A high-precipitation tornadic supercell storm was observed on 29–30 May 2004 during the Thunderstorm Electrification and Lightning Experiment. Observational systems included the Oklahoma Lightning Mapping Array, mobile balloon-borne soundings, and two mobile C-band radars. The spatial distribution and evolution of lightning are related to storm kinematics and microphysics, specifically through regions of microphysical charging and the location and geometry of those charge regions. Lightning flashes near the core of this storm were extraordinarily frequent, but tended to be of shorter duration and smaller horizontal extent than typical flashes elsewhere. This is hypothesized to be due to the charge being in many small pockets, with opposite polarities of charge close together in adjoining pockets. Thus, each polarity of lightning leader could propagate only a relatively short distance before reaching regions of unfavorable electric potential. In the anvil, however, lightning extended tens of kilometers from the reflectivity cores in roughly horizontal layers, consistent with the charge spreading through the anvil in broad sheets. The strong, consistent updraft of this high-precipitation supercell storm combined with the large hydrometeor concentrations to produce the extremely high flash rates observed during the analysis period. The strength and size of the updraft also contributed to unique lightning characteristics such as the transient hole of reduced lightning density and discharges in the overshooting top.
Before applications of total lightning data can be used to diagnose storms in weather operations, it is important to understand how lightning density and evolution relate to the evolving microphysics and kinematic structure of the storm as a whole. There is considerable evidence (e.g., MacGorman and Rust 1998, 65–30, 226–228) that the predominant charging in thunderstorms is due to rebounding collisions between graupel and cloud ice in the presence of liquid water (Takahashi and Miyawaki 2002; Saunders et al. 2006). Sedimentation then separates the charge on precipitation from the charge of opposite polarity on small particles to create regions of net charge, with the maximum electric fields typically between two regions of opposite charge polarity. In analyzing multiple regions of storms at different stages of development, many studies (e.g., Marshall and Rust 1991; Stolzenburg et al. 1998; Wiens et al. 2005; Weiss et al. 2008; Bruning et al. 2010) have concluded that, while charge regions often are treated as simple dipoles or tripoles, they can become quite complex. Much of the charge structure depends on the microphysical and kinematic characteristics of the main updraft region (e.g., Ray et al. 1987) and tends to become more complex in larger, longer-lived storms.
Because lightning is initiated in regions of large electric field, flash initiations tend to cluster in the vicinity of updraft cores, where either sedimentation alone or sedimentation combined with wind shear or turbulence produces gradients in the charged particles, thereby creating maximum electric fields large enough to initiate lightning (e.g., MacGorman and Rust 1998, 192–234; Bruning et al. 2007; Lund et al. 2009). Thus, the electrification process links the microphysical and kinematic evolution of storms with the evolving characteristics of storm charge and lightning.
As one might expect from this process, storms with a large updraft mass flux and large graupel volume typically produce large flash rates (e.g., Wiens et al. 2005; Kuhlman et al. 2006; Tessendorf et al. 2007; Deierling and Petersen 2008). Thus, isolated supercell storms tend to produce much larger flash rates than those of nonsevere isolated storms, often as high as hundreds per minute (e.g., Vonnegut and Moore 1958; Rust et al. 1981; MacGorman 1993; Wiens et al. 2005; Goodman et al. 2005; Steiger et al. 2007).
Some kinematic features typical of supercell storms are reflected in their distribution of lightning. For example, the stronger the updraft, the higher the height of the lightning associated with charge produced by noninductive charging in the mixed phase region, so supercell storms tend to elevate regions of charge higher than in other storms (e.g., MacGorman et al. 1989; Ziegler and MacGorman 1994; Marshall et al. 1995b). This is partly because the release of latent heat by rapid hydrometeor growth in strong updrafts shifts the mixed-phase region to higher altitudes, partly because there is less time at any given level within strong updrafts for sedimentation to macroscopically separate the charge that was transferred to hydrometeors by rebounding collisions, and partly because hydrometeors are lofted much higher in the storm (e.g., MacGorman et al. 1989; Ziegler and MacGorman 1994).
Supercell storms also typically have a transient lightning hole or ring related to the lack of precipitation-size particles, including graupel, in the bounded weak echo region (BWER), which greatly inhibits the noninductive charging mechanism and total lightning activity in that region (e.g., Krehbiel et al. 2000; MacGorman et al. 2005; Steiger et al. 2007; Payne et al. 2010). This feature can also be produced in regions of wet hail growth, which inhibits rebounding from collisions, in strong storms with or without rotation in their updrafts (e.g., Emersic et al. 2011). Storms with updrafts strong enough to produce a prominent overshooting top, typical of supercell storms, often have a secondary maximum in lightning density within the overshooting top (Lhermitte and Krehbiel 1979; MacGorman et al. 2005; Emersic et al. 2011); early observational evidence suggests that the timing of lightning in the overshooting top is associated with new updraft surges.
Heavy precipitation (HP) supercell storms, such as the storm analyzed in this study, are distinguished by the mesocyclone being embedded in heavy precipitation on the west or southwest flanks (Doswell and Burgess 1993; Moller et al. 1994). They tend to be larger than classic supercell storms and can have larger reflectivity values in the hook echo than in the precipitation core. Because HP storms are often larger storms, with more ice and precipitation than in low precipitation (LP) and classic supercell storms, one would expect higher flash rates in HP storms. Indeed, the first studies of Lightning Mapping Array (LMA) observations of HP supercells in northern Alabama have shown flash rates ranging from 150 to 800 min−1 (Bridenstine et al. 2005; Goodman et al. 2005; Schultz et al. 2011). The HP supercell storms are most commonly associated with predominately negative cloud-to-ground (−CG) flashes, as are most storms in the southeastern United States (Branick and Doswell 1992; Knapp 1994; MacGorman and Burgess 1994; Boccippio et al. 2001; Carey and Rutledge 2003). However, there is little documentation of total lightning as it relates specifically to the evolution and structure of HP storms.
