12
13
14
15
16
10
11
1
2
7
8
5
6
3
4
Seasonal Influences upon and Long-Term Trends in the Length of the Atlantic Hurricane
Season
Juliana M. Karloski and Clark Evans
+
9 Atmospheric Science Program, Department of Mathematical Sciences
University of Wisconsin-Milwaukee, Milwaukee, Wisconsin
Submitted to the Journal of Climate for consideration as an Article
1 May 2015
Revised 4 August 2015
+Corresponding Author: Dr. Clark Evans, Atmospheric Science Program, Dept. of Mathematical
Sciences, University of Wisconsin-Milwaukee, P. O. Box 413, Milwaukee, WI 53201. E-mail: evans36@uwm.edu.

17 Abstract
18 Considering a subset of the North Atlantic Ocean south of 30°N and east of 75°W,
19 Kossin (2008) found that the Atlantic tropical cyclone (TC) season increased in length by about
20 two days per year between 1980 and 2007, albeit only to 80-90% confidence. It is uncertain,
21 however, whether the same is true over the entire Atlantic basin or into the present. Separately, it
22 is unclear as to whether meaningful sub-seasonal variability in the environmental factors
23 necessary for TC formation exists between early- and late-starting and -ending seasons.
24 Quantile regression is used to evaluate long-term trends in Atlantic TC season length. No
25 statistically-significant trend in season length exists for the period 1979-2007 when considering
26 the entire Atlantic basin, or for the period 1979-2014 independent of the portion of the basin
27 considered. Linear regression applied to June and November monthly-mean reanalysis data is
28 used to examine sub-seasonal environmental variability between early- and late-starting and -
29 ending seasons. Within an otherwise favorable environment for genesis, early-starting seasons
30 are associated with increased lower tropospheric relative vorticity where most early-season TCs
31 form. Late-ending seasons are associated with La Niña, negative-phase Pacific Decadal
32 Oscillation events, and environmental conditions that promote an increased likelihood of TC
33 development along the preferred genesis pathways for late-season TCs.
34 While confidence in these results is relatively high, they only explain a small portion of
35 the total variation in Atlantic TC season length. More research is needed to understand how
36 variability on all timescales influences Atlantic TC season length and its predictability.
37
1

38 1. Introduction
39 Every year, multiple meteorological organizations, such as NOAA; research groups, such
40 as that at Colorado State University (http://hurricane.atmos.colostate.edu/Forecasts/); and private
41 firms release predictions of the number of tropical storms, hurricanes, and major hurricanes they
42 expect to form that year, particularly within the North Atlantic basin. Other attributes of tropical
43 cyclone (TC) activity, however, such as when the first TC will form or how long the season will
44 last, are typically not forecast. Officially, the Atlantic TC season begins on 1 June and ends on
45 30 November. Over 97% of TC activity within the Atlantic basin occurs between these two dates
46 (HRD 2015). However, as assessed utilizing the National Hurricane Center HURDAT2 database
47 (Landsea and Franklin 2013), the average first TC formation for the period 1979-2014 occurred
48 on 30 June and the average last TC formation occurred on 4 November. The question thus
49 remains: what causes some seasons to be significantly shorter or longer than average?
50 It is worthwhile to consider whether there could be a relationship between the period over
51 which the conditions necessary for TC development are present at one or more locations and the
52 length of the TC season. Gray (1968, 1979) identified six necessary but not sufficient criteria for
53
TC development: sea surface temperatures (SSTs) greater than or equal to 26.5°C, abundant
54 lower- to mid-tropospheric relative humidity, conditional instability through a deep tropospheric
55 layer, large lower tropospheric cyclonic relative vorticity, low vertical wind shear of the
56 horizontal winds (roughly less than 10 m s
-1
between the surface and tropopause), and
57 displacement by at least 5° latitude poleward from the equator. These six conditions may be
58 condensed (e.g., as done by Frank 1987 and others) into four: sufficiently large cyclonic vertical
59 vorticity, weak vertical wind shear of the horizontal winds, SSTs greater than or equal to 26°C,
60 and abundant lower- to middle-tropospheric relative humidity. The SST criterion may
2

61 alternatively be expressed in terms of upper oceanic heat content and/or potential intensity, the
62 latter of which is influenced by both SST and outflow-layer temperatures (e.g., Emanuel 1986).
63 Likewise, the relative humidity and vertical wind shear criteria may alternatively be represented
64 by ventilation, which reflects the extent to which environmental, middle-tropospheric low
65 entropy air may be imported into the circulation of a TC or predecessor disturbance (e.g., Tang
66 and Emanuel 2012, Tang and Camargo 2014).
67 Subsequent investigators (e.g., Bruyere et al. 2012, Camargo et al. 2007, Emanuel 2010,
68 Emanuel and Nolan 2004, McGauley and Nolan 2011, Schumacher et al. 2009, Tang and
69 Camargo 2014, and Tippett et al. 2010) have utilized these criteria to develop, apply, and
70 evaluate genesis potential indices. Such indices are designed to highlight whether conditions
71 favorable for TC development exist in observations or climate model outputs at one or more
72 locations and times. For example, applying the ventilation index of Tang and Emanuel (2012) to
73 eight model outputs from the fifth Coupled Model Intercomparison Project (CMIP5), RCP8.5
74 experiment, Tang and Camargo (2014) identified a potential decrease in the average annual
75 number of days that the ventilation index is favorable for TC genesis in the North Atlantic Ocean
76 of up to two weeks by the end of the 21 st
century. Alternatively, downscaling methods may be
77 applied to climate model outputs to assess projected future changes in TC season length (e.g.,
78 Dwyer et al. 2015).
79 In the context of genesis potential indices, greater lower to middle-tropospheric moisture
80 content, higher upper oceanic heat content and/or potential intensity, lower vertical wind shear,
81 and sufficiently large cyclonic vertical vorticity indicate a greater potential for tropical
82 cyclogenesis (e.g., McGauley and Nolan 2011). Presumably, TC seasons that start earlier or end
83 later than normal are characterized by the presence of the necessary ingredients for TC formation
3

84 at times when they are not typically present. The inverse is presumably true for seasons that
85 begin later and/or end earlier than normal. Herein, we seek to examine the extent to which this is
86 true on sub-seasonal time scales, particularly in the context of the genesis pathways that early-
87 and late-season North Atlantic TCs typically follow (McTaggart-Cowan et al. 2008, 2013). Note
88 that the genesis pathways identified by McTaggart-Cowan et al. (2008, 2013) describe the
89 environments within which the physical processes believed to be responsible for tropical
90 cyclogenesis (e.g., wind-induced surface heat exchange; Emanuel 1986) are triggered so as to
91 lead to the formation of a tropical cyclone.
92 In addition to understanding the synoptic- to climate-scale factors that control the length
93 of an individual TC season, it is worthwhile to consider whether the length of the Atlantic TC
94 season has increased or decreased in recent decades. Focusing on an extended North Atlantic
95
96 main development region (MDR) south of 30°N and east of 75°W where an increase of 0.5-
1.5°C in mean June-November SST was observed between 1950 and 2007, Kossin (2008)
97 utilized quantile regression to identify changes in the annual distribution of Atlantic TC
98 formation dates. Their findings (Figure 4 of Kossin 2008) suggest that the Atlantic TC season
99 has increased in length by ten or more days per decade since 1950, reflective of both earlier starts
100 and later ends to the season, with a level of confidence between 80-90%. This was attributed,
101 albeit also with low confidence, to warmer May SSTs across the extended MDR, with an
102 increase of the area-averaged SST by 1°C corresponding to as much as a twenty day shift in the
103 start and/or end of the season. Though quantile regression is relatively insensitive to individual
104 outliers, to what extent this result is due to several TC seasons during the early-to-mid-2000s in
105 which particularly late-forming TCs occurred within Kossin (2008)’s extended MDR (e.g., 2001,
106 2003, 2005, and 2007; Table 1) remains unclear. In addition, whether this result holds in whole
4

