The finescale three-dimensional structure and evolution of the near-surface boundary layer of a tornado (TBL) is mapped for the first time. The multibeam Rapid-Scan Doppler on Wheels (RSDOW) collected data at several vertical levels, as low as 4, 6, 10, 12, 14, and 17 m above ground level (AGL), contemporaneously at 7-s intervals for several minutes in a tornado near Russell, Kansas, on 25 May 2012. Additionally, a mobile mesonet anemometer measured winds at 3.5 m AGL in the core flow region. The radar, anemometer, and ground-based velocity-track display (GBVTD) analyses reveal the peak wind intensity is very near the surface at ~5 m AGL, about 15% higher than at 10 m AGL and 25% higher than at ~40 m AGL. GBVTD analyses resolve a downdraft within the radius of maximum winds (RMW), which decreased in magnitude when varying estimates for debris centrifuging are included. Much of the inflow (from −1 to −7 m s−1) is at or below 10–14 m AGL, much shallower than reported previously. Surface outflow precedes tornado dissipation. Comparisons between large-eddy simulation (LES) predictions of the corner flow swirl ratio Sc and observed tornado intensity changes are consistent.
The corner and boundary layer regions of tornadoes (e.g., Lewellen 1976, 1993; Snow 1982; Kessler 1986, 197–236; Davies-Jones et al. 2001) are the most difficult and most critical to sample because they are in direct contact with the underlying surface, cause damage, and dictate the vortex structure at low levels and perhaps through a significant depth of the tornado. Although mobile radar observations of the tornado core flow have become more frequent (e.g., Wurman et al. 1996; Wurman and Gill 2000, hereafter WG00; Bluestein et al. 2003; Bluestein et al. 2004; Lee and Wurman 2005, hereafter LW05; Wurman et al. 2007a, hereafter W07; Tanamachi et al. 2007; Kosiba et al. 2008, hereafter K08; Kosiba and Wurman 2010, hereafter KW10; Wakimoto et al. 2012, hereafter W12; Wurman et al. 2013a, hereafter W13; Wurman and Kosiba 2013), they commonly are >40 m AGL and inadequately sample the near-surface tornado boundary layer (TBL) (i.e., the lower portion of the boundary layer that may be impacted directly by frictional effects).
To observe the TBL with radar, it is necessary for that radar to be close to the tornado to mitigate the effects of beam spreading and intervening obstructions (e.g., terrain, structures, and foliage). On only rare occasions (e.g., W07) are observations obtained <10 m AGL. Ideally, near-surface three-dimensional radar observations in combination with surface in situ one-dimensional transects of the TBL would combine to provide a time-varying three-dimensional mapping. However, in situ wind observations are exceedingly difficult and hazardous to obtain due to the violence, unpredictable paths, and typically short lifespans of tornadoes (Wurman et al. 2013b, manuscript submitted to Bull. Amer. Meteor. Soc.). In the rare instances that both in situ and low-level radar wind observations have been obtained (e.g., W13), ground-relative wind speeds near the surface, Vg, are comparable to or even stronger than radar-measured winds, Vd, ~30 m AGL.
Numerical simulations (e.g., Lewellen et al. 1997; Lewellen et al. 2000, hereafter L00; Lewellen and Lewellen 2007, hereafter LL07) and laboratory models (e.g., Church et al. 1979) of tornadoes have been instrumental in understanding the TBL, but these results remain largely unverified observationally. Hoecker's (1960) video analysis of the flow in a tornado attempted to quantify the near-surface structure of a tornado. Using debris and cloud tags as tracers, radial inflow, Vr < 0, was identified below 152 m AGL, and maximized at 53 m AGL. Doppler on Wheels (DOW; Wurman et al. 1997; Wurman 2001, 2008) radar data in one tornado did not reveal inflow ~30 m AGL (WG00), but it was suggested that centrifuging of debris (Dowell et al. 2005) may have masked the convergence signature. DOW data, taken at ~10 and 14 m AGL from two different tornadoes (W07; Alexander and Wurman 2005, hereafter AW05), revealed convergences of 0.10 and 0.06 s−1, respectively, suggesting strong convergence (not fully masked by centrifuging) near the surface.
