Spatial and Temporal Characteristics of Path Direction*

Philip W. Suckling Texas State University–San Marcos

Walker S. Ashley Northern Illinois University

Common perception is that tornadoes travel in paths from the southwest quadrant of directions toward the northeast. This study examines path directions for 6,194 tornadoes that occurred in the eastern two-thirds of the United States during the twenty-three-year period 1980–2002. At the national scale, nearly 70 percent of tornadoes included in the study propagated from the west, west-southwest, and southwest, with west-southwest being the highest frequency origin direction. Nevertheless, distinct seasonal and regional variations were found. In central and northern areas of the country, a more westerly or northwesterly path origin prevails during late spring and summer. The midtropospheric flow, convective typology, and synoptic patterns of tornado outbreaks are thought to contribute to the distributions observed in the climatology. Key Words: natural hazards, tornadoes, tornado path.

Introduction safest location is opposite to the southwest cor- ner since debris actually accumulates nearest to azard mitigation studies indicate that over the initial impact (Eagleman, Muirhead, and Hthe past half century improved warning Williams 1975; Grazulis 2001). technologies and community preparedness have Current tornado safety recommendations for substantially reduced fatalities and injuries as- buildings assert that one should seek shelter in sociated with tornadoes (Boruff et al. 2003). an interior, reinforced room or corridor in the Tornado hazard studies and associated risk-as- lowest level of a structure (AMS 2000; FEMA sessment models frequently focus on spatial 2004; NWS 2004). In addition to this primary characteristics such as track length and track rule, safety guidelines, research, and accounts width, but often overlook path direction (e.g., continue to indicate that the ‘‘downstream’’ Schaefer, Kelly, and Abbey 1986; Grazulis, portion of a building will be somewhat safer Schaefer, and Abbey 1993; Brooks 2004). Tor- than the ‘‘upstream,’’ relative to the approach- nado path direction, however, is not a trivial ing tornado (Eagleman 1967; Eagleman, Muir- matter for tornado hazard mitigation. When head, and Williams 1975; Tornado Project considering the safest location within a building Online 1999; AMS 2000). Clearly, such guide- in the event of a tornado, at one time the south- lines need to be revaluated in light of recent west corner of a basement was considered the wind engineering research, advancements, and safest, on the theory that wind-blown debris techniques. For example, should ‘‘safe rooms’’ from the damaged building would accumulate and shelters be constructed away from the west-, in the northeast corner of the basement as the southwest-, and south-facing walls, which, tornado moved from the southwest toward the based on climatological data, have a higher northeast (Finley 1887; Flora 1954). Although probability of receiving the highest wind loads tornado path direction, as we will show, is in- and therefore have a higher chance of collapse deed generally from the southwest, research by (Eagleman 1967)? Clearly, microscale vortex Eagleman (1967) and others has shown that the and structure interactions make it difficult to

* The assistance of Jamie Dyer (Climatology Research Laboratory, University of Georgia) is gratefully acknowledged. The authors thank the anonymous reviewers, whose comments helped us clarify and improve the manuscript. The Professional Geographer, 58(1) 2006, pages 20–38 r Copyright 2006 by Association of American Geographers. Initial submission, March 2004; revised submission, November 2004; final acceptance, December 2004. Published by Blackwell Publishing, 350 Main Street, Malden, MA 02148, and 9600 Garsington Road, Oxford OX4 2DQ, U.K. Spatial and Temporal Characteristics of Tornado Path Direction 21 determine where debris will fall. However, cli- path direction in a climatological study of tor- matological information such as that derived nado data from the period 1916–1985 and con- from this study can be utilized in future engi- cluded that 59 percent of tornadoes moved from neering studies to investigate the safest location the southwest. in structures for different tornado scenarios. As illustrated in this review, there is limited Additionally, data on tornado path direction research on tornado path direction for the Unit- are also extremely important to public officials ed States, especially during the past twenty and information managers at local emergency years. Recent enhancement of tornado report- operational facilities who must determine the ing procedures by the distribution of resources and storm spotters in (NWS), the development and enhancement of severe weather situations. Results from this an integrated warning system (Doswell, Moller, study show that spotters should be positioned to and Brooks 1999), increased population, in- have an unobstructed view to the west and creased public awareness, and the proliferation southwest horizons, where a majority of torna- of video cameras have led to improvements in does originate. Understanding climatological the accuracy and consistency of the tornado re- features such as predominant tornado path di- port database. Despite inherent limitations of rection can aid in general tornado awareness the raw tornado report database (e.g., see Do- and in obtaining critical information quickly so swell and Burgess 1988; Grazulis 1993; Brooks, that officials can make timely local warning dis- Doswell, and Kay 2003), these recent improve- semination decisions. ments and enhancements in the data allow for Common perception is that U.S. tornadoes detailed examination of the more recent portion travel in paths from the southwest toward the of the database to uncover characteristics of northeast (Eagleman, Muirhead, and Williams tornado path directions. Given the potential 1975; Eagleman 1990; Grazulis 2001). Howev- importance of tornado path direction to hazard er, there has been very limited research on tor- mitigation recommendations, this study seeks nado path direction for the United States, and to provide a comprehensive geographic study of most studies that have examined tornado path path direction for tornadoes occurring in the direction have only considered specific states. central and eastern United States. We also spec- For example, estimates from the early twentieth ulate that the interaction of processes across a century indicated that in Kansas 61 percent of multitude of scales, including storm-scale dy- tornadoes came from the southwest, 16 percent namics and synoptic-scale steering winds, are from the west, and 11 percent from the north- important controls on the distributions de- west (U.S. Department of Commerce 1953). It scribed. was shown that in Illinois more than 80 percent of tornadoes traveled from the southwest Data and Methods through west directions (Wilson and Chang- non 1971). In a study of the state of Iowa based Data regarding U.S. tornadoes are now much on the thirteen-year period 1959–1971, Notis more reliable and complete than ever before and Stanford (1973) reported that almost (Burgess, Donaldson, and Desrochers 1993; 70 percent of Iowa tornadoes traveled from Schaefer et al. 1993; Grazulis 2001). Several the southwest quadrant, but about 30 percent factors account for the improvement in the tor- came from the west or northwest directions. nado report database. These include emphasis The descriptive climatology by Schaefer et al. on and improvement of warning verification (1980) examined national tornado path direc- practices by the NWS since the early 1980s tion, albeit briefly, and concluded that 87 ( Johns and Evans 2000), increased use and im- percent of conterminous U.S. tornadoes travel provement of spotter networks (Doswell, Moll- from the southwest quadrant. Schaefer et al. er, and Brooks 1999), modernization of the (1980) also noted that when tornado motion is NWS during the early 1990s with improved considered on a seasonal basis, only summer- technologies including deployment of Doppler time storms differ markedly from this general radar (Friday 1994), and improved communi- pattern, with one of four summer tornadoes cations (e.g., cell phones) and documenting me- traveling from the northwest quadrant of direc- dia (e.g., video cameras). Despite improvement tions. Fujita (1987) briefly examined tornado in the quality and consistency of the data, there 22 Volume 58, Number 1, February 2006

