A Comparison of Lightning Strikes to Radar
Observations of Thunderstorms on July 9-10, 1993
National Weather Service Forecast Office
National Weather Service Forecast Office
Over the National Weather Service's Automated Field Operations and Services (AFOS) computer network, several graphic products depicting lighting activity are available. The product that is the subject of this study, the Lightning Detection Summary (LDS) chart (Figure 1), is generated every five minutes and depicts all cloud-to-ground lightning activity reported in the continental United States in the preceding five minutes.
Figure 1. Lightning Detection Summary (LDS) for 0013-0017 UTC 10 July 1993.
The purpose of this study was to compare cloud-to-ground lightning activity to radar observed thunderstorm characteristics. It was also desired to evaluate some of the operational uses and limitations of the method of presenting lightning data in the form of the LDS AFOS graphic.
It is desirable to attempt to relate lightning activity with radar observed storm characteristics. These are the two main remote sensing data sources available to operational forecasters in real-time, which can be utilized in assessing thunderstorm activity. Further, while radar generally indicates the ultimate outcome of the processes occurring within a convective element, lightning data may provide insight into these processes as they occur since a lightning stroke represents an immediate resolution of stresses applied to the electrical field within the convective element.
Similar studies in the past have found correlations between lightning frequency and thunderstorm intensity. Reap and MacGorman (1988) found a good correlation between cloud-to-ground lightning frequency and low-level radar echo intensity. Similarly, Holle and Maier (1982) found that the greatest rate of cloud-to-ground lightning activity corresponded to the storms having greatest vertical development. Goodman et al. (1988) suggested that higher lightning rates may be a result of stronger updrafts as the peak flash rates were found to occur in-phase with maximum storm mass, Vertically Integrated Liquid (VIL), echo volume and cloud height.
In comparing lightning frequency to severe thunderstorm potential, several studies have revealed a possible relationship between the two. Kane (1991) suggested that a sharp decline in cloud-to-ground lightning activity following a lighting rate peak may serve as a precursor to severe thunderstorm. Similarly, Goodman et al. (1988) noted that a decrease in lightning activity associated with the collapse of a storm was a precursor to the arrival of a microburst at the surface. Elson (1993) also showed that cloud-to-ground lightning activity holds some potential as a predictor to severe local storms, particularly for high lightning rates related to strong winds.
A. Data Collection
All data for this study were collected from 1950-0130 UTC 9-10 July 1993. The area of data collection covered north and central Indiana, and east-central Illinois. The thunderstorms developed in a warm, humid air mass ahead of a weak surface cold front that extended from northwest Indiana through central Illinois.
Five minute total cloud-to-ground lightning data are available to National Weather Service offices on AFOS as the graphic product LDS. This chart utilizes a 140 x 100 grid with individual box dimensions of 0.43°x 0.43°. Contained in each grid box is a symbol representing a range of lightning strikes that occurred in that box during the preceding five minutes.
The data for the LDS graphic originates from real-time lightning data obtained using the Lightning Position and Tracking System (LPATS) time-of-arrival tracking technique. The National Severe Storms Forecast Center (NSSFC) receives this data from Atmospheric Research Systems, Inc. (ARSI). The graphic is then generated and distributed over AFOS (Lewis 1993, personal communication).
Radar data were collected from WSR-74 C-band radars that use a 5 cm wavelength. Data were primarily gathered from the radar located in Indianapolis, with supplemental observations from South Bend, Indiana.
Storm locations were based on the center of VIP 5 (50-57 dBZ) reflectivities at a 0.5° antenna elevation angle. Echo tops of the VIP levels 1 (0-30 dBZ), 3 (41-46 dBZ), 4 (46-50 dBZ) and 5 were noted for each storm. The severe local storm reports are from Storm Data (1993).
B. Data Analysis
Radar data were gathered for 19 individual storms. To insure the quality of data, storms had to meet objective criteria. First, a minimum of three consecutive radar observations not more than 15 minutes apart were required. Secondly, the storm had to be clearly distinguishable in the lightning data. In effect, no two cells could be in such close proximity that it would be questionable with which storm to associate the lightning data.
The use of grid boxes to summarize data contributes to the problem of differentiating individual cells. Of the data originally gathered for this study, nearly one-half had to be discarded due to an inability to successfully identify individual storms. This was generally attributed to more than one storm being located within a particular grid box. Use of grid boxes also made it difficult to assess total lightning production when a storm straddled or was crossing a boundary between two or more boxes.
