1. Introduction
Distinguishing between tornadic and nontornadic supercell storms, particularly prior to the time at which a tornadic storm produces its first tornado, is an important challenge faced by the research and forecasting communities. Significant efforts have been made to compare and contrast the environments of tornadic and nontornadic storms in observations and models, which have led to important improvements in forecasting. Detailed case studies using very high-spatial-resolution and very high-temporal-resolution data have significantly increased understanding of supercells. Despite these and other efforts, the lead time of tornado warnings (i.e., the difference between issuance and tornado occurrence) has remained relatively unchanged from 1986 to 2011, averaging 18.5 min (Stensrud et al. 2013; Brooks and Correia 2018). One potentially important reason for this lack of progress in tornado warning performance is our incomplete knowledge of tornadogenesis. While a variety of processes have been found that generate broadscale near-ground vorticity in supercells and/or tilt and stretch this vorticity to tornadic magnitude (Rotunno and Klemp 1985; Davies-Jones and Brooks 1993; Straka et al. 2007; Marquis et al. 2012; Kosiba et al. 2013; Schenkman et al. 2014; Markowski and Richardson 2014; Dahl 2015; Davies-Jones 2015; Markowski 2016; Rotunno et al. 2017; Roberts et al. 2020), much remains unknown about the relative importance of these tornadogenesis processes, how they are modulated by dynamically adjacent processes (e.g., inhibition of vertical vorticity stretching by negatively buoyant outflow), how the importance of all of these processes varies with the near-storm environment, and so forth. These knowledge gaps imply the existence of yet-unknown relationships between tornado potential, storm characteristics, and near-storm environment characteristics that could be exploited to improve tornado forecasting.
Likely the most immediate and important obstacle to improving tornado warnings, however, is imposed by our limited observing network. Proximity sounding analyses of observations (Brooks et al. 1994; Rasmussen and Blanchard 1998; Parker 2014) and model output (Markowski et al. 2003; Thompson et al. 2003, 2007; Coffer et al. 2019) have extensively documented the differences between the mesobeta environments accompanying tornadic and nontornadic supercell storms. For example, tornadic supercells are known to be favored by higher 0–1-km storm-relative helicity (SRH) and wind shear, and lower lifted condensation level (LCL). Despite substantial mean differences in such parameters, however, considerable overlap exists between the mesoscale environments of tornadic (especially weakly tornadic) and nontornadic supercells. It therefore often happens that both tornadic and nontornadic supercells occur in proximity to one another in approximately the same large-scale environment. This phenomenon, along with the spatial and temporal gradients observed in composite analyses of significantly tornadic (Potvin et al. 2010) and supercell (Parker 2014) environments, suggest that typically unsampled heterogeneities in the near-storm environment substantially modulate supercell tornado potential. While tornado forecasting techniques based on sounding indices continue to improve (Coffer et al. 2019), our ability to use observed environmental parameters to discriminate between tornadic and nontornadic supercells is fundamentally constrained by the coarse horizontal resolution of our vertical profiling networks.
The aforementioned challenges with leveraging environmental observations for storm discrimination motivates the identification of routinely observable, intrastorm markers or precursors of tornadogenesis. Challengingly, extensive overlap also occurs between the characteristics of tornadic and nontornadic supercells, including at finer scales that are only observed during field campaigns (Trapp 1999; Wakimoto and Cai 2000; Markowski 2008; Klees et al. 2016). However, several storm properties have been found to be substantially better correlated with tornadic than nontornadic supercells.
Supercells containing low-level mesocyclones (identified using radar observations) are much more likely to produce tornadoes than supercells with mesocyclones based at higher altitudes. This is because the presence of a low-level mesocyclone is necessary to produce the strong low-level updraft required to stretch near-ground vertical vorticity to tornadic magnitude in tens of minutes (Markowski and Richardson 2014). Still, at least half of supercells containing low-level mesocyclones do not produce tornadoes (Trapp et al. 2005a), and of those that do, the timing of tornadogenesis remains uncertain. Idealized simulations suggest that one common cause of tornadogenesis failure in supercells containing a low-level mesocyclone is the deceleration of vorticity-rich parcels by negatively buoyant outflow, which reduces vertical vorticity stretching (Snook and Xue 2008; Markowski and Richardson 2009, 2014). This tornadogenesis failure mode is supported by field observations showing that tornadic supercells tend to contain more buoyant rear-flank downdrafts than nontornadic supercells (RFDs; Markowski et al. 2002; Grzych et al. 2007; Hirth et al. 2008). Unfortunately, direct measurements of the thermodynamic characteristics of RFDs are generally unavailable for use in operational tornado prediction.
A second, much more observable, factor in whether a low-level mesocyclone succeeds in stretching near-ground vertical vorticity to tornadic magnitude is the degree of vertical alignment of the low-level mesocyclone with large near-ground vorticity and the midlevel mesocyclone (Wicker and Wilhelmson 1995; Snook and Xue 2008; Skinner et al. 2014; Markowski and Richardson 2014; Guarriello et al. 2018; Brown and Nowotarski 2019). Differences in vertical mesocyclone alignment between nontornadic and tornadic supercells can be inferred from a number of Doppler radar fields and will be a major focus herein.
A third leading contributor to the tornadic potential of a low-level mesocyclone is the orientation of the low-level horizontal vorticity vector relative to the storm inflow. Substantial streamwise (crosswise) vorticity has been found to promote (impede) the development of strong, steady low-level mesocyclones and therefore tornadogenesis in idealized simulations (Davies-Jones 1984; Davies-Jones and Brooks 1993; Wicker 1996; Coffer and Parker 2017) and dual-Doppler wind retrievals of real supercells (Beck et al. 2006; Murdzek et al. 2020). While the vorticity field is often poorly sampled due to radar data resolution limitations, the lack of dual-Doppler coverage at lower and middle altitudes, and the total absence of Doppler velocity data near the ground, the streamwise and crosswise vorticity magnitudes are strongly correlated with low-level SRH, as recently demonstrated in a series of idealized simulation experiments (Coffer and Parker 2017, 2018; Coffer et al. 2017). This relatively direct linkage to tornadogenesis processes explains why SRH is one of the best environmental discriminators between tornadic and nontornadic supercells. Brown and Nowotarski (2019) similarly clarified the dynamical link between tornadogenesis and another well-discriminating environmental parameter: LCL. Their idealized simulations demonstrated that the magnitude of storm outflow buoyancy, which is anticorrelated with the LCL, modulates tornado potential not only through its aforementioned influence on low-level vertical vorticity stretching, but also through its impact on the vertical alignment of the low-level mesocyclone with the midlevel mesocyclone. Such advances in our understanding of the linkages between environmental parameters, storm processes, and tornado potential can improve tornado forecasting by highlighting observable properties and features of storms and near-storm environments that are most directly connected with tornado potential (Coffer et al. 2019).
As discussed above, many previous studies have endeavored to identify differences in storm features between tornadic and nontornadic supercells. The small samples of storms investigated in some of these studies, however, limit the generality of the conclusions that can be drawn from their analyses, and may have prevented the discovery of additional tornado precursors. Radar observations are often the most utilized tool for identifying and evaluating differences between tornadic and nontornadic storms given their ability to sample detailed storm physics and kinematics in both the horizontal and vertical dimensions. However, lightning observations and geostationary satellite imagery have been increasingly used to discriminate between severe and nonsevere storms and, to a lesser extent, tornadic and nontornadic storms using patterns such as the lightning jump (rapid increases in total lightning; e.g., Schultz et al. 2017) and many cloud top features, including overshooting tops (Dworak et al. 2012), rapid cloud top cooling (Cintineo et al. 2013), and above-anvil cirrus plumes (Bedka et al. 2018). Given the increasing quantity and complexity of observations and, thereby, increased burden on forecasters to evaluate these data to aid in warning decision-making, recent efforts have begun to leverage machine learning techniques to synthesize these observations and provide probabilistic forecasts of hazards including tornadoes (e.g., see application of deep learning to radar observations in Lagerquist et al. 2020).
In this study, we generate composite analyses of observations from the NEXRAD WSR-88D network in the contiguous United States (CONUS; Crum and Alberty 1993) over a large number and range of cases to facilitate the confident and comprehensive identification of systematic differences between tornadic and nontornadic supercells. A long record of NEXRAD WSR-88D observations exists, which may offer unique opportunities to advance understanding of differences between supercell types, especially given the recent advancement of the network to dual-polarization (polarimetric) capabilities.
