a. Vertical structure

Figure 5 presents upper sounding data gathered from 2030 to 2202 JST 19 September at Tateno. Since synoptic conditions changed very little and no significant storm nor frontal system passed over Tateno during the observation, the sounding is considered to give a good representation of the environment in which the storms developed on the eastern Kanto Plain.

The Showalter’s stability index (SSI) value calculated from the sounding was about −2 K indicating an unstable environment. CAPE was about 1600 J kg⁻¹, and is comparable to that of the miniature supercell environment (1834 J kg⁻¹) under hurricane Danny in 1985 (McCaul and Weisman 1996). Though this value is significantly larger than the average CAPE value of the hurricane-tornado environment (e.g., 253 J kg⁻¹; McCaul 1987), it is smaller than that of typical supercell environments over the Great Plains (e.g., 2542 J kg⁻¹ for Oklahoma supercells; Bluestein and Jain 1985).

Figure 6 shows the wind hodograph obtained from the same sounding. Strong vertical shear with veering existed below 3 km AGL, which again is similar to that of hurricane Danny’s case, indicated in the same figure by dashed line. The bulk Richardson number (BRN), introduced by Weisman and Kemp (1982), was 10, and is in the lower end of possible range for simulated and observed supercell storm development. Storm relative helicity (SRH), which is suggested to be an important component in quantitative measurements of the storm environment (e.g., Davies-Jones et al. 1990), was 310 m² s⁻² assuming storm motion of 17 m s⁻¹ to the north-northeast. This value is about twice that of the threshold value (150 m² s⁻²) for mesocyclone formation noted by Davies-Jones et al. (1990). Although recent research (e.g., Davies-Jones 1993; Markowski et al. 1998) suggests that significant spatial and temporal variabilities may exist, the estimated CAPE, BRN, and SRH values indicate that the environment would support the development of supercells.

b. Preexisting surface boundary

When the prevailing wind is southerly, a mesoscale surface boundary often forms on the Kanto Plain, which is bordered from west to north by mountains ∼1000–2000 m in height (e.g., Kawamura 1970). The boundary on such an occasion runs from north-northeast to south-southwest, and divides warm air in the southeast and cool stagnant air in the northwest. On the evening of 19 September, such a surface boundary with sharp temperature-gradient and wind shear did exist near the eastern edge of the precipitation region (see Fig. 4) over the western portion of the Kanto Plain.

Figure 7 shows the distribution of surface wind and temperature derived from AMeDAS and EMS at 1900 JST, together with the reflectivity map obtained by the Tokyo radar. The boundary is distinguishable, extending north-northeastward from the northeastern edge of the Izu Peninsula and located in the central Kanto Plain (see Fig. 1). The western side of the boundary contained cool air produced by the precipitation over the western Kanto Plain, and in the east side was warm air advected from south. The boundary had an eastward protuberance near the Inagi City where a tornado occurred between 1900 and 1920 JST. This protuberance may have been caused by outflow from storm D, which produced the Inagi tornado.

From 1800 to 2100 JST the bandlike echo cluster over the western Kanto Plain moved eastward (Figs. 4c–f), the surface boundary also drifted slowly eastward in the northern part of the Kanto Plain. In the southern part of the Plain, on the other hand, the boundary retreated westward. After 2100 JST, the boundary remained stationary, though some fluctuations, related to the passage of storms near or over the boundary, were observed.

Figure 8 shows the traces of wind direction, wind speed, temperature, dewpoint temperature, and pressure from 2000 to 2300 JST at Utsunomiya Local Meteorological Observatory (ULMO, see Fig. 2). Around 2100 JST, a sharp temperature drop of 2 K over 5 min was recorded, caused by a passage of the surface boundary toward the east. The boundary returned from the east around 2140 JST.

Figure 9 shows distributions of surface wind and temperature at 2200 JST. The boundary was located 5∼10 km west of the Mibu Town (solid square in the figure), displaying an eastward protuberance near the storm F, which produced the Mibu tornado around 2215–2200 JST, like the one near storm D in Fig. 7.

