1] (8 CG-, 1 CG+) for A and 4 min[sup-1] (15 CG-) for B during the same 4-min interval. Assuming an NLDN detection efficiency of 90% (Cummins et al. 1998), the intracloud to CG lightning ratio is 7.4 for storm A and 14.1 for the more vigorous storm B. Both numbers exceed the average continental U.S. intracloud to CG ratio of about 2.8 determined by Boccippio et al. (2001). The value for storm B is within the range of high ratios of 10-30 observed within tornadic and severe/intense storms (Goodman et al. 1999; Buechler et al. 2000; Lang et al. 2000).

D. RELATIONSHIPS BETWEEN CG LIGHTNING AND TORNADIC ACTIVITY
    CG lightning time series associated with the strongest tornado of each storm exhibited dissimilar patterns relative to tornado formation time. This behavior is consistent with strong tornadoes examined by Perez et al. (1997). The F3 tornado produced by storm A began about 1 h after the second peak in CG+ during a relatively inactive lightning period [CG- and CG+ rates of 23 and 9 (5 min)[sup-1], respectively]. The CG- and CG+ rates continued to decrease to minimum values of 3 and 4 (5 min)[sup-1], respectively, near the time of tornado dissipation 20 min later. Hail 70 mm in diameter was reported near the time of tornado dissipation at 2340 UTC. A local peak in CG flash rates within about 10 min before tornado formation was not evident for this storm, although a local minimum CG flash rate nearly coincided with tornado formation.
    The F4 tornado associated with storm B formed about 5 min after the first significant CG period, consisting of intense CG+ and CG- rates of 53 and 41 (5 min)[sup-1]. This is in agreement with the results of MacGorman and Burgess (1994) and Perez et al. (1997). The tornado continued through a minimum in CG activity and dissipated during a modest rise in CG-. A local flash-rate minimum did not occur within 10 min of tornado formation.
    The F4 tornado produced by storm C displayed a contrasting behavior. This tornado formed during a small relative peak in CG- rates and a minor peak in CG+ rates of 27 and 5 (5 min)[sup-1], respectively, but again a local flash-rate minimum did not occur near the time of tornado formation. During the 1-h lifetime of this tornado, CG activity of both polarities remained nearly constant until a marked increase to 71 and 25 (5 min)[sup-1] near the time of tornado dissipation.
    Inspection of Figs. 7a-c reveals no systematic relation between weak (F0-F1) tornadoes and CG patterns. Weak tornadoes were not consistently correlated with peaks in CG+ or CG-.
    The observations of strongest tornadic activity described here are consistent with other observations. While two of the three F3-F4 tornadoes formed just after a local peak in CG flash rates (a strong signal for storm B and a weak signal for storm C), storm A displayed a local minimum of CG flash rates coincident with tornado formation. This is in general agreement with Perez et al. (1997) and also with MacGorman (1993), who determined that relationships between relative maxima or declines in CG flash rates and tornadogenesis were apparent, but not consistent. Perez et al. (1997) did not include tornadoes weaker than F4, and recommended this for further studies. In our case, consistent relationships between CG lightning rates and weak (F0-F1) tornadoes were not evident, in contrast to Buechler et al. (1996) who found that tornadogenesis in four southeastern United States storms was often heralded by a decrease in CG-.

E. CONSTANT RANGE-HEIGHT REFLECTIVITY ANALYSIS
    Due to their ENE path and considerable distance (190-300 km) from the BMX radar, the three storms remained at a nearly constant range from the radar for a period of time long enough to analyze Z at nearly constant range (height). During this period the storm echo volume was confined to the lowest three beam elevation angles. At the lowest two beam elevations (0.5° and 1.45°) the estimated area of Z > 65 dBZ (A[sub65]) fluctuated significantly;(FN2) at times no range gates exceeded 65 dBZ. The area of each radar sample volume was calculated using Z sample spacing of 1000 m and the calculated transverse beam dimension. The number of pixels for each storm at consecutive volume scans and one beam elevation angle were summed for three Z bin intervals (45-55, 55-65, and >65 dBZ) to compare with CG flash rates. Pixels in higher bins were not added to those of lower bins, for example, the area of Z between 45-55 dBZ does not include pixels with values >55 dBZ. The results are presented in Figs. 8a-c.
    Table 3 defines the range, physical beam dimension, height span (assuming standard propagation) and corresponding environmental temperatures determined from the thermodynamic sounding in Fig. 2. Fig. 9 provides a schematic showing the beam dimension and height relative to the storm. In the following, we assume that pixels with Z > 65 dBZ signify the presence of large hail.
    Radar sampling at the heights defined in Table 3 and Fig. 9 includes the temperature regime where ice particle interactions in the presence of supercooled water are hypothesized to produce electrification (MacGorman et al. 1989; Saunders et al. 1991). Solomon and Baker (1998) coin this as the charging zone (CZ), a term that we adopt henceforth. In theory the CZ occurs at temperatures colder than -10°C, with the upper bound depending on updraft strength, that is, the height to which the updraft can carry graupel and generate supercooled cloud water. Significant concentrations of graupel likely correspond to a radar reflectivity factor of 40 to 55 dBZ (Doviak and Zrnic 1993). MRX radar measurements of storm A at 0030 UTC, 65-km range, and 11.9° elevation indicated that 50-dBZ radar reflectivity factor extended to a height of about 13.5 km (approximately -65°C). This supports the assumption that the temperature ranges in Table 3 are within the storm CZ. The following analysis of A[sub65] investigates the relationship between lightning and an estimate of hail within the CZ of each storm.
    Figure 8a suggests an inverse relationship between A[sub65] and CG lightning for storm A over the 1940-2100 UTC period. Table 4 indicates a relatively large negative value of -0.74 for the correlation coefficient between A[sub65] and the sum of CG+ and CG-. Significantly smaller (negative) correlation coefficients were computed for the 45-55 and 55-65 dBZ intervals. Increasing CG rates after the start of the constant height observation period at 1930 UTC, to the peak in both CG- and CG+ rates at 2100 UTC, correspond to decreasing A[sub65] over this time. When CG rates decreased from this peak to a relative minimum, A[sub65] increased to a minor secondary peak near 2040 UTC. In addition, smaller CG rate minima and maxima superimposed on this large-scale variation appear to correspond to relative maxima and minima in A[sub65]. Because relative variations in CG- and CG+ flash rates are similar during this period it is difficult to determine whether this relationship is more valid for CG- or CG+ lightning activity, although the correlation coefficients exhibit greater magnitudes for CG+.
    Radar observations of storm B were at a center-beam height of about 1 km less than those of storm A. Much smaller values of A[sub65] are observed in this height range. Signal attenuation (from storm C) and partial beam blockage by an intervening ridge within this azimuth sector are contributing factors to lower A[sub65]. The range of Z values within storm C that intersect radials to the core of storm B are as follows: 30-45 dBZ between 2130 and 2215 UTC, 45-55 dBZ between 2215 and 2300 UTC, 55-65 dBZ between 2300 and 2330 UTC, and 50-61 dBZ between 2330 and 2350 UTC. Nonetheless, the observations again suggest an inverse relationship between A[sub65] and CG lightning, though not as distinct as for storm A. Values of the correlation co-efficient are primarily negative and greatest in magnitude for CG- and 45-55 and 55-65 dBZ echo areas. Coefficients for the >65 dBZ row were not calculated due to insufficient data.
    Storm C displays low values of correlation coefficient for all three Z intervals in Table 4. From 2130 UTC (the start of the observation period) to 2220 UTC, CG rates do not correlate with A[sub65] as in storms A or B, most notably the minimum in CG- rates occurring at the same time as the relative minimum in A[sub65] at about 2208 UTC. After this time, a very weak negative correlation between A[sub65] and total CG is indicated, but is not significant.
    The results of this analysis suggest a relationship between large Z values (inferred to represent large hail) in the CZ and a suppression of CG lightning activity. Storm A displayed a significant negative correlation between A[sub65] and total CG for a 100-min period, but storm C did not. These results are consistent with the observations of Lang et al. (2000) and are discussed further in section 6.