Some efforts have begun to identify polarimetric signatures and their differences in tornadic and nontornadic supercell storms. Kumjian and Ryzhkov (2008b) summarize routinely observed polarimetric signatures in tornadic and nontornadic supercells (18 total), many of which they relate to storm kinematics and microphysical processes (the location of the updraft, hail and graupel formation, growth, and melting, and raindrop size sorting) and the occurrence of tornadoes (lofted debris). The authors found that most of the obvious polarimetric signatures did not differ clearly enough for supercell type discrimination in their small sample, but noted that more extensive research was needed to better elucidate the potential utility of polarimetric observations for tornadic and nontornadic supercell discrimination. Several studies have since investigated supercell storm differences in raindrop size distributions within hook echoes (Kumjian and Ryzhkov 2008a; Kumjian 2011; French et al. 2015), finding that hook echoes in tornadic supercells are often characterized by smaller raindrop distributions than nontornadic supercells, which is promising for operational storm discrimination. Another promising pattern for storm discrimination is the separation between areas of enhanced specific differential phase (KDP) and differential radar reflectivity (ZDR) at low ground-relative altitudes (near 1 km AGL), where the most promising signature of separation is an orientation more perpendicular to storm motion in tornadic supercells and parallel to storm motion in nontornadic supercells (Crowe et al. 2012; Loeffler et al. 2020). Van Den Broeke (2020) recently evaluated quantitative measures of these common polarimetric signatures to distinguish between pretornadic and nontornadic supercells using observations from up to 31 storms of each type. Similar to the earlier results of Kumjian and Ryzhkov (2008b), results from Van Den Broeke (2020) suggest that the spatial extent, magnitudes, and temporal variability of many known signatures broadly overlap between pretornadic and nontornadic supercells, providing minimal utility for operationally discriminating between them. However, it was found that ZDR column size (area of enhanced positive values extending vertically above the environmental freezing level) and the extent of hail identified from the polarimetric radar variables may differ enough to be operationally useful.
This study uses a composite mean approach to identify common differences between tornadic and nontornadic supercells using observations from hundreds of storms. In particular, a probability-matched composite mean (PMM) technique that has proven to be useful for identifying and visualizing systematic differences between simulated storms generated using different models (e.g., Potvin et al. 2019b, 2020) is applied to single- and dual-polarization radar observations from the NEXRAD WSR-88D network. Rather than the tracking of individual signatures and their characteristics that is common in previous work, the PMM approach reduces emphasis on storm-to-storm variability and helps to elucidate prevailing storm structure (horizontal and vertical) and typical storm-relative locations of key microphysical and kinematic signatures and their differences between storm types. The methods used and results summarized in the subsequent sections demonstrate that robust differences in storm structure, kinematics, and microphysics exist between tornadic and nontornadic supercells, and these differences exist both near the time the first tornado is produced in a tornadic storm and at times up to 30 min prior.
2. Data and methods
a. Event selection
Events (severe weather days) analyzed in this study are selected to capture a wide range of environmental conditions and severe weather coverage, and to overlap considerably with the time period during which polarimetric radar data are available. Events are from a 7-yr time period (2011–17) and include 127 severe weather days (81 of which overlap with the polarimetric radar record). Most events analyzed occurred during spring (MAM; ~59%) and summer (JJA; ~29%), reflecting the common seasonality of the frequency of severe weather and supercells. Based on an objective technique applied to the radar data analyzed (outlined below), 478 nontornadic and 294 tornadic supercells are identified for analysis from all available cases (Fig. 1), with 341 nontornadic and 149 tornadic supercells identified from cases that include polarimetric observations.
Locations of tornadic (red circles) and nontornadic (blue circles) supercells analyzed in this study.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
b. Radar
Level II radar data (volume scans) from the NEXRAD WSR-88D network were retrieved from the National Centers for Environmental Information (NCEI; NOAA/NWS/ROC 1991) for all cases on a polar grid with sampling every 0.5° to 1.0° in azimuth, every 250 m in range, and up to 14 elevations (5 is common for clear-air scans, while 14 is common for deep convection). All NEXRAD volume scans include radar reflectivity at horizontal polarization ZH, radial velocity VR, and velocity spectrum width σV. Polarimetric radar data have been available for all NEXRAD WSR-88Ds since early 2013 and include three additional variables: ZDR, differential propagation phase shift ϕDP, and copolar correlation coefficient ρHV.
These NEXRAD WSR-88D observations provide information on the characteristics of hydrometeors and storm kinematics that can be used to elucidate important microphysics and/or dynamics processes taking place within supercell storms. For the microphysical variables, ZH provides information on hydrometeor concentration and size; ZDR, the difference between reflectivity factors at horizontal and vertical polarizations, provides a reflectivity-weighted observation of average particle shape, orientation, and phase composition; ϕDP and one-half its range derivative KDP provide information on the total mass of nonspherical particles; and ρHV provides information on the diversity of particle type and phase within the sample volume. Additional detailed descriptions of the polarimetric variables and their physical meaning can be found in several textbooks (e.g., Doviak and Zrnić 1993; Bringi and Chandrasekar 2001; Ryzhkov and Zrnić 2019) and review papers (e.g., Herzegh and Jameson 1992; Hubbert et al. 1998; Zrnić and Ryzhkov 1999; Straka et al. 2000; Ryzhkov et al. 2005; Kumjian 2013a,b,c).
The kinematic variables measured by the NEXRAD WSR-88Ds, VR and σV, provide information on the motion of hydrometeors toward and away from the radar in a sample volume and its variability, respectively. Azimuthal shear and radial divergence of VR provide estimates of rotation and horizontal divergence in storms, which can help to reveal the location of the convective updraft in a supercell (e.g., Smith and Elmore 2004). For an approximately axisymmetric feature such as an updraft, the radial divergence is typically an estimate of half the horizontal divergence, with greater deviations and errors from this approximation as departures from axisymmetry increase (outside of the updraft) and as the elevation of the radar scan increases and contributions to radial divergence from vertical velocity become more significant [up to 10% at altitudes below 5 km AGL based on O(10) m s−1 horizontal and vertical winds]. Through an assumption of mass continuity for an incompressible or anelastic atmosphere and a surface boundary condition (typically a vertical velocity of zero), the radial divergence can be integrated vertically (upward) to estimate vertical motion (see supplemental material for more detail). The strength of the updraft and rotation within a storm can also be inferred through analyses of σV as a function of altitude (e.g., Doviak and Zrnić 1993; Feist et al. 2019). In particular, turbulence is often the dominant contributor to σV in the middle and upper troposphere and relates well to updraft speed at these altitudes, while rotation (horizontal shear) contributes significantly to σV at low levels.
For analysis of NEXRAD WSR-88D data, all individual volumes are first merged onto a common large-area grid using the Gaussian space- and time-weighted binning algorithm known as Gridded NEXRAD WSR-88D Radar (GridRad), outlined in Homeyer and Bowman (2017). Data out to 300 km in range from each radar contribute to the GridRad data, but data at far ranges carry minimal weight when near-range data is present (true for all cases analyzed in this study). Elevation scans from individual radar volumes are binned into the GridRad volumes if they fall within ±3.8 min of the analysis time and weighted based on the proximity to the analysis time. The resulting multiradar GridRad volumes include up to seven variables on a grid with ~0.02° × ~0.02° longitude–latitude resolution (48 grid points per degree) and a vertical resolution of 0.5 km for altitudes below 7 km above mean sea level (MSL) and 1 km for altitudes between 7 and 22 km. GridRad data are made at 5-min intervals for all analyzed events and include ZH and kinematic variables for all cases, with polarimetric variables included for events occurring in 2013 or later. The seven potential variables included in a GridRad volume are ZH, σV, azimuthal shear, radial divergence, ZDR, KDP, and ρHV.
More detail on the general steps involved in merging individual NEXRAD volumes to create GridRad data can be found in Homeyer and Bowman (2017), but a few important steps used for merging the kinematic and polarimetric variables are briefly summarized here. Merging of azimuthal shear and radial divergence of VR requires additional processing steps. Namely, because VR is often aliased from motions that exceed the maximum detectable velocity of a radar at its operating frequency, winds must first be dealiased prior to computing the azimuthal and radial derivatives of VR for a single radar and, ultimately, merging the data into the GridRad volume. Dealiasing is performed using the “dealias region based” option in the Python ARM Radar Toolkit (PyART; Helmus and Collis 2016). In the authors’ experience, incomplete or failed dealiasing is rare, but may ultimately occur in a handful of the storms analyzed here. In addition to dealiasing, multiple noise reduction methods are applied to VR before the azimuthal shear and radial divergence are calculated to minimize the impact of measurement uncertainty. The resulting azimuthal shear and radial divergence are expected to have a maximum (i.e., worst case) uncertainty of 0.004 s−1, and often much smaller due to the VR noise reduction methods applied. In addition, due largely to the range- and altitude-resolution dependence of the azimuthal shear calculations and the GridRad binning process, the minimum resolvable scales of circulations (i.e., vortices) range from 2 to 6 km. Extensive detail on the steps of radial velocity binning process, limitations, and expected errors are summarized in Sandmæl et al. (2019). Merging of the polarimetric variables involves two unique processing steps: (i) correcting for systematic biases in ZDR due to poor radar calibration, and (ii) calculating KDP from the native ϕDP observations. In addition, when polarimetric data are available, only echoes with ρHV > 0.5 are merged into the GridRad volumes to limit deleterious impacts from contributions of nonmeteorological scatterers. Extensive detail on steps used for merging the polarimetric variables is provided in Homeyer and Kumjian (2015).
