a. The tropical cyclone environment

The tropical cyclone environment is very different from the typical tornadic environment. Shallow supercells are possible in tropical cyclones due to strong lower-tropospheric vertical shear, even though buoyancy is limited because ambient lapse rates are close to moist adiabatic (McCaul and Weisman 1996). McCaul (1991) described the favorable conditions for tornadoes in a tropical cyclone environment. Table 1 lists parameters and their thresholds that can be used to distinguish between a high tornado threat and a low tornado threat in a tropical cyclone environment.

Of the four tropical cyclones that produced tornadoes in central North Carolina in 2004, only one, Gaston, made direct landfall in the Carolinas. The other three made landfall either in Florida or recurved northeast after making landfall along the Gulf of Mexico coast. This placed central North Carolina in the most favorable region of a tropical cyclone for tornadoes, which is the northern to eastern quadrants between 0° and 120° with respect to the tropical cyclone’s motion. In this quadrant, 0–3-km shear and 0–1-km storm-relative helicity are usually highest (McCaul 1991). The storm environment associated with the four tropical cyclones that were investigated in this study exhibited values that were mainly in the high threat category in Table 1.

More recent studies have shown that a low lifted condensation level (LCL) and a small surface dewpoint depression indicate an environment that is favorable for tornadoes. Studies by Markowski et al. (2002) and Thompson et al. (2003) found that increased low-level relative humidity may contribute to increased buoyancy in the rear-flank downdraft, and an increased probability of tornadoes. Simulations of storm structure by McCaul and Cohen (2002) found that a lowered LCL reduced the strength of the surface outflow, prevented outflow dominance, and promoted storm persistence. The LCL height was found to be a strong discriminator between tornadic and nontornadic supercells by Rasmussen and Blanchard (1998). LCL heights below 800 m were associated with tornadic environments 50% of the time in that study. According to Markowski et al. (2002), the mean surface dewpoint depression in tornadic environments was 10°F (5.6 K). In a tropical cyclone environment, the boundary layer relative humidity is typically high, and thus the LCL height and surface dewpoint depressions are typically low. These conditions, along with strong low-level wind shear and sufficient buoyancy, will create an environment favorable for tornadogenesis. It is uncertain whether the critical values of LCL height and surface dewpoint depression in the Great Plains studies by Rasmussen and Blanchard (1998) and Markowski et al. (2002) will differ significantly in the tropical cyclone environment; however, this is beyond the scope of this study and should be addressed in future research. LCL height and surface dewpoint depression could be added to the research by McCaul (1991) to further aid in distinguishing between a low tornado threat and a high tornado threat in the tropical cyclone environment.

Midlevel dry intrusions have been recognized as having an important impact on the tropical cyclone environment. Midlevel dry intrusions can increase convective instability by increasing the environmental lapse rate, enhancing CAPE and surface-based instability, and enhancing evaporative cooling within the rear-flank downdraft (Vescio et al. 1996). The lack of midlevel clouds due to the dry-air intrusion tends to increase diurnal heating and thus surface-based destabilization. Curtis (2004) examined 13 historical tornado outbreaks associated with tropical cyclones. A midlevel dry intrusion was identified with 11 of the outbreaks. The study found that tropical cyclones associated with tornado outbreaks and midlevel dry intrusions had a lower average LCL, had more moisture in the layer from just above the surface through 900 mb, and were much drier at heights above 700 mb than tropical cyclones that did not produce outbreaks. Dry-air intrusions appeared to have played a significant role in the 2004 tropical cyclones that affected central North Carolina. Because three of the four tornado-producing tropical cyclones made landfall along the Gulf of Mexico coast, there was ample time for a drier continental air mass to become entrained into the tropical cyclones before reaching North Carolina. This entrainment of dry air into the tropical cyclone reduced the cloud cover and allowed for greater diurnal heating and surface-based destabilization. Analysis of satellite and upper-air data indicates that all four of the tropical cyclones examined in this study exhibited some degree of dry-air intrusion. Figure 1 is a water vapor image as the remnants of Hurricane Ivan crossed the Raleigh CWA, with 500-mb upper-air data plotted. This image was taken around 1200 UTC 17 September 2004, about 3 h before the time of a tornado touchdown in the northwest corner of the Raleigh CWA. A line of convection developed on the eastern edge of the dry intrusion, and several storms within this line produced tornadoes or straight-line wind damage as they crossed the Raleigh CWA. This dry intrusion pattern was also evident in the other three tropical cyclones that were examined in this study. The tropical cyclones in which the dry-air intrusions were most pronounced (Ivan and Jeanne) coincided with the most severe occurrences of damage.

