Cloud to Ground Lightning Climatology and Hail Prediction in the Mid-South Matthew Reagan Mississippi State University
Objective Create model to predict severe hail events using strictly cloud-to-ground lightning
Hypothesis Cloud to ground lightning activity will change as a hail core descends to the ground
Charge Structure Precipitation Theory –Krehbiel 1986 –Lang and Rutledge 2002 Dipole Structure –Krehbiel 1986 –Dotzek et al –Rakov and Uman 2003 Krehbiel 1986
Cloud-to-Ground Lightning Percentage of Positive and Negative Strikes –Reap and MacGorman 1989 Severe Weather and Positive CG –Reap and MacGorman 1989 –MacGorman and Burgess 1994 –Liu et al. 2006
Return Strokes Krehbiel 1986 Reap and MacGorman 1989
Scope
Lightning Climatology MonthPositive StrikesTotal StrikesPercent Positive January13,29999, February27,246241, March44,114445, April65,672898, May98,9381,907, June53,9991,425, July84,8802,252, August66,0821,840, September18,677381, October19,779293, November27,104276, December22,360144, Total542,15010,205, ,795 strikes per day
Lightning Climatology
Percent Positive
Return Strokes Reap and MacGorman 1989 –Positive: 78% –Negative: 31% Reap & MacGorman –NSSL Network Mid-South Study –NLDN PositiveNegative January February March April May June July August September October November December Total:
Diurnal Climatology
Pulse Thunderstorms
Storm Collection Hail Storms –Storm reports Non hail producing storms –VILD between 2.5 and 3.5 kg/m 3
Extraction
Storm of Interest
Model Creation Lightning in 5 minute increments –Total CG –Total Positive Strikes –Return Strokes Per Strike 355 storms –300 for training –55 for testing Jack knifing –300 times in training –55 times in testing Test finalists on independent data set
Models Used Logistic Regression Artificial Intelligence –Polynomial –Linear –Radial –Sigmoid
What is Artificial Intelligence?
y z x o x x o x o o x y
Percent Correct
Probability of Detection
False Alarm Ratio
Probability of False Detection
Bias
Heidke Skill Score
Eliminated –D2 & D3 Polynomials –Sigmoid –Radial E001 C100 Remaining –D1 Polynomials –Linear –Radial –Logistic
Model Families Radial E01 C10 Radial E001 C1
PCBiasFARPODPOFDHSS Polynomial D1 C Linear C Radial E001 C Logistic Regression PCBiasFARPODPOFDHSS Polynomial D1 C Linear C Radial E001 C Logistic Regression Median Statistics Variance
Independent Data Set storms used for training –355 events storms used for testing –205 events Linear C100 Logistic Regression PC Bias FAR POD POFD HSS
January 20, 2010
January 20, 2010 Event b
January 20, 2010 Event c
EHI and Average Strikes ML CAPE0-3 SRHEHIAverage Strikes Nov. 15, Nov. 29, May 07, May 10, March 29,
Center 7.5 Radius
Limitations Hail Report Bias ¼”Pea ½”Marble ¾”Penny 7/8”Nickel 1"Quarter 1 ¼”Half Dollar 1 ½”Walnut/Ping Pong Ball 1 ¾”Golf Ball 2"Hen Egg 2 ½”Tennis Ball 2 ¾”Baseball 3"Teacup 4"Grapefruit 4 ½”Softball Hail SizeFrequency 1”161 1 ¼”18 1 ½”13 1 ¾”101 2”4 2 ½”2 2 ¾”10 4”3 4.50”1
Limitations Hail Report Bias Non Severe VILD Threshold
Limitations Hail Report Bias Non Severe VILD Threshold Cloud to ground lightning only
Summary CG lightning peaks in summer months Positive CG accounts for 5.31% of CG lightning with a peak in winter months Script has proven flexible with storm mode Logistic regression model out performed artificial intelligence Storm environment factors need to be studied
Questions?