c. Tornado reports
To classify supercells as tornadic, tornado reports from NCEI’s Storm Event Database were used (National Centers for Environmental Information 2020). Each tornado report was connected with the nearest storm tracked in the GridRad data at the time of the report, as long as an updraft was found within 30 km of the tornado’s initial location. This purely objective approach was found to be reliable in previous analyses of ~100 tornadic storms where appropriate storm-report linkages were confirmed by an experienced human analyst (as in Sandmæl et al. 2019).
d. Volume extraction
Storms (both supercell and nonsupercell) are tracked in time in the GridRad data for each event using an echo-top altitude-based method (Homeyer et al. 2017). The tracking technique works by identifying and time-linking local maxima in ZH = 40-dBZ echo-top altitude in the 5-min volumes, which can miss some of the early development and late decay of a storm when ZH is weaker. Following tracking, storms are objectively classified as right-moving supercells using a simple threshold algorithm, slightly expanded on that originally described in Sandmæl (2017). Namely, a candidate storm is classified as a supercell if each of the following criteria are met: 1) maximum midlevel (4–7 km MSL) azimuthal shear exceeds 4 × 10−3 s−1 for at least 40 min, 2) the maximum midlevel azimuthal shear and that at any altitude reaches or exceeds 5 × 10−3 s−1 and 7 × 10−3 s−1, respectively, 3) the maximum radial divergence at any altitude reaches or exceeds 1 × 10−2 s−1, 4) the maximum σV at any altitude reaches or exceeds 13 m s−1, and 5) the maximum 40-dBZ echo-top altitude reaches or exceeds 11 km MSL. These thresholds used for objective supercell classification were subjectively determined via statistical evaluation and comparison of storm track observations from a large set of manually identified supercells and nonsupercells. They are designed to be strict, such that few (if any) nonsupercells satisfy them. Thus, some weak supercells are missed.
To extract individual GridRad storm volumes for PMM analysis, the direction of the 30-min average storm motion for each supercell (from echo top tracking) is used to rotate the storm volumes such that the storm motion vector is aligned with the positive x dimension. This allows for the storms to be analyzed in a consistent storm-relative framework. Data for each storm is extracted in a 60 km × 60 km grid centered on each supercell at all available vertical levels in the GridRad volumes. Since the GridRad data are made in altitude MSL, terrain heights from NCEI (NOAA/NCEI 1988) are used to convert the altitudes to above ground level (AGL) for final PMM analysis.
e. Probability matched composite means
The primary approach to analysis in this study is evaluation of PMMs of the radar variables. A PMM is a unique type of composite mean in that it uses a cumulative distribution of “typical” values for a given variable (found by accumulating the instantaneous data from all individual events) to rescale the smooth (and often reduced) values of a composite mean (Ebert 2001). This process is illustrated in Fig. 2. In the analysis outlined here, probability matching is done for constant altitude maps of each radar variable. That is, individual maps of a single radar variable at a single altitude are accumulated to determine the typical cumulative distribution function of that variable; then, the composite mean of the same variable and altitude is rescaled by its own cumulative distribution function to match the values in the event-based cumulative distribution. To limit the influence of extremes on the rescaling procedure, we first compute a cumulative distribution function at a given altitude and trim values below the 1st and above the 99th percentiles to eliminate outliers. The trimmed distribution is then linearly stretched to the complete 0th–100th-percentile range to rescale the composite means. This approach is applied to the rotated, storm-relative data volumes described in section 2d.
An illustration of the probability matching process for spatial composite mean analysis. Blue and green lines show idealized cumulative frequency distributions of a variable from raw input data (e.g., individual maps of storms) and their corresponding composite mean, respectively. Probability matching replaces the average (often reduced) value of a composite mean at a given cumulative frequency with the “typical” value from the distribution of raw data.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
The strength of PMM analysis for supercell evaluation is that it helps reveal common observational features and their storm-relative positions in a way that is less sensitive to the deleterious impacts of broadening and muting of the typical magnitudes of such features common in traditional composite means. Further, a PMM analysis helps reveal patterns that may be partially or irregularly observed in individual storms (due to poor radar coverage, miscalibration, random errors or artifacts, etc.), but can be gleaned by careful collection of a large number of observations. Increased variability in the storm-relative positions (or even occurrence) of significant features could limit their representation in PMMs, but this is primarily true for fields where competing/compensating features (e.g., maxima and minima of opposite sign) are found or if there are multiple common modes of storm-relative features (for which averaging will reduce emphasis of the modes or provide an intermediate feature that is inconsistent with the modes responsible—a worst case scenario). Otherwise, PMMs are an advantageous pathway for reducing the effects of observational errors or uncertainty in the radar fields to diagnose repeatable and substantial differences between storm types or evolutionary stages. It is important to analyze PMMs in concert with individual storm maps so that variability in and drivers of features and their storm-relative locations can be well contextualized. Individual storm maps for several variables used for PMM analysis are provided as supplemental material.
For PMMs of tornadic supercells, several time periods were examined to investigate at which, if any, point of a tornadic storm’s evolution the supercells showed distinguishing characteristics from nontornadic storms. These periods included 30, 20, 15, 10, and 5 min before and at the time of the closest 5-min radar analysis time to the first tornado reported in a storm. This results in an uncertainty from the radar analysis of up to ±2.5 min relative to the reported tornadogenesis time, but note that the report timing itself typically has a slightly larger uncertainty of up to ±5 min (Stumpf et al. 1998; Witt et al. 1998). The different lead time increments prior to tornadogenesis showed consistent results, so the analysis presented here uses a single lead time to represent the pretornadic characteristics of tornadic supercells. The lead time selected for presentation is 20 min prior to the first tornado (see Table 1), since this is the longest lead time where the number of storms contributing to the PMMs is not too different from that at tornadogenesis, and to reflect what could be expected to be seen at lead times comparable to the average tornado warning lead time found by Brooks and Correia (2018). For nontornadic supercells, two time instances are selected for comparative PMM analysis: 1) the time of maximum 40-dBZ echo-top altitude reached during a storm’s tracked life cycle (based on recent work by Sandmæl et al. (2019) demonstrating tornadic storms reach their maximum 40-dBZ echo-top altitude preceding the first tornado and during tornado occurrence), and 2) the time of maximum midlevel rotation, which is a common choice in prior work and often used for guidance during forecaster decision-making.
For tornadic supercells, the fraction of enhanced Fujita (EF) scale damage rating (0–5 and unknown) assigned to the first tornado in each supercell’s tracked life cycle.
3. Results
Storm-centered PMM analyses of all seven radar variables included are summarized in the following two subsections. For most variables, maps of PMMs are provided at five ground-relative altitude levels (0.5, 1.5, 3, 5, and 10 km AGL). Results of the PMM analyses are first summarized for ZH and the kinematic variables (azimuthal shear, radial divergence, estimated vertical velocity, and σV) since these data are available for all cases analyzed, followed by PMM analysis of the polarimetric variables (ZDR, KDP, and ρHV).
a. Radar reflectivity and kinematics
PMM analysis of ZH (Fig. 3) reveals some expected and well-known signatures in supercell storms. In particular, both nontornadic and tornadic PMMs indicate the presence of a hook echo and echo appendages extending away from the main precipitation core at low levels (≤3 km AGL). Despite the consistency of this feature in all PMM maps, its character differs between the tornadic and nontornadic storms. At the time of tornadogenesis and at 20 min prior, tornadic supercells show a more pronounced echo appendage, a hook echo that is commonly displaced ~10 km right of storm center (relative to storm motion), and a larger area of ZH > 55 dBZ than nontornadic supercells at low levels. For context, we define the hook echo as the “kink” or relatively sharp directional change in low-level reflectivity contour coincident with the storm mesocyclone (right of storm motion in the maps here) and the echo appendage as the extension of low-level echo away from the hook feature and primary precipitation core. The hook echo in nontornadic supercells (both at peak echo-top height and midlevel rotation) is commonly found ~5–10 km down-motion of that in tornadic supercells (i.e., into the lower right storm-relative quadrant). Based on the analysis that follows, these differences in hook echo location likely reflect differences in low-level updraft location. From inspection of individual storm maps, the apparent differences in the echo appendage that extend right of storm motion in the lower left storm-relative quadrant are largely driven by increased variability in the orientation of the echo appendage in the nontornadic storm sample rather than differences in occurrence, magnitude, or size. Apart from these differences in low-level ZH PMM features, the magnitudes and spatial distribution of ZH differ little between nontornadic and tornadic supercells at higher altitudes.
