The authors analyze 105 yr (1901–2005) of tropical cyclone strikes at 45 coastal locations from Brownsville, Texas, to Eastport, Maine, with the primary objective of examining spatiotemporal patterns of storm activity. Interpretation of the data suggests that geographically, three focal points for activity are evident: south Florida, the Outer Banks of North Carolina, and the north-central Gulf Coast. Temporally, clusters of hyperactivity are evident in south Florida from the 1920s through the 1950s and then again during the most recent years. North Carolina was a region of enhanced activity in the 1950s and again in the 1990s. A more consistent rate of occurrence was found along the north-central Gulf Coast; the last two years, however, were active in this region. Return periods of tropical storm strength systems or greater range from a frequency of once every 2 yr along the Outer Banks of North Carolina, every three years on average in southeast Texas, southeastern Louisiana, and southern Florida, and about once every 10–15 yr in northern New England. Hurricane return periods range from 5 yr in southern Florida to 105+ years at several sheltered portions of the coastline (e.g., near Cedar Key, Florida, Georgia, and the northeastern seaboard), where some locations experienced only one strike, or no strikes through the entire period of record. Severe hurricane (category 3–5) return periods range from once every 15 yr in South Florida to 105+ in New England.
Tropical storms and hurricanes are among the most media-focused environmental hazards (Unger 1999). These events, especially major hurricanes such as Andrew or Katrina, result in catastrophic losses of life and property, they interrupt commerce, and they frequently cause dramatic geomorphic change within the coastal zone. Most societal impacts of tropical storms and hurricanes, however, are restricted to rather narrow zones along the coast, including major recreational beach resorts (i.e., Myrtle Beach, South Carolina; Ocean City, Maryland; and Pensacola, Florida), bays and lagoons, estuarine locations, ports (i.e., Miami, Florida; New Orleans, Louisiana; and even New York, New York), and residential areas. For society to better cope with, and plan for these events, it would be advantageous to understand the frequency and magnitude of tropical storms and hurricanes at or near the cities in question along the coast.
One of the earlier efforts to produce hurricane return periods was provided by Simpson and Lawrence (1971), based on hurricane strikes from 1886 to 1970. They analyzed strikes and return periods for all hurricanes and major hurricanes within 80-km coastal segments. For all hurricanes (categories 1–5), return periods ranged from 6 years in southeastern Florida to 85 or more years (no strikes at all) along coastal segments in Georgia, New Jersey, and along the New England coast. A more recent effort was presented by Elsner and Kara (1999) based on strikes from 1900 to 1996. They focused on return frequencies along the coast for each county. Hurricane return periods (all categories 1–5) ranged from 4 years for Monroe County, Florida (Florida Keys), to 97+ years in Georgia and Maine. However, the estimated return periods were somewhat affected by the varying lengths of coastline in each county. For example, the coastline of Monroe County extends approximately 275 km from Key Largo to Everglade City. In contrast, Charlotte County, Florida, also located in southern Florida on the Gulf Coast, has a coastline of only 19 km. Without accounting for the varying lengths of coastline, the aforementioned approach is biased toward giving longer county coastlines more opportunity to experience an event, thereby reducing the length of the return periods.
In this paper, we present a point-based analysis that removes this bias. Although there is utility in these previous research efforts and their applications, we contend that an analysis of tropical cyclone return periods at specific points, rather than sectors or county-length segments of coastline, yields results with even greater utility for local building codes. Furthermore, results from this study are more applicable to understanding the hurricane climatology and its geographic variability along the East and Gulf Coasts of the United States.
The objectives of this research are as follow:
To investigate tropical storm and hurricane strikes for specific coastal places (towns or cities) rather than sectors of the coasts
To depict the geographical and temporal patterns (1901–2005) of tropical storm and hurricane strikes along the entire East and Gulf Coasts of the United States
To calculate return periods for tropical storms, hurricanes, and severe hurricanes
This spatiotemporal approach is an extension of Muller and Stone (2001) by increasing its temporal (updated from 2000) and geographical extension (from Cape Hatteras, North Carolina, to Maine) of tropical storm and hurricane strikes. The combination of a geographical and temporal analysis provides a unique picture of when and where certain portions of the coastline were most active with respect to tropical storms and hurricanes. This research may also shed some light on the debate surrounding the impacts of global warming and the Atlantic Multidecadal Oscillation (AMO) on landfalling tropical storms and hurricanes.
