Interannual and multidecadal extremes in Atlantic hurricane activity are shown to result from a coherent and interrelated set of atmospheric and oceanic conditions associated with three leading modes of climate variability in the Tropics. All three modes are related to fluctuations in tropical convection, with two representing the leading multidecadal modes of convective rainfall variability, and one representing the leading interannual mode (ENSO).
The tropical multidecadal modes are shown to link known fluctuations in Atlantic hurricane activity, West African monsoon rainfall, and Atlantic sea surface temperatures, to the Tropics-wide climate variability. These modes also capture an east–west seesaw in anomalous convection between the West African monsoon region and the Amazon basin, which helps to account for the interhemispheric symmetry of the 200-hPa streamfunction anomalies across the Atlantic Ocean and Africa, the 200-hPa divergent wind anomalies, and both the structure and spatial scale of the low-level tropical wind anomalies, associated with multidecadal extremes in Atlantic hurricane activity.
While there are many similarities between the 1950–69 and 1995–2004 periods of above-normal Atlantic hurricane activity, important differences in the tropical climate are also identified, which indicates that the above-normal activity since 1995 does not reflect an exact return to conditions seen during the 1950s–60s. In particular, the period 1950–69 shows a strong link to the leading tropical multidecadal mode (TMM), whereas the 1995–2002 period is associated with a sharp increase in amplitude of the second leading tropical multidecadal mode (TMM2). These differences include a very strong West African monsoon circulation and near-average sea surface temperatures across the central tropical Atlantic during 1950–69, compared with a modestly enhanced West African monsoon and exceptionally warm Atlantic sea surface temperatures during 1995–2004.
It is shown that the ENSO teleconnections and impacts on Atlantic hurricane activity can be substantially masked or accentuated by the leading multidecadal modes. This leads to the important result that these modes provide a substantially more complete view of the climate control over Atlantic hurricane activity during individual seasons than is afforded by ENSO alone. This result applies to understanding differences in the “apparent” ENSO teleconnections not only between the above- and below-normal hurricane decades, but also between the two sets of above-normal hurricane decades.
The recent 1995–2004 period is the most active Atlantic hurricane decade in the reliable record dating back to 1945, eclipsing even the high levels of activity seen during the 1950s–60s (Landsea 1993; Goldenberg et al. 2001). This has generated a renewed interest in developing a better understanding of the climate factors controlling both multidecadal and interannual fluctuations in Atlantic hurricane activity (Goldenberg et al. 2001; Bell et al. 2004).
During 1995–2004, seasons averaged 13 tropical storms (TSs), 7.6 hurricanes (Hs), and 3.6 major hurricanes (MHs) and every season was above normal (defined in section 2b; Table 1) except for the two El Niño years of 1997 and 2002. These numbers are far above those seen during the 1971–94 period of below-normal activity, when seasons averaged only 9 TSs, 5 Hs, and 1.5 MHs. During that 24-yr period, 13 seasons were below normal and only 3 were marginally above normal (1980, 1988, and 1989).
In the North Atlantic, almost 55% of all hurricanes and 80% of all major hurricanes develop from tropical storms first named in the main development region (MDR; Fig. 1). Most of these systems form during the August–October (ASO) climatological peak of the season. Since 1950, an average of seven–eight tropical storms have formed in the MDR during above-normal seasons, compared with an average of three–four during the near-normal seasons, and an average of only two during below-normal seasons. Named storms originating in the MDR account for nearly the entire difference in hurricanes and major hurricanes between above-normal and below-normal hurricane seasons and decades (Landsea and Gray 1992; Landsea 1993; Goldenberg and Shapiro 1996; Landsea et al. 1996, 1999).
Because many of the hurricanes that form in the MDR move westward toward the United States and Caribbean Islands, there is considerable interannual and multidecadal variability in the number of landfalling hurricanes for these regions. Both areas experience an average of two to three hurricane landfalls in above-normal seasons, compared to an average of one U.S. landfalling hurricane in below-normal seasons, and an average of one Caribbean landfalling hurricane every three below-normal seasons.
Interannual and multidecadal extremes in Atlantic hurricane activity result from a coherent and interrelated set of atmospheric and oceanic conditions in the near vicinity of the MDR. These conditions include anomalies in features such as the upper-level subtropical ridge and tropical easterly jet, the low-level tropical easterlies, the vertical wind shear, the 700-hPa African Easterly Jet (AEJ), sea level pressure, and sea surface temperatures (SSTs; Gray 1984; Goldenberg and Shapiro 1996; DeMaria 1996; Landsea et al. 1998; Bell et al. 1999, 2000, 2004, 2005).