4. THE ROLE OF ANVIL INTERACTIONS IN CG BEHAVIOR

A. WERE UPPER-LEVEL WINDS A FACTOR?
    One intriguing finding in this study is the systematic decrease in total CG (and CG+) activity, from north to south, within the three storms exhibiting similar kinematic structures. One hypothesis for CG+ generation is the "tilted dipole hypothesis" (Brook et al. 1982) in which the combination of midlevel precipitation tilting and displacement of upper positive charge in the anvil (Fig. 9) by strong upper-level winds exposes upper positive charge directly to the surface, resulting in CG+ strikes. If the positively charged anvil area is increased by upper-level wind shear, then it follows that CG+ strikes will be greater for storms with more extensive downshear anvils. Branick and Doswell (1992) have also suggested such a behavior. The north-south gradient in upper-level winds in the study region was not well sampled due to missing upper-tropospheric wind data at BNA above 450 hPa at 1200 UTC and above 700 hPa at 0000 UTC (19 May). The OKO profiler data (Fig. 2b) show 40-50 m s[sup-1] wind speeds at anvil level, but this sample was taken south of all three storms.
    GOES visible images were analyzed for the 1902-2045 UTC period to compare downshear anvil movement of storms A and C. At 1902 UTC, the anvil of storm A was well developed (1.6 h after first echo), while the anvils of B and C were small (0.25 and 0.5 h after first echo). By 2045 UTC the anvils of all three storms had developed significantly (Fig. 11a). Anvil-level winds were calculated by measuring the downshear anvil displacement between 1902 and 2045 UTC. Downshear anvil displacement rates of 27 and 25 m s[sup-1] (ground relative) were calculated for storms A and C, respectively. This small difference is less than the errors involved in this measurement, which assumes that the divergent component of the wind is small when, in reality, it is inversely related to the distance from the storm.
    Given that the estimated upper-level winds near storms A and C are similar, one could argue that the greater CG+ activity in storm A up to 2045 UTC was related to greater anvil divergence prior to this time. However, the CG+ activity within storm B around 2045 UTC (during its early growth stage) was greater than that of storm A. This fact implies that another mechanism produced the greater CG+ rates to the north.

