1. Introduction

In the United States, the impacts from tropical cyclone winds often extend well inland after making landfall along the coast. For example, after the passage of Hurricane Camille (1969), more than 150 deaths occurred in the state of Virginia, some 1300 km inland from where the storm originally made landfall along the Louisiana coast (Emanuel 2005). According to Rappaport (2000), a large portion of fatalities often occurs inland, associated with a decaying tropical cyclone’s winds (falling trees, collapsed roofs, etc.) and heavy flooding rains. In the 1970s, 1980s, and 1990s, freshwater floods accounted for 59% of the recorded deaths from tropical cyclones (Rappaport 2000), and such floods are often a combination of meteorological and hydrological factors. Moisture from tropical cyclones occasionally merges with eastward-moving continental low pressure systems, producing copious amounts of rainfall inland from the coast. Tropical cyclone–induced flooding has devastated communities located many hundreds of miles from the coast (Gibney 2000). Recent examples of storms that caused significant inland impacts include Hurricane Hugo (wind and rain) in 1989, Hurricane Floyd and Tropical Storm (TS) Allison (flooding) in 1999 and 2001, respectively, and Hurricane Fran (wind) in 1996. Each of these storms reveal how the destructive forces of tropical cyclones can impact areas far from the initial landfall point.

Strong winds associated with tropical systems usually diminish quickly once they move ashore, primarily because of the frictional effects of land-based obstructions (topography, forests, urbanized areas) and thermodynamically through a loss of heat energy from the ocean’s surface (Friedman 1975). Emanuel (2005) suggests that the decay rate of a land falling hurricane is rapid, losing half its wind speed value in roughly 7 h, 75% in 15 h, and nearly 90% after just 1 day inland. In numerical simulations, Tuleya et al. (1984) found that more intense hurricanes decay faster upon landfall than do weaker storms. For these reasons, the primary impact areas of tropical cyclones are generally found along coastal (or near coastal) regions.

The National Oceanic and Atmospheric Administration’s (NOAA) National Hurricane Center (NHC) currently develops and maintains a database that tracks those coastal counties that have been impacted by hurricane strikes (see http://www.nhc.noaa.gov/aboutpubs.shtml?#MEMO), based on historical landfall records, storm surge reports, and the tropical cyclones’ wind field (Jarrell et al. 1992). In another study, Keim et al. (2007) produced return periods for hurricane strikes along the U.S. coastline from Texas to Maine. A similar dataset can also be found in Elsner and Kara (1999, their Fig. 8.15), where the coastline east of the Rocky Mountains was subdivided into 2.5° grid boxes and frequency counts of hurricanes appearing in each grid box were tallied. These datasets are limited to the single coastal county or grid box bordering the Gulf of Mexico and/or the Atlantic Ocean. Therefore it is the intent of this study to fill this large gap in our knowledge concerning the inland impacts from historical tropical cyclones.

Most previous studies involving the inland extent of tropical cyclones have generally focused on their expected or modeled rate of decay postlandfall (e.g., Tuleya et al. 1984; Kaplan and DeMaria 1995, 2001), while others have focused on recurrence thresholds or probabilities of landfalls along a given portion of the U.S. coastline (e.g., Bove et al. 1998; Elsner and Bossak 2001; Gray and Klotzbach 2009; Saunders and Lea 2005). Results from Kaplan and DeMaria (1995) showed an idealized scenario for the maximum possible inland wind speed of a decaying tropical cyclone based on both intensity at landfall and forward motion for the Gulf Coast and southeastern United States, and for the New England area (Kaplan and DeMaria 2001). In general they found that for inland locations, the effect of storm motion speed is just as critical as the storm intensity upon landfall, such that a fast-moving category-3 storm on the Saffir–Simpson Scale [hereinafter the Saffir–Simpson Hurricane Wind Scale (SSHWS); Simpson 1974] will have additional inland impacts than a slow-moving category-5 storm.

More recently, Zandbergen (2009) examined the inland exposure of U.S. counties to tropical storm– and hurricane-force winds. It was shown that the inland impacts from tropical cyclones extended across much of the Deep South and into portions of the Ohio valley and New England. Local maxima were evident across much of Florida, eastern North Carolina, and the southern Gulf Coast states. While these results appear realistic at first glance, further investigation reveals various concerns with the methodology that was used to derive the results—namely the use of a symmetric rather than an asymmetric wind swath to estimate the total spatial envelope of a storm’s impact. A more thorough discussion on the differences in methodologies between the Zandbergen (2009) article and those expressed herein follows in section 2.

