A mobile, phased-array Doppler radar, the Mobile Weather Radar, 2005 X-band, Phased Array (MWR-05XP), has been used since 2007 to obtain data in supercells and tornadoes. Rapidly updating, volumetric data of tornadic vortex signatures (TVSs) associated with four tornadoes are used to investigate the time–height evolution of TVS intensity, position, and dissipation up through storm midlevels. Both TVS intensity and position were highly variable in time and height even during tornado mature phases. In one case, a TVS associated with a tornado dissipated aloft and a second TVS formed shortly thereafter while there was one continuous TVS near the ground. In a second case, the TVS associated with a long-lived, violent tornado merged with a second TVS (likely a second cyclonic tornado) causing the original TVS to strengthen. TVS dissipation occurred first at a height of ~1.5 km AGL and then at progressively higher levels in two cases; TVS dissipation occurred last in the lowest 1 km in three cases examined. Possible explanations are provided for the unsteady nature of TVS intensity and a conceptual model is presented for the initial dissipation of TVSs at ~1.5 km AGL.
Recent observational research of supercells using mobile Doppler radars has focused on detecting small-spatial-scale features within supercells not sufficiently resolved by the network of Weather Surveillance Radar-1988 Doppler (WSR-88D). Tornadoes, in particular, occur on spatial scales (Δx = Δy = ~100 m) often much smaller than that which can be resolved by WSR-88D radars, even at close ranges (e.g., Wurman and Gill 2000; Bluestein et al. 2003). The increased spatial resolution of mobile Doppler radar systems such as the Doppler on Wheels (DOW; Wurman et al. 1997) ~3-cm wavelength (X band) systems and the University of Massachusetts ~3-mm wavelength (W band) system (Bluestein et al. 1995) have allowed for unique observations of supercells and tornadoes to be obtained. Examples include observations in tornadoes of multiple vortices (Wurman 2002) and very narrow pendants of reflectivity in hook echoes (e.g., Bluestein and Pazmany 2000), among others.
Tornadoes are thought to evolve over very short time scales, as short as <10 s (e.g., Bluestein et al. 2003, 2010), in addition to the small spatial scales over which they occur. If one is interested in tornado processes or tornado evolution, then increased volumetric temporal resolution commensurate with that scale is desirable. Volumetric updates from the WSR-88D network occur, at a minimum, every ~270 s and most ground-based (airborne), mobile Doppler radar systems have volumetric update times of ~100 (~300) s. Furthermore, to study the volumetric evolution of tornadoes, space-to-time conversions are often needed in constructing RHIs and vertical cross sections. In turn, the conversions require what is likely a dubious assumption in some situations: that the tornado is not evolving significantly over the time it takes to complete a volume. A lack of volumetric observations of tornadoes with update times sufficiently small enough to examine short-time-scale processes represents a shortcoming in severe storms observational research.
In the springs of 2007–11, the first mobile, ground-based, phased-array, Doppler radar used in severe storms research, the Mobile Weather Radar, 2005 X-band, Phased Array (MWR-05XP; Bluestein et al. 2010), was used to investigate the short-time-scale evolution and time–height evolution of four mesocyclone tornadoes. Volume scans from 1° to 20° and to as high as 40° in elevation angle were obtained in as little as 6 s for these datasets. As a result, the data can be used to investigate rapid changes in the strength of tornadic vortex signatures (TVSs; Brown et al. 1978) associated with tornadoes. In addition, the manner in which the MWR-05XP obtains data allows for the height evolution of TVSs to be investigated without the need for steady-state assumptions over the time it takes for a volume scan to be completed. Previously, French et al. (2013) used MWR-05XP data to investigate the volumetric evolution of TVSs during three tornadogenesis cases. Also, two other mobile, “rapid-scan” Doppler radars, the Rapid-Scan DOW (Wurman and Randall 2001) and the Rapid-Scanning, X-band, Polarimetric Doppler radar (RaXPol; Pazmany et al. 2013), have been used to study supercells and tornadoes (e.g., Kosiba et al. 2013; Pazmany et al. 2013).
The following research questions are centered on topics in tornado science that are best addressed by utilizing unique rapid-scan volumetric observations:
What is the time–height distribution of TVS intensity through storm midlevels (2–6 km AGL)?
What is the time–height distribution of TVS position (i.e., vertical orientation) through storm midlevels?
Is there vertical directionality to the TVS decay process?
The above topics are not well represented in past work using mobile Doppler radar data. Past studies tracking changes in tornado intensity in time and/or height have noted the important influence of observed or inferred multiple vortices in modulating tornado flow (e.g., Bluestein and Pazmany 2000; Wurman 2002; Lee and Wurman 2005; Alexander and Wurman 2005; Marquis et al. 2008; Kosiba and Wurman 2010). In addition, it generally has been found that the difference between maximum and minimum radial velocities (ΔV) and/or axisymmetric vertical vorticity (AVV) in tornado signatures decreases with increased height (e.g., Wurman and Gill 2000; Burgess et al. 2002; Alexander and Wurman 2005; Wurman et al. 2007b; Alexander 2010). However, in the above studies, changes in tornado intensity were examined every ~10 s at one level or volumetrically every 60–90 s. We aim to assess changes in TVS ΔV up through storm midlevels every ~10 s with the hope that we can better understand the steadiness (or lack thereof) of tornado intensity.
