Introduction

Dual-polarization radar has been recognized as being an efficient tool for the classification of different hydrometeor types and discrimination between meteorological and nonmeteorological scatterers (Zrnic and Ryzhkov 1999; Vivekanandan et al. 1999). It is natural to assume that tornadic debris is composed of more or less randomly oriented particles with very irregular shapes and a refractive index that is different from that of hydrometeors, thereby producing much different signatures than hydrometeors. Randomly oriented scatterers are characterized by differential reflectivity Z_DR that is equal to zero. If large debris scatterers are not chaotically oriented and possess some degree of common orientation, then their Z_DR might be either positive or negative, depending on their size and the mean canting angle.

The linear depolarization ratio (LDR) and cross-correlation coefficient ρ_hv of tornadic debris should also be quite different from signatures that are associated with hydrometeors. Similar to other nonmeteorological scatterers, such as natural ground cover (trees, grass, etc.), biological scatterers (insects, birds, and bats), and military chaff, tornadic debris is expected to have a significantly higher LDR and lower ρ_hv than that which is typical for liquid or frozen hydrometeors. This is because debris particles have large sizes, very irregular nonspherical shapes, a high refractive index, and a low degree of common alignment.

A polarimetric tornadic debris signature was first reported by Ryzhkov et al. (2002a), based on the measurements with the National Severe Storms Laboratory (NSSL) Cimarron polarimetric radar in one of the storms that constituted the famous central Oklahoma tornado outbreak on 3 May 1999. After tornado touchdown, the signature at the tip of the hook echo for that storm was identified by a Z_DR that was close to zero and anomalously low values of ρ_hv (less than 0.5).

The capability of a dual-polarization radar to detect tornadoes was further tested during the Joint Polarization Experiment (JPOLE), which was designed to evaluate the engineering concept and data quality of the polarimetric KOUN Weather Surveillance Radar-1988 Doppler (WSR-88D) and to demonstrate the utility of polarimetric radar data and products to operational users. Two significant tornadic events were observed with the KOUN radar in the Oklahoma City metropolitan area in May 2003. Analysis of these two events provides evidence of the polarimetric signatures that are associated with tornadic debris. The signatures looked very similar to the previous one observed with the Cimarron radar in 1999.

In this paper, we present polarimetric analyses of the three tornadic supercell storms that occurred in central Oklahoma on 3 May 1999, 8 May 2003, and 9 May 2003. Special attention is given to the polarimetric detection of tornadic touchdown that is associated with lofted debris. We also speculate about the possible interpretation of unusual polarimetric signatures that are observed elsewhere in tornadic storms, which might provide insight into microphysical aspects of tornadogenesis.

Tornadic case on 3 May 1999

Multiple tornadoes occurred in close proximity to the Oklahoma City metropolitan area on 3 May 1999 (Burgess et al. 2002). Approximate damage paths and highest Fujita-scale ratings for multiple storms within the Cimarron radar coverage area southwest of Oklahoma City are shown in Fig. 1. Polarimetric data from the Cimarron radar are available for the period from 2145 to 2322 UTC, after which time the radar went down after being hit by storm B (Fig. 1). Thus, the radar missed the most violent stage of storm A, which eventually struck the Oklahoma City metropolitan area. However, we have 15 volume scans of polarimetric data that include the developing stages of storms A and B, and a less destructive tornado rated as F3 in the Fujita scale (west of Chickasha, Oklahoma, in Fig. 1). This tornado produced an approximately 900-m-wide damage swath and lasted from 2246 to 2310 UTC. The tornado track was at the range of 45–60 km from the radar.

The 10-cm Cimarron radar measured the radar reflectivity factor Z at horizontal polarization, mean Doppler velocity V, Doppler spectrum width σ_υ, differential reflectivity Z_DR, differential phase Φ_DP, and cross-correlation coefficient ρ_hv between radar returns at two orthogonal polarizations (Zahrai and Zrnic 1993). These data were collected at elevations of 0.0°, 0.5°, 1.5°, 2.5°, 4.0°, and 6.0°, with an update time of approximately 6 min. All radar variables were measured with a radial resolution of 0.24 km and an azimuthal resolution of about 1.9° (although the radar beam has 0.9° width). The 1.9° beam spacing resulted from a longer dwell time that was used to reduce statistical errors of polarimetric measurements.