Knowledge of the climatologically favored areas for hurricanes, as provided in the aforementioned studies, combined with a climatological expectation of the inland penetration frequency of these storms can be of tremendous value. Thus, there is a need to accurately document and understand the inland extent of winds associated with tropical cyclones to enhance public awareness, improve forecasting of hazardous conditions, and in turn save lives. To that end, a comprehensive climatology of the inland extent of tropical cyclone winds over the eastern United States was developed at NOAA’s National Climatic Data Center (NCDC). Section 2 describes the data and methods used to construct the climatology, while section 3 discusses the results of the new climatology. The paper concludes with a summary and discussion of the results.

2. Data and methodology

Historical track data for the North Atlantic were obtained from the International Best Track Archive for Climate Stewardship (IBTrACS; Knapp et al. 2010; Kruk et al. 2010). The IBTrACS dataset is a comprehensive collection of all recorded tropical cyclones worldwide, and the Atlantic basin portion of the record is derived from NOAA’s Atlantic Hurricane Database (HURDAT; Jarvinen et al. 1984). In addition, the so-called extended best-track data were also obtained for the North Atlantic (Demuth et al. 2006). Data for the extended best track begin in 1988 and contain not only the time, position, and intensity of the tropical cyclone at 6-h intervals, but also the maximum extent of winds (MEW) of a given threshold (34-, 50-, and 64-kt (1 kt ≈ 0.5 m s⁻¹) sustained wind averaged over 1 min at 10-m height) in four quadrants: northeast, southeast, southwest, and northwest. These wind radii values are the result of operational estimates and became part of the official North Atlantic best-track data after 2004.

The first step toward producing an inland tropical cyclone wind impact climatology was to compute the average distance for each MEW threshold (wind radius) using the extended best-track dataset. This was done by differentiating those storms that were solely tropical (denoted by an asterisk in the dataset) or extratropical¹ (denoted by an “E” in the dataset). Then, for each wind radius and quadrant, a mathematical average was computed and binned by the SSHWS, and included post-tropical storms. This resulted in a matrix of average distances (nm) for each wind radii, SSHWS intensity, and quadrant. However, this matrix was unsuitable for computing the average inland extent of tropical cyclone winds for two reasons.

First, using all data provided in the extended best-track dataset produces an average wind radii distance that is composed of mostly overwater observations and fewer inland observations. Applying this biased result to inland locations would result in inflated frequency counts since the lack of surface friction at sea produces larger wind swaths than those over land. Therefore, it was necessary to recalculate the average wind radii distances by extracting only those portions of storm tracks that were located over land. This also included an extension of the coastline 150 nm out to sea to ensure that coastal counties were included when the spiraling winds around an approaching tropical cyclone were already impacting the coastline. The use of the 150-nm extension is consistent with the approach as used by Ho et al. (1987) and Schwerdt et al. (1979) to assess the climatological characteristics of decaying tropical cyclones.

The results of the average wind radii distances over land by SSHWS intensity for each quadrant and MEW threshold are shown in Table 1. In short, the largest sample size was associated with tropical storms (386), while category-4 and category-5 storms had the fewest number (22). Interestingly, the standard deviation in the computed average wind radii distances (not shown) was highest in the northeast direction, regardless of intensity. For example, for the 34-kt threshold in the northeasterly direction, the standard deviation for post-tropical storms was 139 nm, while it was 64 nm in the southwesterly direction. At the same time, for the 64-kt threshold in the northeasterly direction, the standard deviation for category-1 storms was 28 nm, while it was 11 nm in the southwesterly direction. Reasons for this pattern across the range of intensities are unknown and are beyond the scope of this paper.

Second, geographical information system (GIS) software was used to obtain results, owing to its capability to graphically depict hurricane tracks and the availability of a number of analysis tools, including the use of a “buffering” utility. A GIS is “an integrated collection of computer software and data used to view and manage information about geographic places, analyze spatial relationships and model spatial processes. A GIS provides a framework for gathering and organizing spatial data and related information so that it can be displayed and analyzed” (Wade and Sommer 2006). Buffering in GIS refers to the construction of a polygon surrounding a point, line, or polygon at a specified distance. In our case however, the buffering operation requires that the tropical cyclones’ wind swath around a best-track data point be represented as a “left side” and “right side” only (relative to the storm’s forward motion) and not in four distinct quadrants. To fit this model, a simple average was computed on the distance estimates in Table 1 to derive the left and right buffers for each SSHWS intensity, time step, and wind radius. The right-front and right-rear quadrants of the tropical cyclone compose the right component of the buffer, while the left-front and left-rear quadrants are assumed to be the left component. Table 2 shows the result of this conversion process.

To accurately depict storm asymmetry, a general assumption was applied such that the left side is an average of the northwest and southwest quadrants, and the right side is the average of the northeast and southeast quadrants. This method was selected because it was a straightforward way to obtain results that adequately reflected the climatological mean. Note that the average distances are quite large for post-tropical storms at the 34-kt threshold, but drop off rapidly at the 50- and 64-kt thresholds. This is due to the very small number of post-tropical storms in the extended best-track dataset that had an average wind speed greater than 50 kt (n = 5). The category-3 hurricane, on average, has the largest wind radii of any storm intensity. Category-4 and category-5 storms were lumped together because of the relatively small sample size over land in the extended best-track dataset. The net result of this derivation is that a climatological asymmetry to the tropical cyclone wind field is realized.