Figure 1 Map of the six regions selected for the regional-scale study. Tornadoes that occurred within these six regions were utilized for the national-scale analysis. remain a number of limitations and biases in the reasons: (1) prior to this period the data become raw tornado report dataset. These limitations increasingly suspect and incomplete (e.g., see and biases in the tornado and other severe Kelly et al. 1978) and (2) a twenty-three-year weather datasets have been described elsewhere period should provide a long enough record to (e.g., Doswell and Burgess 1988; Grazulis account for any climatological shifts in favor- 1993; Brooks, Doswell, and Kay 2003; Schaefer, able synoptic patterns for severe weather devel- Weiss, and Levit 2003) and involve such factors opment (e.g., see the effects of favorable, long- as basic errors in reporting or recording of time term severe weather pattern shifts on and location information, spatial and temporal climatologies presented by Bentley and Sparks variability in the efforts to collect severe weather 2003 and Coniglio and Stensrud 2004). reports for warning verification programs, ad- Because tornadoes are rare west of the Con- justments in the nature of detailed damage sur- tinental Divide, we restricted our analysis to a veys, and population changes. thirty-seven-state area across the central and Detailed characteristics of tornadoes in the eastern United States. In an attempt to examine United States are collected by local NWS of- possible region-specific shifts in the climatolo- fices and, thereafter, published in the National gy, we divided that area into six regions (Figure Climatic Data Center (NCDC) publication 1). The regions were divided primarily along Storm Data and made available on the NCDC state lines due to the state-by-state reporting storm events Web page1 or in the SeverePlot strategy utilized by Storm Data and because software package2 (Hart 1993). For this study, mitigation efforts are often implemented on a we restricted our analyses to 1980–2002 for two state-by-state basis. Subdividing the thirty-sev- Spatial and Temporal Characteristics of Tornado Path Direction 23

Table 1 Number of tornadoes during the twenty- Path direction origin is illustrated using a three-year period 1980–2002 figure similar to a wind-rose diagram, known as F-scale Total Track 1.61 km 1.61 km a ‘‘radar diagram,’’ with the frequency of torna- number information path length with do events plotted according to the sixteen car- incomplete length complete dinal directions (e.g., Figure 2). The radar track information diagrams in this study utilize a floating scale of F0 11754 10140 2615 349 actual frequency numbers, instead of percent- F1 7017 4474 4195 2452 ages, for the period in question. The actual F2 2438 826 2060 1598 F3 708 77 691 629 magnitudes are employed in the diagrams F4 158 5 156 153 rather than percentages because they illustrate F5 13 0 13 13 the importance of seasonality and sample size Total 22088 15522 6730 6194 on developing a tornado climatology. None- theless, noteworthy percentages are indicated en-state area into regions was a somewhat sub- throughout the text. jective process and we realize that some differ- ences of opinion may arise as to which states Results: National Scale belong in which regions. To analyze tornado path direction, detailed At the national scale, tornadoes travel in paths latitude and longitude coordinate information dominantly from the W, WSW, and SW direc- for the start and end points of each individual tions (69.3 percent of tornadoes included in the tornado is needed. Many tornadoes in the da- study) with WSW having the modal frequency tabase are either small, short-lived, or lack post- (Figure 2). The frequency of tornadoes trave- storm survey team analysis and, therefore, in- ling from the NW (3.2 percent) and WNW (7.0 clude only ‘‘touchdown’’ locations. In fact, more percent), and from the SSW (8.1 percent) and S than 70 percent of tornadoes in the study period (4.5 percent) is notable, but these directions are lacked full track information (Table 1), and in a far less common than the three main directions number of cases where a track length was indi- of W (21.6 percent), WSW (26.2 percent), and cated no tornado end-point information was SW (21.5 percent). Tornadoesdo travel in paths available. Since 97 percent of the tornadoes with from all nine of the remaining cardinal direc- a length of less than 1.61 km (1 mile) had in- tions, but far less often. In the thirty-seven-state complete track data, we restricted our analyses region, 65.1 percent of all tornadoes move into to tornado events greater than or equal to this the northeast quadrant. This percentage is far length. More than 6,000 tornadoes meeting the less than the 87 percent indicated for nationwide start and end-point and minimal length criteria tornadoes from 1950–1978 by Schaefer et al. were available for analysis (Table 1). Tracks for (1980). From 1980–2002, 28.2 percent of torna- these tornadoes are plotted in Figure 1, reveal- does propagated into the southeast quadrant. ing the distribution of events across the nation. Tornadoes moving toward the west were ex- A tornado can indeed change path direction tremely rare during this period, with only 4.0 during its life span. Nevertheless, for the pur- percent of events moving into the northwest poses of this study, a straight path from start quadrant and 2.7 percent of tornadoes moving location to end location is assumed. We believe into the southwest quadrant. this is a safe assumption because tornadoes The dominant easterly and northeasterly mo- characteristically have much smaller ‘‘side-to- tion of tornadoes described in our twenty- side’’ deviation along the track in comparison to three-year climatology is likely due to several the overall length of the along-track vector. Path factors, including the predominance of the mid- directions according to path origin (i.e., the di- latitude westerlies across the study region, tor- rection from which the tornado is traveling) are nado-producing storm typology, and the categorized into sixteen directions (N, NNE, tendency for tornado outbreaks in the eastern NE, ENE, etc.). Again, these directions refer to two-thirds of the United States to occur in areas the direction from which the tornado is ap- of southwesterly flow at 500 hPa with a long proaching (i.e., WSW means that the tornado is wave trough to the west of the outbreak area traveling from the west-southwest and moving (Beebe 1956; Miller 1972; Galway and Pearson toward the east-northeast). 1981). 24 Volume 58, Number 1, February 2006

