1. Introduction
Typical polarimetric radar features of supercell storms are well known in the literature (Kumjian and Ryzhkov 2008a; Romine et al. 2008; Van Den Broeke et al. 2008; Kumjian et al. 2010; French et al. 2015; Tanamachi and Heinselman 2016; Van Den Broeke 2016; and many others). One recent focus has been to quantify these features (e.g., Kumjian 2011; Snyder et al. 2015; Van Den Broeke 2016) and apply them across the tornado life cycle (e.g., French et al. 2015; Houser et al. 2016; Van Den Broeke 2017). Fewer studies have quantified how tornadic and nontornadic supercell storms may differ polarimetrically (e.g., French et al. 2015). Kumjian and Ryzhkov (2007) examined a sample of nine tornadic and six nontornadic supercell storms and found generally similar and repeatable polarimetric signatures across the sample, though their primary purpose was not to examine differences between tornadic and nontornadic storms. Here the results of a study are reported that seek to separate nontornadic and pretornadic supercell storms using their polarimetric radar characteristics.
An overview of the tornado warning process including current challenges is provided by Brotzge and Donner (2013). One challenge in identifying potentially tornadic storms is that they occasionally occur in such close proximity to nontornadic storms that the observational network is unable to distinguish their environments (e.g., Klees et al. 2016). In this case, radar signatures may be critically important to determine the tornadic potential of individual storms. Severe weather predictability can be extremely sensitive to mesoscale details (e.g., Bluestein and Snyder 2015), and thus any attempt to use radar information for this purpose must be supplemented by available knowledge of the regional- and storm-scale environment. The reader is referred to Klees et al. (2016) for a review of the primary documented kinematic and nonpolarimetric radar differences between tornadic and nontornadic supercell storms.
The radar-observed hook echo or echo appendage (e.g., Stout and Huff 1953; Forbes 1981; Markowski 2002) has been associated with tornadic thunderstorms since the early days of radar meteorology but is not a reliable predictor of tornadic potential (e.g., Forbes 1981; Markowski et al. 2002). Nevertheless, properties of the drop size distribution within the appendage may be a predictor of whether a storm is tornadic. Tornadic appendages have been associated with greater areal predominance of small drops (Kumjian and Ryzhkov 2008b), and small drops may become more numerous in the appendage prior to tornadogenesis (French et al. 2015). Cao et al. (2008, hereafter C08) developed a curve to be placed on a scatterplot of radar reflectivity (ZHH) and differential reflectivity (ZDR) characterizing “typical” rainfall in Oklahoma, which can be used to determine whether median drop size is larger than expected for Oklahoma rainfall (points above the C08 curve) or smaller than expected (points below the C08 curve). Prior work has found that tornadic supercells are more likely than nontornadic supercells to have a larger proportion of appendage points lying below the C08 curve (French et al. 2015), indicating relatively small median drop size.
Environmental characteristics show some skill in determining whether supercells will be tornadic or nontornadic (e.g., Stensrud et al. 1997; Thompson 1998; Coffer and Parker 2017). Tornadic and nontornadic environments are somewhat distinguished by observations of environmental characteristics (e.g., Brooks et al. 1994; Rasmussen and Blanchard 1998; Rasmussen 2003; Thompson et al. 2003; Davies 2004; Esterheld and Giuliano 2008; Mercer et al. 2009; Parker 2014; Klees et al. 2016; and others). For example, prior work has shown that high values of boundary layer–6 km shear reasonably distinguish supercells from nonsupercells, but this quantity does not skillfully distinguish significantly tornadic supercells (those producing tornadoes rated EF2+) from weakly tornadic and nontornadic supercells (Rasmussen and Blanchard 1998; Thompson et al. 2003). Storm relative helicity (SRH) and low-level shear are shown to better distinguish tornadic from nontornadic supercells. The greater applicability of 0–1 and 0–3 km products [e.g., SRH, convective available potential energy (CAPE)] was discussed and quantified by Rasmussen (2003), and mixed-layer parameters were examined by Thompson et al. (2003). Many of these parameters may have operationally meaningful ability to distinguish tornadic from nontornadic environments (e.g., Klees et al. 2016) if the observational network takes samples at sufficient resolution. The high-resolution dataset of Parker (2014) indicates that tornadic supercell environments are distinguished by higher low-level mixing ratio values and a more strongly veering wind profile in the lowest 1 km.
Such environmental differences may translate to differences in supercell inflow, updraft, and microphysical characteristics, in turn leading to variable manifestations of supercell polarimetric features (e.g., French et al. 2015; Van Den Broeke 2016). Since these differences are closely tied to the environment, nontornadic supercells may exhibit different polarimetric characteristics than supercells, which will eventually produce a tornado. If this is the case, polarimetric information may be helpful to determine which storms are most likely to become tornadic in the near future. So far, however, little work has quantified polarimetric signatures and their differences within samples of pretornadic and nontornadic supercells to examine whether these supercell populations may be distinguishable. The goal of this paper is to test several hypothesized differences between nontornadic and pretornadic supercell storms in the 15–30 min prior to initial tornadogenesis, described below:
Hailfall may display greater temporal variability in tornadic supercells than in nontornadic supercells (e.g., Kumjian and Ryzhkov 2008a, 2009; Van Den Broeke 2016). Therefore, pretornadic supercells in this study are hypothesized to be distinguished by a hail signature that exhibits greater temporal variability in size. A higher proportion of analysis periods with relatively small hail extent is also hypothesized to result in smaller median storm-core ZHH values in pretornadic supercells compared to nontornadic supercells.
Nontornadic supercells often exhibit consistent and pronounced ZDR arc disruptions due to hailfall, which is more extensive spatially and temporally than in tornadic supercells (e.g., Kumjian and Ryzhkov 2009). Thus, for this study pretornadic storms are hypothesized to have steadier ZDR arc characteristics (width, size, and ZDR value within the arc) when a larger sample is examined due to less frequent hailfall disruptions.
Given a smaller contribution from melting hail (e.g., as hypothesized by Dawson et al. 2014) and consistent with the findings of Van Den Broeke (2017) for times when a supercell was producing a tornado, the ZDR arcs of pretornadic supercells are hypothesized to contain higher median ZDR values than are nontornadic supercells.
Past numerical studies have indicated a possible but likely relatively weak positive association between low-level vortex intensity and midlevel updraft size for mesocyclone-associated tornadoes (e.g., Trapp et al. 2017; Coffer and Markowski 2018). Radar-based evidence for a possible relationship between vortex intensity and midlevel updraft size, depth, and steadiness was presented by Van Den Broeke (2016, 2017). Thus, updrafts are hypothesized to be larger, deeper, and steadier in pretornadic supercells than in nontornadic supercells, though such associations may be weakened by the relatively poor vertical resolution of operational radar data.