The time–height evolution of tornado position has been studied using both visual and radar observations. Visual observations of tornado tilting in the subcloud layer are ubiquitous, particularly as a tornado nears the end of its life cycle, in the dissipating stage (e.g., Golden and Purcell 1977, 1978; Moller 1978; Wakimoto and Martner 1992). Radar studies of vortex signature position with height generally have found that tornadoes tilt toward the north and either toward the east or west with increased height in the Northern Hemisphere (e.g., Brown et al. 1978; Wakimoto and Martner 1992; Wurman and Gill 2000; Lee and Wurman 2005; Alexander and Wurman 2005; Tanamachi et al. 2012), presumably owing to vertical gradients of horizontal vertical vorticity advection. However, most of the above studies considered data only from the lowest ~2 km AGL. Little has been documented about the vertical orientation of tornadoes in the cloud layer, or perhaps more importantly, how tornado orientation changes in time. The tilt of a tornado in the cloud layer may indicate something dynamically important about features internal to the tornado and/or characteristics of the environment in which the tornado is embedded.
Many observational and numerical modeling studies of tornadoes have discussed factors influencing tornado longevity and/or potential causes of tornado dissipation (e.g., Lemon and Doswell 1979; Brandes 1981; Wicker and Wilhelmson 1995; Dowell and Bluestein 2002a,b; Markowski et al. 2002; Wurman et al. 2007a, 2010; Marquis et al. 2012b). However, there is little past work focused on the time–height evolution of vortex decay. A few early analyses of TVSs associated with tornadoes found that TVSs dissipated at all heights at roughly the same time (e.g., Brown et al. 1978; Vasiloff 1993). Wakimoto and Martner (1992) observed a tornado that dissipated as outflow associated with a rain shower impacted the vortex in the subcloud layer; a visual break in the vortex was observed above the outflow layer at ~1.5 km AGL. However, in most tornado studies utilizing mobile Doppler radar data, the dissipation process either was not captured or the data lacked the vertical–temporal resolution necessary to assess how the tornado dissipated in height. MWR-05XP data of the TVS decay process are used here to investigate the vertical progression of tornado dissipation.
In section 2, some brief background information about the MWR-05XP and the datasets used in this study is provided. Section 3 includes detailed, rapid-scan observations of TVSs associated with four tornadoes. In section 4, possible explanations for the unique radar observations of TVS intensity, position, and dissipation are provided. The results are summarized in section 5.
a. The MWR-05XP
MWR-05XP data of TVSs associated with four tornadoes obtained from 2009–11 are used in this study. The most notable characteristic of the MWR-05XP is its electronic scanning in elevation. In 2009 and 2011, the radar center frequency was ~9.5 GHz. The frequency was altered in 2010 to ~9.9 GHz to prevent interference with other X-band radars being used in the second Verification of the Origins of Rotation in Tornadoes Experiment (VORTEX2; Wurman et al. 2012). Other notable characteristics of the radar system are a peak power of ~15 kW, a gate length of 150 m sampled every ~75 m, and a half-power beamwidth of 1.8° (2.0°) in azimuth (elevation) sampled every ~1.5° (1.5°). Additional information about the MWR-05XP, including details about how electronic scanning is accomplished, can be found in Bluestein et al. (2010).
The MWR-05XP attribute most beneficial to studying vortex signatures associated with tornadoes is the volumetric (i.e., up to storm midlevels or higher) update time of ~10 s. To ensure that volumetric update times are kept at ~10 s, only limited spatial oversampling is employed. As a result, the coarse spatial resolution of the MWR-05XP prevents all but the largest tornadoes from being spatially resolved (Carbone et al. 1985). In this study, azimuthal shear signatures are used to estimate properties of tornado location and intensity as in past studies (e.g., Brown et al. 1978; Trapp et al. 1999), albeit imperfectly. An additional drawback to the MWR-05XP is the lack of truck levelers and a functioning system to measure radar tilt after 2008.1 Deployment sites for the five datasets discussed here were all free of noticeably large changes in elevation (Fig. 1, left). Details about expected errors in vortex positions are discussed in section 3. Estimated errors in the heights of MWR-05XP TVS observations (≤500 m magnitude in most cases) do not significantly change any of the conclusions reached herein. Nonetheless, listed height values are merely approximations and great care should be taken in their interpretation. For a complete positional error analysis, see French (2012).
Data were processed and edited in the same way as that described in French et al. (2013). In addition, great time and care was spent in dealiasing radial velocity data in vortex signatures from the ~6000 plan position indicator (PPI) scans used for this study. The inherent subjectivity in manually dealiasing radar data (automatic algorithms failed to consistently dealias vortex signatures correctly) is an inevitable source of error in the estimation of TVS intensity.
A brief overview of the datasets used in this study is provided (Table 1). The four tornadoes under examination occurred (i) on 5 June 2009 in Goshen County, Wyoming; (ii) on 19 May 2010 near Kingfisher, Oklahoma; (iii) on 25 May 2010 near Tribune, Kansas; and (iv) on 24 May 2011 near El Reno, Oklahoma. The first three tornadoes were observed during VORTEX2 and the fourth during an annual spring field experiment.
The MWR-05XP team was able to capture the entire life cycle of the Goshen County tornado (hereafter GC tornado; Fig. 1a) during year one of VORTEX2. Data from the GC tornado will be used to address all three questions discussed in section 1. Several studies have examined the GC supercell (e.g., Wakimoto et al. 2011, 2012; Markowski et al. 2012a,b; Wurman et al. 2012, 2013; Atkins et al. 2012; Kosiba et al. 2013; French et al. 2013), but the aforementioned studies did not focus specifically on the questions posed in section 1. Furthermore, the MWR-05XP was the only rapid-scan radar that captured the entirety of the tornado’s life cycle and consistently scanned the tornado up through storm midlevels.
