a. Operational network

Data used in this study came in part from the operational, national network of rawinsonde sites and from the Weather Surveillance Radar-1988 Doppler (WSR-88D) network (Crum and Alberty 1993). The former data were accessible in real time from the University Corporation for Atmospheric Research/National Center for Atmospheric Research/Research Applications Program (UCAR/NCAR/RAP) website and some of each was obtained later from the NOAA Storm Prediction Center (SPC) archive and the National Climatic Data Center (NCDC). Surface data were obtained from the Oklahoma Mesonet (Brock et al. 1995;Fiebrich and Crawford 2001) archive and images were prepared using WeatherScope software. Doppler wind profiler data were obtained online from the NOAA archive of the National Profiler Network (Weber et al. 1990;http://www.profiler.noaa.gov/npn/index.jsp). Satellite imagery was obtained in real time from the College of DuPage “NeXt Generation Weather Lab.”

Data were also accessed online from NCDC for Rapid Refresh (RAP) model analyses (Pan et al. 2014;http://www.ncdc.noaa.gov/data-access/model-data/model-datasets/rapid-refresh-rap).

b. Special instrumentation

Mobile radar data from RaXPol were used (Pazmany et al. 2013). A summary of the characteristics of the radar and data collected are given inTable 1. Since the radar scans mechanically at speeds up to just over 180° s⁻¹, frequency hopping is employed to collect enough independent samples to reduce the noise in the estimate of Doppler velocity and other variables. The reader is referred toPazmany et al. (2013) for more technical details on the radar design. On 31 May 2013, the data collection strategy consisted of pulse pairs transmitted consecutively at radar frequencies separated by the pulse bandwidth (i.e., two pulses were transmitted at a given pulse repetition interval at frequency 1, followed by two more pulses at frequency 2, etc., typically using 11 frequencies).

Table 1.

Characteristics of RaXPol and scanning modes employed on 31 May 2013. Here,D_i refers to theith deployment, and h and v refer to the horizontal and vertical channels, respectively.

View Table

The polarimetric variables differential reflectivityZ_DR and copolar cross-correlation coefficientρ_hv are computed. These variables can be useful in identifying the type of scatterers in the radar volume; in particular, they can be used to distinguish between debris (typically,Z_DR < 1 dB andρ_hv < 0.90) and precipitation (typically,Z_DR > 1 dB andρ_hv > 0.90) in tornadoes (e.g.,Ryzhkov et al. 2005;Bluestein et al. 2007a;Snyder et al. 2010;Bodine et al. 2013;Snyder and Bluestein 2014;Wurman et al. 2014).

For this case, most of the time when the tornado was close to the radar, the radial sampling resolution was 30 m, although several minutes of data were collected using 15-m radial sampling resolution. For the latter, only eight pulses were used to calculate radar moments, which is considered marginal for obtaining low-variance estimates (e.g.,Doviak and Zrnic 1993;Bringi and Chandrasekar 2001). There is a trade-off between the number of quasi-independent samples obtained and the radial resolution as a consequence of the relationship between available system bandwidth, decorrelation time of the signal, and the dwell time for each radial.Snyder and Bluestein (2014) have shown how in tornadic debris clouds, where theρ_hv is often very low, data from the horizontal and vertical channels have only a low degree of correlation, so that the effective number of independent samples for pulse-pair covariance processing is increased when data from each channel are averaged before the radial velocity is estimated.

During the time that Doppler velocity data are described in subsequent sections, the Nyquist interval was ±30.8 m s⁻¹. Dealiasing the velocities was relatively easy early on, but became much more difficult when multiple vortices were apparent. Based on temporal and spatial continuity considerations, we believe the dealiased velocities are reliable.

Although the antenna half-power beamwidth is ~1°, some smearing in azimuth occurs as a result of the rapid rotation rate of the antenna. RaXPol is often configured with a dwell time that yields radials every 1°. For rotation of the antenna at 180° s⁻¹, the effective azimuthal resolution is ~1.4°–1.5° [see (7.34) inDoviak and Zrnic (1993) and seePazmany et al. (2013)].

Other special radar data came from KCRI, NOAA’s Radar Operation Center’s legacy radar, a test bed WSR-88D, located near University of Oklahoma Westheimer Airport in Norman (Saxion et al. 2011).

c. Data processing and display

Radar data were edited and displayed using Solo II software from the Earth Observing Laboratory (EOL) at NCAR or Warning Decision Support System–Integrated Information (WDSS-II) (Lakshmanan et al. 2007) software. RAP data were displayed using WDSS-II.

