The majority (61%) of severe storm reports (i.e., large hail and tornado) during the 1989–98 warm seasons (April–September) were associated with predominantly (>90%) negative cloud-to-ground (PNCG) lightning. Across the contiguous United States, only 15% of severe storm reports were characterized by predominantly (>50%) positive CG (PPCG) lightning activity. However, significant regional variability occurred in the relationship between warm season severe storm reports and CG lightning polarity. In the eastern United States, a significant fraction (81%) of severe storm reports occurred nearby PNCG lightning while only 2% of severe storms were associated with PPCG lightning. The CG lightning behavior was quite different over the northern plains; only 28% of severe storm reports were linked with PNCG lightning while 43% were characterized by PPCG lightning. Although the direct physical relationship is still not evident, this regional variability appears to be at least partially explained by differences in the meteorological environment of severe storms producing PPCG and PNCG lightning.
The locations of the monthly frequency maxima of severe storms that produced PPCG and PNCG lightning were systematically offset with respect to the climatological monthly position of the surface θe ridge on severe outbreak days. Severe storms that produced PPCG lightning generally occurred west and northwest of the θe ridge in the upstream θe gradient region. Severe storms generating PNCG lightning were located southeast of the PPCG lightning maxima, closer to the axis of the θe ridge in higher mean values of θe.
Rust et al. (1981a,b) were first to document cloud-to-ground (CG) lightning flashes lowering positive charge beneath severe thunderstorms. Rust et al. (1985) noted that positive ground strokes sometimes dominated the CG lightning activity in vigorous storms that produced large hail or tornadoes, with upward of 50% of the CG lightning being of positive polarity. Indeed, such storms were found to generate +CG lightning flash rates and flash densities comparable to those of negative polarity ground discharges in ordinary warm season thunderstorms (e.g., MacGorman and Burgess 1994; Stolzenburg 1994).
The possibility of a relationship between storm severity and predominantly positive cloud-to-ground (PPCG) lightning activity suggests that the real-time lightning data provided by the National Lightning Detection Network (NLDN) (Cummins et al. 1998) may be useful in the nowcasting of some severe local storms. Indeed, National Weather Service (NWS) and U.S. Air Force forecasters in the central plains region were some of the first meteorologists to note a subjective relationship between +CG lightning frequency and storm severity (e.g., Reap and MacGorman 1989; Branick and Doswell 1992; Knapp 1994; Bluestein and MacGorman 1998; Smith et al. 2000). The potential nowcasting application of +CG lightning data and the desire to understand the cloud electrification mechanisms responsible for this anomalous lightning behavior have led to increased interest in severe storms associated with PPCG lightning.
In many regional and local studies, PPCG lightning activity has been related to mesocyclone development and supercell structure (MacGorman and Nielsen 1991; Branick and Doswell 1992; Curran and Rust 1992; Seimon 1993; MacGorman and Burgess 1994; Carey and Rutledge 1998; Bluestein and MacGorman 1998; Gilmore and Wicker 2002; Carey et al. 2003, hereafter CPR), tornado occurrence (Reap and MacGorman 1989; MacGorman and Nielsen 1991; Seimon 1993; Knapp 1994; MacGorman and Burgess 1994; Perez et al. 1997; Bluestein and MacGorman 1998; Smith et al. 2000; Gilmore and Wicker 2002; CPR), large hail production (Reap and MacGorman 1989; Curran and Rust 1992; MacGorman and Burgess 1994; Stolzenburg 1994; Carey and Rutledge 1998), and the mesoscale environment (Reap and MacGorman 1989; MacGorman and Burgess 1994; Smith et al. 2000; Gilmore and Wicker 2002). As reported in the above studies, the relationship between severe storm morphology, the local environment, and polarity and frequency of ground flashes has varied considerably. As a result, our knowledge of severe storms that generate PPCG lightning should be considered fairly embryonic. Indeed, our understanding of the cloud electrification mechanisms within severe storms producing PPCG lightning remains speculative (e.g., MacGorman and Burgess 1994; Carey and Rutledge 1998; Williams 2001; Gilmore and Wicker 2002; Lang and Rutledge 2002).
Complicating matters further, the relationship between storm severity and positive CG lightning is not a general one. Many severe hailstorms and tornadic storms are dominated by negative CG lightning (e.g., Doswell and Brooks 1993; MacGorman and Burgess 1994; Perez et al. 1997; Bluestein and MacGorman 1998; Galarneau et al. 2000). As pointed out by Branick and Doswell (1992), “before we can use lightning polarity data effectively in warning operations, we must learn why only some severe storms produce high positive CG rates, and, in particular, which storms.”
To date, we still lack basic knowledge across the contiguous United States regarding 1) the frequency, percentage, and geographical and monthly distribution of severe storms that are characterized by a typical percentage (>90%) of negative CG lightning; 2) the frequency, percentage, and geographical and monthly distribution of severe storms that are characterized by a predominant (>50%) percentage and high flash density [≥0.01 km−2 h−1; Stolzenburg (1994)] of positive CG lightning; and 3) the relationship of 1 and 2 to the meteorological environment.
Taking a step toward the first two objectives above, Knapp (1994) analyzed the polarity of CG lightning flashes within the vicinity of 264 tornado reports during the spring of 1991. Knapp (1994) found that only 62 (23%) of the associated tornadic storms were positive strike dominated.1 Interestingly, these positive strike dominated tornadic storms occurred primarily west of the Mississippi River, consistent with most of the case studies mentioned above. Knapp noted that regions characterized by a less (more) vertically saturated environment contained more (less) positive strike dominated storms. Similar comprehensive analyses of large hail reports across the contiguous United States have been lacking up to this point.