It is important that we utilize the radar data collected at 0.0° elevation for which the center of the radar beam is as close to the ground as possible. At such a low elevation, the radar beam is inevitably partially blocked and the power-related radar variables such as Z and Z_DR are biased. Partial blockage, however, does not affect the phase-related variables: Doppler velocity and differential phase. Moreover, it is possible to restore correct values of Z and Z_DR by using the specific differential phase K_DP and the concept of self-consistency among Z, Z_DR, and K_DP in rain (Gorgucci et al. 1999). The self-consistency technique proves to work well even in the presence of severe beam blockage (Ryzhkov et al. 2002b). The Z and Z_DR data collected at 0.0° and 0.5° elevations have been corrected according to such a methodology.

After analysis of all 15 volume scans of data, we have selected the one that started at about 2305 UTC to illustrate tornadic polarimetric signatures. At that time, the F3 tornado was relatively close to the radar (within 55 km). Therefore, small-scale features can be more easily resolved. A combined plot of Z, V, Z_DR, and ρ_hv at the lowest plan position indicator (PPI) scan (0.0°) is shown in Fig. 2. The data in Fig. 2 have a resolution of 0.5 km × 0.5 km. A simple linear interpolation is used to convert the data from polar to Cartesian format in our study. Note that at longer distances from the radar a “physical” spatial resolution determined by the size of the radar sampling volume can be worse than 0.5 km.

At 2305 UTC, a hook echo was well developed and the area of hail mixed with rain is recognized north of the hook at Y > −45 km. The latter is marked with Z exceeding 60 dBZ near ground and 65 dBZ aloft. Maximal radar reflectivity within the hook is slightly below 50 dBZ. Intense cyclonic rotation at the tip of the hook is evident in the Doppler velocity image. Relatively poor azimuthal resolution of the radar data in this particular dataset (about 2°) does not allow us to distinguish the fine structure of velocity field in Fig. 2c. Nevertheless, analysis of individual adjacent radials shows that the azimuthal change in Doppler velocity is about 39 m s⁻¹ across a distance of 2–3 km in the hook area.

As expected, differential reflectivity and the cross-correlation coefficient are anomalously low in the part of the hook where the tornado was detected (according to a ground survey, the tornado track is depicted by a thick dashed line in Fig. 2). Values of Z_DR (in decibels) are slightly positive or even negative at the tip of the hook. Such low Z_DR values can be explained by the presence of lofted debris in the radar resolution volume, and, to some extent, by enhanced differential attenuation along the propagation path that intersects the hail-bearing region northeast of the hook. Normally, Z_DR is corrected for differential attenuation using the empirical relation ΔZ_DR (dB) = 0.004Φ_DP (°), which is valid at the S band for rain in Oklahoma (Ryzhkov and Zrnic 1995). Note the high Z_DR (more than 4 dB) in the area ahead of the forward-flank downdraft (FFD). Analysis of the vertical structure of Z_DR shows that the region of high Z_DR stretches above the freezing level in the updraft region (the “Z_DR column”) and is very shallow (confined to a 1-km-depth layer) in the FFD area. The enhanced Z_DR in the updraft has been referred to as a Z_DR column (Conway and Zrnic 1993; Hubbert et al. 1998; Loney et al. 2002).

The cross-correlation coefficient drops below 0.4 at the inner side of the hook in the vicinity of the tornado track. Its minimal value is 0.25 at X = −20.0 km and Y = −44.5 km. The corresponding values of Z and Z_DR in that pixel are 43.2 dBZ and 0.0 dB, respectively. In pure rain or dry snow, ρ_hv usually varies between 0.980 and 0.997 if a dual-polarization radar is well designed. Because of quantization noise in the Cimarron data processor, the measured values of ρ_hv are negatively biased, and those high values have never been attained. This should be taken into account in interpretation of the Cimarron polarimetric data. Although absolute values of ρ_hv are not reliable, its relative changes are more trustworthy. Notable are lower ρ_hv values (less than 0.7) within the 55-dBZ contour of Z (indicative of a rain–hail mixture) and at weaker reflectivities in the southern part of the storm that is associated with the updraft. The latter signature is very repetitive in the supercell storms and might indicate a mixture of raindrops and light debris (leaves, grass, etc.) being advected into the cloud by strong inflow.