LW05, KW10, W12, and W13 have used the ground-based velocity-track display (GBVTD) (Lee et al. 1999) scheme to resolve three-dimensional tornado structure. LW05 revealed inflow from z = 0 to 1.15 km AGL in a large and violent tornado. W12 applied the GBVTD technique to DOW data from the second Verification of the Origins of Rotation in Tornadoes Experiment (VORTEX2; Wurman et al. 2012) in a tornado at two consecutive times. Initially, there was strong outflow, while 2 min later there was very weak inflow surrounding the tornado. An approximation for centrifuging of hydrometeors weakened the initial outflow and strengthened the later inflow. The large-eddy simulation (LES) of Lewellen et al. (2008) demonstrated that small debris could alter the wind speeds and tornado structure near the surface. Using GBVTD and numerical simulations, Nolan (2013) emphasized the need to represent accurately the low-level inflow and to account for debris centrifuging. The debris centrifuging velocity, Vcf, which depends on debris type, amount, density, shape, recycling, electrical/scattering characteristics, etc., currently is unknown, and can only be approximately accounted for in radar-based analyses.
W13 combined in situ anemometer wind measurements (in the same tornado analyzed by W12) at 3.5 m AGL with Rapid-Scan DOW (RSDOW; Wurman and Randall 2001) observations ≥30 m AGL to reconstruct the low-level tornado vortex structure. At 3.5 m AGL, strong radial inflow (Vr ~−40 m s−1) was present just outside the radius of maximum Vt (RMW), but at ~30 m AGL, radar analyses revealed little to no inflow. Integrated analyses revealed a divided vortex structure with an axial updraft near the surface and a downdraft aloft. No data were available between 3.5 and 30 m AGL, so the details of the vertical profile of the winds in these lowest, critical levels of the TBL are unknown. Peak anemometer-measured winds, Vg = 58 m s−1, were more intense than those measured by radar ~30 m AGL. DOW observations from other tornadoes also suggested that the most intense velocities may occur <30 m AGL (W07). Except for W13, these previous radar-based analyses used conventionally scanned radar data, resulting in limitations due to possible errors resulting from rapid tornado evolution during the several 10s of seconds between observations at the lowest and highest levels (e.g., Wurman et al. 2007b).
2. Deployment overview and data
During the 2012 season of the Radar Observations of Tornadoes and Thunderstorms Experiment (ROTATE; e.g., Wurman 1998, 2008) field campaign, three DOWs, including the RSDOW, and in situ measurement teams deployed in advance of potentially tornadic storms moving toward Russell, Kansas, after dark on 25/26 May 2012. At 0230 UTC 26 May, DOW71 was deployed at 38.84195°N, 98.8524°W and the RSDOW deployed at 38.85638°N, 98.8545°W, establishing a 1.6-km, dual-Doppler baseline (Fig. 1). An instrumented mobile mesonet vehicle (P1) carrying “tornado pods” (not deployed) moved north from DOW7.
A tornado formed unexpectedly at 0235 UTC moving northeast, broadening, and impacting P1 at approximately 0238 UTC. The tornado contracted/reformed to the north and then moved northeast, where P1 drove into its southwestern sector just after 0241 UTC. The tornado continued north, with the center passing 30 m east of the RSDOW, resulting in generator failure and ending RSDOW operations.
The tornado transported a house with its resident,2 located 30 m northeast of the RSDOW, about 30 m eastward, destroying it completely3 [10 degrees of damage (DOD10) on the enhanced Fujita (EF) scale; Edwards et al. (2013); see Fig. 1]. A nearby house and several other structures were also damaged. At 0242:11 UTC, Vg = 43 m s−1 was observed by P14 at 3.5 m AGL, which experienced the weaker, western side of the tornado core flow. Maximum RSDOW-measured winds, Vd, were 64 m s−1 at 6 m AGL5 at 0240:53 UTC, and the peak velocity difference across the vortex was 113 m s−1 at 6 m AGL at 0241:08 UTC. DOW7 observations depicted a low-reflectivity eye and an intermittent debris ring (Fig. 1). The RMW was typically about 50–70 m during the analysis period. The tornado dissipated at ~0243 UTC.