Herein, we have greatly expanded upon the results of Knapp (1994). We have broadened the scope of study to include both large hail and tornado reports and have significantly increased the sample size of severe weather events to include 10 warm seasons (April–September) available during the first decade of operation of the NLDN (1989–98). For over 67 000 severe storm reports, we have determined the nearby positive and negative CG lightning flash density and percentage. This information was compiled as a function of geographic location, month of the year, and storm severity (i.e., hail size and F scale). With this dataset, we can readily address objectives 1 and 2 above.
Several studies have noticed regional variability in the polarity of CG lightning produced by severe storms (e.g., Curran and Rust 1992; Branick and Doswell 1992; MacGorman and Burgess 1994). MacGorman and Burgess (1994) determined that the mesoscale environment in which negative or positive CG lightning dominated severe storms on a given day was consistent from storm to storm. As a result, MacGorman and Burgess suggested that dominant CG lightning polarity was controlled by the mesoscale structure of the atmosphere, possibly through systematic effects on storm properties related to severe weather. Based on the regional behavior of CG lightning during several tornado outbreaks, Smith et al. (2000) proposed a conceptual model that related the CG lightning polarity of tornadic thunderstorms to the relative movement of the storm systems through ridges of near-surface equivalent potential temperature (θe). In Smith et al. (2000), the majority of storms whose CG lightning activity was dominated by negative flashes formed in regions of weak θe gradient, downstream of a θe maximum. The majority of thunderstorms whose initial CG lightning activity was dominated by positive flashes formed in regions of strong θe gradient, upstream of an θe maximum. When some of these storms crossed over the θe maximum, their predominant CG lightning polarity switched from positive to negative. Alternatively, some of these storms moved adjacent to the θe maximum and were dominated by positive CG lightning throughout their life cycle. Gilmore and Wicker (2002) present broadly similar results in their study of supercell CG lightning, radar characteristics, and severe weather behavior relative to a mesoscale outflow boundary during a well-observed tornado outbreak.
To expand upon these intriguing but still limited results, we have tested the conceptual model of Smith et al. (2000) on climatological temporal and spatial scales. We have examined mean monthly frequency of positive and negative CG lightning dominated severe events observed over the 10 warm seasons as they relate to concomitant distributions of near-surface θe [computed from National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis data on a 2.5° × 2.5° grid] on severe weather days. If the Smith et al. (2000) conceptual model is generally valid for explaining the regional behavior of the dominant CG lightning polarity in severe storms, then mean monthly frequency maps of predominantly positive and negative CG lightning dominant severe events should be systematically related to the 10-yr-average monthly pattern of near-surface θe on severe weather outbreak days.
3. Data and method
We have correlated severe storm reports of large hail (diameter ≥1.9 cm or 0.75 in.) and tornadoes (F0–F5), as compiled by the NWS (NOAA 1989–98) and archived at the Storm Prediction Center (SPC),2 to cloud-to-ground lightning data from the NLDN for the contiguous United States during the 1989–98 warm seasons (April–September). Cummins et al. (1998) and Wacker and Orville (1999a,b) report on the design and performance of the NLDN. The CG lightning data utilized in this study include ground strike location, date, time, and polarity. Since positive cloud-to-ground lightning flashes characterized by peak currents less than 10 kA were likely associated with misidentified in-cloud lightning from 1995 to 1998, they were removed from our data sample (Cummins et al. 1998; Wacker and Orville 1999a,b). The location, date, and time of each large hail and tornado report were used to compare to the CG lightning data. Only the beginning point of the first segment of a multiple-segment tornado was included in the data analysis.
Our rationale for using only large hail and tornado reports was threefold: 1) The dynamical–microphysical–environmental causes for severe straight-line winds are so numerous that we felt that the linkage between lightning behavior and severe weather would likely be obscured. For future work, it would be interesting to see if bow echoes, microbursts, derecho, and heat burst events yielded similar results as large hail and tornadoes. 2) Past case studies of severe storms that produced PPCG lightning focused overwhelmingly on the relationship between large hail and tornadoes and CG lightning polarity. Our motivation was to provide a comprehensive, climatological framework within which to interpret these case studies. 3) For practical reasons, we needed to narrow the scope of the study to something manageable.
Since our goal was to characterize the CG lightning polarity and flash density within the vicinity of a large number of severe storm reports (58 019 large hail + 9349 tornado = 67 368 total), a simple and pragmatic approach was developed that could be easily automated. Knapp (1994) used subjective visual criteria for determining the CG lightning activity associated with 264 different tornadoes. Manual analyses were deemed to be impractical for the large number of severe events considered in this study. Instead, we calculated the percentage and flash density (km−2 h−1) for both negative and positive polarity ground discharges occurring within 50 km and 0.5 h of each large hail report and 1 km prior to and within 50 km of an initial tornado report. We chose these temporal and spatial sampling intervals based on a survey of the literature, a subjective visual assessment of the method using national radar summaries and CG lightning data, and sensitivity tests.3
Several case studies of CG lightning behavior in intense tornadic storms suggest that anomalous behavior occurred prior to tornadogenesis (e.g., Seimon 1993; MacGorman and Burgess 1994). This was our justification for using a 1-h window prior to the tornado event time and why we used only the beginning location of the tornado. This temporal window is similar to the one used by Knapp (1994). Similar studies for severe hailstorms have shown anomalous CG lightning behavior prior to, during, and after large hail has been reported at the surface (e.g., Carey and Rutledge 1998). Hence, we utilized a 1-h period centered on each large hail report. Since some severe storms were associated with multiple storm reports, a given lightning flash might be included in the CG statistics associated with more than one severe storm report. For example, if a particular lighting flash fell within the temporal and spatial analysis radii associated with more than one severe storm report, it was included for calculation of CG properties in all such instances.
We then grouped the storms reports into three categories based on the nearby CG lightning behavior: 1) predominantly [>90%; e.g., Orville and Huffines (2001)] negative CG lightning (PNCG), 2) predominantly [>50%; e.g., MacGorman and Burgess (1994); Stolzenburg (1994)] positive CG lightning, and 3) high flash density [HFD; ≥0.01 km−2 h−1; Stolzenburg (1994)] PPCG lightning (Table 1). As defined, PPCG HFD is a subset of the PPCG lightning category.