a. Mesoscale features

Prior to the formation of convective storms in Oklahoma on 31 May 2013, at midday, a weak front was situated over northwestern Oklahoma (Fig. 1a). This front was marked by a wind shift from south, south-southwest, or south-southeast ahead of the front to northeast behind the front, and temperatures were ~5°C cooler northwest of the wind-shift line marking the front. By 2100 UTC, the winds ahead of the front had backed to south-southeasterly (Fig. 1d) and then southeasterly (Figs. 1e,f). In the northern Texas Panhandle, the air was much cooler than to the south and the winds were relatively light (not shown). A dryline was located in far southwestern Oklahoma, where surface dewpoints were ~5°C drier than ahead of it. The boundary between the dryline and the front extended westward through the Texas Panhandle (not shown). The front weakened (as measured qualitatively by the horizontal temperature gradient seen by the change in colors) with time where it abutted against the warmer air mass to the southeast (Figs. 1b–e; the temperature difference across the front in north-central Oklahoma was ~4°–5°C at 1800 UTC and only ~1°–2°C or less at 2200 UTC), while it strengthened with time in western Oklahoma, where surface temperatures rose into the upper 30°C behind the dryline (the temperature difference across the front–dryline border was ~7°C or less at 1800 UTC and ~10°C at 2200 UTC). The warmest air at the surface lay in a tongue oriented in the northeast–southwest direction from west of the dryline in western Texas to east of the dryline in northwestern Texas and far southwestern Oklahoma (not shown). The frontolysis along the eastern portion of the front must have been due mainly to differential diabatic heating, as the air was heated by insolation more rapidly behind the front than ahead of it, because deformation alone should have increased the temperature gradient; the frontogenesis where the dryline abutted against the front must also have been due mainly to differential diabatic heating, where the air was heated more on the warm, dry side, than to the rear of the front; changes in surface temperature on either side of the front could not be accounted for by horizontal advection alone. The meridionally oriented dryline became sharper with time during the afternoon, as surface dewpoints dropped to below 10°C in the hot air, while surface dewpoints remained above 20°C in the moist air mass. The wind shifted to a more westerly direction in far western and southwestern Oklahoma, to the west of the dryline boundary and also just ahead of the front, consistent with the observed (not shown) and inferred formation of a meridionally oriented trough in low pressure where the air was being heated most rapidly. At 2100 UTC (Fig. 1d), prior to storm formation, from the wind field it is seen that qualitatively there was some cyclonic shear along the front, and weak convergence along the front ahead of the dryline.

View Full Size

Hourly analyses of surface data from the Oklahoma Mesonet: temperature (shaded) at 1.5 m AGL (°C; color scale at bottom), dewpoint at 1.5 m (°C; solid contours), and winds at 10 m (half barb = 2.5 and whole barb = 5 m s⁻¹). The thick solid line indicates location of a quasi-stationary front. Shown are images from (a) 1800, (b) 1900, (c) 2000, (d) 2100, (e) 2200, and (f) 2300 UTC 31 May 2013. In (d)–(f), analyses and data are during and after the onset of deep convection. Dots indicate the location of El Reno. (The analyses of temperature and dewpoint in the Texas Panhandle and northwestern Texas are not to be considered because the coverage of the data there is much sparser and data were not incorporated into the objective analyses.)

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

b. Storm formation and evolution

It is seen in satellite imagery that convective storms first began to form along the front in extreme southeastern Kansas and later in north-central Oklahoma, as evidenced by small anvils (Fig. 2a) at 2115 UTC. A line segment of convective clouds appeared along and south of the front along the surface thermal ridge (Figs. 2a and1d) and produced a first echo northwest of El Reno (Fig. 3b). This line segment extended southwestward along the axis of the warmest air behind the dryline as a zone of more sparse convective clouds. The most concentrated line segment, in western Oklahoma, consisted mostly of highly sheared cumulus congestus (Fig. 4). Another zone of convective clouds extended to the south-southwest behind the dryline in the deeply mixed air along the axis of warmest air (Fig. 1d). Cloud bands of stratus oriented north-northwest to south-southeast were seen south of the front in the moist air mass, indicative of very stable conditions. Bands of clear-air echoes oriented in the north-northwest to south-southeast direction were also noted on radar (Fig. 3).

View Full Size

GOES visible satellite imagery for Oklahoma and small portions of surrounding states, every ~10–15 min, depicting storm initiation, organization, and development into mature convective storms. Shown are results from (a) 2115, (b) 2132, (c) 2145, (d) 2155, (e) 2215, and (f) 2240 UTC 31 May 2013. Images were obtained in real time from the College of DuPage website. The approximate location of the front (solid white line) and dryline (dashed white line) are indicated in (a).

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Storm initiation, organization, and evolution (a)–(f) prior to the development of supercells (every ~15 min), and (g)–(i) as supercells evolved (every ~30 min), as depicted by images of radar reflectivity factor (dBZ; color scale is at the bottom), at a 0.5°-elevation angle, from KTLX. Shown are images from (a) 2114, (b) 2128, (c), 2144, (d) 2201, (e) 2215, (f) 2229, (g) 2301, (h) 2328 UTC 31 May, and (i) 0001 UTC 1 Jun 2013. Range rings are plotted every 50 km. Zoomed-in view of the northern and southern fine lines as depicted by radar images of (j) radar reflectivity factor (dBZ) and (k) Doppler velocity (m s⁻¹) at a 0.5°-elevation angle, from KCRI at 2232 UTC 31 May 2013. Color scales for reflectivity and Doppler velocity are shown.

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Photographs of the developing convective storms at (a) 2115 (from El Reno, view to the northwest), (b) 2128 (view to the west), and (c) 2204 UTC (panoramic photographs) 31 May 2013 from just north of El Reno, view from south to north (from left to right); RaXPol is visible on the far right. [Photographs courtesy of H. Bluestein.]