A All Tornadoes N 1800 NNW NNE 1500 NW NE 1200 900 WNW ENE 600 300 W 0 E

WSW ESE

SW SE

SSW SSE S

BCDWeak (F0-F1) Tornadoes Significant (F2+) Tornadoes Violent (F4+) Tornadoes

Figure 2 Radar diagrams illustrating the frequency of tornado path direction origin for (A) all, (B) weak, (C) significant, and (D) violent tornadoes.

The motion of tornadoes is undoubtedly af- northeast assuming that the 500-hPa level is a fected by the parent convective storms that good indicator for steering-level flow. spawn them. In turn, the parent convective In addition, it is important to understand the storms are steered by the prevailing winds in factors that determine cellular and, in particular, which they are embedded (Fujita 1987). Notis supercellular convective propagations since and Stanford (1973) discovered that Iowa tor- most tornadoes are produced by these convec- nado directions of propagation were closely tied tive typologies (Moller et al. 1994; Tessendorf to 500-hPa flow. The mean 500-hPa flow for the and Trapp 2000). Weisman and Klemp (1986) 1,927 days during which tornadoes occurred in determined that supercell motion is primarily our dataset was predominantly westerly to due to the advection of the storm by the mean southwesterly across the United States except wind and the internal storm dynamics that form for a small portion of the north-central region when convective updrafts develop in a vertically (Figure 3A). Most storm systems producing sheared environment (e.g., see Bunkers et al. tornadoes would have propagated to the east or 2000). In many cases, supercells, including Spatial and Temporal Characteristics of Tornado Path Direction 25

Figure 3 Mean 500-hPa heights, winds, and anomalies for days in which a tornado occurred (A) during the entire twenty-three-year period of study, and (B) during the summer months of June, July, and August over the twenty-three-year period.

those producing tornadoes, have been observed for nearly 24 percent of all tornadoes in the moving slower and at 201,301, or greater to the contiguous United States. It is likely that tor- right of the mean wind—so called, right-turn- nadoes spawned by hurricanes may account for ing supercells (Maddox 1976). In fact, the de- some of the more unusual directions identified viant motion (in terms of mean flow) of in our study. supercells is a distinct characteristic of this Finally, eighty-one tornado outbreaks3 oc- storm type and specific techniques for calculat- curred during the study period—an outbreak ing supercell motion have been developed (e.g., being defined as a day with at least five F2 þ tor- see Bunkers et al. 2000). We hypothesize that nadoes and a Destruction Potential Index the unique motion of right-turning supercells (Thompson and Vescio 1998) of 50 or greater. likely accounts for some shift in the path clima- All but ten of these tornado outbreaks (i.e., 86 tology to a more westerly direction. Addition- percent) occurred within 500-hPa flow that ally, there are likely a number of noncellular contained a southwesterly component across tornadoes in our dataset. Tessendorfand Trapp the outbreak region. Nearly 81 percent of all (2000) illustrated that quasi-linear convective tornadoes in the dataset that occurred during systems and hurricane rain bands may account tornado outbreak days propagated from the 26 Volume 58, Number 1, February 2006 southwesterly quadrant, indicating that tornado case, with SW (35.2 percent) and WSW (29.2 directions of motion do appear to be closely tied percent) dominant, followed by SSW (13.2 to midtropospheric flow, as suggested by Notis percent) and W (11.8 percent). and Stanford (1973). The bimonthly and monthly (not shown) Comparing ‘‘weak’’ (F0–F1) tornadoes with analyses illustrate subtle, yet distinct, seasonal significant (F2 þ ) and extreme (F4 þ )events shifts in the tornado path climatology. Tornado illustrates that weak tornadoes have a greater paths propagate from a primarily southwesterly concentration of W path direction than more direction during January, February, and March, severe tornadoes (Figures 2B–2D). Significant then from a predominantly westerly direction and extreme tornadoes are tightly clustered, with for the next six months (April–September), be- most of these events traveling from the WSWor fore returning to again propagating from the SW. Nevertheless, there are several examples of southwesterly direction toward the end of the strong tornadoes that have propagated in direc- annual cycle. Previous research has suggested tions different from these typical path directions. that the seasonal shifts in tornado occurrence For example, the F5 Jarrell, Texas, tornado that are strongly linked to upper-air synoptic-scale occurred on 27 May 1997 propagated from the meteorological patterns (Notis and Stanford northeast toward the southwest. 1973; Bluestein and Golden 1993; Davis, Stan- To illustrate variations through the year, tor- meyer, and Jones 1997; Monfredo 1999; Brown nado path direction origin is plotted in radar 2002). As previously illustrated, Notis and Stan- diagrams for monthly pairs (Figure 4). The di- ford (1973) noted seasonal shifts in Iowa torna- agrams are designed to illustrate dominant di- does and suggested that direction of tornado rection rather than total frequency, therefore movement is intimately related to the 500-hPa the frequency scale varies: the March–April and flow. Indeed, mean 500-hPa charts for the tor- May–June monthly pairs have greater total fre- nado days in the bimonthly cases (not shown) quency of tornadoes, the January–February and indicate a distinct association between the September–October pairs have the least fre- 500-hPa flow across the study domain and the quency. For January–February, tornado paths highest frequency path directions. For example, are dominantly from the SW (31.8 percent), during June, July, and August, tornadoes from followed by WSW (29.5 percent) and, to a lesser the N, NNW, NW,and WNWaccount for 25.3 extent, W (17.9 percent) directions. By March– percent of all events during this period, but dur- April, there is a slight shift to a greater westerly ing the remainder of the year only 8.8 percent of component, with WSW and W accounting for tornadoes propagate from these directions. 31.0 percent and 21.0 percent of all tornado These warm-weather months are the only peri- paths, respectively. As spring progresses toward od where the mean tornado day 500-hPa analysis summer, a shift in dominant origin direction to is from the west-to-northwest across the majority W is evident in the May–June diagram with W of the study area (Figure 3B). Johns (1982) dis- and SW path origins accounting for 25 percent covered that during the summer months a large each. For the summer months of July–August, number of severe weather outbreaks, including tornado paths are primarily from the north- those producing tornadoes, occur with west- western quadrant rather than from the south- northwest and northwest flow in the midtropo- western quadrant. During July–August, the sphere. Hence, the climatological maximum in modal frequency is W (26.0 percent), however northwest-flow severe weather outbreaks during WNW (15.7 percent) and NW (7.8 percent) the summertime (c.f., Figure 3 in Johns 1982) is frequencies remain relatively high. By fall, tor- likely responsible for the distinct shift toward nado path origins shift back to the southwestern more easterly and southeasterly moving torna- quadrant. For September–October, WSW does during this warm-weather period. (21.2 percent) and SW (20.5 percent) are mo- dal, with a wide swath of directions being evi- Results: Regional Scale dent including WNW (5.3 percent) and W (17.4 percent), through SSW (11.5 percent) and The national-scale results illustrate that there S (10.6 percent). For November–December, are potentially important seasonal dynamics in- tornado path direction origins return to a nar- volved with the analysis of tornado path direc- rower range, similar to the January–February tion. Tornado season varies considerably by Spatial and Temporal Characteristics of Tornado Path Direction 27