Although somewhat subjective in nature, these categories are helpful in understanding the range of CG lightning behaviors in severe storms and their geographic variability. The PNCG lightning category is consistent with the behavior of most thunderstorms, both nonsevere and severe storms. The PPCG lightning category was chosen to represent anomalous positive CG lightning behavior. A sensitivity study of the method reveals that our general conclusions regarding anomalous positive CG lightning were insensitive to the choice of percentage of positive CG lightning within a range from 25% to 50%. Finally, the HFD PPCG category is consistent with most documented cases of anomalous positive CG lightning activity in severe storms (e.g., MacGorman and Burgess 1994; Stolzenburg 1994; Smith et al. 2000, among others). However, some recent case studies have shown the existence of a class of severe storms with low CG lightning flash density and anomalous (>25%) positive CG lightning for periods of up to an hour or more (e.g., Carey and Rutledge 1998; Lang et al. 2000; McCaul et al. 2002; Lang and Rutledge 2002). These storms are well represented in our PPCG lightning category.
A graphical example of the procedure is presented in Fig. 1 for a large hail report, which occurred at 1820 UTC on 7 May 1995 over the Texas Panhandle. The associated CG lightning activity from 1750 to 1850 UTC is depicted. All flashes striking ground within 50 km of the hail report were included in the calculation of percent positive and positive CG lightning flash density. In this example, the +CG lightning flash density was 0.02 km−2 h−1 and positive flashes accounted for 88% of the CG lightning activity. Based on the criteria given above, this large hail report was coincident with a HFD PPCG lightning event. This procedure was repeated on all large hail and tornado (with the slight modification discussed above) reports that occurred during the 1989–98 warm seasons. Since there were only minor differences in the results for tornado and large hail reports from 1989 to 1998, results for both types of storm reports were combined into a “severe storm” category and averaged over the 10-yr period.
To explore regional variability, the frequency (per 10 000 km2 area per warm season, April–September) and percentage of severe storm reports that were accompanied by a PNCG, PPCG, and HFD PPCG lightning event were calculated and gridded at 1° × 1° resolution over the contiguous United States. From this gridded data, maps were produced and are presented in section 4a in order to address the first two fundamental goals discussed in the introduction.
The goals of this study were to ascertain the frequency and geographic location of severe storms that are characterized by PNCG and PPCG lightning activity. Unfortunately, severe storm reports are not synonymous with “severe storms.” For example, a single severe storm can result in multiple severe storm reports. Or, large hail and tornadoes can be misreported or even go completely unreported for numerous nonmeteorological reasons (e.g., Kelly et al. 1978, 1985). Despite these problems, severe storm reports still represent the most comprehensive and straightforward record of large hail and tornadic events readily available for research purposes.
In this study, we assumed that errors in reporting and failure to report are about as probable in one particular lightning category of severe storms (e.g. PNCG, PPCG, HFD PPCG) as they are for any other category or for all severe storms in general. As a result, the percentage of severe storm reports that are characterized by high positive or negative CG lightning percentage and/or flash density and their geographical distribution are likely to reflect similar properties for severe storms. Because of incomplete and imperfect reporting, the absolute frequency of each CG lightning property associated with a report is likely not synonymous with the same CG lightning property of the associated severe storm and should be interpreted with some caution.
To address the third primary goal of this study, we follow Smith et al. (2000) who explored the link between storm environment, storm dynamics, and lightning production through analysis of surface θe patterns. Using elementary parcel theory, surface θe can be utilized to estimate the maximum potential updraft speed in the undiluted cores of deep convection (e.g., Darkow 1986). As severe storms move through regions of higher (lower) surface θe and ingest the more (less) buoyant air, their updrafts will intensify (weaken) (all else being equal). Although mesoscale environmental effects on storm structure and lightning production were postulated, the θe analysis in Smith et al. (2000) was accomplished with synoptic surface observations. If their results are generally applicable to severe local storms, then there should be a recognizable relationship between the climatological (1989–98) mean monthly pattern of surface θe calculated from NCEP reanalysis data and the corresponding total monthly frequency of PPCG and PNCG lightning events on severe storm outbreak days.
Ten years (1989–98) of NCEP reanalysis data (2.5° × 2.5° grid) were utilized to compute the monthly mean (0000 UTC) θe pattern from the gridded 0.995 sigma-level temperature, humidity, and pressure data over the contiguous United States from April through September (Bolton 1980). For the purposes of this study, we further restricted each monthly sample to include 0000 UTC data from only those days during which an HFD PPCG lightning event occurred somewhere in the contiguous United States (i.e., an HFD PPCG day). We utilized the closest (i.e., in time) available 0000 UTC reanalysis data to the severe weather outbreak. These severe outbreak days most resembled those studied in Stolzenburg (1994), MacGorman and Burgess (1994), and Smith et al. (2000), allowing for a more direct comparison. There were 3352 HFD PPCG lightning events occurring on 435 HFD PPCG days (about 24% of all days) during the 10 warm seasons. Since a large number of PNCG lightning events (14 483) also occurred on each of these 435 days, we were able to investigate the geographic relationship between near-surface θe, CG lightning polarity, severe weather, and the monthly variability of these relationships. In order to evaluate the conceptual model of Smith et al. (2000) on climatological temporal and spatial scales, we compared monthly (April–September) frequency maps of PNCG and PPCG lightning events to the mean monthly θe pattern for so-called HFD PPCG severe outbreak days (section 4b).