This paper analyzes lightning relative to the kinematic structure of a HP supercell storm that occurred on 29–30 May 2004 during the Thunderstorm Electrification and Lightning Experiment (TELEX; MacGorman et al. 2008) in Oklahoma. The TELEX field experiment provided a uniquely comprehensive dataset to examine how storm electrification and lightning are linked to storm kinematics, microphysics, and dynamics. For this analysis, data from the Oklahoma LMA are combined with dual-Doppler analyses from mobile radar observations to further investigate the relationship of the updraft core and storm kinematics to the inferred charge structure and spatial distribution of lightning.
2. Data and methods
a. Radar data
The use of Doppler radar to infer the dynamical structure of storms has been a common practice since the 1970s (Ray et al. 1975; Brandes 1977). This study primarily utilizes the two C-band Shared Mobile Atmospheric Research and Teaching (SMART) radars, described by Biggerstaff et al. (2005), and occasionally employs data from two S-band radars: KTLX, the National Weather Service (NWS) Weather Surveillance Radar-1988 Doppler (WSR-88D) in Oklahoma City, Oklahoma, and KOUN, the National Severe Storms Laboratory (NSSL) polarimetric research radar located in Norman, Oklahoma. The SMART radars completed volume scans of the storm every 3 min for over 2 h as it passed through central Oklahoma. For the majority of analyses shown in this study, the sector volume scans of approximately 120° were used for both radars, with the radars completing an entire volume scan in 1 min, 50 s. Elevation angles ranged from 0.5° to 59° with increments of 0.3°–3.0°. Both SMART radars had a 1.5° beamwidth (Biggerstaff et al. 2005).
Data were edited manually by using the SOLOII software suite (Oye et al. 1995) to dealias radial velocities and remove ground blockage, range folding, and regions of high noise. With Nyquist velocities for SR1 and SR2 of 21 m s−1, velocity dealiasing for this supercell often involved more than one fold. Ground clutter was removed by identifying areas of near-zero velocity and high returned power at the lowest sweeps. A correction of azimuthal truck orientation of +1.0° for SR1 and −0.8° for SR2 was determined from ground-target echo patterns and the rotation correction was applied to the sweep filter in SOLOII.
Dual-Doppler analysis is a commonly used technique to estimate the three-dimensional wind field of supercell storms (e.g., Brandes 1977; Ray et al. 1980; Ziegler and MacGorman 1994; Dowell and Bluestein 1997; Wurman et al. 2007). The SMART radars were deployed in a configuration suitable for dual-Doppler analysis for the supercell that passed through Geary, Oklahoma (west of Oklahoma City). Data collection started at 2247 UTC 29 May 2004 and ended at 0212 UTC 30 May 2004. Dual-Doppler analyses were completed in this study for nine volume scans spanning the time period in which the storm moved through the TELEX domain: 2320, 2331, 2347, 2355, 0011, 0016, 0027, 0038, and 0052 UTC.
Radial velocity and reflectivity from both SR1 and SR2 were interpolated onto a Cartesian grid with a grid spacing of 0.5 km using a single-pass Barnes analysis (Majcen et al. 2008; Trapp and Doswell 2000; Koch et al. 1983). The analysis employed a weighting function of the following form:
where ωjk,n is the weight assigned to the kth radial observation at the jth grid point on the nth pass, rjk is the distance (km) between the grid point j and observation k, κo is the smoothing parameter (km2), and γ is the convergence parameter (Majcen et al. 2008). The analyses used in the present study employed interpolation parameters of n = 1, κo = 1.343, and γ = 0.3 to provide adequate smoothness for the purposes of the present study. The Barnes filter response (e.g., Majcen et al. 2008) can be described by a ratio of the unfiltered input to the output sinusoidal filtered wave amplitudes (%); for the chosen weighting parameters of this study, the 10% and 50% filter responses occur at wavelengths of 2.4 and 4.4 km, respectively.
The three-dimensional winds were synthesized via an iterative solution of a system composed of two linear equations in u, υ, w, and radial velocity (i.e., one equation for each radar) and the anelastic mass continuity equation:
where u, υ, and w are the x (east), y (north), and z (vertical) components of the airflow, respectively, and κ is the logarithmic change in density with height [i.e., Eqs. (4)–(5) of Ray et al. (1980)]. The precipitation terminal fall speeds were estimated from radar reflectivity and subtracted from total precipitation vertical motion to express the aforementioned linear equations in terms of w. The dual-Doppler analysis was affected at a given grid point only if the between-beam angle ΘBBA at that point was inside the range 20° ≤ ΘBBA ≤ 120° and the maximum elevation angle at that point was less than 58°. Two unique iterative, dual-Doppler airflow solutions were obtained by integrating the continuity equation both upward and downward from boundary conditions of w = 0 at the ground and the local-column storm top (Ray et al. 1980). The two iterative vector airflow solutions were merged into a single weighted-average solution wherein the weight of the airflow component based on upward (downward) integration varied linearly from 1 at the ground (storm top) to 0 at the storm top (ground). The weighted-average airflow analysis retained a satisfactory approximation of Eq. (2).
Small errors may be incorporated in the retrieval of vertical velocity through dual-Doppler analysis due to incomplete sampling, boundary conditions (zero assumed vertical velocity at ground and storm top), and incorrect estimates of storm motion. Following Kessinger et al. (1987), the error analyses for u, υ, and w were linearly weighted versus height (Fig. 1) using a radar baseline equal to that of the radar baseline between SR1 and SR2 and grid location at the center of the updraft core at 2330 UTC. The linear weighting in height of the w-error profiles is the same as the weighted averaging of the upward and downward integration wind syntheses. The expected w-error values on the order of 5 m s−1 or less are small when compared to values of vertical velocity within the supercell main updraft (with smaller errors expected outside the core of the updraft) and horizontal velocity errors are negligible relative to the total horizontal wind.
b. Lightning data
Total lightning activity, both in-cloud and cloud-to-ground, can give insight into storm charge structure. The main tools used to investigate lightning in this study are the Oklahoma LMA and the National Lightning Detection Network (NLDN). Storm charge is also inferred from in-storm soundings by balloon-borne electric field meters.
