1. Introduction
Three-dimensional analyses of radar data from the Verification of the Origins of Rotation in Tornadoes Experiment 1995 (VORTEX; Rasmussen et al. 1994) by Rasmussen et al. (2006) revealed that, in several cases, development of a hook-echo appendage was preceded by the descent of a reflectivity core pendant from an echo overhang (Fig. 1). Descending from 3 to 6 km AGL, these reflectivity cores typically took 5–15 min to descend to the lowest elevation tilt used by the Weather Surveillance Radar-1988 Doppler (WSR-88D). This feature either helped create or intensify the reflectivity in an already present rear-flank echo appendage. Called the descending reflectivity core (DRC), it is often spatially associated with enhanced rear-to-front, low-level flow, on Doppler radar, and is presumed to be associated with outflow from a downdraft. Because the rear-to-front flow is spatially isolated, it is sometimes associated with counterrotating vortices (Fig. 2). In some regards, the DRC resembled the reflectivity “echo dot” appendage Forbes found during the Xenia, Ohio, supercell of 3 April 1974 (Forbes 1978; Fig. 3). Beyond this reference, however, the authors know of no other studies on echoes resembling the DRC. For brevity, a complete literature review on rear-flank appendages is omitted. Instead, the reader is referred to Markowski (2002), who offers a lengthy discussion on hook echoes and rear-flank downdrafts in supercells.
In Rasmussen et al. (2006), the DRC typically occurred prior to tornadogenesis, hence making it a topic worthy of further study. With its temporal and spatial occurrence around tornadoes, the DRC could be related to the onset of tornadogenesis. If this is validated or quantified, the DRC could have use for warning decision making for tornadoes. Analysis of the DRC also may be vital in obtaining a more complete understanding of the morphology of the rear-flank appendage of the supercell and related tornadogenesis.
The primary motivation for this paper is to serve as an extension to Rasmussen et al. (2006) by quantifying how often DRCs occur within a larger sample of supercells. With the inclusion of tornadic supercells, their relationship to tornadoes will also be ascertained. Further motivation extends from the necessity to analyze a large quantity of supercell radar data in three dimensions. While not discussed herein, detailed three-dimensional time analysis of reflectivity data within the rear flank of supercells has not been done before.
This paper is divided in the following manner: section 2 sets forth the methodology of the study and section 3 presents the results. Discussion of the findings is given in section 4, while conclusions and discussion of future research pertaining to the DRC are offered in section 5.
2. Data and methodology
To assemble a climatological sample of DRCs, a systematic process was necessary to make the study as objective as possible. This was accomplished by defining the domain for the dataset, selecting storms of interest, and, finally, analyzing radar data to locate DRCs.
a. Domain
WSR-88D level II data from the National Climatic Data Center (NCDC) were used to find storms of interest. Only radars in and around the southern plains were considered, which includes those within Kansas, Oklahoma, and northern Texas (Fig. 4). Because the DRC is an echo that evolves with height, it is necessary to utilize data from several elevation angles. By limiting the domain to the 20–100-km range, the lowest data were no more than 1.5 km AGL and the data extended upward to at least 6 km AGL for all storms.
To obtain a reasonable sample of supercells, radar data from the month of May in the years 2001–05 were examined. Level II radar data were acquired within ±6 h of either tornado reports or hail reports ≥1 in. in diameter, as obtained from the NCDC Storm Event Database. This process was expedited by quickly filtering storm reports through the use of Geographical Information System (GIS) software and SeverePlot (Hart 1993).
b. Storm selection
To make storm selection as objective as possible, a set of criteria was developed to classify those supercells that should be considered. Supercells were chosen for study by using a combination of both radial velocity and reflectivity signatures on radar as discriminators. While no condition was applied based upon mesocyclone strength, storms had to contain vertical continuity of cyclonic shear over at least three elevation tilts. This region of shear also had to lie within the area one would expect for a mesocyclone—to the right of the main reflectivity core and centered near the echo vault. In addition, these storms had to contain a persistent (existent during several volume scans) rear-flank appendage, a necessary requirement for the DRC. To minimize subjectivity, Forbes’s (1981; Fig. 5) classification scheme for rear-flank appendages of supercells was used. This scheme was originally implemented to discriminate which echoes were associated with tornadoes during the superoutbreak of 3–4 April 1974. To qualify as an appendage, the echo protrusion called the appendage had to be oriented 40° or greater from the storm motion vector. Finally, only isolated storms were considered, meaning those storms that existed without any significant interaction with other storms. Storms or other echoes along the rear flank of supercells often contaminated the reflectivity field and masked the occurrence of a DRC.