B. THE ROLE OF ANVIL INTERACTIONS
    GOES visible (VIS) and IR imagery indicate that the storms initially displayed separate anvils with clear gaps between them, but as anvil ice was transported downstream from storms B and C, it intersected the cores of downstream storms A and B. Anvil interactions affecting storm A are considered first. The anvil of storm B initially impinged on the updraft core of storm A near 2000 UTC. Figures 3a,b indicate that this intersection occurred between 1945 and 2015 UTC. At 2002 UTC, the brightness temperature of the anvil near storm A was 235 K (-38°C) corresponding to an environmental height of about 9.7 km (assuming a short optical depth), as determined from thermodynamic sounding data. At this time, Z values of 10-15 dBZ below the anvil top indicate that precipitating ice extended from storm B and another storm to the north toward storm A. Figures 7a and 10a indicate that the beginning of this anvil impingement corresponds to the initial increase in CG+ rates, representing the first CG+ dominant period associated with storm A. The time lag between anvil arrival and maximum CG rates is about 55 min, but the time uncertainty of anvil intersection (and inferred ingestion of ice into the CZ) is estimated at +/-30 min.
    To further portray this process, Fig. 11 displays GOES visible imagery at 2045 UTC (45 min after anvil arrival), and a corresponding WSR-88D image from BMX. GOES IR imagery (not shown) indicates a relatively dense anvil near storm A, originating from anvils of storm B and C. The Z values of 10 dBZ indicated in the 1.4° PPI image at 2048 UTC (Fig. 11b), with a center-beam height near the convective core of about 11.6 km, shows precipitation extending downstream, in the vicinity of storm A, originating from storms B and C. The radar beam is likely sampling above the maximum reflectivity within the anvil at greater range near storm A. We infer the presence of pristine ice ahead of this (as inferred from Z, below the noise level of ˜0 dBZ at this range, and the presence of visible anvil indicated in the GOES VIS image of Fig. 11a) surrounding the updraft core of A. This pattern is also evident on the 0.5° PPI scan (not shown), corresponding to a beam-center height of approximately 7.0 km. Hence, by 2045 UTC, significant anvil ice surrounded storm A, and both CG- and CG+ rates attained maximum values about 10 min later (Fig. 10a).
    Likewise, the anvil from storm C encroached the convective core of storm B on the low-level inflow (SE) flank by 1945 UTC, 1 h after first echo and 15 min after initial CG. The CG+ dominant period within B occurred 2 h later near 2145 UTC, during a time when it was clearly surrounded by anvil material from storm C. This behavior is consistent with the observations of storm A, but the response from seeding is obscured by potentially rapid storm evolution early in the life cycle of storm B (i.e., nonsteady conditions).
    A similar behavior was observed for storm C. The anvil from an upstream supercell storm approached the convective core of storm C near 2115 UTC (Fig. 3c). This encroachment corresponds to a modest increase in CG- around this time. However, BMX radar data indicate that anvil precipitation (at the 7.4-km level) from this storm did not fully engulf the core of C (particularly the southern flank) until about 2330 UTC. This complete merger occurred near the time of the F4 tornado spawned by storm C. A scenario consistent with that of storm A resulted. Storm C had exhibited very low CG+ activity throughout its lifetime, but near the time of the anvil merger, both CG- and CG+ rates increased significantly (Figs. 7c and 10b). There are some differences in the patterns of storms A and C. The CG- and CG+ rates of storm C did not peak simultaneously, the peak in CG- significantly exceeded the peak in CG+ (in contrast to the behavior exhibited by storms B and C), and the response time was faster in storm C (but within the bounds of uncertainty in timing of anvil impingement).
    Thus, CG enhancements were associated with anvil interaction for all three storms, but the polarity is not consistent (CG+ in two cases and CG- in the other case). Nonetheless, we hypothesize that seeding from anvils of upshear storms is related to enhancement of CG activity for each of the storms studied here. This finding is discussed in more detail next.

5. DISCUSSION
    Two consistent patterns have been observed in this case study:
    1) There appears to be an inverse relation between A[sub65] (the area exceeding 65 dBZ, i.e., the inferred presence of large hail) at storm middle levels and CG activity.
    2) Merger of anvil cloud originating from upwind (upshear) storms with convective cores of downshear storms was associated with enhanced CG lightning, especially CG+.
    We assume that these observations are physically related according to the following hypothesis: Seeding of a storm's middle to upper regions by anvil ice originating from upstream storms increase the number of graupel embryos within strong updrafts, and hence produces increased CG lightning activity (i.e., rate of charge separation).
    This hypothesis requires (a) the existence of an abundant concentration of ice particles within the encroaching anvil, and (b) that these particles are entrained into the updraft of the downshear storm. A conceptual model of this process is presented in Fig. 12. Rasmussen and Straka (1998) have recently (and independently) formulated a similar hypothesis to explain differences in precipitation distributions of isolated LP, CL, and HP supercell storms. They also allude to the possible influence of anvil seeding that we are considering here. The net result is a modification and conceivable increase in precipitation (graupel and small hail) concentration and precipitation efficiency. A corollary to this hypothesis is that anvil seeding increases the precipitation efficiency of the seeded storm, that is, precipitation efficiency and lightning rate are related.
    Two key physical parameters associated with this hypothesis have not been measured in this case. These parameters include (a) the profile of convergence (e.g., entrainment) associated with the updraft, and (b) the vertical distribution of ice particle concentration and size within the seeding anvil. The following sections expand on these two key issues, utilizing more detailed measurements from previous comprehensive studies of supercell storms.

A. UPDRAFT MASS FLUX PROFILES
    A number of studies have examined supercell updraft properties using wind fields derived from multiple-Doppler radar analyses. We define dynamic entrainment as the inflow associated with an accelerating updraft required to maintain mass continuity. The vertical profile of updraft mass flux is therefore a key parameter here. The updraft mass flux, [integral] rhow dA, where rho is air density and w is vertical motion, is related to mass convergence by the anelastic continuity equation
    [df]rhow / [df]z = -[universal quantifier] • rhoV[subh],
    where V[subh] is the horizontal wind vector. Mass convergence therefore exists up to the level of maximum updraft mass flux (where [df]rhow/[df]z = 0, and not necessarily the level of maximum updraft), that is, the level of zero mass divergence. The available observations in Table 5 for supercell/thunderstorm updrafts reveal that this level typically is close to H/2 (where H is storm top height), but the associated updraft typically peaks above this level, near 0.6H to 0.7H.
    In order for anvil ice to systematically enter the updraft, the anvil base must extend below about H/2, which is about 7 km AGL in our case. It is also possible for ice to enter via less systematic, discrete turbulent entrainment processes at higher levels, and for storm-scale circulations (i.e., cloud edge downdrafts) to transport anvil ice to lower convergent levels.
    Comprehensive measurements of a supercell hailstorm observed during the 1981 Cooperative Convective Precipitation Experiment (CCOPE) are used here and in the next section to illustrate flows and precipitation characteristics associated with the convective core and anvil. For brevity we will term this storm 8/1 since it occurred on 1 August 1981. Miller et al. (1990) documented details of the airflow with multiple-Doppler analyses, which revealed a maximum updraft of ˜45 m s[sup-1] at 10 km AGL. Figure 13, taken from that paper, clearly demonstrates a prominent inflow from the upshear direction within the 5-8-km layer, below the updraft maximum analyzed at 10 km. (Incidentally, the level of maximum updraft mass flux for this storm was 2 km below the level of maximum draft speed; see Knupp 1985). Using a Lagrangian precipitation trajectory model, Miller et al., (1990) identified this inflow region as one of the important embryo source regions for hailstones in this storm. Thus, this observation supports the premise that entrainment of anvil from the upshear direction can readily occur at midlevels.