107 or in part for the entire Atlantic basin, in which several TC seasons during the early-to-mid
108 2000s were also associated with particularly late end dates (Table 2), is uncertain. Evaluating
109 long-term trends in Atlantic TC season length when considering the entire basin is particularly
110 important given that many of the earliest- and latest-forming TCs undergo genesis outside of
111
Kossin (2008)’s extended MDR (Figs. 1 and 2).
112 Our research thus has several aims. We first examine whether the positive but non-
113 statistically-significant trend in Atlantic TC season length identified by Kossin (2008) is
114 maintained when the period of record is extended to the present. Next, we examine whether the
115 positive but non-statistically-significant trend in Atlantic TC season length identified by Kossin
116 (2008) is maintained when the set of TCs considered is extended to the entire Atlantic basin,
117 whether through 2007 or the present. Separately, we investigate how the synoptic- to planetary-
118 scale distributions of proxies for the necessary conditions for tropical cyclogenesis vary on sub-
119 seasonal to interannual time scales between early- and late-starting and -ending Atlantic TC
120 seasons, particularly in light of the different genesis pathways followed by early- and late-season
121 TCs. The remainder of this work is structured as follows. Section two outlines the data utilized
122 and methodology followed in this study. Section three presents results regarding long-term
123 trends in Atlantic TC season length for the period 1979-2014. Section four presents results
124 regarding the synoptic- to planetary-scale conditions that characterize early and late starting and
125 ending Atlantic TC seasons. A summary of the key findings of the work and a discussion of their
126 implications is presented in section five.
127 2. Data and Methodology
5

128 In this study, the National Hurricane Center HURDAT2 database (Landsea and Franklin
129 2013) is used to identify Atlantic TC formations, excluding tropical depressions and subtropical
130 cyclones, between 1979 and 2014. A total of 431 TC formation events are identified across the
131 entire basin, with 228 of these being located within the extended MDR considered by Kossin
132 (2008). The year 1979 is chosen as the start of our analysis period for several reasons. First, this
133 is the first year for which ERA-Interim reanalysis data (Dee et al. 2011), the use of which is
134 described below, are available. Second, substantial improvements in the quantity and quality of
135 satellite clear-sky radiance and cloud-track wind observations were realized beginning in 1979,
136 resulting in more robust atmospheric reanalysis products from 1979-onward (e.g., Uppala et al.
137 2005). Third, so as to facilitate direct comparison between our results and those of Kossin
138 (2008), we desire that the start of our period of record closely match that of Kossin (2008). We
139 note that the sensitivity to a starting year of 1979 versus 1980 in the results to be presented is
140 negligible (not shown). Finally, as routine geostationary satellite surveillance of TCs began in
141 the mid-1960s (e.g., Neumann et al. 1999), it is unlikely that TCs were systematically missed
142 over part or all of the period of record considered herein (e.g., Landsea 2007). Landsea (2007)
143 also notes that advances in observational and analysis capabilities appear to be contributing to
144 one additional Atlantic TC event per year since 2002, with many of their cited examples
145 occurring at the start or end of the TC season. While it is possible that this could influence both
146 the results presented herein as well as those of Kossin (2008), no attempt is made to account for
147 this possible change in classification practices.
148 To quantify long-term trends in Atlantic TC season length, we apply the method of
149 quantile regression. In contrast to linear (or least squares) regression, which provides a method
150 by which the nature of an assumed linear relationship between the means of grouped data can be
6

151 evaluated, quantile (or percentile) regression (Koenker and Bassett 1978, Koenker and Hallock
152 2001) provides a method for evaluating the nature of an assumed linear relationship for
153 conditional quantile functions. One common application of quantile regression is median
154 regression, though quantile regression need not be limited exclusively to the median (or 50th
155 percentile) of a data set. Quantile regression has been applied in the atmospheric sciences to
156 examine long-term trends in Atlantic TC season length (Kossin 2008) and the intensity of the
157 most-intense Atlantic TCs (Elsner et al. 2008). Quantile regression considers the topology of the
158 full dataset of TC events, rather than just the first and last TCs for each season, thereby
159 increasing the degrees of freedom of the statistical analysis; furthermore, unlike linear
160 regression, it is relatively insensitive to individual outliers within the data (Kossin 2008).
161 Herein, quantile regression is first utilized to replicate the findings of Kossin (2008),
162 considering only TCs that formed within their extended MDR between 1980 and 2007. Quantiles
163 between the 5 th
and 95 th
percentile, at 5% intervals, are considered. This analysis is then
164 replicated for the period beginning 1979; as noted earlier, the results are found to be insensitive
165 to small differences in the start year. Next, to evaluate whether the positive but non-statistically-
166 significant trend in Atlantic TC season length identified by Kossin (2008) is maintained when the
167 period of record is extended to the present, the analysis is repeated with an ending year of 2014.
168 Subsequently, to evaluate whether the positive but non-statistically-significant trend in Atlantic
169 TC season length identified by Kossin (2008) is maintained when the set of TCs is extended to
170 the full Atlantic basin, the analysis is repeated for the periods of 1979-2007 and 1979-2014 when
171 considering all Atlantic basin TCs. Finally, the sensitivity in how changing the end of the period
172 of record by one year at a time is examined by repeating the analysis, over both Kossin (2008)'s
173 extended MDR and the full basin, for end years between 2001 and 2014. The 90 th
percentile
7

174 confidence interval is computed for each analysis utilizing bootstrapping (e.g., Efron and
175 Tibshirani 1993) with 200 surrogates.
176 To assess how the synoptic- to planetary-scale distributions of the necessary conditions
177 for TC formation vary on sub-seasonal to interannual time scales between early- and late-starting
178 and -ending Atlantic TC seasons, linear regression is utilized. The 10%, 25%, 75%, and 90% TC
179 formation date for each Atlantic TC season is first identified. These values are then linearly
180 regressed against monthly mean fields of SST, 850 hPa relative vorticity, 600 hPa relative
181 humidity, 500 hPa geopotential height, and 850-200 hPa vertical wind shear magnitude.
182 Monthly-mean SST data are obtained from the NOAA Extended Reconstructed Sea Surface
183 Temperature (ERSST) dataset, version 3b, on a 2° latitude x 2° longitude grid (Smith et al.
184 2008). It should be noted that this dataset differs slightly from that described by Smith et al.
185 (2008) in that satellite-derived SST data are not used in its composition so as to remove a cold
186 SST bias introduced by non-clear sky satellite SST retrievals. All monthly-mean atmospheric
187 fields are obtained from the ERA-Interim (Dee et al. 2011) reanalysis on an approximate 0.7°
188 latitude x 0.7° longitude grid. The results are insensitive to one isobaric level (50 hPa) shifts in
189 the isobaric level(s) over which atmospheric fields are considered (not shown). Sensitivity in the
190 results to the choice of SST or atmospheric reanalysis dataset is expected to be small but is not
191 explicitly examined.
192 Herein, linear regression takes the form:

J
 x
 cov( J , x ) var( x )
(1)
8

193 This notation follows that of Torn and Hakim (2008), where the left-hand side of (1) represents
194 the slope of the linear regression line between J , here defined as the n th percentile date, and x ,
195 here defined as the relevant oceanic or atmospheric field. Note that in (1), the long-term (1979 to
196 2014) means for each field are removed from both J and x . Thus, (1) enables the identification in
197 changes in the n th percentile date as a function of changes in a given forcing. In (1), cov refers to
198 the covariance between J and x , while var refers to the variance of x . In the analyses to follow,
199 the right-hand side of (1) is multiplied by the standard deviation of x , such that a one standard
200 deviation change in x is responsible for a N day change in the n th percentile date. A student’s t
201 test is utilized to test whether the slope of the linear regression between J and x is non-zero to
202 90%, 95%, and 99% confidence.
203 Four sets of linear regression analyses between monthly-mean values of each field and
204 n th percentile formation date are conducted: (1) June monthly-mean to 10 th percentile formation
205 date, (2) June and July monthly-means to 25 th
percentile formation date, (3) October and
206 November monthly-means to 75 th
percentile formation date, and (4) November monthly-mean to
207 90 th
percentile formation date. The results are found to be qualitatively similar between linear
208 regressions conducted against the 10 th
and 25 th
percentile formation dates and between linear
209 regressions conducted against the 75 th and 90 th percentile formation dates (not shown).
210 Consequently, the results presented in Section 4 consider only linear regression computed
211 between June monthly-mean fields and the 10 th
percentile formation date and between November
212 monthly-mean fields and the 90 th
percentile formation date. As presented herein, positive values
213 of (1) indicate an earlier start – as measured by the 10 th
percentile formation date – or later end –
214 as measured by the 90 th percentile formation date – of the TC season as a given oceanic or
215 atmospheric field increases in value. All linear regression analyses are conducted globally with
9