In year two of VORTEX2, two partial datasets of tornadoes near Kingfisher (Fig. 1b) and Tribune (Fig. 1c) were obtained. In both cases, data collection began as the tornado likely was in its dissipation phase, so the focus of analysis in both cases will be in addressing the vertical evolution of tornado dissipation through storm midlevels. Finally, in the spring of 2011, data were obtained on a violent tornado near El Reno (Pazmany et al. 2013; French et al. 2013). Two separate deployments were undertaken while the tornado was ongoing. Data from the first deployment (Fig. 1d) and part of the second deployment (Fig. 1e) were used for this study. The deployments covered time periods from just as the tornado formed through part of the tornado mature phase. Therefore, the focus will be on the time–height evolution of tornado intensity and position.
3. Rapid-scan, volumetric observations of tornadoes
MWR-05XP observations of the TVSs associated with four tornadoes described previously are presented in this section. Consistent with French et al. (2013), a shear signature must have gate-to-gate (GTG) ΔV ≥ 20 m s−1 and contain local maxima and minima (in azimuth or range) separated by ≤2 km for it to be considered a TVS in this study. Tornado intensity is estimated by using the maximum ΔV in the TVS associated with the tornado2 in question. For most of the TVS observations, the maximum ΔV also was the maximum GTG ΔV, so the distance between radial velocity maxima is not examined. The lack of TVS spatial information is also why AVV, which requires knowledge about the distance between peak inbound and outbound radial velocities in a TVS, is not used to estimate TVS intensity. The tornado dissipation time is calculated as the radar scan time prior to the shear signature failing to meet the TVS criteria in 60 s of continuous data.
The MWR-05XP’s relatively close range to the tornadoes studied herein (Fig. 2a) and the use of azimuthal oversampling in data collection allows for meaningful information to be extracted from the TVS data. However, owing to the inability to resolve the tornado, the strength of a TVS likely is an indeterminable underestimate of the true strength of the tornado at the level sampled (e.g., Brown et al. 1978). In this study, we are concerned less with the absolute intensity of the tornado than with the relative time–height change in TVS intensity as measured by the MWR-05XP. Furthermore, it has been shown that when the radar aspect ratio (the ratio of radar beamwidth to the radius of the tornado core flow) is relatively large, the radial velocities in shear signatures can change not from physical changes in the tornado, but rather from the position of the radar beam compared to the location of the vortex (Wood and Brown 1997). As a result, caution must be used when interpreting seemingly random, short-time-scale changes in TVS intensity.
Another issue that could impact TVS intensity calculations is the possibility that there is a relative bias in intensity when TVS range, and therefore azimuthal resolution, changes significantly (Fig. 2a) within a deployment. Because of this effect, TVS ΔV for the GC (second El Reno) deployment might become progressively larger (smaller) relative to TVS ΔV at earlier times as the tornado moved toward (away from) the MWR-05XP. Also, as the elevation angle increases, less (more) of the horizontal (vertical) radial component of the wind in a TVS is being sampled. In particular, TVS data from 20° to 40° elevation angle in the first El Reno deployment should be viewed cautiously. All other TVS data used in this study are from ~20° elevation angle and below. Similarly, TVS ΔVs may be deflated owing to radar sampling of a vortex that is not vertically oriented. In particular, this effect could bias the analysis of TVS dissipation if there is a persistent horizontally oriented vortex that appears as a weak or dissipated vertically oriented vortex in MWR-05XP data.
In addition to radial velocity, the other main piece of information recorded in examining single-Doppler radar data is the position of the TVS in space. As mentioned previously, there were no truck levelers and no functioning system to record truck tilt. In 2007–08, it was found that pitch angles3 were small (deployments occurred with the truck parallel to the road) and roll angles rarely exceeded ±3° when the radar was positioned on relatively level roads. In an error analysis performed by French (2012), maximum vertical positional errors were found not to exceed ±750 m for roll angles of ±3° when echoes were within a 20-km range (Fig. 2b); errors in four of the five deployments generally were less than ±500 m and had little dependence on elevation angle. Zonal (i.e., in the horizontal direction orthogonal to the radar truck) positional error magnitudes were found to be ≤300 m at 10°–20° elevation angle (Fig. 2c); minimum (maximum) errors at 1.0° (40.0°) elevation angle were less than 50 m (~800 m). As a result of a possible changing positional error field, the focus in this study will be on large changes in tornado tilt so as to be more certain that the changes are not artifacts caused by unleveled radar sampling.
a. Goshen County tornado on 5 June 2009
The life cycle of the GC tornado extended from 2152 to 2231 UTC (hereafter all times are in UTC). During that time period, the MWR-05XP obtained ~3600 PPI scans of the TVS associated with the tornado at heights ranging from ~100 m up to as high as ~6 km AGL.4 There were three gaps in data collection: at 2157:13 for 2.5 min, at 2216:07 for 74 s, and at 2221:29 for 99 s. Herein, the first and second times are used as approximate cutoff times for the tornadogenesis and mature phases of the tornado, respectively.5 This dataset presents a unique opportunity to assess the steadiness of TVS intensity in time and height every 6–9 s for the entire life cycle of a long-lived tornado (Fig. 3).
During tornadogenesis, the TVS associated with the tornado was identified at progressively higher levels at later times, as discussed in French et al. (2013). When data collection resumed after the data gap (~2200 in Fig. 3), ΔVs typically were 50–60 m s−1. However, TVS intensity oscillated during the 16 min of uninterrupted data collection during the tornado mature phase. In addition, at ~2205 there was a brief period of time in which no TVS was observed at several levels above 2 km. As the tornado entered its dissipation phase (~2217 in Fig. 3), TVS intensity began to decrease first at heights of 1.5–2.5 km and then at lower levels. The TVS criteria were no longer met first at a height of 1.7 km at 2223:52 (indicated by an absence of TVS observations in Fig. 3) and then progressively upward and downward from that level in the next ~3 min. The TVS was followed the longest in the lowest 1 km and the last TVS observation occurred at 2230:08 near the time a condensation funnel no longer could be identified in video stills from photogrammetry data (not shown).