B. CHARACTERISTICS OF ICE PARTICLES AND ICE PROFILES IN SUPERCELL ANVILS
    Radar measurements of Z in our case indicate that significant ice was present within the anvil far downstream. In Fig. 11b a plume of anvil precipitation, centered at 7-8 km AGL and exceeding 24 dBZ, extends from the core of storm C to a point 100 km downstream. The downstream extension of this anvil (Z > 10 dBZ) is located just upstream (˜20 km) from the core of storm A. GOES visible imagery in Fig. 11a indicates small pristine ice anvil material (as inferred from the lack of significant radar reflectivity) surrounded the core of storm A.
    To our knowledge the only quantitative measurements of isolated supercell anvil characteristics are those obtained in the same 1 August Montana supercell storm alluded to earlier (Heymsfield 1986). The anvil in the 1 August storm extended > 100 km downstream due to strong storm-relative winds of 20 m s[sup-1] within the anvil, similar to the storm-relative anvil flow in our study (Fig. 2b). The peak-measured Z within the anvil of the 1 August storm was 20-25 dBZ at 8-9 km AGL, quite similar to the anvil reflectivity in our case. In situ particle measurements acquired at the 8.0- and 9.3-km AGL aircraft flight levels revealed average ice water contents of 0.3-0.6 g kg[sup-1] and peak/average ice concentrations of > 100 L[sup-1] and 20-40 L[sup-1], respectively, in four of the penetrations within this anvil. Ice particle dimensions exceeding 6 mm (major axis) were measured, and many of the particles greater than 2 mm in size were ice aggregates.
    In view of the similarities in the environmental and storm quantities, it is reasonable to assume similar anvil characteristics in our case. The anvil base is dependent on precipitation trajectories (e.g., the trajectory slope is the ratio of particle fall speed to storm-relative wind speed) and relative humidity (Heymsfield 1986), the latter influencing ice sublimation rates and hence ice survival time/distance. In our case the slope is 1:20, assuming a terminal fall speed of 1 m s[sup-1] and anvil-relative flow speed of 20 m s[sup-1]. Thus, over a 60-100-km path, anvil ice would descend 3-5 km from point of origin. Thus, sufficient ice could have been ingested into the updraft below the 7-8-km level.

C. SYNTHESIS
    Radar data indicate that the anvils of storms A, B, and C were of sufficient depth and contained sufficient numbers of large ice particles (given the observed Z) necessary for significant microphysical modification of downstream updraft cores. The link between time of anvil impingement and the time of enhanced CG from the downstream "seeded" storm (Fig. 10) supports our hypothesis. Moreover, the existing (albeit limited) knowledge on supercell storm inflow and anvil microphysical properties also supports this premise. Although we do not have sufficient data to prove or reject our hypothesis, the limited evidence supports its plausibility. The concept of seeding hailstorms to reduce hail size via increased numbers of smaller particles was a primary hypothesis of the National Hail Research Experiment (Knight et al. 1982). Storms producing large hail typically are embryo-starved, and hence contain large bounded weak echo regions (e.g., Miller et al. 1982; Frank and Foote 1982; Knight 1984; Miller et al. 1988). Browning and Foote (1976) introduced this idea in their examination of an intense Colorado hailstorm (exhibiting a significant BWER) in which embryo injection was inferred to be minimal. Storm C also contained a large BWER, based on RHI observations near 2240 UTC from a television-owned radar 30 km away. In situ measurements within a strong supercell BWER have been made only once (Musil et al. 1986), and these were measurements within a marginal BWER that was surrounded by appreciable Z (see also Miller et al. 1988). This BWER exhibited a maximum updraft of ˜50 m s[sup-1], an adiabatic core and very low concentrations of precipitation-sized ice particles. Hailstones to softball size were produced.
    Anvil seeding is likely most effective in cases where an isolated storm is not an efficient "self-seeder". When the low-level relative inflow forms a large (˜180°) angle with the anvil-level relative flow, precipitation recycling (self-seeding) is likely, that is, a storm can readily ingest its own anvil debris. Browning (1977) introduced this concept and inferred that precipitation recycling was not effective for angles of 90° (as in our case) or less. Knupp and Cotton (1982) applied this premise to explain the absence of a WER within a high-precipitation, left-moving supercell storm, which was also a prolific producer of CG lighting (based on visual observations of the first author). The angle was 140° in their case.
    Self-seeding is also the basis of the hypothesis advanced by Rasmussen and Straka (1998), who determined that the storm-relative flow at anvil level was significantly different among LP, CL, and HP storms. They surmise that isolated HP storms ingest their own anvil ice more effectively, while LP storms are not efficient self-seeders.
    Lyons et al. (1998) have shown that unusually high CG+ fraction produced by storms of various types over the Great Plains region was associated with extensive smoke plumes transported from widespread fires in Mexico in 1998. They concluded that the smoke contained ice nuclei, which altered the storm microphysics (enhanced ice nucleation) and associated charge distribution. This observation is consistent with our anvil-seeding hypothesis, since existing ice is the most effective means of nucleating ice from supercooled water. As an aside, we note that Battan (1967) examined CG behavior (visually) of thunderstorms seeded with AgI, but found no statistically significant CG frequency difference between seeded and nonseeded convective clouds over the Santa Catalina Mountains in SE Arizona.
    Charging theory also supports the hypothesis. The rate of charge separation is proportional to the particle collision rate (ice crystal, graupel, supercooled water) within the mixed phase region (Takahashi 1978; Jayaratne et al. 1983; Saunders 1993). Seeding could nucleate a large mass of supercooled water provided that the anvil ice is entrained into the storm. Numerical simulations described by Solomon and Baker (1998), who reported a 10%-20% increase in lightning when the Hallett-Mossop ice multiplication mechanism was included in their simplified numerical model, substantiate this. Aircraft observations reported by Musil et al. (1986) show very low concentrations (<10[sup-3] L[sup-1]) of particles >1 mm within a supercell storm BWER. This compares with much larger measured concentrations (>1 mm) of about 1 L[sup-1] or greater measured in the 8/1 storm anvil (Heymsfield 1986). Thus, significantly increased particle collision rates appear to be plausible from anvil seeding provided that (a) the anvil contains sufficient concentrations of large ice, (b) a large fraction of the ice is entrained by the updraft into the CZ, and (c) the updraft is embryo-starved, that is, the storm is not an efficient self-seeder.
    The effects of anvil seeding on hail concentration and size cannot be addressed here. Limited evidence shows that storm A experienced a decrease in A[sub65] after anvil impingement at 2002 UTC. Storm C was merged with the anvil from an upshear storm near 2300 UTC. This merger was associated with a decrease in A[sub65], followed by an increase during the interaction. Storm B was shrouded by the anvil from storm C before the time of the reflectivity analysis in section 5b, but A[sub65] was consistently less than that of storms A and C. These patterns support the hypothesis that anvil seeding decreases maximum hail size, but these observations are only circumstantial.
    The mechanism(s) favoring the increase in CG+ rates over CG- rates associated with anvil seeding in storms A and B is unknown. In fact, storm C displayed increased CG- rates when the implied anvil seeding occurred. This contrasting behavior may be related to the CL-HP transition inferred at this time, or to changes in storm kinematics or local environment. More definitive answers to these questions require more detailed observations of storm kinematics and microphysics, such as multiple-Doppler radar, polarimetric radar, and aircraft data to monitor the coevolving kinematic and microphysical fields associated with such interactions (e.g., Bringi et al. 1997).
    Preliminary research utilizing the lightning detection and ranging (LDAR) network centered on Cape Kennedy, Florida, has indicated that the relationship between storm attributes and total lightning (CG plus intracloud) displays greater consistency (e.g., Goodman et al. 1999; Williams et al. 1999). Such observations of total lightning would provide clarification of this seeding process.

