1
Tornado Outbreaks Associated with Landfalling Hurricanes in the North
Atlantic Basin: 1954â2004
STEPHANIE M. VERBOUT
1
, DAVID M. SCHULTZ
2,3
, LANCE M. LESLIE
1
,
HAROLD E. BROOKS
2
, DAVID KAROLY
1
,
AND
KIMBERLY L. ELMORE
2,3
1
School of Meteorology, University of Oklahoma, Norman, Oklahoma
2
NOAA/National Severe Storms Laboratory, Norman, Oklahoma
3
Cooperative Institute for Mesoscale Meteorological Studies, University of Oklahoma, Norman,
Oklahoma
Submitted as an article for the Meteorology and Atmospheric Physics Special Issue on Tropical
Cyclones
Submitted 10 June 2006, Revised 14 August 2006
Corresponding author address
: Stephanie M. Verbout, University of Oklahoma, School of
Meteorology/National Weather Center, 120 David L. Boren Blvd, Norman, OK 73072
E-mail: stephanieverbout@hotmail.com
2
ABSTRACT
Tornadoes are a notable potential hazard associated with landfalling hurricanes. The
purpose of this paper is to discriminate hurricanes that produce numerous tornadoes (tornado
outbreaks) from those that do not (nonoutbreaks). The data consists of all hurricane landfalls
that affected the United States from the North Atlantic basin from 1954 to 2004 and the United
States tornado record over the same period. Because of the more than twofold increase in the
number of reported tornadoes over these 51 years, a simple least-squares linear regression (âthe
expected number of tornadoesâ) was fit to the annual number of tornado reports to represent a
baseline for comparison.
The hurricanes were sorted into three categories. The first category, outbreak hurricanes,
was determined by hurricanes associated with the number of tornado reports exceeding a
threshold of 1.5% of the annual expected number of tornadoes and at least 8 F1 and greater
tornadoes during the time of landfall (from outer rainbands reaching shore to dissipation of the
system). Eighteen hurricane landfalls were classified as outbreak hurricanes. Second, 37
hurricanes having less than 0.5% of the annual expected number of tornadoes were classified as
nonoutbreak landfalls. Finally, 28 hurricanes that were neither outbreak nor nonoutbreak
hurricanes were classified as midclass hurricane landfalls.
Stronger hurricanes are more likely to produce tornado outbreaks than weaker hurricanes.
While 78% of outbreak hurricanes were category 2 or greater at landfall, only 32% of
nonoutbreak hurricanes were category 2 or greater at landfall. Hurricanes that made landfall
along the southern coast of the United States and recurved northeastward were more likely to
produce tornadoes than those that made landfall along the east coast or those that made landfall
along the southern coast but did not recurve. Recurvature was associated with a 500-hPa trough
3
in the jet stream, which also contributed to increased deep-layer shear through the hurricane,
favoring mesocyclogenesis, and increased the low-level shear, favoring tornadogenesis. The
origin of the hurricane, date of landfall, and El NiñoâSouthern Oscillation phase do not appear to
be factors in outbreak hurricane creation. The results of this study help clarify inconsistencies in
the previous literature regarding tornado occurrences in landfalling hurricanes.
4
1. Introduction
Wind damage, storm surges, and inland flooding are hazards faced by coastal communities
in the southern and eastern United States due to landfalling hurricanes. Neumann et al. (1999)
found that the United States Atlantic and Gulf coasts experience an average of 1.7 landfalling, or
near-landfalling, hurricanes per year. Hurricane-related property losses accounted for nearly
40% of all insured losses in the United States from 1984 to 1993, exceeding earthquake losses by
a factor of four (Pielke and Pielke 1997).
A hurricane with watches or warnings in effect is considered a land-threatening hurricane
(Franklin et al. 2003). Tornadoes are a notable potential hazard associated with land-threatening
hurricanes. Ten percent of hurricane-related fatalities from 1948 to 1972 were attributed to
tornadoes near or during the time of landfall (Novlan and Gray 1974). Hurricane Allenâs (1980)
landfall in Texas produced well over $70 million in damage associated with hurricane-spawned
tornadoes alone (Gentry 1983). Other examples of tornadoes associated with tropical systems are
described in Gray (1919), Hills (1929), Tannehill (1938, 24â25), Malkin and Galway (1953),
Sadowski (1962), Rudd (1964), Pearson and Sadowski (1965), Orton (1970), Hoadley (1981),
Stiegler and Fujita (1982), McCaul (1987), Grazulis (1993, 124â127), McCaul et al. (2004), and
Watson et al. (2005).
Emergency managers typically treat landfalling hurricanes differently than tornadoes.
When a hurricane threatens landfall, coastal communities are evacuated inland to large buildings
with free-spanning roofs that can house many people, such as school gymnasiums and armories.
On the other hand, these types of buildings are more susceptible to collapse in a tornado. This
quandary proves problematic for organizations such as the Red Cross and the Federal Emergency
Management Agency, who work to shelter those evacuated from coastal regions.
5
Most tornadoes generated by hurricanes occur within 24 h of landfall (e.g., Hill et al. 1966;
Novlan and Gray 1974; Gentry 1983; McCaul 1991). Although some tornadoes can occur within
the inner rainbands or in the eyewall (âcoreâ tornadoes; Gentry 1983; Weiss 1987; McCaul
1991), most tornadoes form in the outer rainbands where convection can be strong (e.g., Hill et
al. 1966; Novlan and Gray 1974). For example, Weiss (1987) found that 74% of postlandfall
tornadoes from 1964 to 1983 were generated by outer-rainband convection.
Tornadoes spawned by hurricanes can have different characteristics than tornadoes
spawned by midlatitude supercells or squall lines. For instance, the parent mesocyclones in
hurricanes tend to have shallower circulations than those associated with midlatitude supercells
in the central United States (average depth of 3â3.5 km vs 6 km, respectively) (McCaul 1987,
1991; McCaul and Weisman 1996; Spratt et al. 1997; McCaul et al. 2004). McCaul and
Weisman (1996) argued that the lack of strong surface cold pools (a source of horizontal
vorticity) associated with supercells in the moist tropical environment may limit the production
of strong tornadoes by tilting. Traditional tornadic radar signatures (e.g., bounded weak echo
regions, hook echoes, and appendages) may be subtle, or even nonexistent for hurricane-
spawned tornadoes (Spratt et al. 1997). In addition, some of the tornado-producing supercells
during Tropical Storm Beryl (1994) did not contain any cloud-to-ground lightning (McCaul et al.
2004). Finally, hurricane-related tornadoes tend to be weaker, and have half the path width and
path length as tornadoes formed by other systems (Smith 1965).
McCaul (1991) examined the number of tornadoes and thermodynamic properties
associated with all tropical cyclones (defined as all hurricanes, tropical storms, tropical
depressions, and subtropical storms) from 1948 to 1986, grouping tornado outbreaks into three
classifications:
minor
(not more than 8 tornadoes),
major
(more than 8 tornadoes), and
severe
6
(more than 24 tornadoes). He found that 413 of the 626 total tornadoes (66%) were produced by
18 major tornado outbreaks, of which four were severe.
One of the most prominent tornadic hurricanes was Hurricane Beulah (1967) (Orton 1970).
Beulah made landfall near Brownsville, Texas, and produced 117 tornadoes in the southern
portion of the state (Fig. 1). The majority of tornadoes occurred on 20 September when 67
tornadoes were reported with Beulahâs landfall (Orton 1970). Of those 67 tornadoes, 21 were
rated F1 and greater on the Fujita scale. Almost 40 years passed since any hurricane produced
nearly 100 tornadoes at landfall. Data from the Storm Prediction Center (SPC) showed that
Hurricane Frances (2004) produced 99 tornadoes and Hurricane Ivan (2004) produced 117
tornadoes (e.g., Watson et al. 2005). The large number of tornadoes associated with Beulah
(1967), Frances (2004), and Ivan (2004) clearly demonstrates the need to examine and
characterize these types of events for future hurricane preparedness.
