Lightning and Severe Weather: A Comparison between Total and Cloud-to-Ground Lightning - PowerPoint Presentation

Lightning and Severe Weather: A Comparison between Total and Cloud-to-Ground Lightning Trends Authors: Christopher J Schultz, Walter A. Petersen, and Lawrence D. Carey Published: Spring 2011

Introduction Proven relation between lightning flashes and kinetic energy of T-storm updraft Stronger updraft  potentially stronger mixed-phase region  more charging/charge separation  Bigger E-fields IC/CG flashes discharge excessive electrical energy buildups Lots of data on CG lightning NLDN, NALDN, etc. Several studies tried to correlate CG flash frequency and thunderstorm intensity  results were mostly inconsistent (even considered reversal of polarity in some studies) VHF total lighting mapping arrays offer a new perspective for comparing lighting and thunderstorm intensity Multiple studies show that total flash rate rapidly increases prior to onset of severe wx But want to perform studies of CG flash rates in the same framework as the total lighting flash rates

Methodology – Case Selections Comparing CG and total lightning flash rates in framework of an application created to test lightning info as an aid for predicting severe wx at the surface Analyzed 711 thunderstorms Used severe (e.g. storms with tornadoes, hail >= 1.9cm D, winds >= 26 m/s) and non-severe thunderstorms 1064 severe wx reports used (133 tornado, 649 hail, 281 wind), combined into 6-minute periods (reduced reports to 784 events)…majority of tornadoes were smaller (76 % of tornadoes were EF0 or EF1) Four different meteorological regimes in the US tested North Alabama, Washington, D.C., eastern Colorado/western Kansas, and Oklahoma All also well within NLDN coverage Used primarily SE US (AL) data and mid-Atlantic (D.C.) data due to ease of access to total lightning data, and to not skew dataset w/ regions of the country known to have higher IC:CG ratios

Methodology – Lightning Data Four LMA (VHF lightning mapping arrays) used to collect total lightning data NALMA, DC LMA, STEPS, OK-LMA Flashes were defined from VHF source points using two flash-clustering algorithms North Alabama storms used McCaul et al. (2005) clustering algorithm Washington D.C., Oklahoma, and eastern Colorado/western Kansas storms used clustering algorithm found in XLMA software package (Thomas et al. (2004)) – similar spatial criteria Main difference between the two – McCaul et al. (2005) doesn’t have upper threshold on temporal length of a lightning flash Lightning flash construction Min of 10 VHF source points Source threshold didn’t affect general trend in total lightning All t-storms within 200 km of center of each LMA network CG flashes determined from NLDN 113 sensors across US, flash detection efficiency of 90%-93% small IC flashes sometimes misclassified as +CG lightning, so 15-kA peak current required for CGs

Methodology – Lightning Jump Algorithm Selection Schultz et al. (2009) tested six lightning jump algorithm configurations on severe/non-severe wx and determined statistically a “2𝛔” configuration worked best The algorithm went as follows… Average the most current two minutes of total lightning data (at t 0 ) from an individual storm  # flashes per minute. Algorithm activates if total flash rate is >= 10 flashes per minute. Divide the 12 minutes of total lightning data prior to the most current two minutes of data, t 0 , into two minute time periods (so, six periods) and average as in step 1. Subtract consecutive periods from the 12 minutes of data. This determines time rate of change of total flash rate (DFRDT). Results in five DFRDT values. Calculate standard deviation from these five DFRDT values. Twice that standard deviation value becomes the “jump threshold” at t 0 . Calculate a new DFRDT value by subtracting the current total flash rate at t 0 from the two minutes of data just prior to most recent time, t -1 .

Methodology – Lightning Jump Algorithm Selection The algorithm continued… Place a “warning” on the t-storm for 45 minutes if a lightning jump occurs (i.e. the new DFRDT value exceeds the calculated “jump threshold”). Repeat calculations of DFRDT values as additional two minute periods of total flash information are received for a particular storm until the total flash rate drops below the activation threshold of 10 flashes per minute. Lightning jump occurs when flash rate exceeds the activation threshold and the DFRDT exceeded 2𝛔 of the mean DFRDT of the previous 10 minutes. Lightning jump ends when DFRDT value <= 0, unless two jumps are separated by six minutes or fewer . To accommodate CG lightning information, algorithm was modified to allow lower-frequency flash rates  made the activation threshold two flashes per minute (based on ordinary convection average CG flash rate)

Methodology – Verification and Statistical Methods Verification of the lightning jump warnings occur if severe wx is observed within the 45 minute time a storm is placed under a severe wx Time between of severe wx reports <= six minutes results in severe wx reports being grouped together Two warnings in place and severe wx observed, verification attributed to earlier warning Unverified warnings = false alarms, severe wx events without lightning jump detected = misses

