1. Introduction
Tornadic thunderstorm outbreaks are frequently attended by interactions between storms, such as cell mergers. A few studies (e.g., Lee et al. 2006b), along with anecdotal evidence, suggest that cell mergers may affect the occurrence and timing of subsequent tornadogenesis. On the other hand, there are also documented instances in which tornado production appears to slow or cease following a merger (e.g., Lindsey and Bunkers 2005), or in which a merger is associated with the disruption of an ongoing tornado (e.g., Wurman et al. 2007).
A relatively small number of formal studies have examined the associations between storm mergers and tornado formation, maintenance, and dissipation (e.g., Lindsey and Bunkers 2005; Lee et al. 2006a,b; Wurman et al. 2007; Hastings et al. 2012). It is increasingly evident from the results of tornado research field projects (e.g., VORTEX and VORTEX2) that the processes governing tornadogenesis potentially occur on time scales of a minute or less (e.g., Dowell and Bluestein 2002b; Wurman et al. 2007; Wakimoto et al. 2011; French et al. 2013). This short time scale is the primary reason why the dynamics of storm mergers and their relationship to tornadoes have not been thoroughly addressed.
Storm mergers have been recognized since at least the 1950s (e.g., Byers and Braham 1949; Browning and Fujita 1965). Within the context of weather radar observations, the consensus appears to be that the union of two previously closed, 35- or 40-dBZ reflectivity contours constitutes a storm merger (Westcott 1984; Bluestein and Parker 1993). In a dynamic sense, a storm merger occurs when two distinct thunderstorm updrafts merge into a single updraft. In supercells, however, reflectivity is a poor proxy for upward vertical motion (Hastings et al. 2010) because the primary updraft is associated with a local minimum in reflectivity (the bounded weak echo region; Lemon et al. 1978). In addition, Rogers and Weiss (2008), studying both tornadic and nontornadic rear-flank storm mergers, found no significant increase in reflectivity in the primary storm up to 15 min after a merger. In this study, we define the storm merger as the process beginning with the union of two reflectivity objects and ending with the union of two updraft objects (with the objects identified by a computer algorithm described below).
Our definition of a storm merger depends on knowledge of the updraft structure. Vertical velocity is not regularly observed by weather radars and usually must be inferred through analysis. Westcott and Kennedy (1989) described two storm merger types based on triple-Doppler analyses (Armijo 1969) from nontornadic storms in Illinois. The first storm merger type was characterized by the development of a “bridging” updraft between two merging updrafts. It was hypothesized that colliding outflow boundaries and resulting convergent uplift generated the bridging updrafts (Simpson 1980; Tao and Simpson 1984, 1989; Wurman et al. 2007). The second merger type was characterized by differential storm motion, that is, one storm overtaking another (Bluestein and Parker 1993). Most of the observational merger studies above were for nontornadic cases.
Because of the rapid changes that can occur within interacting storms, coarse temporal resolution observations impart significant challenges for the analysis of storm mergers and their relationship to tornadoes. In their seminal study of the 16 April 1996 Illinois severe weather outbreak, Lee et al. (2006a,b) examined WSR-88D data collected in 26 individual storm mergers and interactions, 14 of which were associated with tornadoes, and found that 54% of tornadoes occurred within ±15 min of a storm merger. Of this subset, 55% occurred within ±5 min of a storm merger, underscoring the requirement for rapid observations of storm mergers to establish a linkage (if any) to tornadogenesis. Lee et al. (2006b) hypothesized that the midlevel updraft of one storm can enhance preexisting low-level vorticity in a second storm via stretching (Fig. 1a). In a similar vein, Lindsey and Bunkers (2005) speculated that a left-moving Oklahoma supercell disrupted tornado production in a nearby, right-moving supercell for about 20 min after the two storms merged. Rogers and Weiss (2008), studying mergers associated with 10 western Texas tornado days, found that, on average, mesocyclone intensity (as gauged by 0.5° gate-to-gate shear) increased slightly up to 15 min after a merger with an ancillary storm, even if the merger was not associated with a tornado. They posited that rear-flank mergers in which the ancillary storm’s cold pool impinged on the upshear end of the supercell’s low-level updraft (near the inside edge of the hook) were more likely to result in tornadoes than those in which the ancillary storm collided with the center or downshear end of the updraft. They hypothesized that low-level convergence served to amplify low-level vorticity at the upshear end of the updraft, which is already a favored location for tornado production in supercells (Lemon and Doswell 1979). Rogers (2012) sought out storm mergers occurring within ±15 min of significant [greater than category 2 on the enhanced Fujita scale (EF2+)] tornado reports over a 5-yr period, finding that about one-third of these involving discrete storms featured apparent reflectivity bridging above colliding outflow boundaries (Simpson 1980), while the remainder did not. In all of those studies, the relative spatial (1-km range gate spacing) and temporal (~5 min) coarseness of the WSR-88D observations prevented detailed dynamical analysis of the storm mergers. While not specifically focused on storm mergers, Bluestein (2009) highlighted the situational complexity prior to the development of the 4 May 2007 Greensburg, Kansas, tornadic supercell, documenting multiple splitting and merging antecedent storms in a “storm genealogy” based on WSR-88D observations.
Two hypotheses for how a storm approaching a supercell’s inflow sector may change low-level vorticity in the supercell, based on suggestions offered by (a) Lee et al. (2006a) and (b) Wurman et al. (2007). Reflectivity values are approximations used to orient the reader to the storm structure.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
A few studies of storm mergers and tornadoes have incorporated high spatial and temporal resolution mobile Doppler radar data. However, such datasets are rare and, often, of limited areal or temporal coverage owing to challenging deployment conditions, beam blockage, and other factors. Dowell and Bluestein (2002a), analyzing data from an aircraftborne radar, noted (but did not focus on) the interaction of the 8 June 1995 McLean, Texas, tornadic storm with a short-lived reflectivity core with no active updraft. Wurman et al. (2007) analyzed high temporal resolution (~18 s) Doppler On Wheels (DOW) mobile Doppler radar data collected in a tornadic supercell that underwent at least two mergers with other nontornadic storms. They provided circumstantial evidence that the mergers disrupted the tornadoes. The resulting dual-Doppler analysis of single-elevation DOW data was two-dimensional, precluding the retrieval of vertical velocity information that could have indicated the strengths of the updrafts, as well as the timing of updraft unions (if any). Wurman et al. (2007) hypothesized that the mergers served first to enhance existing low-level vorticity through stretching and/or low-level convergence (Finley et al. 2002), but that the storm merger then suppressed the tornadoes as cooler, less buoyant outflow from the merging storms was ingested (Fig. 1b). Hastings et al. (2010), using volumetric DOW data, investigated a case of a storm merger and apparent tornadogenesis failure (only a brief funnel cloud was observed) during VORTEX2 (Wurman et al. 2010).
Cell mergers have also been studied using three-dimensional numerical simulations. Klemp et al. (1980) successfully simulated the interaction (not merger) between the 20 May 1977 Del City, Oklahoma, storm and a nearby, nontornadic hailstorm. Kogan and Shapiro (1996) hypothesized that the apparent “attraction” between symmetric updrafts, initiated in an otherwise horizontally homogeneous environment, resulted from mutual advection of each updraft by the other’s low-level radial inflow. Bluestein and Weisman (2000), simulating storms initiated along a line, found storm merger types roughly analogous to those found by Westcott and Kennedy (1989). They concluded that, in the presence of supercell shear, supercells were more likely when fewer storm interactions occurred. Finley et al. (2001, 2002) documented the upscale growth, resulting from storm mergers, of a weakly tornadic supercell into a high-precipitation supercell and, subsequently, a bow echo, using high-resolution numerical simulations to relate surface convergence to tornadogenesis. Jewett et al. (2002) found that when a line connecting two initialized updrafts was roughly perpendicular to the deep-layer environmental wind shear, storm mergers resulted in stronger rotation than when the updrafts had other initial orientations. In a related experiment, Hastings and Richardson (2010) attempted to artificially instigate storm mergers by “targeting” pairs of simulated storms for collision. They found that mergers between more-mature storms resulted in stronger vertical velocity and vertical vorticity maxima than mergers between younger storms, in spite of the possible mitigating effects of expanding cold pools. In an expansion of that work, R. Hastings (2014, personal communication) sorted simulated mergers between mature and nascent supercells into five categories. In none of the studies above was data assimilation performed.
Operational meteorologists treat strong low-level (≤1 km AGL) rotation (e.g., vorticity of ~0.01 s−1) in supercells as a prognostic indicator of tornadogenesis (e.g., Markowski and Richardson 2009) because it indicates the presence of ample preexisting vertical vorticity (Mitchell et al. 1998; Stumpf et al. 1998). The detection of a low-level mesocyclone (LLM) by a Doppler radar can constitute sufficient grounds to issue a tornado warning (Burgess et al. 1993; Brotzge and Donner 2013; Heinselman et al. 2015). Where Doppler radar data may not be available at low levels owing to range or beam blockage, the detection of a midlevel (3–7 km AGL) mesocyclone (MLM) may suffice (Trapp et al. 2005; Brotzge et al. 2011). About 25% (15%) of detected LLMs (MLMs) are associated with tornadoes (Trapp et al. 2005).