The purpose of this paper is to examine the differences between outbreak and nonoutbreak
hurricane landfalls. Section 2 describes the dataset used in this study. Section 3 discusses our
criteria for identifying days with many tornadoes, as determined by the method in Verbout et al.
(2005). Section 4 outlines the three types of hurricanes identified by this study. Characteristics
such as intensity of the hurricane, landfall location, synoptic pattern, hurricane origin, date of
landfall, and El NiñoâSouthern Oscillation (ENSO) phase are analyzed in section 5.
Lastly,
section 6 summarizes the conditions favoring tornado outbreaks associated with landfalling
hurricanes.
2. Data
7
A main concern surrounding any tornadic event is the accuracy and availability of reports
(e.g., Pearson 1975; Doswell and Burgess 1988; Grazulis 1993, 187â195; Hagemeyer 1998;
Hagemeyer and Spratt 2002). This problem is amplified because hurricane-force winds, heavy
rains, and storm surge can obscure tornado verification. The present study inspected all tornado
reports from the United States tornado database (McCarthy 2003) from 1954 to 2004 two days
prior and four days following the landfall of a hurricane, similar to the criteria chosen in Weiss
(1987) and Fig. 15 in McCaul (1991). Because Weiss (1985) found 85% of tornadoes associated
with tropical systems were attributed to hurricane-strength storms, we chose to focus strictly on
hurricanes (category 1 or greater on the SaffirâSimpson scale). All landfalling hurricanes that
affected the United States originating from the North Atlantic basin from 1954 to 2004 were
identified from the National Hurricane Center/Tropical Prediction Center (NHC/TPC) archives.
These hurricanes included not only those that made landfall in the United States, but also those
that made landfall in northeastern Mexico and produced tornadoes in the United States.
Tornadoes were considered related to the hurricane if the tornado report occurred within 400 km
of the cyclone center. This distance is consistent with Spratt et al.âs (1997) definition of
maximum range. This definition may need to be expanded for future investigations to account
for Hurricane Katrina (2005) because of the extremely large, unparalleled size of Katrinaâs outer
rainbands. At the time of this study, the 2005 hurricane season was ongoing and official
hurricane and tornado data had not been verified.
3. Defining a big tornado day
As stated earlier, the large number of tornadoes associated with Hurricanes Beulah (1967),
Frances (2004), and Ivan (2004) clearly demonstrates the need to examine these types of events
8
for future hurricane preparedness. With the aim of studying tornado outbreaks associated with
landfalling hurricanes, we first examined the daily United States tornado record from 1954 to
2004. Verbout et al. (2005) defined a âbig tornado dayâ as a single day where numerous
tornadoes and/or many tornadoes exceeding a specified intensity threshold were reported. This
term was chosen with the intention of distinguishing a big tornado day from previous
applications of the word outbreak. Overall, the total number of tornadoes in the United States
tornado database (McCarthy 2003) had doubled from roughly 600 per year in the 1950s to
around 1200 per year in the 2000s (Fig. 2). A least-squares linear regression was fit to the
annual number of reported tornadoes over the period 1954â2004 in order to adjust for the general
increase in reports (Verbout et al. 2005). Such a procedure allows comparison of the tornado
record through the decades. Traditionally, the number of F2 and greater tornadoes has been used
to determine the significance of a tornado event. However, Verbout et al. (2005) demonstrated
that over this time period, the F2 and greater series had some overrating problems early in the
dataset. Therefore,
the F1 and greater time series is more stationary over this period in
comparison with the F2 and greater series.
Thus, a big tornado day is determined by a fraction
of the annual expected number of tornadoes associated with the linear regression and/or a
minimum number of F1 and greater tornado reports (Verbout et al. 2005).
In order to assess the daily probability of tornado occurrence in the United States, Verbout
et al. (2005) constructed a statistical model of the daily mean number of tornadoes reported from
1954 to 2004 using a kernel density estimation technique with a Gaussian smoother (
Ï
t
= 15
days) (Brooks et al. 2003). This technique was used to smooth the data in time and space and
estimate the probability of any tornado occurring anywhere in the United States on any given day
(Fig. 3; dashed line). The peak probability of 90% chance of any tornado occurring in the United
9
States is near June 12 (Brooks et al. 2003). However, once the data is constrained by thresholds
with the purpose of determining a big tornado day (for example, 1.5% of the linear regression
value and at least 8 F1 and greater tornadoes), the probability distribution changes. The black
curve in Fig. 3 illustrates the probability of these days from 1954â2004 and shows that the
largest peak occurs in mid-May, nearly three weeks earlier in the year than the any-tornado
curve. Hence, big tornado days are more likely to occur slightly earlier in the year than just any
day with a tornado. [See section 3e in Verbout et al. (2005) for further discussion of this point.]
Tornadoes embedded within hurricanes can occur within a one-day period (e.g. Hurricane
Edith 1971) to over a three-day period (e.g. Hurricane Frances 2004). Because a landfalling
hurricane can spawn tornadoes over several days, this paper considers the number of tornadoes
over the duration of landfall (from outer rainbands reaching shore to dissipation of the system) as
a natural temporal limit.
4. Hurricane classifications
In order to determine if an outbreak hurricane occurred, thresholds must again applied to
the tornado data. As again discussed in Verbout et al. (2005), thresholds must be chosen
arbitrarily depending on the number of events one wishes to analyze.
For this study, a threshold
of 1.5% of the linear regression value and at least 8 F1 and greater tornadoes
was chosen to
distinguish between outbreak and nonoutbreak hurricanes in order to create a sizeable outbreak
dataset. As a result, a total of 83 hurricane landfalls were classified as
outbreak
(18 cases),
nonoutbreak
(37 cases), or
midclass
(28 cases) hurricanes.
4.1 Outbreak hurricanes
10
Hurricanes with tornado reports exceeding a threshold of 1.5% of the annual expected
number and at least 8 F1 and greater tornadoes were classified as
outbreak
hurricanes. For
instance, the 1.5% annual expected value for 1954 was 8; therefore, at least 8 tornadoes were
needed for an outbreak (Fig. 4). For 2003, however, the 1.5% annual expected value was 18
(Fig. 4). These thresholds generated 18 outbreak hurricane landfalls (Table 1). When the 18
outbreak hurricane cases were plotted in Fig. 3 (again using the kernel density estimation
technique; gray curve), the peak occurs in early to mid-September (most active hurricane time in
the North Atlantic basin). These results show that the outbreak hurricane tornadoes peak
probability (gray curve; ~0.005) comprises nearly all the daily big tornado day signal (1.5% and
8 F1+ black curve; ~0.006) for early to mid-September. Thus, if a big tornado day occurred in
early to mid September, chances are it occurred within an outbreak hurricane.
Eleven of the 18 cases defined by McCaul (1991) are included in this study, while seven
additional storms since 1986 have been added to this dataset (Gilbert 1988, Andrew 1992, Opal
1995, Georges 1998, Frances 2004, Ivan 2004, and Jeanne 2004). Hurricane Alicia (1983) was
not considered an outbreak hurricane in our study because only two of the total 22 reported
tornadoes were rated F1 or greater. Similarly, Curtis (2004) created a dataset of 13 outbreak
hurricanes from 1960 to 1999 based on 20 or more reported tornadoes (Galway 1975, 1977).
Ten of the 13 cases in Curtisâs (2004) dataset appeared in our datasetâ the excluded three were
Hurricane Alicia (1983), described above, and Tropical Storms Beryl (1994) and Josephine
(1996). Beryl (1994) and Josephine (1996) were omitted from this dataset because of the strict
requirement of the storm needing to be of hurricane strength (according to SaffirâSimpson scale)
at some point in its lifetime. Additionally, Curtisâs (2004) dataset did not include Hurricanes
Audrey (1957), Cleo (1964), Hilda (1964), and Babe (1977) because of his higher threshold that
11
required at least 20 tornadoes. Finally, 10 of the 13 (77%) outbreaks identified by Curtis (2004)
occurred in the last 25 years (1979 to 2004), indicating that the temporal increase in reports (Fig.
2) was perhaps reflected in his dataset.