Since very little is known about the general behavior of positive CG lightning in severe storms, we begin with histograms of the positive CG lightning percentage (Fig. 2a) and flash density (Fig. 2b) associated with severe storm reports in the contiguous United States. For example, there were a little over 26 000 severe storm reports characterized by a positive CG percentage of 0%–4% (bin center and x axis label of 2%). This corresponded to about 38% of all severe storm reports (Fig. 2b). As shown in Fig. 2a, 61% of severe storm reports in our study were accompanied by PNCG lightning, which is typical for most thunderstorms in the United States (e.g., Orville and Huffines 2001). An overwhelming majority (85%) of large hail and tornado reports that occurred during the 1989–98 warm seasons were associated with ≥50% negative cloud-to-ground lightning. In other words, only 15% of severe storm reports were accompanied by predominantly positive CG (PPCG, >50%) lightning. The distribution in Fig. 2a is strongly positively skewed with a long tail characterized by nearly equal frequency starting at about 25% +CG lightning. This percentage also represents an inflection point in the cumulative frequency histogram after which the line becomes nearly linear. Because of this characteristic of the distribution, the results of our study were insensitive to the definition of PPCG lightning in the range of 25%–50%.
Only 10% of severe storms produced high flash densities (≥0.01 km−2 h−1) of positive CG lightning (Fig. 2b). Only half of that population, or 5% of all severe storm reports, were associated with PPCG HFD lightning. A majority (58%) of severe storms produced low flash densities (<0.002 km−2 h−1) of positive CG lightning. This is equivalent to <16 positive ground flashes over a period of an hour in a circular area with radius of 50 km.
Clearly, the presence of PPCG lightning beneath severe storms is an infrequent event across the contiguous United States during the warm season despite an increasing number of studies documenting this phenomenon. In addition, the occurrence of severe storms dominated by high percentages and flash densities of positive ground strikes appears to be even more rare in spite of several high profile case studies (e.g., Seimon 1993; MacGorman and Burgess 1994; Stolzenburg 1994; CPR). The rarity of severe storms accompanied by PPCG lightning is at odds with the growing number of case studies of the phenomena. A quick review of the studies above sheds some light on a possible reason for the discrepancy. Nearly every severe storm characterized by PPCG lightning in the above studies occurred west of the Mississippi River and most of these unique storms occurred over the high plains. Based on this limited number of studies, it appears that the relationship between positive CG lightning and severe weather may vary geographically.
a. Geographic variability of CG lightning polarity in severe storms during the warm season
We utilized gridded maps over the contiguous United States to explore regional variability in the relationship between severe storm reports and PPCG lightning events. For comparison, we first present the frequency (per 10 000 km2 per warm season) of severe storm reports (large hail and tornadoes) for the 1989–98 warm seasons in Fig. 3.
There was a broad maximum in the frequency of severe weather in the southern to central plains region. The region characterized by >30 severe storm reports per 10 000 km2 per warm season was located over Oklahoma, northern Texas, the Texas Panhandle, and most of Kansas stretching west to the Kansas–Colorado–Nebraska border. This broad frequency maximum is generally consistent with the long-term climatology of tornadoes (Kelly et al. 1978; Concannon et al. 2000) and large hail (Kelly et al. 1985).4 The absolute maxima exceeded 60 severe storm reports per 10 000 km2 per warm season in northwestern Oklahoma.5 In addition to a decrease in severe storm report frequency from the central to eastern United States, there was a clear north-to-south gradient in the frequency of severe storm reports. The mean frequency was 24.8 (10 000 km2)−1 per warm season in the southern plains (e.g., see Fig. 3 for depiction of regions). On the other hand, the mean frequency of severe storm reports decreased to 9.1 (10 000 km2)−1 per warm season in the northern plains. The average frequency in the eastern region was significantly less than the southern and central plains but comparable to the northern plains. See Table 2 for a summary of the mean severe storm report frequency per region.
Gridded maps of the frequency [(10 000 km2)−1 per warm season] and percentage of severe storm reports accompanied by a PNCG lightning event are given in Figs. 4a,b, respectively. Since 61% of all severe storm reports across the contiguous United States were associated with PNCG lightning activity, the geographical pattern in Fig. 4a is comparable to that of Fig. 3. The similarity was especially true in the eastern and southern plains regions where 81% and 63%, respectively, of severe storm reports were accompanied by PNCG lightning in the mean. The exception was in a southwest-to-northeast tilted region in the central and northern plains (e.g., Kansas–Colorado border through Nebraska, eastern South and North Dakota, and into Minnesota), where <40% of severe storm reports were associated with PNCG lightning. The minimum in PNCG percentage (<20%) was located in eastern Nebraska, eastern South Dakota, and western Minnesota, consistent with the results of Zajac and Rutledge (2001).
Gridded maps of the frequency [(10 000 km2)−1 per warm season] and percentage of severe storm reports accompanied by a PPCG lightning event are given in Figs. 5a,b, respectively. As seen in Fig. 5a, the overwhelming majority of severe storm reports associated with ≥50% +CG lightning were located east of the Rocky Mountains and west of the Mississippi River, similar to the positive strike dominated tornadoes in Knapp (1994). A broad region of >2 +CG dominated severe storm reports (10 000 km2)−1 per warm season stretched from eastern Colorado and New Mexico and the Texas Panhandle northeastward to the eastern Dakotas, southern Minnesota, and northern Iowa. The overall frequency maxima of 23 (10 000 km2)−1 per warm season occurred in western Kansas near the Colorado border.6 Several other relative maxima (e.g., >8 per 10 000 km2 per warm season) were located at the Texas Panhandle and north-central Texas, central Oklahoma, eastern Nebraska and South Dakota, and southern Minnesota. Very few severe storm reports in the eastern region were associated with PPCG lightning [i.e., a mean frequency of 0.2 (10 000 km2)−1 per warm season). The central plains was the favored region for PPCG lightning during severe storms with a mean frequency of about 8 reports (10 000 km2)−1 per warm season. The southern and northern plains regions had similar mean frequencies of positive ground flash dominated severe reports [i.e., approximately 3–4 (10 000 km2)−1 per warm season].