Echo appendages of tornadic storms may on average contain smaller drops than nontornadic storms (e.g., Kumjian and Ryzhkov 2008b; French et al. 2015), though this varies by location within the appendage (French et al. 2015). If greater prevalence of melting hail in the appendages of nontornadic storms is responsible as speculated prior (Kumjian and Ryzhkov 2008b), it is here hypothesized that nontornadic appendages will be characterized by higher median ZHH and ZDR values, lower median ρhv, and a higher portion of bins falling above the C08 curve, consistent with larger drops and the presence of mixed-phase precipitation.
2. Data and methods
Samples of 32 pretornadic and 31 nontornadic supercell storms were compared in this study, though for individual radar metrics the number of contributing storms was lower as noted below. The sample of pretornadic storms is described in Table 1. The time of the first tornado report was determined for each of these supercells, and all radar scans beginning ≤30 min prior were defined as pretornado times. To be included in the analysis, the base-scan radar beam centerline throughout the polarimetric signature of interest was required to be ≤1 km above radar level (ARL; assuming beam propagation in a standard atmosphere). This resulted in five–eight sample volumes for each pretornadic supercell. These were further categorized as being 30, 25, 20, and/or 15 min prior to initial tornadogenesis, and separate analyses were conducted for each temporal length to examine whether there may be an optimal length of time prior to tornadogenesis when particular radar metrics are especially able to indicate future tornado potential.
The date, analysis period, representative radar, and unanalyzed metrics for the 32 pretornadic supercell storms included in this study. The unanalyzed metric numbers follow the metric descriptions in section 2 of this paper.
A sample of 31 completely separate nontornadic supercell storms (Table 2) was developed following a similar method, by searching for storms containing radar features described by Thompson et al. (2003) within 100 km of a polarimetric Weather Surveillance Radar-1988 Doppler (WSR-88D) and not connected to other convective cells with ZHH > 20 dBZ. Storms were selected that had base-scan data ≤1 km ARL for ≥20 min so reasonably robust time series of low-level radar metrics could be developed. Analysis periods ranged in length from 24 min to 1 h 45 min, with an average length of 56 min. As with the sample of pretornadic storms, events were chosen without regard to geographic region (Fig. 1) or time of year (Tables 1 and 2). Events were classified as nontornadic if the Storm Events Database (https://www.ncdc.noaa.gov/stormevents/) from the National Centers for Environmental Information (NCEI) indicated that no tornado was associated with the storm of interest at any time during its life. Though the Storm Events Database contains the most rigorously verified tornado dataset available, the absence of tornado reports does not guarantee that a particular storm did not produce a tornado, particularly a weak and/or short-lived tornado (e.g., Tanamachi et al. 2013). A comparison of warning statistics indicated that tornado and/or severe thunderstorm warnings were issued for 93.8% of nontornadic storms but only 81.3% of tornadic storms, indicating that the nontornadic storms were of sufficient intensity to warrant a warning. Tornado warnings were rarely issued for the nontornadic storms (12.5%). Though tornado warnings were often issued for tornadic storms (78.1%), there was still a >20% rate of unwarned tornadic storms, motivating such work as reported here to investigate whether polarimetric radar signatures could increase confidence in a tornado warning decision.
Approximate locations of the 31 nontornadic storms (blue dots) and 32 pretornadic storms (green dots) included in the analysis.
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
Each radar scan included in the analysis window for each pretornadic and nontornadic storm was analyzed by computing values of several metrics, described below. Metrics were calculated using GIS software (for metrics requiring a size calculation), image processing software (for metrics requiring a mean value calculation), and cursor output within radar display software (for ZDR column altitude). Excepting ZDR column size and altitude, all metrics were calculated using 1-km CAPPIs of radar variables derived by inverse distance weighting-squared interpolation, reducing potential error due to metrics being calculated at differing altitudes through time or between storms. Metrics calculated include:
Areal extent of large hail (km2; illustrated by Van Den Broeke 2016) was calculated as the area of the storm core over which ZHH > 55 dBZ was collocated with ZDR between −0.5 and +1.0 dB (e.g., Kumjian and Ryzhkov 2008a; Park et al. 2009). This metric could not be calculated for one pretornadic and eight nontornadic storms because there were not enough successive scans with base-scan hail area ≤1 km ARL. Additionally, in one pretornadic storm the storm-core ZHH was <55 dBZ, though a localized ZDR minimum with −0.5 < ZDR < 1 dB collocated with the highest ZHH values was strong evidence of hail so this case was retained.
Maximum storm-core ZHH (dBZ) was the average of the 10 highest ZHH bin values associated with the storm core, thought to represent precipitation intensity and hail likelihood. Note that Van Den Broeke (2017) used the single highest ZHH value, which is more prone to noise. Pearson’s correlation between results of the two methods was 0.995, with the single-pixel method yielding a value ~1.7 dBZ higher on average. An area-based metric may be optimal (e.g., the top 5% of bin values associated with storm area >35 dBZ) given beam spreading with distance and therefore poorer observation of localized maxima. For this dataset, however, maximum storm-core ZHH did not appear to be affected by distance from the radar (Pearson’s correlation = 0.068). This metric could not be calculated for 2 pretornadic and 7 nontornadic storms because there were not enough successive scans in which the storm core ZHH at base-scan was ≤1 km ARL.
ZDR arc size (km2) was defined as the portion of the ZDR arc (Kumjian and Ryzhkov 2008a) with values ≥3.5 dB (illustrated by Van Den Broeke 2016). This threshold well captured the core of the ZDR arc and its temporal changes across this dataset of storms; sensitivity to the threshold choice is described below. This metric was not calculated for 3 pretornadic and 9 nontornadic storms because there were not enough successive scans with the ZDR arc fully ≤1 km ARL or because data quality was poor in the storm inflow region.
ZDR arc mean value (dB; Van Den Broeke 2016) was calculated as the mean value of all ZDR arc bins with a value ≥3.5 dB. This contrasts with the method of Van Den Broeke (2016), who calculated this metric as the mean of all bins with a value ≥0 dB. This metric could not be calculated for 4 pretornadic and 9 nontornadic storms because there were not enough successive scans in which the ZDR arc was fully ≤1 km ARL or because data quality was poor in the storm inflow region.
ZDR arc width (km) was the width of the 2-dB arc measured perpendicular to the forward flank. Since ZDR arc width often varies downstream from the mesocyclone region, an average value was calculated. This metric could not be calculated for 3 pretornadic and 9 nontornadic storms because there were not enough successive scans in which the ZDR arc was fully ≤1 km ARL or because data quality was poor in the storm inflow region.
ZDR column size (km2), pictured by Van Den Broeke (2016), was calculated as the area of the 0.5-dB ZDR column at an altitude ~1 km above the environmental 0°C level. The 0°C level was determined from archived RAP model output. An altitude ~1 km above this level was required since ZDR columns are generally well-defined by that altitude. A ZDR threshold of 0.5 dB was chosen since ice-phase scatterers at this altitude typically have ZDR < 0.5 dB. The elevation angle providing data nearest the required altitude was selected for analysis. Selected scans were typically ±300 m from the environmental 0°C level, with a maximum allowable value of ±500 m. Features identified as ZDR columns were required to be vertically continuous with the high-ZDR region below and collocated with the storm’s high-ZHH core or with a weak echo region. This metric could not be calculated for 2 pretornadic storms because data quality was poor in the storm updraft region.