January-February Tornadoes March-April Tornadoes

N N NNW 150 NNE NNW 500 NNE 120 400 NW NE NW NE 90 300

WNW 60 ENE WNW 200 ENE 30 100

W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

May-June Tornadoes July-August Tornadoes

N N 800 200 NNW NNE NNW NNE

NW 600 NE NW 150 NE

400 100 WNW ENE WNW ENE 200 50

W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

September-October Tornadoes November-December Tornadoes N N 250 NNW 120 NNE NNW NNE 100 200 NW NE NW NE 80 150 60 WNW WNW 100 40 ENE ENE 20 50 W 0 E W 0 E

WSW WSW ESE ESE

SW SE SW SE

SSW SSE SSW SSE S S

Figure 4 Radar diagrams illustrating the frequency of tornado path direction origin for monthly pairs. location (region) within the country, governed Stanmeyer, and Jones 1997; Monfredo 1999; by changes in synoptic-scale meteorological Brown 2002). Therefore, we repeated the anal- controls (Bluestein and Golden 1993; Davis, ysis of tornado path direction for six similar- 28 Volume 58, Number 1, February 2006

Table 2 Number of tornadoes, by region, during the twenty-three-year period 1980–2002 with path length 1.61 km and complete track information

Region Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Total

South Central 126 52 177 298 465 154 14 20 44 119 119 100 1688 Southeast 80 110 230 223 154 95 27 37 65 56 235 64 1376 Central 7 11 87 325 460 375 122 44 38 76 32 22 1599 Ohio V/Mid Atlantic 11 4 52 78 126 162 91 40 38 37 46 1 686 Northeast 1 1 3 11 41 65 69 22 16 5 12 0 246 North Central 0 0 14 34 100 210 131 52 32 21 5 0 599 Total 225 178 563 969 1364 1061 454 215 233 314 499 187 6194 sized regions (Figure 1). The total number of distribution could be an artifact of the weak tornadoes and the monthly frequency for each tropospheric flow that occurs across this region of these regions are presented in Table 2. before the strong westerlies return during Tornado season peaks in March, April, and the winter season. Supercell motion, when the May for the southernmost regions (South Cen- mean wind is relatively weak, has been observed tral, Southeast), nevertheless many tornadoes to deviate by greater than 301 to the right of are still evident during January and February. the mean wind (Davies 1998; Bunkers et al. Tornadoesalso are experienced throughout the 2000), and that deviant motion might explain fall in both of these regions, with November in this unique distribution. Moreover, tornadoes fact being the overall peak month for tornado generated by tropical storms or hurricanes occurrence in the Southeast. Veryfew tornadoes could account for the increase in tornadoes occur in either region during the summer with a southerly component during this months of July and August. period. Path direction origins are from a nar- In contrast, the northernmost regions (North rower range for the other monthly pairs, espe- Central, Ohio Valley–Mid Atlantic, Northeast) cially November–December. Because summer experience few tornadoes during the winter tornadoes are rare, no radar diagram is present- (December, January, February). Tornadoseason ed for July–August. The rarity of tornadic peaks during spring and summer (May, June, events during this period is likely caused by an July for the North Central region; April through increasingly capped environment in this region July for the Ohio Valley–Mid Atlantic region; (Farrell and Carlson 1989), a decrease in the May through July for the Northeast). An April, thermodynamic instability of the atmosphere May, June peak is evident for the Central region, ( Johns 1982), and a general weakening of the but summer and fall tornadoes are also experi- midtropospheric westerlies leading to an overall enced. Similar to the northernmost regions, the decrease in the deep-layer shear needed to Central region has few winter tornadoes. facilitate the growth of tornado-producing con- In the following sections, diagrams for each vection ( Johns 1982). of the six regions illustrate the seasonal distri- bution of tornadoes, and radar diagrams plot Southeast Region tornado path direction for monthly pairs. For the Southeast region, March–April is again the bimonthly period with the greatest number South Central Region of tornadoes (Figure 6, top left). To an even Spring is the peak tornado season in the South greater extent than in the South Central region, Central region (Figure 5, top left). For March– March–April tornado paths in the Southeast are April, tornado paths are dominantly from the dominantly from the W, WSW, and SW direc- W, WSW, and SW directions (Figure 5), which tions (Figure 6). This pattern is also prevalent is quite similar to the national scale for these for the late fall and winter months of Novem- months illustrated previously in Figure 4. Other ber–December and January–February. monthly pairs illustrated in Figure 5 do not vary Fewer tornadoes occur during the summer, much from the March–April case. However, the and tornado paths are somewhat more variable path origin directions for September–October for May–June (with a W dominance) and Sep- indicate a wide distribution among directions in tember–October (with greater variation from the southwest quadrant. This relatively broad WNW through to SSE and SE). Some of the Spatial and Temporal Characteristics of Tornado Path Direction 29