Figure 2 is a skew T plot of temperature, dewpoint temperature, and winds at 1800 UTC 17 September 2004 during Ivan at Greensboro, North Carolina (GSO). The features of this sounding are very similar to the features noted by Curtis (2004) that distinguish a tropical cyclone environment that is favorable for tornadoes. The sounding from Ivan features a low LCL, saturation in the layer from just above the surface through 900 mb, and dry air at heights above 700 mb. This sounding also shows winds rapidly veering just above the surface, creating a favorable low-level shear profile for tornado formation.

Curtis (2004) also found a strong diurnal signal in the temporal distribution of the tornadoes, with 65% of the tornadoes occurring in the daylight hours. In the current study, all but one of the tornadoes examined occurred during the daylight or early evening hours between 1500 and 0000 UTC. The lack of cloud cover due to the dry intrusions along with surface heating during the day likely increased the CAPE, contributing to an enhanced tornado threat over central North Carolina during the events in this study.

The presence of low-level boundaries has been recognized as an important contributor to tornadogenesis. Although the tropical cyclone environment is often barotropic, tropical cyclones that have made landfall or are undergoing an extratropical transition can contain temperature and moisture discontinuities or wind convergence zones. These boundaries can locally enhance storm-relative helicity, creating a favored area for tornadoes to develop. McCaul et al. (2004) described the importance of a low-level boundary in the tornado outbreak as the remnants of Tropical Storm Beryl crossed South Carolina. In Beryl, most of the tornadoes occurred near or just northwest of a boundary, in slightly cooler surface air with backed surface winds. The presence of boundaries also played an important role in the development of several tornadoes in the current study. A boundary was present across the southern portion of the Raleigh CWA as the remnants of Hurricane Jeanne passed through the area. Numerous cells were translating from south to north, and as they encountered the east–west-oriented boundary, they produced several tornadoes. Boundaries were also present to some extent in Frances, Gaston, and Ivan as well. Knowledge of low-level boundaries can enhance the situational awareness of the warning forecaster, and potentially increase warning lead time and accuracy.

b. Structural characteristics of tropical cyclone supercells

McCaul and Weisman (1996) numerically simulated supercells within tropical cyclone environments. The simulations revealed that even though buoyancy is limited due to lapse rates that are nearly moist adiabatic, shallow supercells are still possible. These shallow supercells may contain updrafts that rival or exceed the intensity of Great Plains supercells. The simulated supercells in the study were low topped, exhibiting a reduction of horizontal and vertical size when compared with the Great Plains supercells. McCaul and Weisman (1996) found that even though parcel buoyancy is often small in tropical cyclone environments, the concentration of the buoyancy in the strongly sheared lower troposphere allows for the development of a perturbation pressure minimum that is similar to those seen in simulated Great Plains supercells. The upward dynamic pressure gradient force contributes much more [at least three times as much according to McCaul and Weisman (1996)] to the updraft speed than does buoyancy in a tropical cyclone–induced supercell. The study also found that peak updraft intensity generally occurs at low levels, usually around 6600 ft (2 km) above the surface.

In the 2004 tropical cyclones that are examined in the current study, the enhanced destabilization from the midlevel dry-air intrusion likely resulted in a greater contribution of buoyancy to the updraft strength than would have occurred without the dry-air intrusion.

c. Radar characteristics of tropical cyclone tornadoes

Previous studies have investigated the characteristics of tropical cyclone tornadoes using Weather Surveillance Radar-1988 Doppler (WSR-88D) data. McCaul et al. (2004) analyzed a severe outbreak of tornadoes associated with Tropical Storm Beryl (1994). The study observed long-lived supercells that contained well-defined mesocyclones in the velocity field, along with hook echoes on their southeastern flanks. The dimensions of the mesocyclones analyzed in that study were considerably smaller than tornadic supercells in the Great Plains. Characteristic rotational velocities in the study were in the 19–29-kt (10–15 m s⁻¹) range, occasionally reaching 39–49 kt (20–25 m s⁻¹). However, the authors acknowledge that the supercells analyzed within Beryl were probably near the top of the spectrum of tropical cyclone convective cell intensity. Given the smaller dimensions and lower rotational velocities of mesocyclones in the tropical cyclone environment, the techniques that are used to warn for non–tropical cyclone tornadoes may not apply in the tropical cyclone environment. Thresholds of rotational velocity would have to be lowered, and reflectivity signatures would be more subtle. In addition, radar sampling limitations may also lead to lower radial velocity values. Reduced radar resolution due to the smaller storm dimensions may also contribute to lower apparent radial velocity values, especially at far distances from the radar.