6. CONCLUSIONS
    This study has examined the CG lightning and supercell storm characteristics of three adjacent tornadic supercell storms (A, B, C), spaced by 60-100 km and aligned NE (storm A) to SW (storm C). While all storms assumed classic supercell (CL) structures, storms A and C transformed to high-precipitation (HP) supercells during later stages. Progressively greater CG rates and a greater fraction of CG+ were observed from SW (upshear) to NE (downshear).
    The primary purpose of this study was to investigate the systematic variation in CG lightning through analysis of radar and CG lighting attributes, with a particular focus on relationships among CG lightning, inferred anvil seeding, and large hail. The basic findings are as follows:
    1) Despite differences in CG lightning behavior, the storms exhibited similar kinematic structures and similar severe storm attributes (i.e., each produced strong tornadoes and large hail).
    2) The CG lightning activity was consistent with known supercell storm precipitation classification and transition between classification types.
    3) Some consistencies between CG lightning activity and strongest tornadic activity were determined among the storms, but no consistent relationship was found for weaker tornadoes.
    4) Impingement of anvil echo (e.g., large ice particles) from upshear storms was temporally related to increases in CG activity, particularly CG+.
    5) The area of Z > 65 dBZ at middle levels (assumed to be an indicator of large hail) was mainly inversely related to CG activity for two of three storms.
    Items 2 and 3 are consistent with previous observations (Branick and Doswell 1992; Curran and Rust 1992; MacGorman and Burgess 1994). Our finding that the formation of strong tornadoes (in two of three cases) followed peak CG flash agrees with Perez et al. (1997), who determined that 75% of the storms with F4 or stronger tornadoes displayed this relationship. Only one of the storms examined here displayed a local minimum of CG flash rates coincident with tornado touchdown, again in general agreement with Perez et al. who found a similar behavior for 50% of tornadic storms studied.
    Items 4 and 5 appear to exhibit an inverse dependence. The presence of large hail (Z > 65 dBZ) displays a modest inverse relation with CG lightning, while the merger of anvil with downshear convective cores is associated with increased CG lightning activity. The two relations may be related: Entrainment of anvil ice by an embryo-starved updraft would provide increased concentrations of precipitation particles necessary for increased rates of charge separation at lower levels. The anvil interactions are especially intriguing, and support the hypothesis that seeding of downwind storms increase charge separation via increased graupel concentration (suppresses large hail) resulting in increased CG lightning activity. This hypothesis requires a comprehensive physical validation, including documentation of the microphysical properties of the anvil (e.g., Heymsfield 1986), the characteristics of the entrainment process, and the subsequent growth of entrained ice within the storm updrafts.
    Future observations of total lighting, utilizing special lightning mapping arrays (e.g., Rison et al. 1999; Goodman et al. 1999) and satellite measurements (Buechler et al. 2000), will assist in clarifications of relationships among storm attributes and lightning behavior. We speculate that total lightning will exhibit a more significant response to anvil seeding than CG lightning.
ADDED MATERIAL
    KEVIN R. KNUPP AND SIMON PAECH
    University of Alabama in Huntsville, Huntsville, Alabama
    STEVEN GOODMAN
    NASA Marshall Space Flight Center, Huntsville, Alabama
    Corresponding author address: Dr. Kevin R. Knupp, Atmospheric Science Department, University of Alabama in Huntsville, Earth System Science Center, Huntsville, AL 35899.
    E-mail: kevin@nsstc.uah.edu
    Acknowledgments. This research was funded by the Institute for Global Change Research and Education, NASA Contract NCC8-22, and by the National Science Foundation under Grant ATM-9704547. We thank Dennis Buechler for assistance in analysis of CG data.
    TABLE 1. Summary of composite radar reflectivity observations for storms A, B, and C.

                        Rightward
         First echo       turn(FN*)  Collision    Dissipation    Duration
Storm      (UTC)         (UTC)        (UTC)         (UTC)         (h)
  A         1725          1930          --           0200         8.5
  B         1845          2115        0115           0300         8.25
  C         1830          2045          --           0300         8.5

FOOTNOTE
    TABLE 2. Cloud-to-ground lightning statistics for storms A, B, and C.