Temperature gradient and vertical vorticity across the boundary are evaluated from the data at 1900 and 2200 JST. The temperature data at Chofu Airport Meteorological Observatory (CAMO in Fig. 2) and Tachikawa (about 11 km apart, not included) show that a temperature gradient across the boundary of at least 3.5 K (10 km)⁻¹ at 1900 JST. A similar value, about 3 K (10 km)⁻¹, was also observed near ULMO at 2200 JST. The vertical vorticities at 1900 and 2200 JST were estimated to be at least 4 × 10⁻⁴ s⁻¹ [∼4 m s⁻¹ (10 km)⁻¹] in the boundary (see Figs. 7 and 9).

a. Tornadoes

Typhoon 9019 produced four tornadoes from 1350 JST on 19 to 0010 JST 20 September (see Fig. 1). All the tornadoes occurred in the northeast quadrant of the typhoon. This fact is consistent with the findings of previous statistical studies: the northeast quadrant of a hurricane (Novlan and Gray 1974; Gentry 1983) or the right-forward quadrant relative to a typhoon/hurricane motion is the preferred location of hurricane/typhoon-associated tornadoes (Fujita et al. 1972; Mitsuta 1983; McCaul 1991). Three of the tornadoes on the Kanto Plain occurred after 1900 JST (see Fig. 1). The Inagi tornado, estimated as F1 in the Fujita scale, occurred between 1900 and 1920 JST. Two F0 damage swaths produced by heavy winds (not identified as tornadoes) occurred in Chiyokawa Village around 2150 and in Shimotsuma City around 2200 JST. The most intense tornado, ranked as F2, occurred in Mibu Town between 2215 and 2200 JST. The Azuma tornado occurred around 0010 JST. Rotation in all the three tornadoes was cyclonic.

In contrast to the Inagi and the Mibu tornadoes, which occurred near/over the surface boundary, the Azuma tornado and damage-producing winds in Shimotsuma City and Chiyokawa Village (S and C in Fig. 1) occurred about 50 km and 25 km east of the boundary, respectively, and did not have relation to the boundary. Damage in Azuma, Shimotsuma, and Chiyokawa, however, was minor when compared to that caused by the Inagi and the Mibu tornadoes.

Characteristics of the three tornadoes and the two heavy winds are summarized in Table 1.

b. Supercell identification

During the night of 19 September 1990, a total of nine supercell storms were identified over the Kanto Plain by the Doppler radar. Recent studies (Weisman and Kemp 1982; Doswell et al. 1990; Doswell and Burgess 1993) suggest that a supercell may be properly defined as a storm that contains a rotating updraft called a mesocyclone. In this paper, we adopt this definition of a supercell. However, the presence of an updraft near a vortex is difficult to identify because vertical velocity information cannot be extracted from a single Doppler radar data with lower elevation angles. Therefore, to identify a supercell, features, such as persistency, weak echo region (WER), bounded weak echo region (BWER), and/or hook echoes are used, in addition to the following mesoscale vortex identification criteria (Donaldson 1970): 1) A pair of maximum and minimum of Doppler velocities is detected on at least two adjacent elevation angles and two successive scans, 2) the angle between beam direction and a horizontal vector perpendicular to the vector pointing out the velocity maximum from the velocity minimum is less than 45°, 3) vertical vorticity is greater than 1.0 × 10⁻² s⁻¹, where the vertical vorticity was calculated by dividing the difference of the maximum and minimum Doppler velocities by their distance and multiplying by a factor of two. A vortex is assumed to be approximated by a Rankine combined vortex and its diameter is defined as a core diameter of the solid rotation, which is estimated by the distance between a Doppler velocity maxima couplet. Throughout this paper, the term “mesocyclone” (MC) will be used to denote either misocyclone or mesocyclone regardless of its diameter for simplicity (cf. Fujita 1981).