In the present study, we investigate in detail the dynamics of the merger of a smaller storm into the inflow sector of a mature tornadic supercell. To do this, we assimilate high temporal resolution radar observations of the 24 May 2011 El Reno, Oklahoma, tornadic supercell (hereafter, the El Reno storm) into a numerical cloud model. The cell merger was associated with the demise of one tornado and the genesis, 4 min later, of a second, more powerful and long-lived tornado. Analyses produced via data assimilation contain a full three-dimensional wind field, including vertical velocity w. The purposes of this paper are 1) to use the radar-derived three-dimensional cloud-scale analyses to construct a coherent sequence of events that qualitatively matches what was observed by radar and spotters and to provide insight into the merger processes and impacts and 2) to determine the relative sources and sinks of vorticity in the El Reno storm’s LLM and MLM before, during, and after the merger.
This study is novel in that volumetric radar observations with 1-min update times are assimilated to produce minute-by-minute analyses. Additionally, the mesocyclone, updraft, and other features of dynamical interest are identified objectively. Similarly, frequent surface thermodynamic observations are used for qualitative verification of the near-surface analyses (Dowell et al. 2004). While the phased array radar (PAR) observations of the El Reno storm are only available at altitudes around 700 m AGL and above, these are sufficiently close to the ground to assess the effect(s), if any, of the merger on the LLM and MLM.
2. Event description
Early on 24 May 2011, a negatively tilted upper-level trough advanced eastward over the Rocky Mountains (Fig. 2). Ahead of the trough, southerly flow over the southern Great Plains increased surface dewpoints in central Oklahoma to 18°–21°C. A nearly saturated layer extended up to 850 hPa. By 0000 UTC 25 May, the trough axis extended from southeastern Montana to the eastern Texas Panhandle. Midlevel lapse rates over most of Oklahoma were nearly dry adiabatic, and convective available potential energy (CAPE) values were estimated at 2500–4000 J kg−1. These, among other factors, prompted the National Weather Service (NWS) Storm Prediction Center (SPC) to assess a “high risk” for severe weather along a corridor extending from north-central Texas to south-central Kansas, encompassing much of Oklahoma (http://www.spc.noaa.gov/products/outlook/archive/2011/day1otlk_20110524_1300.html).
SPC’s 500-hPa analysis at 1200 UTC 24 May 2011. Geopotential height (dam; solid contours, contour interval is 6 dam). Temperature (°C; dashed contours, contour interval is 2°C). Wind barbs are coded as follows: half-barbs, 2.5 m s−1; full barbs, 5 m s−1; and pennants, 25 m s−1.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
In the late morning, a faster short-wave impulse detached from the trough. In addition to mixing at the dry–moist interface, this impulse helped to move the Texas Panhandle dryline rapidly eastward into the moist surface plume over the western half of Oklahoma, increasing low-level convergence along its length. Meanwhile, strong surface heating led to erosion of the capping inversion. A roughly north–south-oriented line of tightly spaced severe convective storms initiated just east of the dryline around 1900 UTC. Those storms north of Interstate 40 (hereafter, I-40) quickly became supercells and began producing tornadoes; those to the south of I-40 tended to remain smaller, weaker, and produced tornadoes later in the outbreak. Numerous storm mergers and splits were noted.
The NWS provides an extensive chronology of the storm events on this day (National Weather Service 2012); we utilize their nomenclature for individual storms and tornadoes in the following discussion. The El Reno storm formed about 150 km west-southwest of Oklahoma City (their storm B), tracked toward the northeast, and produced at least four tornadoes. The first tornado was a category 3 on the enhanced Fujita scale (EF3) that passed close to the town of Lookeba, Oklahoma (2031–2046 UTC; tornado B1). The second tornado was a category 5 on the enhanced Fujita scale (EF5) that crossed I-40 and impacted El Reno, Piedmont, and Guthrie, Oklahoma (2050–2235 UTC; tornado B2) along its 105-km path. We focus on the “handoff” between tornadoes B1 and B2, which occurred just as a weaker, nontornadic cell merged with the El Reno storm (Fig. 3) from the south. The merging storm, which initiated at 2010 UTC about 60 km directly south of the El Reno storm, merged into the El Reno storm’s right flank around 2045 UTC. The El Reno storm’s notch area filled with the smaller storm’s echo, indicating that rain was falling in the El Reno storm’s updraft. A circular area of relatively high reflectivity (“debris ball”; see Pazmany and Bluestein 2011) in the El Reno storm’s hook (Figs. 3c,d) marked the location of tornado B2.
Merger between a tornadic supercell and a nontornadic storm west of El Reno as observed by the NWRT PAR at (a) 2049, (b) 2055, (c) 2102, and (d) 2115 UTC 24 May 2011. An EF3 tornado ended at 2046 UTC, just prior to the time in (a), while an EF5 tornado began at 2050 UTC, between the times in (a) and (b). The debris ball of the developing EF5 tornado is denoted by a dashed white circle in (c) and (d). Reflectivity is indicated by shading (dBZ); the radar elevation angle is 0.5°.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
3. Data and methodology
a. The PAR at the National Weather Radar Testbed
Since the PAR at the National Weather Radar Testbed (NWRT) has been well documented by Zrnić et al. (2007), Heinselman et al. (2008), and Heinselman and Torres (2011), only a brief overview of the instrument and its relevant characteristics are provided. The PAR at the NWRT is an S-band (9.4-cm wavelength) weather research radar situated in Norman, Oklahoma. Its antenna panel comprises 4352 transmitting and receiving elements that can be independently phased to form electronically one or more radar beams with a peak transmitted power of 750 kW. The PAR scans a sector up to 90° wide. The platform can be mechanically steered in azimuth to keep a phenomenon of interest within the sector. Because it scans electronically over a 90° sector, rapid (~1 min) volume updates can be achieved. The received beamwidth varies as a function of the angle between the beam and the plane of the array, from 1.6° at boresight to 2.3° at ±45° off boresight. The range resolution of the system is 240 m.
Because severe storms were expected to develop in western Oklahoma on the afternoon of 24 May 2011, the PAR collected continuous observations in a westward-pointing sector from 1800 UTC 24 May to 0030 UTC 25 May. The volume coverage pattern spanned elevation angles from 0.5° to 52.9° (with a higher concentration of sweeps collected at low angles), extended out to a maximum range of 285 km, and had an update time of around 60 s. In this study, only those sweeps at and below 10° were assimilated, as they spanned most of the depth of the troposphere at the range of the El Reno storm (~90 km). The lowest PAR tilt (0.5°) intersected the El Reno storm around 700 m AGL. Therefore, we focus on the LLM and MLM evolution patterns at and above 1 km AGL.
The PAR Doppler velocity Vr observations were dealiased manually; some ground clutter and other nonmeteorological artifacts were also manually removed (K. Manross 2013, personal communication). Prior to assimilation, the PAR reflectivity Z and Vr observations were objectively analyzed onto a 4-km grid using a Cressman (1959) objective analysis scheme with a 2.8-km radius of influence. The horizontal thinning of the data decorrelated the observation errors, reduced the sample variance, and helped maintain ensemble spread by reducing the total number of observations. The analyzed radar data remained on the coplane surface of the radar sweep so that no vertical interpolation or averaging was introduced.
b. Radar data assimilation with NCOMMAS
To generate analyses of the atmospheric state over western Oklahoma on 24 May 2011, we used a local ensemble transform Kalman filter (LETKF; Ott et al. 2004; Hunt et al. 2007) technique to assimilate the PAR reflectivity and Doppler velocity observations. To generate the very short (1 min) forecasts needed to provide a background (prior) atmospheric state for the data assimilation system, we used the NSSL Collaborative Model for Multiscale Atmospheric Simulation (NCOMMAS; Coniglio et al. 2006). In many regards, our experiment setup is similar to that used by Tanamachi et al. (2013), with details modified for the 24 May 2011 case. A principal methodological difference was the use of LETKF instead of the ensemble square root filter (EnSRF). Thompson et al. (2014) demonstrate that LETKF and EnSRF give comparable results for convective-scale radar data assimilation. The experiment’s parameters are shown in Table 1.
Partial list of NCOMMAS parameters.
The domain (Fig. 4a) was specified so that the El Reno storm’s hook would be near the center at the time of the merger and handoff (Fig. 4b). The horizontal grid spacing (1 km) is insufficient to resolve tornadoes, but is sufficient to resolve mesocyclones (2–6-km diameter). The sounding used to initialize the ensemble was taken from the grid point closest to Binger, Oklahoma, from the 40-km Rapid Update Cycle (RUC) 2000 UTC model run at 2100 UTC (e.g., a 1-h forecast; Fig. 4c). This sounding, which featured surface-based CAPE in excess of 4500 J kg−1, a 0–6-km wind shear magnitude of 27 m s−1 (Fig. 4d), and a 0–6-km bulk Richardson number less than 12, was felt to be representative of the inflow profile of the El Reno storm. To initialize the ensemble of horizontally homogeneous storm environments with some uncertainty, a uniform distribution of random perturbations with amplitudes of 2.5 m s−1 was added to each ensemble member’s zonal u and meridional υ base state winds. To initiate convection, 10 ellipsoidal “perturbation blobs” (within which potential temperature θ, u, υ, w, and water vapor mixing ratio qυ were perturbed by small amounts) were placed in the south-central portion of the domain in a narrow rectangular region where active convection was already occurring.