To explore changes in TVS intensity in more detail, a time series of ΔV using all 3569 TVS observations was constructed (Fig. 4a). There were several time periods, particularly between 2200 and 2216, in which there was significant spread in TVS intensity. When the data are separated out by observations that occurred above and below 2 km (Fig. 4b), it can be seen that the oscillations in ΔV (for the purposes of this study, oscillations had both a decrease and subsequent increase in ΔV of at least 10 m s−1 and a duration of at least 1 min) occurred mostly above 2 km during the tornado mature phase. Below 2 km, TVS ΔV was relatively steady at 55–65 m s−1, but above 2 km, there were five separate oscillations that occurred with an average period of ~3 min.
Time series of ΔV from individual elevation angles are shown for several levels below 1.5 km, above 2 km, and at the “transition” levels in between (Fig. 5). In the lowest 1.5 km (Fig. 5a), there were mostly small-amplitude changes in TVS ΔV of ~5 m s−1. However, there was a steady increase in ΔV at the lowest-observed level beginning at ~2209 consistent with observations from other mobile radars (cf. Fig. 6a in Kosiba et al. 2013). Above the lowest-observed levels (Fig. 5b), there was a transition to progressively larger-amplitude oscillations in ΔV as the heights of the observations increased. In addition, there was a positive time lag in some of the oscillations with increasing height (e.g., at ~2204 and ~2205). Above 2 km (Fig. 5c), ΔV underwent the short-time-scale, large-amplitude (≥10 m s−1) changes discussed previously. Again, in the first three oscillations, changes in ΔV at lower levels slightly led those at higher levels (e.g., at ~2201, ~2206, and ~2207).
The second intensity oscillation (2205–2207 in Figs. 5b,c) is of particular interest because of the aforementioned period in which no TVS was identified. The progression of MWR-05XP radial velocity at three levels during the time period of the second oscillation is shown (Fig. 6). At a height of 200 m (Fig. 6a), the TVS was identifiable continuously and TVS ΔV was consistently 55–65 m s−1. However, on a scale larger than the TVS, there was a decrease in radial velocity over the same time period (e.g., inbounds decreased from ~35 to ~25 m s−1). In contrast, in data from 2.5 km (Fig. 6b), it can be seen that the TVS, initially with ΔV of ~60 m s−1, weakened quickly and considerably. By 2205:30, there was no longer a TVS identifiable at this level. At 2205:37, a TVS again can be identified ~750 m southeast of the location of the previous TVS. This TVS strengthened in the next 20 s and was identifiable until dissipation at that level at 2223:52 (Fig. 3). The same progression, in which the TVS dissipated and another formed southeast of the first one, also occurred in data from higher up in the storm, at 4 km (Fig. 6c). However, dissipation occurred about 40 s after that at 2.5 km (i.e., the “secondary genesis cycle” progressed upward with time). It is likely the two TVSs were separate features because (i) of the short amount of time and relatively large distance between successive TVS identifications (e.g., ~15 s and ~1 km, respectively, at a height of 4 km) and (ii) at three levels, both TVSs were identified in separate locations at the same time (black circles in Fig. 3).
During the time period of the TVS dissipation and secondary genesis above 2 km, the GC tornado underwent a scale contraction in the lowest 200 m and developed a funnel cloud (cf. Fig. 3 in Wakimoto et al. 2011). MWR-05XP observations of decreasing radial velocities at the tornado cyclone scale (~2 km; Fig. 6a) also are consistent with a low-level scale contraction occurring during this time period, though TVS ΔV did not increase in data from 200 m in height (e.g., Fig. 5a). In addition, before and as the TVS aloft was dissipating, it was tilted increasingly downshear (toward the northeast) with height (Fig. 7), particularly above 2 km. However, when the secondary genesis occurred, the resulting TVS was oriented almost vertically upright (e.g., at 2206:39 in Fig. 7).
The four other oscillations in TVS intensity (not shown) did not undergo the cycle of dissipation and secondary genesis as in the second oscillation. The first and fourth ΔV oscillations had changes in TVS intensity of 20–35 m s−1. Also, the first oscillation occurred during a tornado-scale contraction associated with a brief funnel cloud that soon dissipated (cf. Fig. 3 in Wakimoto et al. 2011). The third and fifth oscillations occurred over short time periods of ~2 min and were the smallest in amplitude (intensity changes of 10–15 m s−1). Following the fifth oscillation, ΔV decreased again, but data collection temporarily stopped at 2216:07, so it is not known if there was a subsequent increase in intensity consistent with a sixth oscillation.
During the tornado mature phase, there was a tendency for ΔV to decrease with height (Fig. 8a). However, there is only a weak relationship between ΔV and height above 2 km. The weak correlations are at least partially influenced by the oscillating TVS intensities, as the highly variable ΔVs above 2 km likely mask any strong signal of a vertical ΔV gradient that might exist. To assess whether the intensity oscillations proceeded upward or downward with time, or were stationary, Pearson product moment and Spearman rank correlation coefficients (e.g., Wilks 2006) were calculated between (i) ΔV at a height of 2 km and ΔV at 3 km at various lag times and (ii) 90-s changes in ΔV at 2 and 3 km height using the same 3-km lag times (Fig. 8b). The largest correlation coefficients (~0.55) occurred for the latter comparison at lag times of 0–32.5 s; these values indicate a tendency for 90-s changes in ΔV at 3 km to occur a short time after similar changes are observed at 2 km. However, while the upward progression of, for example, the secondary genesis cycle (Figs. 6b,c) and the ΔV minima during the third oscillation (Fig. 5c), is also apparent in raw data, there were other times (e.g., local minima during the fifth oscillation; Fig. 5c) when this was not the case.
