The NSSL Tornado Detection Algorithm (TDA), and Its Use for the 1996 Warning Decision Support System (WDSS) Proof-of-Concept (PoC) Tests
1. Purpose.
The purpose of this document is to describe the NSSL Tornado Detection Algorithm (TDA), how to interpret the output shown on RADS, and the advantages over and differences between the NSSL TDA and the Tornadic Vortex Signature (TVS) algorithm currently being used to using with the WSR-88D.
Full details of how information is displayed using RADS is contained within the RADS Display Operators Guide. The Operators Guide should be used as a supplement to this document, particularly the sections on overlays, tables, and trends for the TDA.
Also, attached to this documentation is a copy of a manuscript about the NSSL TDA which was presented at the 1995 AMS Radar Conference. It provides more details on the NSSL TDA.
2. The Algorithm.
The Tornadic Vortex Signature (TVS) is classically defined as a locally intense circulation indicated by strong shear on the order of .01 s-1 between two velocity gates which are azimuthally adjacent and constant in range (gate-to-gate) (Brown et al. 1978). Historically, the NSSL TDA was designed to detect such circulations associated with the non-supercell tornadoes sometimes referred to as landspouts. However, since 1991, the NSSL TDA has been designed and tested using data associated with a wide variety of tornadoes, supercell and non-supercell (Vasiloff 1991, Mitchell 1995, Mitchell 1996).
The NSSL TDA addresses the problem of extremely low probability of detection with the WSR-88D TV algorithm (88D TVS) without a high false alarm cost. The NSSL TDA extracts as much information as possible about locally intense circulations within the base velocity data and displays this information via the Radar Algorithm and Display System (RADS). The algorithm then classifies the circulation as either tornadic or non-tornadic near the end of the processing based upon the overall strength and size of the circulation.
The NSSL TDA uses very low velocity difference thresholds (default 15 m/s) to identify the individual gate-to-gate velocity pairs, or pattern vectors, which are characterized by relatively high shear. The pattern vectors (one-dimensional circulations) are conjugated into two-dimensional circulations, and two-dimensional circulations within 2.5 km of each other are then classified as three-dimensional circulations. The attributes for each three-dimensional circulation are passed through a rule base which classifies the circulation as either tornadic, potentially tornadic, or non-tornadic. This classification is based upon the depth, strength, and lower extent of the 3D circulation. Furthermore, past track and future position locations are maintained and computed for each detection. Tracking of persistent circulations allows the extraction of trends of important circulation attributes (e.g. maximum low-altitude gate-to-gate velocity difference, depth, base, etc.). According to a performance evaluation in 1995 using a database of about 50 tornadoes, the NSSL TDA has a Critical Success Index (CSI) of 36%, compared to only 3% for the default (TTS = 72 hr-1) 88D TVS.
3. Functional Comparison of the WSR-88D TVS Algorithm and the NSSL Tornado Detection Algorithm.
The current NSSL TDA works independently from all other algorithms and does not require a mesocyclone detection in order to function. Furthermore, the NSSL TDA examines the difference between gate-to-gate velocities, whereas the 88D TVS algorithm calculates the shear between the maximum inbound and outbound velocities within or very near a mesocyclone, regardless of whether the velocity maxima are gate-to-gate. In order to construct a three-dimensional circulation, the NSSL TDA incorporates a more stringent vertical association criteria between successive two-dimensional features at different elevation angles than does the 88D TVS. The NSSL TDA also has the advantage of retaining past tracks of persistent circulations and computing a forecast track for all circulations. In addition, the NSSL TDA provides trends for important diagnostic parameters. Trends of circulations from their incipient stages are possible within the NSSL TDA since very low velocity difference thresholds are used to identify even the weakest circulations prior to becoming tornadic.
The forecaster must be aware of the fundamental differences between the NSSL TDA and the 88D TVS. Namely, these differences are the examination of gate-to-gate velocity differences and the operation of the TDA independently from other algorithms. The NSSL TDA is much more sensitive to detections of weak incipient circulations prior to becoming tornadic than the default 88D TVS. Therefore, a forecaster must be more judicious in his/her decision making when using the NSSL TDA as a warning tool.
4. The NSSL TDA Output.
Output from the TDA is displayed in three ways: diagnostically in the RADS Tornado Algorithm Output table, as overlays on the radar images, and in trend plots.
The following is a description of each column in the Tornado Algorithm Output table, with units where appropriate.