Maps of lightning channel discharge geometry and flash rates were acquired with the Oklahoma LMA. The Oklahoma LMA (Rison et al. 1999; Thomas et al. 2004; MacGorman et al. 2008) is a global positioning system (GPS) based, time-of-arrival system that maps lightning by measuring the time at which a very high frequency (VHF) electromagnetic signal produced by a developing lightning channel arrives at each station in the array in central Oklahoma. In 2004, the Oklahoma array had 11 stations spaced 10–22 km apart, each of which received signals in a locally unused television channel in the VHF band (channel 3, 60–66 MHz). The LMA can map up to 12 000 sources per second, and for the storm described in this paper, typically mapped ten to several hundred points per flash. For this study, VHF sources that had a reduced χ2 greater than 2, had fewer than 7 contributing stations, or occurred more than 150 km from the center of the network were not used.
Individual flashes were resolved by using the flash algorithm developed by Thomas et al. (2003). The algorithm determines whether an individual VHF source point belongs to a given flash by using time and space constraints (3 km and 150 ms) between adjoining points in a flash and then determines the number of points the flash includes; individual flashes are limited to less than 3 s in duration. When flash rates are large, as was the case with the storm studied here, separation of the many VHF points into individual flashes becomes more difficult because of the continual activity, and so becomes more sensitive to changes in the time and space criteria. Thus, it is important not to focus on exact flash rates, but instead to examine trends in flash rates and the approximate value of the flash rates (Wiens et al. 2005; Murphy 2006; MacGorman et al. 2008). Throughout this study, rates should be regarded as approximate, although trends are likely reliable. As was done by Wiens et al. (2005), multiple flash algorithms, all using a minimum of 10 VHF points per flash, have been tested on the LMA data, and while the specific number may change slightly, the overall trends remained the same.
The location of flash initiations was determined by an algorithm developed by Lund et al. (2009). The algorithm groups the first 10 VHF points of a single flash and determines the average latitude, longitude, height, and standard deviation. If the standard deviation is greater than 0.5 km, an outlier VHF source point is dropped from the averaging until the standard deviation is less than 0.5 km. To reduce misplaced locations, the algorithm requires that at least five source points be used in each calculated flash initiation.
Individual flashes mapped by the LMA can be analyzed to infer the regions of storm charge involved in a flash. Numerical modeling studies (Mansell et al. 2002) and laboratory studies using sparks (Williams et al. 1985) have shown lightning channels are denser and propagate farther in regions of larger charge density. According to the bidirectional model (Kasemir 1960; Mazur and Ruhnke 1993), which has become the paradigm for understanding lightning development, lightning is initiated between regions of opposite charge where the magnitudes of the electric field are near a local maximum. The lightning then propagates bidirectionally into regions of opposite charge, with the negative leader traveling toward and through regions of positive charge and the positive leader traveling toward and through regions of negative charge.
Shao and Krehbiel (1996) and Rison et al. (1999) have demonstrated that VHF mapping systems such as the LMA preferentially map negative leaders (which tend to propagate through positive charge), as negative leaders produce much more noise than positive leaders do at the radio frequencies used by the LMA. Thus, individual flashes can be examined to identify the charge structure of the storm (Wiens et al. 2005; Rust et al. 2005). This method requires a flash-by-flash analysis, subjectively partitioning the first several sources and higher density activity (positive charge) from the sources occurring later and at lower density (negative charge).
3. Observations and analysis
a. Environmental conditions
Conditions across the southern plains were extremely unstable on 29 May 2004 as an upper-level trough moved across the area and enhanced strong vertical shear. A surface low in southwestern Kansas and a strong dryline extending southward across western Oklahoma led to the initiation of thunderstorms in the area (Fig. 2a). Afternoon temperatures and dewpoints east of the dryline were near 30° and 20°C, respectively. Convection was originally delayed by a strengthening cap, but as the upper-level trough deepened, 20–25 m s−1 winds from the south combined with surging dewpoints along an area of convergence along the dryline to initiate the first cells around 1955 UTC. An environmental sounding launched into the region of the storm inflow from Minco, Oklahoma, at 0008 UTC indicated large CAPE (>3000 J kg−1) and veering winds with a peak of 50 m s−1 at 220 hPa (Fig. 3).
Two of the storms initiated by the low-level convergence in western Oklahoma on 29 May 2004 were dominant. The first (storm A) was initiated at approximately 2030 UTC as the southernmost of three cells along the dryline. By 2130 UTC, at least four other storms were north of storm A and two more cells had been initiated south of storm A (Fig. 4a). The focus of this paper, the Geary storm (storm B), initiated at 2130 UTC at the southern end of this line of storms. Storm A remained the strongest of these storms through the next hour and developed supercell characteristics, including midlevel rotation, as it moved northeast at 16 m s−1 through 2230 UTC. During the following hour (2230–2330 UTC), storm A began to turn more eastward and slowed to 12 m s−1. Meanwhile, the Geary storm rapidly intensified and moved northeast, eventually interfering with storm A's low-level inflow from the south. At around 0000 UTC, the motion of the Geary storm slowed and began to turn right (Fig. 2b). After 2330 UTC, the Geary storm was the strongest storm (both in terms of peak reflectivity values and the number and intensity of severe reports) in the domain, and after 0030 UTC, it was the only storm in the TELEX domain.