4.2 Nonoutbreak hurricanes
Hurricanes with less than 0.5% of the annual expected number of tornado reports were
classified as
nonoutbreak
hurricane landfalls (Table 2). For example, no more than three
tornadoes could have occurred in 1954, and no more than six in 2004 (Fig. 4). From 1954 to
2004, there were 37 cases of nonoutbreak hurricane landfalls. Several of these cases include
hurricanes that made multiple landfalls and either had no reported tornadoes or very few (Table
2). Note that five nonoutbreak cases (Diane 1955, Ione 1955, Cindy 1959, Gloria 1985, Charley
1986) narrowly made landfall and quickly turned back out to sea, allowing little time for the
hurricane to produce tornadoes over land (Table 2). Most recently, 1999 had three nonoutbreak
hurricane landfalls (Bret, Dennis, and Irene). Bret (1999) was the first hurricane to make landfall
on the Texas coastline since Hurricanes Chantal and Jerry (1989) and the first intense hurricane
(category 3 or greater on SaffirâSimpson scale) to hit Texas since Hurricane Alicia (1983).
4.3
Midclass hurricanes
Many landfalling hurricanes produced several tornadoes but could not be labeled strictly
outbreak or nonoutbreak. Those cases that fell between the two thresholds were classified as
midclass
hurricanes. There were 28 midclass hurricanes identified from 1954 to 2004 (Table 3).
A few of these midclass hurricanes failed to make outbreak classification because of too few F1
and greater tornadoes reported. For example, Hurricanes Alicia (1983) and Lili (2002) both had
12
the minimum number of tornadoes needed (22 and 26, respectively), yet not enough F1 and
greater tornadoes were reported (2 and 5, respectively). On the other hand, Hurricanes Isbell
(1964) and Juan (1985) did have 8 F1 and greater reported tornadoes, but the 9 and 12 total
tornadoes (respectively), however, fell short of the 1.5% criterion (Fig. 4). Another midclass
case is Hurricane Gracie (1959). Gracie only had six reported tornadoes associated with landfall,
but three of the six tornadoes were rated F3.
5. Discriminating between outbreak and nonoutbreak hurricanes
Of the 83 hurricane landfalls in this study (Tables 1â3), only 14 (17%) had no reported
tornadoes. Of those 14, only four have occurred since 1973, suggesting that early tornado
records, especially for weaker tornadoes, may be incomplete (e.g., Smith 1965; Hill et al. 1966;
Gentry 1983; Hagemeyer 1998; Verbout et al. 2005). Thus, the threat of tornadoes faces
forecasters with nearly every hurricane. What determines the conditions for especially prolific
tornado-producing hurricanes when tornado watches may be required? In this section, we
examine various characteristics that may prove useful in discriminating outbreak cases from
nonoutbreak cases. These include the intensity of the hurricane, landfall location, synoptic
pattern, hurricane origin, date of landfall, and ENSO phase.
5.1 Hurricane intensity
Previous research has indicated that stronger hurricanes are more likely to produce
tornadoes than weaker ones (Hill et al. 1966; Novlan and Gray 1974; Gentry 1983; Weiss 1985;
McCaul 1991). In this study, the SaffirâSimpson scale, a measure of the maximum near-surface
wind speed, was used as a measure of hurricane intensity. The SaffirâSimpson scale ranges from
13
1 to 5, with 5 having the strongest winds. A category 0 was included in this study to account for
sub hurricane-force winds. Most hurricanes are on the low end of the SaffirâSimpson scale. For
example, whereas 34 of 83 hurricane landfalls were category 1 strength at landfall, only two
hurricanes were category 5 at landfall: Hurricane Camille (1969; nonoutbreak case) and
Hurricane Andrew (Florida 1992 landfall; nonoutbreak case) recently upgraded to a category 5
(Landsea et al. 2004).
Figure 5 shows the distribution of SaffirâSimpson category of each classification of
hurricane (outbreak, nonoutbreak, or midclass) at landfall. Forty hurricanes were of at least
category 2 at landfall and 43 were rated categories 1 or 0. Thirty-five percent (14 of 40) of all
category 2 or greater hurricanes at landfall were associated with outbreaks, whereas 30% (12 of
40) were associated with nonoutbreaks (Fig. 5). In addition, 78% (14 of 18) of outbreak
hurricanes were category 2 or greater at landfall, and only 32% (12 of 37) of nonoutbreak
hurricanes were category 2 or greater at landfall (Tables 1 and 2). In order to assess the
statistical significance of hurricane intensity as a discriminator between outbreak and
nonoutbreak cases, a chi-square test was used. In essence, the chi-square test tests associations
between two variables and can determine if the association is stronger than random chance alone
(i.e., statistically significant; Wilks 1995, 133â134). The chi-square test statistic depends on the
degrees of freedom. At one degree of freedom, a statistically significant relationship exists
between outbreak and nonoutbreak hurricanes and SaffirâSimpson categories 1 and 0 and
categories 2 and higher at landfall. A
p
value equal to 0.002 was found; thus indicating that
intensity is a significant parameter at
α
=0.01 level. Clearly, these results show that outbreak
hurricanes are more likely to occur if the hurricane is a category 2 or higher at landfall. On the
14
other hand, hurricane intensity at time of landfall is not the only variable that should be
considered when assessing the risk of tornado outbreaks embedded within the system.
5.2. Landfall location
Previous authors have suggested that hurricane landfalls over the southern United States
are more likely to produce tornadoes than those making landfall over the eastern United States
(e.g., Hill et al. 1966; Novlan and Gray 1974; Gentry 1983; Weiss 1985; McCaul 1991). Such a
relationship, however, was not apparent from the figures comparing tracks for tornado-producing
hurricanes with tracks of non-tornado-producing hurricanes presented by Novlan and Gray
(1975). We wished to examine our dataset for such a purported relationship.
Landfall locations for the 83 hurricanes in our dataset were separated into two categories:
southern-coast landfalls (Texas, Louisiana, Mississippi, Alabama, and west coast of Florida) and
eastern-coast landfalls (east coast of Florida, Georgia, South and North Carolina, Virginia, and
New England states). Overall, 14 of 51 (27%) landfalls on the southern coast were outbreaks,
whereas 4 of 32 (13%) landfalls on the eastern coast were outbreaks. Again, using a chi-square
test with one degree of freedom, a
statistically significant relationship exists between outbreak
and nonoutbreak hurricanes and landfall location at
α
=0.10 level (
p
value=0.06).
Tracks of the various hurricanes for outbreak, nonoutbreak, and midclass categories were
plotted in Fig. 6. When presented in this manner, it is clear that outbreak hurricanes tended to
make landfall on the southern coast compared to the eastern coast (Fig. 6a). In addition, nearly
all landfalling outbreak hurricanes recurved northeastward after landfall, supporting the results of
Smith (1965) and Hill et al. (1966). In contrast, nonoutbreak hurricanes tended to fall into one of
two categories (Fig. 6b). Although the majority of nonoutbreak hurricanes tended to be those
15
that made landfall on the eastern coasts and recurved northward, there were a lesser group of
hurricanes that made landfall over the southern coast, but did not recurve (Fig. 6b). Midclass
hurricanes were more evenly divided between southern and east coast landfalls (Fig. 6c). When
the distinction is drawn between hurricanes that made landfall on the southern coast that
recurved versus hurricanes that made landfall on the southern coast that did not recurve, the
statistics produce a different result. In this case, 10 of 21 (48%) hurricanes that made landfall on
southern coasts and recurved were outbreaks, whereas 4 of 21 (19%) hurricanes that made
landfall on southern coasts and did not recurve were outbreaks.
This presentation suggests that
simple rules of thumb such as âlandfalling hurricanes on the east coast do not produce
tornadoesâ can be a serious oversimplification of the actual results.
To understand why recurving hurricanes that made landfall on the southern coast are more
likely to produce tornado outbreaks, we need to understand why hurricanes spawn tornadoes.