TORNID - tornado identifier, lowercase letters a-z
TYPE - type of tornado detection, classified into two categories:
TVS - Tornado Vortex Signature: A three-dimensional circulation whose base extends to the 0.5 elevation angle or whose base is below 2,000 ft (600 m) AGL. This type of detection is denoted by a red inverted triangle and is coded red in the Cell and Tornado tables.
ETVS - Elevated Tornado Vortex Signature: A three-dimensional circulation whose base does not extend to the 0.5 elevation angle and whose base is above 2,000 (600 m) ft AGL. This type of detection is denoted by a yellow inverted triangle and is coded yellow in the Cell and Tornado tables.
Note: The user may specify whether the ETVS detections are to be displayed with the TVS_PLOT_ETVS_ONLY variable in the ssaparm.dat file. Setting this variable to true (T) specifies that only TVS type detections are to be displayed; setting it to false (F) displays both ETVS and TVS detections.
TVS (and ETVS) detections are displayed in the NSSL Cell Algorithm Table under the CIRC column if the detection is associated with a SCIT-identified storm cell. If a TVS is associated with a mesocyclone, then it is displayed in the Cell table as a TVSMES in the CIRC column.
AZ/RAN - Azimuth and range of the centroid of the base of a TVS or ETVS (degrees/km, nm).
CELLID - Cell ID of the closest storm identified by the NSSL Storm Cell Identification and Tracking (SCIT) algorithm.
MESOID - Mesocyclone ID of the closest associated mesocyclone detected by the NSSL Mesocyclone Detection Algorithm. A cyan-colored box in this column indicates an associated mesocyclone ID; a green-colored box indicates no associated mesocyclone.
BASE - Altitude AGL of the base of a TVS or ETVS (km, kft).
TOP - Altitude AGL of the top of a TVS or ETVS (km, kft).
DEPTH - The overall depth of a TVS or ETVS (km, kft).
LA GTG - The maximum low-altitude gate-to-gate velocity difference at the base of a detection (m/s), color-coded in the following manner:
RED: velocity difference > 40 m/s
YELLOW: velocity difference 20-39 m/s
GREEN: velocity difference < 19 m/s (but not below the minimum velocity difference)
Note: The minimum velocity difference that will be displayed is determined by setting the TDA_GTG_THR variable in the adaptable parameter (ssaparm.dat) file. It is recommended that the minimum gate-to-gate velocity difference threshold not be set below 10 m/s. The default minimum velocity difference threshold is 15 m/s.
MX GTG - Maximum gate-to-gate velocity difference within a three-dimensional circulation (m/s), color-coded in the following manner:
RED: velocity difference > 40 m/s
YELLOW: velocity difference 20 - 39 m/s
GREEN: velocity difference < 19 m/s (See preceding Note)
DIR/SPD - Direction and speed of motion of TVS or ETVS (degrees/m s-1, kts).
TDA detections are sorted in the table based on severity. The order of the sort is such that all TVSs are sorted above ETVSs (if ETVSs are displayed). The secondary sort is based upon the maximum low-altitude gate-to-gate velocity difference.
Each detection location, past track, and forecast track are displayed by overlays onto the radar data images. The current location is indicated by the center of an inverted red triangle for a TVS or a yellow inverted triangle for a ETVS. The past track is indicated by white dots and the forecast track is indicated by the magenta cross-hairs. A maximum of ten past positions (approximately one hour) are displayed for each detection. The forecast positions indicate the computed position at five minute intervals. The number of forecast positions which are displayed (up to six) equals the number of past track positions, for approximately a 30-minute forecast. Currently, the direction and speed are based upon the current and previous signature locations. If the detection is new, the direction and speed are determined (in rank order) from 1) an average motion vector of detections from the previous volume scan, 2) the nearest storm motion vector or 3) a default motion vector. Be aware that the NSSL TDA tracks all identified circulations. Therefore, if the option to display a TVS only is specified then the past identified circulations may have been either TVSs or ETVSs. In other words, a TVS track does not imply that the entire track consists only of TVS classified circulations.
Finally, diagnostic information is provided in trend plots. Trends may give a forecaster valuable guidance about whether a circulation is strengthening, weakening and/or descending or ascending. The TORN trend set contains time-series plots of the altitude of the base, the depth of the detection, the low-altitude gate-to-gate velocity difference, the maximum 3D gate-to-gate velocity difference, and the height of the maximum velocity difference. A trend plot of the altitude of the top of the detection as well as the trends comprising the trend set are also available in the single trend window.