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

By ~2130 UTC, anvils began to emanate from the cumulus congestus, which had grown into narrow cumulonimbus clouds, in western Oklahoma (Figs. 2b and4b). These incipient storms, which were marked by several radar echoes oriented west-northwest to east-southeast or east–west (Fig. 3c), were located near the axis of maximum equivalent potential temperature at the surface (Fig. 5). By 2145 UTC, a broken line of convective storms had extended up to the north-northeast, along the front (Fig. 3c), while new, smaller cells had begun to form to the south, near the dryline (Figs. 1e and3c). The storms near the dryline were more widely scattered and not as concentrated along a line segment as the storms to the north (Fig. 2d). By ~2200 UTC, a line of convective towers appeared to the northeast, along the front, northeast of the previously existing convective storms (Figs. 2d and3d). At this time, the storms to the north took on the appearance of a broken line of convection, while the weaker storms to the south took on the appearance of a more widely scattered set of quasi-equally spaced, narrow radar echoes oriented in a northwest-to-southeast direction (Fig. 3e). About this time, RaXPol was deployed just north of El Reno, where the broken line of convection along the front was clearly visible (Fig. 4c, left of center to right;Fig. 2e), as were the more widely scattered storms along the dryline to the south (Fig. 4c, far left;Fig. 2e).

View Full Size

Analyses of equivalent potential temperature (K) from the Oklahoma Mesonet, just prior to and during storm initiation, at (a) 2100 and (b) 2130 UTC 31 May 2013, respectively. Analyses of (c) mixed-layer CAPE and (d) mixed-layer CIN at 2200 UTC 31 May 2013 from the RAP analyses. Black (white) dot marks the location of El Reno in (c) [(d)].

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

Between 2200 and 2300 UTC, the quasi-stationary front began to move southward as a cold front (Figs. 1e,f). To the northeast of the dryline, the front strengthened, apparently in response to evaporative cooling of precipitation from the convective storms in unsaturated air on the cool side of the front. Strong outflow winds of 20 m s⁻¹ were reported at one Mesonet station (Fig. 1f). However, the portion of the front near and to the north of the hot, dry air mass behind the dryline behaved differently. The wind shift propagated to the south, while the zone of strong horizontal temperature gradient that marked the location of the front remained stationary, behind the wind-shift line. Thus, the wind-shift line in western Oklahoma took on the appearance of a prefrontal wind-shift line, not a front (Hutchinson and Bluestein 1998;Schultz 2004). It might be that the apparent movement of the eastern portion of the front and wind-shift line to the west was caused by outflow produced by the convective storms. At 2229 UTC, two fine lines in the radar imagery appeared to the west and southwest of the broken line of convective storms along the front (Fig. 3f). The fine lines were tracked by reflectivity imagery from the KCRI radar (Figs. 3j,k), which at the time was providing low elevation scans every ~2–2.5 min, which is twice as often as usual, owing to a test of the Supplemental Adaptive Intra-Volume Low-Level Scan (SAILS) technique that was being conducted (http://www.roc.noaa.gov/wsr88d/NewRadarTechnology/NewTechDefault.aspx). The northern fine line, which was associated with approaching Doppler velocities, was moving to the south, while the southern fine line was quasi-stationary. The northern fine line was therefore probably associated with the cold front, while the southern fine line was probably associated with the dryline.

c. Synoptic- and mesoscale environment

The location where storms were first initiated to the west and northwest of El Reno (Figs. 2b,c;3c,d; and4) was marked by a local maximum in equivalent potential temperature (Figs. 5a,b). The sounding nearest in time and space from the storms was at Norman, to the southeast of the front, ~2–3 h after storms had initially formed (Fig. 6a). The moist boundary layer was ~100 hPa deep and was surmounted by a capping inversion, above which it was relatively dry (the relative humidity was ~30%–50%) up to the level of anvils from the storms. The convective temperature for a well-mixed boundary layer was ~38°C, which was not attainable at Norman [convective inhibition (CIN) was ~100 J kg⁻¹]. With both this convective temperature and the observed temperature, convective available potential energy (CAPE) was greater than 3000 J kg⁻¹, which is relatively high for tornadic-supercell environments [e.g.,Rasmussen and Blanchard (1998), their Fig. 13]. Farther to the west, along the dryline, convective temperature had been reached, as evidenced by the storms growing there in an environment of high temperatures (near 38°C) and high dewpoints; convective temperature may not have been reached to the north along the front, where lift may have been necessary to bring air parcels to the level of free convection (LFC).

View Full Size

Upper-air data in central Oklahoma, near the time that the storms were forming/in progress on 31 May 2013. (a) NWS sounding from the Norman WSR-88D (KOUN) at 0000 UTC 1 Jun 2013, released when the El Reno tornado was in progress to the northwest and displayed as a skew (temperature)–log (pressure) diagram. Temperature (°C) is given by the red line, dewpoint (°C) is given by the green line, and winds are plotted to the right (half barb ≈ 2.5, whole barb ≈ 5, and pennant ≈ 25 m s⁻¹). Hodograph is inserted at the upper left; wind speeds are labeled in knots (kt; 1 kt = 0.51 m s⁻¹). (b) Time (UTC; abscissa)–height (km; ordinate) diagram depicting hourly evolution of the vertical wind profile to the southwest of Norman at Washington, Oklahoma (from the “Purcell profiler”). (c) EVAD from RaXol data collected from volume scans averaged at the first deployment site, just north of El Reno, at five time periods, from 2220 to 2228 UTC; rings shown every 10 m s⁻¹.

