Performance and Comparison of the WSR-88D and NSSL's TDA
during the Halloween 2000 Tornado Outbreak

 

David D. Thede
WFO Goodland

 

1. INTRODUCTION

On October 31, 2000 several supercell thunderstorms quickly developed within the WFO GLD county warning area. As these storms moved northeast at nearly 50 mph, they evolved into a broken squall line. Ahead of the developing squall line supercells developed and moved into the WFO LBF county warning area. A total of twelve tornadoes were reported by a variety of sources including NWS trained weather spotters, the public, and law enforcement officials. The KGLD radar was operating in VCP11 with a set of locally derived parameters (Thede, 1999).

2. METHODOLOGY

The twelve tornadoes that occurred within the WFO Goodland and WFO North Platte County Warning Areas were examined using KGLD WSR-88D archive II data. Beginning and ending times were defined as when the thunderstorms that eventually produced the tornadoes entered the Goodland CWA and when they exited the Goodland CWA.

The parameters that were adjusted for the Build 10 and NSSL TDA include the 3D feature depth (depth), minimum 3D feature low-level delta velocity (ladv), and minimum TVS delta velocity anywhere in a 3D feature (mxdv). The range was maximized at 230km. The ROC default, maximized, tropical, and supercell parameter sets were included as well as 100 other parameter sets. For the TDA analysis, algorithm performance was evaluated for each supercell (some containing multiple tornadic events) using WATADS software (NSSL 2000). The Witt (1998) scoring method was used: each tornadic time window included any volume scans occurring up to 20 minutes (three or four volume scans) prior to the beginning time of the event. The leading 20 minutes of the time window is chosen to allow the algorithm to diagnose the detections with sufficient lead time. The trailing 5 or 6 minutes of the time window is to account for any small reporting errors in the ending time of the events (Stumpf et al. 1998). False alarms were defined as any TVS detections not associated with a tornadic cell within the time window.

Prior to use in algorithm scoring, the tornado reports had to be validated. For those reports deemed likely to be in error, report times and/or report locations were modified to make them consistent with the radar data using techniques similar to Stumpf et al. (1998). Scoring techniques were also similar.

The adaptable parameter sets listed in Table 1 were tested. To establish the optimized parameter set for the KGLD radar for this event, various combinations of depth, ladv, and mxdv were used. The minimum 3D feature depth was allowed to vary from 1.0 km to 5.0 km (every 1 km). The minimum 3D feature low-level delta velocity (ladv) was allowed to vary from 26 m s-1 to 56 m s-1(every 10 m s-1). The minimum TVS delta velocity anywhere within the 3D circulation (mxdv) was allowed to vary from 34 m s-1 to 74 m s-1 (every 10 m s-1).

3. RESULTS

Table 1 shows the top 5 TDA performance statistics based on CSI for 100 independent parameter sets and the four ROC approved parameter sets for the NSSL TDA. Table 2 shows the top 5 TDA performance statistics based on CSI for 100 independent parameters and the four ROC approved parameter sets for the WSR-88D Build 10 TDA. A total of 91 volume scans were analyzed.

 

TABLE 1. POD, FAR, and CSI from the NSSL TDA.
RankDepthMin LADVMXDVPODFARCSI
1 1.0 26 74 .314 .238 .286
  1.0 26 64 .314 .238 .286
3 1.0 26 54 .314 .273 .281
4 1.5 25 36 .333 .370 .279
5 1.0 26 34 .333 .414 .270
6 2.0 26 34 .333 .414 .270
others
  1.6 27 27 .333 .528 .243
  3.1 27 30 .275 .533 .209
  5.0 56 74 .000 .000 .000


TABLE 2. POD, FAR, and CSI for the WSR-88D Build 10 TDA.
RankDepthMin LADVMXDVPODFARCSI
1 1.0 26 74 .412 .382 .328
  1.0 26 64 .412 .382 .328
2 1.0 26 54 .412 .400 .323
3 2.0 26 74 .373 .321 .317
4 2.0 26 64 .373 .345 .311
5 1.5* 25 36 .431 .488 .306
* default
others
  2.0(current) 26 34 .392 .444 .299
  1.6(squalline) 27 27 .431 .577 .272
  3.1(supercell) 27 30 .333 .514 .246
  5.0(minmizd) 56 74 .020 .000 .020


Figures 1- 3 compile the POD, FAR, and CSI from Table 1 (NSSL TDA) and compare the results to the Build 10 TDA values.



Figure 1
FIGURE 1. COMPARISON OF TABLE 1 AND VALUES FROM BUILD 10 TDA POD.

 

Figure 2
FIGURE 2. COMPARISON OF TABLE 1 AND VALUES FROM BUILD 10 TDA FAR.

Figure 3
FIGURE 3. COMPARISON OF TABLE 1 AND VALUES FROM BUILD 10 TDA CSI.



4. CONCLUSIONS

The WSR-88D Build 10 TDA had a slight improvement over the NSSL TDA during the October 31, 2000 tornado outbreak. Comparisons of these two algorithms indicate that the optimized WSR-88D TDA CSI had a 4.2% increase over the NSSL TDA, a 9.8% improvement in POD, but a 14.4% increase in FAR. The results also show that the present GLD WSR-88D TDA settings (Thede, 1999) were within 3% of the optimized parameters as were the ROC's default settings. NWS offices that do not perform research into optimizing the WSR-88D TDA for their particular location can feel some security that the ROC's default parameters (installed with their Build 10 upgrade will provide superior performance over the minimized (Build 9.0) parameters.

ACKNOWLEDGEMENTS

The author would like the following WFO Goodland staff: Aaron Johnson (intern) for his assistance with graphics and Llyle Barker (SOO) for his time reviewing this document.

REFERENCES

NOAA, NSSL, 2000: WATADS (WSR-88D Algorithm Testing and Display System) Reference Guide for Version 10.2. National Severe Storm Laboratory, Storm Scale Applications Division, Norman, OK.

Stumpf, G.J., A. Witt, E. D. Mitchell, P. Spencer, J.T. Johnson, M.D. Eilts, K.W. Thomas, D. Burgess, 1998: The National Severe Storms Laboratory Mesocyclone Detection Algorithm for the WSR-88D.Wea. and Forecasting, 13, 304-326.

Thede, D., 1999: Performance of the Build 10.0 Tornado Detection Algorithm (TDA) During Tornadic and Non- Tornadic (Large Hail) Situations. National Weather Service Central Region Technical Attachment 99-01, NWS, Kansas City, Mo.

Witt, A., M.D. Eilts, G.J. Stumpf, E.D. Mitchell, J.T. Johnson, K.W. Thomas, 1998: Evaluating the Performance of the WSR-88D Severe Storm Detection Algorithms. Wea. and Forecasting ,13, 513-518.

 


USA.gov is the U.S. government's official web portal to all federal, state and local government web resources and services.