January-February Tornadoes South Central 500 N 70 450 NNW NNE 60 400 NW 50 NE 350 40 300 30 WNW ENE 250 20 200 10 150 W 0 E 100 Number of Tornadoes 50 0 WSW ESE JFMAMJJASOND SW SE

SSW SSE S

March-April Tornadoes May-June Tornadoes N N NNW 150 NNE NNW 200 NNE NW 150 NE NW 100 NE 100 WNW 50 ENE WNW ENE 50

W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

September-October Tornadoes November-December Tornadoes N N NNW 40 NNE NNW 100 NNE 80 NW 30 NE NW NE 60 20 WNW ENE WNW 40 ENE 10 20 W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

Figure 5 Seasonal distribution of tornadoes (top left) and radar diagrams illustrating the frequency of tornado path direction origin for monthly pairs, for the South Central region. latter events were associated with tropical sys- coming onshore or from tornadoes forming in tem-generated tornadoes rather than being the low-shear environments on convergence lines more common supercell-generated tornadoes. associated with sea breezes (Brooks, Doswell, In addition, the relatively wide variation in path and Kay 2003). Again, no radar diagram is pre- directions during the late warm season in the sented for July–August given the rarity of sum- Southeast may be attributable to mer tornadoes in this region. 30 Volume 58, Number 1, February 2006

January-February Tornadoes Southeast 250 N NNW 60 NNE 50 200 NW NE 40 30 150 WNW 20 ENE 100 10 W 0 E 50 Number of Tornadoes

0 WSW ESE JFMAMJJASOND SW SE

SSW SSE S

March-April Tornadoes May-June Tornadoes N N NNW 200 NNE NNW 80 NNE

NW 150 NE NW 60 NE

100 40 WNW ENE WNW ENE 50 20

W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

September-October Tornadoes November-December Tornadoes N N NNW 25 NNE NNW 120 NNE 20 100 NW NE NW NE 80 15 60 WNW 10 ENE WNW 40 ENE 5 20 W 0 E W 0 E

WSW ESE WSW ESE

SE SW SW SE

SSW SSE SSW SSE S S

Figure 6 Seasonal distribution of tornadoes (top left) and radar diagrams illustrating the frequency of tornado path direction origin for monthly pairs, for the Southeast region.

Central Region cludingW,WSW,andSWdirections,butalso with a significant number from the SSWand some The peak months for tornado events in the Cen- from the S (Figure 7). For the smaller number of tral Region are April, May, and June (Figure 7, top tornadoes occurring in the preceding winter left). Tornado paths for March–April are again months (not shown), tornado paths are from a dominantly from the southwestern quadrant in- very narrow range of directions dominated by SW. Spatial and Temporal Characteristics of Tornado Path Direction 31

March-April Tornadoes Central 500 N NNW 140 NNE 450 120 400 NW 100 NE 350 80 300 WNW 60 ENE 250 40 200 20 150 W 0 E 100 Number of Tornadoes 50 0 WSW ESE JFMAMJ JASOND SW SE

SSW SSE S

May-June Tornadoes July-August Tornadoes N N NNW 250 NNE NNW 50 NNE 200 40 NW NE NW NE 150 30 WNW 100 ENE WNW 20 ENE 50 10 W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

September-October Tornadoes November-December Tornadoes N N NNW 30 NNE NNW 20 NNE 25 NW NE NW 15 NE 20 15 10 WNW ENE WNW 10 ENE 5 5 W 0 E W 0 E

WSW ESE WSW ESE

SE SW SW SE SSW SSE SSW SSE S S

Figure 7 Seasonal distribution of tornadoes (top left) and radar diagrams illustrating the frequency of tornado path direction origin for monthly pairs, for the Central region.

The May–June monthly pair experiences the August summer tornadoes are more numerous most tornadoes for this region. The modal path than in the South Central and Southeast re- direction is from the W, with a notable number gions, although still much smaller in number of events occurring in a wider swath from the than the March–April and May–June time pe- NWand WNW,and from the WSW,SW,SSW riods. Path direction origins are notably differ- through to S directions. For this region, July– ent during the summer months of July and 32 Volume 58, Number 1, February 2006

March-April Tornadoes Ohio Valley - Mid Atlantic 180 N NNW 50 NNE 160 40 140 NW NE 120 30 100 WNW 20 ENE 80 10 60 W 0 E 40 Number of Tornadoes 20 0 WSW ESE JFMAMJJASOND SW SE SSW SSE S

May-June Tornadoes July-August Tornadoes N N NNW 100 NNE NNW 50 NNE 80 40 NW NE NW NE 60 30 WNW 40 ENE WNW 20 ENE 20 10 W 0 E W 0 E

WSW ESE WSW ESE

SE SW SW SE

SSW SSE SSW SSE S S

September-October Tornadoes November-December Tornadoes N N NNW 25 NNE NNW 15 NNE 20 NW NE NW NE 10 15 WNW 10 ENE WNW 5 ENE 5 W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE SSW SSE SSW SSE S S

Figure 8 Seasonal distribution of tornadoes (top left) and radar diagrams illustrating the frequency of tornado path direction origin for monthly pairs, for the Ohio Valley–Mid Atlantic region.