Probability matched composite mean (PMM) maps at altitudes of (from top to bottom) 10, 5, 3, 1.5, and 0.5 km AGL of radar reflectivity at horizontal polarization (ZH) for (from left to right) nontornadic supercell storms at the time of maximum ZH = 40-dBZ echo-top altitude, nontornadic supercells at the time of maximum midlevel rotation, tornadic supercells at the time of tornadogenesis for the first tornado linked with each storm, and tornadic supercells at 20 min prior to the first tornado. Thick, smooth transparent contours atop each map show ZH = 35 dBZ at 1.5 km AGL. Gray “downhill” contours (dashes point toward low values) indicate the number of observations (i.e., storms) contributing to the mean and are given in intervals of 50 up to a maximum of 200.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
The most consistent and apparent differences between tornadic and nontornadic supercell storms of all variables analyzed are found in PMMs of azimuthal shear (Fig. 4). Namely, the location of the mesocyclone is vertically aligned in tornadic supercells both at tornadogenesis and 20 min prior and vertically misaligned (offset spatially in altitude) in nontornadic supercells. In particular, for tornadic supercell storms, the mesocyclone is found in a consistent storm-relative location–within 10 km of storm center and along the right of storm motion axis at all altitudes analyzed. For nontornadic supercells, the low-level (≤3 km AGL) mesocyclone is commonly displaced ~5 km down-motion of that in tornadic supercells and the upper-level (at 10 km AGL) mesocyclone is displaced ~3 km up-motion, leading to an ~8-km horizontal misalignment vertically. Apart from these clear and consistent differences in mesocyclone location as a function of altitude, the magnitudes of azimuthal shear observed at altitudes ≤3 km AGL differ little between tornadic and nontornadic supercells. Larger differences occur at the lowest ground-relative altitude analyzed (0.5 km AGL), consistent with the expectation that many of the nontornadic and pretornadic supercells lack a strong low-level mesocyclone.
As in Fig. 3, but for azimuthal shear (rotation). Mesocyclone locations at each altitude are indicated by the superimposed yellow circles (subjectively drawn to coincide closely with the maximum rotation broader area of positive values).
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
There are also prominent anticyclonic rotation signatures at most altitudes in the PMMs, which are located adjacent to the mesocyclone. For nontornadic supercells, the anticyclone is located primarily in the lower left storm-relative quadrant at low levels (≤3 km AGL) and in one of the upper storm-relative quadrants at higher altitudes. For the tornadic supercells, the anticyclone is less evident at low levels but is located in a consistent storm-relative location with altitude (in the upper storm-relative quadrants). It is not clear based on analysis of azimuthal shear alone why these differences in the orientation of a cyclone–anticyclone dipole relative to storm motion exist, but it may be related to differences in the three-dimensional locations of vertical drafts between supercell types (discussed further in the following paragraphs). At low levels, it may also be indicative of differences in the storm-relative orientation of horizontal vortex lines (or similarly, wind shear) tilted into the vertical dimension by the updraft, thereby producing counterrotating vortices (e.g., Straka et al. 2007). The orientations of the nontornadic and tornadic cyclone–anticyclone dipoles at altitudes ≤3 km AGL are somewhat consistent with that expected from tilting of the common storm-relative low-level wind shear vectors for each supercell type identified in the modeling work of Coffer and Parker (2017).
PMMs for radial divergence (Fig. 5) reveal signatures that are somewhat consistent with the azimuthal shear results. Namely, strong low- to midlevel (0.5–5 km AGL) convergence is largely contained within the same storm-relative sector as the mesocyclone in tornadic supercells, indicating broad overlap between rotation and implied updraft locations. In nontornadic supercells, convergence at the lowest altitude (0.5 km AGL) is found in a comparable storm-relative location to the mesocyclone, but displaced to the left of the mesocyclone location at higher altitudes (especially at 1.5 and 3 km AGL). This displacement implies that the primary storm updraft in nontornadic supercells does not overlay the low-level updraft and mesocyclone, thereby reducing the probability that low-level rotation could be stretched to tornado strength (Markowski and Richardson 2009, 2014). In addition, for nontornadic storms at the time of peak midlevel rotation, convergence near the low-level mesocyclone is the weakest of all storm periods analyzed at 0.5 km AGL. Otherwise, the magnitudes and area covered by low- to midlevel convergence (and upper-level divergence) differ little between tornadic and nontornadic supercells.
As in Fig. 3, but for radial divergence.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
To better assess the likely relationships between azimuthal shear and radial divergence signatures and vertical drafts in these storms, we provide PMMs of estimated vertical velocity (Fig. 6) by applying an anelastic mass continuity assumption to radial divergence volumes. In particular, the radial divergence is treated as half of the horizontal divergence, such that it is first doubled and then vertically integrated from the surface up to 5 km AGL to compute profiles of vertical velocity in individual storm volumes. The surface boundary condition (i.e., at ground level) for vertical velocity integration is 0 m s−1. There are large sources of error in these estimates that predominantly affect the magnitudes of vertical velocity (see supplemental material) such that they facilitate only a qualitative comparison of vertical drafts between supercell storms. PMM analysis of this “retrieved” vertical velocity is restricted here to data that allow for continuous retrieval from 0.5 to 5 km AGL. This approach ensures dynamical consistency in the PMM analysis and primarily reveals the storm updraft (located near radar-diagnosed storm center). A rear-flank downdraft (RFD; downward motion in the rear storm-relative quadrants) is somewhat apparent in the nontornadic supercells at peak echo-top height and in the tornadic storms at tornadogenesis, but inspection of individual storm maps reveals that the location of downdrafts varies considerably in each sample and the resulting PMMs are not representative of typical magnitudes. The retrieved updrafts in both supercell storm types are found at similar storm-relative locations aloft, but differ in location with decreasing altitude. Namely, the area of the low-level updraft in nontornadic supercells is displaced to the right of storm motion relative to the broad midlevel (3–5 km AGL) updraft, but not to the extent at which the mesocyclone is displaced. Similarly, the area of strongest estimated vertical velocity at each altitude in the tornadic supercells is more vertically aligned and broadly overlaps with the vertically aligned mesocyclone.
As in Fig. 3, but for single-Doppler vertical velocity estimates up to 5 km AGL. Vertical velocity is calculated for individual storms by assuming a surface boundary condition of 0 m s−1, applying anelastic mass continuity, and assuming axisymmetry, such that horizontal divergence is equal to twice the radial divergence at each altitude. Only grid columns that have continuous radial divergence observations from 0.5 to 5 km AGL in individual storms contribute to the means here. Midlevel updraft locations in each storm population are enclosed by the orange ellipses (subjectively drawn to encompass the broad maximum in retrieved vertical velocity at 5 km AGL).
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
The final kinematic field used for PMM analysis is σV (Fig. 7). The σV PMMs reveal notable but somewhat inconsistent differences between tornadic and nontornadic supercells. In general, differences in the PMMs are limited to the area encompassed by high σV (≥4 m s−1), which is slightly greater in tornadic supercells at low- to midlevels. The highest σV magnitudes for both supercell types are found at upper levels (10 km AGL), where contributions from turbulence in the strong updrafts of these storms tend to dominate (e.g., Feist et al. 2019). σV at the lowest levels (≤1.5 km AGL) is most variable and there are slight spatial offsets (≤5 km) in the storm-relative location of the greatest magnitudes observed, with enhanced σV more commonly coinciding with the hook echo location at these altitudes in tornadic supercells. Comparing these PMM results with those of the remaining kinematic variables reveals that σV maxima are most consistent with the location of the storm updraft in both supercell types.
As in Fig. 3, but for radial velocity spectrum width (σV).
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
b. Polarimetric characteristics
While based only on slightly more than half of the total number of supercells analyzed, PMMs of the polarimetric variables do reveal some clear and consistent differences between tornadic and nontornadic supercells. PMMs of ZDR (Fig. 8) show the largest differences of the three polarimetric variables analyzed. Namely, at midlevels near the freezing, or 0°C, altitude (which varies from 3.5 to 4.5 km AGL for nearly all storms analyzed here), there exists a ZDR dipole, with a circular feature of large positive values (≥1 dB) and a similar adjacent feature of slightly negative values. This ZDR dipole indicates the presence of two well-known microphysical features in supercell storms: the ZDR column often collocated with the storm updraft (i.e., the positive feature; e.g., Kumjian et al. 2014, and references therein) and scattering from large, wet hail that is often found atop and adjacent to the ZDR column (i.e., the negative feature; e.g., Balakrishnan and Zrnić 1990; Homeyer and Kumjian 2015). Given this analysis is at constant altitude, the negative element of the dipole is most likely indicative of the location of hail fallout as it is advected out of the main storm updraft (indicated by the positive element of the dipole, which is coincident with the updraft location in the diagnosed vertical velocity in Fig. 6). In tornadic storms, the ZDR dipole (positive to negative) is oriented in a direction extending from the lower left to upper right storm-relative quadrants both at tornadogenesis and at 20 min prior (i.e., more parallel to storm motion). In nontornadic storms, the dipole is oriented approximately 45° left of that in tornadic storms: from the lower-left to upper-left storm-relative quadrants (approximately perpendicular to the storm motion vector). This change in dipole orientation likely indicates differences in hydrometeor trajectories and size sorting owing to the differing three-dimensional kinematics of tornadic and nontornadic supercells discussed previously in section 3a and thoroughly diagnosed in previous observational and modeling studies (e.g., Kumjian and Ryzhkov 2008b, 2012; Dawson II et al. 2014). Given the orientation of the dipole and the expectation that hail formation is most efficient atop and immediately adjacent to the ZDR column, this result implies that storm-relative flow is left of storm motion in nontornadic supercells and increasingly down-motion in tornadic storms at these altitudes. While the difference in dipole orientation is apparent in the means, inspection of individual storm maps reveals substantial variability in dipole occurrence. This is not surprising given that ZDR columns often exhibit considerable temporal variability in area and depth due to updraft pulses and hail fallout (e.g., Kumjian et al. 2014; Snyder et al. 2015; Kuster et al. 2019).