216 particular focus given to describing the results across the North Atlantic Ocean, North America,
217 and equatorial Pacific Ocean. This area extends into the equatorial Pacific Ocean so as to reflect
218 the potential influence of the El Niño Southern Oscillation (ENSO; e.g., Bjerknes 1969) upon
219 Atlantic basin TC activity, particularly late in the Atlantic TC season (Dunion 2011; Klotzbach
220 2011a,b).
221 To complement the spatial linear regression analyses described above, we compute the
222 linear correlation coefficient between the 10 th (90 th ) percentile formation date and June
223 (November) monthly mean values of selected teleconnection indices. Teleconnection indices
224 considered include the Arctic Oscillation (AO; e.g., Thompson and Wallace 1998), Atlantic
225 Meridional Mode (e.g., Moura and Shukla 1981), Atlantic Multidecadal Oscillation (AMO; e.g.,
226 Kerr 2000), North Atlantic Oscillation (e.g., Barnston and Livezey 1987), Oceanic Niño Index
227 (ONI; e.g., L’Heureux et al. 2013), Pacific Decadal Oscillation (PDO; e.g., Mantua et al. 1997),
228 Pacific-North American (PNA; e.g., Wallace and Gutzler 1981), Quasi-Biennial Oscillation
229 (QBO; e.g., Baldwin et al. 2001), and Sahel Precipitation Index (SPI; e.g., Haywood et al. 2013).
230 Many of these teleconnection indices have been shown to exert a control on North Atlantic TC
231 activity on subseasonal to interannual scales, particularly for that which occurs at lower latitudes
232 (e.g., Gray 1984, Kossin and Vimont 2007, Kossin et al. 2010, Villarini et al. 2010, Klotzbach
233 2011, Villarini et al. 2012, to cite but a few examples). Only linear relationships that are
234 statistically-significant to ≥ 90% confidence, as assessed utilizing a Student’s t -test, are reported
235 upon in the results that follow.
236 3. Results: Long-Term Trends in Season Length
10

237 Over the period 1979-2007, when considering only TCs that formed within the extended
238 MDR of Kossin (2008), the Atlantic TC season increased in length by approximately 1.5 days
239 per year, as primarily associated with a later end to the season (Fig. 3a). This trend is not
240 statistically-significant to ≥ 90% confidence, however. These results are nearly identical to those
241 obtained for the period 1980-2007 considered by Kossin (2008, their Fig. 4c), with subtle
242 differences resulting from the inclusion of 1979 within our analysis. For the periods 1979-2001
243 and 1979-2002, no trend – statistically-significant or otherwise – in Atlantic TC season length
244
245 can be identified (Figs. 3f,g). There is some indication of a broadening of the peak of the season
– here defined as formation events occurring between the 25 th
and 75 th
percentiles – but this
246 result is also not statistically-significant to ≥ 90% confidence. Introducing subsequent seasons
247 into the analysis (Figs. 3b-e), particularly the active 2005 season (Beven et al. 2008), results in
248 the gradual development of the positive but non-statistically-significant trend identified over the
249 period 1979-2007 (Fig. 3a). This trend primarily results from four seasons (2001, 2003, 2005,
250 and 2007) with TC season end dates – here represented by the 95 th
percentile TC formation date
251 to increase the degrees of freedom of the analysis – two to seven weeks later than the 1979-2014
252 average (Table 1).
253 Extending the end of the analysis through 2014 eliminates any trend in Atlantic TC
254 season length (Fig. 3h). The below-average 2009 season (Berg and Avila 2011) contributes
255 particularly strongly to the elimination of this trend, as manifest by a season end date
256 approximately two-and-a-half weeks earlier than the long-term average (Table 2). This implies
257 that the positive trend identified by Kossin (2008) to between 80-90% confidence is primarily a
258 function of the several abnormally-long TC seasons seen during the early-to-mid-2000s. It
259 should be noted that extending the analysis through 2014 also decreases uncertainty in the
11

260 results, as manifest by a narrower 90% confidence interval at both the start and end of the
261 season.
262 When considering TC formation events across the entire Atlantic basin, no statistically-
263 significant trend in season length is identified for the period 1979-2007 (Fig. 4a) despite
264 numerous late-ending TC seasons in the early-to-mid 2000s (Table 2). For the periods 1979-2001
265 and 1979-2002, a non-statistically-significant trend towards seasons that are shorter by one day
266 per year is identified (not shown). The comparatively longer seasons of 2003 through 2007
267 (Table 2) act to eliminate this trend but are not sufficient to result in an identifiable trend toward
268 longer seasons. Furthermore, no trend – statistically-significant or otherwise – in Atlantic TC
269 season length is identified for the period 1979-2014 (Fig. 4b). As for the Kossin (2008) extended
270 MDR, the uncertainty in this result is somewhat less than that associated with the period 1979-
271 2007 (c.f. Figs. 4a and 4b). Consequently, it is argued that the positive but non-statistically-
272 significant trend toward longer Atlantic TC seasons identified by Kossin (2008) is not indicative
273 of a robust long-term trend but is primarily the result of both the subset of TC formation events
274 considered and end date of the analysis. Fluctuations in season length therefore appear to be
275 primarily controlled by subseasonal to interannual variability in the necessary conditions for TC
276 formation, particularly early and late in the season.
277 4. Results: Seasonal Influences upon Season Length
278 a. Early-starting seasons
279 Tropical cyclones that form before the 10 th
percentile date for each season preferentially
280 do so in the Gulf of Mexico, the northwestern Caribbean Sea, and near the southeastern United
281 States coastline (Fig. 1). Note that the TCs within this composite that form in the MDR are
12

282 almost exclusively associated with later-starting Atlantic TC seasons. The preferred genesis
283 pathway for TCs that form in the preferred regions for early-season development is weak tropical
284 transition, as evaluated utilizing the genesis pathway classification database of McTaggart-
285 Cowan et al. (2013). For reference, tropical transition (TT; Davis and Bosart 2003, 2004) occurs
286 in environments in which organized deep, moist convection upshear of an extratropical cyclone,
287 through the horizontal and vertical redistribution of potential vorticity, acts to weaken initially
288 strong vertical wind shear atop the surface low. If the cyclone is located over sufficiently warm
289 waters (SST ≥ 24-26°C) for a sufficiently long period of time (≥ 24 h) and is of sufficient
290 intensity so as to foster wind-induced surface heat exchange (e.g., Emanuel 1986), an initially
291 cold-core extratropical cyclone may transform into a warm-core TC. In the preferred genesis
292 locations for early-season Atlantic TCs, June monthly-mean SST is between 24-28°C (Fig. 5a),
293 600 hPa relative humidity is between 40-60% (Fig. 5b), 850 hPa relative vorticity is
294 approximately zero (Fig. 5d), and the 850-200 hPa vertical wind shear magnitude is less than 10
295 m s
-1
(Fig. 5e). In the June monthly-mean, as compared to known TC genesis-supporting
296 environments in the deep tropics (e.g., Gray 1968, 1979; Frank 1987) and subtropical to middle
297 latitudes (e.g., Mauk and Hobgood 2012), the SST and deep-layer vertical wind shear criteria for
298 tropical cyclogenesis generally are met, whereas the precursor disturbance and middle
299 tropospheric relative humidity criteria for tropical cyclogenesis generally are not met.
300 Earlier-starting Atlantic TC seasons are associated with a statistically-significant (to ≥
301 90% confidence, as assessed utilizing a Student’s t test) increase in June monthly-mean 850 hPa
302 relative vorticity in the eastern Gulf of Mexico (Fig. 5d). A one standard deviation increase in
303 850 hPa relative vorticity (0.5 x 10 -5 s -1 ; Fig. 7d) is associated with a six to nine day earlier start
304 to the Atlantic TC season relative to its mean. As the June climatological mean 850 hPa relative
13