The tornado times and tracks contained within this study are taken from Storm Data, which combined data from the Doppler on Wheels (DOW), chaser accounts, and field surveys (D. Burgess 2009, personal communication). The strongest tornadoes of interest during the analysis period of the Geary storm include two F2 tornadoes: one at 0017–0038 UTC that traveled from northwest to north of Geary and another at 0038–0111 UTC that moved from north of Geary to the vicinity of Calumet, Oklahoma.
b. Dual-Doppler analysis
The Geary storm had developed into a mature supercell by the time it was in the field of view of both mobile radars; a well-defined midlevel mesocyclone, BWER, and a midlevel updraft exceeding 70 m s−1 are all evident at 2347 UTC (not shown). By 2347 UTC, it had already produced its first brief tornado (F0) at 2330 UTC near the town of Thomas, Oklahoma. A double-hook echo structure was evident in the low-level reflectivity, with rotation evident along the inflow gust front region aligned with the eastern hook echo and along a weaker region wrapped into higher reflectivity in the western hook echo. By 2355 UTC (Fig. 5), a mesocyclone was associated with the eastern hook echo, with vertical vorticity values exceeding 4 × 10−2 s−1 at midlevels, almost double the maximum 8 min earlier. This mesocyclone was collocated with the BWER and peak updraft with vertical velocities reaching 65 m s−1 (Fig. 5).
Tornadogenesis occurred at 0016 UTC for the F2 tornado that struck Geary. The analysis at 1 km at this time depicts the increased strength and extent of the mesocyclone as vertical vorticity values exceeded 2 × 10−2 s−1 and the elongated shape of the mesocyclone, with a southwest-to-northeast orientation (Fig. 6). A single-hook echo replaced the two hook echoes seen previously at 1 km and reflectivity values increased to as high as 60 dBZ in the rear flank downdraft (RFD). Along the gust front, there were two pronounced maxima in vertical velocity, both with continuity extending upward through the middle levels of the storm. Though peak updraft values (50 m s−1 between 7.5 and 8 km) were slightly lower than in the previous analysis at 2355 UTC, the main updraft core grew in areal extent during this time period. Coinciding with this larger updraft region, the BWER also increased in size and reflectivity deficit at this time.
The tornado remained on the ground while the supercell continued to strengthen over the next 10 min. By 0027 UTC (Fig. 7), the mesocyclone had become more symmetrical in shape with vertical vorticity exceeding 4 × 10−2 s−1 at 1 km and greater than 3 × 10−2 s−1 at 6 km. The main updraft core remained continuous from 1 km to storm top at about 17 km, with updraft values exceeding 60 m s−1 between 5.5 and 7.5 km.
The storm maintained this intensity in the period 0027–0052 UTC, after which the mesocyclone became fully occluded (Fig. 8). The mesocyclone was clearly wrapped within the RFD at 1 km and was separate from the BWER and updraft region in the midlevels of the storm. The supercell exhibited a typical cyclic evolution during this time period, as the old circulation became enveloped within the storm core and a new area of rotation simultaneously formed ahead of the occluding hook echo, along the gust front. Peak vertical velocities in the main updraft core remained about 55–60 m s−1 between 5 and 9 km through this period.
c. Lightning and inferred charge structure
1) Evolution of lightning relative to evolving storm characteristics
In storms with high flash rates, such as the Geary storm, it can be difficult to delineate those flashes occurring in close proximity in space and time. As mentioned in section 2, multiple flash algorithms were tested, and while the specific number of flashes varies slightly from algorithm to algorithm, the overall trends remain the same. Furthermore, the flash rates from all algorithms were consistent in showing rates of at least 200 min−1 throughout the analysis period. A minimum in flash rate occurred around 2340 UTC, followed by an increase and leveling off of flash rate through 0000 UTC and a dramatic increase in flash rate to a peak at 0015 UTC (Fig. 9). The dramatic increase in flash rate occurred just prior to the touchdown of the tornado northwest of Geary at 0017 UTC.
Dual-Doppler velocities at the analysis times discussed in section 3b were used to calculate the updraft mass flux through the storm for vertical velocities greater than 5 and 10 m s−1 (Figs. 10 and 11). Thresholds of 5 and 10 m s−1 were used, and have commonly been used in previous studies, as it is believed vertical velocities of at least 5–7 m s−1 are required for the production of graupel in the mixed-phased region requisite to thunderstorm charging (e.g., Zipser and Lutz 1994; Deierling and Petersen 2008). The updraft mass flux (MF, kg s−1) through a level i was calculated according to Emanuel (1994):
where ρi is the density of air at the ith level, σi is the areal coverage of the updraft (m2), and wi is the vertical velocity (m s−1).
Although there are not enough dual-Doppler analyses for robust statistical correlation, the overall flash rate for the storm (Fig. 9) is generally higher near the time of or following increases in the updraft mass flux throughout all heights of the storm (e.g., between 2355 and 0016 UTC) (Fig. 10). Updraft mass flux and flash rate both depict the same decrease at 0040 UTC, followed by an increase near the end of the analysis period after 0052 UTC. The height of updraft mass flux maxima was between 10 and 12 km (where in situ updraft temperatures were near −20° and −40°C) throughout the analysis period while peak vertical velocities varied throughout the column from one time to the next (Fig. 11). The height of the updraft mass flux maxima matched the heights of maximum density of flash initiations and VHF sources (Figs. 12 and 13).
Following the methodology of Deierling and Petersen (2008), similar trends are evident in total updraft volumes for vertical velocity exceeding 5 and 10 m s−1 for the mixed-phase region (between −5° and −40°C) (Fig. 10). The values of updraft volume of this storm are on the same order of magnitude of other simulated and observed supercell storms noted in previous studies (e.g., Kuhlman et al. 2006; Deierling and Petersen 2008).