As might be expected, this study does not offer itself as representative of all supercells in all months, seasons, or geographies. Many storms were not considered due to contamination in the reflectivity field by other cells, falling outside of the 100-km range limit, or not having echo appendages. Further, other storms could not be considered due to missing WSR-88D data. This work should be treated as a statistical study of DRCs within isolated supercells with persistent rear-flank hook echo appendages.
c. Determination and analysis of DRCs
Rasmussen et al. (2006) set forth an objective way to determine which echoes within the rear-flank appendage of a supercell are DRCs. To qualify as a DRC, its echo must first be pendant from the echo overhang in the right-rear flank of the supercell. Once the DRC reaches the lowest elevation tilt, it must be associated with an isolated core of 4 dB greater than the highest value along the path of the appendage leading to the core. The purpose of these requirements is to ensure that the DRC echo is isolated and not just a typical appendage.
Volume scans that met the surface requirements for a DRC were selected using objective analysis. Additional scans were considered 10–15 min prior to detect the descent of the DRC. Data were objectively analyzed using a Barnes (1964) weighting function to smooth the data in a fashion similar to Rasmussen et al. (2006). To summarize this process, a weighting function was chosen to give approximately a 75% response at the worst four-sample wavelength (corresponding to four times the typical horizontal beam-to-beam separation at the range of the storm). Such a technique allowed the smoothing parameter κ to vary based upon the resolution of the WSR-88D data. The cutoff radius was adjusted between 1500 and 3000 m depending on the distance of the storm from the radar. These values offered a good compromise between retaining finescale reflectivity features and generating isosurfaces of reflectivity smooth enough to be readily interpreted. Due to the computational times, domains were constrained to regions of the storm immediately around the weak echo region (WER) and appendage. Typically, the domain was 10–25 km in horizontal extent with a vertical extent of 10–12 km. The grid resolution was fixed at 500 m, which does exclude some finescale reflectivity features near the radar; however, it is adequate for viewing DRCs, which are typically 2–3 km in diameter. As WSR-88D data with finer resolution in the azimuth direction and range [using azimuthal oversampling and removal of 4-bin averaging (Brown et al. 2005)] become available or mobile radar data are acquired from field experiments, smaller-scale DRCs may be discovered.
Tornado intensities, times, and locations were taken from the NCDC Storm Event Database. DRCs were marked tornadic if they reached the base-tilt elevation of the radar data within from 10 min prior to 5 min after tornado formation. A less stringent time period of from 30 min prior to 15 min after was also tested. The latter time period offers a better comparison to the DRCs studied in Rasmussen et al. (2006). No effort was taken to correct reports in cases where tornado times appeared erroneous, although the 15- and 45-min time window should mitigate some of the possibilities of falsely associating a DRC with no tornado.
3. Results
a. Frequency in supercells
Four months of data from 12 WSR-88D radars contained 64 isolated supercells with persistent appendages. Of these, 33 were tornadic, while the remaining 31 did not produce tornadoes. Despite the large percentage of tornadic supercells within this study, the reader should not be alarmed. Considering that the sample is constrained to isolated supercells with persistent appendages, it is not hard to fathom that these storms would be more likely to produce tornadoes; appendages are widely believed to be caused by the advection of hydrometeors around low-level mesocyclones and tornadocyclones (Fujita 1958; Browning 1965; Brandes 1977).
A breakdown of isolated supercells into groups of what phenomena they produced is shown in Tables 1a and 2a. The most common type was those that had both DRCs and tornadoes. The smallest category was storms that produced tornadoes with no DRCs. Nontornadic supercells were split nearly in half between those that did and did not produce DRCs. From the storms in this study, 89 tornadoes occurred. These break down into the classic distribution of exponentially more numbers of weaker tornadoes (Fig. 6). Seventy-one DRCs were recorded from the 39 DRC-producing supercells. Twenty of these 39 produced one DRC, while 18 of the remaining 19 storms had two to three DRCs reported. Two or three DRCs were recorded with 18 of those 19; one outlier produced 10 DRCs (as well as 13 tornadoes) over a span of 7 h across the domains of two radars.