An overview of TVS tilt is shown as a time series of TVS inclination angle, defined here as the angle from the vertical made between the TVS location at 1.0° elevation angle and that at a particular height level,6 at several different heights during the entire life cycle of the tornado (Fig. 9a). TVS tilt varied little in height from 2 to 4 km, so the focus here is on changes in TVS tilt in time. TVS tilt was relatively large (inclination angles of 25°–50°) immediately after tornadogenesis and decreased right before the first MWR-05XP data gap. Tilt again decreased a large amount (from 30° to 10°) during the TVS dissipation and secondary genesis aloft (~2205), as described previously. Once the TVS reformed vertically upright, TVS tilt was steady for ~5 min before there was a large increase of 15°–20° over a ~90-s period at ~2212. Several minutes later, at the time of the first observed TVS dissipation (vertical line), tilt had increased to ~40°, and continued to increase as the TVS dissipated at multiple levels. Also during the tornado dissipation phase, the TVS began to move in vertically disparate horizontal directions (Fig. 9b), consistent with an increase in inclination angle. Above (below) 1.5 km, the TVS moved in a direction similar to that (to the right) of estimated storm motion.7
Dissipation of the GC TVS occurred first at 1.7 km and then at progressively higher (up to 3.5 km) and lower (down to 750 m) levels before dissipation occurred last in the lowest 500 m (Fig. 3). To determine if the vertical progression of TVS dissipation was sensitive to the exact criteria used to define a TVS, the time–height plots were reanalyzed (Fig. 10). Shortening (Fig. 10a) and lengthening (Fig. 10b) the time criterion for TVS dissipation from its baseline value of 60 s did not change the general vertical progression of TVS dissipation but did lead to earlier dissipation times aloft in the former case. Decreasing the TVS minimum ΔV criterion to 15 m s−1 (Fig. 10c) extended TVS observations to later times at most levels above 1 km, but TVS dissipation similarly occurred first at a height of 1.5 km. Using radial velocity observations separated by 500 m or less (Fig. 10d) rather than 2 km or less had almost no impact on the depiction of TVS dissipation. Overall, the vertical progression of TVS dissipation was relatively insensitive to the exact TVS criteria used in the calculations. The estimated height at which the TVS first dissipated (Figs. 3,and 10,) marks the approximate transition level between the levels of the disparate TVS translational motions (Fig. 9b ).
b. Kingfisher and Tribune tornadoes in May 2010
The Kingfisher and Tribune tornadoes were relatively weak tornadoes; each formed before MWR-05XP data collection began. However, in both cases, data collection continued through tornado decay, so the datasets are used here to investigate the vertical directionality of TVS dissipation. The Kingfisher dataset had two data gaps: at 2303:45 for 132 s and 2309:10 for 133 s. The estimated time of Kingfisher tornado dissipation in Storm Data was 2300, though a near-ground TVS was followed in MWR-05XP data to ~2309 (Fig. 11a). The Tribune tornado was very short-lived (~4 min) and remained almost stationary during data collection (Fig. 11b). Another weak, short-lived tornado preceded the Tribune tornado, but was not scanned by the MWR-05XP.
A time–height plot of TVS dissipation is shown for both tornadoes in Fig. 12. Similar to the GC case, the Kingfisher TVS dissipated first at 1.5 km and then at progressively higher levels (up to ~2.5 km) before it dissipated last at 100 m (Fig. 12a). The TVS could not be unambiguously identified in data from above 2.5 km because there was more than one TVS at these levels when data collection began (not shown). Similarly, the Tribune TVS dissipated last near the surface (Fig. 12b) and could be tracked definitively only in the lowest 2 km because there were multiple TVSs at higher levels. As with the GC TVS, both the Kingfisher and Tribune TVSs dissipated in essentially the same manner regardless of the exact TVS criteria (not shown). A plan view of TVS position for the Kingfisher TVS (Fig. 12c) is also very similar to that of the GC TVS. The TVS moved to the right (left) of approximated storm motion below (above) a height of 1.5 km.
c. El Reno tornado on 24 May 2011
The final tornado included in this study is the only violent tornado observed by the MWR-05XP. Two deployments on the long-lived tornado were undertaken. The first deployment began just before the estimated time of tornadogenesis and lasted ~8 min. The second deployment took place ~35 min later8 as the tornado moved toward the northeast away from the radar. During the second deployment, there were two data gaps: for 111 s at 2140:22 and for 112 s at 2149:43. There was another data gap at 2155:01; when data collection resumed, the TVS was beyond 30 km in range from the MWR-05XP. Coarse radar spatial resolution and poor data quality at these ranges precluded further investigation, so the focus here is on TVS intensity and position.
The time–height evolution of TVS intensity (ΔV) is shown for both deployments in Fig. 13. Above 2 km, TVS formation occurred at progressively higher levels at later times [see French et al. (2013) for details about TVS formation] and TVS ΔV was as high as 80 m s−1 by the end of the first deployment. Once data collection resumed 37 min later, TVS intensity was roughly the same, though no clearly defined TVS was identified above 5.5 km. TVS intensity increased greatly at all levels at ~2135 and generally remained very strong until the end of the deployment. Between 2143 and 2149, TVS ΔVs in the lowest 2 km were extremely high at 140–150 m s−1. After the second data gap, TVS intensity decreased in the lowest 1 km as the tornado may have entered its dissipation phase and/or radar spatial resolution became coarser.