Most tornadoes in hurricanes are spawned in the right-front quadrant of the hurricane relative to
the direction of motion of the hurricane (e.g., Malkin and Galway 1953; Pearson and Sadowski
1965; Smith 1965; Orton 1970; Novlan and Gray 1974; Gentry 1983; McCaul 1991; Hagemeyer
and Hodandish 1995), the quadrant where the vertical wind shear and helicity tend to be most
favorable for tornado production (McCaul 1991; Bogner et al. 2000). Thus, hurricanes that place
their right-front quadrant over land for a longer period of time are more likely to produce greater
numbers of tornadoes. Hurricanes approaching southern United States coastlines have more time
to interact with land surfaces and have more exposure for the preferred right half of the hurricane
to generate tornadoes (Sadowski 1962; Smith 1965). Most east coast landfalls, on the other
hand, do not penetrate far inland, recurve back over the ocean more quickly, and are less likely to
produce a large number of tornadoes (Hill et al. 1966; Gentry 1983; McCaul 1991). One notable
16
exception is Hurricane Beulah (1967), which produced 117 tornadoes while tracking cyclonically
(Fig. 1). Hagemeyer and Hodanish (1995) found that 87% of tornado-producing hurricanes that
made landfall in Florida approached from the west coast of Florida, which placed the right-front
quadrant over the state and thus provided more opportunity for the hurricane to produce
tornadoes over land (e.g., Hill et al. 1966). As a result, whether storms recurve over land appears
to be an important distinction in order to determine whether or not an outbreak may occur. To
address the nature of recurving, we look to the next section.
5.3 Synoptic patterns for landfalling hurricanes that affected Texas
As mentioned previously, the ability of hurricanes making landfall over the southern
United States to produce a tornado outbreak apparently is tied to its ability to recurve
northeastward. To explore the reason for this recurvature, we investigate the differences in
composite synoptic patterns between outbreaks and nonoutbreaks associated with landfalling
hurricanes that affected Texas. We chose to perform this composite on Texas hurricanes for two
reasons. First, such hurricanes are relatively abundant in the outbreak and nonoutbreak
categories (5 and 8, respectively). Second, these hurricanes reach landfall in a relatively focused
geographic region, arriving in a nearly consistent storm track from the southeast (Figs. 6a, b).
(Developing such a composite analysis for Florida hurricanes was not robust for this reason.)
These two criteria allow for the construction of meaningful composite patterns. Composite
analyses on the day of landfall were produced using the NOAAâCIRES Climate Diagnostics
Center daily-averaged NCEP/NCAR Reanalysis (Kalnay et al. 1996) web page
(http://www.cdc.noaa.gov/Composites/Day). Composite analysis using the 6-h data did not
produce significantly different results and are not considered here.
17
Of the five outbreaks that affected Texas, there was one category 3 and four category 4.
In contrast, of the eight nonoutbreaks, all were category 1 hurricanes, except one category 4 and
one category 0. Thus, the synoptic composite might also provide some insight into why the
outbreak storms were more intense than the nonoutbreaks.
Although several different composite parameters show differences between outbreaks
and nonoutbreaks, we illustrate these differences with 500-hPa geopotential height and surfaceâ
850-hPa wind shear parameters (Fig. 7). Composite 500-hPa geopotential height indicates a
substantial difference between the Texas outbreaks and nonoutbreaks at landfall (Figs. 7a, b).
Whereas the Texas outbreak composite has a 500-hPa trough in the north-central United States,
the nonoutbreak composite has a ridge where the jet stream is far poleward of the landfalling
hurricanes (Figs. 7a, b). The anomalies of these composite means (based on 1968â1996
climatology) extend over a large area in the northern United States and southern Canada (Figs.
7c, d). Furthermore, the 500-hPa height anomaly over the landfalling hurricane is over twice as
deep, with a gradient of geopotential height (geostrophic wind) twice as great, for the outbreak
events (Figs. 7c, d).
Each of the five outbreak events had a trough in the northern United States and each of
the eight nonoutbreak events had a ridge in the northern United States (not shown), supporting
the composites. However, there remain two questions: 1) is the difference between the means at
each grid point significant (local significance), and 2) does the collection of grid points with
mean differences have field significance? Elmore et al. (2005) presented a nonparametric
approach to this problem. Unlike the data considered in Elmore et al. (2005), however, serial
correlation does not need to be considered in our data. Thus, a permutation test (Efron and
Tibshirani 1995; Elmore et al. 2002) is more accurate than a bootstrap test of the difference
18
between the means. The permutation test is performed at each grid point using 10 000 replicates.
All grid points for which the difference between the means is significant at the
α
p
= 0.05 level
are counted and divided by the total number of grid points, which yields the proportion of the
grid containing significant differences. Following Elmore et al. (2005), spatial correlation is
accommodated using a Monte Carlo process first described in Livezey and Chen (1989), and
dubbed the B method by Wang and Shen (1999). The B method for this work uses 10 000 trials
and is performed by combining the two fields (5 outbreaks and 8 nonoutbreaks) into a single data
matrix with 925 rows and 13 columns. Hence, 13 values are associated with each grid point. The
correlation between a uniform random variate at each grid point, as described in Elmore et al.
(2005), is tested for significance at
α
f
= 0.05. The proportion of grid points for which a
significant correlation occurs purely by chance yields the minimum proportion of grid points that
must possess significant difference between the means for field significance at
α
p
= 0.05. For
the height data, 14.38% of the gridpoints display statistically significant mean differences, and
12.97% are needed for field significance. Thus, the difference between these two fields is
significant at the
α
f
= 0.05 level.
These results suggest that landfalling hurricanes that affected Texas require greater deep-
layer (surfaceâ500-hPa) shear over the right-front quadrant in order to spawn a large number of
tornadoes, as noted by Wills (1969). This greater deep-layer shear is due to several factors. First,
the greater geostrophic wind anomaly at 500-hPa for the outbreak composite (Fig. 7c) suggests
that greater deep-layer shear is present compared with the nonoutbreak events (Fig. 7d). Second,
the jet stream dipping equatorward over the central United States toward Texas (Fig. 7a)
suggests that some interaction between the trough and the hurricane may be helping to produce
more intense hurricanes at landfall. As reviewed in Jones et al. (2003, their section 3d), a
19
moderate amount of shear associated with midlatitude troughs in the westerlies is sometimes
believed to be responsible for hurricane intensification. For nonoutbreak hurricanes, the jet is
much farther poleward, effectively isolating the landfalling storm underneath a large-scale 500-
hPa ridge (Fig. 7b) in much weaker deep-layer shear. Finally, the differences in the 500-hPa
flow over the southeastern United States result in greater deep-layer shear in the outbreak cases
than the nonoutbreak cases. The results from these composites support previous research
showing that the presence of large deep-layer shear is important for the generation of midlevel
mesocyclones in supercell thunderstorms (e.g., McCaul and Weisman 1996; McCaul et al. 2004),
which can spawn tornadoes in the hurricane environment.
Wills (1969), Novlan and Gray (1974), and McCaul (1991) also found that high low-level
wind shear was present in tornado-producing hurricanes. The surfaceâ850-hPa (hereafter, low-
level) wind shear also shows considerable differences between the outbreak and nonoutbreak
composites (Figs. 7e, f). The mean low-level wind shear is 4â8 m s
-1
in an onshore direction
northeast of the landfall location for the outbreak events (Fig. 7e), but is about half as strong (2â
4 m s
-1
) for the nonoutbreak events (Fig. 7f). To test the statistical significance of this data, only
4.0% of the data display significantly different shear values, yet as many as 7.4% could display a
significant difference by chance alone. Hence, the difference between the surfaceâ850-hPa shear
for the outbreak and nonoutbreak cases does not possess field significance.
If we increase
α
p
to
0.10 from 0.05, then field significance is obtained with
α
f
= 0.05.
Thus, the synoptic composites bring some insight into the potential causes for the
outbreaks: greater deep-layer shear and greater low-level shear, both associated with more
intense hurricanes, favors mesocyclogenesis and tornadogenesis, respectively. What causes the
recurvature? The composite of outbreak hurricanes that affected Texas suggests that a trough in
20
the jet stream (Fig. 7a) may transport them northeastward, whereas the nonoutbreaks remain
isolated from the jet (Fig. 7b). To test this hypothesis on a different set of cases, we examine the
strongly recurving nonoutbreak hurricanes over the eastern United States. Seven cases were
chosen from Table 2 and their 500-hPa geopotential height fields were composited (Fig. 8).