(a) Map of western and central Oklahoma (outlined in black) showing the NCOMMAS domain for this study (blue box), the NWRT PAR coverage area (sector outlined in green), tornado tracks (purple outlines, courtesy of the NWS WFO in Norman), and Oklahoma Mesonet stations (gray triangles). The track of tornado B1 is near Binger, while the track of tornado B2 passes by El Reno. Other tornadoes whose tracks are depicted are not the focus of this study. (b) Ensemble mean reflectivity (dBZ; filled color contours) in the domain shown in (a) at 2045 UTC and 0.375 km AGL. County boundaries in (a) and (b) are drawn with thin gray lines. (c) Skew T–logp diagram (where T indicates temperature and p indicates pressure) of the sounding used to generate the ensemble of initial model states. Only every fifth wind barb is plotted. (d) Hodograph for the wind profile shown in (c).
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
After the ensemble was initialized at 1930 UTC, each member was allowed to integrate freely for 10 min. Then, Twin Lakes, Oklahoma, WSR-88D (KTLX) Z and Vr observations were assimilated every 5 min from 1940 to 1955 UTC (i.e., approximately four KTLX volumes), and model integration stopped at 2000 UTC. The early KTLX data assimilation populated the ensemble with heterogeneous, well-spread model states to serve as initial conditions (Stensrud and Gao 2010) for the PAR data assimilation starting at 2000 UTC.
Starting at 2000 UTC, both PAR Z and Vr observations were assimilated synchronously every 1 min. We mitigated detrimental effects of frequent Z assimilation (Dowell et al. 2011) by assuming a relatively large observation error for Z (σZ = 10 dBZ), thereby moderating adjustments to hydrometeor fields by Z assimilation. Ensemble spread was maintained through the application of additive noise every 5 min, while adaptive inflation (Miyoshi 2011) was applied as part of each data assimilation cycle.
c. Object identification and tracking
With NCOMMAS analyses produced every minute, it was necessary to automate objective identification of features of interest (storms, mesocyclones, and updrafts) within the domain that were used to constrain analysis products. In addition, because of the size and intensity differences among the objects being identified (e.g., mature supercells versus developing storms), we wanted to use an automated technique that could satisfactorily identify objects in a manner similar to a trained human observer. We applied the enhanced watershed algorithm of Lakshmanan et al. (2009) to individual horizontal slices of the model analysis volume, identifying Z objects (“storms”), positive w objects (“updrafts”), and vertical vorticity (ζ = dυ/dx − du/dy) objects (“vortices”).
In this algorithm, a two-dimensional variable field (e.g., Z or ζ) is treated as though it were the inversion of a topographic surface. Objects (termed basins) are identified by “flooding” the image starting from extrema and closing off basins once they reach a prescribed areal size (saliency). Additional points with gradients pointing back toward the current basin’s extremum are also reserved as “foothills” of the basin, and are not assigned to any other basin. If a basin does not attain the prescribed saliency, all its points revert back to “unassigned” status, and no foothills are generated. The algorithm proceeds until all the points in the image are assigned, or no more extrema exist from which to begin the flooding process. The reader is referred to Lakshmanan et al. (2009) and Lakshmanan (2012) for additional details about the algorithm. Some experimentation with algorithm parameters (such as the saliency; Table 2) was necessary because the objects to be identified differed markedly in size (e.g., storms were much larger than updrafts).
Object identification algorithm parameters used on the ensemble mean fields.
The extended watershed algorithm produced a set of objects at each analysis time, but no information about how those objects were related in time (tracks). We used the tracking algorithm of Lakshmanan and Smith (2010) (hereafter, NEW) to track identified storms, updrafts, and vortices from 2030 to 2130 UTC. Some of the update-dependent parameters outlined by Lakshmanan and Smith (2010) were modified. For example, the threshold association distance d between projected and actual centroids was reduced from 5 to 3 km based on the 1-min update time and the estimated storm motion [a 0–6-km pressure-weighted mean wind calculated from the initial sounding: (ustorm, υstorm) = (11.75 m s−1, 19.75 m s−1)].
In this tracking algorithm, the estimated storm motion is used to project an object forward to the next analysis time (i.e., from time n − 1 to time n). The projected object’s geometric centroid is tested for proximity to other current centroids. If there is a one-to-one correspondence, then the two objects are associated, and a track is drawn between their centroids. For remaining, unassociated, or ambiguously associated objects, a cost function based on areal coverage and peak intensity [given in Lakshmanan and Smith (2010)] is used to establish associations with other unassociated objects. If an object at time n − 1 is associated with two objects at time n, an object split is identified, with the object with the larger cost identified as a new object. If two objects at time n − 1 are associated with a single object at time n, an object merger is identified, with the combined object taking on the identifier of the object that had the longest track prior to the merger. If no association with previous objects can be made, a new object is identified. In theory, the temporal search could be expanded to include additional times (e.g., time n + 1 or n + 2), but the algorithm exhibited acceptable results and split–merger identifications for our case with only a 1-min forward search. As a last step, we executed a 1-min backward search to identify missed object track termini.
4. Qualitative verification of forecasts
We produced NCOMMAS forecasts and analyses at 1-min intervals from 2000 to 2130 UTC. This 90-min data assimilation period encompasses the end of tornado B1 (2031–2046 UTC) and the start of tornado B2 (2050–2235 UTC). The resulting forecast fields contain two or three powerful supercells in the northern and central portions of the domain (Fig. 4b), including the El Reno storm. As in the actual event, the largest supercells progressed toward the northeast, while smaller, weaker storms moved toward the north-northeast, some colliding with the right flanks of the supercells.
Although the locations and numbers of smaller cells were different in each ensemble member (not shown), the ensemble (Figs. 5g–l and 5s–x) accurately depicts the overall storm location and movement as seen by the PAR (Figs. 5a–f and 5m–r). In particular, the simulated cell merger into the right flank of the El Reno storm (Figs. 5i–l and 5s–x) is well represented in terms of time and location when compared with that in the PAR Z observations (Figs. 3a,b; 5c–f; and 5m–r). For the remainder of this study, we focus primarily on the El Reno storm (as objectively identified) from 2030 to 2120 UTC (e.g., Fig. 5).
(a)–(f),(m)–(r) Observed NWRT PAR reflectivity (dBZ; objectively analyzed onto a 1-km grid) and (g)–(l),(s)–(x) ensemble mean forecast reflectivity (dBZ) during the El Reno storm at 2.0 km AGL plotted at 2-min intervals from 2038 to 2100 UTC. The cell merger under investigation occurs northeast of Binger around 2050 UTC. A subdomain, 100 km on a side, of the domain depicted in Fig. 4 is shown.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
It is accepted that there will be some differences in the analyzed fields and observations resulting from model physics error. However, because of the spatial and temporal density of the PAR dataset for this case, we were able to assimilate a full volume of radar data every minute, strongly constraining the ensemble analysis. Furthermore, we found good correspondence (given the data spacing) between the locations of analyzed updrafts and enhanced differential reflectivity ZDR columns (Kumjian and Ryzhkov 2008; Romine et al. 2008) observed by a dual-polarized WSR-88D that is almost collocated with the PAR in Norman, Oklahoma (KOUN; J. Snyder 2014, personal communication). Thus, in this particular case, we have high confidence that the structures and locations of storm features are reasonably represented.
a. Comparison with rotation tracks
We generated a vorticity swath (Dawson et al. 2012) at 1 km AGL to compare with independent indicators of rotation. The swath can be interpreted as an indicator of the probability of a strong LLM. We have broken the swath up into five segments (labeled V1–V5 in Fig. 6a) for ease of discussion. Overall, the locations of the swath and the surface damage track (Fig. 6a) corresponded well but did exhibit some localized differences. In particular, the swath indicates strong low-level rotation was likely southwest of the start of tornado B1’s damage track (swath V1), where no surface damage was found, as well as directly above the gap between the tracks of tornadoes B1 and B2 (swath V3). Additionally, the vorticity swaths corresponding to tornado B2 (swaths V3–V5; see later discussion) are displaced a few kilometers to the north of the surface damage track. We are not particularly troubled by these displacements because tornadoes have often been observed to tilt with height and become displaced from their parent LLMs, usually as a result of interaction with surface gust fronts (e.g., Wakimoto and Atkins 1996). The resultant vortex stretching can actually result in brief intensification of the tornado’s surface winds even as it shrinks in diameter (Golden and Purcell 1978).