As stated earlier, a related study on the exposure of U.S. counties to tropical storms and hurricanes for the Atlantic basin was conducted by Zandbergen (2009). In that article, GIS was also used in which a symmetrical buffer around the track of the tropical cyclone, as derived from an inland wind decay model, was used to deduce those counties that were influenced by the tropical cyclone. While the use of symmetrical buffers in GIS is convenient, the typical asymmetry of tropical cyclones (with, in the absence of storm motion, the strongest winds occurring in the right-rear quadrant; Elsner and Kara 1999) demands a more robust technique than offered above. In addition, Zandbergen (2009) used a period of record of 1851–2003 for hurricane tracks, while the methodology discussed above includes a period of record beginning in 1900 through the 2008 season. Although the historical best-track data prior to 1900 have been updated by Fernandez-Partagas and Diaz (1996), Zandbergen (2009) removed the overwater portions of all hurricane tracks prior to 1944, yet kept the landfall segments of those early tracks. As Ludlum (1963) points out, the intensity and location of hurricanes at landfall were highly subjective in the early portion of the record since instruments to measure wind speeds were not widespread until the late 1870s. Therefore, our use of the period 1900–2008 removes the early portion of the record and includes more recent memorable storms such as Katrina (2005), Rita (2005), and Ike (2008), all of which produced significant inland impacts.

3. Results

While the use of frequencies is beneficial in ascertaining how often a particular region was impacted by tropical storm– or hurricane-force winds, return intervals are often referred to instead. In this regard, rather than acknowledging a single grid cell having 20 “hits” over the course of the 108-yr period, or 0.185 storms per year, it is often more meaningful to say that grid cell experiences tropical storm–force winds once every 5.4 yr (simply the inverse of the ratio of frequency count over the period of interest). In the discussion that follows, return intervals will be used to summarize the historical inland wind climatology.

Return intervals for post-tropical winds greater than or equal to 34 kt are shown in Fig. 4. The map shows that nearly every state east of the Rocky Mountains has experienced winds associated with a post-tropical cyclone. The most favored areas include the Carolinas northeastward through Virginia and into New England. In these locations, return intervals are one event every 2–5 yr. In more inland areas, the return intervals drop off rapidly northwestward from the coasts. For example, the Ohio valley generally experiences post-tropical storm–force 34-kt winds once every 6–10 yr, while parts of northwestern Missouri do once every 50–100 yr. Other minimums in post-tropical storm-force 34-kt winds are the Louisiana and south Florida coastlines, where such events occur once every 20–30 yr.

Figure 5 is similar to that of Fig. 4, except that it is an examination of 34-kt winds solely from tropical (and not post-tropical) cyclones (Fig. 5a), and all hurricanes with winds greater than 64 kt (Fig. 5b). The figure shows that much of the southern Gulf Coast and Eastern seaboard is regularly affected by 34-kt winds from a tropical cyclone (i.e., one event every 2–5 yr), and 64-kt winds from hurricanes once every 3–5 yr. The return intervals decrease in the New England area, where tropical storm–force (hurricane force) winds are experienced once every 6–10 yr (11–20 yr). The map shows that, in general, warm-core tropical cyclones rarely make it farther north and west than the intersection of 40°N and 85°W.

The distribution of hurricane-force winds is further dissected, splitting them into nonmajors, category-1–2 hurricanes (Fig. 6a), and majors, category-3–5 hurricanes (Fig. 6b). For the nonmajor hurricanes, several distinct maxima are evident, particularly along eastern North Carolina, southeastern Texas, and much of the southernmost Gulf Coast states including nearly the entire state of Florida. In these areas, return intervals are one nonmajor hurricane every 3–10 yr. The return intervals drop off markedly with inland extent, again signaling the often rapid decay associated with landfalling tropical cyclones. For the major hurricanes, the areal distribution is less, owing to their relative low frequency of occurrence (at landfall), however several important inferences can be made. The highest return intervals exist largely across the southern portions of Florida (one major hurricane every 5–10 yr), with a large area of return intervals between 11 and 20 yr blanketing a good portion of the southern Gulf Coast states. Extreme eastern Maine and Massachusetts also experience hurricane-force winds from major hurricanes once every 51–108 yr. This return interval also extends southward into the Carolina piedmont regions.