We examined temporal and geographical variability in tropical storm and hurricanes strikes at 45 places along the Gulf of Mexico and Atlantic Coasts of the United States from Texas to Maine (Fig. 1). These places are not evenly spaced along the coast, but instead, places were selected that are well known regionally as cities, or famous beach resorts along barrier islands. Better-known inland cities are also noted in the figure legend, where appropriate. We anticipate that return periods will be similar, but somewhat longer for a single point than for 80-km segments of coast, or for the coastline of an entire county.
To evaluate storm strikes at various places along the coast, we needed to adapt standard estimates of the average sizes of storms by intensity. Dunn and Miller (1960), Simpson and Riehl (1981), and Merrill (1984) all agree that there is considerable variability in the size of tropical storms and hurricanes, and spatial dimensions are not well correlated with storm intensity. Category-5 storms can be large or small, and this is also true for significantly weaker tropical storms. Pielke and Pielke (1997) provided examples of the varying sizes and diameters of storms, and display the tendency for storm conditions to extend farther to the right than the left of the storm focal point (eye) along the storm track (Fig. 2). The above studies, however, suggest that for an “average” hurricane, hurricane-force winds extend forward and to the right about 50 to 100 km from the eye and 25 to 50 km to the left.
From these estimates of storm properties, coastal storm conditions can be generalized by storm categories, while taking into account that Northern Hemisphere storms are more extensive and severe on the right side of the storm and less developed on the left-hand side. Our model, depicted in Fig. 3, is a relatively conservative representation of average-size storms. The model assumes that with tropical storms, tropical-storm-force winds extend 80 km to the right and 40 km to the left of the track. For category-1 or -2 hurricanes, hurricane conditions are assumed to extend 80 km to the right, and tropical storm conditions an additional 80 km to the right, with hurricane conditions 40 km to the left and tropical storm conditions an additional 40 km to the left. Similarly, for major category-3–5 hurricanes, major hurricane conditions are assumed to extend 80 km to the right, weaker hurricane conditions an additional 80 km to the right, and tropical storm conditions another 80 km to the right, with the respective increments of 40 km each to the left. Hence, for storm strike identification at the 45 points, tropical storms represent a swath 120 km wide along a coast perpendicular to the storm track, category-1 or -2 hurricanes represent a hurricane swath 120 km wide with tropical storm conditions an additional 80 km to the right and 40 km to the left. For major hurricanes the hurricane intensity swath is 240 km wide, again with tropical storm conditions an additional 80 km to the right and 40 km to the left.
The model is applied to tropical storm and hurricane tracks and intensities taken from the 6-hourly data from the National Hurricane Center (NHC) for each storm; these data are available on several Web sites (i.e., www.wunderground.com/tropical/). Strike intensities at the 45 coastal locations are estimated from reported maximum sustained winds when approaching centers were within 80 km of the coastline. This is consistent with our model in Fig. 3 where estimated storm conditions extend outward and forward in the right-front quadrant 80 km. This identification of strike intensity at the coastal sites tends to be a conservative estimate of storm impacts because of very destructive storm surges that are generated by storm intensities well offshore. A tragic example is the record 28-ft storm surge at Bay St. Louis, Mississippi, during Hurricane Katrina; at landfall in Mississippi, Katrina was classified a category-3 storm, but the record surge obviously was a response to its category-5 status for 24 h when the eye of the unusually large Katrina was 250 km and 24 h from landfall (Knabb et al. 2005). In this same report, the authors state that between 1986 and 2005 all 11 hurricanes approaching the central Gulf Coast with minimum pressure less than 973 mb weakened significantly during the final 12 h before landfall. For example, Hurricane Lili off the Louisiana coast in 2002 was a category-4 storm when the center was approximately 250 km off the coast. Within 100 km of the coast, Lili had rapidly weakened to a category-2 storm, before making landfall as a category 1. More recently, Hurricane Rita also weakened substantially before making landfall in western Louisiana during the 2005 season.
