Tornadoes strike Liberty and Moore,
South Carolina, and near Gastonia,
North Carolina, on 5 January 2007
Patrick D. Moore
Vehicles damaged by the tornado that struck Liberty, South Carolina, on 5 January 2007. Image courtesy of and copywright by The Pickens Sentinel, used by permission.
Author's Note: The following report has not been subjected to the scientific peer review process.
A line of strong to severe thunderstorms produced at least three confirmed tornadoes, and several more reports of damaging wind gusts, across upstate South Carolina and the Charlotte metro area during the afternoon of Friday, 5 January 2007. The National Weather Service Office in Greer, South Carolina (GSP), issued five tornado warnings and 14 severe thunderstorm warnings between noon and 6:00 pm on that day. The first tornado touched down briefly at 2:24 pm (1924 UTC) in Liberty, South Carolina, on the campus of Liberty Elementary School, tossing about several vehicles in the parking lot waiting for afternoon dismissal. The second tornado struck an area northwest of Moore, South Carolina, around 3:10 pm (2010 UTC), destroying two sheds, damaging the roof of a mobile home, and snapping several large pine trees. The third tornado touched down briefly at 4:45 pm (2145 UTC) near Gastonia, North Carolina, damaging several roofs in the Autumn Acres subdivision. [All times in this document from this point onward are referred to in Universal Time Coordinated (UTC), which is Eastern Standard Time plus five hours.] The Liberty Tornado was rated F1, while the Moore Tornado and the Gastonia Tornado were rated F0 on the Fujita Scale.
(Click here to view a summary of severe weather reports for 5 January 2007.)
Figure 1. Wind damage, large hail, and tornado reports for 5 January 2007. Click on image to enlarge.
The meteorology leading up to the events on 5 January 2007 was well understood and well anticipated by the GSP Weather Forecast Office and the Storm Prediction Center (SPC). While the storm that produced the Moore Tornado appeared to be a classic example of a "broken-S" type of quasi-linear convective system (McAvoy et al. 2000), the storm that produced the Liberty Tornado had more subtle features which made the warning process particularly challenging. The Gastonia storm showed few signs that a tornado was imminent, at least not according to our current understanding of tornadogenesis in high shear and weak instability environments.
2. Synoptic Features and Pre-Storm Environment
A dynamic closed upper low was located over the Mississippi Delta region on the 500 mb analysis at 1200 UTC on 5 January (Fig. 2), and could also be seen on the GOES-12 water vapor imagery. Although the upper low was expected to open and deamplify during the day, strong forcing at mid- levels was expected to continue with an 80-100 kt jet streak wrapping around the forward edge of the short wave trough. Additional forcing aloft was anticipated by midday with the arrival of the left exit region of a jet streak at 300 mb, seen moving northward from the Gulf of Mexico at 1200 UTC (Fig. 3). In fact, a linear convective system had already developed by 1200 UTC across eastern Alabama in this strongly forced environment.
Figure 2. SPC objective analysis of 500 mb geopotential height, temperature, and wind at 1200 UTC 5 January. Click on image to enlarge.
Figure 3. SPC objective analysis of 300 mb isotachs, streamlines, and wind divergence at 1200 UTC 5 January. Click on image to enlarge.
A southerly low level jet of 40-45 kt at 850 mb (Fig 4) and winds of 35-40 kt at 925 mb across Georgia and the western Carolinas were favorable for a continuation of organized severe thunderstorms. The environment ahead of the squall line, as sampled by the upper air sounding at Peachtree City, Georgia (FFC), at 1200 UTC (Fig. 5), was characterized by strong deep layer shear on the order of 60 kt, surface to 1 km shear of 30-35 kt, and surface to 1 km storm relative helicity on the order of 200 m2/s2 were all favorable for the development of tornadoes. In fact, the upper air sounding taken at Greensboro, North Carolina (GSO), at 1200 UTC (Fig. 6) showed a surface to 3 km shear of 40 kt, which is known to be conducive to the formation of tornadoes in quasi-linear convective systems. In spite of extensive low clouds, daytime heating was expected to raise Convective Available Potential Energy to around 800 J/kg ahead of the line. This measure of updraft potential, combined with the strong shear, would be sufficient to maintain the squall line as it moved east. For that reason, the SPC placed the area roughly to the south of a line from Athens, Georgia, to McCormick, South Carolina, to Wadesboro, North Carolina, in a Slight Risk in the Day 1 Convective Outlook issued at 1229 UTC.