305 vorticity is anticyclonic across the Gulf of Mexico (Fig. 5d), the presence of increased 850 hPa
306 relative vorticity in the eastern Gulf of Mexico in June indicates a greater likelihood that a
307 seedling disturbance for TC formation exists in the Gulf of Mexico in earlier-starting Atlantic TC
308 seasons. Given that most early-forming Atlantic TCs do so via tropical transition (Fig. 1), it is
309 believed that a stationary frontal boundary along and ahead of the June climatological mean 500
310 hPa trough in the far western Atlantic (Fig. 5c) is the likely source of such a disturbance.
311 However, it is not immediately clear as to what is the driving force behind increased 850 hPa
312 relative vorticity in the eastern Gulf of Mexico during earlier-starting Atlantic TC seasons.
313 Though 500 hPa geopotential height is typically lower across the Gulf of Mexico and
314 northwestern Caribbean Sea during earlier-starting Atlantic TC seasons, this correlation is not
315 statistically-significant to ≥ 90% confidence (Fig. 5c). It is possible that this result is a reflection
316 of the earlier-developing TCs themselves. However, this is unlikely given a typically short
317 duration of TCs near land compared to the duration of June and the lack of a similar relationship
318 between higher 600 hPa relative humidity and earlier-starting Atlantic TC seasons (Fig. 5b).
319
320
Though few coherent, statistically-significant (to ≥ 90% confidence) linear correlations between sub-seasonal variability and the 10 th
percentile Atlantic TC formation date exist within
321 the North Atlantic basin, the 10 th percentile Atlantic TC formation date is associated with
322 statistically-significant linear correlations to large-scale variability in the Southern Hemisphere
323 (Fig. 6). This is particularly manifest by variability in the 500 hPa longwave pattern (Fig. 6c) and
324 its impacts upon and relationship with atmospheric and oceanic variability in the remaining
325 fields considered (Fig. 6a,b,d,e). It is unclear, however, as to whether the relationship between
326 such variability, whether in whole or in part, and the 10 th percentile Atlantic TC formation date is
327 causal or merely associative in nature. The cause of this variability is also uncertain, as it does
14

328 not closely resemble that associated with any predominant planetary-scale modes of climatic
329 variability known to the authors. Therefore, further research is necessary to identify what is
330 responsible for the synoptic- to planetary-scale variability associated with early-starting Atlantic
331 TC seasons.
332 b. Late-ending seasons
333 Tropical cyclones that form after the 90 th
percentile formation date for each season
334 preferentially do so in the western Caribbean Sea and subtropical western Atlantic (Fig. 2). Late-
335 season TCs that form in the subtropical western North Atlantic almost exclusively are the result
336 of tropical transition, whereas TCs that form in the western Caribbean Sea either result from
337 tropical transition or form in non-baroclinic environments. In the preferred genesis locations for
338 late-season Atlantic TCs, November monthly-mean SST is between 24-28°C (Fig. 8a), 600 hPa
339 relative humidity is between 30-60% (Fig. 8b), 850 hPa relative vorticity is approximately zero
340 (Fig. 8d), and the 850-200 hPa vertical wind shear magnitude is between 10-20 m s
-1
(Fig. 8e). In
341 the November monthly-mean, as compared to known TC genesis-supporting environments, SST
342 generally is sufficiently warm, deep-layer vertical wind shear is marginally favorable, and both
343 middle tropospheric relative humidity and lower tropospheric relative vorticity are insufficiently
344 large so as to promote tropical cyclogenesis.
345
346
To first order, late-ending Atlantic TC seasons preferentially occur during cool-phase (La
Niña) ENSO events. This is manifest by a strengthened Walker circulation, as reflected in both
347 November monthly-mean SST (Figs. 8a and 9a) and 600 hPa relative humidity (Figs. 8b and 9b)
348 fields across the equatorial Pacific. This is also reflected by a statistically-significant (to ≥ 90%
349 confidence, as assessed utilizing a Student’s t -test) inverse linear relationship between the 90 th
15

350 percentile formation date and the November mean value of the ONI ( R = -0.33). In the context of
351 the linear regression analyses presented in Figs. 8 and 9, a one standard deviation reduction in
352 SST (1-2°C; Fig. 10a) and 600 hPa relative humidity (7.5-12.5%; Fig. 10b) in the central to
353 eastern equatorial Pacific is associated with a six to nine day extension of the Atlantic TC season
354 relative to its mean.
355
356
The increased likelihood of late-season TC formation in the western Caribbean Sea during La Niña events is consistent with Dunion (2011) and Klotzbach (2011a,b). Together,
357 these studies found that atmospheric conditions known to be hostile for TC development –
358 namely, increased vertical wind shear, increased static stability, and decreased middle-
359
360 tropospheric relative humidity – occur in the western Caribbean Sea during October and
November more (less) frequently during El Niño (La Niña) events. To that end, increased
361 November monthly-mean 600 hPa relative humidity (Fig. 8b), increased November monthly-
362 mean 850 hPa relative vorticity (Fig. 8d), and decreased November monthly-mean 850-200 hPa
363 vertical wind shear (Fig. 8e), relative to their respective climatologies, in the Caribbean Sea
364 during later-ending Atlantic TC seasons are consistent with the influence of La Niña. It is
365 uncertain, however, as to the extent that an increased (a decreased) likelihood of late-season TC
366 formation at higher latitudes is directly attributable to La Niña (El Niño) and its influence upon
367 the global circulation during boreal fall (e.g., Horel and Wallace 1981).
368 Late-ending Atlantic TC seasons also preferentially occur during negative-phase PDO
369 events. This is manifest by abnormally cold November monthly-mean SST along the west coast
370 of North America and abnormally warm November monthly-mean SST in the central North
371 Pacific Ocean (Figs. 8a and 9a). This is also reflected by a statistically-significant (to ≥ 90%
372 confidence, as assessed utilizing a Student’s t -test) inverse linear relationship between the 90 th
16

373 percentile formation date and the November mean value of the PDO index ( R = -0.34). In the
374 context of the linear regression analyses presented herein, a one standard deviation increase in
375
376
SST across the central North Pacific (0.5-1°C; Fig. 10a) and decrease in SST along the west coast of the United States (≈ 0.5°C; Fig. 10a) are associated with an approximate six day increase
377
378 in the length of the Atlantic TC season relative to its mean. The relationship between late-season
Atlantic TC activity and the PDO has, to the authors’ knowledge, not yet been documented
379 within the refereed literature, and identification of the physical underpinnings behind this
380 relationship is planned for further study.
381 In closer proximity to where late-season Atlantic TCs form, late-ending Atlantic TC
382 seasons are associated with an eastward shift of the November monthly-mean 500 hPa longwave
383 pattern across the United States and subtropical western North Atlantic Ocean (Fig. 8c). A one
384 standard deviation (30-40 m; Fig. 10c) increase (decrease) in 500 hPa geopotential height across
385 the central United States (subtropical western North Atlantic Ocean) is associated with a three to
386 nine day extension of the Atlantic TC season relative to its mean. Though not statistically-
387 significant to ≥ 90% confidence, late-ending Atlantic TC seasons are also associated with
388 increased 500 hPa geopotential heights near Greenland (Fig. 8c), wherein a one standard
389 deviation (≥ 50 m; Fig. 10c) increase in 500 hPa geopotential height is associated with an
390 approximate three day extension of the Atlantic TC season relative to its mean. The occurrence
391 of below-normal 500 hPa geopotential heights in the subtropical western North Atlantic Ocean
392 with above-normal 500 hPa geopotential heights near Greenland is consistent with an elevated
393 likelihood of cut-off extratropical cyclones in the subtropical western North Atlantic. Further,
394 given sufficiently warm SSTs, this is consistent with an increased likelihood of late-season
395 tropical transition (TT) occurrence in the subtropical western Atlantic so long as deep, moist
17