In a ΔV time series from the second deployment (Fig. 14a), it can be seen that the elevation-angle-averaged ΔV increased by ~50 m s−1 from 2135 to 2139, became nearly constant, and then underwent a slow decrease until the end of data collection. The observations were separated out by height and averaged to assess TVS intensity differences at different heights (Fig. 14b). There were two time periods that exhibited large, rapid changes in intensity. First, ΔV increased by over 40 m s−1 at all height levels in ~2 min at ~2135. Second, below 4 km, ΔV increased by 25–40 m s−1 over a 3-min period at ~2143 followed by a similarly steep decline over the next 4 min. Also, the increase at these levels was followed by two brief oscillations in ΔV. In general, during the tornado mature phase, ΔV decreased with height in the lowest 6 km (Fig. 14c).
Radial velocity data from near the surface (200 m) during the time period of the first intensity increase are shown in Fig. 15a. The ΔV increase occurred as the El Reno TVS merged with a second strong TVS that was likely associated with a second tornado.9 During MWR-05XP data collection, the secondary TVS moved from a location southeast to northeast to north of the main TVS before the merger occurred. The secondary TVS was identifiable in data from all levels at which the El Reno TVS was observed (e.g., Fig. 15b). Once the two TVSs merged, both the size and intensity of the resulting TVS (still referred to here as the “El Reno TVS”) increased.
From 2143 to 2149, TVS ΔV increased by as much as 40 m s−1 in the lowest 4 km and underwent apparent rapid oscillations in the lowest 2 km (Fig. 14b). For example, ΔV decreased from ~150 to ~125 m s−1 at ~2145 (Fig. 16). It is plausible that these oscillations are an example of radial velocities being significantly affected by the position of the radar beam centerline relative to the tornado center for the underresolved vortex (Wood and Brown 1997). For example, at 2144:36, the strongest inbound and outbound radial velocities were in adjacent gates (Fig. 16a). About 30 s later, ΔV had decreased by ~25 m s−1 and the strongest radial velocities were now one gate apart (Fig. 16b). Another ~20 s later, ΔV had increased by ~40 m s−1 and the strongest radial velocities again were in adjacent gates (Fig. 16c). Radial velocities over 90 m s−1 and some as high as 105 m s−1 were observed (e.g., Fig. 16c), values consistent with those obtained by RaXPol earlier in the tornado’s life cycle (Pazmany et al. 2013). Therefore, changes in observed TVS intensity of 25–50 m s−1 from a changing beam position relative to the tornado is reasonable considering the strong winds in the tornado and the relatively coarse spatial radar resolution.
Estimated TVS tilt was calculated at several levels during both deployments (Fig. 17). Upon TVS formation, tilt was large, often at 30°–40° inclination angle at the three levels sampled (Fig. 17a). When data collection resumed at ~2134, TVS inclination angle had decreased to 10°–20° (Fig. 17b). Subsequently, low-level tilt (from near the surface to a height of 2 km) increased to near 30° before the TVS became nearly vertically upright at all levels sampled by ~2146. As the tornado continued to move away from the MWR-05XP, low- and midlevel TVS tilt increased back to 30°–40°, the highest observed values since TVS formation in the first deployment. As in the GC case, the TVS tilt was consistently toward the northeast (downshear) with increasing height.
a. Tornado intensity
Unique observations of the time–height progression of TVS intensity were made in two long-lived tornadoes by the MWR-05XP. For both the GC and El Reno tornadoes, there were periods in which TVS intensity changed rapidly in time and height. In the latter case, the immediate increase in TVS size and ΔV during the vortex merger (Figs. 13 and 15a) is strong evidence that the TVS merger caused an increase in TVS intensity. To the authors’ knowledge, this is the first time a tornado merger has been documented in a mobile Doppler radar dataset. Based on the increasingly large number of tornadoes that have been sampled by mobile Doppler radars (~200), it is likely that a merger process that increases TVS intensity is a rare event. Nonetheless, it is one process by which TVS intensity can change rapidly.
In the former case, there is no clear answer as to what caused the height-dependent, oscillatory changes in TVS intensity. Despite the large amount of data obtained during VORTEX2, rapid-scan volumetric (i.e., up to 3–6 km) observations of the tornado were obtained only by the MWR-05XP. Efforts to relate ΔV oscillations in the GC tornado to storm-scale features observed by radars with higher spatial resolution than the MWR-05XP were unsuccessful (K. Kosiba 2012, personal communication) because such data were available only every 2 min (e.g., Kosiba et al. 2013), a time period longer than that necessary to resolve temporally any processes involved in the rapid intensity changes. Single-Doppler data from the MWR-05XP were investigated to identify features such as rear-flank gust fronts (RFGFs) and small-scale vorticity maxima that may have been associated with the changes in TVS intensity, but no obvious cause-and-effect relationships were found.10 Therefore, using the available information, we merely speculate as to possibilities.
One potential cause of the height-dependent intensity oscillations is a sudden change in vertical vorticity stretching above the LFC (~2 km; cf. Fig. 2 in Markowski et al. 2012a). In such a scenario, vertical vorticity stretching in the tornado would have to be increasingly affected by buoyancy relative to vertical pressure gradient forces as this level was reached. Then, a tornado ingesting more negatively (positively) buoyant air would realize a (an) decrease (increase) in vorticity stretching that weakened (strengthened) the tornado at all levels but substantially more at or near the LFC and above (e.g., Fig. 5c). However, there are no adequate data available to test such a hypothesis. Recent data assimilation studies have estimated the average LFC height in this case at 2.3 km (Marquis et al. 2012a), though the true LFC height likely varied considerably in time.