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

The RAP model (Pan et al. 2014;http://www.ncdc.noaa.gov/data-access/model-data/model-datasets/rapid-refresh-rap) analysis (not shown), however, indicated the 100-hPa mean-layer/mixed-layer convective temperature was only ~28°–30°C northwest of Oklahoma City along the front and these temperatures were reached by midafternoon. The highest mixed-layer CAPE at the time of storm initiation (~2200 UTC) was in excess of 5000 J kg⁻¹ in a localized area just to the south-southwest of El Reno (Fig. 5c) while the mixed-layer convective inhibition was less than 100 J kg⁻¹ and near zero in a swath extending to the south-southwest of El Reno.

The vertical shear in the lowest 6 km was ~25–30 m s⁻¹, which is well in the range supportive of supercells (e.g.,Markowski et al. 2003;Thompson et al. 2003,2007;Mead and Thompson 2011). The sharply curved hodograph at low levels and greater than 400 m² s⁻² 0–3-km storm-relative helicity is supportive of tornadic supercells (e.g.,Markowski et al. 2003;Thompson et al. 2003,2007;Mead and Thompson 2011). Based on data from the nearby profiler southwest of Norman, it is seen that winds backed to southeasterly at ~2100 UTC and picked up in speed thereafter, while the winds at 6 km increased in speed to 25 m s⁻¹ by 0000 UTC 1 June 2014 (Fig. 6b). The vertical wind shear profile and thermodynamic environment were therefore supportive of supercells and tornadoes.

A vertical shear profile even closer in time and space than that derived from the Norman sounding at 0000 UTC was obtained just north of El Reno by RaXPol during the first deployment (Table 1,Fig. 7) while the storms were organized in a broken line segment along the front (Fig. 6c). This hodograph was obtained as an extended velocity azimuth display (EVAD;Matejka and Srivastava 1991) east of the convective storms (Figs. 3e–f) averaged from five volume scans at 2220, 2222, 2224, 2226, and 2228 UTC. Most of the EVAD was taken in clear air at low levels or stratiform precipitation aloft. The vertical shear in the lowest 6 km was just over 20 m s⁻¹, which is supportive of supercells, but was not extreme. The EVAD from RaXPol is quite similar to the hodograph from the Norman sounding, with differences likely attributable to differences in time (e.g., the Norman sounding was launched later than the RaXPol data used to construct the EVAD, when the low-level jet likely had intensified more) and space.

View Full Size

Upper-air maps for 0000 UTC 1 Jun 2013, at the time of the El Reno tornado at (a) 850, (b) 700, (c) 500, and (d) 250 hPa. All images are from the NOAA SPC. Shown are temperature (°C; red), dewpoint (°C; green), and height [m; without leading “1” in (a) and dam in (b) and (c); purple]; height contours are plotted as black lines. In (d), streamlines are analyzed and blue shading denotes area of wind speeds >37.5 m s⁻¹ (half barb ≈ 2.5, whole barb ≈ 5, and pennant ≈ 25 m s⁻¹).

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

The surface front visible in detail inFig. 1 is also seen at 0000 UTC at 850 hPa, mainly as a wind-shift line extending from southeastern Kansas west-southwestward across northern and northwestern Oklahoma (Fig. 7a); the temperature gradient associated with the front is not well represented in southwestern Kansas or the Texas Panhandle, owing to the elevated terrain. The front is also seen at 700 hPa, where there is a strong north-to-south temperature gradient and zone of confluence (Fig. 7b).

At 0000 UTC 1 June 2013, a major synoptic-scale cyclone was located near the northeastern corner of South Dakota at all levels. A front also is apparent in the midtroposphere at 500 hPa in southern Kansas, as the temperature varies from approximately −7°C south of the front in northwestern Oklahoma to −18°C in central Colorado (Fig. 7c). Such a midtropospheric front is characteristic of the cyclonic-shear side of a jet in the baroclinic westerlies (Bluestein 1993); at 500 hPa the airflow was from north-central Utah, southeastward into southwestern Kansas and the northern Texas Panhandle, and then into a jet streak with wind speeds of 45 m s⁻¹ over eastern Kansas and southern Iowa. A consequence of a midtropospheric front on the environment is that it was relatively warm on the anticyclonic-shear side of the jet, which lay over Oklahoma. Often subsidence warming on the anticyclonic-shear side of the jet reduces CAPE and results in an environment less conducive for supercell development. In this case, however, the surface temperature was so warm and the surface dewpoint so high that the relatively warm temperature at 500 hPa of −7°C did not prevent high CAPE from being realized.

Along with the midtropospheric jet, an upper-level jet (>37.5 m s⁻¹) was evident at 250 hPa (Fig. 7d) flowing around the base of the cyclone over northeastern South Dakota. Oklahoma was on the anticyclonic-shear side of the jet, so that anvil material from convective storms was not being evacuated from its updraft as vigorously as it would have been farther to the north in Kansas, so that the intensity of the storms in Oklahoma was probably limited.

While CAPE was relatively high and the vertical shear in the lowest 6 km was supportive of supercells, the shear was not as high as observed during some tornado outbreaks (e.g.,Grams et al. 2012), and the storms formed on the relatively warm, anticyclonic-shear side of the mid- and upper-tropospheric jet. The storms in central Oklahoma formed near the right-entrance region (right-rear quadrant) of a jet streak at 250 hPa, which was located in northeastern Kansas and northern Missouri (Fig. 7d), and near the right-front exit-region (right front) quadrant of a jet streak at 500 hPa, which was located in the northern Texas Panhandle, as evidenced by the relatively strong wind at Amarillo (Fig. 7c). The location of the storms with respect to the 250-hPa jet streak is consistent with the findings ofRose et al. (2004) andClark et al. (2009). The location of warmer air at 500 hPa in central Oklahoma is consistent with the finding of weak sinking motion in the midtroposphere in the composite right-front quadrant of a jet streak (Clark et al. 2009); in this case, the jet streak was located at 500 hPa in the northern Texas Panhandle.