Vertical cross sections of the three radar variables along the 203° azimuth (marked by a straight solid line in Fig. 2) are also revealing (see Fig. 3). A radar reflectivity maximum centered at a 1.5-km height (42-km range) is accompanied by low Z_DR and high ρ_hv. This combination of radar variables might indicate pure hail. In the area underneath, Z_DR sharply increases but ρ_hv decreases, which likely point to a mixture of hail and big raindrops with ice cores inside.

Within the hook, a tiny, shallow signature centered at 49.5 km from the radar and extending to less than 1 km above ground is visible. This signature is characterized by Z_DR close to 0 dB and ρ_hv less than 0.4. Very close proximity to the tornado track on the ground suggests that this signature is very likely associated with tornadic debris. Indeed, this is exactly what is expected for randomly oriented nonmeteorological scatterers with an irregular shape and high refractive index.

Tornadic case on 8 May 2003

During JPOLE, polarimetric data were collected with the KOUN WSR-88D—a prototype of a future polarimetric WSR-88D. The KOUN radar experiences much less blockage at lower elevations than the Cimarron radar and surpasses the latter in the quality of the polarimetric data. Values of ρ_hv measured by KOUN reach theoretical limits for rain (0.997–0.998) and confirm the high quality of the radar engineering design and radar data processor. Higher values of ρ_hv ensure lower statistical errors in the estimates of all of the polarimetric variables—Z_DR, ρ_hv, Φ_DP, and K_DP—for the same dwell time (Bringi and Chandrasekar 2001). One of the basic requirements for the operational demonstration was the compatibility of the KOUN antenna-scanning strategy with the standard volume coverage patterns (VCPs) that are currently employed for Next-Generation Weather Radar (NEXRAD) operations. Most data during the spring of 2003 were collected following the VCP-11 scanning strategy, which includes 14 elevation sweeps from 0.5° to 19.5° and a volume update time of about 6 min.

As part of JPOLE, considerable KOUN data were acquired in tornadic storms. In particular, May 2003 was an active storm month with several damaging tornadoes occurring near KOUN. Most notable were the afternoons and evenings of 8 and 9 May when violent tornadoes struck the Oklahoma City metropolitan area.

On 8 May 2003, a destructive F4 tornado hit Moore, southeast Oklahoma City, Midwest City, and Choctaw, Oklahoma, creating a 27-km damage path (see Fig. 4). The tornado was spotted from 2210 to 2238 UTC [1610–1638 central standard time (CST)]. The KOUN radar recorded three full volume scans of data during this period, but they were not continuous because of a power outage. The beginning of tornado at 2210 UTC was well captured by the radar. However, because the first tornado touchdown occurred at a distance of only about 10 km from the radar, the data at the lowest sweeps were heavily contaminated by ground clutter.

A good volume scan of data began at 2229 UTC when the tornado was about 20 km from the radar. The tornado location on the damage track at that time is indicated by the cross in Fig. 4. A composite plot of Z, V, Z_DR, and ρ_hv at 1.5° elevation is shown in Fig. 5. A tornadic signature at the tip of the hook is marked by Z exceeding 50 dBZ, an obvious presence of a vortex in the Doppler velocity field, Z_DR close to zero, and anomalously low ρ_hv (less than 0.5). These components of the tornado signature are very similar to what was observed by the Cimarron radar on 3 May 1999. Outside the hook, the highest values of Z_DR are associated with low to moderate values of Z in the inflow region, which is an indication of pronounced drop sorting. This is discussed in more detail in section 6.