ZDR column depth (km) was calculated as the height at the top of the 1-dB ZDR column minus altitude of the environmental 0°C level (Van Den Broeke 2016). The 1-dB threshold is consistent with the algorithm of Snyder et al. (2015). Column identification again required vertical continuity and collocation with the storm’s high-ZHH core or with a weak echo region. Times were removed for which storms were too close to the radar for the ZDR column to be fully sampled (e.g., the top of the column fell within the cone of silence). Since there are often large differences in beam centerline altitude between successive elevation angles, error in estimates of this metric is potentially large, especially at large range. This metric could not be calculated for two pretornadic storms because data quality was poor in the storm updraft region.
Mean appendage ZHH was calculated using the method of French et al. (2015) to identify the echo appendage, and the mean ZHH value within that area was calculated subject to a ZHH threshold of 30 dBZ and ρhv threshold of 0.95 to exclude tornadic debris and dry hail. To be consistent with the appendages over which ZHH–ZDR plots were compared (metric 11 below), mean appendage ZHH was only calculated if the appendage ZDR was not influenced by nonmeteorological scatterers (e.g., tornadic debris) or other degrading features such as beam blockage. This led to the exclusion of 13 pretornadic and 13 nontornadic cases.
Mean appendage ρhv was calculated over the same area as defined for other echo appendage metrics, with a ZHH threshold of 30 dBZ and a mean appendage ρhv threshold of ≥0.95 again applied. Given data quality issues with ρhv, this metric could not be computed for 13 pretornadic and 13 nontornadic storms (see Van Den Broeke 2016 for a list of reasons why cases were removed).
Mean appendage ZDR (dB) was calculated using the same appendage area as for metrics 8 and 9. As with mean appendage ρhv, cases were excluded if tornadic debris or dry hail were indicated. In total, 13 pretornadic and 13 nontornadic cases were excluded.
Proportion of ZHH–ZDR points within the echo appendage falling above the C08 curve was calculated for the same appendage regions as in metrics 8–10. As in French et al. (2015), a ρhv lower threshold of 0.95 was used. As with mean appendage ZDR, 13 pretornadic and 13 nontornadic storms were excluded when this metric was calculated.
It is clear from the number of storms excluded for some of the metrics, especially those related to echo appendage characteristics, that this should be seen as a preliminary study with a need for additional work utilizing a larger number of storms. In addition, some of the resulting n values are as low as 18 individual storms. A larger number of storms will increase the statistical robustness of such comparisons.
Once all values for a given metric for a single storm were calculated, a median value of that metric was calculated for that single storm. Mean metric values were also calculated for each storm using the population of all values from that storm. The mean values are not shown in this paper, but the interpretation of results was identical whether the mean or median values were used. Thus, the median was selected as a more outlier-resistant measure. Areal extent variables were normalized by storm size [defined as storm area with ZHH > 35 dBZ, as by Van Den Broeke (2017)], though both normalized and nonnormalized measures were retained for the analysis. Also as in Van Den Broeke (2017), the coefficient of variation was used to quantify the variability of each radar metric through the analysis window for each storm. Given a set of values, the coefficient of variation is defined as the set’s standard deviation normalized by the set’s mean value, allowing comparison of the variability between storms with widely differing mean values of a particular metric. The coefficient of variation was only calculated for the 30 and 25 min prior to tornadogenesis, because for shorter analysis periods the standard deviation was often very high for strongly variable radar metrics and was not comparable with others, which were based on a larger sample size.
ZDR calibration was achieved for all storms following the scatterer-based method of Picca and Ryzhkov (2012) and applied by Van Den Broeke (2017)—the resulting calibration factor was applied to all radar metrics that require ZDR. For example, the 3.5-dB threshold used to define the ZDR arc core was adjusted upward or downward for each individual storm to account for that storm’s calibration factor. For the nontornadic storms, the mean calibration factor was −0.103 dB, comparable to the value of −0.07 dB for a sample largely overlapping the pretornadic storms in this study, reported by Van Den Broeke (2017). Among the nontornadic storms, the calibration factor exceeded 0.2 dB in 50% of events and 0.5 dB in 6% of events.
Wilcoxon–Mann–Whitney (WMW) statistics were utilized for the comparison of pretornadic and nontornadic storms, with a null hypothesis that radar metric median or coefficient of variation values from each set of storms could have come from the same distribution. This test was used to generate p values assuming that populations (e.g., of radar metric medians) may be non-Gaussian and that sample sizes may be small (e.g., Corder and Foreman 2014). By themselves p values are not authoritative evidence for or against a particular hypothesis, but they can increase confidence that a hypothesized physical mechanism is contributing (e.g., Wasserstein and Lazar 2016).
Environments of nontornadic supercells were characterized using the same variables as described by Van Den Broeke (2017) for the pretornadic storms (see Table 2 of that paper for a list of environmental variables included). In summary, the closest hourly Rapid Update Cycle (RUC) or Rapid Refresh (RAP) model initialization was selected relative to the center of the analysis period of each storm. Soundings were selected to be representative of the undisturbed far-field environment (e.g., as in Thompson et al. 2003, 2007). Error could have been introduced by the presence of undetected environmental heterogeneities (which were carefully screened for using satellite, surface, and radar data) and by the biases inherent to RUC and RAP (e.g., Benjamin et al. 2016). Storm-scale environmental variations, which may be critically important (e.g., Parker 2014), were also not able to be detected. Thus, soundings selected are generally representative of the synoptic-scale environment in which storms were embedded. Environments of tornadic and nontornadic storms were compared to get a sense for how well environmental characteristics distinguish these storm categories among this sample of storms and to ensure that the environments of these samples of storms are representative of ‘typical’ pretornadic and nontornadic environments.
3. Environments of supercells analyzed
This section describes differences between the environments of pretornadic and nontornadic supercell storms used in this study, to show that the environments of these storm samples are generally consistent with known differences between tornadic and nontornadic supercells. Accordingly, the radar signatures of these storms are thought to be a representative sample. Though a large set of environmental variables was examined [Table 2 of Van Den Broeke (2017)], only those with a WMW p value <0.01 are discussed here (see Table 3). The pretornadic and nontornadic storm datasets were generally distinguishable using environmental factors, as found in prior studies. Environmental differences such as those described here may influence storm structural and microphysical characteristics and therefore their radar signatures, so some of the radar differences noted in the following section may result from environmental differences, emphasizing the value of a future study examining radar signatures between storms in similar environments.
Environmental variables with an ability to separate pretornadic from nontornadic storms in this sample (WMW p value ≤ 0.01).