August with W, NW, and even a N spike in the breaks across the region during this period (see, number of events. The unique shift to more e.g., Figures 4, 7, and 8 in Johns 1982). By fall, westerly and northwesterly path origins during there is a dramatic shift back to tornado paths these summer months is likely due to the cli- from the WSW through S directions. Similarly, matological maximum in northwest flow out- the mean midtropospheric flow during this time Spatial and Temporal Characteristics of Tornado Path Direction 33

Northeast 80 70 60 50 40 30 20

Number of Tornadoes 10 0 JFMAMJJASOND

May-June Tornadoes July-August Tornadoes

N N NNW 50 NNE NNW 30 NNE 40 25 NW NE NW NE 20 30 15 WNW 20 ENE WNW ENE 10 10 5 W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

Figure 9 Seasonal distribution of tornadoes (top left) and radar diagrams illustrating the frequency of tornado path direction origin for monthly pairs, for the Northeast region. period for tornado days in the dataset shifts back and WNW directions during July and August is predominantly from the southwest, indicating likely the result of an increase in northwest flow that this shift in the path direction climatology severe weather outbreaks that tend to occur across is governed primarily by the upper-level steer- the region during these two months (e.g., see ing flow. Finally, although November–Decem- Figures 7 and 8 in Johns 1982). By fall, path ber tornadoes are rare, they primarily have path directions shift back to the southwestern quadrant origins from the WSW and SW in this region. for both the September–October and November– December radar diagrams (the latter diagram Ohio Valley–Mid Atlantic Region really represents November since there is only Tornado events for the Ohio Valley–Mid At- one December event). Tornadoes are also quite lantic region are most numerous during the rare during the winter months of January and spring and early summer months of April, May, February. June, and July (Figure 8, top left). Notable numbers of tornadoes also are evident for the Northeast Region months of August through November, but win- Illustrative of the climatological return of low- ter tornadoes are rare. level moisture and low static stability to the Path direction origins illustrated in the radar Northeast during late spring and summer, tor- diagram sequence indicate a gradual shift from the nado events for this region are most numerous March–April case dominated by the Wand WSW during the months of May, June, and July (Fig- directions to more westerly to northwesterly or- ure 9, top left). Throughout May and June in the igin directions in July and August (Figure 8). Northeast, tornado paths originate from pri- Similar to the Central region, the increase in W marily the W. During July and August, the tor- 34 Volume 58, Number 1, February 2006 nado path direction shifts, with the modal origin North Central Region becoming WSW. Nevertheless, a large number of Northeast tornadoes arrive from the north- Tornadoes are very rare during the winter half of west quadrant of motion during the summer. the year for the North Central region (Figure 10, Tornadoes during the remainder of the year in top left). Only two monthly pairs (May–June, this region are rare. July–August) have a sizeable number of torna- North Central 250

200

150

100

50 Number of Tornadoes

0 JFMAMJJASOND

March-April Tornadoes May-June Tornadoes

N N NNW 15 NNE NNW 80 NNE

NW NE NW 60 NE 10 40 WNW ENE WNW ENE 5 20

W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

July-August Tornadoes September-October Tornadoes

N N NNW 60 NNE NNW 20 NNE 50 NW NE NW 15 NE 40 30 10 WNW ENE WNW ENE 20 5 10 W 0 E W 0 E

WSW ESE WSW ESE

SW SE SW SE

SSW SSE SSW SSE S S

Figure 10 Seasonal distribution of tornadoes (top left) and radar diagrams illustrating the frequency of tornado path direction origin for monthly pairs, for the North Central region. Spatial and Temporal Characteristics of Tornado Path Direction 35 does. Nevertheless, four radar diagrams are pre- ospheric flow to the west-to-northwest across sented in order to illustrate changes in tornado the study area during summer may be the pri- path direction for this region (Figure 10). mary contributing factors leading to this ob- The small numbers of events that occur dur- served shift in tornado path directions. ing March–April have paths from the SSW, SW, Regionally, important variations occur in tor- WSW,and, to a lesser extent, W.For May–June, nado path direction origins. In the North Cen- when there are far more tornadoes, origins shift tral and Northeast regions, most tornadoes a bit and are primarily from the W and WSW. occur in late spring (May–June) and summer This shift to a more westerly and northwesterly ( July–August). Tornadoesduring these month- direction continues through July and August, ly pairs travel dominantly from the west and where W and WNW are the predominant path west-southwest during May and June, and pri- origin directions. The modal travel direction marily from the northwest quadrant of direc- for these summer events is from the WNW, tions during July and August. Additionally, with paths ranging from the NNW and NW although tornado path direction is most com- through to W, WSW, SW, and SSW. Unique to monly from the southwestern quadrant for both this region, the primary path origin for the six- the Central and Ohio Valley–Mid Atlantic re- month period May–October is from the W.The gions, variations do occur. The months of April, predominance of westerly midtropospheric May, and June represent peak tornado season flow (Figure 3) and the influence of northwest- for these regions. By May–June, the modal di- erly flow severe weather outbreaks are the prob- rection for tornado path origin is west or west- able reasons for this unique distribution. southwest, similar to the case for the North Central and Northeast regions. Although sum- Conclusions mer tornadoes are far less frequent in the Cen- tral and Ohio Valley–Mid Atlantic regions, path Although forecasters and atmospheric scientists directions are dominantly from the west and the are aware of the generalization that tornadoes northwest quadrant of directions when they do typically propagate from the southwest quad- occur. For the remaining seasons in these re- rant, no study has comprehensively examined gions, the commonly perceived southwest the characteristics of U.S. tornado path direc- quadrant of directions dominates. tions. Such research has important implications For the southernmost regions (South Central for hazard mitigation and engineering studies. and Southeast), spring and winter tornadoes For example, knowledge about tornado path dominate. Summer tornadoes are very rare. direction can be useful when deciphering the Tornadopath direction, as expected, is predom- safest location within a structure. inantly from the southwest quadrant; therefore Results from this study illustrate that the the common perception that tornadoes travel common perception that most U.S. tornadoes from the southwest quadrant of directions is travel from the southwest quadrant of directions most valid for these southern regions. is generally valid. For the thirty-seven-state re- Seasonal shifts in tornado occurrence are gion, tornadoes moving from the west-south- strongly linked to upper tropospheric synoptic- west toward the east-northeast were the most scale meteorological patterns (Bluestein and frequent, with tornadoes traveling from the Golden 1993; Davis, Stanmeyer, and Jones southwest and west directions also having high 1997; Monfredo 1999; Brown 2002). Notis frequencies. Even on this national scale, subtle and Stanford (1973) demonstrated that tornado seasonal shifts in tornado path direction are ev- path direction is closely linked to the 500-hPa ident. In late spring (May–June), west and west- level flow pattern. The results from our study southwest are the modal origin directions for support Notis and Stanford’s conclusions and tornado paths. By the summer months ( July– illustrate further that the midtropospheric flow August), when tornadoes are far less common is the primary control on the overall tornado nationally, path directions are dominantly from direction climatology. In addition, more than 23 the west, west-northwest, and northwest direc- percent of tornadoes in the dataset were asso- tions. The climatological maximum in north- ciated with one of eighty-one tornado out- west flow severe weather outbreaks ( Johns breaks. In a large majority of these outbreaks, 1982) and a shift in the mean tornado day trop- the steering-level flow was from the southwest. 36 Volume 58, Number 1, February 2006