The broad frequency maxima of severe storms dominated by +CG lightning in the central southern plains was located in the general vicinity of the dryline during the spring and early summer months (e.g., Hagemeyer 1991; Bluestein 1993). Several of the first case studies of positive strike dominated storms were low-precipitation (LP; Bluestein and Parks; 1983) supercells, which developed along the dryline (e.g., Branick and Doswell 1992; Curran and Rust 1992).
It is important to note that the dividing line, between relatively few +CG dominated severe storms in the east and relatively numerous ones in the west, varied with latitude. South of about 40° latitude, the north–south boundary between relatively frequent and rare +CG dominated severe storms in this study was actually closer to the 96° longitude (i.e., eastern Kansas–Oklahoma–Texas). For example, very few +CG dominated severe storms occurred in Missouri, Arkansas, and Louisiana. On the other hand, there were numerous PPCG lightning producing severe storms to the north of this region in Minnesota and Iowa and a modest number stretching eastward even into Wisconsin.
As shown in Fig. 5b, the percentage of severe storm reports associated with ≥50% +CG lightning was a strong function of latitude and longitude. There is both an east-to-west and a north-to-south gradient in Fig. 5b. A higher percentage of severe storms were dominated by +CG lightning in the Great Plains/Midwest than the East. In the Great Plains/Midwest area, a higher percentage of severe storms produced PPCG lightning in the north than the south. Very few (e.g., about 2%) severe storm reports in the eastern United States were accompanied by PPCG lightning. Similarly, only a small minority (e.g., 11%) of severe storms in the southern plains was associated with PPCG lightning. The Texas–Oklahoma panhandles were an exception in the southern plains region with 15%–40% of severe storm reports accompanied by PPCG lightning. The highest percentage of +CG dominated severe storms occurred in the central and northern plains. In a large portion of these regions, more than half of all severe storm reports were accompanied by PPCG lightning. Throughout the central United States, a southwest-to-northeast orientation in the percentage of severe storm reports accompanied by PPCG lightning is readily apparent (Fig. 5b). Orville and Huffines (2001) and Zajac and Rutledge (2001) also found a southwest-to-northeast-oriented region of elevated positive polarity percentage (≥10%) in their analysis of annual NLDN CG lightning behavior. The maximum in the percentage of severe storm reports associated with PPCG lightning activity was collocated with the minimum in the percentage of severe storm reports accompanied by PNCG lightning (cf. Figs. 4b and 5b in the central and northern plains).
The frequency and percentage of severe storm reports accompanied by a high flash density PPCG lightning event are presented in Figs. 6a,b, respectively. The eastern region was nearly devoid of positive strike dominated severe storms. Most high flash density PPCG lightning events associated with severe weather occurred in the Great Plains/Midwest; the highest frequencies were located in the northern and central plains region (see Table 2). The overall frequency maximum [>6 (10 000 km2)−1 per warm season] was found in western Kansas, the Nebraska panhandle, and eastern South Dakota. A broad maximum extended from the Texas Panhandle and Oklahoma–Texas border through western Kansas and Nebraska and then turn northeastward toward eastern Nebraska and South Dakota and southern Minnesota and North Dakota. A similar southwest-to-northeast tilt occurred across the central United States in the frequency (Fig. 6a) and percentage (Fig. 6b) of severe storm reports associated with high flash density PPCG lightning. The percentage of severe storm reports dominated by high flash density +CG lightning was significantly higher in the northern plains than the central and southern plains (Fig. 6b, Table 2). A large fraction of the northern plains region (i.e., Nebraska, South Dakota, North Dakota, Minnesota) had >20% of severe storm reports associated with a high flash density PPCG lightning event.
b. Mean monthly behavior and the relationship to near-surface θe
Given the regional nature of the relationship between severe storm reports and CG lightning polarity during the warm season, it is important to pose two basic questions: 1) what is the monthly (i.e., temporal) nature of this spatial relationship? and 2) how are these patterns related to the large-scale meteorological environment? To begin addressing these fundamental questions, we chose to extend the results of Smith et al. (2000), who compared the juxtaposition of tornadic storm location, CG lightning polarity, and the surface θe pattern, to climatological temporal (mean monthly 1989–98) and spatial (2.5° × 2.5°) scales.
Six maps of mean monthly near-surface θe during the warm season (April–September) for severe outbreak days containing at least one PPCG HFD event during 1989–98 are presented alongside the frequency of severe storm events associated with PPCG (Figs. 7a–12a, top panels) and PNCG (Figs. 7b–12b, bottom) lightning. As expected, the source of the buoyant, high-θe air originates from the Gulf of Mexico and penetrates into the central United States along the low-level jet (e.g., Bluestein 1993). The monthly position of the ridge of enhanced surface θe (i.e., θe ridge) for severe outbreak days migrated northwestward from the spring (e.g., Figs. 7, 8) to the summer months (e.g., June–July; Figs. 9, 10). Since severe local storms in general are associated with elevated surface θe, frequency maxima of severe storm reports associated with both PPCG (Figs. 7a–10a) and PNCG (Figs. 7b–10b) migrated northwestward along with the θe ridge from spring to summer. However, note that the location of the frequency maxima of severe storm reports associated with PPCG and PNCG lightning were systematically offset for each month.
For each month, the severe local storms associated with PPCG lightning generally occurred to the west through north of the θe ridge in a strong gradient region of θe on the dry (upstream) side of the θe maximum. The absolute frequency maximum of severe storm reports associated with PPCG lightning was often located at the top of the surface θe ridge axis, where high-θe air had made its farthest northwestward progression (note especially Figs. 8a, 9a, and 10a). Severe storms producing PNCG lightning generally occurred southeast of the PPCG maxima, closer to the axis of the θe ridge, and in higher mean values of θe. The absolute frequency maximum of severe storm reports associated with PNCG lightning was systematically displaced southeastward along the axis of the θe ridge, toward higher mean values of θe and away from the high gradient region of θe (note especially Figs. 8a,b, 9a,b, and 10a,b). For each month, the highest percentage of severe storm reports associated with PPCG lightning was consistently located in a region of strong θe gradient to the northwest of the θe ridge axis (Figs. 7a–12a). Not surprisingly, the relationship between surface θe and PPCG/PNCG storm events was still noisy, suggesting that other atmospheric parameters in addition to surface θe are also influencing the location and frequency of severe PPCG storms.