Figure 4. SPC objective analysis of 850 mb geopotential height, temperature, dewpoint, and wind barbs at 1200 UTC 5 January. Click on image to enlarge.
Figure 5. Skew-T log P diagram (upper left) and hodograph (upper right) for upper air sounding at FFC at 1200 UTC 5 January. The tables at the bottom summarize several objective parameters used by the SPC to determine severe weather potential. Click on image to enlarge.
Figure 6. As in Figure 5, except at GSO. Click on image to enlarge.
At the surface, a cold front stretched along the Mississippi-Alabama border (Fig. 7), with the air mass ahead of the squall line characterized by temperatures in the middle to upper 60s with dewpoints in the lower to middle 60s. Regional surface analyses at 1200 UTC and 1300 UTC indicated the presence of a weak warm front lifting north across extreme northeast Georgia and the western part of upstate South Carolina, with dewpoints behind the front climbing into the lower 60s and surface winds veering to the southeast (Fig. 8). It was quickly surmised that the environment across the Lower Piedmont and the Lakelands would be at least as favorable as that across the central Savannah River valley and the Midlands of South Carolina. As as result, a Severe Weather Outlook was issued at 1339 UTC for the area south of the North Carolina border, mentioning the possibility of damaging wind gusts and the potential for a brief, isolated tornado.
Figure 7. Hydrometeorological Prediction Center surface fronts and pressure analysis at 1200 UTC 5 January. Click on image to enlarge.
Figure 8. Regional surface observations plots at 1200 UTC (left) and 1300 UTC (right), 5 January. Note how the winds veer at AND and GSP and the increasing dewpoint at AND as the warm front passes between 1200 UTC and 1300 UTC. Click on images to enlarge.
The pre-storm environment was sufficient to re-intensify the northern end of the squall line as it crossed the Atlanta metro area through the middle part of the morning. The SPC adjusted the Slight Risk area northward to include the area generally along and south of Interstate 85 on the updated Day 1 Convective Outlook issued at 1629 UTC. The Severe Weather Outlook was updated at 1647 UTC to follow suit, and continued to mention the potential for brief, isolated tornadoes. As the squall line approached extreme northeast Georgia, the air mass to the south of the North Carolina border was weakly unstable with a CAPE of 300-400 J/kg, but strongly sheared with surface to 1-km shear values of 30-35 kt, as noted by a Mesoscale Discussion issued by the SPC at 1706 UTC. After coordination with the SPC, a Tornado Watch was issued at 1740 UTC for most of northeast Georgia, upstate South Carolina, and the Charlotte metro area. By that time, the leading edge of the squall line was poised to enter the WFO GSP County Warning Area.
3. Radar observations
The squall line (otherwise known as a quasi-linear convective system, or QLCS for short) had a history of producing strong wind gusts as it moved across north central Georgia, including a gust of 54 mph in Gainesville (Hall County). The GSP Weather Forecast Office issued a Severe Thunderstorm Warning for Rabun, Habersham, Stephens, and Franklin counties in Georgia, and for Oconee County in South Carolina, at 1737 UTC as the leading edge of the line was poised to move into Habersham County (Fig. 9). Wind damage was reported across southern Habersham County, Stephens County, and western Franklin County as the line of thunderstorms passed. Additional damage was reported in western Elbert County near Bowman.
Figure 9. Radar reflectivity on 0.5 degree scan from the KGSP WSR88-D radar at 1738 UTC. The radar is located at the upper right edge of the image. Click on image to enlarge.
a. The Liberty Tornado
The leading edge of the QLCS stretched from central Oconee County just west of Seneca, south across the extreme western tip of Anderson County, to Hartwell, Georgia, at 1839 UTC (Fig. 10). A notch of lower radar reflectivity developed in the rear flank of the QLCS near Fair Play at a height of 4.5 km above ground level between 1839 UTC and 1843 UTC, seen on the 3.0 degree scan (Fig. 11). A relative maximum in inbound radial velocity (i.e. toward the radar) of 50 knots or greater was observed within the reflectivity notch. This indicated the presence of a rear inflow jet. Over the next four minutes, a similar low reflectivity notch developed at the back edge of the QLCS on the 0.5 degree scan. This was seen as an indentation of lower reflectivity at a height of approximately 1 km above ground level near Townville (Fig. 12). By 1900 UTC, the notch was located along the Pickens-Anderson county line to the south of Clemson (Fig. 13). The radial velocity increased from at least 26 kt to greater than 36 kt, indicative of strengthening rear-to-front flow. The appearance of the notch on the 3.0 degree scan four to eight minutes prior to its appearance at the lowest elevation scan strongly suggests a subsident component to the rear-to-front flow in the storm.