(a) Probabilistic vorticity swath generated from the 48-member ensemble at 1 km AGL. Red shading denotes the probability that the vorticity at that grid point exceeded 0.02 s−1 at some time between 2030 and 2130 UTC. The swath is broken up into segments, labeled V1–V5. Tornado tracks and county outlines are plotted as in Fig. 4. The plot shows a subdomain, 100 km on a side, of the domain displayed in Fig. 4. (b) Low-level rotation track product generated from KTLX observations of the El Reno storm, consisting of accumulated max azimuthal shear observed in the 0–3-km AGL layer (Miller et al. 2013). This product can be obtained online (https://www.nssl.noaa.gov/about/history/2011/). Swath segments from (a) are overlaid on (b) for comparison. Tornado tracks are outlined in black; counties are outlined in cyan. Distances (km) are relative to the southwestern corner of the model domain (Fig. 4b).
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
As an independent proxy for the LLM track, we used a radar-derived, low-level (0–3 km) rotation track (Miller et al. 2013) generated primarily from KTLX and Vance Air Force Base, Oklahoma, WSR-88D (KVNX) velocity observations (their Figs. 16e,f and 17; which we have enlarged in Fig. 6b to show detail). Because the rotation track is a time-integrated azimuthal shear maximum over the 0–3-km layer, we cannot know at exactly what altitude or time the maximum shear occurred. However, the observations used in the calculation of the azimuthal shear in the El Reno storm during the first two tornadoes were collected with beam center heights between 800 m and 3 km, so the product was restricted to the layer containing the LLM and not the tornado itself.
Overall, the low-level rotation track (Fig. 6b) also exhibited good correspondence with, and appears to support a number of features of, the vorticity swath (Fig. 6a). First, the rotation track, like the vorticity swath, was displaced slightly north of the surface damage track of tornado B2, indicating that tornado B2 did indeed tilt northward with height. Second, the rotation track indicates the presence of a strong LLM southwest of tornado B1’s surface damage track, prior to tornadogenesis (2031 UTC). We therefore consider the simulation’s portrayal of this pretornadic circulation (swath V1; Fig. 6a) to be accurate. Third, the rotation track is more or less continuous over the gap between the surface damage tracks of tornadoes B1 and B2, although the model’s portrayal of a highly probable LLM intensification over the gap (swath V3) is only weakly supported.
b. Comparison with Oklahoma Mesonet observations
Twenty-eight Oklahoma Mesonet (Brock et al. 1995) stations were operating in the analysis domain on 24 May 2011 (Figs. 4a,b), recording observations of relative humidity RH, aspirated air temperature, wind speed, wind direction, and atmospheric pressure every 1 min. Of particular interest are the data from the El Reno Mesonet station (Fig. 7), as tornado B2 passed very close to the instrumented tower and damaged part of the site (K. Ortega 2014, personal communication; http://ticker.mesonet.org/select.php?mo=05&da=27&yr=2011). As the tornado passed, the atmospheric pressure decreased 17 hPa at 2120 UTC (Fig. 7c), while the 1-min-average wind speed at 10 m AGL increased from 10 to 51 m s−1 (Fig. 7d). The wind direction changed from easterly to southerly until 2120 UTC as the inflow sector approached the station, before abruptly switching to northerly when the tornado passed (Fig. 7e). The station recorded a maximum wind gust of 67 m s−1 at 2121 UTC (not shown). If it could be shown that this gust was sustained for 3 s, this observation would correspond to EF3 tornadic winds (McDonald and Mehta 2006).
El Reno Mesonet measurements (thick black lines) of (a) air temperature (°C) at 9 m AGL, (b) RH (%) at 1.5 m AGL, (c) atmospheric pressure (hPa), and horizontal wind (d) speed (m s−1) and (e) direction (°) at 10 m AGL, overlaid on top of the ensemble of simulated near-surface variables (red lines) taken at the model grid point closest to El Reno Mesonet station at the lowest scalar level (125 m AGL). Because of the 360° wraparound, ensemble wind directions are plotted as points instead of lines. The thin black line in the middle of each bundle of red lines (or points) is the ensemble mean.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
We derived simulated near-surface observations for comparison by interpolating the model variables at 125 m AGL from the grid points closest to the El Reno Mesonet station (Fig. 7). We adjusted the simulated ensemble of pressure traces from the lowest model scalar level (125 m AGL) to the surface using the hydrostatic equation and the surface pressure in the initial sounding (947 hPa). This adjustment added about 13.5 hPa to the ensemble of pressure traces (Fig. 7c). The simulated El Reno storm cold pool is close to the observed temperature, but much drier (Figs. 7a,b). The near-surface temperature decreased from 26° to about 22° or 23°C as the mesocyclone passed (Fig. 7a). The ensemble mean near-surface RH corresponds well to the observations initially, with both tracking around 75% until 2105 UTC (Fig. 7b). During the storm passage, the Mesonet RH increased to 96%, while the ensemble mean near-surface RH decreased to less than 50%. It is believed that the low humidity in the modeled cold pools resulted from downward advection of dry midlevel air (Fig. 4c) (e.g., Dawson et al. 2010).
The passages of the El Reno tornado and mesocyclone, which appear as a sharp decrease (increase) in the observed pressure (wind speed) and an abrupt change in the wind direction around 2120 UTC, are represented in the simulations as more gradual changes in these near-surface quantities, with the extrema muted (Figs. 7c,d). The relatively coarse model grid spacing (1 km) smoothed the sharp pressure gradients responsible for the observed rapid changes. The tornado vortex passage occurs about 4 min early in our ensemble of simulations. However, this event occurs after more than 90 min of radar data assimilation and model error buildup, so we would not necessarily expect the simulation to phase perfectly with the observations.
Overall, we found reasonable agreement (given the model’s limitations) between the El Reno Mesonet observations and the simulated observations. Although the cold pool generated by the thunderstorm was too dry, it was close to the right temperature. Although the passage of the El Reno tornado-like vortex in the model occurred about 4 min early and the gradients were not as steep as in the actual tornado passage, the mesocyclone passage was clearly reflected in the wind and pressure fields.
5. Analysis of the merger
We applied the object identification and tracking algorithms to the ensemble mean forecast (Table 2). The El Reno storm was unambiguously identified as a single reflectivity object at 125 m AGL, and tracked from 2005 to 2130 UTC (Fig. 8a). Throughout the following discussion, “the El Reno storm” refers to this reflectivity object, which is used to constrain most of the analysis products. For our purposes, the “merger process” starts when the reflectivity objects first unite and ends when the updraft objects unite.
Objects identified in the ensemble mean fields using the extended watershed algorithm and tracked using the NEW algorithm, plotted at 5-min intervals: (a) the El Reno storm at 125 m AGL from 2005 to 2130 UTC, (b) the low-level (1 km AGL) mesocyclone from 2030 to 2130 UTC, and (c) the midlevel (5 km AGL) updraft from 2030 to 2130 UTC. The color corresponds to the time of identification. (See the vertical color bar.) County boundaries are drawn in thin gray lines. Tornado tracks are outlined in purple.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
The vortex object corresponding to the El Reno storm’s LLM was tracked as a single object from 2030 to 2117 UTC (Fig. 8b), with a brief discontinuity in the track at 2101 UTC (not shown). The MLM object track (not shown) also exhibited a discontinuity at 2106 UTC (not shown). In these instances, we manually combined separate tracks into a single track. The El Reno storm’s primary midlevel updraft (MLU) was somewhat difficult to track automatically because of its time-varying morphology (Fig. 8c), which resulted from its vertically pulsing character and mergers with nearby updrafts (to be discussed later). We evaluated several updraft tracks in close proximity to the El Reno storm’s hook and manually combined four tracks into the one pictured in Fig. 8c.
The simulated El Reno storm remained robust throughout the analysis period in the ensemble mean, with reflectivity in the core exceeding 60 dBZ (Figs. 5g–l and 5s–x) and mean maximum w reaching as high as 64 m s−1 (Fig. 9a). A cold pool beneath the rear flank of the El Reno storm (Fig. 10) formed a gust front that occasionally surged east of the hook and triggered smaller cells (e.g., Fig. 10b), some of which merged with the El Reno storm.
Time–height plots of ensemble mean max (a) vertical velocity (filled blue contours) and (b) vertical vorticity (filled red contours) in the El Reno storm (reflectivity object dilated by three grid points to ensure inclusion of the bounded weak-echo region). The times of tornadoes B1 and B2 are annotated along the x axes as solid black lines.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
Virtual potential temperature (K; filled color contours, the contour interval is 1 K), reflectivity (dBZ; gray contours at 35 and 55 dBZ), vertical vorticity (s−1; thick black contours, the contour interval is 0.01 s−1 starting from +0.01 s−1, with negative and zero contours suppressed for clarity), horizontal divergence [s−1; thin green solid (dashed) contours, the contour interval is +0.01 s−1 (−0.01 s−1)], and storm-relative horizontal wind vectors at 125 m AGL. Times shown are (a) 2027 UTC (when the near-surface vortex first intensified), (b) 2050 UTC (during the premerger stage), (c) 2100 UTC (about halfway through the merger process), and (d) 2110 UTC (after the merger). Tornado tracks are outlined in purple. County outlines are drawn in thin gray lines. Distances (km) are relative to the southwestern corner of the model domain (Fig. 4b).