Other possible causes of the oscillations include multiple vortices, centrifugal waves, and symmetric oscillations. Unobserved multiple vortices could cause significant changes in TVS intensity; past observational studies have used similar tornado ΔV oscillations as those seen in the GC TVS as indirect evidence of multiple vortices (Alexander and Wurman 2005). However, the oscillations in this case occurred every 2–5 min, while multiple vortices, even in large tornadoes, have been observed to modulate tornado flow over a much shorter period of time (e.g., Wurman 2002). It is also possible that the causes of the oscillations were axially traveling centrifugal waves induced by perturbations in the flow downstream of the LFC (A. Shapiro 2012, personal communication). Estimates of vertical vorticity and wavelength associated with the oscillations were used to calculate expected phase speeds and periods of the waves (not shown) derived by Shapiro (2001). The calculated phase speeds (periods) of the waves were off by a (an) factor of ~2 (order of magnitude) compared to that derived from ΔV observations.11 A final idea is that the height-dependent changes in TVS strength were caused by wave-induced symmetric oscillations (Nolan 2012) of the flow. However, most symmetric oscillations diagnosed in idealized tornado simulations have been stationary or progressed downward with time (e.g., Nolan and Farrell 1999; Nolan 2012) whereas at least some of the GC TVS intensity changes progressed upward with time (e.g., Figs. 6b,c).
b. Tornado tilt
In both the GC and El Reno cases, TVS tilt was consistently in the downshear direction with height. Also, TVS tilt increased in the minutes prior to the decay of the GC tornado. For both cases, there was an inverse relationship between TVS inclination angle and intensity (Fig. 18). All of these observations are consistent with past studies that have documented tornado tilt summarized in section 1. However, of note is that in both cases, TVSs tilted a great amount at other times during their life cycles. Both TVSs had inclination angles over 30° upon TVS formation (Figs. 9a and 17a) and GC TVS tilt was highly variable during the tornado mature phase (Fig. 7), all occurring well before dissipation. In addition, Pearson product moment and Spearman rank correlation coefficients between 30-, 60-, and 90-s changes in TVS ΔV and changes in inclination angle over the same time periods varied only from −0.2 to 0.1 (not shown); for all but the 60- and 90-s correlations in the El Reno case, a null hypothesis that the two variables are uncorrelated could not be rejected at the 1% level. Though TVS tilt was negatively associated with TVS intensity for these two cases, it was too variable in time to be a reliable short-time-scale (i.e., on the order of minutes) predictor of future TVS intensity.
c. Tornado dissipation
In the GC and Kingfisher tornadoes, TVS dissipation occurred first at ~1.5 km, then at progressively higher levels, and last in the lowest 1 km. The dissipation of the TVS associated with the Tribune tornado also was observed to occur in a top-down manner in the lowest 1.5 km. In addition, in both the GC and Kingfisher cases, the TVS moved to the right (left) of storm motion below (above) the 1.5-km height level during dissipation and dissipated first at the interface between these two translational directions (e.g., Fig. 19).
In studying the demise of the GC tornado, Richardson et al. (2012) noted that, near the end of the tornado’s life cycle, there was an enhanced area of reflectivity that rotated around the tornado at ~2224 and an associated rain shaft that undercut the tornado at low levels by ~2226. The rain shaft likely was associated with enhanced RFD outflow, which advected the tornado away from the location of the midlevel mesocyclone causing it to dissipate. In data from the MWR-05XP in the lowest 1 km, there are areas of enhanced reflectivity likely associated with a rear-flank gust front that moved southeastward (Figs. 20a,b). At a height of 2 km, a similar reflectivity feature was located ~2 km west-southwest of the TVS (Fig. 20c) because of the northeastward TVS vertical tilt (e.g., Fig. 9a); at these higher levels, the TVS likely was not subjected to the southeastward advection. In the Kingfisher case, there were areas of radial convergence, likely associated with RFGFs, oriented from northeast to southwest in close proximity to the TVSs in the lowest 1 km (Figs. 20d,e); both the areas of radial convergence and the TVSs moved eastward and southeastward. Again, aloft (~2 km), these features were either not identified or were located much farther from the TVS because of the northeastward TVS tilt with height (Fig. 20f).
A conceptual model of the theorized dissipation process in these two cases is shown in Fig. 21. It is argued that both tornadoes became occluded in the lowest 1.5 km and moved in a direction similar to that of the mean flow they were embedded in (toward the southeast) behind RFGF boundaries. Above these levels, the tornadoes were not occluded and/or boundaries were either weaker, did not exist, or were too far away from the tornadoes to advect them southeastward. At the level just above where the tornadoes were no longer advected southeastward, the tornadoes weakened and then dissipated as vertical vorticity generation decreased in an area where the vortices became tilted vertically and stretched horizontally [cf. Fig. 1 in Bluestein et al. (1988) for an example of what this might look like visually] and turbulent mixing dissipation effects dominated. The dissipation process continued from this level in an upward direction as tornado inflow was cut off. Near the surface, the tornadoes continued moving toward the southeast as they became completely occluded and eventually dissipated. The cause and initial height of dissipation in this model is also consistent with the combined radar and visual observations of tornado dissipation made by Wakimoto and Martner (1992; see their Fig. 9i) discussed in section 1.