These results show that a 500-hPa trough lies to the northwest of the hurricane at landfall (Figs.
8a, c), which, by a day after landfall, has moved eastward (Figs. 8b, d). These results are
reminiscent of the northwest composite of extratropical transition over the western North Pacific
noted by Harr et al. (2000). Furthermore, the results of Fig. 8 support those of the synoptic
composites from the Texas storms in Fig. 7 that hurricane interaction with the jet stream results
in recurvature. Whereas this interaction and recurvature may lead to tornado outbreaks in the
Texas cases, the right-front quadrant tends to be offshore in the east-coast recurvature cases, so
tornadoes are not likely to be observed over land, if they are even produced at all.
To summarize, simple statistical tests on the location of landfall only provide part of the
explanation for whether an outbreak occurs. When the statistical methodology is combined with
synoptic composites, physical insight is obtained. Despite the relative success of these
composites on the synoptic scale, mesoscale and storm-scale effects may produce an
environment favorable for tornadoes or a tornado outbreak in an atypical synoptic-scale
environment, as might have been the case with Beulahâs unusual track.
Before leaving this discussion, this paper is not the first attempt to relate the track of
landfalling hurricanes to tornadogenesis. Novlan and Gray (1975) plotted landfall location and
tracks for hurricanes that produced tornadoes versus those that did not. They did not see as clean
a distinction as we did. This apparent discrepancy suggests that our approach of distinguishing
21
outbreak from nonoutbreak is more successful than attempting to distinguish tornado from no-
tornado.
5.4 Hurricane origin and date of landfall
Three areas of the North Atlantic Basin were identified as hurricane origin sites. First, the
Atlantic Ocean was defined as the area north and east a diagonal line from Key West, FL,
southeastward towards Cuba, Haiti, and the Dominican Republic ending near Trinidad and
Tobago. Next, the Caribbean Sea was defined as the area south and west of the same diagonal
line and south of ~23°N latitude near the Yucatan Peninsula. Last, the Gulf of Mexico was
identified as the area north and west of the Yucatan Peninsula and Cuba and waters west of Key
West, FL. The Atlantic Ocean is the largest and most active body of water and produced 38
United States hurricane landfalls from 1954â2004. Furthermore, the Caribbean Sea and Gulf of
Mexico formed 34 hurricane landfalls that affected the United States. Hurricane origin was
investigated as a possible parameter that may distinguish between outbreak and nonoutbreak
hurricanes, yet no significant relationship was found.
Additionally, date of landfall was inspected in order to determine if outbreak and
nonoutbreak hurricanes favor a certain time of year. To distinguish between outbreak hurricanes
and nonoutbreak hurricanes, Fig. 9 shows the ranked distribution of landfall date for each
classification. There appears to be a small gap in outbreak hurricanes between June and August
(Fig. 9), consistent with Hagemeyer (1997) who showed that the months of July and August had
a relative deficiency in tornado reports for Florida peninsular outbreaks. One possible
explanation for this shortage was noted by Hagemeyer and Schmocker (1991). They found that,
in the middle of the wet season (July and August in Florida), tornado environments were
22
characterized by weak lower-tropospheric winds and very low shear; consequently contributing
to the lack of tornadogenesis in the region. Even though date of landfall was not found to be a
statistically significant parameter in distinguishing between outbreak and nonoutbreak
hurricanes, it is interesting to note the general lack of outbreak cases in July.
5.5 El Niño-Southern Oscillation (ENSO) phase
There have been many studies on the effect of ENSO on hurricanes in the North Atlantic
basin (e.g., Gray 1984; Bove et al. 1998; Pielke and Landsea 1999; Larson et al. 2005). These
studies show that El Niño events suppress hurricane and tropical storm activity because of the
abnormally strong upper-tropospheric westerlies in the western Atlantic and Caribbean. La Niña
events, on the other hand, increase the number of hurricanes and tropical storms in the Atlantic
basin, increase the frequency of damaging landfalling storms, and increase the amount of
damage per storm. Hagemeyer (1999) found the relationship between ENSO and hurricane-
spawned tornadoes in Florida to be unclear. To attempt to resolve some of these discrepancies,
we examined the effect of ENSO on tornado outbreaks associated with hurricanes with our
dataset.
Rasmusson and Carpenter (1982) demonstrated that sea-surface temperature (SST)
anomalies can identify the onset of an El Niño event in late spring to early summer, and the peak
in SST anomalies occur in December, January, and February (i.e., Northern Hemisphere winter).
To determine the phase of ENSO during the months of peak hurricane activity (August,
September, and October), we used the Climate Prediction Centerâs Oceanic Niño Index (ONI)
for
Niño
3.4
SST
anomalies
based
on
the
1971â2000
period
(http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml). The
23
ONI indicates an El Niño event when the Niño 3.4 SST anomaly is greater than or equal to
0.5
°
C, and a La Niña event when the anomaly is less than or equal to â0.5
°
C. Anomalies
between â0.5
°
C and 0.5
°
C are considered neutral years. Figure 10 illustrates each hurricane
classification and the respective El Niño, neutral, or La Niña phase during August, September,
and October of the hurricane year. More hurricanes occurred during non El Niño years (26 in
neutral phases; 28 in La Niña phases) than in El Niño years (18), in agreement with previous
studies (e.g., Gray 1984; Bove et al. 1998; Pielke and Landsea 1999; Larson et al. 2005). Six
outbreak hurricanes occurred during El Niño events and seven occurred during La Niña events.
Therefore, if a landfalling hurricane occurred during an El Niño year, the event had a higher
chance of being associated with a tornado outbreak (6 outbreaks of 18, or 33%), as opposed to if
the hurricane occurred during non El Niño (neutral or La Niña) years (12 outbreaks of 54, or
22%). However, a statistically significant relationship was not found between outbreak and
nonoutbreak hurricanes and El Niño versus non-El Niño years using a chi-square test and one
degree of freedom (
p
value=0.26). A much longer period of observations will be required to
assess whether this relationship is significant or just due to random variations and the small
sample size.
6. Summary and Conclusions
This study examined all landfalling hurricanes that affected the United States from 1954 to
2004. Criteria developed by Verbout et al. (2005) was used to examine days with a large number
of tornadoes across the United States (big tornado days) and later to develop a dataset of tornado
outbreaks associated with hurricanes and nonoutbreak hurricanes. The results of the present
study, as well as a review of the previous literature, indicates the following characteristics are
24
capable of distinguishing landfalling hurricanes that produce tornado outbreaks from those that
do not.
More intense hurricanes are more likely to produce tornadoes (Novlan and Gray 1974;
Gentry 1983; McCaul 1991; this study). Hurricanes that are weakening are more likely to
produce tornadoes (Hill et al. 1966; Novlan and Gray 1974), although this effect may be more
likely due to the fact that most hurricanes are weakening as they approach land.
Previous studies have argued that hurricanes recurving to the northeast were more likely to
produce tornadoes than those moving westward (Smith 1965; Hill et al. 1966; Novlan and Gray
1974). By exception, one of the most prolific tornado-producing storms, Hurricane Beulah
(1967), did not behave in this manner (Fig. 1). Although there is some skepticism about the
validity of this rule (Novlan and Gray 1974), nearly all our outbreak cases (Fig. 6a) exhibited
recurvature over the southern or eastern United States. By contrast, nearly all our nonoutbreak
cases (Fig. 6b) did not recurve or recurved along or off the eastern United States. Composite
synoptic analyses for landfalling hurricanes that affected Texas show that distinct synoptic
patterns distinguish outbreaks from nonoutbreaks due to the location of the jet stream.
Outbreaks were associated with a 500-hPa trough over the north-central United States, allowing
the jet stream to dip equatorward over the central United States (Fig. 7a). With such a synoptic
pattern, landfalling hurricanes that affected Texas experienced recurvature quickly (Fig. 6a).