This coherent variability suggests a strong relationship to the larger-scale climate signal. For example, ENSO causes large interannual fluctuations in Atlantic hurricanes and major hurricanes through its impacts on the upper-level circulation and vertical wind shear in the MDR (Gray 1984). Multidecadal fluctuations in Atlantic hurricane activity have been linked to regional climate phenomena such as the Atlantic multidecadal mode (Landsea et al. 1999; Goldenberg et al. 2001; Vitart and Anderson 2001) and West African monsoon variability (Gray 1990; Gray et al. 1992; Landsea and Gray 1992; Landsea et al. 1992; Goldenberg and Shapiro 1996). These works establish a strong body of evidence indicating that multidecadal fluctuations in Atlantic hurricane activity are not simply random collections of above-normal or below-normal seasons (e.g., Gray et al. 1996; Chelliah and Bell 1998, 1999; Bell and Chelliah 1999; Landsea et al. 1999).
The Atlantic multidecadal mode reflects SST fluctuations in the area south of Greenland and in the MDR (Delworth et al. 1997; Kushnir 1994; Hansen and Bezdek 1996; Enfield and Mestas-Nuñez 1999). It is thought to influence hurricane formation by modulating the local boundary layer and vertical wind shear in the MDR (Landsea et al. 1999; Vitart and Anderson 2001). Its warm phase is associated with above-normal hurricane decades, and its cold phase is associated with the below-normal period 1971–94.
However, neither these physical links nor the extent to which the Atlantic multidecadal mode actually influences Atlantic hurricane activity is well understood. For example, during the above-normal hurricane seasons the circulation anomalies in the MDR are generally much larger than can be accounted for by the SST anomalies, which typically average less than +0.5°C. Goldenberg and Shapiro (1996) suggest that Atlantic SSTs may only play the secondary role of influencing hurricane activity once the main atmospheric anomalies in the MDR are already established.
The circulation patterns associated with multidecadal extremes in Atlantic hurricane activity also exhibit a strong link to West African monsoon variability, which is the second dominant climate factor thought to influence Atlantic hurricane formation on multidecadal time scales (Hastenrath 1990; Landsea and Gray 1992; Goldenberg and Shapiro 1996). These studies indicate that increased Sahel rainfall contributed to reduced vertical wind shear and enhanced tropical cyclogenesis in the MDR during the 1950s–60s, while a prolonged drought in the African Sahel and Sudan regions contributed to the below-normal Atlantic hurricane activity during 1971–94 (Nicholson 1980; Thaiw et al. 1998; Ward, 1998).
While there is a strong contemporaneous correlation between the Atlantic multidecadal mode and West African monsoon variability, the cause of this relationship is not well understood. Some argue that low-frequency fluctuations in the Atlantic thermohaline circulation are ultimately the source of the variability (Gray et al. 1996). However, it is not clear how this variability actually causes fluctuations in the West African monsoon system, whose associated circulation is very large and extends well into the Southern Hemisphere.
Looking at even larger scales, Gray et al. (1996), Bell and Chelliah (1999), and Chelliah and Bell (1998, 1999) have related the low-frequency fluctuations in Atlantic hurricane activity to the Tropics-wide climate variability. These results were expanded by Chelliah and Bell (2004), who related fluctuations in both the West African monsoon and the Atlantic multidecadal mode during June–August to anomalous convection and surface temperatures in other regions such as the Amazon basin (Chen et al. 2001; Chu et al. 1994), the central equatorial Pacific (Morrissey and Graham 1996), and the Indian Ocean (Kawamura 1994; Hoerling et al. 2001). They showed that coherent fluctuations in all of these regions were captured by the leading Tropics-wide mode of convective rainfall variability [hereafter referred to as the tropical multidecadal mode (TMM)]. This finding is consistent with Gray et al. (1996), who noted that “multi-decadal fluctuations in intense Atlantic hurricane activity are but one manifestation of an extensive array of regional and global climate trends.” These studies necessitate a closer examination of the links between the larger-scale climate variability and regional climate fluctuations including Atlantic hurricane activity.