Figure 10. Radar reflectivity on the 0.5 degree scan from the KGSP radar at 1839 UTC. The radar is located in the upper right corner to the right of the last 'e' in Greenville. Click on image to enlarge.
Figure 11. Radar reflectivity (top) and radial velocity (bottom) on the 3.0 degree scan from the KGSP radar at 1843 UTC. Negative values (green shades) indicate motion toward the radar and positive values (red shades) indicate motion away from the radar. The arrow points to the notch of lower reflectivity at the back edge of the QLCS corresponding to higher inbound velocity near the town of Fair Play (point FP). Note the red colors near the tip of the arrow in the bottom figure represent velocities that have been improperly dealiased. The actual velocity is a value greater than 50 kt toward the radar. Click on images to enlarge.
Figure 12. As in Fig. 10, except for 1847 UTC. The arrow denotes the rear inflow notch. Click on image to enlarge.
Figure 13. Radar reflectivity (top) and radial velocity (bottom) on the 0.5 degree scan from the KGSP radar at 1900 UTC. The arrow points to the notch of lower reflectivity at the back edge of the QLCS south of Clemson. Click on images to enlarge.
Between 1847 UTC and 1909 UTC, the southern (or leading) segment of the QLCS accelerated east across Anderson County, leaving the northern (or trailing) segment behind across southwestern Pickens County (Fig. 14). This evolution was similar to other QLCSs in high shear environments. However, in this case a clean break in the line into southern (leading) and northern (trailing) segments was not readily apparent, at least not at the resolution of the data and the color table that was used. A channel of weak reflectivity began to wrap cyclonically around the appendage of high reflectivity to the east of Clemson at 1909 UTC, which was even more apparent at 1913 UTC (Fig. 15). A weak low-level mesocyclone appeared to form by 1913 UTC on the storm relative motion image as the rotational velocity at 0.5 degrees strengthened to 30 kt by this time. This was an increase of about one-third over the previous scan. In a storm-relative sense, convergence is implied from the southeast corner of the reflectivity appendage, south along the back edge of the storm over Anderson County. A Severe Thunderstorm Warning was issued for southern Pickens County at 1915 UTC.
Figure 14. As in Fig. 10, except for 1909 UTC. The gray arrow denotes the developing weak echo channel. White lines indicate the leading and trailing segments of the quasi-linear convective system. Click on image to enlarge.
Figure 15. Radar reflectivity (top) and storm relative motion (bottom) on the 0.5 degree scan from the KGSP radar at 1913 UTC. The arrow points to the channel of weak reflectivity. Click on images to enlarge.
Some portion of the rear-to-front flow appeared to wrap completely around the reflectivity appendage to the west-southwest of Liberty by 1917 UTC, as evidenced by the development of an area of outbound radial velocity (i.e. away from the radar) northeast of the reflectivity appendage (Fig. 16). It is interesting to note the developing outbound radial velocity appeared directly beneath a new area of higher reflectivity aloft on the 1.8 degree scan, which suggests either a developing updraft in the storm or a strongly tilted one. The mesocyclone remained broad but cyclonically convergent on the 0.5 degree scan. At this time, the rotational velocity on the 1.2 degree scan jumped by 50 percent over the previous scan (from 21 kt to 31 kt) and was also cyclonically convergent, suggestive of an upward development of the mesocyclone (Fig. 17).
Figure 16. Radar reflectivity on the 1.8 degree scan (upper left) and the 0.5 degree scan (lower left), storm relative motion on the 0.5 degree scan (upper right), and radial velocity on the 0.5 degree scan (lower right) from the KGSP radar at 1917 UTC. Click on images to enlarge.
Figure 17. As in Figure 15, except for the 1.3 degree scan at 1917 UTC. Click on images to enlarge.