Nonoutbreaks, on the other hand, were associated with a 500-hPa ridge over the north-central
United States and a jet stream in southern Canada (Fig. 7b). As such, recurvature was less likely,
and the tracks of Texas nonoutbreak hurricanes showed a westward or northwestward track once
onshore (Fig. 6b).
25
Hurricanes that recurve along or off the eastern United States do not produce tornado
outbreaks because their right-front quadrant is not over the land for an extended period of time
(Hill et al. 1966; Novlan and Gray 1974; McCaul 1991; this study). Most tornadoes are found in
the right-front quadrant relative to the motion of the hurricane (Malkin and Galway 1953;
Pearson and Sadowski 1965; Smith 1965; Orton 1970; Novlan and Gray 1974; Gentry 1983;
McCaul 1991). This quadrant is where the greatest deep-layer shear and helicity are found
(McCaul 1991; Bogner et al. 2000), typical ingredients for mesocyclogenesis and
tornadogenesis, respectively. Air parcels in the right-front quadrant also have a recent history of
being over water, and the right-front quadrant is where convection tends to be strongest (e.g.,
Gentry 1983). Outbreak hurricanes have stronger low-level wind shear than nonoutbreak
hurricanes (Wills 1969; Novlan and Gray 1974; McCaul 1991; this study). This result is
consistent with the requirement for strong low-level wind shear for tornadogenesis.
Gray (1984), Bove et al. (1998), Pielke and Landsea (1999), Larson et al. (2005) and this
study have shown that Atlantic basin hurricanes tend to occur during non El Niño years. The
number of outbreak hurricanes, however, was nearly evenly split between El Niño and La Niña
years. This study also found the following characteristics were not useful for distinguishing
outbreaks from nonoutbreaks: storm origin and date of landfall.
The results of this study allow forecasters to identify the factors that affect tornado
outbreaks with landfalling hurricanes, with the goal of being better able to anticipate these
events. As our results show, the statistics suggest better discrimination between outbreak and
nonoutbreak hurricanes rather than the occurrence of tornadoes or no tornadoes. This effect may
explain some contradictory results in the previous literature.
26
Acknowledgments
. We thank Dan McCarthy and NOAA/NWS/Storm Prediction Center for
providing the tornado data and the NOAA/NWS/Tropical Prediction Center for providing the
hurricane data. Thanks to Brad Barrett for his assistance with the hurricane data. Schultz thanks
Paul Hoffmann (MIT) for his early collaboration on this topic. This manuscript benefited from
the comments from Richard Thompson and several anonymous reviewers. Verbout was funded
by a research assistantship through U.S. Office of Naval Research grant N00014-0021-1-0181.
Funding for Schultz was provided by NOAA/OAR/NSSL under NOAAâOU Cooperative
Agreement NA17RJ1227.
27
REFERENCES
Bogner PB, Barnes GM, Franklin JL (2000) Conditional instability and shear for six hurricanes
over the Atlantic Ocean. Wea Forecasting 15: 192â207
Bove MC, Elsner JB, Landsea CW, Niu X, OâBrien JJ (1998) Effect of El Niño on U. S.
landfalling hurricanes, revisited. Bull Amer Meteor Soc 79: 2477â2482
Brooks HE, Doswell CA III, Kay MP (2003) Climatological estimates of local daily tornado
probability for the United States. Wea Forecasting 18: 626â640
Curtis L (2004) Midlevel dry intrusions as a factor in tornado outbreaks associated with
landfalling tropical cyclones from the Atlantic and Gulf of Mexico. Wea Forecasting 19: 411â
427
Doswell CA III, Burgess DW (1988) On some issues of United States tornado climatology. Mon
Wea Rev 116: 495â501
Efron B, Tibshirani RJ (1993) An Introduction to the Bootstrap. Chapman and Hall 436 pp
Elmore KL, Stensrud DJ, Crawford KC (2002) Ensemble cloud model applications to forecasting
thunderstorms. J Appl Meteor 41: 363â383
28
______, Baldwin ME, Schultz DM (2005) Field significance revisited: Spatial bias errors in
forecasts as applied to the Eta model. Mon Wea Rev in press
Franklin JL, McAdie CJ, Lawrence MB (2003) Trends in track forecasting for tropical cyclones
threatening the United States, 1970â2001. Bull Amer Meteor Soc 84: 1197â1203
Galway JG (1975) Relationship of tornado deaths to severe weather watch areas. Mon Wea
Rev
103: 737â741
_____ (1977) Some climatological aspects of tornado outbreaks. Mon Wea Rev 105: 477â484
Gentry RC (1983) Genesis of tornadoes associated with hurricanes. Mon Wea Rev 111: 1793â
1805
Gray RW (1919) A tornado within a hurricane area. Mon Wea Rev 47: 639â639
Gray WM (1984) Atlantic seasonal hurricane frequency, Part I: El Niño and 30 mb quasibiennial
oscillation influences. Mon Wea Rev 115: 1649â1668
Grazulis TP (1993) Significant Tornadoes, 1680â1991. Environmental Films 1326 pp
Hagemeyer BC (1997) Peninsular Florida tornado outbreaks. Wea Forecasting 12: 399â427
29
_____ (1998) Significant tornado events associated with tropical and hybrid cyclones in Florida.
Preprints 16th Conf on Weather Analysis and Forecasting Phoenix, AZ Amer Meteor Soc: 4â6
_____ (1999) El Nino and significant tropical and hybrid cyclone tornado events in Florida.
Preprints 23rd Conf on Hurricanes and Tropical Meteorology Dallas, TX Amer Meteor Soc:
415â418
_____, Hodanish SJ (1995) Florida tornado outbreaks associated with tropical cyclones.
Preprints 21st Conf on Hurricanes and Tropical Meteorology Miami, FL Amer Meteor Soc: 312â
314
_____, Schmocker GK (1991) Characteristics of east central Florida tornado environments. Wea
Forecasting 6: 499â514
_____, Spratt SM (2002) Thirty years after Hurricane AgnesâThe forgotten Florida tornado
disaster. Preprints 25th Conf on Hurricanes and Tropical Meteorology San Diego, CA Amer
Meteor Soc: 422â423
Harr PA, Elsberry RL, Hogan TF (2000) Extratropical transition of tropical cyclones over the
western North Pacific. Part II: The impact of midlatitude circulation characteristics. Mon Wea
Rev 128: 2634â2653
30
Hill EL, Malkin W, Schulz WA Jr (1966) Tornadoes associated with cyclones of tropical originâ
practical features. J Appl Meteor 5: 745â763
Hills GB (1929) The September 28, 1929, tornado in Fort Lauderdale, Fla. Mon Wea Rev 57:
420â421
Hoadley DK (1981) A Tropical Storm David tornado in Fairfax CountyâSeptember 1979. Bull
Amer Meteor Soc 62: 498â507
Jones SC, Coauthors (2003) The extratropical transition of tropical cyclones: Forecast
challenges, current understanding, and future directions. Wea Forecasting 18
:
1052â1092
Kalnay E, Coauthors (1996) The NCEP/NCAR 40âYear Reanalysis project. Bull Amer Meteor
Soc 77: 437â471
Landsea CW, Coauthors (2004) A reanalysis of Hurricane Andrewâs intensity. Bull Amer
Meteor Soc 85: 1699â1712
Larson J, Zhou Y, Higgins RW (2005) Characteristics of landfalling tropical cyclones in the
United States and Mexico: Climatology and interannual variability. J Climate 18
:
1247â1262
Livezey RE, Chen WY (1983) Statistical field significance and its determination by Monte Carlo
techniques. Mon Wea Rev 111: 46â59
31
Malkin W, Galway JG (1953) Tornadoes associated with hurricanes as illustrated by Franconia,
Va., tornado, September 1, 1952. Mon Wea Rev 81: 299â303
McCarthy DW (2003) NWS tornado surveys and the impact on the national tornado database.
