1. Evaluation of Lightning Jumps as a Predictor of Severe Weather in the Northeastern United States
Pamela Eck, Brian Tang, and Lance Bosart
University at Albany, SUNY
Northeast Regional Operational Workshop 17
Wednesday, 3 November 2016

2. Motivation
Terrain can play an important role in the evolution of severe convection in the northeastern United States
Great Barrington, MA (1995)
F4 tornado
3 killed, 24 injured
11.5 mile track
Springfield, MA (2011)
EF3 tornado
3 killed, 200 injured
Mechanicville, NY (1998)
F3 tornado
30.5 mile track
Duanesburg, NY (2014)
10-cm-diameter hail
Springfield, Massachusetts on 1 June 2011
Elevation (m)

3. Background Upslope flow + convection =
Markowski and Dotzek 2011
Upslope flow + convection =
locally enhanced updraft and increased probability of severe weather
(a)
qr = model rain fall at 1km (shading), w5km = vertical velocity at 5km (closed contours) A B u
(c)
(b)
windward
leeward
NOTE: -1K potential temperature perturbation contours indicate approximate location of the gust front A B

4. Background Lack of surface observations
Inability of radar beams to sample behind mountains
Alternative method = sudden increase in total lightning (“lightning jump”)
A lightning jump is indicative of a strengthening updraft and an increase in the probability of severe weather
Valatie supercell (19 July 2015)
Flash rate increased by 2 standard deviations (σ) in 10 minutes
Time series of flash rates
MENTION VALATIE SUPERCELL
Minimum threshold of
10 flashes/min

5. Background Schultz et al. 2011
Severe wind producing thunderstorm on 20 June 2000 in western Kansas
Purple = total lightning
Red = cloud-to-ground lightning
wind reports
lightning jumps

6. Methodology
New England (CT, MA, RI, VT, ME, NH), New York, Pennsylvania
8-km resolution grid spacing
July & August 2014–2015 (1,791 storm tracks)
Lightning data from the National Lightning Detection Network (NLDN)
Total lightning = intracloud (IC) and cloud-to-ground (CG)
Severe weather reports from the Storm Prediction Center (SPC)
Wind, hail, and tornado
Severe weather day = 12Z–12Z
2σ lightning jump algorithm
Minimum threshold of 5 flashes/min
NLDN has a detection efficiency of 40–60% for intracloud lightning
Mention that we chose an 8-km resolution grid spacing to mimic that of the GOES-R LMA...
The image in the bottom right corner of instantaneous flash density over the past 2 minutes depicts a pseudo-GLM
Valatie supercell

7. Severe Weather Report) NO Severe Weather Report)
Verification
Lightning jumps were verified against severe weather reports taken from the SPC
If a lightning jump occurred within 45-minutes prior to a severe storm report, it verified as a hit
Calculated false alarm rate (FAR) and probability of detection (POD)
Lightning Jump
YES NO
A. Hit
(Lightning Jump +
Severe Weather Report)
B. False Alarm
NO Severe Weather Report)
C. Miss
(NO Lightning Jump
D. Correct Null
FAR = B / ( A + B )
POD = A / ( A + C )
YES NO
Severe Weather Report

8. Results: Part I
Schultz et al (Sigma Level = 2.0, Flash Rate = 10 flashes/min)
POD = 79%; FAR = 36%
Our results (Sigma Level = 2.0, Flash Rate = 5 flashes/min)
POD = 83%; FAR = 84%
FAR is too high… Why?

9. Upslope Case studies indicated…
Lightning jumps are occurring in sub-severe storms
Markowski and Dotzek 2011
Upslope flow + convection = locally enhanced updraft  increased probability of severe weather
Hypothesis: Conditioning on an upslope variable will reduce the FAR!
Calculating upslope:
Upslope (Λ) = v ∙ ∇zs > 0
v = u & v component of the 80-m wind
∇zs = gradient of terrain height
High Resolution Rapid Refresh (HRRR) model data
2015–2016 convective seasons
July 2015 (1, 9, 14, 18, 19, 24, 26, 28)
Chose 80-m wind because it has a stronger signal

10. Upslope Two criteria must be determined…
What is the minimum upslope flow threshold?
Top 5%
Specify that we calculated upslope along storm tracks
Lightning was used to calculate each storm track

11. Upslope Two criteria must be determined…
What is the minimum upslope flow threshold?
How many points along a storm track must meet the minimum upslope threshold?
Start: 1936Z
End: 2228Z
~ 50 miles
Top 5%
Specify that we calculated upslope along storm tracks
Lightning was used to calculate each storm track
= 1
= 74
19 July 2015
Λ >= 0.10 ms-1
Λ < 0.10 ms-1