To ascertain the relative significance of the surface θe pattern on PPCG HFD severe outbreak days, we calculated the mean monthly (1989–98) surface θe for all days in which PPCG HFD lightning events did not occur (Figs. 13a–f). There were significant differences in mean monthly (and warm season; not shown) patterns of surface θe between PPCG HFD days and non-PPCG HFD days. In general, the magnitude of the mean surface θe at a given grid location over the central United States was substantially higher (by more than 3° and as much as 22°) for the PPCG HFD severe outbreak days compared to the composite of all other days. For most months, the mean surface θe pattern on non-PPCG HFD lightning days was characterized by either no ridging or a lower-amplitude ridge (i.e., more zonal), when compared to the PPCG HFD lightning event composite (cf. Figs. 7–12 with Figs. 13a–f). Interestingly, the differences in surface θe pattern and magnitude were most dramatic for those months that had the clearest relationship between θe and CG lightning polarity in severe storms (e.g., April, May, June, and September in that order as seen in Figs. 7a,b–9a,b, 12a,b). The differences were more muted for the midsummer months (i.e., July and August as seen in Figs. 10a,b and 11a,b), which were also characterized by more noisy relationships between surface θe and CG lightning polarity. This is likely further evidence that the surface θe pattern had either only an indirect control on storm kinematics, microphysics, and lightning behavior or that it had a direct role but was only one of several possible environmental controls.
c. Relationship between CG lightning polarity and storm severity
Since our 10-yr sample of reports includes a wide range of storm severity (i.e., about 2–18-cm-diameter hail and F0–F5-rated tornadoes), we have explored the relationship between storm severity and the mean percentage of positive polarity CG lightning for both hail (Fig. 14a) and tornado (Fig. 14b) reports as a function of geographic region. We excluded severity categories for a given region that contained less than 25 samples.
There was a positive trend between mean hail size and mean positive CG percentage up to 8 cm for the entire contiguous United States (“ALL” in Fig. 14a). The mean percentage of positive CG lightning increased by a factor of 1.7 as hail diameters increase from 2 to 8 cm (Fig. 14a). At about 8 cm, the mean trend became flat to slightly decreasing with hail size. The relationship between mean hail size (up to 8 cm) and mean positive CG percentage was about the same for all four subregions. Since the sample size decreased as a function of hail size, it is possible that the flattening to slight decreasing trend at diameters beyond 8 cm is the result of an insufficient sample size. For the whole contiguous United States, there were only 390 hail reports for D > 8 cm (i.e., about 0.7% of the total severe hail sample size).
The relationship between mean tornado F scale and mean positive CG lightning percentage was less clear. For all regions except the northern plains (NP), the trend in positive CG lightning percentage was flat from F0 to F2 (Fig. 14b). From F2 to F3, the mean positive CG lightning percentage increased substantially (10%) in all geographic regions. However, the mean positive CG percentage for all F4 tornadoes (58 sampled) actually decreased from the F3 value by about 10%. Although not shown, the mean percentage of positive CG lightning for the eight sampled F5 tornadoes across the contiguous United States during 1989–98 was 20% higher than the corresponding value for F4 tornadoes. Given the small sample size for violent tornadoes (F4, F5) during 1989–98, it was not possible to determine conclusively the actual trend. However, the upward tendency in the mean positive CG percentage from F2 to F3 damage rating appears to be real and significant.
As could be anticipated by the earlier results, perhaps the most obvious trend in Figs. 14a,b is the systematic regional variation of the mean positive CG lightning percentage associated with storm reports regardless of severity. The percentage of positive CG lightning consistently ranked from highest to lowest in all severity categories as follows: northern plains, central plains, southern plains, and eastern United States. Regardless of any mean trend between severe storm severity and the percentage of positive CG lightning, regional differences dominated.
The mean trends discussed above mask an important characteristic of the percentage of positive CG lightning as a function of hail size and F scale. As shown in Figs. 15a,b, the joint probability density functions of percent positive lightning versus hail size and F scale were bimodal for the severe storms characterized by hail diameter ≥6 cm and F scale ≥F3. The dominant population is characterized by <25% positive CG lightning. A second population is characterized by >50% positive CG lightning. As a result, the generally increasing trend of percent positive CG lightning versus hail size and especially F scale depends strongly on the emergence of the second population of reports associated with >50% positive CG lightning for ≥6 cm hail and F3 tornadoes. Furthermore, the emergence of this second population at high positive CG percentages was largely confined to the northern and central plains and to a lesser extent the southern plains. So although there were some positive trends between storm severity and positive CG percentage, the scatter in the raw data was large, actually bimodal and largely controlled by region.
The polarity of cloud-to-ground lightning produced by severe storms is a function of both season (or month) and geographic location. Based on these results, PPCG lightning beneath severe storms is (is not) atypical beneath severe storms in the eastern (northern and central plains) region of the United States during the warm season. Similarly, severe storms producing PPCG lightning are not (are) uncommon in the southern plains during the spring (summer) months.