As in Fig. 3, but for differential radar reflectivity (ZDR). Thin white contours outline areas of ZDR = 0 dB at altitudes of 3, 5, and 10 km AGL. Thick white and black contours at 1.5 km AGL outline areas of enhanced KDP (≥2° km−1) and ZDR (≥3 dB) at this altitude, respectively.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
Comparative analysis of PMMs of ZDR and KDP (Fig. 9) at low levels (highlighted by superimposed contours at 1.5 km AGL) reveals separation of enhanced areas of each variable that is consistent with previous studies. Namely, the orientation of the separation is more perpendicular to storm motion in tornadic supercells both at and prior to tornadogenesis, and is more parallel to storm motion in nontornadic supercells, as recently found in Loeffler et al. (2020) and largely attributed to differences in hydrometeor size sorting between supercell types (as previously argued for the midlevel ZDR dipole). While no apparent differences in the extent of separation (or overlap) of these enhanced features exists, inspection of PMM ρHV (Fig. 10) provides new insight into this polarimetric signature. In particular, it is found that a ρHV minimum at 1.5 km AGL (and additional nearby altitudes) is roughly collocated with the enhanced KDP feature in nontornadic supercells and along the interface between enhanced KDP and ZDR features in tornadic supercells. A collocation of the ρHV minimum with the KDP maximum implies a greater likelihood of small and/or melting hail in nontornadic storms, supported further by higher ZDR collocated with the KDP maximum than that found in tornadic supercells. The ρHV minimum is also found to be lower in tornadic supercells than that in nontornadic supercells, also indicating a greater likelihood of large hail given its overlap with the maximum ZH found at these altitudes (e.g., Ortega et al. 2016). Apart from these low-level polarimetric signatures and the midlevel ZDR differences identified, the polarimetric variables at remaining altitudes differ little between nontornadic and tornadic supercells.
As in Fig. 3, but for specific differential phase (KDP). As in Fig. 8, areas of enhanced KDP and ZDR at 1.5 km AGL are outlined.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
As in Fig. 3, but for copolar correlation coefficient (ρHV). As in Fig. 8, areas of enhanced KDP and ZDR at 1.5 km AGL are outlined.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
c. Case study
To determine if the PMM results are repeatable in individual supercells and operationally useful, three objectively identified case studies (out of a total of six found in our database) of neighboring tornadic and nontornadic supercells that initiate and mature in close proximity to one another in time and space are examined. Such neighboring supercells are desired for analysis to minimize mesoscale environmental differences in our comparison of tornadic and nontornadic storm structure. We focus only on mesocyclone alignment here because the remaining key discriminating features identified (viz., the midlevel ZDR dipole and the low-level enhanced ZDR–KDP separation) are confined to relatively shallow layers and not well-sampled by single radars throughout a storm’s lifetime, such that they cannot be reliably evaluated. The three cases chosen for analysis and presentation span time of year, location, and significance. The cases include supercells in the vicinity of NEXRAD WSR-88Ds in Frederick, Oklahoma (International Civil Aviation Organization code KFDR): southwestern Oklahoma on 18 May 2017 (Fig. 11), Sioux Falls, South Dakota (KFSD): northwestern Iowa on 18 June 2014 (Fig. 12), and Oklahoma City, Oklahoma (KTLX): southeastern Oklahoma on 14 April 2011 (Fig. 13). The storms in the 18 May 2017 and 14 April 2011 cases were long-lived supercells, with the nontornadic and tornadic supercells having radar-tracked lifetimes of 320 and 295 min and 255 and 330 min, respectively. The supercells in the 18 June 2014 case were relatively short lived, with radar-tracked lifetimes of 70 min (nontornadic) and 120 min (tornadic). The first tornadoes produced by the tornadic supercells (the times relative to which are those focused on here) include a 9-min EF-1 (18 May 2017), a 2-min EF-0 (18 June 2014), and a 2-min EF-1 (14 April 2011).
Maps of (left) radar reflectivity at horizontal polarization ZH, (center) dealiased radial velocity VR, and (right) azimuthal shear for select times during the case study of NEXRAD WSR-88D observations from the Frederick, Oklahoma, site (KFDR) on 18 May 2017. (top two rows) The tornadic supercell analyzed is denoted by the red arrows and labels and (bottom two rows) the nontornadic supercell analyzed is denoted by the blue arrows and labels. All maps show observations from the 0.5° elevation scan in the volume (the lowest available). The plus symbol in each panel denotes the radar location and the black circles indicate distance from the radar (every 50 km). State county boundaries are outlined in black.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
As in Fig. 11, but from the Sioux Falls, South Dakota, site (KFSD) on 18 Jun 2014.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
As in Fig. 11, but from the Oklahoma City, Oklahoma, site (KTLX) on 14 Apr 2011.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
To diagnose mesocyclone alignment in each supercell, VR observations from the radar volumes are first dealiased following the approach used in GridRad data creation outlined in section 2b. Dealiased VR data from each elevation scan in each radar volume are then subjectively evaluated to determine mesocyclone locations. Namely, through tracking of ZH hook echoes, dealiased VR inbound-outbound couplets, and positive azimuthal shear objects, mesocyclone locations are subjectively identified as the approximate centroid of the azimuthal shear object. Figures 11–13 show ZH, dealiased VR, and azimuthal shear at the lowest elevation scan (0.5°) for four select times from the analysis period of each case, with either the tornadic or nontornadic supercells labeled in each panel. The exact time periods analyzed vary slightly for each case, with the tornadic storms spanning 20 min before to 20 min after the tornado, and periods for nontornadic storms spanning an equivalent time window, but centered on the time of maximum midlevel rotation.
For ease of comparison between the PMM results and the mesocyclone structures in these cases, subjectively tracked mesocyclones are binned into three altitude layers: low level (0.5–3 km AGL), midlevel (4–7 km AGL), and upper level (≥8 km AGL). Once binned and time sorted, mesocyclone locations in each altitude layer are interpolated to a regular 1-min time series and smoothed using 5-min boxcar averaging to minimize potential location errors from the subjective identification procedure. Figure 14 shows density plots of the resulting differences in the diagnosed location of the low- and midlevel mesocyclones for each supercell. From a visual comparison of the tornadic and nontornadic density plots, the tornadic supercells more frequently show alignment of the low- and midlevel mesocyclones than the nontornadic supercells. In fact, 58.57%, 36.84%, and 39.58% of all 1-min midlevel mesocyclones exist within a 4 km × 4 km box bounding the low-level mesocyclone location in the tornadic supercell storms for the 2017, 2014, and 2011 events, respectively, compared to 35.71%, 34.33%, and 22.92% for the nontornadic storms. These results support the general result in Fig. 4; that is, the low- and midlevel mesocyclones are often well aligned in tornadic storms, and often misaligned in nontornadic storms.
Density plots of the differences in location of the manually tracked midlevel (4–7 km AGL) and low-level (0.5–3 km AGL) mesocyclones for the (left) tornadic and (right) nontornadic supercells for each of the case studies summarized in Figs. 11–13: (top) 18 May 2017, (middle) 18 Jun 2014, and (bottom) 14 Apr 2011. Differences are shown in distance relative to the low-level mesocyclone position and densities are the fraction of 1-min mesocyclone locations during the analysis period (~45 min for each storm).
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
To examine the tornadic supercell life cycles in greater detail and at similar times to those used for the PMM analysis, Fig. 15 breaks down the density plots from Fig. 14 into analysis time periods 20 min before, during, and 20 min after the tornado produced by each storm. For the time periods before and during the tornado, nearly all 1-min midlevel mesocyclone observations in each event exist within a 4 km × 4 km box bounding the low-level mesocyclone location and it is rare to find deviations more than ~3 km from the low-level mesocyclone location, indicating persistent vertical alignment of mesocyclones during these time periods. Conversely, the alignment of low- and midlevel mesocyclones appears to break down after tornado cessation, especially in the storms with short-lived tornadoes (the 2014 and 2011 events), with a much larger fraction of midlevel mesocyclone observations found outside the bounding box and in more variable low-level mesocyclone-relative locations. While not shown, a similar breakdown of the nontornadic supercell life cycles reveals that misalignment is a consistent feature in those storms (i.e., there is little time sensitivity).