Tornadic environments are often distinguished by low mixed-layer lifting condensation level (MLLCL) height (e.g., Thompson et al. 2003), which was also the case for this sample of storms (Fig. 2a, Table 3)—mean MLLCL height was more than twice as high among the nontornadic storms. Pretornadic storms also tended to be associated with warmer environments, evidenced by their warmer MLLCL temperatures (Table 3; Fig. 2b). These results did not appear to be due to storms occurring at different times of year—median Julian day was similar [139.5 for pretornadic storms (mid-May) and 150.0 for nontornadic storms (late May)].
Comparison of environmental variables between the pretornadic and nontornadic storms: (a) MLLCL height (m), (b) MLLCL temperature (°C), (c) 0–1 km shear (m s−1), (d) 0–1 km storm relative helicity (m2 s−2), and (e) effective-layer storm relative helicity (m2 s−2). The bottom of each blue box is the first quartile, and the top of each blue box is the third quartile. The purple bar indicates the median value, and the light blue “+” indicates the mean value. Outlying horizontal bars are at the 9th and 91st percentiles, with outliers indicated as filled circles. Numbers below the x axis indicate the number of values used to construct each boxplot.
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
As expected given prior studies (e.g., Thompson et al. 2003; Parker 2014; Tochimoto and Niino 2016; Taszarek et al. 2017), wind variables in the lowest 3 km strongly separated pretornadic and nontornadic supercells. Vertical wind shear was much stronger in the 0–1 km (Fig. 2c) and 0–3 km (Table 3) layers among pretornadic storms. SRH estimated using Bunkers storm motion (Bunkers et al. 2000) was also significantly higher in the 0–1 km (Fig. 2d) and 0–3 km (Table 3) layers among pretornadic storms. For both wind shear and SRH variables, the near-surface layer (0–3 km) had the best ability to separate pretornadic from nontornadic storms. When SRH in an effective inflow layer was examined (e.g., Thompson et al. 2007), the same strong distinguishing ability existed (Fig. 2e, Table 3). The significant tornado parameter (STP; Rasmussen and Blanchard 1998; Parker 2014), a composite parameter incorporating wind and instability information, was 3.5 times larger among pretornadic storms (Table 3).
In summary, environmental variables well-distinguished these samples of pretornadic and nontornadic storms, and the variables allowing separation are as expected from prior studies. This indicates that the sample’s radar signatures are also likely representative.
4. Differences between pretornadic and nontornadic supercell storms
a. Storm size, hail areal extent, and storm-core ZHH values
It was hypothesized that pretornadic supercells would be characterized by a more variable hail signature and therefore a smaller mean hail areal extent and lower storm-core ZHH values. Hail extent was normalized by storm size (km2, defined as area of ZHH > 35 dBZ). Among this sample of storms, storm size was ~5.6% larger for nontornadic storms, but statistically similar (Table 4; Fig. 3a). Variability of storm size was also statistically similar between pretornadic and nontornadic storms (Table 5).
Left two columns of values indicate the median (MED) values of polarimetric radar metrics compared between samples of nontornadic (NT) and pretornadic (PT) supercell storms in the 30 min prior to initial tornadogenesis. Following columns indicate the WMW p values for a comparison of the distributions of medians when pretornadogenesis analysis periods of 30, 25, 20, and 15 min are used for the PT storms. “N” next to the variable name indicates normalization by storm size. WMW p values < 0.01 are boldface and italic and WMW p values 0.05 > p > 0.01 are boldface only. Units are indicated in brackets next to the name of each variable; “[ ]” indicates a unitless quantity.
Violin plots showing the distribution of (a) storm size (km2), (b) maximum storm-core ZHH values (dBZ), (c) raw hail areal extent (km2), and (d) normalized hail areal extent, all at an altitude of 1 km ARL. Left column of each pair is for pretornadic storms (PT), and right column is for nontornadic storms (NT). PT values are valid for the 30-min period prior to initial tornadogenesis. Number of storms contributing to each violin indicated within each violin, and raw values indicated as points. In each violin, the black line is the mean and the yellow line is the median. The p value for each comparison is included in each panel.
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
As in Table 4, but for coefficient of variation (CoV) in and 30 and 25 min prior to tornadogenesis.
Storm-core maximum ZHH, calculated as the mean of the 10 highest-valued bins, can be an indicator of hail size and/or the presence of melting hail (e.g., Ryzhkov et al. 2010). Given the potential importance of large and melting hail to low-level storm thermodynamics and potential differences in hail characteristics of tornadic and nontornadic storms observed prior (e.g., Kumjian and Ryzhkov 2008a, 2009), it is hypothesized here that this variable may be indicative of whether a storm is pretornadic or will remain nontornadic. Polarimetric measurements such as hail differential reflectivity (Depue et al. 2007) may also be insightful but were not investigated here. Over all 440 pretornadic and nontornadic analysis periods, the 10th percentile was 58.4 dBZ with a mean value of 65.3 dBZ—while ZHH < 55 dBZ can indicate hail, this is usually assumed to indicate relatively small wet hail (e.g., Straka et al. 2000). Median storm-core maximum ZHH was 2.4 dBZ higher in the nontornadic storms with a pretornado lead time of 30 min (Table 4; Fig. 3b), a statistically significant result (p = 0.023–0.038 at lead times of 15–30 min; Table 4). Figure 4 shows a representative distribution of ZHH for a pretornadic (Fig. 4a) and a nontornadic (Fig. 4b) storm. Variability of this metric did not distinguish pretornadic and nontornadic storms (Table 5). The degree to which this variable provides similar information to hail areal extent was briefly investigated. Though there is a positive relationship between storm-core maximum ZHH and hail areal extent, Pearson’s correlation between the two variables indicates that they may contain separate, useful information (p = 0.596 [pretornadic] and p = 0.453 [nontornadic]).
1-km comparison of pretornadic and nontornadic supercell storms. (a),(c) Tulsa, Oklahoma, radar (KINX; at 2039 UTC 20 May 2013), and range to storm core is approximately 63 km. (b),(d) Bismarck, North Dakota, radar (KBIS; at 0553 UTC 4 Sep 2014); range to storm core is approximately 66 km. White outlines represent areas of polarimetrically inferred hailfall; ZDR calibration factors of −0.05 dB (KINX) and −0.62 dB (KBIS) have been applied to the data shown here.