The focus of this analysis is primarily on landfalling tropical cyclones, or at least those that traversed close enough to coastal communities to fall within the constraints of the model noted above. The record for tropical cyclones in the North Atlantic Basin dates back to the nineteenth century (Neumann et al. 1993) and even earlier (i.e., Ludlum 1989). However, this record has large inhomogeneities over time, which complicates interpretations of long-term trends and variability in frequencies and intensities (Solow and Moore 2002). Empirical observations of pressure provide a fairly reliable record of landfalling storms in the United States back to around the turn of the twentieth century (Landsea et al. 1999). As a result, our dataset begins in 1901 and extend through 2005.
3. Spatiotemporal strike patterns
The National Hurricane Center dataset listing tropical and hurricanes tracks for the entire Atlantic Basin includes 421 tropical storms and 555 hurricanes, a grand total of 976 cyclones for the 105 yr between 1901 and 2005 (www.wunderground.com/tropical/). For each storm track, pressure, wind speed, Saffir–Simpson category, and latitude–longitude are given at 6-h intervals. The same dataset also provides Saffir–Simpson categories at landfalls in the continental United States. For these years, the dataset shows 354 landfalls, 176 at tropical storm status, 111 category 1 or 2, and 67 major hurricanes, categories 3–5. Our analysis of strikes at the 45 coastal locations includes 194 individual tropical storms, 117 category-1 and -2 hurricanes, and 57 major hurricanes for a grand total of 368 storms, 14 more than in the National Hurricane Center dataset. The reason for the discrepancy stems from storms that tracked near the coast, but did not make landfall. Our analysis includes those storms that grazed the coast and fell within the domain of our model, but were not counted as landfalls by the National Hurricane Center.
Using the criteria noted above, Fig. 4 provides a unique assessment of tropical storm and hurricane strikes along the Gulf and Atlantic Coasts by merging time, geography, and storm intensity. Based on these 45 time series, there are three broad geographical areas that are highly active: 1) the northern Gulf Coast from Galveston, Texas, to Panama City Beach, Florida, 2) south Florida from Marco Island on the Gulf Coast around to Vero Beach on the Atlantic Coast, and 3) North Carolina, especially the Outer Banks, from Wrightsville Beach to Nags Head. Two intervening coastal sections exist with noticeably fewer strikes from Apalachicola, Florida, southeastward to south of Cedar Key, Florida, on the Gulf Coast, and on the Atlantic Coast from northern Florida northeastward to include most of the Georgia coast and centered especially on St. Simons Island, in the vicinity of Brunswick, Georgia. From Virginia Beach northeastward, tropical cyclone strikes become increasingly uncommon, except for a modest increase along the more-exposed coasts of eastern Long Island (Montauk Point), Rhode Island (Newport), and Cape Cod (Chatham). The model indicates no major (category-3 and above) hurricane strikes north of Cape Hatteras, no hurricane strikes at Ocean City, Maryland, and only one hurricane strike, in 1903, at Atlantic City and Sandy Hook in New Jersey. Because storm tracks are most often northeastward and parallel with the coast north of Cape Hatteras, in Fig. 4 it is evident that the same storm can result in strikes at many coastal locations for more than 1500 km from Cape Hatteras to Eastport, Maine. Notable examples include the 1943 hurricane, 1954 Hurricane Carol, 1960 Hurricane Donna, 1996 Hurricane Bertha, and 1999 Hurricane Floyd. Although hurricane frequencies decrease northward from Cape Hatteras there is an increasing frequency of potentially damaging winter and spring midlatitude cyclones, or nor’easters; often 10 to 20 per season (i.e., see Hirsch et al. 2001).