396 convection associated with a given extratropical cyclone is able to locally reduce the 850-200
397 hPa vertical wind shear on meso- to synoptic time scales (Davis and Bosart 2003, 2004).
398 c. Representativeness of linear regression analyses
399 Spatial linear correlation is used to quantify the extent to which the linear regression
400 analyses presented in Figs. 5-6 and 8-9 are representative of the variability in monthly-mean
401 fields observed with both early- and late-starting and -ending seasons. Herein, spatial linear
402 correlation is computed over the region 10°S-70°N, 150°-0°W, consistent with the region
403 depicted in Figs. 5 and 8. The statistically-significant linear correlations between the 90 th
404 percentile Atlantic TC formation date and boreal autumn atmospheric and oceanic variability
405 associated with both the ENSO and PDO (Section 4b) motivate the selection of the southern and
406 western extents of this domain. The genesis locations for many of the earliest- and latest-forming
407 Atlantic TCs (Figs. 1 and 2), many of which originate from precursor disturbances of mid-
408 latitude origin, motivate the selection of the northern extent of this domain. Similar qualitative
409 insight to that described below is obtained when spatial linear correlation is computed only over
410
411 the tropical North Atlantic (0°-30°N, 95°-10°W) or over a global domain (not shown). As a result, only the results for the region 10°S-70°N, 150°-0°W are presented below.
412 The results of the spatial linear correlation analyses are presented in Tables 3 and 4. To
413 first order, the linear regression analyses are to some extent representative of variability in
414 monthly-mean fields observed with both early- and late-starting and -ending seasons. Anomalies
415 in the monthly-mean fields for the five earliest-starting and latest-ending Atlantic TC seasons are
416 generally positively correlated with the linear regression fields, while anomalies in the monthly-
417 mean fields for the five latest-starting and earliest-ending Atlantic TC seasons are generally
18

418 negatively correlated with the linear regression fields. However, correlations are generally weak,
419 highly variable between individual Atlantic TC seasons within a given five-member composite,
420 and stronger for some variables than others. This implies a limit upon the predictive ability of the
421 linear regression analyses.
422 For example, consider the Atlantic TC seasons with the five latest 90 th percentile
423 formation dates, as listed in Table 4. Previously, La Niña and negative-phase PDO events were
424 demonstrated to be associated with later-ending Atlantic TC seasons. Of the five-latest ending
425
426
Atlantic TC seasons, as classified using the ONI, four were associated with neutral ENSO conditions and one (1994) was associated with El Niño conditions. Likewise, as classified by the
427 PDO index, two of the five latest-ending Atlantic TC seasons were associated with weak
428 positive-phase PDO conditions. Thus, while La Niña and negative-phase PDO events appear to
429 increase the likelihood of a later-than-normal end to the Atlantic TC season, they do not
430 represent the only pathways to a later-than-normal Atlantic TC season end. Further, slower-
431 evolving fields with significant synoptic-scale structure in the linear regression analyses –
432 particularly SST, but also 600 hPa relative humidity, 500 hPa geopotential height and 850-200
433 hPa vertical wind shear magnitude – are generally associated with stronger correlations. The
434 more rapidly-evolving 850 hPa relative vorticity field, with primarily mesoscale structure in its
435 linear regression analyses, is generally associated with weaker correlations. This implies that
436 meso- to synoptic-scale variability not captured by the linear regression analyses, particularly
437 with respect to whether or not a predecessor disturbance from which a TC can originate exists,
438 contributes to variability in Atlantic TC season length.
439 5. Summary and Discussion
19

440 The length of the Atlantic TC season varies from year to year with some seasons
441 significantly shorter or longer than the average. Albeit only to between 80-90% confidence,
442 Kossin (2008) suggested that the duration of the year over which TCs formed within an extended
443 main development region of the Atlantic basin has increased by 1-2 days per year over the course
444 of the last several decades. Herein, it was first determined whether this result is maintained when
445 TC genesis events occurring over the full Atlantic basin are considered or when the period of
446 record is extended to the present. Subsequently, monthly-mean synoptic- to planetary-scale
447 conditions associated with early and late starting and ending Atlantic TC seasons were examined.
448 Quantile regression was applied to National Hurricane Center HURDAT2 data (Landsea
449 and Franklin 2013) for the 36 year period between 1979 and 2014 to first replicate and
450 subsequently expand upon Kossin (2008)’s findings. Extending the end of the analysis through
451 2014 eliminated the positive but non-statistically-significant trend toward longer Atlantic TC
452 seasons identified by Kossin (2008). Rather, this trend appears to be a function of several
453 abnormally-long Atlantic TC seasons – 2001, 2003, 2005, and 2007 – that occurred during the
454
455 early-to-mid-2000s (Tables 1 and 2). Of these particular seasons, all but 2003 were associated with weak to moderate La Niña and negative-phase PDO conditions, each of which were found
456 to be associated with later ends to the Atlantic TC season. However, the precise reasons as to
457 why these seasons were considerably longer than normal remains to some extent unclear, and
458 further study is planned to address this question. Furthermore, when TC formation events across
459 the entire Atlantic basin were considered, no statistically-significant trend in Atlantic TC season
460 length was identified for the period 1979-2007 and no trend whatsoever in Atlantic TC season
461 length was identified for the period 1979-2014. Variations in season length therefore appear to
20

462 be primarily controlled by interannual variability in the necessary conditions for TC formation,
463 particularly early and late in the Atlantic TC season.
464 Utilizing linear regression applied to June and November monthly-mean reanalysis data,
465 synoptic- to planetary-scale variability associated with early and late starting and ending Atlantic
466 TC seasons was identified. Earlier starting Atlantic TC seasons are associated with increased
467 June monthly-mean 850 hPa relative vorticity across the eastern Gulf of Mexico, indicative of a
468 greater likelihood that a precursor disturbance for TC formation exists in the Gulf of Mexico in
469 June in earlier-starting Atlantic TC seasons. Given that most early-forming Atlantic TCs do so
470 via tropical transition, it is believed that a stationary frontal boundary along and ahead of the
471 June climatological mean 500 hPa trough in the far western Atlantic is the likely source of such a
472 disturbance. There also exist statistically-significant correlations between the 10 th
percentile
473 Atlantic TC formation date and large-scale variability within the Southern Hemisphere.
474 However, whether these correlations are causal or merely associative is uncertain, as are the
475 causes of such variability.
476 Late-ending Atlantic TC seasons primarily occur during La Niña and negative-phase
477 PDO events. The former is consistent with several previous studies (e.g., Dunion 2011;
478 Klotzbach 2011a,b) and is believed to be the primary influence upon late-forming TCs
479 (independent of genesis pathway) in the western Caribbean Sea. Further investigation into the
480 physical connection(s) associated with the latter is planned for further study. A later end to the
481 Atlantic TC season is also associated with an eastward shift in the November monthly-mean 500
482 hPa longwave pattern across the United States and subtropical western North Atlantic Ocean. It
483 is this modulation of the atmospheric pattern that is hypothesized to increase the likelihood of
484 late-season TT events across the subtropical Atlantic Ocean, although it is uncertain as to
21

485 whether this result can also be attributed to modulation of the mid-latitude atmospheric pattern
486 by the ENSO and/or PDO.
487 While the linear regression analyses are to some extent representative of variability in
488 monthly-mean fields observed with both early- and late-starting and -ending seasons, there exist
489 multiple pathways to an early- or late-starting and/or -ending Atlantic TC season. Consequently,
490 while associated with statistically-significant (to ≥ 90% confidence) linear relationships, these
491 results only explain a small portion of the total variation in Atlantic TC season length. For
492 instance, in order for TCs to form, environmental conditions must locally be favorable on shorter
493 time scales of several hours to a few days, such as may be associated with a transient Madden-
494 Julian Oscillation event (e.g., Klotzbach 2010) and/or a convectively-coupled atmospheric
495 Kelvin wave (e.g., Schreck 2015). Thus, while the results elucidate variability in the large-scale
496 conditions necessary to support TC formation associated with early- and late-starting and -ending
497 Atlantic TC seasons, variability on smaller spatiotemporal scales must be known in order to
498 more completely quantify variability in Atlantic TC season length.
499 The results presented in this manuscript motivate a number of future studies aimed both
500 at better understanding past, present, and future variability in TC season length and at
501 quantifying the large-scale conditions that promote early-starting and late-ending TC seasons.
502 For instance, what results in multiple successive short or long Atlantic TC seasons, such as was
503 observed in the early 2000s? Note that while no statistically-significant linear relationship exists
504 between the 10 th
percentile ( R = -0.18) or 90 th
percentile ( R = 0.11) formation dates and annual
505 Atlantic TC count, the 90 th
minus 10 th
percentile formation date ( R = 0.43) is strongly linearly
506 correlated to annual Atlantic TC count. Thus, to some extent, more (less) active Atlantic TC
507 seasons are also longer (shorter) Atlantic TC seasons, consistent with the hypothesis of Dwyer et
22