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
The size of the inferred polarimetric hail signature was compared between pretornadic and nontornadic supercells; for consistency with past studies, both raw values and values normalized by storm size were compared (Table 4; Figs. 3c,d). Both had about the same ability to distinguish whether a storm was pretornadic or nontornadic (p = 0.014–0.020 for normalized hail signature size and p = 0.014–0.022 for raw hail signature size for 15–30 min prior to tornadogenesis). Hail extent tended to be most diagnostic in the 15–20 min prior to initial tornadogenesis (Table 4). Though variability of hail extent was larger in pretornadic storms, this was not statistically significant (Table 5). These findings support the hypothesis that nontornadic supercells contain hail over a larger relative area but do not statistically support that hailfall is less variable in nontornadic storms. Figure 4 shows a typical example of hailfall in pretornadic (Figs. 4a,c) and nontornadic (Figs. 4b,d) supercells—the nontornadic storm has a large region with hail (indicated by the white outline in Figs. 4b and 4d), while the pretornadic storm has a much smaller hail region (Figs. 4a,c). Hail areal extent never exceeded 112 km2 (and rarely 80 km2) among the pretornadic storms (Fig. 3c), and hail extent this large among nontornadic storms was often accompanied by a storm-core ZHH maximum value >62 dBZ (not shown).
b. ZDR arc characteristics
ZDR arcs are indicative of hydrometeor size sorting in storm-relative inflow winds along the forward flank. In this paper they are hypothesized to be less variable in width and mean value in pretornadic supercells than in nontornadic supercells due to less consistent hailfall disruptions. ZDR within the arcs of pretornadic storms is also hypothesized to be higher given a possible smaller contribution from melting hail. In prior studies ZDR arc characteristics have been related to tornado life cycles (e.g., Palmer et al. 2011; Crowe et al. 2012), and some preliminary differences have been noted between samples of tornadic and nontornadic storms (Van Den Broeke 2016). The magnitude of ZDR values within the arc has also been related to the magnitude of low-level shear and SRH (e.g., Kumjian and Ryzhkov 2009; Dawson et al. 2015). Among this sample of storms, mean ZDR value within the arc was weakly correlated to 0–1 and 0–3 km shear (r = 0.169 and r = 0.267, respectively) and weakly correlated to 0–1 km SRH and 0–3 km SRH (r = 0.216 and r = 0.268, respectively), generally in support of these prior findings.
To test the appropriateness of the 3.5-dB ZDR threshold used to define the ZDR arc, varying ZDR thresholds were tested on a random sample of 7 pretornadic and 8 nontornadic storms from this dataset. When thresholds of 1.5, 2, 2.5, 3, 3.5, and 4 dB were selected, the typical pattern was for low thresholds to produce very large areal extents that include much of the supercell area. Some intermediate thresholds produce roughly parallel areas, while the highest threshold (4 dB) occasionally produced no area. Pearson’s correlation was calculated between arc area from successive ZDR thresholds, and comparisons yielding correlation ≥0.70 were interpreted as indicating that a similar feature (the true ZDR arc) was being identified. The minimum and maximum threshold values associated with a correlation of at least 0.7 were recorded for each storm. Across all storms, the average minimum threshold value that well characterized the ZDR arc was 2.03 dB, and the average maximum threshold value was 3.47 dB. These closely correspond to the prior-chosen subjective values (e.g., Van Den Broeke 2017) of 2 dB for ZDR arc width and 3.5 dB for ZDR arc area. Thus, these two threshold values appear to reasonably well characterize the ZDR arcs of the storms examined.
Variability of ZDR arc metrics was generally statistically similar between pretornadic and nontornadic storms, except for ZDR arc mean bin value which was more variable in nontornadic storms (p = 0.029–0.034 at lead times of 25–30 min; Table 5). Not all ZDR arc metrics were more variable in nontornadic storms (Table 5), lending little support to the hypothesis that the arc is more consistently disrupted by hail in nontornadic storms. The median value of ZDR within the ZDR arc was effectively the same between pretornadic and nontornadic storms (Table 4; Fig. 5), not supporting the hypothesis that ZDR values within the arc should be larger in pretornadic storms. These findings generally indicate that pretornadic and nontornadic storms may not be separable using basic ZDR arc characteristics. Though not statistically significant, ZDR arcs were often larger in pretornadic storms (by an average of ~48%; Table 4; Figs. 5a,b), an observation that may warrant further investigation.
As in Fig. 3, but for (a) raw ZDR arc size (km2), (b) normalized ZDR arc size, (c) ZDR arc width (km), and (d) mean ZDR value within the ZDR arc (dB).
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
c. ZDR column characteristics
ZDR columns are representative of the updraft region in deep convective storms including supercells (e.g., Kumjian and Ryzhkov 2008a; Kumjian et al. 2010, 2014; Houser et al. 2015; Van Den Broeke 2016; and others), and updraft strength changes may be inferred polarimetrically by utilizing characteristics of this signature. Given prior observations (e.g., Van Den Broeke 2016, 2017), it is hypothesized that updrafts may be larger, deeper, and steadier in pretornadic supercells. In contrast to ZDR arc characteristics, ZDR columns did distinguish pretornadic and nontornadic supercells in this sample of storms.
The size of the 0.5-dB ZDR column (corrected for calibration offset) approximately 1 km above the ambient 0°C level was ~55% larger in pretornadic storms (using data from the 30 min prior to tornadogenesis) when the value was normalized by storm size (p = 0.001–0.002 at lead times of 15–30 min; Table 4; Fig. 6b). Figure 7 shows an example of typical ZDR column appearance in pretornadic (Figs. 7a,c) and nontornadic (Figs. 7b,d) supercells (white outline indicates ZDR column location). At 1 km above the ambient 0°C level, ZDR columns of pretornadic storms were more likely to be nearly the size of the central high-ZHH region (as in Figs. 7a,c), while in nontornadic storms the ZDR column was often more isolated to small portions of the high-ZHH region (as in Figs. 7b,d). This observation may be related to possible masking of the column by the large quantity of hail in many nontornadic storms. Supporting this possibility, a correlation value of −0.356 was found between storm-mean normalized hail extent and normalized ZDR column area using all storms in this dataset (larger hail extent generally corresponded to a smaller ZDR column). Median pretornadic ZDR column depth was also ~0.27 km greater over the 30 min prior to tornadogenesis (Table 4; Fig. 6c), though this difference was not significant. Measures of ZDR column size and depth variability statistically separated pretornadic and nontornadic storms and were larger among nontornadic storms (Table 5). ZDR column size normalized by storm area was ~58% more variable and ZDR column depth ~42% more variable among nontornadic storms when compared with the 30 min prior to tornadogenesis. These differences did not appear to be a function of environmental CAPE, which was statistically similar between the two populations of storms (p = 0.39). In aggregate, these findings support the hypothesis that updrafts are steadier, deeper, and spatially larger in pretornadic storms.
As in Fig. 3, but for (a) ZDR column size at 1 km above the ambient 0°C level (km2), (b) normalized ZDR column size, and (c) ZDR column depth (km).
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
Comparison of pretornadic and nontornadic supercell storms at an elevation ~1 km above the ambient 0°C level. (a),(c) Tulsa, Oklahoma, radar (KINX; at 2039 UTC 20 May 2013); elevation angle is 4.00°, and range to storm core is approximately 62 km (altitude of beam center ~4.55 km). (b),(d) are from the Bismarck, North Dakota, radar (KBIS; at 0553 UTC 4 Sep 2014); elevation angle is 5.10°, and range to storm core is approximately 53 km (altitude of beam center ~4.91 km). White outlines represent ZDR columns. ZDR calibration factors of +0.404 dB (KINX) and −0.62 dB (KBIS) have been applied to the data shown here.