As in Fig. 14, but for the tornadic supercell only at times relative to the tornado generated by the storm (left) 20 min before, (center) during, and (right) 20 min after tornado occurrence.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
4. Conclusions and discussion
This study used a large set of radar observations and a unique composite mean analysis technique (probability matched means or PMMs) to compare and contrast common kinematic and microphysical characteristics in tornadic and nontornadic supercell storms. Single- and dual-polarization radar variables from events occurring throughout a 7-yr period (2011–17) contributed to PMM analysis, which was performed for storm volumes centered on the 40-dBZ echo-top altitude maximum and rotated relative to the diagnosed storm motion vector. The resulting PMMs of microphysical (ZH, ZDR, KDP, and ρHV) and kinematic (azimuthal shear, radial divergence, estimated vertical velocity, and σV) variables revealed several robust differences between supercell types, with little sensitivity in the results to the lead time prior to a storm’s first tornado used (up to 30 min). Case studies of neighboring tornadic and nontornadic supercells within a comparable mesoscale environment revealed signatures consistent with these PMM results, indicating that differences found may be useful for short-term forecasting and warning decision-making and may also help to identify transitions to time periods with reduced tornado potential in tornadic supercells. Figure 16 provides a summary schematic of the greatest differences found between tornadic and nontornadic supercells, with all notable results summarized as follows:
Tornadic supercells are characterized by a low-level ZH hook echo aligned with and ~10 km right of storm center, while the hook echo in nontornadic supercells is displaced 5–10 km down-motion.
The mesocyclone (and storm updraft) is vertically aligned in tornadic supercells and vertically misaligned in nontornadic supercells. In particular, the low-level mesocyclone in nontornadic storms is displaced toward the forward flank of the storm and away from the primary storm updraft (that at mid- and upper levels). This result is consistent with recent modeling work and with past observational studies that identified alignment of low- and midlevel mesocyclones with the storm updraft to be important to tornadogenesis and maintenance in supercells (e.g., Dowell and Bluestein 2002; Marquis et al. 2012; Tanamachi et al. 2012; French and Kingfield 2019).
A ZDR dipole exists at midlevels (near and in a shallow layer above the 0°C isotherm) in supercells, for which its orientation relative to storm motion differs by approximately 45° between tornadic and nontornadic storms. In particular, the dipole is mostly perpendicular to storm motion in nontornadic supercells and (from positive to negative ZDR) rotates toward the storm motion vector in tornadic supercells. The positive ZDR node of this dipole represents the location of the well-known ZDR column that coincides with the storm updraft (e.g., Kumjian et al. 2014) and the negative ZDR node is a signature of hail fallout (e.g., Balakrishnan and Zrnić 1990).
The separation between enhanced ZDR and KDP at low levels is more perpendicular to storm motion in tornadic supercells and parallel to storm motion in nontornadic supercells, as recently found in Loeffler et al. (2020). In addition, it is revealed for the first time (to the authors’ knowledge) that a ρHV minimum is commonly found near and within the ZDR–KDP separation, being roughly collocated with the enhanced KDP feature in nontornadic supercells and along the overlapping boundaries of enhanced ZDR and KDP features in tornadic supercells.
Summary schematic of key distinguishing characteristics in radar observations of tornadic and nontornadic supercell storms.
Citation: Monthly Weather Review 148, 12; 10.1175/MWR-D-20-0136.1
The differences in tornadic and nontornadic supercells summarized here motivate a number of future efforts to increase understanding of these storms and our ability to forecast them. For example, an immediate step that could be undertaken is to evaluate output from convection-allowing models to determine whether or not they can replicate observations. Sensitivity tests could be performed to determine what model elements (e.g., horizontal grid spacing) are most important to prediction of supercell type. Systematic environmental differences between the tornadic and nontornadic supercells could be diagnosed (e.g., using the NOAA Storm Prediction Center Hourly Mesoscale Analysis archive) and evaluated in the context of previous proximity sounding studies. Our tornadic supercell sample could be further stratified into weak versus significant tornado producers to examine whether systematic differences from nontornadic supercells amplify for supercells that produce stronger tornadoes. Moreover, the operational utility of all supercell storm features and their differences identified in the PMMs should be better evaluated with a much larger set of case studies and single-radar observations since only the vertical mesocyclone alignment was evaluated here.
Multiple feature analysis could be used to determine how often the tornado precursors and markers highlighted in the PMMs occur in conjunction with one another, which is vital to adequately assess their potential operational utility. Similarly, multivariable techniques such as hydrometeor classification algorithms applied to the polarimetric variables could be used to improve understanding of potential differences in the three-dimensional hydrometeor distributions between supercell types, for which differences in the areal extent of hail fall are expected to be revealed given the recent results from Van Den Broeke (2020) and the low-level polarimetric differences identified here. Finally, machine learning techniques (including deep learning via convolutional neural networks) have received increasing attention and use in the atmospheric sciences, and could be trained on and applied to radar observations and other datasets to forecast tornadogenesis (e.g., Lagerquist et al. 2020). Analyses similar to those carried out in this study are useful to these machine learning applications because they help to minimize data input to observations that have the greatest potential for discrimination and model convergence (e.g., select altitudes, select variables). Such an approach is also vital to potential improvements in forecaster decision-making in the future given that available data is becoming increasingly varied and time-consuming to evaluate.
Another promising focus of future work is to apply the PMM technique used in this study to radar observations of alternative storm types. The focus on supercells in this study was motivated by their production of most significant tornadoes (Trapp et al. 2005b), but many tornadoes occur in alternative storms, of which quasi-linear convective systems (or QLCSs) are most common. Thus, applying the PMM technique to a large set of tornadic and nontornadic convective cells in QLCSs is warranted. Application of the PMM technique to radar observations categorized with respect to alternative hazards (e.g., severe hail or straight-line winds) is also expected to be worthwhile.
Despite the use of a large number of radar observations in this study, there are some caveats that could be improved upon in future work. Namely, while these results are consistent at different times of the supercell life cycle, the results are likely less clear than reality. In particular, some of the nontornadic supercells are likely misclassified since many tornadoes are unobserved and/or unreported, especially in rural areas (Potvin et al. 2019a). Further, variations in storm size are not taken into account, which could result in reduced emphasis on important storm-relative features in the PMMs. Since it is also possible that features in the PMM fields may be produced by contributions from multiple independent feature modes, it is also important for future work to confirm the reliability and occurrence of the PMM features in high-resolution individual storm observations and examine further their detailed characteristics (e.g., measuring the orientation of the midlevel ZDR dipole and the distance between mesocyclone locations as a function of altitude). Finally, while a large storm sample was analyzed here, it is possible that differences in the polarimetric variables between tornadic and nontornadic storms could be made clearer with more observations (since such data was only available for approximately half of all storms analyzed here). As the availability of polarimetric data continues to increase, future work should make greater use of these data to help clarify potential differences between tornadic and nontornadic supercell storms, especially at low levels.
Acknowledgments
All graphics in this paper were designed to be equally interpretable to individuals with full color vision and individuals with color vision deficiency. This work was supported by the National Science Foundation (NSF) under Grant AGS-1802627.
Data availability statement
All data used in this study are freely available online in archives hosted by the National Centers for Environmental Information (NCEI). All processed datasets produced by the authors for this study are available upon request.
REFERENCES
Balakrishnan, N., and D. S. Zrnić, 1990: Use of polarization to characterize precipitation and discriminate large hail. J. Atmos. Sci., 47, 1525–1540, https://doi.org/10.1175/1520-0469(1990)047<1525:UOPTCP>2.0.CO;2.
Beck, J. R., J. L. Schroeder, and J. M. Wurman, 2006: High-resolution dual-Doppler analyses of the 29 May 2001 Kress, Texas, cyclic supercell. Mon. Wea. Rev., 134, 3125–3148, https://doi.org/10.1175/MWR3246.1.
Bedka, K., E. M. Murillo, C. R. Homeyer, B. Scarino, and H. Mersiovsky, 2018: The above anvil cirrus plume: An important severe weather indicator in visible and infrared satellite imagery. Wea. Forecasting, 33, 1159–1181, https://doi.org/10.1175/WAF-D-18-0040.1.
Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar. 1st ed. Cambridge University Press, 636 pp.
Brooks, H. E., and J. Correia, 2018: Long-term performance metrics for National Weather Service tornado warnings. Wea. Forecasting, 33, 1501–1511, https://doi.org/10.1175/WAF-D-18-0120.1.
Brooks, H. E., C. A. Doswell III, and J. Cooper, 1994: On the environments of tornadic and nontornadic mesocyclones. Wea. Forecasting, 9, 606–618, https://doi.org/10.1175/1520-0434(1994)009<0606:OTEOTA>2.0.CO;2.
Brown, M., and C. J. Nowotarski, 2019: The influence of lifting condensation level on low-level outflow and rotation in simulated supercell thunderstorms. J. Atmos. Sci., 76, 1349–1372, https://doi.org/10.1175/JAS-D-18-0216.1.