508 al. (2015). Case studies of both early- and late-starting and -ending seasons may provide further
509 insight toward addressing this question. Furthermore, the methods utilized in this study can be
510 readily adapted to understand long-term trends in season length and large-scale variability
511 contributing to early- or late-starting and -ending seasons in other oceanic basins that feature
512 well-defined TC seasons. These methods may also be readily adapted to subsets of cyclones –
513 e.g., only TCs with maximum sustained surface winds of ≥ 33 m s
-1
– for any basin in which TCs
514 occur. Finally, given appropriate downscaling and statistical sampling methods, these methods
515 may be applied to climate model outputs to evaluate potential future changes in both TC season
516 length and early- and late-season genesis pathways under a wide range of emissions scenarios.
517 The results presented herein also motivate an investigation into the predictability – or
518 lack thereof – of the large-scale conditions that promote early- or late-starting and -ending TC
519 seasons, whether in the Atlantic basin or elsewhere. In other words, are shorter- and longer-than-
520 normal TC seasons primarily driven by synoptic-scale variability? Or, are such events primarily
521 driven by sub-seasonal to climate-scale variability that evolves more slowly (and is thus more
522 predictable) than that on the synoptic-scale? The strong linear correlations of Atlantic TC season
523 length with the ENSO, PDO, and seasonal Atlantic-basin TC count argue in favor of sub-
524 seasonal to climate-scale controls upon Atlantic TC season length. Likewise, we note that linear
525 regression analyses conducted between the 10 th
percentile formation date and May monthly-
526 mean fields and between the 90 th
percentile formation date and October monthly-mean fields
527 bear some resemblance to those presented in Figs. 5-6 and 8-9, respectively, particularly for the
528 slower-evolving SST and 600 hPa relative humidity fields (not shown). However, the limited
529 overall extent to which the linear regression analyses are representative of observed variability in
530 both early- and late-starting and -ending Atlantic TC seasons argues against dominant climate-
23

531 scale controls, though it remains possible that such relationships may exist if combinations of
532 multiple modes of climate variability are considered. Further investigation is planned to address
533 the intrinsic predictability of comparatively short and long Atlantic TC seasons.
534 Acknowledgements
535 We acknowledge fruitful discussions with Rob Hodges, Sergey Kravtsov, and Kyle
536 Swanson during the course of this research. We are indebted to Ron McTaggart-Cowan for
537 making available to us the genesis pathway classification database described in McTaggart-
538 Cowan et al. (2013). This manuscript benefitted greatly from constructive review comments
539 provided by Phil Klotzbach and two anonymous reviewers. Monthly mean ERA-Interim data
540 were obtained from the ECMWF. NOAA ERSST v3b SST data were obtained from the
541 NOAA/OAR/ESRL Physical Sciences Division. Teleconnection index data were obtained from
542 NOAA/NCEP/Climate Prediction Center (ONI, NAO, AO, PNA, QBO), the University of
543 Washington (PDO, SPI), and NOAA/OAR/ESRL Physical Sciences Division (AMM, AMO).
544 Quantile regression computations were carried out utilizing MATLAB code provided by Aslak
545 Grinsted.
546
24

547 References
548 Baldwin, M. P., and coauthors, 2001: The Quasi-Biennial Oscillation. Rev. Geophys.
, 39 , 179-
549 229.
550 Barnston, A. G., and R. E. Livezey, 1987: Classification, seasonality and persistence of low-
551 frequency atmospheric circulation patterns. Mon. Wea. Rev.
, 115 , 1083-1126.
552 Berg, R. J., and L. A. Avila, 2011: Atlantic hurricane season of 2009. Mon. Wea. Rev.
, 139 ,
553 1049-1069.
554 Beven, J. L. II, and coauthors, 2008: Atlantic hurricane season of 2005. Mon. Wea. Rev.
, 136 ,
555 1109-1173.
556 Bjerknes, J., 1969: Atmospheric teleconnections from the equatorial Pacific. Mon. Wea. Rev.
, 97 ,
557 163-172.
558 Bruyère, C. L., G. J. Holland, and E. Towler, 2012: Investigating the use of a genesis potential
559 index for tropical cyclones in the north Atlantic basin. J. Climate , 25 , 8611–8626.
560 Camargo, S. J., K. A. Emanuel, and A. H. Sobel, 2007: Use of a genesis potential index to
561 diagnose ENSO effects on tropical cyclone genesis. J. Climate , 20 , 4819–4834.
562 Davis, C. A., and L. F. Bosart, 2003: Baroclinically induced tropical cyclogenesis. Mon. Wea.
563 Rev.
, 131 , 2730-2747.
564 Davis, C. A., and L. F. Bosart, 2004: The TT problem: forecasting the tropical transition of
565 cyclones. Bull. Amer. Meteor. Soc.
, 85 , 1657-1662.
25

566 Dee, D. P., and coauthors, 2011: The ERA-Interim reanalysis: configuration and performance of
567 the data assimilation system. Quart. J. Roy. Meteor. Soc.
, 137 , 553-597.
568 Dunion, J. P., 2011: Rewriting the climatology of the tropical North Atlantic and Caribbean Sea
569 atmosphere. J. Climate , 24 , 893-908.
570 Dwyer, J. G., S. J. Camargo, A. H. Sobel, M. Biasutti, K. A. Emanuel, G. A. Vecchi, M. Zhao,
571 and M. K. Tippett, 2015: Projected 21 st
century changes in the length of the tropical
572 cyclone season. J. Climate , in press.
573 Efron, B., and R. Tibshirani, 1993: An Introduction to the Bootstrap . Chapman and Hall, 436pp.
574 Elsner, J. B., J. P. Kossin, and T. H. Jagger, 2008: The increasing intensity of the strongest
575 tropical cyclones. Nature , 455 , 92-95.
576 Emanuel, K. A., 1986: An air-sea interaction theory for tropical cyclones. Part I: steady-state
577 maintenance. J. Atmos. Sci.
, 43 , 585-605.
578 Emanuel, K. A., 2010: Tropical cyclone activity downscaled from NOAA-CIRES reanalysis,
579 1908–1958. J. Adv. Model. Earth Syst.
, 2 (1) , doi:10.3894/JAMES.2010.2.1.
580 Emanuel, K. A., and D. S. Nolan, 2004: Tropical cyclone activity and global climate. Preprints,
581 26th Conf. on Hurricanes and Tropical Meteorology , Miami, FL, Amer. Meteor. Soc.,
582 240–241.
583
Frank, W. M., 1987: “Tropical cyclone formation.” Chapter 3,
A Global View of Tropical
584 Cyclones , Office of Naval Research, Arlington, Virginia, 53-90.
26

585 Gray, W. M., 1979: Hurricanes: Their formation, structure and likely role in the tropical
586 circulation. Meteorology over the Tropical Oceans , D. B. Shaw, Ed., Royal
587 Meteorological Society, 155–218.
588 Haywood, J. M., A. Jones, N. Bellouin, and D. Stephenson, 2013: Asymmetric forcing from
589 stratospheric aerosols impacts Sahelian rainfall. Nature Climate Change , 3 , 660-665.
590 Horel, J. D., and J. M. Wallace, 1981: Planetary-scale atmospheric phenomena associated with
591 the Southern Oscillation. Mon. Wea. Rev.
, 109 , 813-829.
592
Hurricane Research Division (HRD), cited 2015: “Tropical Cyclone FAQ: When is hurricane
593 season?” [Available online at http://www.aoml.noaa.gov/hrd/tcfaq/G1.html
].
594 Kerr, R. A., 2000: A North Atlantic climate pacemaker for the centuries. Science , 288 , 1984-
595 1985.
596 Klotzbach, P. J., 2010: On the Madden-Julian Oscillation-Atlantic hurricane relationship. J.
597 Climate , 23 , 282-293.
598
Klotzbach, P. J., 2011a: The influence of El Niño-Southern Oscillation and the Atlantic
599 multidecadal oscillation on Caribbean tropical cyclone activity. J. Climate , 24 , 721-731.
600 Klotzbach, P. J., 2011b: Forecasting October-November Caribbean hurricane days. J. Geophys.
601 Res.
, 116 , D18117, doi:10.1029/2011JD016146.
602 Koenker, R., and K. F. Halloch, 2001: Quantile regression. J. Econ. Perspect.
, 15 , 143-156.
603 Koenker, R., and G. Bassett, 1978: Regression quantiles. Econometrica , 46 , 33-50.
27