The mesocyclone reached its peak intensity at 1922 UTC in a purely rotational sense at a height of 1.2 km above the ground (Fig. 18). The mesocyclone continued to grow upward as suggested by a plot of rotational velocity on the 0.5 degree, 1.2 degree, and 2.4 degree scans (Fig. 19), and the WSR-88D mesocyclone algorithm. The distance between the maximum outbound and inbound storm relative velocity was nearly 3 nautical miles at a range of 24 nautical miles from the radar, which is considered broad and weak by the mesocyclone nomogram. Nevertheless, a tornado developed and touched down on the campus of Liberty Elementary School at 1924 UTC, as reported by several eyewitnesses.
Figure 18. Storm relative motion on the 0.5 degree scan from the KGSP radar at 1922 UTC. Click on image to enlarge.
Figure 19. Rotational velocity in the Liberty storm. Click on image to enlarge.
The storm that produced the Liberty tornado moved quickly northeast across the eastern part of Pickens County over the next 30 minutes. The low level rotation in the mesocyclone showed signs of strengthening as it passed to the north and northeast of Easley after 1939 UTC, but by that time the storm was collapsing. No other reports of damage were received along the storm's path.
b. The Moore Tornado
Shortly after the Liberty Tornado, the QLCS gained strength as it moved across Greenville County. By 1947 UTC, a line of very high reflectivity was oriented from north to south across southern Greenville County (Fig. 20), moving rapidly to the east. The evolution of the reflectivity pattern in the severe storm that produced the tornado to the northwest of Moore exhibited a classic "Broken-S" radar presentation as it moved into western Spartanburg County. The "Broken-S" radar reflectivity signature has been linked to the occurrence of non-supercell tornadoes in QLCSs in similar environments over the western Carolinas in the past (Lane and Moore 2006, McAvoy et al. 2003).
Figure 20. Radar reflectivity on the 0.5 degree scan from the KGSP radar at 1947 UTC. The radar is located near the center of the image, just to the right of the e in Greenville. Click on image to enlarge.
An examination of the reflectivity on the 1.2 degree scan from the KGSP WSR-88D radar (approximately 300 to 500 meters above ground level at that range) revealed a break in the line segment between 2000 UTC and 2008 UTC (Fig. 21). The line segment showed broad curvature in the form of an S in the high reflectivity (35 dBZ and greater) over western Spartanburg County at 2000 UTC, which wrapped tighter to the west of Moore by 2004 UTC. A fracture in the high reflectivity line segment occurred over the area to the northwest of Moore and west of Roebuck at 2008 UTC, just two minutes before the tornado touched down. The fracture appeared as a north to south oriented minimum in reflectivity (30 dBZ and less) west of Moore (Fig. 21, bottom).
Figure 21. Radar reflectivity on the 1.2 degree scan from the KGSP WSR-88D at 2000 UTC (top), 2004 UTC (middle), and 2008 UTC (bottom), on 5 January. The point labeled KGSP is the location of the radar and the point labeled T is the approximate location of the tornado damage. Click on images to enlarge.
The radial velocity showed the rapid development of the mesocyclone associated with the Moore Tornado between 2000 UTC and 2008 UTC (Fig. 22). The rotation in the developing mesocyclone can be inferred from the couplet of inbound velocity (green shades) and outbound velocity (red shades) to the west of Moore at 2000 UTC. The couplet strengthened through 2004 UTC, to the point where the maximum inbound velocity was on the order of 20 kt and the maximum outbound velocity was greater than 64 kt to the northwest of Moore at 2008 UTC. The storm relative velocity showed the rotational couplet with even greater clarity at 2008 UTC (Fig. 23). Interpretation of these images allowed the warning forecaster to issue a Tornado Warning before the tornado touched down northwest of Moore.
Figure 22. Radial velocity on the 1.2 degree scan from the KGSP WSR-88D at 2000 UTC (top), 2004 UTC (middle), and 2008 UTC (bottom), on 5 January. Targets moving toward the radar are indicated by green shades and targets moving away from the radar are shown by red shades. The point T is the approximate location of the tornado damage. Click on images to enlarge.
Figure 23. Storm relative motion on the 1.2 degree scan from the KGSP radar at 2008 UTC. Click on image to enlarge.