Cintineo, J. L., M. J. Pavolonis, J. M. Sieglaff, and A. K. Heidinger, 2013: Evolution of severe and nonsevere convection inferred from GOES-derived cloud properties. J. Appl. Meteor. Climatol., 52, 2009–2023, https://doi.org/10.1175/JAMC-D-12-0330.1.
Coffer, B. E., and M. D. Parker, 2017: Simulated supercells in nontornadic and tornadic VORTEX2 environments. Mon. Wea. Rev., 145, 149–180, https://doi.org/10.1175/MWR-D-16-0226.1.
Coffer, B. E., and M. D. Parker, 2018: Is there a “tipping point” between simulated nontornadic and tornadic supercells in VORTEX2 environments? Mon. Wea. Rev., 146, 2667–2693, https://doi.org/10.1175/MWR-D-18-0050.1.
Coffer, B. E., M. D. Parker, J. M. L. Dahl, L. J. Wicker, and A. J. Clark, 2017: Volatility of tornadogenesis: An ensemble of simulated nontornadic and tornadic supercells in VORTEX2 environments. Mon. Wea. Rev., 145, 4605–4625, https://doi.org/10.1175/MWR-D-17-0152.1.
Coffer, B. E., M. D. Parker, R. L. Thompson, B. T. Smith, and R. E. Jewell, 2019: Using near-ground storm relative helicity in supercell tornado forecasting. Wea. Forecasting, 34, 1417–1435, https://doi.org/10.1175/WAF-D-19-0115.1.
Crowe, C. C., C. J. Schultz, M. Kumjian, L. D. Carey, and W. A. Petersen, 2012: Use of dual-polarization signatures in diagnosing tornadic potential. Electron. J. Oper. Meteor., 13, 57–78.
Crum, T. D., and R. L. Alberty, 1993: The WSR-88D and the WSR-88D operational support facility. Bull. Amer. Meteor. Soc., 74, 1669–1687, https://doi.org/10.1175/1520-0477(1993)074<1669:TWATWO>2.0.CO;2.
Dahl, J. M. L., 2015: Near-ground rotation in simulated supercells: On the robustness of the baroclinic mechanism. Mon. Wea. Rev., 143, 4929–4942, https://doi.org/10.1175/MWR-D-15-0115.1.
Davies-Jones, R., 1984: Streamwise vorticity: The origin of updraft rotation in supercell storms. J. Atmos. Sci., 41, 2991–3006, https://doi.org/10.1175/1520-0469(1984)041<2991:SVTOOU>2.0.CO;2.
Davies-Jones, R., 2015: A review of supercell and tornado dynamics. Atmos. Res., 158–159, 274–291, https://doi.org/10.1016/j.atmosres.2014.04.007.
Davies-Jones, R., and H. Brooks, 1993: Mesocyclogenesis from a theoretical perspective. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., Vol. 79, Amer. Geophys. Union, 105–114, https://doi.org/10.1029/GM079p0105.
Dawson, D. T., II, E. R. Mansell, Y. Jung, L. J. Wicker, M. R. Kumjian, and M. Xue, 2014: Low-level ZDR signatures in supercell forward flanks: The role of size sorting and melting of hail. J. Atmos. Sci., 71, 276–299, https://doi.org/10.1175/JAS-D-13-0118.1.
Doviak, R. J., and D. S. Zrnić, 1993: Doppler Radar and Weather Observations. Academic Press, 562 pp.
Dowell, D. C., and H. B. Bluestein, 2002: The 8 June 1995 McLean, Texas, storm. Part II: Cyclic tornado formation, maintenance, and dissipation. Mon. Wea. Rev., 130, 2649–2670, https://doi.org/10.1175/1520-0493(2002)130<2649:TJMTSP>2.0.CO;2.
Dworak, R., K. Bedka, J. Brunner, and W. Feltz, 2012: Comparison between/textitGOES-12 overshooting-top detections, WSR-88D radar reflectivity, and severe storm reports. Wea. Forecasting, 27, 684–699, https://doi.org/10.1175/WAF-D-11-00070.1.
Ebert, E. E., 2001: Ability of a poor man’s ensemble to predict the probability and distribution of precipitation. Mon. Wea. Rev., 129, 2461–2480, https://doi.org/10.1175/1520-0493(2001)129<2461:AOAPMS>2.0.CO;2.
Feist, M. M., C. D. Westbrook, P. A. Clark, T. H. Stein, H. W. Lean, and A. J. Stirling, 2019: Statistics of convective cloud turbulence from a comprehensive turbulence retrieval method for radar observations. Quart. J. Roy. Meteor. Soc., 145, 727–744, https://doi.org/10.1002/qj.3462.
French, M. M., and D. M. Kingfield, 2019: Dissipation characteristics of tornadic vortex signatures associated with long-duration tornadoes. Mon. Wea. Rev., 58, 317–339, https://doi.org/10.1175/JAMC-D-18-0187.1.
French, M. M., D. W. Burgess, E. R. Mansell, and L. J. Wicker, 2015: Bulk hook echo raindrop sizes retrieved using mobile, polarimetric Doppler radar observations. J. Appl. Meteor. Climatol., 54, 423–450, https://doi.org/10.1175/JAMC-D-14-0171.1.
Grzych, M. L., B. D. Lee, and C. A. Finley, 2007: Thermodynamic analysis of supercell rear-flank downdrafts from project ANSWERS. Mon. Wea. Rev., 135, 240–246, https://doi.org/10.1175/MWR3288.1.
Guarriello, F., C. J. Nowotarski, and C. C. Epifanio, 2018: Effects of the low-level wind profile on outflow position and near-surface vertical vorticity in simulated supercell thunderstorms. J. Atmos. Sci., 75, 731–753, https://doi.org/10.1175/JAS-D-17-0174.1.
Helmus, J. J., and S. M. Collis, 2016: The Python ARM Radar toolkit (Py-ART), a library for working with weather radar data in the Python programming language. J. Open Res. Software, 4, e25, https://doi.org/10.5334/jors.119.
Herzegh, P. H., and A. R. Jameson, 1992: Observing precipitation through dual-polarization radar measurements. Bull. Amer. Meteor. Soc., 73, 1365–1374, https://doi.org/10.1175/1520-0477(1992)073<1365:OPTDPR>2.0.CO;2.
Hirth, B. D., J. L. Schroeder, and C. C. Weiss, 2008: Surface analysis of the rear-flank downdraft outflow in two tornadic supercells. Mon. Wea. Rev., 136, 2344–2363, https://doi.org/10.1175/2007MWR2285.1.
Homeyer, C. R., and M. R. Kumjian, 2015: Microphysical characteristics of overshooting convection from polarimetric radar observations. J. Atmos. Sci., 72, 870–891, https://doi.org/10.1175/JAS-D-13-0388.1.
Homeyer, C. R., and K. P. Bowman, 2017: Algorithm description document for version 3.1 of the three-dimensional gridded NEXRAD WSR-88D radar (GridRad) dataset. Tech. Rep., 23 pp., http://gridrad.org/pdf/GridRad-v3.1-Algorithm-Description.pdf.
Homeyer, C. R., J. D. McAuliffe, and K. M. Bedka, 2017: On the development of above-anvil cirrus plumes in extratropical convection. J. Atmos. Sci., 74, 1617–1633, https://doi.org/10.1175/JAS-D-16-0269.1.
Hubbert, J. C., V. N. Bringi, L. D. Carey, and S. Bolen, 1998: CSU-CHILL polarimetric radar measurements from a severe hail storm in eastern Colorado. J. Appl. Meteor., 37, 749–775, https://doi.org/10.1175/1520-0450(1998)037<0749:CCPRMF>2.0.CO;2.
Klees, A. M., Y. P. Richardson, P. M. Markowski, C. Weiss, J. M. Wurman, and K. K. Kosiba, 2016: Comparison of the tornadic and nontornadic supercells intercepted by VORTEX2 on 10 June 2010. Mon. Wea. Rev., 144, 3201–3231, https://doi.org/10.1175/MWR-D-15-0345.1.
Kosiba, K., J. Wurman, Y. Richardson, P. Markowski, P. Robinson, and J. Marquis, 2013: Genesis of the Goshen County, Wyoming, tornado on 5 June 2009 during VORTEX2. Mon. Wea. Rev., 141, 1157–1181, https://doi.org/10.1175/MWR-D-12-00056.1.
Kumjian, M. R., 2011: Precipitation properties of supercell hook echoes. Electron. J. Severe Storms Meteor., 6 (5), https://ejssm.org/ojs/index.php/ejssm/article/viewArticle/93.
Kumjian, M. R., 2013a: Principles and applications of dual-polarization weather radar. Part I: Description of the polarimetric radar variables. J. Oper. Meteor., 1, 226–242, https://doi.org/10.15191/nwajom.2013.0119.