In the current study, this was indeed found to be the case with most of the tornadic cells that were analyzed. Rotational velocity values were lower and reflectivity signatures were less distinct when compared with non–tropical cyclone tornado signatures. In the current study, there were a few long-lived supercells that produced multiple tornadoes, but this was the exception rather than the rule. Most of the tornadic storms observed in central North Carolina in 2004 had shallow rotation, often observed only in the lowest one or two radar elevation angles, and usually extending no higher than 15 000 ft (4.5 km). A deep persistent mesocyclone often did not precede tornado touchdown. Most of the tornadoes were very short lived with rotation signatures often lasting only two to four volume scans (about 10–20 min), much shorter than those observed in McCaul et al. (2004). However, the rotational velocity values typical of tornadic supercells in McCaul et al. (2004) [19–29 kt (10–15 m s⁻¹)] were comparable to those found in this study.

Spratt et al. (1997) investigated radar signatures from several tornadic storms that occurred during Tropical Storm Gordon (1994) and Hurricane Allison (1995). Although reflectivity features such as inflow notches and hooks were more subtle, these traditional severe weather indicators were occasionally observed during several of the Gordon and Allison tornado events. Values of rotational velocity that were observed in mesocyclones in that study were in the 13–23-kt (7–12 m s⁻¹) range. This threshold value agrees closely with the findings of McCaul et al. (2004), as well as with the findings of the current study. Spratt et al. (1997) also analyzed the echo-top product, and found that trends of decreasing echo tops corresponded with a time period when the depths of the rotational velocities were determined to be increasing. The current study also examined the potential to use echo tops to predict the onset of tornadogenesis. Typical echo-top values analyzed with the storms that produced tornadoes were between 25 000 and 45 000 ft (7.6 and 13.7 km). However, with most of the storms there appeared to be no discernable pattern to the echo tops before, during, or after tornado touchdown.

a. Rotational velocity

All 12 cells that produced tornadoes had parent circulations with at least 15-kt (7.7 m s⁻¹) rotational velocity and less than 3 n mi diameter calculated at the 0.5° elevation angle. The average lead time from the first appearance of this threshold to the tornado touchdown was 28.5 min. However, there were eight cells that had a Vr of 15 kt (7.7 m s⁻¹) but did not produce a tornado. Using this threshold alone would produce a high probability of detection, but also a high rate of false alarms.

Raising the Vr to 20 kt (10.3 m s⁻¹) with less than 3 n mi diameter resulted in an average lead time of 25 min, with 10 detections, six false alarms, and two missed events when viewing the storms from the KRAX radar. It is important to examine these two missed events more closely. One of the missed tornadoes occurred in Anson County, which is located in the far southwest corner of the Raleigh CWA (cell 5 in Table 2). At a distance of 90 n mi (167 km), the lowest that KRAX can sample a cell is around 10 000 ft (3.1 km) AGL. The maximum computed Vr from KRAX with this cell before tornado touchdown was 15 kt (7.7 m s⁻¹). The radar that was able to sample the cell the lowest was in Columbia, South Carolina, around 7500 ft (2.3 km) AGL. From this radar, the Vr was calculated to be 25 kt (12.9 m s⁻¹). In this particular tornado event, it is clear that data from neighboring radars are of critical importance to obtaining the best velocity sampling possible. Other cells that were analyzed had circulations that were confined mainly to the lowest levels of the storms, usually extending no higher than 15 000 ft (4.5 km).

The second missed event was a tornado that touched down in Harnett County, about 32 n mi (60 km) southwest of KRAX (cell 2 in Table 2). The maximum Vr that was computed with this cell before the tornado touchdown was 19 kt (9.8 m s⁻¹). Figure 3 shows the 0.5° storm-relative velocity image about 10 min before tornado touchdown. Range folding in the velocity data may also have influenced the Vr calculation. However, this cell exhibited a distinct hook signature in the reflectivity data 10 min before the tornado event. Figure 4 shows the reflectivity image at the same time as the storm-relative velocity image in Fig. 3. This event stresses the need to consider multiple datasets in the warning decision-making process, and not focus on one threshold that must be met before issuing a warning. Both storm-relative velocity and reflectivity data must be examined closely to make a fully informed warning decision.