In fact, 90 percent of all tornadoes that occurred Literature Cited during these outbreaks propagated from the SSW, SW, WSW, or W, indicating that the American Meteorological Society (AMS). 2000. Pol- icy statement: Tornado preparedness and safety. midlevel tropospheric flow is a controlling fac- Bulletin of the American Meteorological Society tor in tornado path direction in these scenarios. 81:1061–65. Finally, right-turning supercells and their asso- Beebe, R. G. 1956. Tornadocomposite charts. Month- ciated tornadoes in the climatology are specu- ly Weather Review 84:127–42. lated to have contributed to a greater easterly Bentley, M. L., and J. A. Sparks. 2003. A 15 yr propagation of direction of motion. However, climatology of derecho-producing mesoscale con- until a climatology of supercells and their var- vective systems over the central and eastern United ious typologies is developed, such hypotheses States. Climate Research 24:129–39. will remain unsupported. Bluestein, H. B., and J. H. Golden. 1993. A review of Additional study, incorporating synoptic cli- tornado observations. In The tornado: Its structure, matological methods, could be useful in iden- dynamics, prediction, and hazards, ed. C. Church, D. tifying the progression of atmospheric features Burgess, C. Doswell, and R. Davies-Jones, 319–52. that lead to the character and seasonality iden- Geophysical Monograph 79. Washington, DC: tified in the climatological distributions shown American Geophysical Union. Boruff, B. J., J. A. Easoz, S. D. Jones, H. R. Landry, in this research. Such research would be useful J. D. Mitchem, and S. L. Cutter. 2003. Tornado in confirming the potential consistency of tor- hazards in the United States. Climate Research nado path direction in summer versus other 24:103–17. times of the year. Brooks, H. E. 2004. On the relationship of tornado This study concludes that the common per- path length and width to intensity. Weather and ception that tornadoes travel from the south- Forecasting 19:310–19. west quadrant of directions (especially SW, Brooks, H. E., C. A. Doswell, and M. P. Kay. 2003. WSW, and W) is generally valid for the south- Climatological estimates of local daily tornado ern United States, and also valid for other parts probability for the United States. Weather and of the country especially during spring, fall, and Forecasting 18:626–40. winter. However, by late spring, a more westerly Brown, M. E. 2002. The spatial, temporal, and ther- component often dominates tornado path di- modynamic characteristics of southern-Atlantic rection origin. Where summer tornadoes occur United States tornado events. Physical Geography in significant numbers, a westerly to north- 23:401–17. Bunkers, M. J., B. A. Klimowski, J. W. Zeitler, R. L. westerly directional origin is most prevalent. Thompson, and M. L. Weismanm. 2000. Predict- Since most tornadoes in the North Central re- ing supercell motion using a new hodograph tech- gion occur during late spring and summer, tor- nique. Weather and Forecasting 15:61–79. nadoes in this area often travel from the Burgess, D. W., R. J. Donaldson Jr., and P. R. Des- northwest quadrant of directions (including rochers. 1993. Tornado detection and warning by west). It is essential that these seasonal and re- radar. In The tornado: Its structure, dynamics, predic- gional variations in tornado path directions be tion, and hazards, ed. C. Church, D. Burgess, C. incorporated into future engineering and haz- Doswell, and R. Davies-Jones, 203–21. Geophysi- ard studies to further improve tornado mitiga- cal Monograph 79. Washington, DC: American tion efforts.’ Geophysical Union. Coniglio, M. C., and D. J. Stensrud. 2004. Interpret- ing the climatology of . Weather and Fore- casting 19:595–605. Notes Davies, J. M. 1998. On supercell motion in weaker wind environments. Preprints, 19th Conference on 1 Available online at http://www4.ncdc.noaa.gov/cgi- Severe Local Storms, 685–88. Minneapolis, MN: win/wwcgi.dll?wwEventStorms. American Meteorological Society. 2 SeverePlot v2 is freely available online at http:// Davis, R. E., T. M. Stanmeyer, and G. V. Jones. www.spc.noaa.gov/software/svrplot2/. 1997. A synoptic climatology of tornadoes in Vir- 3 These tornado outbreak days have been identified by ginia. Physical Geography 18:383–407. John Hart of the NWS’s . Doswell, C. A., III, and D. W.Burgess. 1988. On some The list is available online at http://www.crh.noaa. issues of United States tornado climatology. Month- gov/mkx/climate/torout.htm. ly Weather Review 116:495–501. Spatial and Temporal Characteristics of Tornado Path Direction 37