A cursory glance at Tables 1a and 2a fails to reveal any immediate conclusions about the relationship between supercells and the types of phenomena they produce. Treating Table 1a as contingency table, however, allows one to predict if the DRC can correctly forecast whether a supercell will be tornadic. Values of 0.59 for the hit rate, 0.70 for probability of detection (POD), 0.41 for false alarm rate (FAR), and a critical success index (CSI) of 0.47 were calculated. If the occurrence of an appendage is used as the forecast, however, CSI actually improves to a value of 0.52 although the FAR also increases to 0.48. This is most likely a by-product of using a dataset where every supercell contains an appendage.
A more significant test involves using the chi-square test of association. In this instance, the null hypothesis is whether the type of supercell is independent of DRC production. The 2 × 2 contingency table for expected frequency is given in Table 1b while statistics for three methods are contained in Table 1c. For one degree of freedom, χ2.05 = 3.841 and χ2.1 = 2.706. In both cases, χ2 is smaller than these values and, thus, at the 0.05 and 0.1 significance levels, the null hypothesis cannot be rejected. The contributions of the residuals to the χ2 statistic are listed in Table 1d.
The chi-square test was also performed for Table 2a. Tables 2b and 2c display the 3 × 3 contingency table for expected frequencies and standardized residuals. For four degrees of freedom, χ2.05 = 9.488 and χ2.1 = 2.706. With χ2 = 4.84, the null hypothesis can once again not be rejected at the 0.05 and 0.1 significance levels. In this case, P = 0.35 for the Fisher exact probability test.
b. Association with tornadoes
By determining the time separating DRC occurrences and reported touchdowns of tornadoes, associations were determined between the two events. A DRC was associated with a tornado if it occurred within a time period of from 10 min prior to 5 min after a tornado report. The reasoning for this timing was based on the fact that the vast majority of volume scans analyzed were spaced by 5 min. A window of 15 min allows for errors in reported tornado times as well as allowing time for the DRC to impact the surface flow.
Observed frequencies of tornadic DRCs are given in Table 3. During this study, 21 of 71 DRCs were associated with tornadoes. Increasing the time period to from 30 min prior to 15 min after tornado onset resulted in 29 DRCs being associated with tornadoes. Nearly half of the tornadoes within the study (42 of 59) were included with this larger time period. Multiple tornadoes within time periods of several DRCs accounted for the larger increase in included tornadoes. While 45 min is too long of a time period with regards to supercell evolution, these numbers are included since they are believed to be the most comparable to the sample included in Rasmussen et al. (2006).
A worthwhile test to determine the practicality of the DRC as an aid for tornado warning nowcasting is to produce a 2 × 2 contingency table based on whether the occurrence or nonoccurrence of a DRC results in a tornado. Unfortunately, such a table could not be completed because the nonoccurrence of DRCs and tornadoes was unknown. POD, FAR, and CSI were computed since these are not dependant on this value. Assuming a tornado warning was issued for each occurrence of an appendage with a DRC, the POD was 0.3, FAR was 0.7, and the CSI was 0.15.
4. Discussion
The DRC is a precipitation protuberance that is pendant from the echo overhang in the right-rear quadrant of the supercell. With time, this echo descends until it appears as an isolated “bright” spot of higher reflectivity within the supercell appendage at the base scan of radar data. Similar to Rasmussen et al. (2006), this study found that DRCs occurred in both tornadic and nontornadic supercells. A substantial amount of supercells (41%) did not produce DRCs, although the smallest minority (16%) of the study was tornadic, non–DRC producing supercells. Within a small sample of storms, Rasmussen et al. (2006) found DRCs descended prior to every tornado. Even though this study has shown that the Rasmussen et al. sample was somewhat unrepresentative, 30% of DRCs descended within from 10 min prior to 5 min after reported tornado onset. While this may seem insignificant, the occurrence of the DRC was a better indicator of tornadogenesis when considering FAR or POD compared to the hook echo or other appendages; only 20% of 950 hook echoes at any given time, as defined by Forbes (1981), were associated with a tornado. How these two numbers relate is questionable; while the hook echo is determined at a single point in time, the DRC descends over a time period of 10–15 min. Statistical testing of the data displayed mixed results; while contingency tables produced a lackluster CSI, the chi-square-associated test failed to reject the null hypothesis that the occurrence of the DRC was independent of whether a supercell produced tornadoes.
Further complicating the situation was the discovery that some DRCs may be fundamentally different than others. Whereas some DRCs descended nearly vertically or toward the front flank of the storm into the hook echo, other DRCs descended toward the rear of the supercell into an appendage (Fig. 7). Related to this orientation, the DRC appeared to descend from kinematically different regions of the supercell at 3–6 km AGL. Even though a small sample size prevents any meaningful statistical testing, it may be possible that the orientation of the descent is important to whether the DRC prompts tornadogenesis. Such a hypothesis bears further study in the future.