There is, however, a problem in the estimation of MC diameter and velocity difference. Brown and Lemon (1976) demonstrated that measurement of MC size and velocity maxima depend significantly upon the vortex size to the effective radar beam width. The estimated values of vertical vorticity presented in this paper will be underestimated because the diameter of MC was nearly comparable to the azimuthal resolution of the radar beam. This effect causes not only underestimation of vertical vorticity, but also a failure to detect a distant MC from the radar. Although this effect deteriorates the accuracy of the estimation, trend of the vertical vorticity of MC may be judged reliable because the movement of the MC corresponds smoothly with the effective beam width during a volume scan.

Figure 10 shows the path of each detected MC center within the supercells observed during a period from 1922 JST 19 September to 0100 JST 20 September. All of the supercell storms, labeled A–I, moved ∼N–NNE at average speeds of 16–19 m s⁻¹.

Storms A, C, D, and I showed little jumps in MC paths toward the east, likely indicating cyclic generation of MC cores (e.g., Brandes 1977; Burgess et al. 1982; Johnson et al. 1987). In fact, multiple cores were detected in storm I at 2343 JST. It is noted, however, that not all the storms observed during the period were supercells (see Fig. 4). The average movement vector of ordinary storms during the period is also denoted for reference in the upper-left corner of Fig. 10, with an absolute value of 25 m s⁻¹. The movement of the supercell storms was about 30% slower than that of the ordinary storms, but in nearly the same direction as that of the ordinary storms. The movement of supercell storms showed little deviation to the right of the density-weighted mean wind below 10 km AGL (see Fig. 6). A similar feature in the storm motion was also noted by McCaul (1987) for miniature supercells associated with hurricane Danny in 1985 and by Omoto (1982) for the tornado-producing storm associated with typhoon 7912.

c. Inagi storm and Mibu storm

Storms D and F, respectively, the parent storm of the Inagi and Mibu tornados, displayed similar life cycles. Both storms were first detected near the Ohshima Island (see Figs. 4b and 4e) as isolated cells at 1645 and 1953 JST by Tokyo radar, moving north-northeastward at average speeds of 18 m s⁻¹ and 17 m s⁻¹, respectively. As they moved to the north, their northern parts merged with the bandlike echo cluster over the western Kanto Plain while keeping their identities. They were tracked until 2100 and 2300 JST, respectively.

The maximum radar reflectivity cores of the storms D and F were located near their southern extents, exhibited elongated beanlike or even ribbonlike (Figs. 7, 9) shapes, and had reflectivity intensities of about 50 dBZ and 55 dBZ, respectively. The Haneda radar and the MRI Doppler radar several times identified a hook-shaped echo, a WER, and/or a BWER near at the southern end of the storms D and F. Echo top heights of both storms were approximately 10 km AGL. An MC was detected in the midlevels (2.1–4.0 km AGL) of the southeastern edge of the storm D at 1922 JST lasting until 2040 JST.

Figure 11 shows a PPI display of Doppler velocity and reflectivity fields of storm F with elevation angle of 1.4° at 2216 JST, nearest time of the Mibu tornado. A pair of local maximum (+42 m s⁻¹) and minimum (−1 m s⁻¹) of Doppler velocities is seen. The diameter of the MC was 1.5 km and the vertical vorticity 3.3 × 10⁻² s⁻¹. The reflectivity core of the storm was kidney-bean like in shape and a WER was located at the northern edge of the MC. The open triangle on the tornado path shows the tornado’s location at 2219 JST, verified by damage to a TEPCO power line. The solid triangle shows an estimated location of the MC center at the same time, assuming the translational speed of 17 m s⁻¹ to the north-northeast. The tornado appears to be located near the southwest edge of the MC core and was not coaxial.

One-minute averaged surface wind relative to the storm motion observed at ULMO were space-to-time converted and plotted in Fig. 11 by assuming the same translational speed and direction. These values clearly present a cyclonic circulation around the MC and the passage of a rear flank gust front (RFG). The temperature drop associated with the RFG was about 0.4 K. This RFG could not be identified in Doppler velocity fields around the time of Mibu tornado. The reason may be that Mibu town was too distant (approximately 54 km at this time) to observe the wind field near the surface due to the high beam altitude (approximately 0.66 km AGL) and the broad beamwidth.