Doswell, C. A. III, A. R. Moller, and H. E. Brooks. vective systems in the central and eastern United 1999. Storm spotting and public awareness since States, 1986–95. Part I: Temporal and spatial dis- the first tornado forecasts of 1948. Weather and tribution.’’ Bulletin of the American Meteorological Forecasting 14:544–57. Society 81:1049–54. Eagleman, J. R. 1967. Tornado damage pattern in Kelly, D. L., J. T. Schaefer, R. P. McNulty, C. A. Topeka,Kansas, June 8, 1966. Monthly Weather Re- Doswell III, and R. F. Abbey Jr. 1978. An augment- view 95:370–74. ed tornado climatology. Monthly Weather Review ———. 1990. Severe and unusual weather, 2nd ed. Le- 106:1172–83. nexa, KS: Trimedia Publishing. Maddox, R. A. 1976. An evaluation of tornado prox- Eagleman, J. R., V. U. Muirhead, and N. Williams. imity wind and stability data. Monthly Weather Re- 1975. , tornadoes, and building damage. view 104:133–42. Lexington, MA: D. C. Heath. Miller, R. C. 1972. Notes on the analysis and severe-storm Farrell, R. J., and T. N. Carlson. 1989. Evidence for forecasting procedures of the Air Force Global Weather the role of the lid and underunning in an outbreak of Central. Air Weather Service Tech.Rept. 200 (Rev.), tornadic thunderstorms. Monthly Weather Review Air Weather Service, Scott Air Force Base, IL, 190 117:857–71. pp. FEMA (Federal Emegency Management Associa- Moller, A. R., C. A. Doswell III, M. P. Foster, and G. tion). 2004. Surviving the storm, tornadoes. Avail- R. Woodall. 1994. The operational recognition of able online at: http://www.fema.gov/pdf/hazards/ supercell environments and storm survtorn.pdf (last accessed 13 October 2005). structures. Weather and Forecasting 9:327–47. Finley, J. P. 1887. Tornadoes. New York: The Insur- Monfredo, W. 1999. Relationships between phases of ance Monitor. Available online at: http://hsci.cas. the El Nino–southern oscillation and character of ou.edu/exhibits/exhibit.php?exbid=30&exbpg (last the tornado season in the south-central United accessed 13 October 2005). States. Physical Geography 20:413–21. Flora, S. D. 1954. Tornadoes of the United States. Notis, C., and J. L. Stanford. 1973. The contrasting Norman: University of Oklahoma Press. synoptic and physical character of northeast and Friday, E. W. 1994. The modernization and associat- southeast advancing tornadoes in Iowa. Journal of ed restructuring of the National Weather Service: Applied Meteorology 12:1163–73. An overview. Bulletin of the American Meteorological NWS (National Weather Service). 2004. Thunder- Society 75:43–52. storm, tornadoes, lightning: A preparedness guide. Fujita, T. T. 1987. U.S. tornadoes Part 1 70-year sta- Available online at: http://www.nws.noaa.gov/om/ tistics. Satellite and Mesometeorology Research brochures/ttI.pdf (last accessed 13 October 2005). Project (SMRP) Research Paper No. 218, Univer- Schaefer, J. T., D. L. Kelly, and R. F. Abbey. 1986. A sity of Chicago, 122 pp. minimum assumption tornado-hazard probability Galway, J. G., and A. Pearson. 1981. Winter tornado model. Journal of Climate and Applied Meteorology outbreaks. Monthly Weather Review 109:1072–80. 25:1934–45. Grazulis, T. P. 1993. Significant tornadoes 1680–1991. Schaefer, J. T., D. L. Kelly, C. A. Doswell III, St. Johnsbury, VT: Environmental Films. J. G. Galway, R. J. Williams, R. P. McNulty, L. R. ———. 2001. The tornado: Nature’s ultimate windstorm. Lemon, and B. D. Lambert. 1980. Tornadoes. Norman: University of Oklahoma Press. When, where, how often: A study of 29 years of Grazulis, T.P., J. T.Schaefer, and R. F.Abbey Jr. 1993. tornado data. Weatherwise 33:52–9. Advances in tornado climatology, hazards, and risk Schaefer, J. T.,R. L. Livingston, F.P.Ostby, and P.W. assessment since tornado symposium II. In The tor- Leftwich. 1993. The stability of climatological tor- nado: Its structure, dynamics, prediction, and hazards, nado data. In The tornado: Its structure, dynamics, ed. C. Church, D. Burgess, C. Doswell, and R. Da- prediction, and hazards, ed. C. Church, D. Burgess, vies-Jones, 408–26. Geophysical Monograph 79. C. Doswell, and R. Davies-Jones, 459–66. Geo- Washington, DC: American Geophysical Union. physical Monograph 79. Washington, DC: Hart, J. 1993. SVRPLOT: A new method of access- American Geophysical Union. ing and manipulating the NSSFC severe weather Schaefer, J. T., S. J. Weiss, and J. J. Levit. 2003. database. Preprints, 17th Conference on Severe Local The frequency of severe thunderstorm winds over Storms, 40–41. St. Louis: American Meteoro- the contiguous United States. 11th International logical Society. Conference On Wind Engineering, Lubbock, TX, Johns, R. H. 1982. A synoptic climatology of north- 8 pp. west flow severe weather outbreaks. Part I: Nature Tessendorf, S. A., and R. J. Trapp. 2000. On the cli- and significance. Monthly Weather Review 112: matological distribution of tornadoes within quasi- 449–64. linear convective systems. Preprints, 20th Conference Johns, R. H., and J. S. Evans. 2000. Comments on ‘‘A on Severe Local Storms, Orlando, FL, American Me- climatology of derecho-producing mesoscale con- teorological Society, 134–73. 38 Volume 58, Number 1, February 2006

Thompson, R. L., and M. D. Vescio. 1998. The De- Wilson, J. W., and S. A. Changnon Jr. 1971. Illinois struction Potential Index—A method for compar- tornadoes. Circular 103. Champaign: Illinois State ing tornado days. Preprints, 19th Conference on Severe Water Survey. Local Storms, Minneapolis, MN. American Meteor- ological Society, 280–82. PHILIP W. SUCKLING is Professor and Chair of Tornado Project Online. 1999. Safety. Available on- the Department of Geography at TexasState Univer- line at: http://www.tornadoproject.com/safety/ sity–San Marcos, San Marcos, TX 78666. E-mail: safety.htm (last accessed 13 October 2005). [email protected]. His research interests involve cli- U.S. Department of Commerce. 1953. Tornado occur- matology and natural hazards. rences in the United States. Technical paper no. 20. Washington, DC: U.S. Government Printing Of- WALKER S. ASHLEY is Assistant Professor in the fice. Meteorology Program, Department of Geography at Weisman, M. L., and J. B. Klemp. 1986. Character- Northern Illinois University, DeKalb, IL 60115. istics of convective storms. In Mesoscale meteorology E-mail: [email protected]. His research interests involve and forecasting, ed. P.S. Ray, 331–58. Boston: Amer- synoptic and mesoscale climatology/meteorology, ican Meteorological Society. especially with respect to severe weather.