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
d. Echo appendage characteristics
In this paper it is hypothesized that echo appendages of pretornadic storms contain less melting hail, which may be detectable as smaller mean ZHH and ZDR, larger mean ρhv, and fewer bins falling above the C08 curve. Among this sample of storms, median ZHH was ~2.3 dBZ larger in the nontornadic appendages compared to the 30 min prior to tornadogenesis (Fig. 8a; Table 4), which often appeared to be related to increased prevalence of small hail in appendages of nontornadic storms, similar to a hypothesis of Kumjian and Ryzhkov (2008b). This pattern was not, however, consistent across the set of nontornadic storms. Note that the ρhv threshold of 0.95 used to filter appendage bins for the purpose of averaging would generally remove dry hail, but might not exclude all bins with wet/melting hail. Median appendage ρhv was similar between pretornadic and nontornadic storms (Fig. 8b; Table 4), not as hypothesized. Median ZDR was 0.54 dB higher in pretornadic storms (Fig. 8c), indicating larger drops. This result was significant (Table 4; p = 0.010–0.042 at 20–30 min). Future work is needed, as this finding contrasts with French et al. (2015) and Kumjian and Ryzhkov (2008b), who found evidence for smaller drops in the appendages of tornadic storms. The difference between pretornadic and nontornadic storms was not significant for a lead time of 15 min (Table 4). As expected given the larger ZDR values in pretornadic storms, the percentage of appendage ZHH–ZDR points above the C08 curve (indicating a drop distribution with large median diameter) in these storms was also ~14% larger in the 30 min prior to tornadogenesis (Fig. 8d; Table 4; p = 0.031–0.032 at leads times of 25–30 min), not supporting the hypothesis of larger mean drops in nontornadic storms. A relatively large percentage of bins above the C08 curve in both supercell categories indicate appendages dominated by relatively large drops, consistent with the findings of Kumjian and Ryzhkov (2008b).
As in Fig. 3, but for (a) mean appendage ZHH (dBZ), (b) mean appendage ρhv, (c) mean appendage ZDR (dB), and (d) percentage of points falling above the C08 curve (e.g., 0.5 = 50%).
Citation: Monthly Weather Review 148, 4; 10.1175/MWR-D-19-0296.1
In aggregate and for this sample of storms, these results provide evidence against the hypothesis that nontornadic appendages contain larger median drops. Given these and prior findings (e.g., French et al. 2015; Kumjian and Ryzhkov 2008b), additional research seems warranted to construct a climatology of appendage characteristics in supercell storms. Additional understanding of microphysics in this storm region (e.g., Kumjian 2011; Kumjian et al. 2015) is also needed.
5. Summary and discussion
Using a population of 19–30 pretornadic and 18–31 nontornadic supercells, several radar-derived metrics (Table 4) and their variability (Table 5) appear to separate supercells that are pretornadic from those that remain nontornadic. Some radar metrics did not show significant differences between pretornadic and nontornadic storms. It is important to note that such metrics should not be discarded from future supercell tornadogenesis studies since this study did not examine trends leading up to tornadogenesis. Even though temporal changes are somewhat conveyed by the analysis of different pretornadogenesis temporal windows in this study, a more detailed examination of trends would be needed. For instance, there may be value in examining whether the magnitude of the metrics is increasing or decreasing, how large these changes are through time, and if these changes are accelerating. Such a study could also be insightful if combined with a comparison of “pretornadogenesis failure” periods in nontornadic supercell storms. The relatively small number of storms included for some of the metrics, especially those characterizing the echo appendage, also points to the need for future work, which examines much larger populations of storms in a climatological sense.
As hypothesized, pretornadic storms appear to have a smaller area of polarimetrically inferred hailfall, lower storm-core ZHH values, and ZDR columns that are larger, deeper, and steadier. ZDR arc characteristics did not distinguish pretornadic and nontornadic storms, contrary to hypotheses. Variability of radar characteristics was generally slightly larger for pretornadic storms, though pretornadic storms exhibited less variability of ZDR column size and depth. Updrafts in aggregate appeared to be larger, deeper, and steadier among the pretornadic storms—this was the case even though measures of instability (e.g., CAPE) were statistically similar between the pretornadic and nontornadic storms, though measures of wind shear were larger in pretornadic storm environments (Table 3). This suggests that more updraft ascent is dynamically driven among the pretornadic storms in this sample. Nontornadic storms were characterized by smaller/shallower ZDR columns, larger hail areal extent, and larger storm-core ZHH values—it is possible that the large hail content in nontornadic storms partially obscures their ZDR columns. The difference in hail extent between pretornadic and nontornadic storms may be partially a function of the lower mean environmental 0°C level among the nontornadic storms, allowing more hail mass to survive to lower altitude. Pearson’s correlation between normalized hail areal extent and environmental 0°C altitude for this dataset was −0.43 (r2 = 0.185).
ZDR arc characteristics did not well-distinguish pretornadic and nontornadic storms in this sample. Arcs were larger among pretornadic storms, which while not statistically significant may warrant additional investigation using a much larger sample of storms. Orientation of the ZDR arc, while not investigated in this study, could be a beneficial focus of future research given the promising findings of prior studies (e.g., Palmer et al. 2011), as could separation of the ZDR–KDP maxima (e.g., Crowe et al. 2012; Loeffler and Kumjian 2018).
After additional research, several preliminary results presented in this paper may eventually be operationally useful to identify pretornadic storms, though application may be challenging because of radar data quality issues (e.g., ZHH and ZDR calibration). Some of the results noted here may eventually lead to guidance suggesting which storm(s) to focus on for future tornadic potential, given situations with numerous storms in the domain of a single radar. Less value may be realized when a single supercell is present, since in that situation it is not possible to compare metric values between storms. Such methods may also lead more to retrospective results than to prognostic value (as described by Kumjian and Ryzhkov 2008a). Future work examining tornadic and nontornadic supercells in close proximity, which may lead to clearer operational guidance, is warranted. The most promising topics for additional research suggested by the results in this paper include:
Smaller hail areal extent in pretornadic storms. It will be necessary to test whether this finding remains valid when tornadic and nontornadic storms are concurrently present. Such methods need to be carefully designed to ensure that natural variability is not having too large of an impact—e.g., by basing the radar time series on an analysis period of sufficient length, optimally over at least 20 min.
Higher storm-core maximum ZHH in nontornadic storms. This could be particularly valuable in the domain of a single radar where calibration does not affect the results. As with hail extent, this result needs to be tested in situations when tornadic and nontornadic storms are present. Future work should consider using a percentile approach to define which bins are averaged, and should examine hail distributions within the appendages of pretornadic and nontornadic storms.
Substantially larger and steadier ZDR columns among pretornadic storms. Though increased depth in pretornadic storms was not statistically significant, it may be worthy of additional investigation alongside other ZDR column metrics.
