A major hurricane [96+ knots (kt; 1 kt = 0.51 m s–1) of maximum sustained wind] has not made landfall in the United States since Wilma (2005). Recent elegant stochastic–statistical modeling estimates the return period of a 9-yr streak for this metric as 177 yr, suggesting extraordinary rarity, especially in the context of the length of the record (1851–2014). Current awareness of the drought is increased given that the 2015 hurricane season is expected to be suppressed from El Niño and the recent anniversaries of several noteworthy landfalls.
Yet, here we show that the significance or even existence of the current 9-yr drought is highly dependent on the metric used. Acknowledging that wind intensity estimates are binned every 5 kt and have approximate 10-kt uncertainty, we examine the same record using landfall thresholds of 95–105 kt. Using 105-kt landfall, 1993–2003 becomes a previously unreported yet more remarkable 11-yr drought and 1981–88 becomes an 8-yr drought. Further, landfall minimum sea level pressure is more reliably estimated than maximum sustained wind speed. For landfall intensities stronger than 960 hPa (a climatological threshold for 100 kt), the current drought disappears because of Irene (2011) and Sandy (2012). A coastline-independent yet nearby proximity metric is tested and reveals a nonexistent drought.
Accordingly, this study suggests the following: 1) Caution is advised when identifying a hurricane drought and its historical significance. 2) Using hurricane landfall statistics to infer a climate signal is fraught with issues (threshold, coastline, and potentially nonscientific contributions), regardless of intensity metric. 3) From a societal context, human and financial losses matter most, and Irene [2011; $8 billion (U.S. dollars)] and Sandy (2012; $88 billion) occurred during the current drought.
Although nine years have passed since the last U.S. major hurricane landfall, the existence and relative significance of the current drought are largely artifacts of the chosen metric.
At the time of this article’s preparation, Saffir–Simpson hurricane wind scale (SSHWS; Simpson 1974) category 4 Hurricane Joaquin was devastating the Bahamas (see title page). It was also threatening landfall in the United States with a projected path by most models—but not all models—disturbingly similar to Sandy (2012), including the rare turn northwest then west into the coast. This sudden threat was a stark contrast to the lack of U.S. major1 hurricane landfalls [classified as 96 knots (kt; 1 kt = 0.51 m s−1) or greater maximum sustained wind VMAX] in recent memory. Indeed, there has been increasing awareness of this major hurricane landfall drought in the United States, as the last major hurricane to landfall in the United States was Wilma on 24 October 2005.
The current gap of 9 yr (potentially extending to 10 with the close of the 2015 season) is the longest since at least 1900, with the next longest gap of only 5 yr (twice: 1901–05 and 1910–14). Recently Hall and Hereid (2015, hereafter HH15) elegantly estimated the wait time between 9-yr droughts as 177 yr using a stochastic–statistical model. While this estimate is striking for its magnitude [in particular because it is similar to the length of Atlantic tropical cyclone (TC) records going back to 1851], those authors also conclude that the current drought is likely “a matter of luck.” Our work complements their results and makes evident that the existence or historical significance of the current period of inactivity varies substantially depending on the choice of metric used. Thus, the definition of drought is largely arbitrary, although the potential impacts of that drought (or lack thereof) are obviously not arbitrary or trivial.
DEFINITION OF LANDFALL.
The May 2015 version of the revised Atlantic hurricane database (HURDAT2; Landsea and Franklin 2013)—an updated version of Jarvinen et al. (1984)—is used here, starting in 1900. That year has been identified as the start of the reliable record of Atlantic hurricane landfalls (Landsea 2007), prior to which potential significant undetected hurricane activity is likely (Landsea 2007; Truchelut et al. 2013). For the period 1900–55, for Camille (1969), and for 1983 onward, HURDAT2 has an explicit landfall record identified (location, time, and intensity) that was used for this analysis. However, for the period 1956–82, minus Camille (1969), no such HURDAT2 identifier exists, although the ongoing HURDAT reanalysis project aims to fill that gap (Landsea et al. 2004b). Accordingly, for the period 1956–82, excluding Camille (1969), we use the intensity at the last 6-h synoptic time before landfall. We note that while this choice may introduce a slight bias (toward stronger intensity) in the landfalls during that period (Rappaport et al. 2010), alternative choices (temporal interpolation to the landfall point or using the first synoptic time after landfall) would introduce even larger, negative intensity biases (C. Landsea 2015, personal communication). Finally, if a TC made multiple landfalls in the United States [e.g., Andrew (1992) and Katrina (2005)], only the most intense landfall intensity–location combination was used. The locations of the TC landfalls used in this paper are given in Fig. 1, the details of which will be discussed later.