Preprints First Symposium on F-scale and Severe Weather Damage Assessment Long Beach, CA
Amer Meteor Soc CD-ROM: 3.2
McCaul EW Jr (1987) Observations of the Hurricane âDannyâ tornado outbreak of 16 August
1985. Mon Wea Rev 115: 1206â1223
______ (1991) Buoyancy and shear characteristics of hurricane-tornado environments. Mon Wea
Rev 119: 1954â1978
______, Weisman ML (1996) Simulations of shallow supercell storms in landfalling hurricane
environments. Mon Wea Rev 124: 408â429
______, Buechler DE, Goodman SJ, Cammarata M (2004) Doppler radar and lightning network
observations of a severe outbreak of tropical cyclone tornadoes. Mon Wea Rev 132: 1747â1763
Neumann CJ, Jarvinen BR, McAdie CJ, Hammer GR (1999) Tropical Cyclones of the North
Atlantic Ocean, 1871â1998. Historical Climatology Series 6-2. National Climatic Data Center
32
Asheville, NC in cooperation with the Tropical Prediction Center/ National Hurricane Center
Miami, FL 206 pp
Novlan DJ, Gray WM (1974) Hurricane-spawned tornadoes. Mon Wea Rev 102: 476â488
______ (1975) Reply. Mon Wea Rev 103: 1021â1023
Orton R (1970) Tornadoes associated with Hurricane Beulah on September 19-23, 1967. Mon
Wea Rev 98: 541â547
Pearson A (1975) Comments on âHurricane-spawned tornadoes.â Mon Wea Rev 103: 1021
______, Sadowski AF (1965) Hurricane-induced tornadoes and their distribution. Mon Wea Rev
93: 461â464
Pielke RA Jr, Landsea CW (1999) La Niña, El Niño, and Atlantic hurricane damages in the
United States. Bull Amer Meteor Soc 80: 2027â2033
______, Pielke RA Sr (1997)
Hurricanes: Their Nature and Impacts on
Society
. John Wiley and
Sons Press 279 pp
Rasmusson EM, Carpenter TH (1982) Variations in tropical sea surface temperature and surface
wind fields associated with the Southern Oscillation/El Niño. Mon Wea Rev 110: 354â384
33
Rudd MI (1964) Tornadoes during Hurricane Carla at Galveston. Mon Wea Rev 92: 251â 254
Sadowski AF (1962) Tornadoes associated with Hurricane Carla, 1961. Mon Wea Rev 90: 514â
516
Smith JS (1965) The hurricane-tornado. Mon Wea Rev 93: 453â459
Spratt SM, Sharp DW, Welsh P, Sandrik A, Alsheimer F, Paxton C (1997) A WSR-88D
assessment of tropical cyclone outer rainband tornadoes. Wea Forecasting 12: 479â501
Stiegler DJ, Fujita TT (1982) A detailed analysis of the San Marcos, Texas tornado induced by
Hurricane âAllenâ on 10 August, 1980. Preprints 12th Conf on Severe Local Storms Amer
Meteor Soc: 371â374
Tannehill IR (1938) Hurricanes: Their Nature and History
.
Princeton University Press 257 pp
Verbout SM, Brooks HE, Leslie LM, Schultz DM (2005) Evolution of the United States tornado
database: 1954â2003. Wea Forecasting 21: 86-93
Wang X, Shen SS (1999) Estimation of spatial degrees of freedom of a climate field. J Climate
12: 1280â1291
34
Watson AI, Jamski MA, Turnage TJ, Bowen JR, Kelley JC (2005) The tornado outbreak across
the north Florida panhandle in association with Hurricane Ivan. Preprints 32d Radar Conf Amer
Meteor Soc CD-ROM: 10R.3
Weiss SJ (1985) On the operational forecasting of tornadoes associated with tropical cyclones.
Preprints 14th Conf on Severe Local Storms Indianapolis, IN Amer Meteor Soc: 293â296
______ (1987) Some climatological aspects of forecasting tornadoes associated with tropical
cyclones. Preprints 17th Conf on Hurricanes and Tropical Meteorology Miami, FL Amer
Meteor Soc: 160â163
Wilks DS (1995) Statistical Methods in the Atmospheric Sciences. Academic Press 467 pp
Wills TG (1969) Characteristics of the tornado environment as deduced from proximity
soundings. Preprints 6th Conf on Severe Local Storms Chicago, IL Amer Meteor Soc: 222â229
35
TABLE CAPTIONS
Table 1. Outbreak cases listed by landfall date. Hurricane name, landfall date, SaffirâSimpson
category at time of landfall, total number of reported tornadoes, and number of F1 and greater
tornadoes are given.
#
Indicates landfall that affected Texas (section 5c); * indicates multiple
landfalls and location.
Table 2. Same as Table 1 except for nonoutbreak cases. ! Indicates hurricanes that narrowly
made landfall;
#
indicates landfall that affected Texas;
&
indicates east coast recurver (section
5c).
Table 3. Same as Table 1 except for midclass cases.
36
FIGURE CAPTIONS
Fig. 1. Track of Hurricane Beulah (1967) from 20â22 Sept. Hollow circles indicate center of
circulation at 0000 UTC on each day. Thin line denotes a distance of 100 nm from shore.
Individual tornado reports are marked with small plus signs.
Fig. 2. Annual number of reported tornadoes (triangles) and annual number of F1 and greater
reported tornadoes (circles) from 1954 to 2004. Black line and dashed line indicate linear
regression fit to each series.
Fig. 3. Dashed line (scale on left side of graph) is the probability of having any tornado reported
somewhere in the United States by day of year, based on the 1980â1999 record (Brooks et al.
2003). Solid black line (black scale on right side of graph) displays the probability of a single
day having at least 1.5% of the expected annual value and at least 8 F1 and greater tornadoes
reported, based on the 1954â2004 record. Solid gray line (gray scale on right side of graph)
displays the probability distribution of all outbreak hurricane cases from 1954â2004. Solid black
curve is scaled by a factor of 9 and solid gray curve is scaled by a factor of 30. Curves were
generated using kernel density estimation with a Gaussian smoother with
Ï
t
= 15 days (Brooks et
al. 2003).
Fig. 4. Graph of the number of tornadoes associated with 1.5% of the linear regression value
(triangles) and 0.5% of the linear regression value (circles) for each year (1954â2004).
37
Fig. 5. Number of outbreak (black), midclass (gray hatched), and nonoutbreak (gray) hurricanes
listed by SaffirâSimpson category (1-5) at time of landfall. Category 0 represents storms that
were not hurricane strength at time of landfall.
Fig. 6. Tracks of hurricanes that affected the United States (1954â2004). (a) outbreak hurricane
tracks; (b) nonoutbreak hurricane tracks; and (c) midclass hurricane tracks.
Fig. 7. Composite fields for landfalling hurricanes that affected Texas with tornado outbreaks
(left) and nonoutbreaks (right) (Tables 1 and 2). (a) and (b) mean 500-hPa geopotential height
(every 25 m); (c) and (d) anomaly 500-hPa geopotential height from 1968â1996 climatology
(every 10 m); and (e) and (f) mean surfaceâ850-hPa wind-shear magnitude (shaded every 2 m s
-
1
) and direction (vectors). Composite maps were provided by the NOAAâCIRES Climate
Diagnostics Center, Boulder, Colorado (http://www.cdc.noaa.gov).
Fig. 8. Composite fields for east coast landfalling hurricanes associated with nonoutbreaks for
(a) and (c) day of landfall and (b) and (d) day after landfall (Table 2). (a) and (b) mean 500-hPa
geopotential height (every 25 m); (c) and (d) anomaly 500-hPa geopotential height from 1968â
1996 climatology (every 10 m). Composite maps were provided by the NOAAâCIRES Climate
Diagnostics Center, Boulder, Colorado (http://www.cdc.noaa.gov).
Fig. 9. Ranked distribution of outbreak (triangles) and nonoutbreak (circles) hurricanes by date
from 1954 to 2004.
38
Fig. 10. Number of outbreak (black), midclass (gray hatched), and nonoutbreak (gray)
hurricanes by ENSO phase. ENSO phase was determined by the Climate Prediction Centerâs
Niño 3.4 sea surface temperature anomalies for Aug, Sep, and Oct.
39
Table 1. Outbreak cases listed by landfall date. Hurricane name, landfall date, SaffirâSimpson
category at time of landfall, total number of reported tornadoes, and number of F1 and greater
tornadoes are given.