In summary, storms formed along a weak stationary front, where lift likely initiated storms, and southwest of the frontal zone, along a dryline, which was marked by a zone of strong horizontal dewpoint gradient, but not a sharp, convergent wind-shift line. Since the first storms did not form along the front, but along a zone of high equivalent potential temperature near the dryline where CIN was close to zero, it is concluded that these storms were probably triggered mainly as a result of air reaching convective temperatures; we cannot quantify from the observations, however, to what extent storm formation may have been caused by a vertical circulation along the dryline.

a. Tornadogenesis and evolution of the parent supercell and other supercells

Although there was a brief tornado reported at 2255–2256 UTC (Fig. 8), the most intense and longest lived of all the tornadoes on 31 May 2013 produced by the El Reno supercell (marked with an A inFig. 12) began at 2304 UTC to the west-southwest of our deployment spot; the tornado moved initially to the southeast. The entire genesis phase of this tornado was documented by volume scans (Table 1;Fig. 10). By 2305 UTC, the tornado was marked by a weak-echo hole (WEH)¹ (e.g.,Wakimoto and Martner 1992;Bluestein et al. 2004,2007b;Wakimoto et al. 2011;Tanamachi et al. 2012,2013), thought to be caused by centrifuging of the most massive radar scatterers (Dowell et al. 2005), at the tip of a hook echo. The storm as it approached at 2258 UTC had a laminar and striated base, suggestive of forced, stable lift (Fig. 11a) (Bluestein 2013). The beginning of the main tornado at 2305 UTC appeared as a large, cone-shaped condensation funnel (Fig. 11b). (Other tornado tracks are indicated inFig. 8.)

View Full Size

Radar reflectivity images (dBZ) from RaXPol at a 4°-elevation angle approximately every 5 min during tornadogenesis of the large, cyclonic tornado at (a) 2256, (b) 2301, (c) 2306, (d) 2311, (e) 2315, and (f) 2324 UTC 31 May 2013. Range rings are shown every 5 km.

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Photographs taken from RaXPol’s second deployment location (seeFig. 8) at (a) 2258 UTC (a wide-angle view of the entire storm cloud base looking to the west) and (b) 2305 UTC 31 May 2013 (a zoomed-in view, during the early stage of the large tornado, looking to the southwest). [Photographs courtesy of H. Bluestein.]

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

The parent tornadic supercell was semidiscrete when the main tornado formed (Fig. 12a), but separate cells developed upstream of the existing supercell by 2316 UTC (Fig. 12b). By 2329 UTC (Fig. 12c), when the tornado was still large and violent, even newer echoes appeared to the west and southwest. By 2346 UTC (Fig. 12d) a new supercell (marked with a B inFig. 12) had evolved just 20–30 km to the west of the supercell that spawned the large tornado near El Reno; the tornado had dissipated several minutes earlier. The new supercell tracked eastward (Figs. 12e,f); by 0029 UTC 1 June 2013 a third supercell (marked with a C inFig. 12) formed ~30 km to west of the second supercell, each of which had its own hook echo. Each of the three supercells evolved from cells that developed at the same location, ~80 km to the west-northwest of the Twin Lakes radar (KTLX), to the southwest of El Reno, near the cold front, which was reinforced by an outflow boundary from the convective storms. This behavior—the formation of successive supercells at a fixed location (colloquially referred to as an anchor point)—has been noted by storm chasers in other cases of tornadic-supercell development near the dryline and its intersection with a front or outflow boundary, but to the best of the authors’ knowledge has not been well documented in the refereed literature.²Bluestein and Parker (1993) found that backbuilding, discrete propagation in the direction opposite to that of storm motion (e.g.,Newton and Fankhauser 1964), often occurs near the dryline, and has been associated with successive development of convective storms in developing squall lines (e.g.,Bluestein and Jain 1985). Between 2300 and 0100 UTC, the supercells became increasingly separated from the line of convection to the north (Figs. 3g–i).

View Full Size

Radar reflectivity images (dBZ) from KTLX at a low elevation angle depicting the evolution of the El Reno tornadic supercell (marked with an A) and neighboring storms (marked with a B and C), every ~15 min, at (a) 2302, (b) 2316, (c) 2329, (d) 2346 UTC 31 May 2013, (e) 0001, (f) 0015, (g) 0029, and (h) 0042 UTC 1 Jun 2013. Range rings are shown every 10 km. White (black) dot denotes location of the El Reno Mesonet site (RaXPol deployment site). Arrows point to hook echoes. Elevation angles are (a) 0.9°, (b) 1.3°, (c) 0.5°, (d) 1.3°, (e) 0.5°, (f) 0.9°, (g) 0.9°, and (h) 0.5°.