We emphasize that the HURDAT2 is a fluid database undergoing reanalysis and revision. For example, Hurricane Andrew (1992) was upgraded a decade later to SSHWS category 5 landfall (Landsea et al. 2004a), joining Camille (1969) and Labor Day (1935). Moreover, the 1960s have yet to be reanalyzed. Thus, it is possible that further reanalysis may upgrade or downgrade landfall intensities based upon new information and scientific understanding, leading to changes in the various drought measurements presented next. Of course, drought length changes induced solely by HURDAT2 changes would only emphasize further the uncertainties in the historical intensity database and thus the arbitrariness of the definition of landfall drought.
DROUGHT SENSITIVITY TO LANDFALL VMAX THRESHOLDS.
As mentioned, the definition of a major hurricane in the Atlantic basin (SSHWS category 3 or higher) is VMAX ≥ 96 kt. Given that operational hurricane intensities are binned every 5 kt, this threshold is more commonly given as VMAX ≥ 100 kt. Although the intensities are binned every 5 kt, this bin size does not imply an equivalent level of precision. Indeed, Torn and Snyder (2012) and Landsea and Franklin (2013) found that the typical VMAX uncertainty is of the order 10 kt, which on average implies that distinctions in VMAX for a given storm between, for example, 100 and 105 kt may be small compared to the observational uncertainty.2
Given this uncertainty, it is reasonable to examine how the current drought and its historical significance are affected by modifying the landfall intensity threshold by an amount within this uncertainty window. Accordingly, Fig. 2a illustrates the history of landfalls since 1900 as a function of landfall intensity from 95 to 105 kt. For the standard threshold of 100 kt, the widely cited gap of 9 yr is immediately evident, particularly in sharp contrast to the prior 115 yr during which no drought of comparable length exists, consistent with HH15.
However, for a mere 5-kt increase to 105 kt for landfall intensity threshold, though the current drought length remains at 9 yr, a longer 11-yr drought emerges from 1993 to 2003. Most striking about that record drought from 1993 to 2003 is that it may have gone completely undetected until now. This lack of detection is perhaps because three 100-kt landfalls of note occurred in the absence of a single 105-kt landfall, despite the fact that the 100-kt threshold is arbitrary for drought definition and a 5-kt difference is within the observational uncertainty.
Meanwhile, when lowering the landfall threshold from 100 to 95 kt, Hurricane Ike (2008) qualifies, and although the current drought is still the longest on record, its length is reduced from 9 to 6 yr. That a change of only 5 kt either side of 100 kt can turn an unprecedented drought (by a large margin) into a drought that has been exceeded previously by a substantial margin (2 yr) in the modern era suggests that hurricane landfall drought definitions are somewhat arbitrary indeed.
DROUGHT SENSITIVITY TO LANDFALL PMIN THRESHOLDS.
While not the official definition of TC intensity, minimum sea level pressure PMIN is a more stable measurement of TC intensity given 1) PMIN is an integrated measure of the storm’s horizontal wind field (e.g., Knaff and Zehr 2007) as well as the vertical structure and thus is expected to be less prone to sampling error and noise, 2) PMIN is better sampled both by reconnaissance flights (through dropsondes within the eye) as well as at landfall by the plethora of barometers (although some extrapolation is often necessary to get the best estimate), and 3) uncertainty estimates for PMIN are comparable to (Landsea and Franklin 2013) or even 40% less than (Torn and Snyder 2012) that for VMAX. It is also worth noting that the landfall PMIN is often recorded to a precision of 1 hPa, in comparison to 5 kt for VMAX. Accordingly, it makes sense to perform a similar landfall analysis using PMIN, and Fig. 1b denotes the landfall locations when using PMIN.3
Figure 2b provides this analysis of TC landfall based upon PMIN, noting that the bins of PMIN have been chosen to match the climatological relationship between VMAX and PMIN [consistent with Dvorak (1984) and verified later in Fig. 6; see also www.ssd.noaa.gov/PS/TROP/CI-chart.html]. For example, VMAX = 100 (105) kt corresponds to PMIN of approximately 960 (955) hPa in the entire Atlantic HURDAT2 TC existence since 1950. When landfall PMIN ≤ 960 hPa is examined, it becomes clear that during the current 9-yr drought there have been four qualifying landfalls [Gustav (2008), Ike (2008), Irene (2011), and post-tropical Sandy (2012)], thus resulting in a trivial 2-yr PMIN drought. Further, changing PMIN from 960 to 955 hPa (effectively in the mean from VMAX = 100 to 105 kt) does not change meaningfully the drought lengths discussed above; the same is true for a 965-hPa threshold. Finally, lowering the threshold further to 950 hPa (the black bars in Fig. 2b) also does not change the drought length.4
Although VMAX and PMIN differ in what precisely they represent,5 both are valid metrics for the intensity of a TC, particularly given that ultimately the specific interest in landfalling TCs is tied directly to societal impacts. Indeed, PMIN is as good as if not better than VMAX as a predictor of historical economic damage (Mendelsohn et al. 2012). One particularly poignant example of this relationship is Sandy (2012), whose relatively modest VMAX yet record low PMIN translated to one of the most economically devastating storms in U.S. history, in large part because of the storm surge.