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
a. The premerger stage (2030–2055 UTC)
Prior to the genesis of tornado B1, the MLM was displaced a few kilometers west of the LLM and near-surface vortex (NSV; Fig. 11a). By 2030 UTC, the vortex became vertically stacked (Fig. 11b), close to the NWS start time of tornado B1 (2031 UTC). The first NSV occurring in conjunction with the hook echo and exceeding 0.02 s−1 appeared at 2027 UTC, at the junction between the rear-flank downdraft (RFD) cold pool and the forward-flank downdraft (FFD) cold pool (Fig. 10a).
Ensemble mean surface reflectivity (dBZ; gray contours at 35 and 55 dBZ), vertical velocity at 625 m AGL (m s−1; filled color contours, the contour interval is 3 m s−1), and vertical vorticity (s−1; contours, the contour interval is 0.01 s−1 starting from +0.01 s−1) near the surface (blue; the NSV), at 1 km AGL (green; the LLM), at 3 km AGL (yellow; an intermediate level between the LLM and MLM), and at 5 km AGL (red; the MLM) plotted at (a) 2020, (b) 2030, (c) 2040, (d) 2045, (e) 2050, and (f) 2100 UTC. For clarity, only positive vorticity contours are plotted. Tornado tracks are outlined in purple. County outlines are drawn in thin gray lines. Distances (km) are relative to the southwestern corner of the model domain (Fig. 4b).
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
The NSV stayed close to the surveyed track for tornado B1 as it progressed northeastward (Figs. 11b–d). The vortex began to tilt northward with height at 2040 UTC, midway through its life cycle (Fig. 11c). Three minutes later, an RFD surge (a small remainder of which can be seen in Figs. 10b and 11d) wrapped entirely around the southern side of the LLM, and pushed the NSV and LLM toward the east with respect to the MLM. During this occlusion process, the NSV weakened (Fig. 9b), as the MLM began to stretch and elongate toward the northwest (Fig. 11d). The El Reno storm’s updraft weakened at all levels around 2046 UTC (Fig. 9a) as the ensemble mean NSV began to reintensify, producing swath V3 (Fig. 6a).
In the ensemble mean, only weak (~5 m s−1) midlevel updrafts were associated with the merging storm until 2047 UTC, when its northern outflow boundary collided with the rear-flank gust front of the El Reno storm. Because the merging storm was relatively immature, it did not possess an expansive cold pool (Fig. 10c). The near-surface storm-relative outflow from the merging storm was nearly stagnant but, nonetheless, served to enhance convergence along the rear-flank gust front (Fig. 10c). An updraft pulse of 24 m s−1 occurred on the northwestern side of the merging storm’s reflectivity core, centered at 5 km AGL (Fig. 12b). This updraft pulse was associated with generation of enhanced near-surface ζ (≥0.01 s−1) along the colliding boundaries owing to horizontal convergence [near (x, y) = (110 km, 100 km) in Figs. 10c and 11e], which is consistent with hypothesis 2 (Fig. 1b; Wurman et al. 2007), but this area of enhanced ζ did not merge into the NSV. The merging storm (which had previously been too small to meet the saliency criterion) was identifiable as a separate reflectivity object in the ensemble mean from 2048 to 2054 UTC. In radar observations and the simulations (Fig. 12a), the merging storm appeared to be a developing supercell, with weak midlevel rotation [near (x, y) = (100 km, 90 km) in Fig. 11e] and, later, low-level rotation [near (x, y) = (115 km, 105 km) in Fig. 11f] in its western appendage. Its MLU weakened as it approached the El Reno storm, but persisted through the merger.
Ensemble mean (a) reflectivity (dBZ; colored isosurfaces) and (b) vertical velocity (m s−1; blue isosurfaces, in increments of 5 m s−1, starting from 10 m s−1) at 2047 UTC, overlaid on surface reflectivity (dBZ; filled color contours on the lowest plane), and surface horizontal velocity vectors (black arrows on the lowest plane). The merging storm is at the lower rhs. The rendering is confined to a subdomain, 50 km on a side, of the domain shown in Fig. 4b. The El Reno storm’s updraft and the merging storm’s updraft are shown (labeled 1 and 2, respectively). (c),(d) As in (a),(b), but for values at 2054 UTC. Individual updraft elements are numbered (1–4) for later detailed discussion.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
The NSV attained ζ of 0.02 s−1 by 2047 UTC and reached a maximum ζ of 0.03 s−1 by 2050 UTC (Fig. 9b). This intensification is interpreted as the model representation of the genesis of tornado B2, although it occurs about 4 min early relative to the NWS start time (2050 UTC). Concurrently, the LLM also intensified, while the MLM weakened (Fig. 9b). The MLM began to split, with its strongest lobe displaced west of the LLM (Fig. 11e).
b. The merger stage (2055–2105 UTC)
Starting at 2055 UTC, the merging storm and the El Reno storm were considered a single reflectivity object by the enhanced watershed algorithm (Figs. 13a,b). By 2100 UTC, the MLM had split into two lobes. The western lobe decoupled from and moved west of the LLM and NSV, while the eastern lobe remained stacked vertically on top of them (Fig. 11f). Near the surface, the cold pool completely occluded the vortex, limiting its access to buoyant inflow (Fig. 10c).
Reflectivity objects at 125 m AGL for the El Reno storm (labeled 1) and the merging storm (labeled 2) (a) before the merger and (b) at the beginning of the merger phase. Updraft objects at 5 km AGL (c) at the beginning of the merger, (d) partway through the merger, and (e) after the merger. Objects not associated with the merger process are not plotted. Tornado tracks are outlined in purple. County outlines are drawn with thin gray lines. Distances (km) are relative to the southwestern corner of the model domain (Fig. 4b).
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
The merger process coincided with an overall weakening of the El Reno storm’s MLU during its interaction with several nearby updrafts. In the following discussion, we refer to four separate MLUs (numbered 1–4; Figs. 12c,d, Figs. 13c–e, and Fig. 14), which were identified by the enhanced watershed algorithm. The primary MLU of the El Reno storm (merging storm) at 2055 UTC is MLU1 (MLU2). Throughout the merger process and beyond, additional MLU pulses developed above the El Reno storm’s forward-flank gust-front uplift zone (Figs. 10b and 11e) and moved rearward with time toward the primary MLU. MLU3 was one such pulse; it developed about 8 km east of MLU1 at about 2053 UTC (Fig. 12d and a few minutes later in Fig. 13c), and moved westward along the forward-flank gust front toward MLU1 (Figs. 14a–d).
Ensemble mean reflectivity (dBZ; gray contours at 35 and 55 dBZ), vertical velocity (m s−1; filled color contours, the contour interval is 4 m s−1), and vertical vorticity (s−1; thick black contours, the contour interval is 0.01 s−1, with the zero contour suppressed for clarity and dashed contours representing negative values) at 5 km AGL plotted at (a) 2056, (b) 2057, (c) 2058, (d) 2059, (e) 2104, and (f) 2107 UTC. Updrafts (Figs. 12, 13c,e) have been numbered (1–4) for reference. Tornado tracks are outlined in purple. County outlines are drawn with thin gray lines. Distances (km) are relative to the southwestern corner of the model domain (Fig. 4b).
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
At 2056 UTC, a distinct, new updraft pulse (MLU4) developed between MLU1 and MLU2, which were separated by approximately 12 km (Fig. 12d). This “bridging” updraft formation between storm updrafts separated by more than 10 km is consistent with one category of simulated supercell–nonsupercell merger outcomes described in a forthcoming study by R. Hastings (2014, personal communication). A pair of shallow, counterrotating midlevel vortices straddled MLU4 (Fig. 14a), with the cyclonic (anticyclonic) vortex on the right (left) side of MLU4 with respect to storm motion (which was toward the northeast). We presume that this vortex pair originated from MLU4’s upward bending of a preexisting, crosswise, horizontal vortex line (Davies-Jones 1984). The two vortices then revolved cyclonically around MLU4, with MLU4 eventually merging into the eastern side of MLU3, and its cyclonic vortex eventually merging into the eastern lobe of the El Reno storm’s MLM at 2100 UTC (Figs. 14b–d and 11f). MLU3 merged into MLU1 at 2059 UTC (Fig. 14d). The overall result was that the El Reno storm’s MLM grew and elongated toward the northeast at 2100 UTC (Fig. 11f), having absorbed both the additional updraft area of MLU3 and the cyclonic vortex generated by MLU4. At 2102 UTC, the LLM and MLM began to strengthen again. The terms in the vorticity tendency equation (e.g., Dowell and Bluestein 2002b), evaluated inside the LLM (Fig. 15a), indicate that the dominant contributor to ζ at this level was tilting of horizontal vorticity. (While backward trajectories could have shown the source of ζ in the LLM more definitively, none were available for these simulations.) It appears that the merger may have been at least in part responsible for the northeast jog in the NSV track, as multiple MLMs and MLUs interacted, changing the orientation of the vortices.
Terms in the vorticity tendency equation are evaluated in the objectively identified (a) LLM and (b) MLM. The solenoidal generation term (not shown) is negligibly small at these altitudes. The approximate times at which vorticity swaths V1–V5 were produced (Fig. 6a) are annotated below.