Shortly after the Moore Tornado lifted, the severe thunderstorm produced a microburst near the town of Roebuck at 2015 UTC. Several trees were uprooted and snapped. A few trees fell across roadways and on top of at least one house. Reports of funnel clouds were received as the storm tracked over eastern Spartanburg County and into Cherokee County, but apparently no other damage occurred.
c. The Gastonia Tornado
The QLCS crossed the line between Cherokee County and York County, South Carolina, around 2100 UTC. A fracture in the convective line occurred over northwestern York County between 2108 UTC and 2112 UTC, prompting the issuance of a Tornado Warning. A second-hand report of a tornado was received from the area near Clover, South Carolina, but the report has not been substantiated.
The system continued to move east northeast as two distinct segments over the next half hour. The southern (or leading) segment moved along the North and South Carolina line and the northern (or trailing) segment moved over Gaston County, North Carolina. As the system moved past Gastonia between 2129 UTC and 2138 UTC, the two segments became more separated as lower reflectivity moved around the southern end of the trailing line segment to the southeast of Gastonia. The KGSP radar showed the trailing segment of the QLCS extending to the south of Gastonia at 2133 UTC, but only weak rotation was seen in the storm relative motion on the lowest elevation scan (Fig. 24). However, it should be noted that the center point of the radar beam was 1.6 km above ground level in the vicinity of Gastonia. This put the KGSP radar at a disadvantage when attempting to detect low level features. The radar signatures associated with the Liberty Tornado were confined to levels below 1.5 km. This suggested the KGSP radar was too far away from the storm to be able to detect the important low level features that were associated with a developing tornado.
Figure 24. Reflectivity and radial velocity on the 0.5 degree scan from the KGSP radar at 2133 UTC. Click on images to enlarge.
This was not the case with the Terminal Doppler Weather Radar (TDWR) located north of the Charlotte - Douglas International Airport. Whereas the KGSP radar was located about 100 km to the west of the QLCS, the Charlotte TDWR (TCLT radar) was only 30 km distant. The favorable location and the smaller beamwidth of the TCLT radar allowed for a more detailed view of the storm. A comparison between the 0.5 degree scan from the KGSP radar (beam center point about 1700 m AGL) with the 2.4 degree scan from the TCLT radar (beam center point about 1200 m AGL) at 2134 UTC showed the difference in resolution (Fig. 25). Note the appearance of the channel of lower reflectivity that extended to the north of the Gastonia airport (KAKH) on the TCLT scan. The poorer resolution of the KGSP radar at that distance smeared out the appearance of the low reflectivity channel.
Figure 25. Radar reflectivity from the KGSP radar on the 0.5 degree scan at 2133 UTC (top) and from the TCLT radar on the 2.4 degree scan at 2134 UTC (bottom). The location of TCLT is shown in the upper right. Click on images to enlarge.
The favorable location of the TCLT radar allowed it to see more of the low level structure of the QLCS as well. A loop of the 1.0 degree reflectivity from TCLT showed the initial break in the QLCS over northwestern York County, South Carolina. By 2133 UTC, the QLCS had evolved to where the break in the two segments was well defined, with a channel of weak reflectivity having wrapped around the southern end of the northern (or trailing) segment (Fig. 26). The only significant rotation observed by the TCLT radar prior to the tornado occurred on the 1.0 degree scan at this time. It is interesting to note that the rotational couplet was not even directly associated with the southern end of the trailing reflectivity segment, but was embedded within a notch in the rear flank of the southern (or leading) reflectivity mass. The 0.2 degree scans from TCLT were even more revealing. Between 2130 UTC and 2135 UTC, a small mass of reflectivity peeled away from the rear flank of the larger reflectivity mass over the southeast corner of Gaston County. This smaller mass moved on a more northeasterly track and merged with the trailing reflectivity segment to the east of Gastonia around 2139 UTC. The tornado touched down at approximately 2141 UTC about four miles east of Gastonia. The tornado appeared to be associated with the southern end of the trailing line segment (Fig. 27), but the storm relative motion was contaminated with improperly dealiased velocities.
Figure 26. Radar reflectivity (top) and storm relative motion (bottom) from the TCLT radar on the 1.0 degree scan at 2134 UTC. The location of TCLT is shown in the upper right. Click on images to enlarge.