Although WSR-88D data quality issues such as range folding and noise precluded a thorough study of velocity data surrounding the DRC, an attempt to subjectively analyze these data was made. The magnitude of outflow in the rear flank of the supercell for DRCs within 60 km (where the beam height is approximately 1 km AGL) was analyzed prior to and after descent of DRCs. Whereas some DRCs did not appear to influence single-Doppler velocities at the base scan, the majority (65%) did. This included many DRCs that were not associated with tornadoes (Fig. 8). When this outflow was enhanced, outflow representative of the RFD was already present; the DRCs occurred after the initial development of the RFD. For storms that had a location nearly due east or west of the radar, this outflow resembled a region of enhanced rear-to-front flow like Rasmussen et al. (2006) noted. Once again, this flow was typically straddled by counterrotating vortices, with the cyclonic vortex sometimes associated with the incipient tornado cyclone. Whereas this occurs with the majority of DRCs, some cases do not exhibit such characteristics. At the present time, we are unable to provide reasoning or speculation on why this may be the case. The dataset of single-Doppler results limits the amount of information that can be gained on the kinematic structure of DRCs and their parent supercell.
5. Conclusions
Given the issues listed above it is concluded that the exact statistical relationship between the DRC and tornadogenesis remains, for the most part, unknown. DRCs were found to be quite common in both tornadic and nontornadic supercells. The data suggest forecasters should have heightened awareness when both phenomena are observed within an isolated supercell. The reader is reminded, however, that DRCs should not be used as a condition necessary for tornado onset; a large number of tornadoes occurred outside of from 10 min before to 5 min after reported tornado onset. Further complicating the matter are the conditions set forth for this study. Even if DRCs had a perfect relationship with tornadogenesis in isolated supercells with appendages, this knowledge would only be useful in a minority of situations where tornado warnings are issued. The fact remains that the issuance of tornado warnings will always be limited by the resolution and height of the radar data. Given the improvements to the WSR-88D network in the upcoming years, it will be necessary to reexamine the utility of the DRC with the focus being on whether one can discriminate between those that precede tornadogenesis by their location and descent orientation. Ideally, this future study will include the number of null events by determining whether individual low-level mesocyclones with and without DRCs produced tornadoes. Such a study would also include warning verification statistics from weather service offices to determine whether DRCs that were tornadic also enhanced lead time.
Many other questions pertaining to DRCs remain unanswered. The microphysical makeup of these reflectivity protuberances is unknown, although research using dual-polarimetric radar is currently under way. The dynamical significance of the DRC is also undetermined. Can the DRC instigate (or even hinder) tornadogenesis, or is it a mere association? How does the location (classification) of the DRC impact supercell dynamics? It is hoped that these questions may be answered with future work involving dual-Doppler analysis of supercells as well as idealized numerical simulations. An additional path of research involves the use of mobile Doppler radar platforms such as the Doppler on Wheels (DOWs; Wurman et al. 1997) and the Shared Mobile Atmospheric Research and Teaching radars (SMART-Rs; Biggerstaff et al. 2005). Such radars offer superior resolution to the WSR-88D network and could answer the question of whether we can or cannot see some DRCs due to limitations in the WSR-88D network.
Acknowledgments
This work was funded by NSF Grant ATM 0340693. The authors thank Jim Ladue and two anonymous reviewers for their excellent comments. Their suggestions significantly clarified the text.
REFERENCES
Barnes, S. L., 1964: A technique for maximizing details in numerical weather map analysis. J. Appl. Meteor., 3 , 396–409.
Biggerstaff, M. I., and Coauthors, 2005: The Shared Mobile Atmospheric Research and Teaching radar: A collaboration to enhance research and teaching. Bull. Amer. Meteor. Soc., 86 , 1263–1274.
Brandes, E. A., 1977: Flow in a severe thunderstorm observed by dual-Doppler radar. Mon. Wea. Rev., 105 , 113–120.
Brown, R. A., Flickinger B. A. , Forren E. , Schultz D. M. , Sirmans D. , Spencer P. L. , Wood V. T. , and Ziegler C. L. , 2005: Improved detection of severe storms using experimental fine-resolution WSR-88D measurements. Wea. Forecasting, 20 , 3–14.
Browning, K. A., 1965: Some inferences about the updraft within a severe local storm. J. Atmos. Sci., 22 , 669–677.