Citation: Weather and Forecasting 30, 3; 10.1175/WAF-D-14-00164.1
Around 2105 UTC, MLU1 and MLU2 finally joined at their easternmost contact point, restoring the comma-shaped updraft envisioned in the supercell conceptual model by Lemon and Doswell (1979), and consolidating the four separate MLUs into a single, larger MLU (shown shortly thereafter in Fig. 14f). Overall, the updraft and vortex structure of the El Reno storm remained relatively disorganized at midlevels throughout the merger process. Once the MLU merger was complete, the NSV’s access to low-level buoyant air in the inflow sector was restored (Fig. 10d). Shortly thereafter, the entire mesocyclone consolidated, became vertically stacked, and intensified rapidly (Fig. 9b), with its motion becoming more easterly as it did so (Fig. 6a).
To summarize mesocyclone evolution during the premerger and merger stages, the entire deep convective vortex (i.e., the MLM, LLM, and NSV) initially moved in tandem with the storm, shedding a lobe of the MLM rearward with respect to storm motion between tornadoes B1 and B2 (Fig. 11). As this occurred, relatively cool, near-surface air completely surrounded the NSV. This evolution is suggestive of occluding, cyclic midlevel mesocyclogenesis followed by cyclic, low-level mesocyclogenesis. This cycling behavior does not appear to be associated in any appreciable way with the cell merger that occurred starting at 2055 UTC, as the MLM split and tornado handoff (2046 to 2050 UTC) preceded it by 5 min or more.
c. The postmerger phase (2105–2120 UTC)
Following the storm merger and consolidation of the mesocyclone around 2105 UTC, w and ζ in the El Reno storm increased (Figs. 9 and 15), as the El Reno storm’s NSV regained access to relatively uncontaminated, buoyant near-surface air in the inflow sector (Fig. 10d). Allowing for some variability owing to rear-flank internal surges (e.g., Lee et al. 2012), the cold pool of the merged storm (Figs. 10c,d) was similar in temperature and structure to the El Reno storm’s premerger cold pool, probably because of the immaturity of the merging storm. The vortex associated with tornado B2 became vertically stacked (Fig. 11f), and ζ in its NSV reached nearly 0.05 s−1 (Fig. 9a). Vortex stretching, a result of the intensifying updraft in the El Reno storm, was primarily responsible for the increase in low-level vorticity during this period (Fig. 15a).
6. Conclusions
High spatiotemporal resolution PAR observations of the 24 May 2011 El Reno, Oklahoma, tornadic storm were assimilated into a numerical cloud model using an ensemble data assimilation technique, generating three-dimensional kinematic and thermodynamic analyses of the storm every 1 min at 1-km horizontal resolution from 2000 to 2130 UTC. Derived products generated from the ensemble mean analyses were corroborated by independent observations (Figs. 6 and 7), lending confidence that the analyses are robust.
The rapid, 1-min update time of the PAR observations and concurrent data assimilation cycling (Fig. 5) enabled the analysis, association, and tracking of features that would not have been possible with conventional, ~5-min WSR-88D volume scans (e.g., Lee et al. 2006b) and similarly frequent analyses. In particular, we were able to verify qualitatively the model representation of the storm and vortex passage over a surface station with 1-min observations (Fig. 7), examine the minute-by-minute evolution of updrafts and vortices (Fig. 9), and diagnose the consolidation of multiple MLUs and splitting and merging midlevel vorticity maxima (Figs. 11 and 14). We were also able to automatically associate storms, updrafts, and vortices between analysis times using the enhanced watershed object identification and NEW tracking algorithms (Figs. 8 and 13). These results clearly demonstrate the value added to analyses of rapidly evolving convective storms through the collection and assimilation of rapid radar volume updates.
We conclude with high confidence that the storm merger (2055–2105 UTC) did not cause the handoff between tornadoes B1 and B2 (2046–2050 UTC), not least because the latter preceded the former by five or more minutes. The handoff was associated with a split and rearward motion of part of the MLM, in accordance with models of occluding midlevel mesocyclone cycling put forth by Burgess et al. (1982), Dowell and Bluestein (2002b), Adlerman and Droegemeier (2005), Beck et al. (2006), and French et al. (2008), among others.
The El Reno storm merger process did not conform clearly to either of the merger models posited by Lee et al. (2006b) or Wurman et al. (2007) (Fig. 1); additional mechanisms were also at work. In particular, the collisions between the inbound (with respect to the El Reno storm) outflow boundary from the merging storm and the El Reno storm’s gust fronts generated additional updraft pulses. The El Reno storm’s updraft weakened in the minutes leading up to the merger (Fig. 9a; Lindsey and Bunkers 2005), and at least three separate updrafts merged into it over the 10-min merger process. In addition, a bridging midlevel updraft appears to have tilted preexisting horizontal vorticity between the El Reno storm and the merging storm, generating a new, small MLM. This new MLM eventually merged with and augmented the MLM already present in the El Reno storm (Fig. 14).
We speculate that the El Reno storm merger event may not be representative of mergers between a tornadic and nontornadic storm, for two primary reasons. First, the details uncovered here are suggestive of a complex series of interactions between several dynamical features. Second, mergers between tornadic and nontornadic storms have been associated with a range of outcomes, including the demise of the tornadic storm. It seems likely that similar studies of additional storm merger events will need to be analyzed before a generalized conceptual model (or models) of such interactions can be synthesized.
Acknowledgments
This work was supported by a National Research Council Research associateship awarded to the first author. Funding was provided by NOAA/Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA11OAR4320072, U.S. Department of Commerce. Manual dealiasing and quality control of the NWRT PAR data were performed by Brandon Smith and Lamont Bain (students of Kevin Manross), and Dr. Michael French. We are particularly indebted to Dr. Valliappa Lakshmanan for his assistance in implementing the enhanced watershed and NEW tracking algorithms. Dr. Lakshmanan, Dr. Patrick Skinner, and two additional anonymous reviewers kindly suggested improvements to this manuscript. Oklahoma Mesonet data with 1-min time resolution were provided by Dr. Kevin Kloesel, Dr. Chris Fiebrich, and Cindy Luttrell. Dr. Daniel T. Dawson II contributed the code to plot the ensemble vortex swath. Dr. Ryan Hastings, Dave Priegnitz, David Andra, Don Burgess, Andrew MacKenzie, and Jaret Rogers provided helpful discussion.
REFERENCES
Adlerman, E. J., and Droegemeier K. K. , 2005: The dependence of numerically simulated cyclic mesocyclogenesis upon environmental vertical wind shear. Mon. Wea. Rev., 133, 3595–3623, doi:10.1175/MWR3039.1.
Armijo, L., 1969: A theory for the determination of wind and precipitation velocities with Doppler radars. J. Atmos. Sci., 26, 570–573, doi:10.1175/1520-0469(1969)026<0570:ATFTDO>2.0.CO;2.
Beck, J. R., Schroeder J. L. , and Wurman J. M. , 2006: High-resolution dual-Doppler analyses of the 29 May 2001 Kress, Texas, cyclic supercell. Mon. Wea. Rev., 134, 3125–3148, doi:10.1175/MWR3246.1.
Bluestein, H. B., 2009: The formation and early evolution of the Greensburg, Kansas, tornadic supercell on 4 May 2007. Wea. Forecasting, 24, 899–920, doi:10.1175/2009WAF2222206.1.
Bluestein, H. B., and Parker S. S. , 1993: Modes of isolated, severe convective storm formation along the dryline. Mon. Wea. Rev., 121, 1354–1372, doi:10.1175/1520-0493(1993)121<1354:MOISCS>2.0.CO;2.
Bluestein, H. B., and Weisman M. L. , 2000: The interaction of numerically simulated supercells initiated along lines. Mon. Wea. Rev., 128, 3128–3149, doi:10.1175/1520-0493(2000)128<3128:TIONSS>2.0.CO;2.
Brock, F. V., Crawford K. C. , Elliott R. L. , Cuperus G. W. , Stadler S. J. , Johnson H. L. , and Eilts M. D. , 1995: The Oklahoma Mesonet: A technical overview. J. Atmos. Oceanic Technol., 12, 5–19, doi:10.1175/1520-0426(1995)012<0005:TOMATO>2.0.CO;2.
Brotzge, J., and Donner W. , 2013: The tornado warning process: A review of current research, challenges, and opportunities. Bull. Amer. Meteor. Soc., 94, 1715–1733, doi:10.1175/BAMS-D-12-00147.1.
Brotzge, J., Erickson S. , and Brooks H. , 2011: A 5-yr climatology of tornado false alarms. Wea. Forecasting, 26, 534–544, doi:10.1175/WAF-D-10-05004.1.
Browning, K. A., and Fujita T. , 1965: A family outbreak of severe local storms - A comprehensive study of the storms in Oklahoma on 26 May 1963, Part I. AFCRL-65-695(I), Special Rep. 32, Office of Aerospace Research, U.S. Air Force, Bedford, MA, 356 pp.
Burgess, D. W., Wood V. T. , and Brown R. A. , 1982: Mesocyclone evolution statistics. Preprints, 12th Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 422–424.