The proximity of the RSDOW to the tornado (~130 m to the circulation center at the last data collection time, 0242:32 UTC) and six simultaneous radar beams resulted in mapping of the TBL with unprecedented spatial (25-m range and 3-m beamwidth with 1-m oversampling) and temporal resolution (7-s volumes) and proximity to the ground (simultaneous slices as low as 4, 6, 10, 12, 14, and 17 m AGL), well resolving the three-dimensional structure and evolution of the TBL for the first time (Figs. 2 and 3).
GBVTD analyses using 7-s RSDOW volumes from 0240:53 to 0242:25 UTC yielded independent three-dimensional wind fields at 14 times (Fig. 5).6 The sensitivity of the results to objective analysis parameters, lower boundary conditions, and errors in the center location7 was explored. The vortex structure was not altered substantially by changes in the grid spacing8 from 2 to 10 m, with only a slight reduction in Vt at grid spacing = 10 m. Small changes in tornado center determination also did not impact the results substantially; an alternate, no-slip, lower-boundary condition reduced Vt below the lowest observation level, but did not alter the vortex structure. The sensitivity to a simple inclusion of debris centrifuging is discussed below.
3. Results and discussion
The maximum Vd from each RSDOW scan9 and objectively analyzed grid was normalized to Vd (10 m AGL). Peak Vd was observed at the lowest radar-observed level, 4–6 m AGL, at all times except 0241:14 and 0241:21 UTC, when the maximum was observed at 10 m AGL (Fig. 4). At most times (and in the median and mean profiles), peak Vd decreased ~15% from 5 to 10 m AGL, then gradually decreased another ~10% up to 25–40 m AGL, matching the trend noted from 14 to 100 m AGL in W07, and consistent with Vg (3.5 m AGL) > Vd (30 m AGL) reported in W13. This indicates that the “nose” (maximum) of Vg was very near 5 m AGL and that the true shallow TBL radial inflow was mapped here in detail for the first time. The strongest inflow may even exist below 5 m AGL, where further frictional reduction of Vt could lead to an imbalance in the pressure-gradient force, causing enhanced inflow. Temporal variations in Vd likely were due to true changes in vortex intensity as the tornado is observed by both the DOW7 and RSDOW to rapidly strengthen then rapidly dissipate. In situ anemometer observations from P1, with peak Vg = 43.2 m s−1 at the edge of the western (weak) edge of the core flow region,10 were roughly consistent with or slightly higher than observational-angle-adjusted Vd at 5 m AGL over P1 (Fig. 2e),11 corroborating that the maximum Vg occurred near or possibly below 5 m AGL. Critically, this result shows that radar observations of Vd at ~50 m AGL (e.g., WG00, AW05, W07, W13) are likely underestimations of the Vg typically occurring at 5–10 m AGL. However, comparing radar-measured Vd to hypothetical “standard” 3-s-averaged anemometer-measured Vg is problematic.12
At all analysis times, GBVTD results revealed a two-celled tornado vortex structure, with a downdraft inside the RMW extending from the top of the domain to the surface. Near-surface inflow, just outside of the RMW, was present in all but the last three analysis times (Figs. 5 and 6). Maximum inflow occurred at 0240:53 UTC, near the time of maximum Vd and Vt. At 0240:53 UTC, Vr < −1 m s−1 was present ≤ 14 m AGL, while Vr < −5 m s−1 was confined to ≤8 m AGL, with the strongest inflow (Vr of approximately −7 m s−1) at 5 m AGL. P1 observations at 0241:14 UTC, adjusted for tornado translation and motion, result in a tornado-relative velocity of 42.9 m s−1 from 281° (Fig. 2), suggesting slight inflow, Vr < 0, while the GBVTD at this time suggests slight outflow, Vr > 0 (Figs. 5 and 6). This discrepancy could be caused by inadequate compensation for debris centrifuging, errors in correcting for tornado movement, asymmetries in the tornado, stronger inflow <5 m AGL, errors in the wind or radar measurements, or a combination of these factors.