Forbes, G. S., 1978: Three scales of motions associated with tornadoes. NUREG/CR-0363, U.S. Nuclear Regulatory Commission, 359 pp. [NTIS PB288291.].
Forbes, G. S., 1981: On the reliability of hook echoes as tornado indicators. Mon. Wea. Rev., 109 , 1457–1466.
Fujita, T. T., 1958: Mesoanalysis of the Illinois tornadoes of 9 April 1953. J. Meteor., 15 , 288–296.
Hart, J. A., 1993: SVRPLOT: A new method of assessing and manipulating the NSSFC Severe Weather data base. Preprints, 15th Conf. on Severe Local Storms, Baltimore, MD, Amer. Meteor. Soc., 40–41.
Markowski, P. M., 2002: Hook echoes and rear flank downdrafts: A review. Mon. Wea. Rev., 130 , 852–876.
Rasmussen, E. N., Straka J. M. , Davies-Jones R. P. , Doswell C. E. III, Carr F. , Eilts M. , and MacGorman D. R. , 1994: Verification of the Origins of Rotation in Tornadoes Experiment: VORTEX. Bull. Amer. Meteor. Soc., 75 , 995–1006.
Rasmussen, E. N., Straka J. M. , Gilmore M. S. , and Davies-Jones R. , 2006: A preliminary survey of rear-flank descending reflectivity cores in supercell storms. Wea. Forecasting, 21 , 923–928.
Wurman, J., Straka J. M. , Rasmussen E. N. , Randell M. , and Zahrai A. , 1997: A portable pencil-beam pulsed Doppler radar. J. Atmos. Oceanic Technol., 14 , 1502–1512.
Three-dimensional perspective view from the south of the 40-dBZ isosurface prior to tornado formation in the Dimmett, TX, tornadic storm of 2 Jun 1995. Times are (a) 0042, (b) 0047, (c) 0052, and (d) 0057 UTC. The descending reflectivity core is marked DRC. [Figure adapted from Rasmussen et al. (2006)]
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
DRC evolution for the case within Fig. 4 from Rasmussen et al. (2006) (left) The 0.5° PPI reflectivity (dbZ) and (right) the ground-relative Doppler velocity (m s−1) at five times as labeled. Scale is located on the right. (b) The radar pointing angle is the thin arrow. The DRC is marked with a solid black circle in the reflectivity data, and a white circle at the same position in the velocity data. The sense of Doppler shear is illustrated with the curved cyan arrows. [Figure adapted from Rasmussen et al. (2006)]
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
Hook echo initiation by way of the “echo dot.” [From Forbes (1978)]
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
Spatial domain for the study. The range rings are 20 km and 100 km distant from each WSR-88D site.
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
Forbes’s classification scheme for supercell echo appendages. Angle was determined by measuring the difference between the storm motion and orientation of the appendage. For angles greater than 40°, 60°, and 80°, echoes were classified as appendages, hooklike echoes, and hook echoes, respectively. [Adapted from Forbes (1981)]
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
Histogram of the number of tornadoes vs F-scale rating. As expected, there were exponentially more tornadoes of weaker intensity.
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
DRCs that descended in a variety of ways with respect to the storm motion vector: (top) forward, (middle) vertical, and (bottom) rearward.
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
Examples of nontornadic DRCs with an increase in the magnitude of velocities within the rear-flank outflow at the 0.5° elevation angle. Radar images are PPIs of base-tilt reflectivity, storm-relative velocity, and the 45-dBZ 3D reflectivity isosurface for the corresponding volume scans. Times increase from top to bottom. (a) Volume scans from 2229, 2234, and 2239 UTC 6 May 2001 at Twin Lakes, OK (KTLX). (b) Volume scans from 0153, 0158, and 0203 UTC 16 May 2002 at Dodge City, KS (KDDC).
Citation: Weather and Forecasting 22, 6; 10.1175/2007WAF2006095.1
Table 1a. A 2 × 2 contingency table for actual observed frequencies of tornadoes and DRCs in supercells.
Table 1b. A 2 × 2 contingency table for expected frequencies of tornadoes and DRCs in supercells.
Table 1d. Standardized residuals for the 2 × 2 contingency table.
Table 2a. A 3 × 3 contingency table for observed frequencies of tornadoes and DRCs in supercells.
Table 2b. A 3 × 3 contingency table for expected frequencies of tornadoes and DRCs in supercells.
Table 2c. Standardized residuals for the 3 × 3 contingency table.
Observed frequencies of tornadic DRCs and the number of tornadoes included within the designated time interval.