Burgess, D. W., Donaldson R. J. , and Desrochers P. R. , 1993: Tornado detection and warning by radar. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., Vol. 79, Amer. Geophys. Union, 203–221.
Byers, H. R., and Braham R. R. , 1949: The Thunderstorm Project. U.S. Government Printing Office, 287 pp.
Coniglio, M. C., Stensrud D. J. , and Wicker L. J. , 2006: Effects of upper-level shear on the structure and maintenance of strong quasi-linear mesoscale convective systems. J. Atmos. Sci., 63, 1231–1252, doi:10.1175/JAS3681.1.
Cressman, G. P., 1959: An operational objective analysis system. Mon. Wea. Rev., 87, 367–374, doi:10.1175/1520-0493(1959)087<0367:AOOAS>2.0.CO;2.
Davies-Jones, R., 1984: Streamwise vorticity: The origin of updraft rotation in supercell storms. J. Atmos. Sci., 41, 2991–3006, doi:10.1175/1520-0469(1984)041<2991:SVTOOU>2.0.CO;2.
Dawson, D. T., II, Xue M. , Milbrandt J. A. , and Yau M. K. , 2010: Comparison of evaporation and cold pool development between single-moment and multimoment bulk microphysics schemes in idealized simulations of tornadic thunderstorms. Mon. Wea. Rev., 138, 1152–1171, doi:10.1175/2009MWR2956.1.
Dawson, D. T., II, Wicker L. J. , Mansell E. R. , and Tanamachi R. L. , 2012: Impact of the environmental low-level wind profile on ensemble forecasts of the 4 May 2007 Greensburg, Kansas, tornadic storm and associated mesocyclones. Mon. Wea. Rev., 140, 696–716, doi:10.1175/MWR-D-11-00008.1.
Dowell, D. C., and Bluestein H. B. , 2002a: The 8 June 1995 McLean, Texas, storm. Part I: Observations of cyclic tornadogenesis. Mon. Wea. Rev., 130, 2626–2648, doi:10.1175/1520-0493(2002)130<2626:TJMTSP>2.0.CO;2.
Dowell, D. C., and Bluestein H. B. , 2002b: The 8 June 1995 McLean, Texas, storm. Part II: Cyclic tornado formation, maintenance, and dissipation. Mon. Wea. Rev., 130, 2649–2670, doi:10.1175/1520-0493(2002)130<2649:TJMTSP>2.0.CO;2.
Dowell, D. C., Zhang F. , Wicker L. J. , Snyder C. , and Crook N. A. , 2004: Wind and temperature retrievals in the 17 May 1981 Arcadia, Oklahoma, supercell: Ensemble Kalman filter experiments. Mon. Wea. Rev., 132, 1982–2005, doi:10.1175/1520-0493(2004)132<1982:WATRIT>2.0.CO;2.
Dowell, D. C., Wicker L. J. , and Snyder C. , 2011: Ensemble Kalman filter assimilation of radar observations of the 8 May 2003 Oklahoma City supercell: Influences of reflectivity observations on storm-scale analyses. Mon. Wea. Rev., 139, 272–294, doi:10.1175/2010MWR3438.1.
Finley, C. A., Cotton W. R. , and Pielke R. A. , 2001: Numerical simulation of tornadogenesis in a high-precipitation supercell. Part I: Storm evolution and transition into a bow echo. J. Atmos. Sci., 58, 1597–1629, doi:10.1175/1520-0469(2001)058<1597:NSOTIA>2.0.CO;2.
Finley, C. A., Cotton W. R. , and Pielke R. A. , 2002: Tornadogenesis in a simulated HP supercell. Preprints, 21st Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., P10.1. [Available online at https://ams.confex.com/ams/pdfpapers/47216.pdf.]
French, M. M., Bluestein H. B. , Dowell D. C. , Wicker L. J. , Kramar M. R. , and Pazmany A. L. , 2008: High-resolution, mobile Doppler radar observations of cyclic mesocyclogenesis in a supercell. Mon. Wea. Rev., 136, 4997–5016, doi:10.1175/2008MWR2407.1.
French, M. M., Bluestein H. B. , PopStefanija I. , Baldi C. A. , and Bluth R. T. , 2013: Reexamining the vertical development of tornadic vortex signatures in supercells. Mon. Wea. Rev., 141, 4576–4601, doi:10.1175/MWR-D-12-00315.1.
Gaspari, G., and Cohn S. E. , 1999: Construction of correlation functions in two and three dimensions. Quart. J. Roy. Meteor. Soc., 125, 723–757, doi:10.1002/qj.49712555417.
Gilmore, M. S., Straka J. M. , and Rasmussen E. N. , 2004: Precipitation and evolution sensitivity in simulated deep convective storms: Comparisons between liquid-only and simple ice and liquid phase microphysics. Mon. Wea. Rev., 132, 1897–1916, doi:10.1175/1520-0493(2004)132<1897:PAESIS>2.0.CO;2.
Golden, J. H., and Purcell D. , 1978: Life cycle of the Union City, Oklahoma, tornado and comparison with waterspouts. Mon. Wea. Rev., 106, 3–11, doi:10.1175/1520-0493(1978)106<0003:LCOTUC>2.0.CO;2.
Hastings, R. M., and Richardson Y. P. , 2010: Storm mergers. Part 1: Preliminary numerical investigations of merger events. Preprints, 25th Conf. on Severe Local Storms, Denver, CO, Amer. Meteor. Soc., 7A.5. [Available online at https://ams.confex.com/ams/pdfpapers/176048.pdf.]
Hastings, R. M., Richardson Y. P. , Markowski P. M. , and Wurman J. , 2010: Storm mergers. Part 2: Observations of merger events from VORTEX2. Preprints, 25th Conf. on Severe Local Storms, Denver, CO, Amer. Meteor. Soc., P8.16. [Available online at https://ams.confex.com/ams/pdfpapers/176073.pdf.]
Hastings, R. M., Richardson Y. P. , and Markowski P. M. , 2012: Mergers in supercell environments. Part II: Tornadogenesis potential during merger as evaluated by changes in the near-surface low-level mesocyclone. Preprints, 26th Conf. on Severe Local Storms, Nashville, TN, Amer. Meteor. Soc., P143. [Available online at https://ams.confex.com/ams/26SLS/webprogram/Paper212524.html.]
Heinselman, P. L., and Torres S. M. , 2011: High-temporal-resolution capabilities of the National Weather Radar Testbed Phased-Array Radar. J. Climate Appl. Meteor., 50, 579–593, doi:10.1175/2010JAMC2588.1.
Heinselman, P. L., Priegnitz D. L. , Manross K. L. , Smith T. M. , and Adams R. W. , 2008: Rapid sampling of severe storms by the National Weather Radar Testbed phased array radar. Wea. Forecasting, 23, 808–824, doi:10.1175/2008WAF2007071.1.
Heinselman, P. L., LaDue D. , Kingfield D. M. , and Hoffman R. , 2015: Tornado warning decisions using phased array radar data. Wea. Forecasting, 30, 57–78, doi:10.1175/WAF-D-14-00042.1.
Hunt, B. R., Kostelich E. J. , and Szunyogh I. , 2007: Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter. Physica D, 230, 112–126, doi:10.1016/j.physd.2006.11.008.
Jewett, B. F., Lee B. D. , and Wilhelmson R. B. , 2002: Numerical simulation of cell interaction. Preprints, 21st Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., P6.6. [Available online at https://ams.confex.com/ams/pdfpapers/47219.pdf.]
Klemp, J. B., Ray P. S. , and Wilhelmson R. B. , 1980: Analysis of merging storms on 20 May 1977. Preprints, 19th Conf. on Radar Meteorology, Miami Beach, FL, Amer. Meteor. Soc., 317–324.
Kogan, Y. L., and Shapiro A. , 1996: The simulation of a convective cloud in a 3D model with explicit microphysics. Part II: Dynamical and microphysical aspects of cloud merger. J. Atmos. Sci., 53, 2525–2545, doi:10.1175/1520-0469(1996)053<2525:TSOACC>2.0.CO;2.
Kumjian, M. R., and Ryzhkov A. V. , 2008: Polarimetric signatures in supercell thunderstorms. J. Climate Appl. Meteor., 47, 1940–1961, doi:10.1175/2007JAMC1874.1.
Lakshmanan, V., 2012: Identifying objects. Automating the Analysis of Spatial Grids: A Practical Guide to Data Mining Geospatial Images for Human & Environmental Applications, V. Lakshmanan, Ed., Geotechnologies and the Environment Series, Vol. 6, Springer, 173–221.
Lakshmanan, V., and Smith T. , 2010: An objective method of evaluating and devising storm-tracking algorithms. Wea. Forecasting, 25, 701–709, doi:10.1175/2009WAF2222330.1.
Lakshmanan, V., Hondl K. , and Rabin R. , 2009: An efficient, general-purpose technique for identifying storm cells in geospatial images. J. Atmos. Oceanic Technol., 26, 523–537, doi:10.1175/2008JTECHA1153.1.
Lee, B. D., Jewett B. F. , and Wilhelmson R. B. , 2006a: The 19 April 1996 Illinois tornado outbreak. Part I: Cell evolution and supercell isolation. Wea. Forecasting, 21, 433–448, doi:10.1175/WAF944.1.