7 April 2006, Tornadic Supercell in N. Alabama Rep. “typical” total and CG flash rates in a supercell t-storm Peak total flash rate of 154 flashes per minute, peak CG flash rate only 14 flashes per minute Total lightning p rovided avg of 23.57 minutes lead time Total lightning trend provided lead times on 14 of 16 severe wx events, CG lightning trend provided lead times on 13 of 16 severe wx events CG lightning p rovided avg of ~19 minutes of lead time Red bars = total lighting jump times, green bars = DFRDT values not reaching the jump threshold, orange line = 2𝛔 jump threshold, x = CG lightning jumps, red diamond = tornado, green asterisk = hail, blue square = wind

20 June 2000, Supercell in W. Kansas Red bars = total lighting jump times, green bars = DFRDT values not reaching the jump threshold, orange line = 2𝛔 jump threshold, x = CG lightning jumps, red diamond = tornado, green asterisk = hail, blue square = wind Initially, total flash rates below 10 flashes per minute, with no CG lightning Activation threshold met, and lightning jump detected at 0045, ending at 0049 No CG detected until 0059 UTC At 0104 UTC wind gust of 75 kt (38.6 m/s) recorded Four more lightning jumps detected – 0103, 0119, 0141, and 0157 UTC Additional severe wind damage reported at 0124, 0142, and 0154 UTC CG flash rate never went above 2 flashes per minute, and there were no CG lightning jumps Two false alarms recorded With 2𝛔, total lightning jumps occurred prior to all severe wx events Avg lead time was 33 minutes with total lightning info…versus none with CG info

18 April 2006, Hail-Producing Supercell Thunderstorm in E. Alabama Red bars = total lighting jump times, green bars = DFRDT values not reaching the jump threshold, orange line = 2𝛔 jump threshold, x = CG lightning jumps, red diamond = tornado, green asterisk = hail, blue square = wind Around 2140 UTC, the seemingly ordinary thunderstorm rapidly developed vertically Between 2146 and 2148 UTC, total flash rate increased and a lightning jump triggered No CGs evident at this point 2215 UTC report of 2.54 cm hail 2218 UTC another lightning jump occurred, while CG rate still remained too low for jump activation At 2244 UTC, 3.81 cm hail observed Two more lightning jumps observed at 2258 and 2346 UTC Large hail reports following at 2313, 2326, 2342, and 0008 UTC (19 April) Wind damage reported at 2326 and 0005 UTC All six severe wx events identified prior to occurrence, with avg lead time of 27 minutes from total lightning data

Summary of Performance POD of 79%, FAR of 36%... 40 % of false alarms were additional lightning jumps occurring prior to severe wx, when another “warning” was already in place Remove those and the FAR drops to 22 % Average lead time prior to onset of all severe wx was 20.65 minutes, s.d. of 15.05 minutes…average lead time for tornadoes was 21.32 minutes, s.d. of 15.15 minutes! CG lightning information for same set of 711 storms was not awful, but worse than total lightning information POD of 66 %, FAR of 53 %...16 % of false alarms were like the ones mentioned above, remove them and FAR reduces to 44 % Average lead time prior to onset of all severe wx was 13.54 minutes, s.d. of 15.07 minutes, and on just tornadoes was 15.24 minutes, s.d. of 15.27 minutes  

Discussion Total lightning data use with the 2𝛔 algorithm outperforms a similar algorithm implementation with only CG lightning data Did identify some regimes where total lightning data use needs work Nearly 40% of all missed events by total lightning 2𝛔 algorithm were from a combo of tropical rainband, cold season, and low-topped storms (39% of all CG lightning 2𝛔 algorithm misses were from these too) These types of storms only represent about ~20% of severe wx in this study With 8% of severe wx in the study having 0 lightning flashes, see not all severe wx associated with lightning CG lightning also missed big portion (46%) of severe wx events in supercellular storms Portion was in earlier life cycles of storms Performed best with storms that were longer lived, or produced severe wx at end of their life cycles

Discussion (cont.) Obviously, need to continue to work on an operationally-useful total lightning jump algorithm, but need to fix the algorithm to account for low-flashing severe wx environments Could lower the threshold to activate the algorithm F or land-falling hurricanes, any observed total lightning info highlighted the most hazardous cells in the outer rainbands CG networks will still provide info on CG ground flash lightning locations and rates Still useful in meteorological applications since it provides info for human safety, forest fire applications, and infrastructure protection! But results here have have implications for use of ground-based CG detection networks relative to ground- or space-based total lightning detection equipment! Incorporation of total lightning info should improve warnings, lead times, save more lives, and protect more property from damage Hence, need things like GLM on GOES-R (Now GOES-16) are necessary for improving severe wx warning decisions

Conclusions Goals of the study were two-fold 1) Contrast utility of total and CG lightning occurrence trends in specific nowcasting application 2) Confirm a methodology for applying total lightning flash information to the severe weather nowcasting problem Examined 711 t-storms from four regions of US and found total lightning trend information outperforms CG lightning trend information Because of the physical relationship between the UD, precipitation-sized ice, supercooled water, and the initial production of IC flashes, obviously, total lightning info is the best EARLY indicator of a strengthening UD Ground-based total lightning networks while accurate and useful for regional use are limited in range and costly to expand Results of the study show the need for deployment/use of geostationary lightning mappers aboard future GOES satellite platforms