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

Surges of outflow from the first two supercells were recorded at the El Reno Mesonet site several minutes after 2300 UTC and also just after 2345 UTC (Fig. 13). Each surge was associated with a maximum wind speed of ~20 m s⁻¹ from the north-northeast. The first surge occurred just after the pressure had begun to rise rapidly (~3 hPa in ~20 min) and both temperature and dewpoint had begun to fall as the air became nearly saturated. FromFig. 12a it is seen that the first surge occurred within the forward-flank core of the first supercell, while the tornado was ~10 km to the southwest. It is therefore likely that the first surge marked the passage of a gust front originating in the forward-flank downdraft region of the supercell. Visual evidence of a gust front/relatively cool outflow to the north in the forward-flank downdraft region is found inFig. 14a: looking to the west and northwest of our deployment spot toward a region just to the south of the El Reno Mesonet site, ~5 min or so after the north-northeasterly surge had passed by the site, a tail cloud is seen extending to the north of the supercell’s wall cloud (which had rapid southward-moving cloud tags) and a greenish-tinged precipitation area is seen to the north. Tail clouds are thought to be evidence of cooler air behind the forward-flank precipitation region of a supercell rising up into the storm’s main updraft (Rotunno and Klemp 1985;Atkins et al. 2014). There is numerical evidence that developing supercells that encounter a cool boundary (e.g., from outflow from earlier convective storms or from a preexisting cold front) produce low-level mesocyclones more quickly and these low-level mesocyclones are more persistent, so that the potential for tornadogenesis is greater (Atkins et al. 1999). There is also numerical evidence that when the parent supercell tracks along the boundary or more toward the warm side, low-level mesocyclones are more intense (Atkins et al. 1999). In the case of the El Reno supercell of 2013, it did interact with outflow from the north and it did track along or to the right of the cooler air, but the outflow may have been only on the storm scale, since there was no direct evidence of any meso- or larger-scale boundary.

View Full Size

(top) Temperature (°C; red line) at 1.5 m AGL, dewpoint (°C; blue line) at 1.5 m AGL, and station pressure (hPa; black line). (bottom) Wind speed (m s⁻¹; blue line) and direction (°; red line) at 10 m AGL, at the El Reno Mesonet site as a function of time (UTC) on 31 May 2013, from well before storms formed to after the El Reno tornadic storm had passed.

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

As inFig. 11, but for (a) 2309 (a wide-angle view to the west of the tail cloud at the right and a tornado indicated with an arrow) and (b) 2314 UTC. Vortices are indicated by arrows within or near the large tornado, to the southwest. One of these may be the satellite vortex noted inFig. 15. [Photographs courtesy of H. Bluestein.]

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

The second surge occurred when the pressure was rising, but not rapidly; the air was nearly saturated and both the temperature and dewpoint rose slightly rather than fell rapidly. FromFig. 12d, it is seen that the second surge occurred within the forward-flank core of the second supercell. So, in both instances the surge was associated with outflow under the forward flank of the core of a supercell, but the thermodynamic behavior was different. In this case, unlike the first surge, the second surge occurred to the rear of a preexisting cold pool. The second surge is reminiscent of secondary, internal downdraft momentum surges in supercells, but in this case the second surge came from the forward flank of a different supercell rather than from the rear flank of the same supercell [e.g., see the summary inSkinner et al. (2014)]. It may be that this surge was dynamically driven, but we find evidence only that it was probably not thermodynamically driven.

It is also noted fromFig. 13 that while convective storms were evolving ~20–30 km to the northwest, west, and southwest of the El Reno Mesonet site (Fig. 3), a relatively slow drop in pressure (~3 hPa between 1800 and 2130 UTC) was followed by a rapid drop in pressure of ~2.5 hPa in just 30 min prior to 2200 UTC. No significant changes in temperature, dewpoint, or wind direction were noted, but there was a brief maximum in wind speed of ~13 m s⁻¹ while the pressure was falling. It is hypothesized that the rapid pressure fall might have been due to the buoyancy associated with latent heat release in the growing updrafts in the developing convective storms downstream with respect to the surface wind.

b. Interaction between the main tornado and a satellite tornado

The short-term evolution, beginning ~10 min after the first surge at the El Reno Mesonet site, of the hook-echo region of the first supercell, is seen in RaXPol data depicted inFig. 15. A WEH is seen, surrounded by a ring of enhanced reflectivity thought to be associated with the tornado debris cloud (e.g.,Wurman and Gill 2000;Bluestein et al. 2007a). A primary band of reflectivity, some of which is broken up into smaller segments, is wrapped around an outer ring of enhanced reflectivity. Preceding the northeastern portion of this band is a narrow zone of weak reflectivity or a notch (e.g.,Bluestein et al. 2007a,b). Near the region of the hook echo, where the notch becomes so narrow that it is difficult to resolve anymore, a WEH appeared at 2313:54 UTC (Fig. 15a), coincident with a tornado vortex signature in Doppler velocity (not shown), ~3–4 km from the center of the main tornado. This WEH therefore seemed to be associated with a second, satellite tornado (marked by an arrow inFig. 15), which rotated cyclonically around the main tornado and dissipated as it appeared to be absorbed into it. The second tornado therefore appeared to form separately from the main tornado (Edwards 2014) and moved southward and southeastward around the main vortex.

View Full Size

Revolution of a satellite tornado around the main tornado as depicted by radar reflectivity imagery (dBZ) from RaXPol at a 4°-elevation angle approximately every 15 s, at (a) 2313:54, (b) 2314:09, (c) 2314:25, (d) 2314:41, (e) 2314:57, and (f) 2315:12 UTC 31 May 2013. The WEH associated with the satellite tornado is indicated with an arrow. Range rings are shown every 1 km.