The vertical extension of the debris signature in the hook is about 500 m, as the composite RHI at azimuth = 25° demonstrates (see Fig. 6, 19–20 km from the radar). Among other notable features in the vertical cross section are the Z_DR column at the periphery of a hail core and extensive region of low ρ_hv (less than 0.90–0.95) stretching up from the tornado on the ground to the height of 7 km in the updraft portion of the storm. We do not exclude that this unusually low ρ_hv might be attributed to the mixture of meteorological particles and light debris that are lofted to the storm’s midlevel height by a strong updraft. A vertical column of specific differential phase K_DP (Fig. 6c) at distances of 29–31 km from the radar is associated with a major precipitation shaft that is loaded primarily with raindrops and possibly some hail, as can be concluded from the vertical distribution of Z and Z_DR (Loney et al. 2002).

A tornadic vortex is a very localized feature. Because of spatial smoothing of radar data as part of data processing and conversion of the data from a polar to Cartesian grid, the corresponding values of radar variables may, therefore, not be correctly represented in the PPI and RHI composite images presented in Figs. 5 and 6. Radial profiles of raw (unprocessed) data, although more affected by measurement noise, better represent extreme values of radar variables that are associated with tornadic debris. An example of such profiles of Z, Z_DR, and ρ_hv is presented in Fig. 7. The tornadic signature at the distance of 20 km from the radar is characterized by a local reflectivity maximum of 53 dBZ, combined with an unprecedented drop of the cross-correlation coefficient to a level of 0.2! Because of extremely low ρ_hv, the corresponding differential reflectivity is quite noisy, but it is definitely lower than in surrounding areas. Additional spatial averaging of Z_DR reveals this drop quite clearly (Fig. 5).

Tornadic case on 9 May 2003

The next day another strong tornado struck northeast Oklahoma City, Witcher, and rural parts of Jones and Luther, Oklahoma, over a 29-km damage path. The path of the F3 tornado is shown in Fig. 8. According to ground information, the tornado started at 2129 CST and ended at about 2206 CST (0329–0406 UTC 10 May 2003). During this time interval, the tornado was at distances of 35–55 km from the radar. The KOUN radar provided uninterrupted flow of polarimetric data throughout the entire lifetime of the tornado. A tornadic signature was identified during successive volume scans, which were updated every 6 min from 0334 to 0358 UTC.

The tornado was relatively far from the radar, and thus the data at the lowest elevation tilt of 0.5° were not contaminated by ground clutter. The fields of Z, V, Z_DR, and ρ_hv at the lowest radar scan at 0346 UTC are displayed in Fig. 9. At that moment, a strong classical hook echo had developed with all of the indications of tornado occurrence at the tip of the hook (X = 6.5 km, Y = 38.5 km)—increased Z, a Doppler vortex, anomalously low ρ_hv, and negative Z_DR.

A remarkable tornadic signature at distances of 39–40 km from the radar is evident in the vertical cross section through the hook echo (Fig. 10). The columns of negative Z_DR, negative K_DP, and low ρ_hv extend vertically from the ground up to a height of 2 km (4 km for ρ_hv). The corresponding Z is between 50 and 55 dBZ. There is no doubt that the radar echo in this region is dominated by nonmeteorological scatterers, that is, debris. Negative values of Z_DR and K_DP might be attributed either to a certain degree of vertical common orientation of the scatterers (if they are relatively small) or to their large size (provided that their orientation is not totally chaotic). In the Mie regime of scattering when the size of scatterers is much larger than the radar wavelength, both Z_DR and K_DP might have a negative sign even if the horizontal dimension of the scatterers is larger than vertical.

As can be seen from Fig. 10, two distinct columns of enhanced Z that are closely connected aloft and are separated by a “vault” at altitudes below 7.5 km have strikingly different polarimetric attributes. The left column is associated with the rear-flank downdraft (RFD) and exhibits an impressive tilted column of high positive K_DP, low values of ρ_hv (in comparison with its right counterpart), and highly variable low to moderate Z_DR. The right column represents the main precipitation core, consisting of rain below 2–2.5 km and hail above, as can be inferred from Z_DR, K_DP, and ρ_hv. A spectacular Z_DR column with maximal Z_DR values approaching 6.5 dB is observed in the weak echo region that is associated with the storm updraft.