These observations collectively are consistent with the relationship between convective-scale charge separation and updraft mass flux and updraft volume (e.g., Wiens et al. 2005; Deierling et al. 2005) that is expected if noninductive charging dominates thunderstorm electrification (e.g., Takahashi and Miyawaki 2002; Saunders et al. 2006). Throughout the analysis period, the Geary storm contained consistent maximum vertical velocities around 45–55 m s−1 at heights of 6–10 km though the flash rate varied by more than 200 min−1. Thus, as seen in previous studies (e.g., Kuhlman et al. 2006), there is further evidence in the lack of correlation between variations in the maximum updraft velocity and the flash rate. It appears that the rate of thunderstorm electrification is not so much proportional to the vertical velocity of hydrometeors, but to the total graupel number concentration and the rate of rebounding collisions with cloud ice (represented by updraft mass flux or updraft volume).
Cloud-to-ground lightning in the 29 May 2004 storm was usually dominated by −CG flashes (Fig. 9). The only exceptions were the 2340–2350 and 0025–0035 UTC periods during which the −CG flash rates dropped rapidly to become comparable to or less than the +CG flash rate (Fig. 9). At its maximum, the CG flash rate (for flashes greater than 10 kA) reached 20 min−1, but the overall average was closer to 9.4 min−1 for the 2-h period between 2300 and 0100 UTC. The peaks in CG flash rate at 0003, 0047, and 0110 UTC (17, 20, and 18 flashes, respectively) were before, during, and just after tornadoes, respectively. Thus, as found previously (MacGorman et al. 1989; MacGorman and Nielsen 1991; Schultz et al. 2011), the CG flash rate showed no consistent correlation to tornadogenesis or to other types of severe weather at the ground.
One aspect of the cloud-to-ground lightning that remained relatively consistent throughout the analysis is the predominance of −CG flashes in the RFD and mesocyclone region. While the forward-flank downdraft (FFD) was also dominated by −CG flashes, typically a much larger percentage of +CG flashes occurred here. At times (e.g., between 0010 and 0030 UTC), the two polarities were evenly split in the FFD. Less than 10% of the ground flashes in the mesocyclone/RFD region lowered positive charge to ground while approximately 35% lowered positive charge in the FFD region. The −CG flash rate in the RFD greatly increased at the time of the mesocyclone occlusion around 0050 UTC, growing from 1–2 min−1 between 0010 and 0040 UTC to 4–5 min−1 between 0040 and 0100 UTC.
Throughout most of the analysis period, the percentage of CG flashes remained less than 3%–5% of the total lightning flash rate. Strong storms on the southern plains typically have a low percentage of CG flashes due to the large amount of charge centered at higher altitudes, having been lofted by the large, intense updrafts of these storms (MacGorman et al. 1989; Ziegler and MacGorman 1994; Stolzenburg et al. 1998; Boccippio et al. 2001; Wiens et al. 2005; Steiger et al. 2007). However, the percentage of CG flashes for this storm is smaller even than for most other documented supercells. The maximum CG percentage (7%) during the analysis period occurred at 0045 UTC at the time of the mesocyclone occlusion. The −CG flash rate increased rapidly, primarily in the region of the RFD, doubling from 10 to 20 min−1 at a time when the total flash rate decreased to 300–350 min−1.
While the storm was in three-dimensional (3D) LMA range, the majority of lightning was initiated near 10–11 km in height and in temperatures between −20° and −30°C in the storm updraft and between −39° and −41°C outside (Fig. 12). Compared to other storms, particularly other supercells, it is unusual for such a large number of flashes to be initiated this high. This is likely due to the continued strength and size of the updraft (as determined from the updraft mass flux and volume) throughout the analysis period. Additionally, this likely contributed to the low CG percentage in this storm, as suggested by MacGorman et al. (2011). Each of the dual-Doppler analyses computed vertical velocities of at least 45–60 m s−1 through the midlevels of the main updraft (6–10 km in height) throughout the analysis period and the largest values of updraft mass flux and updraft volume both occurred above 9 km. Following a brief lull and decrease in initiation heights from 0003 to 0011 UTC, the highest density of flash initiations occurred just after 0012–0020 UTC between 10 and 11 km, immediately before and during the time of the tornado northwest of Geary. Flash initiations and VHF source density then increased at 8 km (−8°C in the updraft, −25°C otherwise) between 0025 and 0035 UTC, providing a bilevel maxima in the density patterns in Figs. 12 and 13. The storm continued to have bilevel maxima throughout the rest of the analysis period (i.e., through 0100 UTC). There was a simultaneous increase in the CG rate coinciding with this increase of flash initiations in lower regions of the storm (Figs. 9 and 12). It appears that the development of a lower charge area strong enough to participate in lightning activity may have increased the storm's overall ability to produce CG lightning.
The persistent large number of initiations at an altitude between 10 and 11 km indicates it was the region in which electric field magnitudes tended to be largest. Because the largest electric fields are typically between two charge regions of opposite polarity, this suggests this region was a preferred altitude for macroscopic charge separation in this storm. Temperatures in the updraft at this altitude were between −20° and −30°C and this was the level of maximum updraft mass flux values in almost all of the dual-Doppler analyses. Lightning structure indicated the region above 11 km (likely composed of cloud ice and snow, colder than −30°C in the updraft sounding, −40°C outside the updraft) consistently contained negative charge. The region below 10 km prior to 2350 UTC (or below 11 km afterward) contained predominately positive charge (likely on larger ice hydrometeors, −20° to −30°C in the updraft core). Later, following the increasing number of initiations between 8 and 9 km and the secondary maxima in source density just below that, a secondary region of positive charge seems to have formed at lower levels in the storm, producing at least four main regions of charge vertically. While this is a good conceptual model of the vertical charge near the main updraft region, the actual charge structure was likely much more complex, with small pockets of opposite charge existing in close proximity horizontally as well as vertically. If the charge structure were as simple as the conceptual model suggests, flashes in the core would likely be similar to flashes in the anvil region or stratiform region of an MCS with long and extensive leader breakdown. The nature of flashes in the core of the storm points to small, compact regions of similar charge with little room for leader breakdown before reaching an area of unfavorable charge. Further details of the charge structure and its relationship to storm dynamics will be examined in a later paper employing radar data assimilation into a numerical model and utilizing explicit parameterizations for electrification and lightning.