For those locations that are not shaded, the maps imply that, in the recorded history, these regions have not experienced tropical storm- or hurricane-force winds. For example, in Fig. 6b, southern Georgia is not shaded (i.e., no major hurricanes have crossed those inland areas in the last 108 yr), but this is realistic owing to the tendency for storms to either recurve away from the coast (Colon 1953; George and Gray 1977; Hodanish and Gray 1993; Evans and McKinley 1998; Elsner et al. 2000) or rapidly dissipate after landfall. However, this does not suggest tropical cyclone winds cannot occur in the future. Moreover, while the areas shaded in dark blue may have only experienced one storm over the course of the 108-yr period used in this study, the inferred risk in these areas is nonzero, owing to the long-term climatological mean that indicates that inland impacts are possible, though rare, events. Again, these maps are a climatological guide for inland wind impacts from tropical cyclones, and a zero percent occurrence does not mean “never going to happen,” but that it is “highly unlikely.”

To some extent, geography helps explain the heightened exposure of certain regions of the U.S. mainland. Areas that jut out into the Atlantic and Gulf of Mexico are more likely to be affected by North Atlantic and Gulf tropical cyclones. Florida, eastern North Carolina, and Massachusetts are examples of areas that frequently find themselves in the path of tropical storms and hurricanes simply because of the seaward extent of their land area (Gibney 2000).

4. Summary

Tropical cyclones pose a significant threat to life and property along coastal regions of the United States. As these systems move inland and dissipate, they can also pose a threat to life and property, through heavy rains, high winds, and tornadoes. While many studies have focused on the impacts from tropical cyclones on coastal counties of the United States, there was a need for a detailed climatology of the inland penetration of tropical cyclone wind fields.

NOAA/NCDC has developed a comprehensive climatology of the inland winds from tropical cyclones for the eastern United States. This was done by using the historical Atlantic basin track data (Jarvinen et al. 1984) in concert with the historical extended best-track dataset (Demuth et al. 2006) to compose an average maximum extent of the winds at 34-, 50-, and 64-kt thresholds according to storm intensity. A unique storm-relative asymmetrical buffer was generated for each storm and was overlaid on a 30-km equal-area grid using GIS to depict those regions of the eastern United States that have historically been impacted by tropical cyclones.

The results indicate that nearly every state east of the Rocky Mountains has experienced 34-kt winds associated with a post-tropical cyclone. The most favored areas include the Carolinas northeastward through Virginia and into New England. However, for warm-core tropical systems, much of the southern Gulf Coast and Eastern seaboard states are regularly affected by 34-kt winds from a tropical cyclone (one event every 2–5 yr). The return intervals decrease in the New England area, where tropical storm–force 34-kt winds are experienced once every 6–10 yr. For hurricane-force 64-kt winds, several distinct maxima were apparent, primarily eastern North Carolina, central and northwestern Florida, and southeastern Texas. There is an apparent minimum in hurricane-force winds in central Georgia as well as the states of Pennsylvania and New York. The relative minimum regarding the inland penetration of 64-kt winds from these strong hurricanes implies that they decay rapidly after landfall. The results indicate that warm-core tropical cyclones rarely make it farther north and west of the intersection of 40°N and 85°W.

Future work should examine the computation and addition of storm movement to the extended best-track data. Then each of the four radii could be more precisely classified as left or right halves. Thereafter, statistics such as the mean and variance could be computed for the left and right sides. This would be compatible with the left and right buffers in GIS since the software indicates the storm direction of movement. The variance would provide an idea of the uncertainty or quality of the wind swath model.

Additional future work may focus on the creation of a shapefile for each record in the extended best-track dataset. Each shapefile would contain four features, the four quadrant radii, and would be broken into the 34-, 50-, or 64-kt intervals. Once this is known, the GIS software analyst can draw a contour for speed and convert to a polygon, after which all polygons can be merged together to derive a single storm wind swath. This technique should result in wind swaths that would match the original extended best-track dataset.

There are, however, some limiting factors to efficiently applying both concepts as there are a number of underlying complexities that must be worked out prior to their implementation. Each storm has its own unique set of behaviors, wind swaths, and caveats that demand manual intervention at each step of the process. In addition, any verification of the actual wind data as compared to the wind swath will most likely be done using surface wind reports, which are not always sited in the best possible locations for capturing the true force of the wind (at 10-m height). Ideally each storm and corresponding wind swath would be verified through some sort of in situ wind observing network, either satellite derived or surface based.

Finally, while the results are shown as return intervals, the focus of this study was to produce a comprehensive climatology of the extent of inland winds from tropical cyclones in the eastern United States. While some locations east of the Rocky Mountains have never experienced such winds, the analysis presented here is purely climatological in nature and no predictive trends or assumptions are provided or inferred. However, for those locations with higher return intervals, knowledge of where tropical storm–force winds or hurricane-force winds have most commonly occurred can better prepare local forecasters, emergency managers, county planners, and others to be even more vigilant against the myriad of threats tropical cyclones pose and recognize that their impacts often extend well inland from the coast.