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

RaXPol at this time was northeast of the tornado (Figs. 8a and9c), and it was assumed that the tornado would continue moving to the southeast; we realized that we were at risk from strong outflow from the north and large hail, and we would lose sight of the tornado inside a core of precipitation if we remained where we were. We chose not to relocate RaXPol to the south, because we were concerned about the possibility that we would cross directly into the path of the tornado. We also recognized that a continuing intercept strategy to the south would require traveling on minor, rural roads and would have limited options if the tornado eventually crossed the Canadian River to our southeast (see the southern edge ofFig. 8a). Instead, we chose to drive east and north to Interstate-40 (I-40) and move quickly to the east for safety and to reposition to the east or northeast of the tornado, which would not be visible during our repositioning owing to intervening precipitation. After the third deployment (Fig. 8), we were surprised to see that the tornado vortex signature was due west of us on I-40 (Figs. 8 and9d,e), not to our southwest as anticipated. The first author mistakenly thought that the original tornado may have dissipated and a new one had formed to the east or northeast, something that happens during some types of cyclical tornadogenesis (Adlerman and Droegemeier 2005). In fact, the main tornado, which had been moving to the southeast, turned to the left and moved eastward and then northeastward while we were in transit between the second and third deployment locations.

c. Multivortex behavior

As the isolated satellite vortex presented inFig. 15 had revolved partway around the main tornado, several subtornado-scale vortices (subvortices) appeared (Fig. 14b) near the time the crew moved to the third deployment spot (Fig. 8). [From chaser videos (not shown) it is seen that there were multiple vortices during our second deployment, though RaXPol was apparently not close enough to resolve any of the subvortices then.] The satellite vortex and a second, weaker vortex were resolved by RaXPol while the radar truck was in transit (Fig. 16). There are two extrema in inbound Doppler velocity to the east of the center of the tornado (Fig. 16c). The maximum in inbound Doppler velocity farther from the center of the tornado is of tornadic strength (~55–60 m s⁻¹) and covers a relatively large area outside of the smaller-scale vortex, whose asymmetric Doppler velocity couplet is only ~250–300 m in diameter; this outer wind maximum may be associated with the rear-flank downdraft (Fig. 16c, the blue region at the right). The asymmetric flow at the greater ranges from the center of the tornado seems to indicate that the rear-flank downdraft was extremely intense, although fromρ_hv data (Fig. 16b), in which the debris signature is centered on the two vortices, it appears that the outer inbound velocity maximum may have been associated with a larger-scale vortex as winds associated with the rear-flank outflow were superimposed on a 2.5–3-km-diameter, larger-scale vortex of tornadic intensity.

View Full Size

Radar imagery from RaXPol at a 3°-elevation angle, at 2315:10 UTC 31 May 2013, depicting two subvortices in the El Reno tornado at a low elevation angle. Range rings are shown every 1 km. (a) Radar reflectivity factor (dBZ); (b) copolar cross-correlation coefficient; (c) dealiased Doppler velocity (m s⁻¹), for NCP > 0.2, with the motion of the radar truck removed; and (d) spectrum width (m s⁻¹). The circles surround the subvortices in the multivortex tornado based on weak-echo holes/curls in reflectivity [in (a)], lobes of reducedρ_hv [in (b)], cyclonic vortex signatures [in (c)], and lobes of enhanced spectrum width [in (d)].

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

A subvortex was seen at 2325 UTC on the northern fringe of the main tornado (Fig. 17). The condensation funnel associated with this subvortex appeared to lean radially outward with height, as is seen in some multivortex tornadoes and laboratory models of multivortex tornadoes [e.g.,Bluestein and Pazmany (2000), their Fig. 2;Church et al. (1979)], probably as a result of frictional inflow near the ground.Rotunno (1984) showed that subvortices should lean outward slightly and in the direction of the azimuthal component of the wind; it may be inferred then that projected onto a vertical plane they can appear to lean outward with height. The broad extent of the main tornado condensation funnel is striking and will be discussed in more detail in another paper.

View Full Size

Photograph of the large tornado to the southwest, from the deployment location of RaXPol, at 2325 UTC 31 May 2013. The arrow points to a subvortex on the north side of the main tornado. [Photograph courtesy of H. Bluestein.]

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

At the start of our third deployment, the RaXPol Doppler velocity field exhibited characteristics mainly of a single-vortex tornado. Following the brief third deployment, data were collected as the radar was in motion to the east. Near 2328 UTC, while the radar truck was moving, the tornado contained multiple vortices (Fig. 18). High-resolution RaXPol data show two WEHs and a third enhanced area of reflectivity (Fig. 18a), each of which is associated with a separate cyclonic vortex signature (Fig. 18b). Each subvortex was characterized by relatively high spectrum width (Fig. 18c); the maxima in spectrum width together formed a ring of relatively high spectrum width (Fig. 18c). High spectrum width can indicate turbulence and high shear within each radar volume. A tornado debris signature defined by lowρ_hv (Fig. 18d) was coincident with the ring of enhanced radar reflectivity associated with the WEHs, etc., with the cyclonic vortex signatures, and with the ring of enhanced spectrum width; it seems likely that the presence of debris of varying sizes, shapes, and movement speeds resulted in a very complex Doppler spectrum. Convex lobes of lowρ_hv (Fig. 18d) were coincident with the cyclonic vortex signatures, including a fourth, much weaker one in the southeastern quadrant of the tornado. There is therefore evidence of four subvortices within the eastern half of the parent tornado vortex. The detailed behavior of these multiple vortices is currently being analyzed and will be reported upon in future publications.Wurman et al. (2014) attribute the deaths of storm chasers to a subvortex (or subvortices) similar to those shown inFig. 18. The subvortices were embedded within a larger-scale cyclonic vortex, whose Doppler velocity couplet was ~1.5 km wide (Fig. 18b).