As hinted at above, there was a distinct difference in the frequency and size of flashes as one moves from near the turbulent storm core out to the far anvil region (Fig. 14). Lightning that initiated in or near the main updraft in the storm core was shorter (typically <50 points total) and more frequent, averaging 157.3 min−1 within a 20-km radius of the storm updraft (Fig. 15), between 2230 and 0100 UTC. Flashes mapped by the LMA within this region usually extended no longer than 5–10 km horizontally and 3–5 km vertically. Meanwhile, lightning in the anvil frequently contained over 100 points per flash, spanned greater than 30 km in horizontal extent, and was much less frequent, averaging only 2.68 flashes per minute during the analysis period (Kuhlman et al. 2009 and Weiss et al. 2012 present additional details about the anvil lightning in this storm). Generally, the frequency of flashes decreased and the horizontal extent increased as the distance from the storm updraft increased.
The spatial distribution of flashes across the storm core and anvil was consistent throughout the 1.5-h analysis period (e.g., Fig. 14). The smallest flashes were most prevalent in the hook echo region, the backsheared anvil, and the updraft region. The flashes farther downshear from the main updraft and through the FFD toward the anvil were longer in length, both in terms of the number of points and the distance spanned by each flash.
The nearly continuous flashes (averaging greater than one flash per second) in the vicinity of the updraft core are confirmation of the sustained electrification in this region. If hydrometeors become charged by noninductive charge exchange during rebounding collisions between riming graupel and cloud ice, as is discussed above, it is natural to expect the greatest charging rates to exist in regions where the greatest number of rebounding collisions are occurring. Because actively riming graupel implies a mixed-phase region with supercooled liquid water, the greatest charging rates would be expected in and around updrafts, where such conditions are present. The highest activity of lightning should occur in this region and just downstream of the updraft, in which we would expect the rate of charge replenishment to be greatest. Two features in the distribution of lightning in the 29–30 May 2004 supercell illuminate the relationship with updrafts further: 1) the absence of lightning collocated with the updraft area and BWER and 2) the existence of lightning in the overshooting top, directly above the main updraft core. The next two sections will examine these phenomena further.
2) Lightning holes or rings
Lightning holes (Krehbiel et al. 2000) are regions of substantially reduced lightning density, likely containing large updraft speeds. The BWER associated with these strong updraft speeds indicates a shift in precipitation growth to higher altitudes. Additionally, a strong updraft allows little time for sedimentation leading to macroscopic charge separation to occur. Besides the short residence times for charging, wet growth of hail in strong updrafts may also reduce the amount of charge in the region, as was hypothesized by Emersic et al. (2011), though this explanation is unlikely to apply to this particular storm. It is likely that little lightning propagates through this region because lightning preferentially travels through regions of greater charge density (MacGorman et al. 1981, 2001; Williams et al. 1985; Mansell et al. 2002). Lightning holes have also been referred to in the literature as “lightning rings” (e.g., Payne et al. 2010) when the emphasis is on the ring of higher lightning density instead of the lack of activity at the center. However, the focus for this analysis will be the minimum.
Though lightning holes are typically a transient feature (duration ≤20 min) approximately 5–10 km in diameter (Wiens et al. 2005; Zhang et al. 2004; MacGorman et al. 2005), some type of lightning hole or minimum existed throughout the analysis period of this storm (2347–0052 UTC). However, the size and shape of the lightning minima varied from one 3-min volume scan to the next, though typically remaining between 7 and 12 km in diameter. To determine the relationship between dual-Doppler characteristics and the corresponding lightning density, LMA data were limited to the 4 min surrounding each of the dual-Doppler analyses, providing roughly 30 s of time on either side of the radar volume scans. The lightning minima were most evident at midlevels of the storm where relatively few precipitation-sized particles existed in the updraft core. Additionally, the feature was three-dimensional in nature and could be obscured by higher-altitude lightning channels in a two-dimensional plan plot encompassing all heights of the storm (Fig. 16). This was especially true during periods when lightning within the overshooting top is included, much of which was directly above the lightning hole.
As in the storm observed by McCaul et al. (2002), the hole was consistently collocated with and extended slightly above and downshear from the BWER, but observations from the 29 to 30 May 2004 storm show that the shape of the hole differed somewhat from that of the BWER and updraft (Figs. 16 and 17a–c). Although the lightning hole typically was within updrafts ≥20–25 m s−1, the largest updraft speeds were typically just upshear of it. Perhaps the kinematic feature corresponding best with the lightning hole was the horizontal velocity field at 10 km (which also was the height of largest density of mapped VHF sources). As shown in Fig. 17, the divergence and circulation of horizontal wind were roughly centered on the lightning hole in each of the analyses. This suggests that charged hydrometeors were actually flowing around the hole, as suggested by Payne et al. (2010). Payne et al. (2010) noted that rings of lightning activity were associated with Zdr and ρhv rings and thus inferred that the lightning ring was associated with graupel circulating around the mesocyclone. During the occlusion of the mesocyclone (i.e., around 0050 UTC), lightning channels became more numerous and filled in older parts of the hole, and new regions of lightning holes appeared as new updrafts and mesocyclones formed.
