Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 Lightning flashes (N = 1088) associated with 24 thundersnow events in the central United States were analyzed to document flash polarity, signal strength, and multiplicity. Negative lightning flashes (N = 872; 80%) dominated positive flashes (N = 216; 20%) with wintry precipitation in this study, which stands in contrast to the majority of the research done on winter thunderstorms (primarily in Japan). Otherwise, limited work has been done, although thundersnow has been documented in the mid-latitudes of North America, Europe and Asia. Statistics on peak amplitude were determined for negative (positive) flashes, yielding mean and standard deviation values of −24 kA ± 22 kA (+38 kA ± 34 kA). A subset of winter lightning events (N = 16) were then sought that occurred with banded (single or multiple) snowfall, as banding often denotes greater organization in the atmosphere (e.g., a jet streak aloft to aid in ascent, or a low level jet streak to aid with moisture and thermal transport) and thus the potential for deeper snow totals. Radar reflectivity values were recorded at the location of each lightning flash, as well as the maximum radar reflectivity within the associated snow band. The location of the lightning activity within the snow band was also noted as being either leading edge (LE), trailing edge (TE), core (C), or not correlated (NC), with respect to the motion of the parent band. The majority of lightning flashes were found downstream of areas of highest radar reflectivity with respect to the motion of the snow bands, and not with the highest reflectivity values. If one uses the highest reflectivity values in a snowband as a proxy for the greatest surface snowfall intensity, then the ground terminus of a cloud-to-ground lightning (CG) flash is often not co-located with the heaviest snowfall rates. However, the work completed here does place the location of the typical CG flash ∼15 km downstream of the snowband location, so one could use the occurrence of lightning as a nowcasting tool for impending snow intensification at the site of the CG flash.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Anecdotal evidence exists to suggest that lightning activity within winter storms should be associated with elevated snowfall rates. Recent work by Crowe et al.  has established that heavy snowfall totals (over the life cycle of an entire storm) in non-lake effect, synoptic-scale systems are often observed in the vicinity of where lightning had been observed. Furthermore, it seems intuitive that thundersnow (very likely a convective phenomenon) should be accompanied by higher precipitation rates than nearby, surrounding areas. Yet, no previous study has quantified these assertions.
 Several issues are addressed with this work. First, the basic characteristics (number, polarity, etc.) of lightning flashes associated with snowfall in the central United States were examined. Second, those lightning flashes that also occurred with respect to an organized snowband (a radar signature of enhanced reflectivity, perhaps within a larger precipitation area, having an ellipticity of 2:1 or greater and a duration of ≥30 minutes for the 25 dBZ region) were documented. Here, the radar reflectivity associated with the lightning flash was recorded as well as the regional maximum radar reflectivity (which was not always the same location). In this way, expectations about lightning and its correlation to higher snowfall rates could be quantified, validated, or refuted.
 Two primary sources of data drove this study. Data (location, polarity, amplitude) for cloud-to-ground lightning flashes were obtained from the National Lightning Detection Network (NLDN) [e.g., Cummins et al., 1998]). Estimates of snowfall intensity were derived from Level III radar base reflectivity data from the Weather Surveillance Radar - 1988, Doppler (WSR-88D) radar network (essentially a mosaic of the 0.5° angle scans). These radar data have been averaged spatially to regions of 1 km2 and the reflectivity binned every 5 dBZ. Temporally, NLDN data are available down to the second (and even to the millisecond), while a new WSR-88D data scan begins every ∼6 minutes. Thus, the lightning flashes were matched to the time period of each radar volume scan. As a matter of quality control, only flashes in a broad ring between 20 km and 70 km from each specific WSR-88D site were retained in this portion of the analysis. This approach eliminates potential ground clutter problems and examines the reflectivity near the ground (200 m above ground level [AGL] at a range of 20 km to 1000 m AGL at 70 km) before the curvature of the radar beam allows it to ascend too far above the earth's surface (e.g., false returns and aliasing among others; see Reinhart , especially chapter 6, for a comprehensive account).
 Animated loops of radar base reflectivity in precipitation mode were examined to determine band existence, orientation, and direction of motion. Distances were then calculated from each lightning flash to 1) the maximum reflectivity in the band, and 2) the center of the precipitation band (where the heaviest snowfall is assumed to exist at the ground), perpendicular to the band orientation (Figure 1a). Given that the band dimensions often exceeded the resolution of the raw radar beam, and the 1 km2 resolution of the Level III product, we have high confidence in the identification and sampling of most of the bands in this study.