View Full Size

Radar imagery from RaXPol at a 2°-elevation angle, at 2327:54 UTC 31 May 2013, depicting the multivortex stage of the El Reno tornado at a low elevation angle. Range rings are shown every 1 km. (a) Radar reflectivity factor (dBZ); (b) dealiased Doppler velocity (m s⁻¹), for NCP > 0.2; (c) spectrum width (m s⁻¹); and (d) copolar cross-correlation coefficient. The arrows point to the subvortices in the multivortex tornado based on WEHs/enhanced reflectivity [in (a)], cyclonic vortex signatures, lobes of enhanced spectrum width [in (c)], and lobes of reducedρ_hv [in (d)].

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

d. Formation of an anticyclonic tornado

The last feature to be noted in this paper is a rare anticyclonic tornado (labeledT₃ inFig. 8). This anticyclonic tornado began near 2329 UTC while RaXPol was moving eastward on I-40. It was first spotted to our southwest as a funnel cloud, underneath which there was a rotating debris cloud (not shown). Data collection at the fourth deployment location began ~3 min after the anticyclonic tornado had formed, though data were collected during tornadogenesis while the radar truck was in transit (not shown). At the fourth deployment location, the weakening, yet extremely large, main, cyclonic tornado and the anticyclonic tornado were both probed by RaXPol (Figs. 9f and19).

View Full Size

Radar imagery from RaXPol at a 3°-elevation angle, at 2332:19 UTC 31 May 2013, depicting the coexisting large cyclonic tornado (marked with a C) and smaller anticyclonic tornado (marked with an A) at a low elevation angle. Range rings are shown every 2 km. (a) Radar reflectivity factor (dBZ), (b) differential reflectivity (dB), (c) copolar cross-correlation coefficient, and (d) dealiased Doppler velocity (m s⁻¹). In (c), the width of the tornadic debris signature is in excess of 5 km along the radial direction. The effects of attenuation through rain at X band (Snyder et al. 2010) have been approximately compensated for using the ZPHI rain-profiling algorithm [in (a);Testud et al. 2000] and a relationship between specific attenuation and specific differential attenuation [in (b)].

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

The cyclonic tornado was very broad, marked by a ~5-km-wide ring of enhanced reflectivity (Fig. 19a) and a polarimetric debris signature of lowZ_DR (Fig. 19b) andρ_hv (Fig. 19c). The cyclonic vortex signature associated with the cyclonic tornado was very broad: the distance between in- and outbound maxima was ~2 km. Based on Doppler velocity (the width of the approximately ±30 m s⁻¹ velocity couplet) and the width of the polarimetric debris signature (an upper bound), the tornado was at least 4 km wide, or at least 2.5 mi wide, making it one of the widest, if the not the widest, ever documented. Centrifuging radially outward of debris (Dowell et al. 2005) could have caused the unusually wide debris signature.

There is some hint inFig. 19c of the vestige of a smaller-scale, but weak Doppler velocity couplet (brown–green couplet near the center), which may be the remains of the main tornado seen inFig. 16. It appears as if a mesocyclone-scale tornado persisted, while a small-scale tornado apparent at the center, seen inFig. 16, had weakened considerably or dissipated by the time of theFig. 19 scan.

The anticyclonic tornado formed along the rear-flank gust front, which was marked by swirls in reflectivity (Fig. 19a) and by vortex signatures (Fig. 19d). The southernmost vortex had a strong anticyclonic vortex signature along with associated polarimetric debris signatures in bothZ_DR andρ_hv. Like other observed cyclonic–anticyclonic tornado/vortex pairs in supercells (e.g.,Wurman and Kosiba 2013), these counterrotating tornadoes moved along different paths (e.g.,Snyder et al. 2007). The anticyclonic tornado turned to the right and curved to the southeast and south. From the standpoint of tornado warnings to the public, it can be confusing to issue warnings for both tornadoes.Figure 9f shows that the area warned for tornadoes included both the cyclonic and anticyclonic tornadoes, but there was no indication that the motion of each tornado differed from that of the other.

A view of RaXPol data zoomed in on the anticyclonic tornado is shown inFig. 20. The reflectivity pattern exhibited an anticyclonic hook with an inner debris ring (Fig. 20a).

View Full Size

As inFig. 19, but for the (zoomed in) anticyclonic tornado.

Citation: Weather and Forecasting 30, 3;10.1175/WAF-D-14-00152.1

• Download Figure
• Download figure as PowerPoint slide

Tornadic debris signatures of lowZ_DR (Fig. 20b) and lowρ_hv (Fig. 20c) were coincident with the tornado, which was marked by Doppler velocity shear of ~75 m s⁻¹ over ~500 m (i.e., ~0.3 s⁻¹ vertical vorticity). Based on the width of the debris signature inρ_hv and the width of the Doppler wind speeds in excess of ~35 m s⁻¹ (Figs. 20c,d), the anticyclonic tornado at this time was ~750 m wide.