In Fig. 4 the most dramatic interdecadal variability has occurred in southern Florida on both east and west coasts; Vero Beach on the Atlantic around to St. Petersburg and Cedar Key on the Gulf of Mexico. There is clustering of events for the 25 yr between the mid-1920s and 1950, and several events during the mid-1960s. However, with the exception of Hurricane Andrew in 1992, there are almost no hurricane strikes, and only a few tropical storm strikes over the most recent 40 yr until 2004 and 2005. Along the Outer Banks of North Carolina, there has also been two temporal clusters of strike events, in the 1950s and again in the 1990s, with relatively few strikes in between. Figure 4 also suggests that there are clusters of strikes in one region that may be associated with infrequent strikes in adjacent regions. Examples include the clusters of strikes in southern Florida, 1903–10, 1924–50, and 1964–67, as contrasted to infrequent strikes along the eastern half of the northern Gulf Coast for the same runs of years. Similarly, the clusters of strikes along the Outer Banks of North Carolina during the 1950s and 1990s can be contrasted with very inactive years over southern Florida. Figure 4 also shows how active the Central Gulf and Florida coasts have been in 2004 and 2005.
The dataset in its totality depicts a temporal pattern beginning in the mid-1920s, extending into the mid-1960s, that was very active for landfalling storms in the United States. A relatively inactive period is evident from the mid-1960s through the mid-1990s, which again becomes active from 1995 to 2005. These patterns appear consistent with documented fluctuations in the AMO (Elsner 2006; Vermani and Weisberg 2006), though the longevity and destructiveness of storms as noted by Emanuel (2005) is not taken into consideration here. However, despite the fact that U.S. landfalling hurricanes only comprise 25% of North Atlantic hurricanes and 3% of global hurricanes (Curry et al. 2006), these results seem consistent with Landsea (2005).
4. Storm return periods
Return periods at the 45 coastal locations for the 105-yr period (1901–2005) are rounded to the nearest year (Fig. 5). These average return periods are derived by dividing the 105-yr period of record by the total number of strikes at the respective locations. We acknowledge, however, that any location can have multiple strikes in one year, and none for several consecutive years, and the variations in the number of storms per annum can easily be modeled using the Poisson distribution.
In Fig. 5, return periods for all tropical cyclones are represented in the inner tier; the middle tier for all hurricanes; the outer tier only for major hurricanes—categories 3, 4, and 5. Tropical cyclone return periods are as low as once every 2 yr on the average along the Outer Banks of North Carolina, once every 3 yr along the north-central Gulf Coast as well as in southern Florida from Cocoa Beach to St. Petersburg. Tropical cyclone return periods are 10 yr or more in southern Georgia, New Jersey, and northern New England.
For all hurricanes, return periods are shortest, about 5 yr, in southern Florida from Palm Beach to Key West, and the Outer Banks of North Carolina, and 10 yr or less along the northern Gulf Coast from Galveston eastward to Apalachicola with the exception of a 15-yr return period at Cameron, in southwestern Louisiana, just east of where Hurricane Rita made landfall in 2005. Less-frequent hurricane strikes and much longer return periods are evident along the northeastern Gulf of Mexico in the vicinity of Cedar Key, Florida (35 yr), and 52 yr at St. Simons Island on the coast of Georgia, with both coastal regions often sheltered from direct storm strikes by the configurations of adjacent coastlines. North of Nags Head, North Carolina, return periods for all hurricanes range between 21 and 35 yr at the more exposed locations between Jones Beach, near the western end of Long Island, and Chatham, Massachusetts, on Cape Cod. Return periods extend to 52 or more years at more sheltered locations in this region. Our strike model indicates no hurricane strikes at Ocean City, Maryland, for the 105-yr record starting with 1901.