This geographic and monthly behavior of CG lightning polarity in severe storms is systematically correlated to the location of the near-surface ridge of θe. As in the Smith et al. (2000) study of three tornadic outbreaks, there was a strong preference for severe storms producing PPCG lightning in our climatological study to occur upstream (west to northwest) of the θe maximum in a high-θe gradient region. Interestingly, each monthly frequency maximum for severe storms producing PPCG lightning crowned the northern and western extent of the surface θe ridge. Severe storms producing PNCG lightning were systematically displaced southeastward of severe storms generating PPCG lightning along the θe ridge axis, in a low-θe gradient region and in higher mean values of θe. When this monthly behavior is averaged over the warm season, the familiar southwest-to-northeast tilted region of PPCG lightning behavior emerges.
When applied to the climatological pattern of surface θe as in our study, the conceptual model of Smith et al. (2000) correctly predicts this southwest-to-northeast tilt of PPCG lightning activity. Storms generally move from west to east and the source of high-θe air is advected northward from the south (i.e., Gulf of Mexico) along the low-level jet. In the Smith et al. model, storms initially producing PPCG lightning that also have a southerly component to their motion would eventually move into higher mean values (and lower gradient) of surface θe and would become PNCG lightning storms. On the other hand, storms that formed near the crown of the θe ridge and move northeastward (i.e., nearly parallel to θe surfaces and within a high gradient region of θe) would continue to produce PPCG lightning. This fundamental difference in CG lightning polarity behavior relative to the climatological position of the surface θe ridge may explain the familiar southwest-to-northeast tilted region in the central to northern plains that encompasses the majority of PPCG lightning storms.
Although our observations generally confirm the Smith et al. (2000) conceptual model, there were subtle differences between the location of our climatological PNCG lightning maximum versus θe ridge and the Smith et al. (2000) conceptual model. The center of our PNCG lightning maximum appears to be shifted upstream of the corresponding PNCG lightning location in Smith et al. (2000). Differences in methodology and data resolution may account for this discrepancy. There are other possible explanations. Our sample included only CG lightning immediately in the vicinity (i.e., 1 h total and 50 km) of severe weather (large hail and tornadoes). Smith et al. (2000) presented cumulative CG lightning activity for long periods of time and large geographic areas that each included a tornado outbreak of interest. In Smith et al., it appears that all CG lightning was included, regardless of its association with severe weather. Because of the mesoscale ingredients required for severe weather [e.g., high CAPE, midlevel dry air, low-level convective inhibition (CIN), large low to midlevel shear], we believe it is consistent for severe weather in general to be biased to the upstream side of θe ridges. Hence, our PNCG lightning maxima would be shifted relative to the negative lightning in Smith et al. (2000). Some examples in Smith et al. (2000), such as their Fig. 2, match their conceptual model well, with most negative lightning occurring downstream of the θe ridge. Other data, such as their Fig. 7, match it less well. The latter example clearly shows negative CG lightning occurring just upstream, in, and downstream of the θe ridge. Since Smith et al. (2000) only show data from three tornadic outbreak cases using synoptic data, it is possible that the finescale behavior of CG polarity relative to the θe ridge during severe weather is still poorly defined. Our sample has significantly more cases than Smith et al. (2000) but suffers from poor resolution. Mesoscale analysis of a large number of cases is required to further validate the Smith et al. (2000) conceptual model.
The physical relationship (direct or indirect) between the surface θe ridge and CG lightning polarity is unknown. Possible climatological influences on CG flash polarity may be tied to preferred spatial patterns of storm-scale charge generation and structure. Both Smith et al. (2000) and Gilmore and Wicker (2002) argue that storms moving through a high-θe gradient region toward the maximum would experience a rapid acceleration in their updrafts. The accelerating updrafts in combination with increasing values of near-surface water vapor mixing ratio would result in higher values of supercooled cloud liquid water aloft. According to the noninductive charging theory (e.g., Saunders et al. 1991), higher values of cloud liquid water could result in the enhancement of the lower-level positive charge region by allowing the transfer of positive charge to hail over a deeper region of the storm (i.e., the charge reversal temperature decreases).
Assuming hail size can be used as a rough proxy for updraft strength (e.g., Williams 2001), our results confirm that increasing updraft strength (i.e., hail size) was associated with higher mean values of positive CG lightning percentage. However, this increase was associated with the emergence of a bimodal distribution of positive CG percentage in the northern and central plains. Large hail was often associated with <25% positive CG lightning. If hail size is even a crude proxy for updraft strength, then some other characteristic associated with the surface θe ridge must be invoked to explain the observed variance in CG lightning polarity in severe storms.
Other possible effects of the surface θe ridge are 1) the entrainment of dry air aloft in θe gradients and the influence on cloud particle distributions and charge transfer (e.g., Avila and Pereyra 2000); 2) variability in cloud-base height and associated impact on the droplet size distribution and charge transfer; 3) the effect of θe (buoyancy) gradients on storm location, dynamics, updraft strength, microphysics, and hence charging (e.g., Rasmussen et al. 2000; Gilmore and Wicker 2002); and 4) the role of aerosol convergence along boundaries that may also affect cloud particle distributions and charge transfer (e.g., Rosenfeld and Lensky 1998). These are topics to be addressed in furture research.
Interestingly, the southwest-to-northeast-oriented PPCG lightning feature is remarkably similar in shape and location to the region of a high percentage (say ≥10%) of positive CG lightning found in the Great Plains and Midwest region when analyzing annual NLDN data from 1989 to 1998 (Orville and Huffines 2001). This feature also resembles the regional distribution of large peak current positive CG lightning flashes noted by Lyons et al. (1998). As noted by Boccippio et al. (2001), this PPCG lightning feature is also collocated with a climatological maximum in the in-cloud to cloud-to-ground lightning ratio (IC/CG). The correspondence between high IC/CG and elevated positive CG lightning percentage has been identified in individual case studies of severe storms in this region (e.g., Carey and Rutledge 1998; Lang and Rutledge 2002).