A CLIMATE PERSPECTIVE: GENERALIZATION TO BASIN ACTIVITY AND LAND PROXIMITY.
One critical aspect of the ongoing discussion is whether landfall droughts are tied to climate variability, either at the decadal+ time scale (e.g., Schlesinger and Ramankutty 1994; Goldenberg et al. 2001; Klotzbach and Gray 2008) or because of anthropogenic influences (Emanuel 2013). Additionally, the clustering of regional landfalls in some years, while suppressed for other sets of years (e.g. Jagger and Elsner 2012), further motivates the question of factors driving landfall and the time scale of such climate factors. However, these questions implicitly assert that the statistics of hurricane landfalls can provide meaningful information about climate-induced variability in the spatial pattern of TC activity. Given that most TCs form over the open Atlantic Ocean and track westward thereafter, the statistics of TC landfall are arguably a convolution of the climatic control of hurricane activity (genesis, track, and intensity) and the orientation and details of the U.S. coastline, topics examined in greater detail by Truchelut (2015).
Accordingly, it is relevant to next ask whether the current drought of major hurricane landfall activity in the United States is simply a result of reduced overall Atlantic basin major hurricane activity. In other words, perhaps the drought is not limited to U.S. landfall? Figure 3 addresses this possibility by analyzing the counts of Atlantic basinwide hurricanes of at least 95-, 100-, or 105-kt intensity, in comparison to the fraction of those hurricanes that made landfall in the United States at those same intensity thresholds. It is clear that during the drought since 2005 (all the dots plotted at a fraction of zero since 2006), there has been no shortage of major or near-major hurricanes in the Atlantic. Thus, the U.S. landfall drought is occurring during a time when overall major hurricane activity is not suppressed. We note that there may be additional suitable measures of landfall drought that may account for variability in overall activity such as the number of consecutive major Atlantic hurricanes that have occurred without impacting the United States at that intensity (P. Klotzbach 2015, personal communication) but are not examined here for brevity.
In light of Fig. 3, we seek to assess the dependence of landfall statistics on the nature of the coastline by exploring these statistics with respect to a simple idealized geographic benchmark in lieu of the complex U.S. coastline. Figure 4 displays historical U.S. landfall probability as a function of TC location. We next count TCs having VMAX ≥ 100 kt while simultaneously existing poleward and westward of 20°N, 70°W, independent of landfall; this geographic benchmark corresponds to the 50% U.S. TC landfall probability contour in Fig. 4 and is also (perhaps not coincidentally) the approximate latitude of entry into the Gulf of Mexico and longitude of Cape Cod, Massachusetts. As such, this endeavor may be interpreted as evaluating the frequency of major hurricanes “on approach” for U.S. landfall, without concern for whether landfall on an irregular coastline actually occurs. Indeed, HH15 noted that Cuba has taken the brunt of multiple recent major hurricane landfalls. This may imply that hurricanes making landfall in the United States might have done so at higher intensity had this interaction with Cuba not occurred, further providing relevance to the geographic benchmark chosen here.
The results are shown in Fig. 5, which does not indicate a meaningful drought for VMAX ≥ 100 kt in the far western Atlantic at any time over the past two decades. Further, this suggests that the current drought may be in part dependent on the orientation of the coastline—or interaction with nearby landmasses (HH15)—highlighting the additional complexity inherent in using TC landfall to attempt to infer climate signals. When Figs. 3, 4, and 5 are compared, they also highlight the need to understand the factors that drive production of Atlantic basin major hurricane activity and major hurricane activity in proximity to the United States, while also suppressing landfall in the United States at that threshold.
POTENTIAL FOR A PATHOLOGICAL CONTRIBUTION.
Given the impact that 5-kt shifts in VMAX near 100 kt have and 5-hPa shifts in PMIN near 960 hPa do not have on the length of the drought, it is prudent to take a step back and examine the distribution of TC intensity generally. From a theoretical perspective, there is no reason to expect that the frequency of landfall should increase as the landfall intensity increases (once hurricane status is reached, anyway). In other words, more intense hurricanes at landfall are less common than weaker hurricanes, given a long enough record.