 The definition of a snowband used for this work is necessary at this point. From the work of Novak et al.  the snowband definition had specific criteria: for single bands, >250 km in length, 20–100 km in width, and a region of 30 dBZ base reflectivity maintained for ≥2 hr; for multiple bands, at least 3 were required to be present with periodic spacing, widths of 5–20 km, and a region of >10 dBZ above the background maintained for ≥2 hr. However, the present research, having been driven first by the presence of lightning and secondly by banding, often included snowbands of lesser dimensions. As mentioned previously, an ellipticity of 2:1 (or greater) and a duration of ≥30 minutes for the 25 dBZ region were used (similar to Novak et al.  but lowered slightly to include more bands in the study). Indeed, length and width were based on the area of the 25 dBZ reflectivity. Moreover, a minimum ellipticity of 2:1 was used to ensure a band-like structure in the shortest of the bands examined. By using this approach we were not limited to only the largest and/or most intense bands, but were free to examine the lightning behavior associated with the spectrum of band sizes (which ranged below the thresholds imposed by Novak et al. ). Finally, these snowbands are associated with transient, synoptic-scale, extratropical cyclones, and not with lake-effect snow events or instances of orographic lifting. Details on snowband dimensions are given in the ensuing section.
 Finally, four categories were established, which indicated where each lightning flash was located within the snowband: leading edge, trailing edge, core, and not correlated. This assessment was done in two different ways. First, the area of reflectivity ≥25 dBZ in each snow band was divided geographically into thirds (understanding that each individual band may have a different width), with respect to the snowband motion. Leading edge (LE) referred to a lightning flash that was located in the front 1/3 portion of the band. Trailing edge (TE) referred to a lightning flash that was located in the rear 1/3 of the snowband. Core (C) referred to a lightning flash that was located inside the inner 1/3 of the snowband (Figure 1b). Not correlated (NC) indicated that the lightning flash was not easily correlated to any of these areas; these were isolated snowbands where lightning occurred near the band, but away from measurable radar reflectivity (<5 dBZ). Secondly, each flash was assessed with regard to the very highest reflectivity in the band. In this instance, leading edge (LE) referred to a lightning flash that was located ahead of the area of strongest radar reflectivity. Trailing edge (TE) referred to a lightning flash that was located behind the area of strongest radar reflectivity. Core (C) referred to a lightning flash that was located with the area of strongest radar reflectivity. As with the assessment of where each lightning flash was located within the snowband, not correlated (NC) indicated that the lightning flash was not easily correlated to any of these areas (Figure 1c).
 1088 flashes were examined from 24 different cases with banded snowfall in the central United States (Table 1). Most of these cases came from the winter seasons (October–April) of 2004–2005 and 2005–2006. Some events produced only one flash in the winter precipitation, while the event with the most occurred on 01 January 2005 when 295 flashes occurred with snowfall over several hours.
Table 1. Date, Location, and Time of Each Snow Band Studieda
Date is in MM/DD/YYYY format, location is by United States Postal Service state abbreviation, and time is Universal Coordinated Time. The final column shows the number of lightning flashes observed. The dates for those cases further selected for banding analysis are denoted by an asterisk after the date.
 Negative flashes dominated the cases studied here, with 872 (80%) of the total. These flashes had a mean peak amplitude of −24 kA (s.d. ±22 kA). Positive flashes comprised 216 (20%) of the total flash count in the study with a mean peak amplitude of +38 kA (s.d. ±34 kA). That negative flashes dominated in these cases stands in contrast to the bulk of the previous work on lightning with winter precipitation (typically snow), most of which was conducted in Japan [e.g., Takeuti et al., 1978; Suzuki, 1992; Takahashi et al., 1999] but with similar, albeit limited, results from two cases in the central United States [Holle and Watson, 1996]. However, a seasonal peak is noted in February for the positive flashes, a result that does correspond to the conclusions of Orville and Huffines . One possible explanation is that a comparison is being made between a coastal/maritime setting versus a continental one. As such, there are large and fundamental variations in nuclei and aerosol contributions that could be relevant here. Of course, this would also be true of an urban region versus a rural or suburban area, for which no control exists in this study. Yet, no data were collected with regard to sources, types, properties, and/or characteristics of the constituent nuclei or aerosols. Thus, the relation of the resulting cloud/hydrometeor processes (and populations) and electrification is only speculation.