#
Indicates landfall that affected Texas (section 5c); * indicates multiple
landfalls and location.
Name
Date of Landfall Cat. @ Landfall
Total Tors
F1+
#
Audrey
6/27/1957
4
22
17
#
Carla
9/11/1961
4
20
19
Cleo
8/27/1964
2
12
9
Hilda
10/4/1964
2
12
10
#
Beulah
9/20/1967
4
117
21
Edith (LA) *
9/16/1971
2
16
15
Agnes (FL) *
6/19/1972
1
28
22
Babe
9/5/1977
1
14
13
David
9/3/1979
2
34
26
#
Allen
8/10/1980
3
29
19
Danny
8/15/1985
1
42
29
#
Gilbert
9/16/1988
4
41
13
Andrew (LA) *
8/26/1992
3
56
23
Opal
10/4/1995
1
32
8
Georges
9/28/1998
2
46
21
Frances
9/5/2004
2
99
36
Ivan
9/16/2004
3
117
68
Jeanne
9/26/2004
3
40
14
40
#
Texas landfall
* multiple landfall
41
Table 2. Same as Table 1 except for nonoutbreak cases. ! Indicates hurricanes that narrowly
made landfall;
#
indicates landfall that affected Texas;
&
indicates east coast recurver (section
5c).
Name
Date of Landfall Cat. @ Landfall
Total Tors
F1+
#
Alice
6/25/1954
1
0
0
Hazel
10/15/1954
2
0
0
!&
Diane
8/17/1955
1
1
1
!&
Ione
9/19/1955
1
0
0
!&
Cindy
7/9/1959
1
1
1
#
Debra
7/25/1959
1
1
0
Donna (S. FL)*
9/10/1960
4
0
0
Donna (CT)*
9/12/1960
1
0
0
#
Cindy
9/17/1963
1
0
0
Dora
9/10/1964
2
3
2
Isbell (NC)*
10/16/1964
0
0
0
Betsy (S. FL)*
9/8/1965
3
2
1
Alma
6/9/1966
1
4
2
Gladys
10/19/1968
1
2
2
Camille
8/18/1969
5
3
1
Fern (LA)*
9/4/1971
0
0
0
#
Fern (TX)*
9/10/1971
1
3
2
#
Edith (S. TX) *
9/14/1971
0
0
0
Agnes (NY) *
6/22/1972
0
0
0
42
&
Diana
9/13/1984
1
1
0
Bob (S. FL)*
7/23/1985
0
2
0
Bob (SC)*
7/24/1985
1
3
1
!&
Gloria
9/27/1985
1
2
1
Kate
11/21/1985
2
0
0
!&
Charley
8/17/1986
1
5
5
Florence
9/10/1988
1
4
0
#
Chantal
8/1/1989
1
2
1
&
Hugo
9/22/1989
4
2
2
Bob (MA)*
8/19/1991
2
3
2
Andrew (FL) *
8/24/1992
5
0
0
Fran
9/6/1996
3
2
1
#
Bret
8/23/1999
4
5
0
Dennis
9/5/1999
0
2
1
Irene
10/15/1999
1
5
4
Gordon
9/18/2000
0
5
1
#
Claudette
7/15/2003
1
0
0
Isabel
9/18/2003
2
0
0
!
narrowly made
landfall
#
Texas landfall
&
east coast recurver
* multiple landfall
43
Table 3. Same as Table 1 except for midclass cases.
Name
Date of Landfall Cat. @ Landfall
Total Tors
F1+
Connie
8/12/1955
1
4
4
Flossy
9/24/1956
1
5
4
Gracie
9/29/1959
3
6
6
Donna (NC)*
9/11/1960
2
4
4
Ethel
9/15/1960
1
5
4
Isbell (S. FL)*
10/14/1964
3
9
8
Betsy (LA)*
9/10/1965
2
4
3
Celia
8/3/1970
1
8
5
Carmen
9/8/1974
3
5
3
Eloise
9/23/1975
3
5
4
Bob
7/11/1979
1
8
5
Frederic
9/13/1979
2
10
4
Alicia
9/18/1983
3
22
2
Elena
9/2/1985
3
9
6
Juan
10/29/1985
1
12
8
Bonnie
6/26/1986
1
5
5
Jerry
10/16/1989
1
7
1
Bob (NC)*
8/18/1991
3
6
4
Gordon
11/16/1994
0
6
3
Allison
6/5/1995
0
6
5
Erin
8/2/1995
1
12
2
44
Bertha
7/13/1996
2
14
6
Danny
7/18/1997
1
12
6
Bonnie
8/27/1998
2
8
3
Floyd
9/16/1999
2
16
4
Lili
10/3/2002
1
26
5
Charley (S. FL)*
8/13/2004
4
12
6
Charley (SC)*
8/14/2004
1
7
1
* multiple landfall
45
Fig. 1. Track of Hurricane Beulah (1967) from 20â22 Sept. Hollow circles indicate center of
circulation at 0000 UTC on each day. Thin line denotes a distance of 100 nm from shore.
Individual tornado reports are marked with small plus signs.
46
Fig. 2. Annual number of reported tornadoes (triangles) and annual number of F1 and greater
reported tornadoes (circles) from 1954 to 2004. Black line and dashed line indicate linear
regression fit to each series.
47
Fig. 3. Dashed line (scale on left side of graph) is the probability of having any tornado reported
somewhere in the United States by day of year, based on the 1980â1999 record (Brooks et al.
2003). Solid black line (black scale on right side of graph) displays the probability of a single
day having at least 1.5% of the expected annual value and at least 8 F1 and greater tornadoes
reported, based on the 1954â2004 record. Solid gray line (gray scale on right side of graph)
displays the probability distribution of all outbreak hurricane cases from 1954â2004. Solid black
curve is scaled by a factor of 9 and solid gray curve is scaled by a factor of 30. Curves were
generated using kernel density estimation with a Gaussian smoother with
Ï
t
= 15 days (Brooks et
al. 2003).
48
Fig. 4. Graph of the number of tornadoes associated with 1.5% of the linear regression value
(triangles) and 0.5% of the linear regression value (circles) for each year (1954â2004).
49
Fig. 5. Number of outbreak (black), midclass (gray hatched), and nonoutbreak (gray) hurricanes
listed by SaffirâSimpson category (1-5) at time of landfall. Category 0 represents storms that
were not hurricane strength at time of landfall.
50
Fig. 6. Tracks of hurricanes that affected the United States (1954â2004). (a) outbreak hurricane
tracks; (b) nonoutbreak hurricane tracks; and (c) midclass hurricane tracks.
51
Fig. 7. Composite fields for landfalling hurricanes that affected Texas with tornado outbreaks
(left) and nonoutbreaks (right) (Tables 1 and 2). (a) and (b) mean 500-hPa geopotential height
(every 25 m); (c) and (d) anomaly 500-hPa geopotential height from 1968â1996 climatology
52
(every 10 m); and (e) and (f) mean surfaceâ850-hPa wind-shear magnitude (shaded every 2 m s
-
1
) and direction (vectors). Composite maps were provided by the NOAAâCIRES Climate
Diagnostics Center, Boulder, Colorado (http://www.cdc.noaa.gov).
53
Fig. 8. Composite fields for east coast landfalling hurricanes associated with nonoutbreaks for
(a) and (c) day of landfall and (b) and (d) day after landfall (Table 2). (a) and (b) mean 500-hPa
geopotential height (every 25 m); (c) and (d) anomaly 500-hPa geopotential height from 1968â
1996 climatology (every 10 m). Composite maps were provided by the NOAAâCIRES Climate
Diagnostics Center, Boulder, Colorado (http://www.cdc.noaa.gov).
54
Fig. 9. Ranked distribution of outbreak (triangles) and nonoutbreak (circles) hurricanes by date
from 1954 to 2004.
55
Fig. 10. Number of outbreak (black), midclass (gray hatched), and nonoutbreak (gray)
hurricanes by ENSO phase. ENSO phase was determined by the Climate Prediction Centerâs
Niño 3.4 sea surface temperature anomalies for Aug, Sep, and Oct.