Surface mesolows were detected in storms D and F through analysis of surface meteorological records. As storm F passed 2–3 km west of ULMO around 2222 JST, a pressure drop of 1 hPa over 10 min at 2220, a gust of 32.6 m s⁻¹, and a clockwise rotation of the wind direction were observed. This pressure drop is thought to have been caused by a mesolow associated with the MC. The diameter of the mesolow estimated from the pressure change was at least 11.8 km, assuming a transitional speed of 17 m s⁻¹ as the storm motion. The diameter of the mesolow at 2222 JST is about twice the diameter of the MC’s core (approximately 6 km), as estimated through the Doppler velocity field and assuming a Rankine combined vortex. The difference in the estimated diameters of the mesolow and the MC may be explained by a formation of a downshear mesolow, a feature often observed in supercell storms in the United States (cf. Charba and Sasaki 1971) or an extended surface low pressure beyond the MC core region. Around 2035 JST, prior to the detection of MC in storm F, a mesolow was also observed at Yokohama Local Meteorological Observatory (YLMO, see Fig. 2) located about 80 km south-southwest of the MRI radar. That no MC was detected by the MRI radar at this time suggests that the storm was too distant and the MC too small for adequate radar resolution due to the beam-broadening effect.

Similar meteorological records were observed at the CAMO (see Fig. 4d) around 1905 JST when the storm D passed over the observatory. A sharp pressure drop of 1.5 hPa over 7 min at 1859 JST, followed by a pressure rise of 2.3 hPa at 1906 JST, a temperature drop of about 0.5 K, anticlockwise rotation of the wind direction from 1858 to 1910 JST, and a gust of 12.3 m s⁻¹ at 1906 JST were observed. These changes infer the passage of a mesolow with a RFG east of CAMO.

Figure 12 is a time-height cross section of vertical vorticity and diameters of the MC in storm D estimated from the Doppler radar data assuming a Rankine combined vortex. As mentioned in 5a, the vertical vorticity presented in the figure may be somewhat underestimated because of the beam-broadening effect (Brown and Lemon 1976). Therefore, estimated vertical vorticity can be used as a minimum bound of the true vertical vorticity.

The region with vertical vorticity greater than 1.0 × 10⁻² s⁻¹ seems always confined to the levels below 4 km AGL. It reached the first maximum of 3.9 × 10⁻² s⁻¹ at 1.6 km AGL 15 min prior to the touch down of the tornado, decreasing slightly but then again strengthening, reaching the second maximum 3.3 × 10⁻² s⁻¹ right before touch down. It is noteworthy that, immediately prior to tornadogenesis, the vorticity was largest at the lowest elevation angle (at 600 m AGL) and decreased monotonically upward. A similar intensification of low-level circulation prior to tornado touch down was often observed in typical supercell storms (JDOP staff 1979). The MC in the storm weakened by 2229 JST, shortly before the storm moved out of the detection range of the Doppler radar.

d. Mini supercells

Excluding storms A, B, and C, most of the supercells observed during the night of 19 September had a similar morphology to that of typical supercell storms over the Great Plains. Common features included an MC detectable by a Doppler radar, a hook-shaped echo, a WER, and/or a BWER. Most MCs were located near the southeastern edge of reflectivity cores. Storms E–I were isolated cells or cells whose southern half (core regions) were surrounded by an echo-free region throughout most of their lifetime. Cyclic MC core generation was observed in some storms; and the average lifetime was 3 h.