At this time, it is not clear whether severe PPCG lightning storms are largely responsible for this well-known annual maximum in the percentage of positive CG lightning or whether nonsevere storms also contribute significantly to this feature. Since uniquely identifying each CG lightning flash with a severe or nonsevere storm is problematic, it is difficult to partition positive CG lightning in this fashion. Nonetheless, we found that 30%–40% of all warm season positive CG lightning over this region was contained within our severe storm analysis rings defined above. Since many severe storms were likely not reported and since portions of many severe storms were not included in our analysis rings, this percentage is likely an underestimate of the actual contribution of severe storms to the production of positive CG lightning in the central and northern plains. As a result, we suggest that severe storms play a significant, if not dominant, role in the creation of this southwest-to-northeast tilted maximum of positive CG lightning percentage in the central and northern plains. This speculation is supported by the analysis in Zajac and Rutledge (2001) suggesting that PPCG lighting storms were produced primarily during the summer in the hours around sunset by isolated storms and convective lines that had not yet fully developed into mesoscale convective systems (MCSs).
Having analyzed 10 yr (1989–98) of severe storm reports (large hail and tornadoes) and NLDN cloud-to-ground lightning measurements during the warm season, the following conclusions can be made.
As in most typical warm season thunderstorms, a large majority (61%) of severe local storms in the contiguous United States produced predominantly (>90%) negative cloud-to-ground (PNCG) lightning.
Severe storms that produced predominantly (>50%) positive cloud-to-ground (PPCG) lightning represented a small minority (15%) of all severe storms across the contiguous United States. Severe storms that produced high flash density (HFD, >0.01 km−2 h−1) PPCG lightning (e.g., Stolzenburg 1994; MacGorman and Burgess 1994) were even more infrequent (i.e., only 5% of all severe storms).
There was significant regional variability in the percentage of positive CG lightning produced by severe storms during the warm season. In the eastern United States, an overwhelming percentage (81%) of severe storms produced PNCG lightning while only 2% of severe storms produced PPCG lightning. The CG lightning behavior was quite different over the northern plains; only 28% of severe storms there generated PNCG lightning while 43% of them produced PPCG lightning.
Throughout the north-central United States, a southwest-to-northeast tilted frequency and percentage maxima of severe storms accompanied by PPCG lightning was readily apparent (Fig. 5b). The maximum in the percentage of severe storms reports associated with PPCG lightning activity was collocated with the minimum in the percentage of severe storm reports accompanied by PNCG lightning.
Severe storms that produced PPCG and PNCG lightning migrated northwestward from spring to midsummer months along with the climatological position of the surface θe ridge.
Our climatological results were generally consistent with the empirical model of Smith et al. (2000) relating the pattern of surface θe to CG lightning polarity beneath severe storms. The monthly frequency maxima of severe storms that produced PPCG and PNCG lightning were systematically offset with respect to the surface θe ridge. PPCG lightning associated with severe thunderstorms generally occurred west and northwest of the θe ridge in either a θe gradient region along what may have been a dryline, or along the top of the θe ridge in regions that were still under the influence of polar fronts during the warm season (e.g., most obvious in midsummer months). PNCG lightning associated with severe thunderstorms were located southeast of the PPCG lightning maxima, closer to the axis of the θe ridge in higher mean values of θe.
Across the contiguous United States, the percentage of positive CG lightning increased noticeably as hail size increased from 2 to 8 cm and as tornado damage increased from F2 to F3. However, this increase was largely due to the emergence of a bimodal distribution in the percentage of positive CG lightning associated with severe storms for very large hail and intense tornadoes [i.e., a primary (secondary) maximum occurred at <25% (>50%) positive CG lightning]. Furthermore, this secondary maximum was evident primarily in the northern and central plains.
Storm severity, as judged by hail size and tornado F scale, explained only a small portion of the variance in the percentage of positive CG lightning. Regional differences in CG lightning polarity, regardless of storm severity, were dominant.
We gratefully acknowledge the NASA Lightning Imaging Sensor (LIS) instrument team and the LIS data center via the Global Hydrology Resource Center (GHRC) located at the Global Hydrology and Climate Center (GHCC) in Huntsville, Alabama, for providing the NLDN lightning data through a license agreement with Global Atmospherics, Inc. The NLDN data from the GHRC are restricted to LIS science team collaborators and to NASA EOS and TRMM science team members (S. A. Rutledge). NWS/SPC and NCDC provided the tornado and large hail historical archive. NCEP reanalysis data were obtained from the NOAA/Climate Diagnostics Center. We are grateful to Drs. Donald MacGorman, Stephan Smith, Earle Williams, Daniel Rosenfeld, and Mr. Bard Zajac for their insightful comments and helpful suggestions on this research. We gratefully acknowledge the National Science Foundation (NSF) for supporting this research under Grants ATM-9726464 and ATM-9912051 (S. A. Rutledge, W. A. Petersen, and also L. D. Carey while conducting a significant portion of the research at Colorado State University). Partial support for L. D. Carey was also provided by North Carolina State University.
Corresponding author address: Dr. Lawrence D. Carey, Dept. of Marine, Earth and Atmospheric Sciences, North Carolina State University, Campus Box 8208, Raleigh, NC 27695. Email: email@example.com
The temporal and spatial clustering radii tested were 0.25, 0.5, and 0.75 h and 25, 50, and 75 km. The choice of clustering radii had little impact on the overall conclusions of the study regarding the geographical and monthly pattern of severe storms producing PPCG and PNCG lightning and their relationship to patterns of surface θe.
The geographic distribution of severe storm reports is likely modulated by population density and other nonmeteorological effects (e.g., Kelly et al. 1978, 1985). We describe the pattern in order to provide a basis for understanding the geographic distribution of CG lightning polarity associated with these storm reports.
The Severe Thunderstorm Electrification and Precipitation Study (STEPS, May–July 2000) investigated the dynamics, microphysics, and electrical structure of severe thunderstorms in the positive anomaly region near Goodland, Kansas. More details can be found on the Internet (http://www.mmm.ucar.edu/community/steps.html).