 Of the 24 cases discussed previously, a subset of 16 cases (906 flashes) had radar data and/or sufficiently organized banding structure in the snowfield to extend the analysis. As stated previously, the band length and width values that follow are based upon the areal extent of the 25 dBZ region. The bands in this study had a mean length of 198 km (s.d. ±158 km). The median band length was 168 km with a minimum of 5 km, and a maximum of 878 km. The mean width of banded snow structures in this study was 61 km (s.d. ±54 km) with a median of 35 km, and a minimum and maximum width of 3 km and 212 km, respectively. The typical range of length-to-width ratios are well-represented by the mean (3:1) with longer bands better represented by the median (5:1). A graphical summary is also provided of band length (Figure 2a) and band width (Figure 2b) measured with each lightning flash. Clearly, there were many shorter, more narrow snowbands studied in this work.
 Lightning flashes in these banded events failed to be associated with the highest reflectivity a majority of the time. Indeed, lightning flashes were found 15 km (s.d. ±17 km) from the center of the snowband along a line perpendicular to the major band axis (Figure 1a). Clearly, the standard deviation suggests some flashes with a distance so great as to question the relationship between the lightning flash and the band identified with it. Indeed, there is often significant spatial separation between the location of lightning flashes and the center of the parent snowband. Yet, the median value for this distance is 8 km, suggesting a stronger link between flashes and their parent snowbands in the majority of cases.
 Of perhaps greater interest is the spatial relation between the lightning location and that of the highest reflectivity in the parent snowband. Although the distance values do increase, the change is minor. In this case, the mean distance between each cloud-to-ground flash and the highest reflectivity is 17 km (s.d. ±25 km) with a median distance of 9 km. The standard deviation is of particular interest here, as it does suggest great distances between flash locations and the maximum band reflectivity that are found in some snow bands. In any event, the reflectivity data also suggest a significant separation. The mean (median) radar reflectivity associated with the geographic location of the cloud-to-ground flash terminus was 29 dBZ (30 dBZ), while the mean (median) of the maximum values in the band core was 43 dBZ (45 dBZ). Thus, the relationship between lightning flashes and the maximum snowfall rate does not appear to be a strong one. Yet, the recent emergence of lightning mapping arrays [e.g., Krehbiel et al., 2000; Goodman et al., 2005; Riousset et al., 2007], which yield high-resolution, three-dimensional information on the morphology of individual lightning flashes holds the promise of refining these findings.
 However, an examination of the location of lightning flashes with respect to the band and its motion shows that the anecdotal does occur, but it is the exception and not the rule. Figure 3a shows the spatial distribution of lightning flashes with respect to their parent snowband as a whole. While about one third of the flashes studied occurred in proximity to the center, the majority did not, with a fairly even distribution on both the leading edge and the trailing edge of the band. When compared to the very highest reflectivity in the band, more flashes were then found to occur downstream of the band core, toward the leading edge (Figure 3b). These data do show that lightning flashes can and do occur with the highest reflectivity (and presumably the highest surface snowfall rates, snow drift between the cloud base and its surface deposition having been neglected here), although that is more the exception and not the rule. Moreover, the very highest radar reflectivities occur upstream of nearly two thirds of the lightning flashes in the bands studied.
 The basic characteristics of lightning were examined with snow and found to be primarily of negative polarity. This finding stands in contrast to the bulk of the previous work on winter lightning (see the survey chapter of Rakov and Uman  and Holle and Watson ) that suggests that positive flashes should be dominant, but is consistent with the recent study of many cases of lightning with snow in the central United States [Pettegrew et al., 2008]. Also, this analysis has established that, while lightning flashes can and do occur concurrently with the highest reflectivities, such a 1:1 correlation only happened in this dataset about 6% of the time. Yet, the mean (median) distance from the typical lightning flash to the maximum reflectivity in a snowband is 17 km (9 km). Thus, we see that a pattern similar to that of Crowe et al.  for lightning activity with respect to storm total snowfall amounts, where the lightning is frequently found in the vicinity of the area of highest instantaneous snowfall.
 This work is supported by the National Science Foundation (NSF), award ATM-0239010. Any opinions, findings, conclusions or recommendations expressed herein are those of the authors and do not necessarily reflect the views of the NSF. The authors would also like to thank Ronald Holle, Anthony Lupo, and Neil Fox for many helpful discussions about this work. Scott Rochette also has our gratitude for reading initial drafts of this manuscript. The constructive exchanges with two anonymous reviewers were especially helpful in improving the final manuscript.