Analysis of raw data along the radial through the tip of the hook at elevation = 0.5° shows that Z varies between 50 and 57 dBZ, ρ_hv drops to approximately 0.60–0.65, and Z_DR ranges from −2 to 0 dB in the tornado location at 39–40 km from the radar (Fig. 11). Of note is a tremendous change of Z_DR from −2 dB in the hook echo to 6.4 dB at the edge of the main updraft area and precipitation core (44–45 km).

Temporal evolution of the tornadic signatures

For scientific and practical reasons it is definitely desirable to relate tornado signatures to the life cycle of the phenomena. This, however, is beyond the scope of our preliminary investigation. Instead, we document herein the temporal evolution of the signatures for the three tornado cases. Thus, we have examined several consecutive scans of radar data at the lowest available elevations where ground clutter contamination was absent. For each scan, we analyzed the 0.5 km × 0.5 km pixels of Z, Z_DR, and ρ_hv data in a 20 km × 20 km area that is centered on the hook echo. A pixel where 45 dBZ < Z < 55 dBZ and ρ_hv < 0.8 is classified as a “ρ_hv debris signature.” A “Z_DR debris signature” is defined with 45 dBZ < Z < 55 dBZ and Z_DR < 0.5 dB. Because all ρ_hv data from the Cimarron radar are negatively biased because of a quantization problem in the radar processor, the ρ_hv threshold is lowered to 0.6 for the 3 May 1999 case. Such a threshold, although quite subjective, allows for the elimination of the majority of pixels containing rain, hail, and less reliable data in the weak reflectivity regions at the periphery of the storm where both ρ_hv and Z_DR can be biased by noise. After this designation was made, we counted the number of pixels identified as ρ_hv and the Z_DR debris signatures and estimate minimal and mean values of ρ_hv and Z_DR for such pixels [N_rhv, N_zdr, ρ^(min)_hv, 〈ρ_hv〉, Z^(min)_DR, and 〈Z_DR〉, respectively].

For the three tornadic storms, temporal dependencies of all six parameters characterizing the intensity of the polarimetric debris signature are displayed in Fig. 12. Time intervals during which a tornado was spotted on the ground are shown by gray thick lines at the top of each panel. Radar data were updated every 6 min. As mentioned before, three volume scans are missing between 2210 and 2228 UTC because of a power outage on 8 May 2003, and because the Cimarron radar lost power after 2322 UTC 3 May 1999. Despite these gaps in the data collection, the general evolution of radar signatures is consistent with the results of ground survey. The parameters ρ^(min)_hv, 〈ρ_hv〉, Z^(min)_DR, and 〈Z_DR〉 reach their minima close to the middle of the time intervals when tornadoes were observed, which was also the time of peak tornado intensity. In all three cases the radar did not detect debris before first tornado touchdown occurred. However, on both 8 and 10 May 2003 the radar debris signatures were detected 6–10 min after tornado ended, which is most likely because lofted debris was still in the air.

Discussion

Analysis of the three tornadic cases that is presented in this study shows that the polarimetric debris signature is a repeatable feature. The signature exists throughout a tornado’s lifetime, provided that the tornado has an intensity of at least F3 according to the Fujita scale. A cursory analysis of other tornadic storms indicates that the majority of the weak tornadoes did not produce definable signatures. One possible reason for this is that wind speeds in weak tornadoes are not sufficient to significantly damage structures and loft debris. Another feasible explanation is that some of the weaker tornadoes may be too short lived. Therefore, a debris signature might have been missed because of coarse temporal sampling.

On the other hand, our study of all of the significant nontornadic supercell storms that were observed during JPOLE does not reveal such a signature. Although Z_DR and ρ_hv can drop considerably in the middle of hail cores, it is almost impossible to confuse hail and tornado designations because of their location in the storm and the depth of the ρ_hv dip. In hail, ρ_hv usually does not drop below 0.85 even if hail is large. The only exception in our dataset is an extreme hail event that occurred on 14 May 2003 for which the measured ρ_hv in the center of the hail core dropped to as low as 0.75. That was an exceptional hailstorm that produced hail of a 13-cm size!