Figure 6a demonstrates that while this is indeed true, there are some very interesting departures from that expectation. The red curve in Fig. 6a is the distribution (as a fraction of the peak occurrence) for all landfalls from 1900 to 2014 as defined previously. The intensity at peak landfall frequency is the tropical storm stage, with on average frequency decreasing as one ascends the SSHWS. However, the occurrence of 100 kt at landfall (0.6 of peak) is twice that of 95 kt at landfall (0.3 of the same peak), oddly suggesting that 100 kt at landfall is twice as common as 95 kt at landfall. The VMAX = 100 kt landfall is also 1.5 times more likely than 105 kt, also suggesting a curious propensity for 100-kt TCs at landfall. Note that when the green and black lines (more confident yet smaller subsets) are compared to the prior analysis, the same result occurs, suggesting that when those TCs without an explicit HURDAT2 landfall record are ignored, the relationship does not weaken but instead becomes stronger.
Meanwhile, when this distribution is expanded to include all HURDAT2 times back to 1950 regardless of landfall (gray bars in Fig. 6a), the local maximum at 100 kt decreases markedly, suggesting that this effect at the 100-kt threshold is likely specific to the time of landfall. Of course we note the relatively small sample sizes here: 18 (8) TC landfalls at 100 (95)-kt intensity for the 1900–2014 period. Perhaps with a longer dataset, these differences would decrease and the curves in Fig. 6a would more closely follow the expected Poisson distribution for TC statistics (Elsner and Kara 1999; Elsner 2003; Parisi and Lund 2008). Nonetheless, this result points to the potential existence of bias in landfall intensity at the major hurricane threshold that may arise intentionally for societal awareness and preparedness reasons. However, in principle this effect should be reduced as a result of the postseason best-tracking process when such real-time factors are largely eliminated.
Additionally, in Fig. 6a (gray bars) there exists a striking local maximum at VMAX = 65 kt. This anomalous maximum occurrence at 65 kt was previously noted by Emanuel (2000) and was then speculated to be partially a result of forecaster desire to increase societal awareness implied by the status change to hurricane intensity. If this is in fact the case, it lends support to the explanation for the 100-kt landfall intensity anomaly discussed earlier and potentially further undermines the objectivity and scientific significance of the 100-kt landfall drought (Fig. 6a).
Yet, the existence of local maxima in the VMAX distribution may also owe to the climatological VMAX associated with each current intensity (CI) number in the Dvorak technique (Dvorak 1984). Only a fraction of the 5-kt intensity bins shown in Fig. 6a is represented in the CI–VMAX association, including (when rounded) 55, 65, 75, 90, 100, 115, 125, and 140 kt, thus making those intensities more likely at advisory time in the absence of in situ data. Finally, for completeness, Fig. 6b shows the same analysis in Fig. 6a, except using PMIN instead of VMAX.
SUMMARY AND IMPLICATIONS.
We summarize in Fig. 7 most of the various metrics previously discussed. For this figure, the distribution of drought length for each metric is given by the colorized bars. The arrows on the abscissa are where the current drought (ending in 2014) resides based upon (and also colorized by) metric. The pairs of years labeled above the corresponding bar of the same color are the years of the maximum drought length by metric. Thus, if an arrow of a given color lies left of the labeled year pair of the same color, the current drought is not the longest drought since 1950 by that metric. Across all of the different methodologies shown in Fig. 7, the current drought is unprecedented only in the case of the landfall VMAX ≥ 100 kt threshold. Although not shown in Fig. 7, for landfall VMAX ≥ 95 kt the drought is still the longest, although its length is reduced to 6 yr. We reiterate that variations in VMAX and PMIN thresholds applied here lie within the range of observational uncertainty in the measurements themselves. Moreover, extending Fig. 7 to use the record back to 1900 does not qualitatively change the figure. The only noteworthy changes are that the periods of maximum drought length by metric, for the nonwind metrics, all fall during the 1900s to 1920s during a less reliable observational period, hence the reason for showing the 1950–2014 period in Fig. 7 instead of 1900–2014.
This work complements the findings of HH15 regarding the statistics of hurricane droughts and identifies a substantial dependence of the widely cited current drought on the common yet largely arbitrary definition of major hurricanes. Because of the sensitivity of landfall drought statistics to both the metric and threshold, we conclude that caution is warranted when identifying a hurricane drought and its historical significance. Moreover, the use of hurricane landfall statistics to infer climate signals is fraught with additional complexities (coastline geometry, potentially nonscientific contributions) that warrant additional research.