For major hurricanes, return periods are shortest, 13–18 yr on average, in southern Florida from Marco Island on the Gulf to Palm Beach on the Atlantic, and 21 to 26 yr from Galveston eastward to Pensacola Beach; exceptions to these patterns are at Cameron and Gulfport, which is somewhat protected by the delta of the Mississippi River. Major hurricane return periods are 35 yr at both Cape Hatteras and Wrightsville Beach in North Carolina. The strike model indicates no major hurricane strikes during the 105 yr period at Port Aransas, Texas (Corpus Christi), Apalachicola, Florida, from Cocoa Beach, Florida, northward to Tybee Beach (Savannah), Georgia, except at Jacksonville Beach, and from Nags Head, North Carolina, to the Canadian border. Although the National Hurricane Center has reported that the Hurricane of 1938 came ashore in New England as a category 3 (http://www.nhc.noaa.gov/pastint.shtml), the 6-hourly data show that the storm “weakened” to less than category 3 approximately 500 km prior to landfall, to a category-2 hurricane approximately 200 km out, and then to a category-1 storm shortly after making landfall. Based on these data, we conclude that this hurricane made landfall as a minimal hurricane, though the system generated a surge indicative of a much stronger storm (Zielinski and Keim 2003), similar to the large storm surge associated with Hurricane Katrina.
A comparison of hurricane return periods as presented in the present study with Elsner and Kara (1999) and Simpson and Lawrence (1971) shows modest differences between datasets (Fig. 6). Note that the periods of record differ in the three respective studies (1901–2005, Keim et al.; 1900–96, Elsner and Kara; and 1886–1970, Simpson and Lawrence). The common period for the three studies is the 70 yr between 1901 and 1970. Simpson and Lawrence include a 15-yr period between 1886 and 1900, when hurricane activity was 5.7 yr−1, slightly above our long-term average of 5.3 yr−1 between 1901 and 2005, and excludes the 35 yr beginning with 1971 when the average was 6.1 yr−1. Elsner and Kara do not include the 9 yr between 1997 and 2005, when the landfall average was especially high at 8.2 hurricanes per year. Additionally, for locations with infrequent strikes, one new strike changes the return period substantially; for example, one strike between 1901 and 1999 yields a return period of almost 100 yr, whereas a second strike in 2000 reduces the return period to 50 yr. Because of the differences between the study periods, Simpson and Lawrence (1971) show return periods that are slightly longer, with a few exceptions, than for the two more recent studies.
Given that our dataset covers a more similar time span to Elsner and Kara (1999), we restrict comparisons to all hurricane strikes in these two studies. For the 45 points, our return periods are longer than Elsner and Kara at 26 locations, the same at 5, and shorter at 14. Having more locations with longer recurrence intervals in our study was expected because our point analysis is a smaller target than an entire county length of coastline. For the 33 locations where our return periods are 26 yr or less, four hurricane strikes in 105 yr, we average 1.3 yr longer than Elsner and Kara (1999), more than 10% longer. Almost all of the locations where our return periods are shorter (more frequent storms) than Elsner and Kara (1999) occur along the northern Gulf of Mexico and southern Florida where there were more frequent hurricane strikes between 1997 and 2005. We also compute longer return periods at all of the locations north of Nags Head, where Elsner and Kara (1999) identify more frequent strikes than we found using our model.
The application of the return-period data for future tropical storm and hurricane strikes assumes stationarity—a more-or-less random distribution of strikes through the decades. However, perusal of the temporal patterns of strikes for individual places in Fig. 4 indicates significant variability through the decades. In addition, there is strong suggestion that the recent pattern of enhanced storm frequencies, beginning in 1995, may persist for another decade or more (Goldenberg et al. 2001) as it is probably related to the shifts in Atlantic sea surface temperature and surface pressure anomalies over multidecadal time scales (Landsea et al. 1999).