Diagnosing whether a storm is more likely to become tornadic or remain nontornadic using the metrics described herein relies on careful assessment of data quality and storm behavior. For many of the metrics it is critical to determine a ZDR calibration offset that is applied across the data, though insight is also possible if proximate storms can be compared (e.g., a similar calibration offset applies to all storms). Other data quality issues, such as three-body scatter downradial of hail, also need to be filtered out to avoid erroneous values in time series of some of the radar metrics. Finally, temporal variability of many of the radar metrics is large enough that median values should be determined over a minimum of ~20 min. Thus, while data from one or a few scans may be used to indicate potential that a storm is pretornadic, confidence in this diagnosis increases with the length of time over which a storm is analyzed. The results here indicate that hail variables may be useful in the 15–20 min prior to tornadogenesis, while ZDR column and echo appendage characteristics may be most useful ~25 min prior to tornadogenesis. With additional work focused on volumes containing both tornadic and nontornadic storms, the results presented here may form a basis for additional operational guidance regarding the use of polarimetric radar metrics to assess the tornadic potential of particular supercell storms.
Acknowledgments
The author was supported by NSF Grant IIA-1539070, NOAA Grant NA18OAR4590307, and an academic appointment at the University of Nebraska–Lincoln. Nick Humrich and Matt Wilson are acknowledged for code contributions. The author acknowledges the reviews of three anonymous peer reviewers who substantially improved the focus and quality of early drafts.
REFERENCES
Benjamin, S. G., and Coauthors, 2016: A North American hourly assimilation and model forecast cycle: The Rapid Refresh. Mon. Wea. Rev., 144, 1669–1694, https://doi.org/10.1175/MWR-D-15-0242.1.
Bluestein, H. B., and J. C. Snyder, 2015: An observational study of the effects of dry air produced in dissipating convective storms on the predictability of severe weather. Wea. Forecasting, 30, 79–114, https://doi.org/10.1175/WAF-D-14-00065.1.
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.
Brotzge, J., and W. Donner, 2013: The tornado warning process: A review of current research, challenges, and opportunities. Bull. Amer. Meteor. Soc., 94, 1715–1733, https://doi.org/10.1175/BAMS-D-12-00147.1.
Bunkers, M. J., B. A. Klimowski, J. W. Zeitler, R. L. Thompson, and M. L. Weisman, 2000: Predicting supercell motion using a new hodograph technique. Wea. Forecasting, 15, 61–79, https://doi.org/10.1175/1520-0434(2000)015<0061:PSMUAN>2.0.CO;2.
Cao, Q., G. Zhang, E. Brandes, T. Schuur, A. Ryzhkov, and K. Ikeda, 2008: Analysis of video disdrometer and polarimetric radar data to characterize rain microphysics in Oklahoma. J. Appl. Meteor. Climatol., 47, 2238–2255, https://doi.org/10.1175/2008JAMC1732.1.
Coffer, B. E., and M. D. Parker, 2017: Simulated supercell 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 P. M. Markowski, 2018: Comments on “The regulation of tornado intensity by updraft width.” J. Atmos. Sci., 75, 4049–4056, https://doi.org/10.1175/JAS-D-18-0170.1.
Corder, G. W., and D. I. Foreman, 2014: Nonparametric Statistics: A Step-by-Step Approach. 2nd ed. Wiley, 288 pp.
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.
Davies, J. M., 2004: Estimations of CIN and LFC associated with tornadic and nontornadic supercells. Wea. Forecasting, 19, 714–726, https://doi.org/10.1175/1520-0434(2004)019<0714:EOCALA>2.0.CO;2.
Dawson, D. T., 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.
Dawson, D. T., E. R. Mansell, and M. R. Kumjian, 2015: Does wind shear cause hydrometeor size sorting? J. Atmos. Sci., 72, 340–348, https://doi.org/10.1175/JAS-D-14-0084.1.
Depue, T. K., P. C. Kennedy, and S. A. Rutledge, 2007: Performance of the hail differential reflectivity (HDR) polarimetric radar hail indicator. J. Appl. Meteor. Climatol., 46, 1290–1301, https://doi.org/10.1175/JAM2529.1.
Esterheld, J. M., and D. J. Giuliano, 2008: Discriminating between tornadic and non-tornadic supercells: A new hodograph technique. Electron. J. Severe Storms Meteor., 3 (2), http://www.ejssm.org/ojs/index.php/ejssm/article/viewArticle/33.
Forbes, G. S., 1981: On the reliability of hook echoes as tornado indicators. Mon. Wea. Rev., 109, 1457–1466, https://doi.org/10.1175/1520-0493(1981)109<1457:OTROHE>2.0.CO;2.
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.
Houser, J. L., H. B. Bluestein, and J. C. Snyder, 2015: Rapid-scan, polarimetric, Doppler radar observations of tornadogenesis and tornado dissipation in a tornadic supercell: The “El Reno, Oklahoma” storm of 24 May 2011. Mon. Wea. Rev., 143, 2685–2710, https://doi.org/10.1175/MWR-D-14-00253.1.
Houser, J. L., H. B. Bluestein, and J. C. Snyder, 2016: A finescale radar examination of the tornadic debris signature and weak-echo reflectivity band associated with a large, violent tornado. Mon. Wea. Rev., 144, 4101–4130, https://doi.org/10.1175/MWR-D-15-0408.1.
Klees, A. M., Y. Richardson, P. Markowski, C. Weiss, J. Wurman, and 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.
Kumjian, M. R., 2011: Precipitation properties of supercell hook echoes. Electron. J. Severe Storms Meteor., 6 (5), http://www.ejssm.org/ojs/index.php/ejssm/article/view/93/65.
Kumjian, M. R., and A. V. Ryzhkov, 2007: Polarimetric characteristics of tornadic and nontornadic supercell thunderstorms. 33rd Conf. on Radar Meteorology, Cairns, Queensland, Australia, Amer. Meteor. Soc., P10.1, https://ams.confex.com/ams/33Radar/techprogram/paper_122882.htm.
Kumjian, M. R., and A. V. Ryzhkov, 2008a: 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, 2008b: 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, 2009: Storm-relative helicity revealed from polarimetric radar measurements. J. Atmos. Sci., 66, 667–685, https://doi.org/10.1175/2008JAS2815.1.
Kumjian, M. R., A. V. Ryzhkov, V. M. Melnikov, and T. J. Schuur, 2010: Rapid-scan super-resolution observations of a cyclic supercell with a dual-polarization WSR-88D. Mon. Wea. Rev., 138, 3762–3786, https://doi.org/10.1175/2010MWR3322.1.
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.
Kumjian, M. R., Z. J. Lebo, and H. C. Morrison, 2015: On the mechanisms of rain formation in an idealized supercell storm. Mon. Wea. Rev., 143, 2754–2773, https://doi.org/10.1175/MWR-D-14-00402.1.