Not only were flashes unlikely to begin within the lightning hole, individual leaders were unlikely to propagate through it. Flashes that began elsewhere and propagated toward the hole typically curved left or right to go around the hole as they approached it. Flashes that began near the ring of lightning around the hole initiated near the outer edge of the ring and propagated inward or began on the interior edge of the ring and propagated either around the hole or outward. Flashes that initiated in the lightning ring had no preferred direction of propagation either upshear or downshear surrounding the lightning hole. However, the longest flashes were typically initiated either south-southeast of the hole (between 9 and 11 km) or north to northeast of the hole (between 5 and 9 km) and propagated downshear. This is likely reflective of large-scale net charging and charge replenishment in the region and localized transport of this charge away from the updraft core.
The duration and spatial extent of flashes contained in the region of the peak updraft, southwest of the lightning hole tended to be quite short, containing only one to five mapped points. Occasionally (2–3 min−1 on average), a longer flash moved through the region. Given the small size and duration of flashes in and around the region of the peak updraft, it is possible that these flashes produced too little light and transferred too little charge to have been detected by more traditional means, such as electric field change meters or optical sensors. In contrast, regions in and near updrafts ≥20 m s−1 (but usually outside regions of ≥40 m s−1) consistently contained relatively large densities of flash initiations (Fig. 15), as well as relatively large VHF source densities (Figs. 16 and 17) throughout the analysis period. This is consistent with the hypothesis that the periphery of the updraft core is a favored region for noninductive charging, consistent with the modeling results of Kuhlman et al. (2006) and as inferred from aircraft observations by Dye et al. (1986).
3) Lightning in overshooting top
High-altitude lightning, occurring at 15–16 km, above the region of the lightning hole, has been noted in previous supercell storms occurring over LMA networks (Lhermitte and Krehbiel 1979; Krehbiel et al. 2000; Bruning et al. 2010; Emersic et al. 2011). An increasing trend of lightning in the overshooting top is expected to correspond generally to rapid vertical growth of the storm, which would likely be related to pulses of increasing updraft velocities and increasing updraft mass flux.
In the 29 May 2004 storm, the majority of high-altitude VHF sources do not occur in episodic bursts, but occur nearly continually at a very low rate of at least six to eight separate VHF source points spread across every second. These singular VHF sources are too far apart in time and space to be considered a single flash, each failing criteria of distance or time (e.g., MacGorman et al. 2008; Lund et al. 2009) for associating it with other points to form a flash. Furthermore, there typically is a minimum in VHF source density between 13 and 16 km and VHF sources above 16 km do not correspond systematically to flashes lower down. Thus, it appears likely that flashes contained in the overshooting top are a separate entity from those in the storm core just below.
A comparison with dual-Doppler analyses indicates that these isolated points were concentrated directly above the main updraft and BWER, as well as downshear of the updraft near the upper level of the 20-dBZ reflectivity contour (Figs. 18 and 19). A sounding through the updraft core revealed temperatures falling below −40°C at 11.7 km, more than 3 km below the height of the lightning activity. The region likely consists of only small, pristine ice crystals and no liquid water, limiting the amount of “traditional” charge separation in the region.
It is unknown why exactly the electrical activity in this region consists of relatively isolated continual VHF sources. However, because these source points were located near the cloud top and along the edge of the large reflectivity gradient above and downshear from the overshooting top, it is possible that the activity involves charge produced in the updraft interacting with a screening layer charge near the upper cloud boundary (Taylor et al. 1984; Proctor 1991). If screening layer charge at the boundary of the overshooting top is folded into the storm by turbulent eddies, it could decrease the distance to charge advected by the updraft and, therefore, increase the electric field.
It is likely that these upper discharges are influenced at least in part by the lower threshold for electric field breakdown at these heights (e.g., Marshall et al. 1995a, 2005). Although there is uncertainty about the details of lightning initiation, all mechanisms suggested thus far have a lower threshold of electric field magnitude required for flash initiation at higher altitudes. The breakeven electric field (McCarthy and Parks 1992) decreases with increasing altitude due to the decreasing density of air [see Eqs. (4) and (5); Fig. 20):
An electric field meter that was launched at 2346 UTC into the back of the storm core had magnitudes (≥100–140 kV m−1) approaching and surpassing the threshold of the breakeven electric field between 6 and 12 km (Fig. 20), similar electric field magnitudes ≥Ez have been measured previously (e.g., Marshall et al. 1995a, 2005). Electric field measurements were not obtained higher in the storm, but at the altitude of these continual VHF emissions (15–18 km), the breakeven electric field decreases to 23–33 kV m−1.
As stated earlier, periods of storm intensification or increased updraft strength can sometimes be diagnosed by periods of high lightning activity in time series plots of lightning density (e.g., Fig. 13). There are several periods in which additional increases in lightning activity occurred above 16 km, either as a rising extension of lightning from the main body of the storm [similar to the rising lightning maxima first noted by Ushio et al. (2003)] or as a completely detached enhancement within the overshooting top. For example, the 0010 UTC peak at 16 km followed an increase in lightning density that rose from roughly 10 km at 0005 UTC through to 12 km at 0007 UTC. There were several previous rising maxima that behaved similarly. On the other hand, at 0052 UTC there was an increase in lightning around 12 km as well as an increase at lower levels, but little or no activity above 16 km in the overshooting top. During most of the 2-h analysis period, there was a secondary maximum of VHF source density, with the largest densities above 16 km occurring 0010–0040 UTC. After this period the maximum altitude began lowering again.