The value of Vt was maximum in the TBL, below ~6 m AGL (Fig. 6). By 0241:00 UTC, the depth of the inflow layer (Vr < −1 m s−1) had decreased to 10 m AGL and the layer containing Vr < −4 to −5 m s−1 decreased to <6 m AGL. From 0241:00 to 0241:22 UTC, the depth of the inflow layer was ~10 m AGL, and Vr and Vt generally decreased in magnitude. At 0241:29 UTC, there was a brief increase in the Vr < 0 and Vt, subsequently followed by a decrease in Vt until 0242:11 UTC, when Vt increased until the end of the analysis. Although Vt increased from 0242:11 to 0242:25 UTC, outflow Vr > 0 was present outside the RMW and the tornado dissipated approximately 60 s later. Strong Vt immediately prior to tornado dissipation has been documented before (e.g., WG00, Kosiba et al. 2013).
Since there is not adequate information on precisely how debris can bias Doppler velocity measurements, an exact quantification of debris centrifuging is beyond the scope of this study. However, using the results of Dowell et al. (2005) and W12, Nolan (2013) provides a simple estimate of debris-centrifuging biases on retrieved axisymmetric winds. Following Nolan (2013) [Eq. (3.1)], and consistent with the values presented in W12 and Dowell et al. (2005), Vc = 8 m s−1 was used in this study since this was a weak tornado and likely only hydrometeors and small debris, such as dirt, gravel, and grass13 were present throughout the GBVTD analysis interval as the tornado crossed open farm and grassland14 prior to impacting the structures near the RSDOW. Including an estimate for debris centrifuging moved the location of maximum convergence inward (Fig. 6) and caused the weak central downdraft near the surface to become a weak central updraft at 0241:22–0241:43 and 0241:57–0242:04 UTC. Indeed, with the inclusion of a simple debris bias, the inward progression of the radial winds is consistent with the LES results of LL00. The sign of Vr outside of the RMW was not changed by the inclusion of debris (Fig. 5). Since one unknown in accounting for debris centrifuging is the size, shape, and density distribution of the debris–hydrometeor mix, and hence Vcf, sensitivity of the results to varying Vcf (4 and 12 m s−1) was explored (Figs. 6c and 6d). As expected, the higher (lower) value increased (decreased) inflow and allowed the inflow to penetrate farther (less far) inward. The central updraft–downdraft responded according to changes in the inflow.
The depth of the inflow layer was much shallower than has been reported in previous studies (e.g., LW05, K08, KW10, W12), but these results were based completely or primarily on data well above the shallow TBL revealed here. W12, KW10, K08, and LW05 employed vertical grid spacings of 50, 40, 25, and 30 m, respectively, so these likely did not resolve very-near-surface inflow, but instead captured the large-scale inflow associated with the wind field outside the core flow region of the tornado, or a smoothed representation of inflow below their lowest analysis level. The weaker inflow revealed here compared to the analyses of LW05, K08, KW10, and W13 may be due to the resolution of the true TBL or because the current tornado was weaker and short lived, with low-level inflow not conducive to sustaining an intense, long-lived vortex (L00).
In the LESs of L00 and LL07, a corner flow swirl ratio15 Sc has been defined to characterize the impacts of near-surface, tornado-scale flow on vortex structure and intensity. While Sc does not replace the outer-scale swirl ratio So (e.g., Church et al.1979; Lee and Wurman 2005), Sc underscores the importance of near-surface flow structure, which is not captured by So.16 At the “optimal” Sc value (Sc = 0.70–1.38; LL07), the ratio of the average maximum swirl velocities (Vmax) in the surface layer to the average swirl velocities above the surface layer (Vc) is 2.5 (L00). In the current analysis, values of Vmax (z = 4 m AGL)/Vc (z = 20 m AGL) < 1.5. The highest values (Vmax/Vc = 1.37–1.47) occurred at the first three analysis times. The lowest values (~1.0) were observed at 0241:22, 0242:04, 0242:18, and 0242:25 UTC, when the angular momentum outside of the RMW was reduced in the TBL, perhaps indicating an influx of far-field lower angular momentum fluid near the surface, decreasing the tornado intensity. Using quantities available in the GBVTD domain, and only for times when Vr < 0 was observed, Sc ≫ 1,17 so Sc would have to either decrease (assuming an unchanged flow aloft) or stayed fixed (if the core radius aloft was reduced) in order for the vortex to intensify.18
The TBL and inflow layer were much shallower than previously documented, confined below 10–14 m AGL. While it is possible that more intense tornadoes may have deeper boundary and inflow layers, it is likely that prior observational studies characterized the inflow layer of the larger-scale circulation, not of the tornado itself. Estimation of Sc predictions from the current data are consistent with LES results. A two-celled vortex structure is revealed by GBVTD analysis without the inclusion of debris centrifuging, and outflow outside the RMW was observed just prior to tornado dissipation. Inclusion of varied estimates for debris centrifuging changed weak downdrafts to very weak updrafts inside the RMW and the flow direction outside of the RMW was unaltered.