                                      Storm A       Storm B      Storm C
Lightning duration (min)                 480           430          350
Total CG (flashes)                      3894          2732         1259
Fraction CG+ (%  )                        43            34           20
CG-/CG+ (flashes)                  2200/1694      1811/921     1003/256
Mean CG-/CG+ rate [(5 min)[sup-1]]      23/17         21/11         14/4
Max. CG-/CG+ rate [(5 min)[sup-1]]      56/81         53/73        71/25
CG+ dominant time (%  )                   23            15           10

    TABLE 3. Radar and environmental temperature details for constant height-range reflectivity analysis.

                               Lateral
                                beam
           Beam        Avg     dimen-     Beam height
         elevation    range     sion         span          Environmental
Storm     (°)       (km)      (km)         (km)        temperature (°C)
  A        0.5         283      4.7         5.2-9.5        -6 to -35
  B        0.5         261      4.3         4.5-8.5        -2 to -27
  C        1.45        195      3.2         5.5-8.7        -8 to -30

    TABLE 4. Correlation coefficient matrix for radar reflectivity area and CG frequency for storms A, B, and C.

                 CG+           CG-       Total CG
              Storm A (1941-2058 UTC)
  >65 dbZ       -0.75          -0.62       -0.74
55-65 dbZ       -0.23          -0.18       -0.22
45-55 dbZ       -0.39          -0.10       -0.32
              Storm B (2110-2221 UTC)
  >65 dbZ          --             --          --
55-65 dbZ        0.07          -0.41       -0.14
45-55 dbZ       -0.15          -0.51       -0.32
              Storm C (2227-2334 UTC)
  >65 dbZ       +0.25          -0.23       -0.08
55-65 dbZ       -0.18          +0.32       +0.18
45-55 dbZ       -0.26          -0.14       -0.21

    TABLE 5. Thunderstorm updraft properties determined from multiple-Doppler analyses.

                                   Maximum w                            Maximum {Begin Greek}S r{End Greek}w
        Study                Value (m s[sup-1])/scale height        Value (kg s[sup-1])/scale height
Knupp and Cotton 1982           30 m s[sup-1]/0.6H                     6 X 10[sup8] kg s[sup-1]/0.5H
Miller et al. 1982             >40 m s[sup-1]/0.5H                   3.6 X 10[sup8] kg s[sup-1]/0.55H
Frank and Foote 1982         ˜35 m s[sup-1]/0.6-7.0H               3.3 X 10[sup8] kg s[sup-1]/0.5H
Knupp and Cotton 1987           20 m s[sup-1]/0.5H                     4 X 10[sup8] kg s[sup-1]/0.45H
Miller et al. 1988             >40 m s[sup-1]/0.55H                   >4 X 10[sup9] kg s[sup-1]/0.45H
Miller et al. 1990              47 m s[sup-1]/0.7H                     6 X 10[sup9] kg s[sup-1]/0.5H(FN*)

FOOTNOTE
FIG. 1. Synoptic analysis, 0000 UTC 19 May 1995, of (a) mean sea level pressure (solid lines, 4-hPa interval) and 1000-500-mb thickness (dashed lines, 60-m interval); and (b) 500-mb heights (solid, 60-m interval), absolute vorticity (long dashed and shaded), and temperature (short dashed, 5°C interval).
FIG. 2. (a) Thermodynamic soundings for Redstone Arsenal, AL (RSA) at 1200 UTC 18 May 1995 (thin lines) and Nashville, TN (BNA) at 0000 UTC 19 May 1995. The boundary layer of the BNA sounding has been modified with observed surface conditions (gray lines). An ascent path of a surface-based parcel (theta[sube] = 350 K) represents the greatest CAPE for this sounding. (b) Wind hodographs obtained from the 404-MHz NOAA wind profiler at Okolona, MS (location shown in Fig. 5b) for 2000 UTC (black) and 2200 UTC (gray). Data were absent above 8.5 and 10.5 km due to dry upper-tropospheric conditions, as depicted in Fig. 3. The hodograph labels represent height in km AGL and the asterisk denotes storm motion.
FIG. 3. GOES 6.7-µm water vapor images depicting the encroachment of anvil from storm B on the upshear side of storm A between 1945 and 2015 UTC. By 2115 UTC the convective cores of storms A and B are surrounded by anvils from storms B and C, respectively.
FIG. 4. (a) Reported wind and hail events for the 24-h period of 18-19 May 1995. Crosses show wind events and other symbols represent hail events of varying size as defined in the key. (b) Reported tornado events for the time period in Fig. 4a. Tornado events are shown as triangles, and paths as lines. Tornadoes associated with storms A, B, and C are labeled accordingly with number labels indicating F-scale values.
FIG. 5. Total CG lightning generated from storms (top) A, (middle) B, and (bottom) C. Negative (CG-) and positive (CG+) flashes are denoted by dots and plus signs, respectively. Locations of the Okolona, MS (OKO) 404-MHz profiler, and the soundings from RSA and BNA are shown in Fig. 5b.
FIG. 6. Plan position indicator (PPI) plots of reflectivity factor (Z) from (a)-(c) the Huntsyille, AL (HSV) WSR-74C radar and (d) the Morristown, TN (MRX) WSR-88D. Range rings are drawn every 50 km for all panels.
FIG. 7. Time series of radar-derived parameters, GOES IR minimum brightness temperature, and 5-min counts of CG lightning flashes for storms (a) A, (b) B, and (c) C. Dashed lines refer to left-hand scale and CG- lightning. Times and F-scales of tornadoes, and transition from level-III data from the BNA WSR-88D to level-II data from the Birmingham, AL (BMX) WSR-88D are indicated.
FIG. 8. Cross-sectional areas of (top) reflectivity factor (Z) at various thresholds and (bottom) CG-/CG+ lightning rates [(5 min)[sup-1]] for storms (a) A, (b) B, and (c) C. Vertical lines represent time intervals for which correlation coefficients between the two time series were calculated (Table 4).
FIG. 9. Idealized conceptual model of a supercell storm showing typical charge regions (after Stolzenburg et al. 1998), temperature regimes, and estimated height intervals sampled by the BMX WSR-88D radar for storms A, B, and C.
FIG. 10. Relation between time of anvil arrival (t = 0) and subsequent CG frequency for storms (a) A and (b) C. The anvil arrival time corresponds to the intersection of an upstream anvil with a downstream convective core, as determined from GOES visible images.
FIG. 11. (a) GOES visible satellite image at 2045 UTC, and (b) 1.4° PPI from the BMX WSR-88D radar at 2048 UTC. The reflectivity grayscale in dBZ is defined at the bottom of (b).
FIG. 12. Conceptual model of general storm properties and interaction via ice seeding of a downshear storm from the anvil of an upshear storm. Reflectivity factor contours are drawn at 5, 20, 35, 45, 55, and 65 dBZ. Bold arrows, separated by the horizontal gray line, denote the heights of updraft inflow and outflow.
FIG. 13. Storm-relative airflow retrieved from a multiple-Doppler radar analysis within the updraft core of a Montana supercell storm on 1 Aug 1981. This section is oriented from SW (upshear) to NE (downshear), approximately along the environmental wind shear vector. Inflow is apparent up to 9 km AGL in this plane. Reflectivity factor contours are drawn every 10 dBZ, beginning at 5 dBZ. Updraft greater than 10 m s[sup-1] is stippled. Adapted from Miller et al. (1990).