We again stress that all of the metrics discussed previously ignore the most important measure of impact—that of human and financial loss (Pielke et al. 2008; Pielke 2012; Chavas et al. 2013). Hurricane Sandy (2012) was barely a hurricane by landfall VMAX. Hurricane Irene (2011) produced historic flooding in parts of New England and yet was a strong tropical storm by landfall VMAX there because of the greater than normal reduction of wind to the surface (Avila and Stewart 2013). Nonetheless, Sandy produced $88 billion (U.S. dollars) in losses and the largest loss of life for an East Coast hurricane in over four decades (Blake et al. 2013). Irene produced $8 billion in loss (Avila and Stewart 2013) despite being a storm that would not count toward the major hurricane landfall analysis in terms of landfall VMAX. Both of these significant if not extreme losses occurred during the very middle of the current landfall drought as measured by landfall VMAX. Thus, we must not lose sight of the fact that our formal definitions of hurricane intensity do not readily scale to loss of life or damage, and in fact there can be considerable outliers.6
This manuscript was motivated by discussion among the authors at and following a landfall risk workshop organized and sponsored by the Risk Prediction Initiative of the Bermuda Institute of Ocean Sciences in London in June 2015. While the views presented herein are those of the authors, we acknowledge further motivating discussion during that workshop provided by Kerry Emanuel, Robert Muir-Wood, Amato Evan, and Tim Doggett. This manuscript has benefited from discussions with Andrew Hazelton, Ryan Truchelut, Levi Cowan, and Phil Klotzbach. We appreciate the recommendations by Christopher Landsea on the determination of landfall intensity using the HURDAT2 database and the constructive feedback of three anonymous reviewers. This work was partially supported by the Risk Prediction Initiative of the Bermuda Institute of Ocean Sciences. Title page imagery: NOAA GOES-East satellite via the NASA GSFC GOES project (source: www.nasa.gov/feature/goddard/joaquin-atlantic-ocean).
CURRENT AFFILIATION: Department of Earth, Atmospheric, and Planetary Sciences, Purdue University, West Lafayette, Indiana
The origins of the word “major” in describing category 3+ hurricanes appear to be later than the establishment of the Saffir–Simpson scale itself (Simpson 1971, 1974), as those publications make no mention of this distinction. However, by 1978 (Hebert and Taylor 1978) the term was in use, although we cannot be certain that the distinction did not appear in an interim work. A much later but comprehensive climatology of major hurricanes in the Atlantic (Landsea 1993) did not discuss the origins of the distinction.
This is not to say that a 5-kt difference in actual intensity does not have real physical consequences. Indeed the kinetic energy (power dissipation; Emanuel 2005) difference between 100 and 105 kt is approximately 10% (15%). The argument here is that with a 10-kt mean uncertainty in landfall VMAX, basing droughts solely upon the length of a single intensity threshold is potentially an unstable measure of landfall drought.
We note that four TCs since 1900 did not have a corresponding PMIN at the time of reported landfall VMAX (gray shading in Fig. 1b). Three of these four did have a PMIN within 12 h of landfall (before or after) that would qualify for Fig. 2. However, all three occurred during years (1906, 1909, and 1959) when other qualifying TCs were occurring, thus not changing the drought climatology. The fourth TC (number 2 in 1941) had a PMIN = 989 hPa one day after landfall and thus cannot be assumed to be representative of landfall PMIN.
This additional threshold (PMIN ≤ 950 hPa) is also provided as the mean landfall PMIN is approximately 5 hPa lower than that for the full HURDAT2 database since 1950.
We have not performed any pressure scaling for environment or latitude but recognize that a Northeast U.S. landfalling TC may exist at a similar wind intensity but with deeper pressure (see prior footnote) compared to the surrounding environment than a TC farther south. Accordingly, pressure gradient matters more than PMIN for determining VMAX. Nonetheless, using PMIN as a simple comparative metric to VMAX can be instructive if not more explanatory of resulting damage (Mendelsohn et al. 2012), and thus PMIN should not be normalized by latitude.
As a postscript, we note that after the final preparations for this manuscript, Hurricane Joaquin (2015) 1) easily satisfied the metric shown in Fig. 5 (reaching 931 hPa while existing northwest of 20°N, 70°W), 2) is one of the most intense hurricanes to ever strike the Bahamas, but 3) turned out to sea only 600 km from U.S. landfall. The 100-kt U.S. landfall drought reached 10 yr on 24 October 2015.