Eight storms met the selection criteria. These storms represented various stages of a thunderstorm's life cycle. In some cases data were collected over the complete life cycle, while in others only part of the life cycle could be analyzed.
With VIL coming into increasingly widespread use in such radar systems as the WSR-88D, it was desirable to approximate storm mass. As the collected data were insufficient to compute VIL, a separate value to approximate storm mass was created. This computed value was termed "Storm Mass", and is simply the sum of the tops of the VIP 5, 4 and 3 levels. This was then divided by 1000 for simplicity.
Three key points were addressed in the analysis. The times of peak lightning activity were compared to the times of maximum storm size, five-minute lightning counts were compared to storm size, and lightning characteristics were compared with severe weather occurrences.
To clearly define the time of peak lightning activity for each cell, a three-point moving average was computed. The lightning activity recorded during a given five-minute period was averaged with the preceding and subsequent five-minute counts. In cases of multiple peaks of equal value, the time of the initial peak was used as the time of peak lightning activity. This time was compared with the time of occurrence of maximum storm top, maximum height of the VIP 5, and maximum value of storm mass. An exception to this analysis was Storm 2. Since there was no discernible peak in the lightning activity, Storm 2 was omitted from this part of the analysis.
In comparing lightning counts with storm size, storms were categorized as "large" or "small" based on threshold values for each characteristic. Storms were defined as large in the lightning category, if the peak five-minute lightning count equaled or exceeded 50 strikes. To be considered large for the storm tops category, maximum tops had to equal or exceed 50,000 feet AGL, the height of the VIP 5 reflectivity had to be at least 30,000 feet AGL, and the storm mass needed to be 100 or greater.
Temporal lightning characteristics were compared with the times of occurrence of severe weather. The National Weather Service definition of a severe thunderstorm was used, namely hail size 3/4" in diameter or larger, and wind damage implying or measured winds of 50 knots or greater (WSOM 1995). The only severe thunderstorm observed was number one.
3. STORM CHARACTERISTICS
A summary of the characteristics of the individual storms is found in Tables 1 through 4. In the characteristic column of each table, terms used were defined as follows. "Steady state" meant no appreciable (maximum total change in lightning was less than 20 strikes, or maximum change in radar heights was less than 15 percent) increase or decrease in lightning activity or storm size was noted. "Peak" indicated a distinct maximum of lightning activity or storm size. This was further classified by the time of the peak with respect to the observational period. "Decreasing" indicated a downward trend in the observed values. "Cyclic" was used when at least two distinct peaks were noted. "Early", "Middle" and "Late" are relative terms used with respect to the period of observation.
In Table 1, lightning values are given as the number of strikes counted in a five-minute period. On the LDS chart, the number is rounded down to the nearest ten, with a few exceptions. For strike counts below ten, the number of strikes was counted as five, except in the case where no lightning activity was reported. All counts exceeding 99 were labeled as 100+.
Tables 2, 3 and 4 give specific information on individual storm size and trends for storm tops, VIP 5's, and Storm Mass respectively. The percent increase means the greatest total increase in value leading up to the highest peak. The percent decrease is the greatest total decrease noted after the highest peak.
|Storm #||Characteristic||Maximum Increase||Maximum Decrease||Maximum Value|
|Storm #||Characteristic||% Increase Tops||% Decrease Tops||Maximum Top|
|1||Peak Early||37||35||52,000 ft|
|2||Peak Late||38||38||47,000 ft|
|3||Peak Middle||79||28||50,000 ft|
|4||Steady State||8||N/A||42,000 ft|
|5||Steady State||5||9||44,000 ft|
|6||Steady State||7||7||46,000 ft|
|7||Peak Late||9||16||51,000 ft|
|Storm #||Characteristic||% VIP 5's||% Decrease VIP 5's||Maximum Height|
|1||Peak Middle||72||58||31,000 ft|
|3||Peak Middle||77||13||23,000 ft|
|5||Steady State||8||11||28,000 ft|
|6||Steady State||15||3||31,000 ft|
|7||Peak Late||35||37||35,000 ft|
|Storm #||Characteristic||% Increase Mass||%Decrease Mass||Greatest Value|
Storm 1 (Figure 2) was the only "severe weather" producer of the cells examined. The storm produced one inch diameter hail and winds that downed trees in South Bend, Indiana at approximately 2100 UTC 9 July. This storm met the criteria for large storms in all three radar observational categories, as well as in the lightning category. The lightning peak preceded maximum storm top by 10 minutes, while the highest VIP 5 and Storm Mass followed peak lightning by 40 minutes. The severe weather occurred 50 minutes after peak lighting.