The most intense winds in the tornado occurred ~5 m AGL, extending surfaceward the results of previous studies sampling above the TBL (e.g., W05, W07) or sparsely within the TBL (W13). Radar-observed Vd a few 10 s of meters AGL are shown to be representative or even underestimates of Vg approximately <10 m AGL. More observations are needed over a range of vortex structures and intensities to characterize the TBL more generally.
Analysis supported by NSF-AGS-0910737 and DOWs by NSF-AGS-0801041. We thank Curtis Alexander, Paul Robinson, and Rachel Humphrey.
DOWs have 3-dB beamwidths of ~0.9°, oversampled at 0.3°–0.4°, gating at 25 m (RSDOW) and 30 m (DOW7).
P1 crew discovered the resident ~100 s after the event, sitting on the rubble of the transported/destroyed house, suffering from lacerations and a broken collarbone; the crew assisted in her extraction.
The National Weather Service (NWS) rated the tornado EF2.
P1 hosts an R. M. Young 5103 propeller-type anemometer.
Reported DOW observations height adjusted +3 m, accounting for antenna height AGL.
RSDOW data were objectively analyzed onto a Cartesian grid using a Barnes scheme (Barnes 1964), grid spacing (radius of influence) of 5 (10) m and 2 (3) m in the horizontal and vertical directions, respectively. Tornado translational motion was subtracted from Vd. Tornado centers were vertically aligned, eliminating tilting errors, but not correcting for flow asymmetries. Vertical winds (w) were derived by the upward integration of the continuity equation, assuming w(z = 0) = 0 m s−1 and free slip.
Errors in vortex center location were <7 m.
Grid spacing (δ) = Δ/2.5 (where Δ is the raw data spacing) (Koch et al. 1983).
It is likely that Vg > Vd because Vd only captures the component of motion toward–away from the radar.
P1 drove into the western side of the tornado after the tornado center had moved east of the road.
P1 is visible as moving ground clutter in the lowest RSDOW sweeps.
Radar measurements are instantaneous and represent reflectivity-weighted averages of scatterer (rain/debris, with associated terminal/centrifuging velocities) motion, usually determined using pulse-pair or spectrally based signal processing techniques, over spatial volumes not equal to 3-s air parcel trajectories.
Since the observations were close to the surface, dirt, gravel, and/or grass likely were appreciable scatter types in the radar sample volumes. The in situ P1 crew did not report airborne debris hitting their vehicle.
The surface roughness length was approximated as 0.1 m, characteristic of “openly rough” terrain (Davenport et al. 2000).
Here, Sc = rcΓ∞/ϒ, where rc is a characteristic radius of Vt(max), Γ∞ is the ambient angular momentum, and ϒ is the depleted angular momentum flux in the surface layer.
Note that So characterizes the flow in which the tornado forms. Both So and Sc are difficult to define in nature since flow parameters are not easily defined. In addition, So cannot be approximated in the current study since the analysis domain is too small to define the large-scale flow in which the tornado is embedded.
Assuming rc = 60 m, Γ∞ = 3.0 × 103 m2 s−1, and ϒ = 1.0 × 103 m5 s−2.
Values of parameters composing Sc are uncertain, but reasonable choices yield Sc ≫ 1.