Loeffler, S. D., and M. R. Kumjian, 2018: Quantifying the separation of enhanced ZDR and KDP regions in nonsupercell tornadic storms. Wea. Forecasting, 33, 1143–1157, https://doi.org/10.1175/WAF-D-18-0011.1.
Markowski, P. M., 2002: Hook echoes and rear-flank downdrafts: A review. Mon. Wea. Rev., 130, 852–876, https://doi.org/10.1175/1520-0493(2002)130<0852:HEARFD>2.0.CO;2.
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.
Mercer, A. E., C. M. Shafer, C. A. Doswell III, L. M. Leslie, and M. B. Richman, 2009: Objective classification of tornadic and nontornadic severe weather outbreaks. Mon. Wea. Rev., 137, 4355–4368, https://doi.org/10.1175/2009MWR2897.1.
Palmer, R. D., and Coauthors, 2011: Observations of the 10 May 2010 tornado outbreak using OU-PRIME: Potential for new science with high-resolution polarimetric radar. Bull. Amer. Meteor. Soc., 92, 871–891, https://doi.org/10.1175/2011BAMS3125.1.
Park, H., A. V. Ryzhkov, D. S. Zrnić, and K.-E. Kim, 2009: The hydrometeor classification algorithm for the polarimetric WSR-88D: Description and application to an MCS. Wea. Forecasting, 24, 730–748, https://doi.org/10.1175/2008WAF2222205.1.
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.
Picca, J., and A. Ryzhkov, 2012: A dual-wavelength polarimetric analysis of the 16 May 2010 Oklahoma City extreme hailstorm. Mon. Wea. Rev., 140, 1385–1403, https://doi.org/10.1175/MWR-D-11-00112.1.
Rasmussen, E. N., 2003: Refined supercell and tornado forecast parameters. Wea. Forecasting, 18, 530–535, https://doi.org/10.1175/1520-0434(2003)18<530:RSATFP>2.0.CO;2.
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.
Romine, G. S., D. W. Burgess, and R. B. Wilhelmson, 2008: A dual-polarization-radar-based assessment of the 8 May 2003 Oklahoma City area tornadic supercell. Mon. Wea. Rev., 136, 2849–2870, https://doi.org/10.1175/2008MWR2330.1.
Ryzhkov, A., D. Zrnić, J. Krause, M. Kumjian, and S. Ganson, 2010: Discrimination between large and small hail, final report. NOAA/NSSL Rep., 18 pp., https://www.nssl.noaa.gov/publications/wsr88d_reports/FINAL_HailSize.doc.
Snyder, J. C., A. V. Ryzhkov, M. R. Kumjian, A. P. Khain, and J. Picca, 2015: A ZDR column detection algorithm to examine convective storm updrafts. Wea. Forecasting, 30, 1819–1844, https://doi.org/10.1175/WAF-D-15-0068.1.
Stensrud, D. J., J. V. Cortinas Jr., and H. E. Brooks, 1997: Discriminating between tornadic and nontornadic thunderstorms using mesoscale model output. Wea. Forecasting, 12, 613–632, https://doi.org/10.1175/1520-0434(1997)012<0613:DBTANT>2.0.CO;2.
Stout, G. E., and F. A. Huff, 1953: Radar records Illinois tornadogenesis. Bull. Amer. Meteor. Soc., 34, 281–284.
Straka, J. M., D. S. Zrnić, and A. V. Ryzhkov, 2000: Bulk hydrometeor classification and quantification using polarimetric radar data: Synthesis of relations. J. Appl. Meteor., 39, 1341–1372, https://doi.org/10.1175/1520-0450(2000)039<1341:BHCAQU>2.0.CO;2.
Tanamachi, R. L., and P. L. Heinselman, 2016: Rapid-scan, polarimetric observations of central Oklahoma severe storms on 31 May 2013. Wea. Forecasting, 31, 19–42, https://doi.org/10.1175/WAF-D-15-0111.1.
Tanamachi, R. L., H. B. Bluestein, M. Xue, W.-C. Lee, K. A. Orzel, S. J. Frasier, and R. M. Wakimoto, 2013: Near-surface vortex structure in a tornado and in a sub-tornado-strength convective-storm vortex observed by a mobile, W-band radar during VORTEX2. Mon. Wea. Rev., 141, 3661–3690, https://doi.org/10.1175/MWR-D-12-00331.1.
Taszarek, M., H. E. Brooks, and B. Czernecki, 2017: Sounding-derived parameters associated with convective hazards in Europe. Mon. Wea. Rev., 145, 1511–1528, https://doi.org/10.1175/MWR-D-16-0384.1.
Thompson, R. L., 1998: Eta model storm-relative winds associated with tornadic and nontornadic supercells. Wea. Forecasting, 13, 125–137, https://doi.org/10.1175/1520-0434(1998)013<0125:EMSRWA>2.0.CO;2.
Thompson, R. L., R. Edwards, J. A. Hart, K. L. Elmore, and P. Markowski, 2003: Close proximity soundings within supercell environments obtained from the Rapid Update Cycle. Wea. Forecasting, 18, 1243–1261, https://doi.org/10.1175/1520-0434(2003)018<1243:CPSWSE>2.0.CO;2.
Thompson, R. L., C. M. Mead, and R. Edwards, 2007: Effective storm-relative helicity and bulk shear in supercell thunderstorm environments. Wea. Forecasting, 22, 102–115, https://doi.org/10.1175/WAF969.1.
Tochimoto, E., and H. Niino, 2016: Structural and environmental characteristics of extratropical cyclones that cause tornado outbreaks in the warm sector: A composite study. Mon. Wea. Rev., 144, 945–969, https://doi.org/10.1175/MWR-D-15-0015.1.
Trapp, R. J., G. R. Marion, and S. W. Nesbitt, 2017: The regulation of tornado intensity by updraft width. J. Atmos. Sci., 74, 4199–4211, https://doi.org/10.1175/JAS-D-16-0331.1.
Van Den Broeke, M. S., 2016: Polarimetric variability of classic supercell storms as a function of environment. J. Appl. Meteor. Climatol., 55, 1907–1925, https://doi.org/10.1175/JAMC-D-15-0346.1.
Van Den Broeke, M. S., 2017: Polarimetric radar metrics related to tornado life cycles and intensity in supercell storms. Mon. Wea. Rev., 145, 3671–3686, https://doi.org/10.1175/MWR-D-16-0453.1.
Van Den Broeke, M. S., J. M. Straka, and E. N. Rasmussen, 2008: Polarimetric radar observations at low levels during tornado life cycles in a small sample of classic Southern Plains supercells. J. Appl. Meteor. Climatol., 47, 1232–1247, https://doi.org/10.1175/2007JAMC1714.1.
Wasserstein, R. L., and N. A. Lazar, 2016: The ASA’s statement on p-values: Context, process, and purpose. Amer. Stat., 70, 129–133, https://doi.org/10.1080/00031305.2016.1154108.