FOOTNOTES

REFERENCES
    Battan, L. J., 1967: Cloud seeding and cloud-to-ground lightning. J. Appl. Meteor., 6, 102-104.
    Bluestein, H. B., and G. R. Woodall, 1990: Doppler-radar analysis of a low-precipitation severe storm. Mon. Wea. Rev., 118, 1640-1664.
    Bluestein, H. B., and D. R. MacGorman, 1998: Evolution of cloud-to-ground lightning characteristics and storm structure in the Spearman, Texas, tornadic supercells of 31 May 1990. Mon. Wea. Rev., 126, 1451-1467.
    Boccippio, D. J., and Coauthors, 2000: The Optical Transient Detector (OTD): Instrument characteristics and cross-sensor validation. J. Oceanic Atmos. Technol., 17, 441-458.
    Boccippio, D. J., K. L. Cummins, H. J. Christian, and S. J. Goodman, 2001: Combined satellite- and surface-based estimation of the intracloud-cloud-to-ground lightning ratio over the continental United States. Mon. Wea. Rev., 129, 108-121.
    Branick, M. L., and C. A. Doswell III, 1992: An observation of the relationship between supercell structure and lightning ground-strike polarity. Wea. Forecasting, 7, 143-149.
    Bringi, V. N., K. R. Knupp, A. Detwiler, L. Liu, I. J. Caylor, and R. A. Black, 1997: Evolution of a Florida thunderstorm during the Convection and Precipitation/Electrification Experiment: The case of 9 August 1991. Mon. Wea. Rev., 125, 2131-2160.
    Brook, M., M. Makano, P. Krehbiel, and T. Takeuti, 1982: The electrical structure of the Hoikuriku winter thunderstorms. J. Geophys. Res., 87, 1207-1215.
    Browning, K. A., 1977: The structure and mechanism of hailstorms. Hail: A Review of Hail Science and Hail Suppression, Meteor. Monogr., No. 38, Amer. Meteor. Soc., 1-43.
    Browning, K. A., and G. B. Foote, 1976: Airflow and hail growth in supercell storms and some implications for hail suppression. Quart. J. Roy. Meteor. Soc., 102, 499-533.
    Buechler, D. E., S. J. Goodman, E. W. McCaul, and K. Knupp, 1996: Cloud-to-cloud lightning activity within tornadic supercells in the Tennessee Valley. Preprints, 18th Conf. on Severe Local Storms, San Francisco, CA, Amer. Meteor. Soc., 499-503.
    Buechler, D. E., K. T. Driscoll, S. J. Goodman, and H. J. Christian, 2000: Lightning activity within a tornadic thunderstorm observed by the Optical Transient Detector (OTD). Geophys. Res. Lett., 27, 2253-2256.
    Cummins, K. L., M. J. Murphy, E. A. Bardo, W. L. Hiscox, R. B. Pyle, and A. E. Pifer, 1998: A combined TOA/MDF technology upgrade of the U.S. National Lightning Detection Network. J. Geophys. Res., 103, 9035-9044.
    Curran, E. B., and W. D. Rust, 1992: Positive ground flashes produced by low-precipitation thunderstorms in Oklahoma on 26 April 1984. Mon. Wea. Rev., 120, 544-553.
    Doswell, C. A., III, and D. W. Burgess, 1993: Tornadoes and tornadic storms: A review of conceptual models. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 161-172.
    Doviak, R. J., and D. S. Zrnic, 1993: Doppler Radar and Weather Observations. Academic Press, 562 pp.
    Frank, H. W., and G. B. Foote, 1982: The 22 July case study: Storm airflow, updraft structure, and mass flux from triple Doppler radar measurements. Hailstorms of the Central High Plains, Vol. 2, C. Knight and P. Squire, Eds., University Press of Colorado, 131-162.
    Goodman, S., D. Buechler, S. Hodanish, D. Sharp, E. Williams, R. Boldi, A. Matlin, and M. Weber, 1999: Total lightning activity associated with tornadic storms. Preprints, 11th Int. Conf. on Atmospheric Electricity, Guntersville, AL, NASA, NASA CP/-1999-209261, 515-518.
    Heymsfield, A. J., 1986: Ice particle evolution in the anvil of a severe thunderstorm during CCOPE. J. Atmos. Sci., 43, 2463-2478.
    Jayaratne, E. R., C. P. R. Saunders, and J. Hallet, 1983: Laboratory studies of the charging of soft-hail during ice crystal interactions. Quart. J. Roy. Meteor. Soc., 109, 609-630.
    Knight, C. A., 1984: Radar and other observations of two vaulted storms in northeastern Colorado. J. Atmos. Sci., 41, 258-271.
    Knight, C. A., P. Smith, and C. Wade, 1982: Storm types and some radar reflectivity characteristics. Hailstorms of the Central High Plains, Vol. 1, National Hail Research Experiment, C. A. Knight and P. Squires, Eds., University Press of Colorado, 81-94.
    Knupp, K. R., 1985: Precipitating convective cloud downdraft structure--A synthesis of observations and modeling. Ph.D. dissertation, Dept. of Atmospheric Science, Colorado State University, Fort Collins, CO, 296 pp.
    Knupp, K. R., and W. R. Cotton, 1982: An intense, quasi-steady thunderstorm over mountainous terrain. Part II: Doppler radar observations of the storm morphological structure. J. Atmos. Sci., 39, 343-358.