Summarizing our observations of these three events, we can tentatively formulate the following five criteria for polarimetric tornado detection: 1) the presence of a hook echo, 2) ρ_hv < 0.8, 3) a pronounced vortex signature in the Doppler velocity field, 4) Z_DR < 0.5 dB, and 5) Z > 45 dBZ. If conditions 2–5 are satisfied in the hook area, then it is very likely that a tornado is lofting considerable debris. Among criteria 2–5, criterion 2 probably has the best discriminating power. The cross-correlation coefficient is the most attractive variable because, unlike Z_DR, it is not affected by radar miscalibration, attenuation in precipitation, and partial radar beam blockage, provided that the signal-to-noise ratio is sufficiently high. The LDR, considered as a proxy for ρ_hv, is vulnerable to all of these conditions.

One reservation regarding the use of ρ_hv is that the magnitude of the cross-correlation coefficient is affected by the variability of the differential phase within the radar resolution volume. If a gradient of Φ_DP across the radar beam is high and the radar sampling volume is too large, then ρ_hv noticeably decreases. This factor explains an observed general decrease of ρ_hv with distance, especially if a propagation path contains a large amount of precipitation. Hence, the ρ_hv threshold in criterion 2 might depend on range and Φ_DP.

The debris signatures can be very useful to confirm tornado warnings and tornado damage, and to pinpoint the current tornado location. Although tornado detection is important, its prediction and early warning are even more important. A cursory look into evolution of the 3D pattern of polarimetric variables in a tornadic supercell reveals quite unusual and intriguing polarimetric signatures aloft, and, in the near proximity of the storm, that might be related to tornado development. Similar to the debris signature, these polarimetric patterns are also repetitive and require microphysical interpretation.

Very high values of Z_DR exceeding 4 dB are regularly observed at the periphery of high-reflectivity areas ahead of the FFD and in the inflow region (see Figs. 2, 5 and 9). As was already mentioned, the Z_DR signature in the FFD area is usually shallow, whereas the one associated with inflow and updraft has a large vertical extent (Figs. 3, 6 and 10). This can be seen very well in a PPI at the higher elevation of 3.5° for the case of 9 May 2003 (Fig. 13). At midlevel heights (2.5–3 km) outside of the inflow and updraft regions, low values of Z_DR signify hail, graupel, and snow. A comma-shaped high-Z_DR signature in the updraft region is likely attributed to large supercooled drops or melting hailstones.

Anomalously high Z_DR can be explained by intense drop sorting in the presence of strong wind shear and mesoscale rotation. Indeed, cloud and precipitation particles (liquid drops, graupel, or hail) that originate in high-reflectivity regions aloft fall to the ground with very different trajectories, depending on their inertia, that is, size (Browning 1965). Smaller hydrometeors follow the airflow quite well, whereas the larger ones can fall “against the flow” in a highly sheared wind environment. For example, smaller hydrometeors that fall out from the overhang region aloft are easily advected or recirculated back to the cloud by a strong updraft (Conway and Zrnic 1993). In contrast, big particles continue to fall down despite the updraft. Similarly, if the hydrometeors travel in the presence of strong air circulation, then the most massive particles have much better chance to be “centrifuged” away from the center of rotation (Snow 1984; Dowell et al. 2004). The comma-shaped Z_DR signature in Fig. 13 that apparently “wraps” a mesoscale vortex points to such a possibility.

In both situations, strong size sorting of hydrometeors takes place and, as a result, the areas adjacent to strong reflectivity cores contain a small number of very large particles (big raindrops or melting hailstones) and lack any small particles. A combination of high oblateness and low concentration produces low Z and high Z_DR. A degree of size sorting is directly related to kinematic properties of the storm. Thus, polarimetric signatures might be very useful in the interpretation of dynamical processes in supercell storms.