12. Results: Part II
Without upslope filter (Sigma Level = 2.0, Flash Rate = 5 flashes/min)
POD = 83%; FAR = 84%
With upslope filter (Sigma Level = 2.0, Flash Rate = 5 flashes/min)
POD = 73% ; FAR = 80%
FAR is still too high… This is not very promising

13. Random Forest
Conditioning on upslope does not necessarily yield more skill
What other products might be useful in differentiating between severe and non-severe?
Maximum Reflectivity
Vertically Integrated Liquid (VIL)
Lightning Density
Echo Top – 50 or 60 dBZ
Let’s test them ALL using a random forest!
What is a random forest?
An ensemble learning method for classification that operates by constructing a multitude of decision trees
Think of the trees as deterministic models and the forest as an ensemble…

14. Decision Tree Dataset is broken into two parts: 2/3 is for training
1/3 is for testing
SPECIFY THAT THIS EXAMPLE IS FAKE

15. Decision Tree Training Nodes partition using best split
Dataset is broken into two parts:
2/3 is for training
1/3 is for testing
Upslope
Maximum Reflectivity
VIL
>= 0.10 ms-1
< 50 dBZ
>= 50 dBZ
< 50 mm
>= 50 mm
< 0.10 ms-1
SPECIFY THAT THIS EXAMPLE IS FAKE

16. Decision Tree Training Nodes partition using best split
Variables are weighted differently based on importance
Training
Nodes partition using best split
Dataset is broken into two parts:
2/3 is for training
1/3 is for testing
Upslope
Maximum Reflectivity
VIL
< 50 dBZ
>= 50 dBZ
< 50 mm
>= 50 mm
< 0.10 ms-1
>= 0.10 ms-1
SPECIFY THAT THIS EXAMPLE IS FAKE

17. Decision Tree Training Nodes partition using best split
Variables are weighted differently based on importance
Testing
Each tree “votes” for a class… The forest chooses the classification with the most votes
Non-Severe | Severe
8 | 5
Upslope
Maximum Reflectivity
VIL
< 50 dBZ
>= 50 dBZ
3 | 1
2 | 0
1 | 4
< 50 mm
>= 50 mm
< 0.10 ms-1
>= 0.10 ms-1
SPECIFY THAT THIS EXAMPLE IS FAKE

18. Decision Tree Training Nodes partition using best split
Variables are weighted differently based on importance
Testing
Each tree “votes” for a class… The forest chooses the classification with the most votes
How well did this forest do? Calculate verification metrics!
Non-Severe | Severe
8 | 5
Upslope
Maximum Reflectivity
VIL
< 50 dBZ
>= 50 dBZ
3 | 1
2 | 0
1 | 4
< 50 mm
>= 50 mm
< 0.10 ms-1
>= 0.10 ms-1
SPECIFY THAT THIS EXAMPLE IS FAKE

19. Decision Tree Severe Non-Severe Hit False Alarm 4 1 Miss Correct Null
Miss Correct Null
1+0+0 = = 7
Upslope
Maximum Reflectivity
VIL
< 50 dBZ
>= 50 dBZ
Non-Severe | Severe
8 | 5
3 | 1
2 | 0
1 | 4
< 50 mm
>= 50 mm
Non-Severe Severe
< 0.10 ms-1
>= 0.10 ms-1
FAR = 0.2
POD = 0.8
This was a pretty good example! Now let’s try it with some real data…
SPECIFY THAT THIS EXAMPLE IS FAKE

20. Decision Tree Severe Non-Severe Hit False Alarm ? ? Non-Severe Severe
? ?
Miss Correct Null
? ?
Upslope
No Jump | Jump
? | ?
Non-Severe Severe
FAR = ???
POD = ???
SPECIFY THAT THIS EXAMPLE IS FAKE
? | ?

21. Decision Tree Severe Non-Severe Hit False Alarm 369 151
Miss Correct Null
No Jump | Jump
Non-Severe Severe
Upslope
151 | 71
When… Sigma Level = 2.0
Flash Rate = 5 flashes/min
FAR = 0.29
POD = 0.84
SPECIFY THAT THIS EXAMPLE IS FAKE
151 | 369

22. Results: Part III Sigma Level = 2.0, Flash Rate = 5 flashes/min
POD = 84% ; FAR = 29%
FAR is so much better!!!
Correlation Coefficient = 0.46

23. Conclusions
Lightning jumps can be a valuable tool for diagnosing severe weather in regions of complex terrain
High false alarm rates suggest that lightning jumps are occurring in sub-severe storms
Random forests provide useful method for incorporating other variables such as upslope, maximum reflectivity, and VIL that may help to differentiate between severe and sub-severe events
Results support the hypothesis that upslope and lightning jumps are correlated, which verifies the findings of Markowski and Dotzek 2011
Thank you!