Figure 2. Storm 1: Cloud-to-Ground Lightning Strokes, Storm Tops and Height of VIP 5 Levels for 1950-2130 UTC 9 July 1993.
The weak lightning activity of Storm 2 did not show a distinct peak. This storm also failed to meet any of the large storm criteria.
The highest storm top for Storm 3 qualified it as a large storm in the tops classification. It did not meet large storm thresholds in the other categories. Maximum tops, VIP 5, and Storm Mass coincided with the lightning peak.
Large storm criteria were not met in any category by Storm 4. A time-lag of ten minutes was noted between the peak lightning activity and the times of maximum storm top, VIP 5 and Storm Mass.
Also failing to meet the large storm criteria was Storm 5 (Figure 3). The lightning peak with this cell occurred in conjunction with the maximum levels of all 3 radar observed characteristics.
Figure 3. Storm 5: Cloud-to-Ground Lightning Strokes, Storm Tops and Height of VIP 5 Levels for 0000-0045 UTC 10 July 1993.
Storm 6 was considered a large storm with respect to lightning, and with respect to the VIP 5 and Storm Mass size. However, the maximum storm top did not qualify this cell as a large storm. There was no apparent time-lag between the peak lightning and the maximum storm top, VIP 5 or Storm Mass.
Qualifying as a large storm in all three radar categories, along with the lightning category, was Storm 7. The largest time-lags between the lightning peak and the maximum storm tops, VIP 5 and Storm Mass, 45 minutes, were observed in this case.
Observations for Storm 8 included only the dissipating stage. Maximum values in lightning counts and storm size were noted in the initial observations, with decreasing values after that. As a result, no time-lag was computed.
A. Storm Size Characteristics
Key results are summarized in Table 5. There is a good correlation between storms categorized as large lightning producers, and large storm size, as observed by radar. Storms 1 and 7 qualified as large in all three radar categories. Additionally, both storms were classified as large lightning producers. Storm 8 also met all three radar criteria for a large storm. Since it appears likely that a greater lightning maximum was reached with Storm 8 before initial observations, a conclusion could not be made.
|Time Lag (min) from Peak Lightning||Large Storm Criteria Met|
|Storm #||Max Tops||Max VIP 5||Max Mass||Svr Wx||Ltg||Tops||VIP 5||Mass|
The only other storm that met two of the three radar criteria for a large storm was Storm 6. This was the only remaining storm classified as a large lightning producer.
Of the remaining storms, none were large lightning producers. In only one cell were any of the radar criteria for large storms met: Storm 3 had a maximum top of 50,000 ft.
B. Temporal Characteristics
For storms considered large, there was a significant time-lag between lightning peaks and radar observed storm size peaks. For other smaller storms, no significant time-lag was consistently present.
Of the two storms that were classified as large in all three radar categories and the lightning category, both exhibited large lag times between the time of lightning peak and the time of peak VIP 5 reflectivity and Storm Mass. Both were more than one-half hour. Additionally, for both storms the time of lightning peak preceded time of maximum storm tops; Storm 1 by 10 minutes and Storm 7 by 45 minutes.
Storm 1 was the only severe storm. Severe local storm events followed the lightning peak by 50 minutes, maximum top by 40 minutes, and peak VIP 5 and Storm Mass by 10 minutes. It is notable that in only this storm was there a progression from lightning peak to maximum top to peak VIP 5 and Storm Mass to severe weather events.
The only other storm in which a large lightning count was noted was Storm 6. No time-lag between the peak lightning activity and the radar observed peak storm sizes was observed.
Of the remaining storms in which a lightning peak was observed, only one showed a time-lag from time of lightning peak to the times of maximum storm size. Storm 4, which was not classified as large in any category, had a time-lag of 10 minutes for all three radar categories.
In the cases examined here, high cloud-to-ground lightning rates were generally associated with the largest radar-observed storms. The three storms exhibiting peak lightning rates more than 50 strikes per five minutes were all considered large storms by the defined radar criteria. The only storm that became severe showed a large peak lightning value. Holle and Maier (1982), and Alleca (1988) also found that lightning activity was maximized near storms with the greatest vertical development.
Also of significance is that the largest storms showed a notable time-lag between time of peak lightning and time of peak radar-observed storm size. Two of the three large lightning producers showed time-lags in excess of 30 minutes to the times of maximum VIP 5 heights and maximum storm mass. Storms that did not produce a large amount of lightning strikes generally lacked this significant time-lag. A significant time-lag of 50 minutes was also seen between the time of peak lightning and the only occurrence of "severe weather".