Figure 27. As in Figure 26, except for 2140 UTC. The approximate location of the Autumn Acres subdivision, where the tornado struck, is shown by point A. Click on images to enlarge.
Both radars showed a low reflectivity notch in the rear flank of the QLCS as it moved into York County, South Carolina. However, the radial velocity data from KGSP did not show higher wind speeds associated with the notch and the data from TCLT was contaminated by range folding. The QLCS fracture occurred about 30 minutes before the tornado touched down, similar to the Bessemer City storm nearly a year ago to the day. The KGSP radar gave little indication that a tornado was imminent on the 2138 UTC scan, with broad and weak rotation indistinguishable from other parts of the scan. Although the TCLT radar was much closer, and therefore able to see the low level features in the storm, it also gave little indication that a tornado was imminent on the scans leading up to tornado formation. Only one elevation scan in the 2134 UTC volume indicated significant rotation. The role played by the weak cell merger in the minutes prior to tornado formation is unknown. The parent storm weakened rapidly through 2200 UTC and no additional severe weather was reported.
4. Discussion and Summary
The forecasters at GSP had a relatively high degree of situational awareness leading up to the event and expected a few tornadoes as the squall line moved across the Upstate and Piedmont. However, that knowledge did not make the warning process any less challenging. With the benefit of hindsight, an examination of the radar data yields several subtle clues in the minutes leading up to each tornado. What is not known is the relative significance of any of the subtle radar signatures, or if similar signatures could be detected in any of the storms that did not produce a tornado on this day. The fact that hindsight magnifies the apparent importance of any radar clues should not be lost by the casual reader.
In the case of the Liberty Tornado, confidence was not high that a tornado was about to form. The QLCS did not exhibit the distinct break in the line that forecasters have observed in other cases. While some rotation was noted in the storm at low levels, it was thought to be too broad and too weak to be indicative of an impending tornado. This suggests that the threshold for issuing tornado warnings based on low level rotation might need to be lowered for similar high shear, low instability environments. The radar observations of the descending rear inflow jet support conclusions about low level vorticity generation drawn by earlier modeling work on QLCSs. The detection of rear inflow jets in similar systems warrants further study.
The Moore Tornado followed the classic radar evolution of a tornado-producing QLCS in this type of environment. This allowed the forecaster to issue an effective Tornado Warning. The close proximity to the KGSP radar yielded a high quality data set, which makes this case very valuable for future study.
As for the Gastonia Tornado, the radar signatures were the most subtle and thus the least understood. The poor representation on the KGSP radar suggests that early detection of important low-level signatures might be nearly impossible at ranges beyond 100 km. Fortunately, the TCLT radar greatly improves the capability to see the low level structure of similar storms across the North Carolina Piedmont.
More images of damage produced by the Liberty Tornado. The concession stand at the Liberty High School football field was ripped from its mooring and overturned. Part of a fence was also knocked down. Relatively little tree damage was noted owing to the location of the very narrow and short tornado path. Click on images to enlarge.
More images of the damage produced by the Liberty Tornado in the parking lot of Liberty Elementary School. Images are courtesy of the Pickens Sentinel and are subject to copywright.
Damage to outbuildings and a barn caused by the Moore Tornado.
Damage to a home and a tree in the Autumn Acres subdivision east of Gastonia, caused by the Gastonia Tornado. Click on image to enlarge.
Sandy Foster, editor of The Pickens Sentinel, provided the images of the damaged vehicles in Liberty, South Carolina. The upper air analysis and sounding graphics were obtained from the Storm Prediction Center. The surface analysis graphic was obtained from the Hydrometeorological Prediction Center. The regional surface plots, satellite imagery, and radar mosaic images were obtained from the University Corporation for Atmospheric Research. The images from the KGSP radar were made using the Java NEXRAD Viewer from the National Climatic Data Center.
Lane, J. D., and P. D. Moore, 2006: Observations of a non-supercell tornadic thunderstorm from a Terminal Doppler Weather Radar. Preprints, 23rd Conf. on Severe Local Storms, St. Louis, MO, Amer. Meteor. Soc. McAvoy, B. P., W. A. Jones, and P. D. Moore, 2000: Investigation of an unusual storm structure associated with weak to occasionally strong tornadoes over the Eastern United States. Preprints, 20th Conf. on Severe Local Storms, Orlando, FL, Amer. Meteor. Soc., 182-185.