Last, past studies have documented the tendency for tornadoes involved in the cyclic tornadogenesis process to move rearward and to the left of storm motion at low levels during the dissipation process (e.g., Burgess et al. 1982; Dowell and Bluestein 2002a,b; Tanamachi et al. 2012). In the above two cases, the TVSs associated with the tornadoes did not move rearward, but toward the right of storm motion, similar to that simulated for mesocyclones in Adlerman and Droegemeier (2005) in their “nonoccluding cyclic mesocyclogenesis” cases. One potential difference in the mode of dissipation is the strength of RFGF outflow (Fig. 21, bottom). Cyclic tornadogenesis has been hypothesized to occur when there is a mismatch between tornado horizontal motion and that of the main storm updraft–downdraft (Dowell and Bluestein 2002b). One such scenario is that when relatively weak storm outflow allows tornadoes and/or mesocyclones to be advected rearward into the storm away from their sources of vertical vorticity generation (Dowell and Bluestein 2002b; French et al. 2008; Marquis et al. 2012b). In the two cases above, areas of enhanced reflectivity and stronger rear-flank outflow likely were involved in advecting the tornado toward the right of storm motion. It is plausible that there is an optimal amount of rear-flank outflow that contributes to tornado maintenance by keeping it at a location close to where vertical vorticity can be produced and/or existing vertical vorticity can be concentrated (e.g., Marquis et al. 2012b). RFGF outflow less (more) than the optimal amount disrupts tornado maintenance such that cyclic tornadogenesis (occluded tornado dissipation) occurs; it is likely that internal RFD momentum surges play a role in this process as well. It is hoped that future studies of tornado dissipation can be used to investigate the impact of rear-flank outflow strength on the mode of tornado decay.
We believe data from the MWR-05XP can be used to argue the following:
Tornado intensity can be unsteady in time and disparate in height, even for long-lived tornadoes during their mature phases. In some cases, the tornado may not even be temporally or vertically continuous aloft at these times. As a result, the evolution of the tornado above 2 km may not be a reliable indicator of tornado evolution in the lowest 1 km, even in a relative sense.
Separate tornadoes may, on rare occasions, merge, resulting in a tornado that is both larger and stronger than either of the original two tornadoes.
Tornadoes can tilt significantly in height at all times during the tornado life cycle. Large tornado inclination angle is not necessarily indicative of tornado weakening or impending dissipation.
In some cases, tornado dissipation may occur in a vertically “inside out” manner, first at ~1.5 km, then at progressively higher and lower levels, and last near the ground. In these cases, the level where dissipation occurs first is located above the level in which tornado motion is influenced heavily by strong RFGF outflow.
Tornado dissipation likely occurs when a tornado is displaced from a location of favorable vertical vorticity generation by a storm inflow–outflow imbalance, as proposed in previous studies (Dowell and Bluestein 2002b; French et al. 2008; Marquis et al. 2012b). The exact nature of the imbalance determines whether tornado cycling or occluded dissipation occurs.
The authors thank Philip Chilson, Richard Doviak, and Alan Shapiro, who reviewed early drafts of this work within the first author’s Ph.D. dissertation at the University of Oklahoma. Thanks also to Jeffrey Snyder, Alex Schenkman, Matthew Kumjian, Lou Wicker, Karen Kosiba, Yvette Richardson, David Nolan, and Curtis Alexander for useful discussions regarding the results of this research. The latter also provided significant computational assistance in the use of the DREADER software. We are grateful to Jana Houser, Paul Buczynski, Randy George, and the VORTEX2 crews, particularly Josh Wurman, David Dowell, and Erik Rasmussen for their assistance in data collection. Comments from Jim Marquis and two anonymous reviewers greatly enhanced this manuscript. This study was supported by NSF Grants ATM-0637148 and ATM-0934307.
Current affiliation: NOAA/National Severe Storms Laboratory, Norman, Oklahoma.
MWR-05XP truck levelers are strong enough only to level the truck frame and are used in deployments solely to stabilize the truck and prevent it from shifting during data collection. An inclinometer system designed to record pitch and roll angles failed early in the spring of 2009. Efforts are ongoing to update the MWR-05XP with a leveling system.
To avoid measuring the intensity of larger-scale phenomena (e.g., mesocyclones), in cases when the maximum ΔV was located at a diameter > 2 km, the largest GTG ΔV was used to estimate the intensity of the TVS. In most such cases, GTG ΔV differed from maximum ΔV by <5 m s−1.
The roll, pitch, and heading angles are defined as the rotation of the truck around the positive x, y, and z axes, respectively. A truck with a positive pitch (roll) angle will have its back (right) side lower than its front (left) side. The heading angle is measured in the x–y plane in a clockwise direction from the positive y axis to the antenna beam.
Hereafter, all heights given are approximate values AGL neglecting differences between the elevation of the MWR-05XP and that of the TVSs under examination (estimated as <150 m for all five deployments used in this study). Changes in TVS elevation within a deployment likewise were small.
It is likely that the position of the TVS at 1.0° elevation angle (height of 100–300 m) was similar to that of the tornado at the surface. Also, the observation closest to the level stated was used; if no observation was within 250 m of that level, the inclination angle was not calculated at that time. Because of occasionally large vertical gradients in tilt, interpolation was not used to fill in data gaps.
In the GC and Kingfisher cases, storm motion was estimated by tracking WSR-88D data of the forward-flank reflectivity maximum over a ~20-min time period centered on the first dissipation time.
The long period of time between mobile radar deployments on a single tornado is unusual. However, there is considerable evidence that the deployments spanned one continuous tornado, including the damage survey used for the Storm Data entry (Table 1) and data of the tornado obtained by RaXPol during the time period between the first and second deployments of the MWR-05XP (Pazmany et al. 2013).
The secondary TVS is also easily identifiable beginning at ~2117 in data from the KTLX WSR-88D (not shown) and is consistent with several visual reports of a second tornado near the El Reno tornado (J. Snyder 2013, personal communication).
It is likely that using data from a radar with relatively poor sensitivity and coarse spatial resolution prevented such features from being identified and studied. For example, often it was difficult to identify internal RFD momentum surges even though they were observed in other data (e.g., Kosiba et al. 2013). These problems emphasize the need (i) for radar data that has both 10–20-s volumetric update times and relatively fine spatial resolution, such as RaXPol and (ii) to attempt synthesis of rapid-scan, dual-Doppler analyses when such data are available.