Despite the small data set, this compares favorably with previous findings. Goodman et al. (1988) concluded that a decrease in total flash rate associated with a storm collapse serves as a precursor to the arrival of the microburst at the surface. Kane (1991) suggested that a sharp decline in cloud-to-ground flashes following the peak may serve as a precursor to severe events, while Elson (1993) found a strong tendency for damaging winds at the surface to occur during or within sixty minutes after a period of relatively high cloud-to-ground lightning activity.
The time-lag is in direct contrast to MacGorman et al. (1989), who noted the ground flash rate in tornadic storms peaked a few minutes after the peak in low level reflectivity. In comparing with tornadic storms though, it should be noted that most studies comparing cloud-to-ground lightning rates to tornadic activity have found little correlation. MacGorman et al. (1989) found no obvious correlation between the time of tornadoes and the ground flash rate, but did note that the ground flash rate was actually higher after the tornadic stage ended. Both Maier and Krider (1982) and Elson (1993) showed tornadoes tend to occur during periods of relatively low cloud-to-ground lightning activity.
The LDS chart can be a valuable tool in assessing storm characteristics and tendencies. It is not without its limitations, however. The five-minute interval between lightning data sets is incomparably better than the fifteen minute intervals previously available to National Weather Service forecasters. Although a shorter time would be preferable, we found the five-minute interval appears adequate for judging storm trends.
Like all forms of remote sensing, it can be difficult to differentiate between storm cells near each other. This problem is compounded by using grid boxes that are larger in scale than the convective processes involved. The grid boxes also created difficulties in determining lightning characteristics of storms crossing grid box boundaries.
Overall, the lightning data as presented in the LDS graphic appeared a useful aid in diagnosing subsequent thunderstorm tendencies. From the cases examined here, storms that produced the greatest peaks in lightning activity could be expected to become the largest storms. Results presented here support previous work that suggest that storms observed to continue intensifying on radar after the peak in cloud-to-ground lightning activity could be expected to ultimately become relatively strong.
The authors wish to thank the staff at NWSFO Indianapolis for their help and support, in particular Bill Gery, Mike Sabones, and John Curran. Thanks to Terry Click (NWSO South Bend) for providing supplemental radar data. Reviews by Jan Lewis (NSSFC) and Ed Berry (CRH) were also appreciated.
Alleca, J.P., 1988: A Case Study of a Frequent Lightning Event. Preprints: 15th Conference on Severe Local Storms. Baltimore, AMS (Boston), 488-495.
Department of Commerce, National Oceanic and Atmospheric Administration, 1993: Storm Data, 35, No. 7, 63.
____________, ____________, 1995: Weather Service Operation Manual (WSOM) (95-2), Severe Local Storm Watches, Warnings, and Statements Section 4, 8.
Elson, D.B., 1993: Relating Cloud-to-Ground Lightning to Severe Weather in Indiana on 2 June 1990, Nat. Wea. Dig., 18, 15-21.
Goodman, S.J., D.E. Buecheler, P.D. Wright, and W.D. Rust, 1988: Lightning and Precipitation History of a Microburst-Producing Storm, Geophys. Res. Lett., 15, 1185-1188.
Holle, R.L., and M.W. Maier, 1982: Radar Echo Height Related to Cloud-Ground Lightning in South Florida, Preprints: 12th Conference on Severe Local Storms, San Antonio, AMS (Boston), 330-333.
Kane, R.J., 1991: Correlating Lightning to Severe Local Storms in the Northeastern United States, Wea. and Forecasting, 6, 3-12.
MacGorman, D.R., D.W. Burgess, V. Mazur, W.D. Rust, W.L. Taylor, and B.C. Johnson, 1989: Lightning Rates Relative to Tornadic Storm Evolution on 22 May 1981, J. Atmos. Sci., 46, 221-250.
Maier, M.W., and E.P. Krider, 1982: A Comparative Study of Cloud-to-Ground Lightning Characteristics in Florida and Oklahoma Thunderstorms, Preprints: Twelfth Conference on Severe Local Storms, San Antonio, AMS (Boston), 334-337.
Reap, R.M., and D.R. MacGorman, 1988: A Comparison of Cloud-to-Ground Lightning to Analyzed Model Fields, Radar Observations, and Severe Local Storms, Preprints: 15th Conference on Severe Local Storms, Baltimore, AMS (Boston), 505-510.