Lee, B. D., Jewett B. F. , and Wilhelmson R. B. , 2006b: The 19 April 1996 Illinois tornado outbreak. Part II: Cell mergers and associated tornado incidence. Wea. Forecasting, 21, 449–464, doi:10.1175/WAF943.1.
Lee, B. D., Finley C. A. , and Karstens C. D. , 2012: The Bowdle, South Dakota, cyclic tornadic supercell of 22 May 2010: Surface analysis of rear-flank downdraft evolution and multiple internal surges. Mon. Wea. Rev., 140, 3419–3441, doi:10.1175/MWR-D-11-00351.1.
Lemon, L. R., and Doswell C. A. III, 1979: Severe thunderstorm evolution and mesocyclone structure as related to tornadogenesis. Mon. Wea. Rev., 107, 1184–1197, doi:10.1175/1520-0493(1979)107<1184:STEAMS>2.0.CO;2.
Lemon, L. R., Burgess D. W. , and Brown R. A. , 1978: Tornadic storm airflow and morphology derived from single-Doppler radar measurements. Mon. Wea. Rev., 106, 48–61, doi:10.1175/1520-0493(1978)106<0048:TSAAMD>2.0.CO;2.
Lin, Y.-L., Farley R. D. , and Orville H. D. , 1983: Bulk parameterization of the snow field in a cloud model. J. Appl. Meteor., 22, 1065–1092, doi:10.1175/1520-0450(1983)022<1065:BPOTSF>2.0.CO;2.
Lindsey, D. T., and Bunkers M. J. , 2005: Observations of a severe, left-moving supercell on 4 May 2003. Wea. Forecasting, 20, 15–22, doi:10.1175/WAF-830.1.
Markowski, P. M., and Richardson Y. P. , 2009: Tornadogenesis: Our current understanding, forecasting considerations, and questions to guide future research. Atmos. Res., 93, 3–10, doi:10.1016/j.atmosres.2008.09.015.
McDonald, J. R., and Mehta K. C. , 2006: A recommendation for an enhanced Fujita scale (EF-scale). Texas Tech University, Lubbock, TX, 111 pp. [Available from http://www.depts.ttu.edu/nwi/Pubs/FScale/EFScale.pdf.]
Miller, M. L., Lakshmanan V. , and Smith T. M. , 2013: An automated method for depicting mesocyclone paths and intensities. Wea. Forecasting, 28, 570–585, doi:10.1175/WAF-D-12-00065.1.
Mitchell, E. D. W., Vasiloff S. V. , Stumpf G. J. , Witt A. , Eilts M. D. , Johnson J. T. , and Thomas K. W. , 1998: The National Severe Storms Laboratory tornado detection algorithm. Wea. Forecasting, 13, 352–366, doi:10.1175/1520-0434(1998)013<0352:TNSSLT>2.0.CO;2.
Miyoshi, T., 2011: The Gaussian approach to adaptive covariance inflation and its implementation with the local ensemble transform Kalman filter. Mon. Wea. Rev., 139, 1519–1535, doi:10.1175/2010MWR3570.1.
National Weather Service, 2012: Information for the tornado outbreak of May 24, 2011. [Available online at http://www.srh.noaa.gov/oun/?n=events-20110524.]
Ott, E., and Coauthors, 2004: A local ensemble Kalman filter for atmospheric data assimilation. Tellus, 56A, 415–428, doi:10.1111/j.1600-0870.2004.00076.x.
Pazmany, A., and Bluestein H. B. , 2011: A mobile, rapid–scanning, X-band, polarimetric (RaXPol) Doppler radar system. Preprints, 35th Conf. on Radar Meteorology, Pittsburgh, PA, Amer. Meteor. Soc., 16B.2. [Available online at https://ams.confex.com/ams/35Radar/webprogram/Paper191294.html.]
Rogers, J. W., 2012: Significant tornado events associated with cell mergers. Preprints, 26th Conf. on Severe Local Storms, Nashville, TN, Amer. Meteor. Soc., 9.4. [Available online at https://ams.confex.com/ams/26SLS/webprogram/Paper211575.html.]
Rogers, J. W., and Weiss C. C. , 2008: The association of cell mergers with tornado occurrence. Preprints, 24th Conf. on Severe Local Storms, Savannah, GA, Amer. Meteor. Soc., P3.23. [Available online at https://ams.confex.com/ams/pdfpapers/141784.pdf.]
Romine, G. S., Burgess D. W. , and Wilhelmson R. B. , 2008: A dual-polarization-radar-based assessment of the 8 May 2003 Oklahoma City area tornadic supercell. Mon. Wea. Rev., 136, 2849–2870, doi:10.1175/2008MWR2330.1.
Simpson, J., 1980: Downdrafts as linkages in dynamic cumulus seeding effects. J. Appl. Meteor., 19, 477–487, doi:10.1175/1520-0450(1980)019<0477:DALIDC>2.0.CO;2.
Stensrud, D. J., and Gao J. , 2010: Importance of horizontally inhomogeneous environmental initial conditions to ensemble storm-scale radar data assimilation and very short-range forecasts. Mon. Wea. Rev., 138, 1250–1272, doi:10.1175/2009MWR3027.1.
Stumpf, G. J., Witt A. , Mitchell E. D. , Spencer P. L. , Johnson J. T. , Eilts M. D. , Thomas K. W. , and Burgess D. W. , 1998: The National Severe Storms Laboratory mesocyclone detection algorithm for the WSR-88D. Wea. Forecasting, 13, 304–326, doi:10.1175/1520-0434(1998)013<0304:TNSSLM>2.0.CO;2.
Tanamachi, R. L., Wicker L. J. , Dowell D. C. , Bluestein H. B. , Dawson D. T. II, and Xue M. , 2013: EnKF assimilation of high-resolution, mobile Doppler radar data of the 4 May 2007 Greensburg, Kansas, supercell into a numerical cloud model. Mon. Wea. Rev., 141, 625–648, doi:10.1175/MWR-D-12-00099.1.
Tao, W.-K., and Simpson J. , 1984: Cloud interactions and merging: Numerical simulations. J. Atmos. Sci., 41, 2901–2917, doi:10.1175/1520-0469(1984)041<2901:CIAMNS>2.0.CO;2.
Tao, W.-K., and Simpson J. , 1989: A further study of cumulus interactions and mergers: Three-dimensional simulations with trajectory analyses. J. Atmos. Sci., 46, 2974–3004, doi:10.1175/1520-0469(1989)046<2974:AFSOCI>2.0.CO;2.
Thompson, T. E., Wicker L. J. , Wang X. , and Potvin C. K. , 2014: A comparison between the local ensemble transform Kalman filter and the ensemble square root filter for the assimilation of radar data in convective-scale models. Quart. J. Roy. Meteor. Soc., doi:10.1002/qj.2423, in press.
Trapp, R. J., Stumpf G. J. , and Manross K. L. , 2005: A reassessment of the percentage of tornadic mesocyclones. Wea. Forecasting, 20, 680–687, doi:10.1175/WAF864.1.
Wakimoto, R. M., and Atkins N. T. , 1996: Observations on the origins of rotation: The Newcastle tornado during VORTEX 94. Mon. Wea. Rev., 124, 384–407, doi:10.1175/1520-0493(1996)124<0384:OOTOOR>2.0.CO;2.
Wakimoto, R. M., Atkins N. T. , and Wurman J. , 2011: The LaGrange tornado during VORTEX2. Part I: Photogrammetric analysis of the tornado combined with single-Doppler radar data. Mon. Wea. Rev., 139, 2233–2258, doi:10.1175/2010MWR3568.1.
Westcott, N. E., 1984: A historical perspective on cloud mergers. Bull. Amer. Meteor. Soc., 65, 219–226, doi:10.1175/1520-0477(1984)065<0219:AHPOCM>2.0.CO;2.
Westcott, N. E., and Kennedy P. C. , 1989: Cell development and merger in an Illinois thunderstorm observed by Doppler radar. J. Atmos. Sci., 46, 117–131, doi:10.1175/1520-0469(1989)046<0117:CDAMIA>2.0.CO;2.
Wurman, J., Richardson Y. , Alexander C. , Weygandt S. , and Zhang P. F. , 2007: Dual-Doppler and single-Doppler analysis of a tornadic storm undergoing mergers and repeated tornadogenesis. Mon. Wea. Rev., 135, 736–758, doi:10.1175/MWR3276.1.
Wurman, J., Wicker L. J. , Richardson Y. P. , Rasmussen E. N. , Markowski P. M. , Dowell D. C. , Burgess D. B. , and Bluestein H. B. , 2010: An overview of the VORTEX2 field campaign. Preprints, 25th Conf. on Severe Local Storms, Denver, CO, Amer. Meteor. Soc., 5.1. [Available online at https://ams.confex.com/ams/pdfpapers/176068.pdf.]
Zrnić, D. S., and Coauthors, 2007: Agile-beam phased array radar for weather observations. Bull. Amer. Meteor. Soc., 88, 1753–1766, doi:10.1175/BAMS-88-11-1753.