5. Tropical hazard index
It is generally accepted that wind power increases nearly as a cubic function of wind speed (Stadler and Hughes 2005). As a result, even modest increases in wind speed (of only a few percent) can result in dramatic increases in potential storm damage (Dorland et al. 1999). Realized damage, however, is not only dependent upon wind speed, wind duration, and wind power, but also on the type and density of development being exposed to storm winds, and with tropical storms and hurricanes, storm surge is usually the most destructive component of the storms. Furthermore, even the threat of a landfalling storm can cause major disruptions in coastal communities in the form of preparation of homes for the storm, loss of electricity, and phone service, business closings, as well as evacuations. Using these basic principles of wind and potential wind and surge damage, compounded by the level of disruption to society, a simple tropical hazard index (THI) is developed to provide an additional geographic perspective on the vulnerability of the Atlantic and Gulf Coast regions. For the entire 105-yr period of analysis, strikes of tropical storm intensity are assigned two points, category-1 and -2 hurricanes four points, and major hurricanes (categories 3–5) eight points. This provides a relative index for the total exposure of a coastal community to storms, taking into account both the frequency and intensity of storms. We fully realize the exponential component of potential wind damage as wind speeds increase; however, given the total disruption caused by storms (including preparation, damage, and other inconveniences), we feel that a weak hurricane represents twice the nuisance of a tropical storm, and severe hurricane is twice the inconvenience of a weaker hurricane. We chose two, four, and eight points rather than one, two, and four so that summed totals over 100 would stand out to denote exceptional locations. We stress that the THI represents a very simple comparative geographical index of the storm hazards among coastal regions, and the THI should neither be related to specific sites along the coasts nor to any one of the environmental and socioeconomic impacts (Pielke and Pielke 1997).
THI values range from highs of 152 at Key Largo, Florida, and 150 at Cape Hatteras, and 130 at Boothville, south of New Orleans. In terms of tropical storm and hurricane strikes, the count at Cape Hatteras is 47, at Boothville 41, and Key Largo only 35. The highest THI at Key Largo results from its exposure to major hurricane strikes, eight, mostly coming westward from the Atlantic and Caribbean, and the very warm water offshore. Cape Hatteras is very exposed to the tracks of tropical storms and hurricanes moving around the western margins of the subtropical, Bermuda high, but somewhat cooler surface waters help to diminish the intensity of these tropical cyclones as they progress into middle latitudes. Boothville is also in an exposed warm-water location on the northern coast of the Gulf of Mexico, but many strikes here are due to tropical storms and category-1 and -2 hurricanes that develop over the southern Gulf of Mexico, often without opportunities to further intensify during the shorter tracks over the Gulf to landfall. In South Florida from Sanibel Island around to Vero Beach, THI values are all over 100, as are all four North Carolina places, as well as along the central Gulf Coast from Galveston, Texas, eastward to Destin, Florida, with the unexplained exception of Cameron, Louisiana, where the THI is 84, increased in 2005 by the major Hurricane Rita strike.
The lowest THI values (20) occur along the Maine coast (Fig. 7), and all of the places northward from Ocean City, Maryland, have THI scores less than 50, due to the obvious impacts of cooler ocean waters on storm intensities at these middle-latitude locations. Other places with lower THI scores include sheltered portions of the coastlines as they relate to the preferred parabolic shape of storm tracks around the Bermuda high over the North Atlantic (see Neumann et al. 1993). These places include Atlantic City (22) and Sandy Hook, New Jersey (26), St. Simons Island, Georgia (44), and Cedar Key, Florida (66).
6. Summary and conclusions
This paper provides a unique view of tropical cyclone strikes along the Gulf and Atlantic coasts by merging time, geography, and storm intensity. It also provides a potentially important operational baseline for assessing return periods of tropical cyclones along the Gulf and Atlantic Coasts. One of the primary differences between this research and that of Elsner and Kara (1999), and Simpson and Lawrence (1971), is that our study incorporated an analysis of geographical points (mostly famous beaches) rather than segments (e.g., county lengths) of coastline. Our data also suggests that the active period of hurricanes from the late 1920s through the mid-1960s, and again from 1995 to 2005, likely has an association with positive anomalies in sea surface temperature and the Atlantic Multidecadal Oscillation mode. In addition, this study also depicts temporal shifts in favored landfall areas, that is, in south Florida from the late 1920s to 1950, North Carolina in the 1950s, and south Florida during the early 1960s. The increase in frequency and intensity of events beginning in 1995 is also evident, with a penchant for North Carolina early in this period, as well as the north-central Gulf Coast and the central coasts of the Florida peninsula in the latter portion. The estimated return periods that are derived from these 45 time series should also have utility for purposes of planning within these coastal communities.
Corresponding author address: Barry Keim, Department of Geography and Anthropology, Louisiana State University, Baton Rouge, LA 70803. Email: firstname.lastname@example.org