4. Other information.
It is important to remember that algorithm products are not available to the user until the end of the volume scan, which is 5-6 minutes later than the lowest-tilt image which can be viewed from that volume scan. Thus, the corresponding low-altitude centroid position (the position of the TVS closest to the ground) is also 5-6 minutes old when it is available for display. This restriction should be taken into consideration when forecasting the position of the radar-indicated tornadoes in warnings. For example, use the pink cross-hairs to forecast the position of an element at time HHMM, and not "xx minutes from now".
As mentioned previously, the forecaster must exercise caution when interpreting the NSSL TDA output. Because the TDA is more sensitive to detection of weak circulations than the 88D TVS, forecasters must be more judicious in deciding whether to issue severe thunderstorm or tornado warnings. It is highly advisable that *ALL* available data be considered along with the TDA guidance when making a warning decision. Other information such as the presence of a pendant/hook echo protruding from the rear flank of the storm, (bounded) weak echo region [(B)WER] and the merger/collision between a storm and a surface boundary i.e. storm outflow gust fronts, mesoscale boundaries, etc. Remember, the NSSL TDA does NOT consider reflectivity structures.
The algorithm has been tested using a dataset consisting of tornadoes within 150 km of a WSR-88D. However, it is not necessarily recommended that the NSSL TDA be used to identify tornadic circulations at this extreme range. Instead, 100 km may be a more appropriate maximum range for the NSSL TDA to more reliably identify tornadic circulations.
Also, remember that due to limitations of the radar, especially beam broadening with range, that the actual vortex associated with the tornado may not be sampled. For example, a mesocyclone at far ranges (> 100 km) may appear "TVS-like." If the tornado is very large and close (within 15 km) to the radar the tornado may be sampled by more than two adjacent radar beams and appear "mesocyclone-like".
Recent evidence has led NSSL scientists to believe that the tornado may only be sampled in rare cases and that it is the tornado cyclone (an intermediate circulation between the tornado and mesocyclone) or the mesocyclone which is actually detected by the NSSL TDA (Mitchell and Stumpf 1996, Rasmussen and Straka 1996, Straka et al. 1996). The case study by Mitchell and Stumpf (1996) observed a small-scale circulation (presumably the tornado cyclone) embedded within a mesocyclone and other observations of near-range tornadoes appear to support such structure.
In the case of near-range tornadoes that may go undetected by the NSSL TDA, it is suggested that a forecaster pay particular attention to any mesocyclone or mesocyclone-like (tornado cyclone) signatures in the base velocity and reflectivity data. The NSSL Mesocyclone Detection Algorithm may be useful in these cases in identifying the broad-scale circulations (tornado cyclone or mesocyclone) and thus may provide useful guidance about the significance of the circulation and its tornadic potential.
The NSSL TDA and RADS contain many useful features which are unavailable in the current WSR-88D package. The goal of creating these algorithms is to provide better information to help guide the warning decisions faced by NWS forecasters. Any feedback from users concerning the usefulness of this document and the NSSL TDA is encouraged and greatly appreciated. Your efforts will enhance the creation of future versions of the algorithm.
5. References.
Brown, R.A., L.R. Lemon and D.W. Burgess, 1978: Tornado detection by pulsed Doppler radar. Mon. Wea. Rev., 106, 29-38.
Burgess, D.W., L.R. Lemon and R.A. Brown, 1978: Tornado characteristics revealed by Doppler radar. Geophys. Res. Lett., 2, 183-184.
Mitchell, E.D., 1995: An enhanced NSSL tornado detection algorithm. Preprints, 27th Conference on Radar Meteorology, Vail, CO, Amer. Meteor. Soc., 406-408.
Mitchell, E.D., 1996: OSF Final Report.
Mitchell, E.D., G.J. Stumpf, 1996: The 19 April 1995 Fort Worth/Dallas tornado event: Implications for automated vortex recognition. Preprints, 18th Conf. on Severe Local Storms, San Francisco, CA, Amer. Meteor. Soc., 574-576.
Rasmussen, E.N. and J.M Straka, 1996: Mobile mesonet observations of tornadoes during VORTEX-95. Preprints, 18th Conf. on Severe Local Storms, San Francisco, CA, Amer. Meteor. Soc., 1-5.
Straka, J.M., J. Wurman, E.N. Rasmussen, 1996: Observations of the low-levels of tornadic storms using a portable X-Band Doppler radar. Preprints, 18th Conf. on Severe Local Storms, San Francisco, CA, Amer. Meteor. Soc., pp. 11-16.
Vasiloff, S.V., 1991: The TDWR tornadic signature detection algorithm. Preprints, 4th International Conference on Aviation and Weather Systems, Boston, MA, Amer. Meteor. Soc., J43-J48.