A Vr threshold of 25 kt (12.9 m s⁻¹) or greater was observed with seven tornado events. There was one cell that met or exceeded this threshold but did not produce tornado, while five tornadoes occurred below this threshold. While this threshold produced a low false alarm rate, the probability of detection was unacceptably low.

There were five storms analyzed that exhibited a Vr of 15 kt (7.7 m s⁻¹) or higher, but their rotation was broad, greater than 3 n mi (5.5 km) diameter. These storms are not listed in Table 2. When this was observed, no tornado occurred. Only when the rotation tightened to less than 3 n mi (5.5 km) diameter did a tornado touchdown follow. This is similar to the findings of Spratt et al. (1997), where a tightening of the rotation signature was found to be a precursor of tornadogenesis.

b. Velocity signatures at higher elevation angles

Of the 15 tornado events reviewed, only 3 showed rotation at or above the 1.5° elevation angle before the tornado touchdown. Strong and deep mesocyclones were rarely observed in these storms. When rotation was observed aloft, it was most often broad, with the inbound and outbound velocities separated by 3 n mi (5.5 km) or greater. However, there was another signature noted at higher-elevation angles that preceded 14 tornado events. This signature will be referred to as a velocity enhancement signature (VES). This signature appeared as a small area of enhanced radial velocity of 30 kt (15.4 m s⁻¹) or greater. It was observed between 7000 (2.1 km) and 14 000 ft (4.2 km) AGL. The average lead time from the first appearance of a VES to the first tornado touchdown was 27 min, with the range of lead times from 10 to 49 min. In all of its occurrences, the VES was located above the low-level inflow region of the storm. When a VES was observed along with a hook and a low-level rotation signature, these features were vertically collocated; the VES was directly above the reflectivity hook and the low-level rotation.

There was only one tornado event that was not preceded by a VES. Thus, the VES offers a high probability of detection. There were 5 out of 12 cells that did have a VES but that did not produce tornadoes. This may suggest that the VES may also have a high false alarm rate.

Figure 5 is a storm-relative velocity image at 1.5° from cell 15 in Table 2, demonstrating the appearance of a VES. In this case, the VES appears as a small area of inbound velocities exceeding 30 kt (15.4 m s⁻¹) in northwest Hoke County. The height of this VES is about 12 500 ft (3.8 km) AGL. Figure 6 is another example of a VES, taken from the same storm at the same time as Figs. 3 and 4 (cell 2 in Table 2). In Fig. 6, the height of the VES is about 10 000 ft (3 km) AGL. A comparison of the location of the VES to the rotation in Fig. 3 and the hook signature in Fig. 4 shows that the VES is vertically collocated with these features.

The physical mechanisms within the cell that cause a VES are unclear, and should be the topic of future research. We speculate that the VES is related to the formation of the rear-flank downdraft due to its location relative to the mesocyclone and other storm signatures. It may be a storm-scale inflow jet, which is being driven by shear-induced pressure perturbations. This inflow jet may lead to or enhance the decent of the rear-flank downdraft. Whatever its cause, it is apparent that the presence of a VES can be a strong indicator of tornadogenesis, especially when observed in conjunction with reflectivity signatures and rotational velocity.

c. Reflectivity signatures

As was observed in the study by Spratt et al. (1997), the reflectivity signatures were more subtle than in non–tropical cyclone tornadoes, but they were discernable. On a few occasions, hook signatures were quite distinct, as was the case with the cell in Fig. 4. In the current study, reflectivity signatures were found to be a reliable precursor to tornado touchdown. The average lead time from the appearance of a hook or appendage to tornado touchdown was about 18 min. Of the 11 times a hook or appendage signature was present in the reflectivity data, 9 were followed by a tornado touchdown. Thus, these reflectivity signatures gave a low rate of false alarms. There were three tornadoes that touched down that were not associated with a discernable hook or appendage signature (cells 3, 5, and 11 in Table 2). An explanation for two of these missed events may be due to the far distance from the radar that the tornadoes occurred. Cells 5 and 11 occurred about 90 n mi (167 km) from the KRAX radar, and it is likely that the radar was sampling the storms well above where a hook or appendage could be seen. Cell 3 was located about 48 n mi (88.5 km), and had a maximum rotational velocity of 26.5 kt and a VES. Relying on reflectivity signatures alone would likely result in several missed tornado events, especially at far distances from the radar. The best warning decision is made when multiple signatures are considered.