Intense size sorting affects the cross-correlation coefficient, as well. Drop size distributions that are skewed toward larger sizes are usually characterized by lower ρ_hv. In addition, the presence of nonmeteorological scatterers (including debris) in a mixture with hydrometeors further depresses ρ_hv. Very low measured values of ρ_hv in the updraft regions for all three examined tornadic storms indicate that large portions of updraft contain light debris, such as leaves, grass, and dust. The magnitude of ρ_hv and the vertical extent of the low ρ_hv signatures in updrafts provide indirect measure of their strength, which is quite difficult to estimate from conventional Doppler measurements. One has to be cautious, however, regarding the interpretation of ρ_hv that can be negatively biased because of enhanced gradients of differential phase within the radar resolution volume.

Once light debris is lofted to higher levels in a tornadic storm, it takes some time (tens of minutes) for debris to sediment to the ground (Magsig and Snow 1998). Suspended light debris is the most reasonable explanation of the supercell storm “wake” signature that is usually observed in the wake of the strong low-level wind field behind the storm (see Fig. 9). It is characterized by low Z (less than 30 dBZ), low ρ_hv (less than 0.7), and mean Z_DR varying between 1 and 2 dB. Low values of ρ_hv point to nonmeteorological scatterers as a source of echo. Ground clutter is excluded because the Doppler velocity is far from zero, and Z_DR is mainly positive, whereas it is usually slightly negative for ground targets. Biological scatterers, like insects and birds, have a much higher Z_DR and quite a different differential phase upon scattering δ. The observed δ in the “wake” echo is about 50°–60°, which is different from the one that is typical for insects (10°–40°) and birds (70°–100°) (Zrnic and Ryzhkov 1998; Schuur et al. 2003). Hence, lofted light debris with a certain degree of common alignment (probably leaves and grass) remains the only feasible explanation for such an echo.

Conclusions

All previous observational studies of tornadoes that were made with Doppler radars emphasized the kinematic properties of storms (see a review by Markowski 2002). For the first time we have obtained strong evidence that a dual-polarizaton radar can effectively complement Doppler information and provide additional tornado detection capabilities. Three major tornadic storms that hit the Oklahoma City metropolitan area in recent years all exhibit well-defined polarimetric debris signatures that are characterized by an unprecedented drop in the cross-correlation coefficient ρ_hv and differential reflectivity Z_DR in the hook echo. Such signatures are less pronounced for weaker tornadoes, but reliably identify tornadoes rated as F3 in the Fujita scale.

The debris signature that is associated with tornadic touchdown is quite small with a horizontal size of about 1 km and vertical extent of 1–3 km. Doppler measurements require good spatial resolution in order to resolve a small tornado vortex, whereas identification of polarimetric signatures can be accomplished with a coarser resolution. Moreover, these signatures are “isotropic” in their nature. That is, as opposed to Doppler velocities, they do not depend on a viewing angle.

Although a very small tornado signature might not be well resolved at long distances from the radar, larger-scale polarimetric signatures, associated with light debris (leaves, grass, etc.) that is lofted in the cloud by a strong updraft, as well as intense size sorting of hydrometeors might be helpful to diagnose the current state of the supercell storm and its potential ability to produce a tornado. Light debris in the inflow region of the storm and in its wake is associated with low values of ρ_hv and a sizeable differential phase upon scattering, whereas size sorting is manifested by very high values of Z_DR. Both larger-scale polarimetric signatures provide indirect estimates of the strength of vertical flows and circulation within the storm.

In cases in which traditional Doppler tornado-warning signatures are absent or overlooked by forecasters, the polarization tornado signature might be very valuable in preventing what otherwise might have been a missed warning. This signature might also be very helpful in issuing accurate severe weather–warning updates to pinpoint the current tornado location and confirm the occurrence of damage (based on debris).

A cursory look into evolution of the 3D pattern of polarimetric variables prior to tornadic touchdown reveals quite unusual and intriguing polarimetric signatures aloft that might be related to the development of a subsequent tornado. Understanding and interpretation of these signatures could provide insight into microphysical aspects of tornadogenesis.