AN ANALYSIS OF NSSL WDSS CIRCULATION
PARAMETERS DURING1999-2000 TORNADO EVENTS
IN THE NWS JACKSON, MISSISSIPPI, COUNTY WARNING AREA
Eric E. Carpenter, Alan E. Gerard, and
Ed R. Agre NOAA/NWS, Jackson, MS
The Jackson, MS National Weather Service (NWS) forecast office utilizes data from three Weather Surveillance Radar - 88 Doppler (WSR-88D) radars in their warning and forecast operations. The National Severe Storms Laboratory's (NSSL) Warning Decision Support System (WDSS) has become the primary method of analyzing this data stream. WDSS provides, through several algorithms, many parameters to the forecaster which enhance the analysis of convection and assist in the warning process. In order to better utilize data provided by the WDSS Mesocyclone Detection Algorithm (MDA), several parameters from this algorithm were analyzed from a collection of severe weather events in the Jackson County Warning Area (CWA) during 1999 and 2000. Statistical analyses of these parameters are being undertaken in an effort to find trends which might indicate improved warnings for tornadoes. Extensive algorithm analysis, including MDA analysis, has already been conducted by NSSL (Stumpf et al. 1999) on a national level. Some correlation of the NSSL findings are being made to the findings in this study.
2. DATA COLLECTION
JAN 88-D Archive II data was processed and displayed through WATADS, a program that runs the RADS image display program. The study focused on five parameters from NSSL's MDA. These parameters were maximum gate to gate velocity difference (MxGtG), maximum shear (MxShear), maximum rotational velocity (MxRotv), strength rank, and mesocyclone strength index (MSI). Values in the initial data set were associated with identified mesocyclones. This set of data was correlated with verified tornadoes listed in Storm Data, thus reducing the data set to algorithm output values associated primarily with tornadic mesocyclones.
For the purpose of this study, mesocyclones are considered tornadic if associated with ground truth tornadoes. This correlation includes the three volume scans prior to tornado touchdown. All other mesocyclone data are considered non-tornadic.
A large majority of the algorithm-identified mesocyclones incorporated into this study developed in highly sheared and slightly to moderately unstable environments. Nearly all occurred during the evening hours and over rural land at ranges of 45 to 185 km from the KJAN radar. The data set is based on the following storm data dates: 1-21-99, 2-27-99, 4-14-99, 12-10-99, and 1-3-00. The data set includes approximately 143 volume scans.
The data analysis is dependent upon accurate timing of tornado events. While the correlated tornadoes were well-documented, Storm Data isn't always exact. Furthermore, with most of the studied mesocyclones occurring over sparsely populated land after dark, it is hypothesized that some tornadoes may have been missed. As of this writing, the data set is rather small.
As mentioned earlier, five parameters from NSSL's MDA were included in the analysis. These parameters were MxGtG velocity difference (m s-1), MxShear (10-3 s-1), MxRotv (m s-1), strength Rank, and MSI. Among these parameters, strength rank and MSI are unique to the algorithm. Strength Rank is defined by NSSL (1998) as a non-dimensional number based on the following strength parameters: rotational velocity, shear, and GtG velocity difference. Table 1 shows the direct relationship between Rank numbers and mesocyclone strength. The MSI is also a non-dimensional value and is based on the vertical integration of the same three strength parameters that are incorporated into Rank calculations. The vertical integration is divided by the depth of the circulation and is weighted by density. This results in more of an emphasis on data from the lowest levels of the mesocyclone. Table 2 shows the direct relationship between MSI values and mesocyclone strength.
3 5 7 9 Weak to Minimal Moderate Strong Very Strong
Weak to Minimal
4.1 Subjective Correlation
The statistical interpretation of this small data set is somewhat subjective, and it doesn't take long for one to realize the variable nature of the data (see Figures 1,2,3,4,5). It is no surprise that statistical calculations reveal large deviations and variances for all analyzed parameters, thus severely thwarting attempts to find any magical algorithmic values for the warning decision-maker.
Of course, it is of great importance to compare the previously defined tornadic (T1) and nontornadic (T0) data subsets. Despite the data variability, discovering a unique sample would certainly lend to improved guidance for tornado warning decisions. In other words, is there a threshold for any parameter that may strongly suggest warning or no warning? In prior analysis, directly comparing the T1 and T0 subsets for all parameters proved to be statistically ambiguous for all but MSI. Scrutiny of the MSI data T1 and T0 subsets indicated that nearly half (48%) of the MSI T1 values were greater than the largest MSI T0 value of 3906. In comparison to other MDA parameters, this stood out as a useful threshold. However, recent additions to the data set have altered this finding. A non-tornadic mesocyclone from 3 January 2000 contained eight consecutive MSI values from eight consecutive volume scans greater than 3906 with values ranging from 4199 to 6788.
With much emphasis being placed on lowering of tornado warning false alarms, a search was made for minimum threshold values of tornado occurrences. In all, very little was found with MSI coming out on top. Even so, it's minimum T1 value of 1777 only topped a few T0 Values making it worthless statistically or otherwise.
4.2 Temporal Analysis
With direct comparisons of the T1 and T0 data not proving fruitful, a different perspective was taken by sub-dividing the T1 data temporally and focusing on time series trends. Fig. 6 illustrates parameter behavoir before and during tornado touchdowns. Notice that all parameters increased with time. However, the MxShear subset stood out with sharp increases (101% on average) in the fifteen minutes prior to tornado touchdown. The MxGtG subset showed lesser increases, but perhaps more importantly, indicated better lead time with the largest increases (37% on average) coming in the 15 to 10 minute time frame. These findings suggest that sharp increases in the MDA MxGtG and MxShear values may be red flags for tornado warning issuance.
5. FURTHER RESEARCH
Of most importance concerning the future of this local study is adding more data. Incorporation of WDSS data from KGWX and KMOB is desired along with data from closer ranges (<50 km). All mesocyclone data to this point have been in closer proximity to KJAN than either KGWX and KMOB, thus precluding the addition of data from either of the latter radars.
Also, it may be prudent to add additional MDA parameters to the study, particularly those defined by the lowest elevation scans (ie. low level shear), given shallow nature of cool-season meso- cyclones in the Jackson, MS area (eg. see Gerard et al. 2000).
For a complete listing of references, please contact the first author, Eric E. Carpenter1.
Fig. 1 Distribution of MxShear values for nontornadic and tornadic mesocyclones.
|Fig. 2 Same as Fig. 1 except for Mx Rotv|
|Fig. 3 Same as Fig. 1 except for Mx GtG|
|Fig. 4 Same as Fig. 1 except for Rank|
Fig. 6. Time intervals are represented by volume scan intervals: v-3 to v-2 (5 min), v-3 to v-1 (10 min), and v-3 to TD (15 min).