604 Kossin, J. P., 2008: Is the North Atlantic hurricane season getting longer? Geophys. Res. Lett.
,
605 35 , L23705.
606 Landsea, C. W., 2007: Counting Atlantic tropical cyclones back to 1900. Eos, Trans. Amer.
607 Geophys. Union , 88 , 197-203.
608 Landsea, C. W., and J. L. Franklin, 2013: Atlantic hurricane database uncertainty and
609 presentation of a new database format. Mon. Wea. Rev.
, 141 , 3576-3591.
610 L’Heureux, M. L., D. C. Collins, and Z.-Z. Hu, 2013: Linear trends in sea surface temperature of
611
612 the tropical Pacific Ocean and implications for the El Niño-Southern Oscillation. Clim.
Dyn.
, 40 , 1223-1236.
613 Mantua, N. J., S. R. Hare, Y. Zhang, J. M. Wallace, and R. C. Francis, 1997: A Pacific
614 interdecadal oscillation with impacts on salmon production. Bull. Amer. Meteor. Soc.
, 78 ,
615 1069-1079.
616 Mauk, R. G., and J. S. Hobgood, 2012: Tropical cyclone formation in environments with cool
617 SST and high wind shear over the northeastern Atlantic Ocean. Wea. Forecasting , 27 ,
618 1433-1448.
619 McGauley, M. G., and D. S. Nolan, 2011: Measuring environmental favorability for tropical
620 cyclogenesis by statistical analysis of threshold parameters. J. Climate , 24 , 5968–5997.
621 McTaggart-Cowan, R., G. D. Deane, L. F. Bosart, C. A. Davis, and T. J. Galarneau, Jr., 2008:
622 Climatology of tropical cyclogenesis in the North Atlantic (1948-2004). Mon. Wea. Rev.
,
623 136 , 1284-1304.
28

624 McTaggart-Cowan, R., T. J. Galarneau, Jr., L. F. Bosart, R. W. Moore, and O. Martius, 2013: A
625 global climatology of baroclinically influenced tropical cyclogenesis. Mon. Wea. Rev.
,
626 141 , 1963-1989.
627 Moura, A. D., and J. Shukla, 1981: On the dynamics of droughts in northeast Brazil:
628 observations, theory, and numerical experiments with a general circulation model. J.
629 Atmos. Sci.
, 38 , 2653-2675.
630 Neumann, C. J., B. R. Jarvinen, C. J. McAdie, and G. R. Hammer, 1999: Tropical cyclones of
631 the North Atlantic Ocean, 1871-1998. Historical Climatology Series 6-2 , 206 pp.
632 Schreck, C. J., 2015: Kelvin waves and tropical cyclogenesis: a global survey. Mon. Wea. Rev.
,
633 in press.
634 Schumacher, A. B., M. DeMaria, and J. A. Knaff, 2009: Objective estimation of the 24-h
635 probability of tropical cyclone formation. Wea. Forecasting , 24 , 456-471.
636
637
Smith, T. M., R. W. Reynolds, T. C. Peterson, and J. Lawrimore, 2008: Improvements to
NOAA’s historical merged land-ocean surface temperature analysis (1880-2006).
J.
638 Climate , 21 , 2283-2296.
639 Tang, B., and S. J. Camargo, 2014: Environmental control of tropical cyclones in CMIP5: A
640 ventilation perspective, J. Adv. Model. Earth Syst.
, 6 , 115-128.
641 Tang, B., and K. Emanuel, 2012: A ventilation index for tropical cyclones. Bull. Amer. Meteor.
642 Soc.
, 93 , 1901-1912.
643 Thompson, D. W. J., and J. M. Wallace, 1998: The Arctic Oscillation signature in the wintertime
644 geopotential height and temperature fields. Geophys. Res. Lett.
, 25 , 1297-1300.
29

645 Tippett, M. K., S. J. Camargo, and A. H. Sobel, 2011: A Poisson regression index for tropical
646 cyclone genesis and the role of large-scale vorticity in genesis. J. Climate , 24 , 2335–
647 2357.
648 Torn, R. D., and G. J. Hakim, 2008: Ensemble-based sensitivity analysis. Mon. Wea. Rev.
, 136 ,
649 663-677.
650 Uppala, S. M., and coauthors, 2005: The ERA-40 re-analysis. Quart. J. Roy. Meteor. Soc.
, 131 ,
651 2961-3012.
652 Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field during
653 the Northern Hemisphere winter. Mon. Wea. Rev.
, 109 , 784-812.
654
30

655 List of Tables
656 Table 1: 5 th
and 95 th
percentile TC formation dates for TCs forming within the extended MDR
657
658 of Kossin (2008) for the years 2001-2007. For reference, the 1979-2014 average 5 th
and
95 th
percentile TC formation dates are provided in the bottom row.
659 Table 2: As in Table 1, except for TCs forming within the full Atlantic basin.
660 Table 3: Linear correlation coefficient, computed over the region 10°S-70°N, 150°-0°W,
661
662 between the departure in the June monthly-mean field from the 1979-2014 mean June monthly-mean field and the slope of the best-fit linear regressions between the 10 th
663
664
665 percentile Atlantic TC formation date and June monthly-mean fields for the Atlantic TC seasons with the five-earliest and five-latest 10 th
percentile formation dates. Positive
(negative) linear correlation coefficients indicate that the anomaly field in question is of
666 like (opposite) sense to the regression field.
667 Table 4: As in Table 3, except for November monthly-mean fields and the 90 th
percentile
668 Atlantic TC formation date.
669
31

670
2006
2007
2008
2009
2010
2011
Season
2001
2002
2003
2004
2005
2012
2013
2014
1979-2014 average
5 th Percentile Formation Date 95 th Percentile Formation Date
August 19 November 14
August 30
July 26
August 11
July 9
August 8
August 19
July 20
August 13
August 13
August 7
August 3
July 13
August 6
August 9
September 22
November 19
September 19
December 8
September 24
November 28
October 13
September 29
October 29
October 21
October 17
October 15
October 12
October 13
671
672 Table 1: 5 th and 95 th percentile TC formation dates for TCs forming within the extended MDR
673 of Kossin (2008) for the years 2001-2014. For reference, the 1979-2014 average 5 th
and 95 th
674 percentile TC formation dates are provided in the bottom row.
675
32

676
2006
2007
2008
2009
2010
2011
Season
2001
2002
2003
2004
2005
2012
2013
2014
1979-2014 average
5 th Percentile Formation Date 95 th Percentile Formation Date
July 16 November 11
July 27
June 12
August 6
June 30
June 27
July 10
June 25
August 13
August 1
July 16
May 26
June 13
July 12
July 15
September 23
December 5
October 27
November 27
September 22
November 12
October 20
October 23
October 29
October 25
October 22
November 2
October 22
October 24
677
678 Table 2: As in Table 1, except for TCs forming within the full Atlantic basin.
679
33

680
Year
2012
1986
2013
1981
2003
1992
1984
1999
1998
1983
Maximum
Minimum
SST
0.04
0.23
0.14
0.54
0.36
-0.36
-0.26
-0.14
-0.47
-0.47
0.54
-0.47
600 hPa relative humidity
500 hPa geopotential height
850 hPa relative vorticity
0.09
0.27
0.29
0.29
0.01
-0.33
-0.33
-0.08
-0.37
-0.37
0.29
-0.37
0.21
0.02
0.07
0.05
0.31
-0.36
-0.26
-0.05
-0.24
-0.35
0.37
-0.42
0.18
-0.01
0.08
0.03
0.09
-0.04
-0.20
-0.13
-0.17
-0.11
0.18
-0.20
850-200 hPa vertical wind shear magnitude
0.21
0.30
0.24
0.18
0.17
-0.34
-0.24
-0.27
-0.35
-0.28
0.30
-0.35
681
682 Table 3: Linear correlation coefficient, computed over the region 10°S-70°N, 150°-0°W,
683 between the departure in the June monthly-mean field from the 1979-2014 mean June monthly-
684 mean field and the slope of the best-fit linear regressions between the 10 th
percentile Atlantic TC
685 formation date and June monthly-mean fields for the Atlantic TC seasons with the five-earliest
686 and five-latest 10 th
percentile formation dates. Positive (negative) linear correlation coefficients
687 indicate that the anomaly field in question is of like (opposite) sense to the regression field.
688
34