Kumjian, M. R., 2013b: Principles and applications of dual-polarization weather radar. Part II: Warm and cold season applications. J. Oper. Meteor., 1, 243–264, https://doi.org/10.15191/nwajom.2013.0120.
Kumjian, M. R., 2013c: Principles and applications of dual-polarization weather radar. Part III: Artifacts. J. Oper. Meteor., 1, 265–274, https://doi.org/10.15191/nwajom.2013.0121.
Kumjian, M. R., and A. V. Ryzhkov, 2008a: Microphysical differences between tornadic and nontornadic supercell rear-flank downdrafts revealed by dual-polarization radar measurements. 24th Conf. on Severe Local Storms, Savannah, GA, Amer. Meteor. Soc., 3B.4, https://ams.confex.com/ams/24SLS/techprogram/paper_141912.htm.
Kumjian, M. R., and A. V. Ryzhkov, 2008b: Polarimetric signatures in supercell thunderstorms. J. Appl. Meteor. Climatol., 47, 1940–1961, https://doi.org/10.1175/2007JAMC1874.1.
Kumjian, M. R., and A. V. Ryzhkov, 2012: The impact of size sorting on the polarimetric radar variables. J. Atmos. Sci., 69, 2042–2060, https://doi.org/10.1175/JAS-D-11-0125.1.
Kumjian, M. R., A. P. Khain, N. Benmoshe, E. Ilotoviz, A. V. Ryzhkov, and V. T. J. Phillips, 2014: The anatomy and physics of ZDR columns: Investigating a polarimetric radar signature with a spectral bin microphysical model. J. Appl. Meteor. Climatol., 53, 1820–1843, https://doi.org/10.1175/JAMC-D-13-0354.1.
Kuster, C. M., J. C. Snyder, T. J. Schuur, T. T. Lindley, P. L. Heinselman, J. C. Furtado, J. W. Brogden, and R. Toomey, 2019: Rapid-update radar observations of ZDR column depth and its use in the warning decision process. Wea. Forecasting, 34, 1173–1188, https://doi.org/10.1175/WAF-D-19-0024.1.
Lagerquist, R., A. McGovern, C. R. Homeyer, D. J. Gagne II, and T. Smith, 2020: Deep learning on three-dimensional multiscale data for next-hour tornado prediction. Mon. Wea. Rev., 148, 2837–2861, https://doi.org/10.1175/MWR-D-19-0372.1.
Loeffler, S. D., M. R. Kumjian, M. Jurewicz, and M. M. French, 2020: Differentiating between tornadic and nontornadic supercells using polarimetric radar signatures of hydrometeor size sorting. Geophys. Res. Lett., 47, e2020GL088242, https://doi.org/10.1029/2020GL088242.
Markowski, P. M., 2008: A comparison of the midlevel kinematic characteristics of a pair of supercell thunderstorms observed by airborne Doppler radar. Atmos. Res., 88, 314–322, https://doi.org/10.1016/j.atmosres.2007.11.026.
Markowski, P. M., 2016: An idealized numerical simulation investigation of the effects of surface drag on the development of near-surface vertical vorticity in supercell thunderstorms. J. Atmos. Sci., 73, 4349–4385, https://doi.org/10.1175/JAS-D-16-0150.1.
Markowski, P. M., and Y. P. Richardson, 2009: Tornadogenesis: Our current understanding, forecasting considerations, and questions to guide future research. Atmos. Res., 93, 3–10, https://doi.org/10.1016/j.atmosres.2008.09.015.
Markowski, P. M., and Y. P. Richardson, 2014: The influence of environmental low-level shear and cold pools on tornadogenesis: Insights from idealized simulations. J. Atmos. Sci., 71, 243–275, https://doi.org/10.1175/JAS-D-13-0159.1.
Markowski, P. M., J. M. Straka, and E. N. Rasmussen, 2002: Direct surface thermodynamic observations within the rear-flank downdrafts of nontornadic and tornadic supercells. Mon. Wea. Rev., 130, 1692–1721, https://doi.org/10.1175/1520-0493(2002)130<1692:DSTOWT>2.0.CO;2.
Markowski, P. M., C. Hannon, J. Frame, E. Lancaster, A. Pietrycha, R. Edwards, and R. L. Thompson, 2003: Characteristics of vertical wind profiles near supercells obtained from the Rapid Update Cycle. Wea. Forecasting, 18, 1262–1272, https://doi.org/10.1175/1520-0434(2003)018<1262:COVWPN>2.0.CO;2.
Marquis, J., Y. Richardson, P. Markowski, D. Dowell, and J. Wurman, 2012: Tornado maintenance investigated with high-resolution dual-Doppler and EnKF analysis. Mon. Wea. Rev., 140, 3–27, https://doi.org/10.1175/MWR-D-11-00025.1.
Murdzek, S. S., P. M. Markowski, Y. P. Richardson, and R. L. Tanamachi, 2020: Processes preventing the development of a significant tornado in a Colorado supercell on 26 May 2010. Mon. Wea. Rev., 148, 1753–1778, https://doi.org/10.1175/MWR-D-19-0288.1.
National Centers for Environmental Information, 2020: NOAA’s severe weather data inventory. National Centers for Environmental Information, accessed May 2017 to May 2018, https://www.ncdc.noaa.gov/severe-weather/severe-weather-data-inventory.
NOAA/NCEI, 1988: ETOPO5 5-minute gridded elevation data. Data announcement 88-MGG-02, digital relief of the surface of the Earth, NOAA/National Geophysical Data Center, accessed June 2018, https://www.ngdc.noaa.gov/mgg/global/etopo5.HTML.
NOAA/NWS/ROC, 1991: NOAA Next Generation Radar (NEXRAD) Level 2 base data. NOAA/National Centers for Environmental Information, accessed May 2017 to May 2018, https://doi.org/10.7289/V5W9574V.
Ortega, K. L., J. M. Krause, and A. V. Ryzhkov, 2016: Polarimetric radar characteristics of melting hail. Part III: Validation of the algorithm for hail size discrimination. J. Appl. Meteor. Climatol., 55, 829–848, https://doi.org/10.1175/JAMC-D-15-0203.1.
Parker, M. D., 2014: Composite VORTEX2 supercell environments from near-storm soundings. Mon. Wea. Rev., 142, 508–529, https://doi.org/10.1175/MWR-D-13-00167.1.
Potvin, C. K., K. L. Elmore, and S. J. Weiss, 2010: Assessing the impacts of proximity sounding criteria on the climatology of significant tornado environments. Wea. Forecasting, 25, 921–930, https://doi.org/10.1175/2010WAF2222368.1.
Potvin, C. K., C. Broyles, P. S. Skinner, H. E. Brooks, and E. Rasmussen, 2019a: A Bayesian hierarchical modeling framework for correcting reporting bias in the U.S. tornado database. Wea. Forecasting, 34, 15–30, https://doi.org/10.1175/WAF-D-18-0137.1.
Potvin, C. K., and Coauthors, 2019b: Systematic comparison of convection-allowing models during the 2017 NOAA HWT spring forecasting experiment. Wea. Forecasting, 34, 1395–1416, https://doi.org/10.1175/WAF-D-19-0056.1.
Potvin, C. K., and Coauthors, 2020: Assessing systematic impacts of PBL schemes in the NOAA Warn-on-Forecast system. Mon. Wea. Rev., 148, 2567–2590, https://doi.org/10.1175/MWR-D-19-0389.1.
Rasmussen, E. N., and D. O. Blanchard, 1998: A baseline climatology of sounding-derived supercell and tornado forecast parameters. Wea. Forecasting, 13, 1148–1164, https://doi.org/10.1175/1520-0434(1998)013<1148:ABCOSD>2.0.CO;2.
Roberts, B., M. Xue, I. Dawson, and T. Daniel, 2020: The effect of surface drag strength on mesocyclone intensification and tornadogenesis in idealized supercell simulations. J. Atmos. Sci., 77, 1699–1721, https://doi.org/10.1175/JAS-D-19-0109.1.
Rotunno, R., and J. Klemp, 1985: On the rotation and propagation of simulated supercell thunderstorms. J. Atmos. Sci., 42, 271–292, https://doi.org/10.1175/1520-0469(1985)042<0271:OTRAPO>2.0.CO;2.
Rotunno, R., P. M. Markowski, and G. H. Bryan, 2017: “Near ground” vertical vorticity in supercell thunderstorm models. J. Atmos. Sci., 74, 1757–1766, https://doi.org/10.1175/JAS-D-16-0288.1.
Ryzhkov, A. V., and D. S. Zrnić, 2019: Radar Polarimetry for Weather Observations. 1st ed. Springer, 497 pp., https://doi.org/10.1007/978-3-030-05093-1.
Ryzhkov, A. V., T. J. Schuur, D. W. Burgess, P. L. Heinselman, S. E. Giangrande, and D. S. Zrnić, 2005: The Joint Polarization Experiment: Polarimetric rainfall measurements and hydrometeor classification. Bull. Amer. Meteor. Soc., 86, 809–824, https://doi.org/10.1175/BAMS-86-6-809.