    Knupp, K. R., and W. R. Cotton, 1987: Internal structure of a small mesoscale convective system. Mon. Wea. Rev., 115, 629-645.
    Lang, T. J., S. A. Rutledge, J. E. Dye, M. Venticinque, P. Laroche, and E. Defer, 2000: Anomalously low negative cloud-to-ground lightning flash rates in intense convective storms observed during STERAO-A. Mon. Wea. Rev., 128, 160-173.
    Lemon, L. R., 1980: Severe thunderstorm radar identification techniques and warning criteria: A preliminary report. NOAA Tech. Memo. NWS NSSFC-1, 60 pp. [NTIS PB273049.]
    Lyons, W. A., T. E. Nelson, E. R. Williams, J. A. Cramer, and T. R. Turner, 1998: Enhanced positive cloud-to-ground lightning in thunderstorms ingesting smoke from fires. Science, 282, 77-80.
    MacGorman, D. R., 1993: Lightning in tornadic storms: A review. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 173-182.
    MacGorman, D. R., and K. E. Nielsen, 1991: Cloud-to-ground lightning in a tornadic storm on 8 May 1986. Mon. Wea. Rev., 119, 1557-1574.
    MacGorman, D. R., and D. R. Burgess, 1994: Positive cloud-to-cloud lightning in tornadic storms and hailstorms. Mon. Wea. Rev., 122, 1671-1697.
    MacGorman, D. R., D. W. Burgess, V. Mazur, W. D. Rust, W. L. Taylor, and B. C. Johnson, 1989: Lightning rates relative to tornadic storm evolution on 22 May 1981. J. Atmos. Sci., 46, 221-250.
    McCaul, E. W., Jr., D. E. Buechler, S. Hodanish, and S. J. Goodman, 2002: The Almena, Kansas, tornadic storm of 3 June 1999: A long-lived supercell with very little cloud-to-ground lightning. Mon. Wea. Rev., 130, 407-415.
    Miller, L. J., F. I. Harris, and J. C. Fankhauser, 1982: The 22 June case study: Structure and evolution of the internal airflow. Hailstorms of the Central High Plains, Vol. 2, C. Knight and P. Squire, Eds., University Press of Colorado, 35-59.
    Miller, L. J., J. D. Tuttle, and C. A. Knight, 1988: Airflow and hail growth in a severe northern High Plains supercell. J. Atmos. Sci., 45, 736-762.
    Miller, L. J., F. I. Harris, and G. B. Foote, 1990: Precipitation production in a large Montana hailstomr: Airflow and particle growth trajectories. J. Atmos. Sci., 47, 1619-1646.
    Moller, A. R., C. A. Doswell, M. P. Foster, and G. R. Woodall, 1994: The operational recognition of supercell thunderstorm environments and storm structures. Wea. Forecasting, 9, 327-347.
    Musil, D. J., A. J. Heymsfield, and P. L. Smith, 1986: Microphysical characteristics of a well-developed, weak echo region in a High Plains supercell thunderstorm. J. Appl. Meteor., 25, 1037-1051.
    Perez, A. H., L. J. Wicker, and R. E. Orville, 1997: Characteristics of cloud-to-ground lightning associated with violent tornadoes. Wea. Forecasting, 12, 428-437.
    Rasmussen, E. N., and J. M. Straka, 1998: Variations in supercell morphology. Part I: Observations of the role of upper-level storm-relative flow. Mon. Wea. Rev., 126, 2406-2421.
    Rison, W., R. J. Thomas, P. R. Krehbiel, T. Hamlin, and J. Harlin, 1999: A GPS-based three-dimensional lightning mapping system: Initial observations in central New Mexico. Geophys. Res. Lett., 26, 3573-3576.
    Saunders, C. P. R., 1993: A review of thunderstorm electrification processes. J. Appl. Meteor., 32, 642-653.
    Saunders, C. P. R., W. D. Keith, and R. P. Mitzeva, 1991: The effect of liquid water on thunderstorm charging. J. Geophys. Res., 96, 11 007-11 017.
    Seimon, A., 1993: Anomalous cloud-to-ground lightning in an F5-tornado-producing supercell thunderstorm on 28 August 1990. Bull. Amer. Meteor. Soc., 74, 189-203.
    Solomon, R., and M. Baker, 1998: Lightning flash rate and type in convective storms. J. Geophys. Res., 103, 14 041-14 057.
    Stolzenburg, M., 1994: Observations of high ground flash densities of positive lightning in summertime thunderstorms. Mon. Wea. Rev., 122, 1740-1750.
    Stolzenburg, M., W. D. Rust, and T. C. Marshall, 1998: Electrical structure in thunderstorm convective regions. 3. Syntheses. J. Geophys. Res., 103, 14 097-14 108.
    Takahashi, T., 1978: Riming electrification as a charge generation mechanism in thunderstorms. J. Atmos. Sci., 35, 1536-1548.
    Weisman, M. L., and J. B. Klemp, 1984: The structure and classification of numerically simulated convective storms in directionally varying wind shears. Mon. Wea. Rev., 112, 2479-2498.
    Williams, E. R., and Coauthors, 1999: The behavior of total lightning activity in severe Florida thunderstomrs. Atmos. Res., 51, 245-265.