Evaluating the LMA densities against the dual-Doppler analyses reveals that the largest density of points exists downstream and above updraft velocities ≥25 m s−1 between 12 and 18 km (Fig. 18). The horizontal position of the highest density of points in the upper area of the storm is within regions of larger updraft speeds or in divergent flow above the larger updraft speeds. Several studies (e.g., Williams et al. 1989; Ziegler et al. 1991) have pointed out that a local relative maximum in charge density in the vertical will tend to occur near a balance level as the updraft decreases with height and particles eventually stop rising. This behavior probably explains the rounded cap or mantle of VHF source density between 16 and 18 km in the overshooting top, with the highest density directly above the region of largest, most vertical updraft and tilting downward east of the overshooting top (Figs. 19a–c), along the vertical gradient in reflectivity. The north cross sections in Figs. 19d–f best depict the separation of the peak activity in the overshooting top into a double-pronged region, one corresponding to the divergent flow out of the updraft with a higher area of larger VHF density to the north where updraft speeds tended to be largest and a secondary peak in somewhat weaker updraft speeds to the south (Figs. 19d–f). (The relationship with the southern VHF maximum cannot be confirmed for 0010–0020 UTC because it was beyond the region where the mobile radars were scanning to storm top across the southern part of the storm.)
4. Summary and conclusions
This supercell produced extremely large flash rates, dominated by flashes of small spatial extent and duration in the vicinity of deep updrafts. Regions surrounding the main updraft core consistently contained the largest densities of flash initiations and VHF sources throughout the analysis period. This is consistent with the hypothesis that the periphery of the updraft core is a favored region for noninductive charging by rebounding graupel-ice collisions (Takahashi and Miyawaki 2002; Emersic and Saunders 2010), consistent with the modeling results of Kuhlman et al. (2006) and as inferred from aircraft observations by Dye et al. (1986). The resulting horizontal structure of lightning can be described as a lightning ring around a lightning hole. The lightning hole was roughly collocated with the BWER and the core of stronger updrafts, though its shape was somewhat different from both. The horizontal dual-Doppler velocities depicted divergence and curvature around the lightning hole at 10 km, suggesting that charged hydrometeors were actually being swept out and around this area. A similar inference was made by Payne et al. (2010) from the evolution observed in polarimetric radar measurements.
From the small flash size in and near the updraft core, we inferred that the vicinity of the updraft core produced relatively small pockets of charge, perhaps caused by turbulence resulting from the large horizontal shear in updraft speeds as modeled by Klemp and Wilhelmson (1978). The close proximity of rapidly generated, oppositely charged pockets likely produced the strong electric fields that caused frequent flash initiation. Lightning is initiated in a region of strong field, with positive channels at one end then tending to propagate through negative charge and negative channels at the other end tending to propagate through positive charge (Kasemir 1960; MacGorman et al. 1981, 2001; Mazur 1989; Coleman et al. 2003). Because a channel of one polarity tends to avoid charge of the same polarity, we suggest the small spatial extent of flashes near strong updrafts is caused by the small size there of charge pockets of a given polarity. Farther from the updraft, the horizontal extent of regions of charge tended to become larger, and this allowed flashes to travel farther before terminating.
Another behavior related to the strong, persistent updraft was the elevated height of flash initiations. The peak region of flash initiations was centered near 10 km, higher in altitude than what is typically observed in less intense storms, as suggested by MacGorman et al. (1989) and observed by Stolzenburg et al. (1998).
Another likely reason large updrafts led to complex charge structure was related to variations in the polarity of charge exchange during rebounding collisions between graupel and cloud ice. Laboratory experiments indicate that the polarity depends on both temperature and either liquid water content or the relative growth rates of graupel and cloud ice (e.g., Takahashi and Miyawaki 2002; Emersic and Saunders 2010), with graupel tending to gain positive charge at from roughly 0° to −10°C with moderate liquid water contents or at large liquid water contents regardless of temperature in the mixed phase region. Variations in latent heating and in the entrainment of drier air across a strong, broad updraft can cause large variations in both temperature and liquid water content as one traverses the updraft, as observed by Musil et al. (1986).
In recent years, several studies have examined storms whose vertical distribution of charge polarity is inverted from the usual distribution, with positive charge at altitudes usually having a negative charge and vice versa (e.g., Rust and MacGorman 2002; Williams et al. 2005; MacGorman et al. 2011; Bruning et al. 2012). The highly complex storm charge structure makes it impossible to characterize the whole storm as having either a normal or an inverted distribution of polarity, and this was reflected in the different polarity of midlevel charge lowered to ground by lightning in different regions. The rear-flank downdraft and hook region of the storm was dominated by −CG flashes; approximately 90% of ground flashes in this region lowered negative charge to ground. Meanwhile, 30%–50% of the ground flashes in the forward-flank downdraft region lowered positive charge to ground, though the exact percentage varied throughout the storms lifetime.
A secondary maximum of mapped VHF sources extended into the overshooting top shown by radar data. However, the discharges at these high altitudes appeared quite different from those lower in the storm, in that they tended to be composed of continual, singular, VHF sources that often were too far apart in time and space to be associated as a single flash. Furthermore, these VHF sources above 16 km appeared not to be associated with flashes lower in the storm. A comparison with dual-Doppler data indicated that these isolated points were concentrated in and downstream of the overshooting top above the main updraft and BWER. Typically between 16 and 18 km, the region containing these upper LMA points appears to be associated with a balance level, with the highest cluster of points directly above the region of largest, most vertical updraft, and points then tilting downward east of the overshooting top, along the vertical gradient in reflectivity. Note that all hypotheses for flash initiation indicate that the required electric field magnitude decreases with altitude and appears to be a few tens of kV m−1 at 15–18 km. While the specific cause of the high VHF sources is still uncertain, we suggest that the lower threshold for initiation, combined with the charge lofted by the updraft core into the overshooting top and interacting with the screening layer charge, contributed to the discharges that produced the high VHF sources in this storm.
This research was supported in part by National Science Foundation Grants ATM-0233268, ATM-1063537, and ATM-0802717. All environmental surface data were provided by the Oklahoma Mesonet. The authors thank Jerry Straka, Stu Ryan, William Beasley, and three anonymous reviewers for their comments and suggestions regarding this work. This study would not have been possible without the participants in the TELEX field program including more than 30 students from the University of Oklahoma.