Figures 13a,b represent an example of a typical PPI display of radar reflectivity (Fig. 13a) and Doppler velocity (Fig. 13b) for storms E and G at 2146 JST. The MCs, indicated by open circles, are found near the southeastern edge of the strong echo region. The RHI display of the reflectivity field (Fig. 13c) derived at 2153 JST displays a vaultlike weak echo region capped by a strong echo of 3 km AGL in storm G. Corresponding Doppler velocity field (Fig. 13d) shows strong radial convergence approximately 2 × 10⁻² s⁻¹ [∼40 m s⁻¹ (2 km)⁻¹] below 2 km AGL and divergence between 2 and 3.5 km AGL. Figure 13d indicates a strong updraft in the vault similar to those found in typical supercell storms (e.g., Browning 1964). If we assume axisymmetry of the divergence field around the updraft and the continuity equation of fluid, this vertical distribution of convergence and divergence indicates that upward velocity increases with height below 2 km, proceeds to a maximum value, and then decreases above. These observed vertical structures are almost identical to that of a simulated supercell in hurricane Danny’s environment by McCaul and Weisman (1996).

Average MC diameter ranged between 1 and 5 km, with most smaller than 4 km. Maximum vertical extent of Doppler radar–detected MCs were less than 5 km. According to Burgess et al. (1982, 1995), averaged diameters and depth of the traditional mesocyclone during mature stage are 6.0 km (midlevel) and 9.2 km, respectively. The values for the present mini supercells are almost identical to averages for WSR-88D-detected mini-supercell MCs examined by Burgess et al. (1995), but are significantly smaller than those of the MCs in typical supercells over the Great Plains. The reflectivity data (e.g., Fig. 13a) indicate that horizontal length is comparable but vertical dimension and width of these storms is much smaller than those of the typical supercell storms (e.g., schematic figures in Doswell and Burgess 1993). Some features found in the mini supercells on the night of 19 September, such as hook echo, longer life time, and smaller size compared to typical supercells, are reported by McCaul (1987) for supercellular storms that developed in Danny’s environment, though, no Doppler data was available.

Several MCs passed over or close to surface meteorological stations where temperature data were available. Temperature deficit of 0.6 K at most were measured during the passage of MCs, while that of 2 K was observed across the surface boundary.

The above features are summarized in Table 2, together with features of typical supercells and supercellular storms in Danny’s environment, and as schematic illustration in Fig. 14. These features demonstrate that, while their dimensions were small, the supercell storms over the Kanto Plain observed during the night of 19 September contained all the functions of typical supercells over the Great Plains.

a. Spatial scale of mesocyclones in the mini supercells

The confinement of vertical vorticity to the low-level is the one of the most remarkable feature of the MCs in the presented storms. This confinement is also demonstrated in the mini supercells generated in numerical simulations under Danny’s environment by McCaul (1993) and McCaul and Weisman (1996).

The reason for this confinement may be explained as follows. As shown in Fig. 6, vertical wind shear was very strong below 1 km AGL and relatively weak between 1 and 7 km AGL. As explained in many previous studies (e.g., Rotunno and Kemp 1985), vertical vorticity of a midlevel mesocyclone developed in a strong vertical wind shear is mainly generated through the tilting of ambient horizontal vorticity to vertical and the following amplification by stretching. As seen in section 5d, the vertical distribution of divergence in the mini supercells indicates strong convergence below 2 km and divergence aloft. Therefore, vertical vorticity generation by the tilting of ambient horizontal vorticity by the storm updraft would be fairly small above 2 km AGL. In the present case, generated vertical vorticity at 1–2 km AGL may have been advected upward and decreased with height above the 2 km by the stretching effect of a divergent updraft.

The small diameter of MCs in the mini supercells may be explained as follows. Since the midlevel MC is considered to be generated through tilting of horizontal vorticity in the environmental wind, the horizontal scale of the MC would nearly coincide with that of the updraft. In dry convection (i.e., Rayleigh-Benard convection), the horizontal dimension of an updraft tends to be scaled by the vertical dimension. Since the vertical scale of the mini supercell is considerably smaller than that of the typical supercell, the horizontal scales of the updraft and MC of the mini supercell are considered to be smaller than those of the typical supercell over the Great Plains.

b. Rightward deviation

As mentioned in section 5b, the movement of mini-supercell storms on the night showed little deviation to the right of the density-weighted mean wind below 10 km AGL (see Fig. 6). And mini-supercell storms and ordinary storms moved to nearly the same direction.