689
Year
2005
2003
1980
1994
2001
1983
1997
1993
2006
1979
Maximum
Minimum
SST
0.75
0.50
-0.21
-0.24
0.59
-0.26
-0.77
-0.47
-0.29
-0.40
0.75
-0.81
600 hPa relative humidity
500 hPa geopotential height
850 hPa relative vorticity
0.33
0.09
0.16
0.18
0.16
-0.23
-0.48
0.06
-0.28
-0.25
0.38
-0.51
0.40
-0.07
0.04
0.02
0.23
-0.14
-0.62
-0.32
0.02
-0.31
0.44
-0.62
0.18
0.10
0.12
0.14
0.10
-0.07
-0.28
-0.03
-0.16
-0.18
0.18
-0.28
850-200 hPa vertical wind shear magnitude
0.37
-0.14
0.10
-0.17
0.38
0.08
-0.60
-0.51
-0.33
-0.55
0.68
-0.60
690
691 Table 4: As in Table 3, except for November monthly-mean fields and the 90 th
percentile
692 Atlantic TC formation date.
693
35

694 List of Figures
695 Figure 1: Formation location, as assessed utilizing National Hurricane Center “best track” data,
696 of all Atlantic TCs forming on or prior to the 10 th
percentile formation date for each
697
698
Atlantic TC season. For TC formation events between 1979 and 2010, the highestprobability genesis pathway determined following McTaggart-Cowan et al. (2013) is
699
700 illustrated by the appropriate symbol. The genesis date within two week bins is referenced by the color given to each symbol.
701 Figure 2: As in Figure 1, except for Atlantic TCs forming on or after the 90 th
percentile
702 formation date for each Atlantic TC season.
703 Figure 3: Trend, in days per year, in the 5 th
to 95 th
percentile Atlantic TC formation dates
704
705 between (a) 1979-2007, (b) 1979-2006, (c) 1979-2005, (d) 1979-2004, (e) 1979-2003, (f)
1979-2002, (g) 1979-2001, and (h) 1979-2014 for TCs that formed within the Kossin
706
707
(2008) subset of the full Atlantic basin. The blue line indicates the 90% confidence interval.
708 Figure 4: As in Figure 3, except for TCs that formed anywhere within the North Atlantic basin,
709 for (a) 1979-2007 and (b) 1979-2014.
710 Figure 5: Slope of the best-fit regression line between the 10 th percentile Atlantic TC formation
711
712
713
714
715 date and June monthly-mean (a) sea surface temperature (°C), (b) 600 hPa relative humidity (%), (c) 500 hPa geopotential height (m), (d) 850 hPa relative vorticity (x10
-5
s
-
1
), and (e) 850-200 hPa vertical wind shear magnitude (m s
-1
) over the domain 10°S-
70°N, 150°-0°W. A one standard deviation increase in a given monthly-mean field is associated with an n day change (shaded; positive = earlier) in the 10 th percentile Atlantic
36

716
717
718
TC formation date. The June monthly-mean field is contoured in each panel. Crosshatching is utilized to identify regions where the slope of the best-fit regression line is non-zero to at least 90% (light gray), 95% (dark gray), and 99% (black) confidence.
719 Figure 6: As in Figure 5, except globally.
720 Figure 7: Standard deviation of the June monthly-mean (a) sea surface temperature (°C), (b) 600
721
722 hPa relative humidity (%), (c) 500 hPa geopotential height (m), (d) 850 hPa relative vorticity (x10
-5
s
-1
), and (e) 850-200 hPa vertical wind shear magnitude (m s
-1
).
723 Figure 8: Slope of the best-fit regression line between the 90 th
percentile Atlantic TC formation
724
725
726
727 date and November monthly-mean (a) sea surface temperature (°C), (b) 600 hPa relative humidity (%), (c) 500 hPa geopotential height (m), (d) 850 hPa relative vorticity (x10
-5
s
-
1
), and (e) 850-200 hPa vertical wind shear magnitude (m s
-1
) over the domain 10°S-
70°N, 150°-0°W. A one standard deviation increase in a given monthly-mean field is associated with an n day change (shaded; positive = later) in the 90 th
percentile Atlantic 728
729 TC formation date. The November monthly-mean field is contoured in each panel. Cross-
730
731 hatching is utilized to identify regions where the slope of the best-fit regression line is non-zero to at least 90% (light gray), 95% (dark gray), and 99% (black) confidence.
732 Figure 9: As in Figure 8, except globally.
733 Figure 10: As in Figure 7, except for November monthly-mean fields.
734
37

735
736 Figure 1: Formation location, as assessed utilizing National Hurricane Center “best track” data,
737 of all Atlantic TCs forming on or prior to the 10 th
percentile formation date for each Atlantic TC
738 season. For TC formation events between 1979 and 2010, the highest-probability genesis
739 pathway determined following McTaggart-Cowan et al. (2013) is illustrated by the appropriate
740 symbol. The genesis date within two week bins is referenced by the color given to each symbol.
741
38

742
743 Figure 2: As in Figure 1, except for Atlantic TCs forming on or after the 90 th
percentile
744 formation date for each Atlantic TC season.
745
39

746
747 Figure 3: Trend, in days per year, in the 5 th
to 95 th
percentile Atlantic TC formation dates
748 between (a) 1979-2007, (b) 1979-2006, (c) 1979-2005, (d) 1979-2004, (e) 1979-2003, (f) 1979-
749 2002, (g) 1979-2001, and (h) 1979-2014 for TCs that formed within the Kossin (2008) subset of
750 the full Atlantic basin. The blue line indicates the 90% confidence interval.
40

751
752 Figure 4: As in Figure 3, except for TCs that formed anywhere within the North Atlantic basin,
753 for (a) 1979-2007 and (b) 1979-2014.
754
755
756
41

757
758 Figure 5: Slope of the best-fit regression line between the 10 th percentile Atlantic TC formation
759 date and June monthly-mean (a) sea surface temperature (°C), (b) 600 hPa relative humidity (%),
42

760 (c) 500 hPa geopotential height (m), (d) 850 hPa relative vorticity (x10
-5
s
-1
), and (e) 850-200
761 hPa vertical wind shear magnitude (m s -1 ) over the domain 10°S-70°N, 150°-0°W. A one
762 standard deviation increase in a given monthly-mean field is associated with an n day change
763 (shaded; positive = earlier) in the 10 th
percentile Atlantic TC formation date. The June monthly-
764 mean field is contoured in each panel. Cross-hatching is utilized to identify regions where the
765 slope of the best-fit regression line is non-zero to at least 90% (light gray), 95% (dark gray), and
766 99% (black) confidence.
767
43

768
769 Figure 6: As in Figure 5, except globally.
44

770
771 Figure 7: Standard deviation of the June monthly-mean (a) sea surface temperature (°C), (b) 600
772 hPa relative humidity (%), (c) 500 hPa geopotential height (m), (d) 850 hPa relative vorticity
773 (x10 -5 s -1 ), and (e) 850-200 hPa vertical wind shear magnitude (m s -1 ).
45

774
775 Figure 8: Slope of the best-fit regression line between the 90 th
percentile Atlantic TC formation
776 date and November monthly-mean (a) sea surface temperature (°C), (b) 600 hPa relative
46

777 humidity (%), (c) 500 hPa geopotential height (m), (d) 850 hPa relative vorticity (x10
-5
s
-1
), and
778 (e) 850-200 hPa vertical wind shear magnitude (m s -1 ) over the domain 10°S-70°N, 150°-0°W. A
779 one standard deviation increase in a given monthly-mean field is associated with an n day change
780 (shaded; positive = later) in the 90 th
percentile Atlantic TC formation date. The November
781 monthly-mean field is contoured in each panel. Cross-hatching is utilized to identify regions
782 where the slope of the best-fit regression line is non-zero to at least 90% (light gray), 95% (dark
783 gray), and 99% (black) confidence.
784
47

785
786 Figure 9: As in Figure 8, except globally.
48

787
788 Figure 10: As in Figure 7, except for November monthly-mean fields.
49