Rotunno and Kemp (1982) demonstrated that tilting of the horizontal vorticity of the vertically sheared basic flow by the storm updraft produces a midlevel vortex and that the vertical pressure gradient force produced by the vortex is responsible for generating a new updraft on the right-hand side with respect to the shear vector, resulting in a rightward deviation of the storm motion. Therefore, it may be appropriate to discuss the mini-supercell storm motion in reference to the mean shear vector between surface and 1–2 km AGL, where the vertical vorticity in the MC is generated through the tilting of the ambient horizontal vorticity. In this mean, a mini-supercell storm does move rightward relative to the mean wind shear vector.

In the above paragraph, rightward deviated motion of a mini-supercell storm motion relative to the mean wind shear was explained. Then, a question may arise on the finding that the ordinary storms also moved nearly the same direction as the mini-supercell storms. Powell (1990) found that cells in the outer rainbands of hurricane Floyd (1981) moved down the band, and the rainbands oriented along the density-weighted mean wind speed of the 0.2–6-km layer. In the present case, direction of the ordinary storms motion and of the density-weighted mean wind of the layer were the almost same. Therefore, this similarity in the storms’ motion between mini supercells and ordinary cells on this night is likely to be a coincidence.

c. Evaluation of vorticities in the surface boundary

As noted in section 5a, two of the stronger tornadoes, the Inagi and the Mibu, occurred when the storms were moving along the surface boundary. This suggests the possible contribution to the storms development of stretching of the preexisting vertical vorticity and the baroclinically generated horizontal vorticity.

First, the possibility of preexisting vertical vorticity amplification in an updraft is examined. The vertical vorticity equation for an inviscid incompressible homogeneous fluid is written as follows:

View Expanded

where ζ is the vertical vorticity, d/dt is the time derivative following the fluid parcel, ω_h is the horizontal component of vorticity vector ω, ∇_h is the horizontal derivative vector, and w is the vertical component of the velocity vector (u, υ, w). The first term on the right-hand side of Eq. (1) is the tilting term. The second term on the right-hand side is the stretching term.

After Kemp and Rotunno (1983), change over time of the vertical vorticity in an updraft with a constant convergence (denoted by w_z) and a constant tilting term (denoted by T) is evaluated. Under this assumption, one can easily integrate Eq. (1) to obtain (Kemp and Rotunno 1983),

View Expanded

where t₀ is the beginning time of the integration and ζ₀ is the vertical vorticity at t = t₀. As mentioned in section 4b, vertical vorticity in the surface boundary is on the order of 4 × 10⁻⁴ s⁻¹. Doppler radar observation (see Fig. 13d) found a strong radial convergence reaching on the order of 1 × 10⁻² s⁻¹. If this value is used as the w_z in Eq. (2), it takes 320 s for the vertical vorticity to be amplified to 1 × 10⁻² s⁻¹ solely by the convergent stretching. Evaluated vertical vorticity of the order of 10⁻² s⁻¹ seems still insufficient to explain the rapid amplification of low-level vertical vorticity before the tornado. However, these values are not so small as to be neglected. There is a possibility that additional vorticity generated by the above mechanism makes a contribution to the amplification of supercell development and resultant tornadogenesis as in a nonsupercell tornado development mechanism (Wakimoto and Wilson 1989).

Second, horizontal vorticity generated by baroclinic effect is roughly evaluated with a calculation after Kemp and Rotunno (1983). The estimated value is approximately a quarter of averaged environmental horizontal vorticity on the order of 2 × 10⁻³ s⁻¹ between the surface and 1 km (Fig. 6). It indicates that streamwise horizontal vorticity generated in the baroclinic zone along the forward flank gust front is less important in the case of our mini-supercell storms than in the case of typical supercell storms, even though this estimation may have errors because the data available were insufficient to describe temperature field in the storms.