Other unresolved issues related to the multidecadal signal include 1) the extent to which this variability masks or accentuates ENSO teleconnections in the MDR during the peak months of the hurricane season, and 2) its contribution to the observed hurricane activity during individual seasons. In particular, the levels of activity during both El Niño and La Niña episodes vary considerably between above-normal and below-normal hurricane decades. During the below-normal period of 1970–94, every El Niño was associated with a well-below-normal hurricane season, and the La Niña episodes of 1984/85 and 1988/89 were associated with only near-normal and slightly above-normal seasons, respectively. Conversely, the only below-normal season during 1995–2004 resulted from the record 1997 El Niño (Bell and Halpert 1998), and the 1998–99 La Niña episode was associated with well-above-normal hurricane seasons in both years.
These issues lead to the fundamental question guiding this research: What are the underlying climate modes controlling both the regional MDR and larger-scale conditions associated with interannual and multidecadal extremes in Atlantic hurricane activity?
The paper is organized with seasonal activity and hurricane season classifications discussed in section 2. The three leading tropical modes of convective rainfall variability (which include the TMM and ENSO) in the National Centers for Environmental Research–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996) are identified in section 3. These three tropical modes are shown to represent the underlying climate modes associated with interannual and multidecadal extremes in Atlantic hurricane activity. In section 4 the relative contributions of these modes to Atlantic hurricane activity are discussed. Their associated circulation anomalies are then examined in section 5 and shown to capture many of the key regional and larger-scale features associated with Atlantic hurricane extremes. Conclusions and a discussion of results are presented in section 6.
2. NOAA's Accumulated Cyclone Energy (ACE) index and seasonal classifications
a. The ACE index
A useful measure of seasonal activity is the Accumulated Cyclone Energy (ACE) index (Bell et al. 2000, 2004), which accounts for the combined strength and duration of tropical storms and hurricanes during a given season. This wind energy index is calculated by summing the squares of the estimated 6-hourly maximum sustained surface wind speed in knots (Vmax2) for all periods while the system is either a tropical storm or hurricane. The ACE index is calculated from the HURDAT dataset (Jarvinen et al. 1984) and does not include periods when systems are classified as subtropical or extratropical.
The ACE index (Fig. 2a) is correlated at approximately 0.95 with other measures of seasonal activity such as the Hurricane Destruction Potential (HDP) index (Gray et al. 1992) and the Net Tropical Cyclone (NTC) index (Gray et al. 1998). It is correlated with seasonal hurricane totals at 0.87 and with major hurricane totals at 0.84. Since the ACE index includes the contribution from tropical storms, it is preferred over the HDP index, which does not. The ACE index is also preferred over the NTC index, which includes multiple samplings of some parameters.
Regional ACE indices for the MDR (Fig. 2b), the extratropics (Fig. 2c), and the Gulf of Mexico (Fig. 2d) are calculated based on the region in which the tropical storm is first named. If a storm is first named in one region and then moves into another, the storm total ACE value gets attributed to the region in which it was named.
By summing the regional and basinwide ACE indices over the period 1950–2004, it is found that the MDR-based index accounts for 71% of the basinwide total, with most of this contribution centered on the ASO season. The extratropics and Gulf of Mexico account for 25% and 5% of the basinwide total, respectively. This disproportionately large contribution from the MDR is consistent with the high percentage of hurricanes and major hurricanes that develops from disturbances forming in the deep Tropics (Landsea 1993; Shapiro and Goldenberg 1998).
The MDR-based ACE index is correlated with the basinwide index at 0.96, and with seasonal hurricane and major hurricane totals at 0.76 and 0.80, respectively. These parameters exhibit substantial multidecadal variability, with large values during 1950–69 and 1995–2004 and much lower values during 1971–94. The MDR accounts for almost 95% of the difference in the basinwide ACE index between the two sets of multidecadal extremes (132.8 × 104 kt2 compared to 64.5 × 104 kt2), and for nearly all of the difference in the number of hurricanes and major hurricanes (e.g., Goldenberg et al. 2001).
b. Seasonal classifications
The ACE index exhibits a highly skewed distribution (not shown), with peaks corresponding to seasons with low and moderate levels of activity, and a broad tail corresponding to very active seasons. Therefore, an approximate tercile distribution is not evenly centered about the mean (93.4 × 104 kt2) or median (87.5 × 104 kt2).
The National Oceanic and Atmospheric Administration (NOAA) uses the basinwide ACE index to classify Atlantic hurricane seasons, combined with seasonal departures from the average in each storm type (tropical storms, hurricanes, and major hurricanes; Bell et al. 2004). Above-normal seasons (Table 1) are defined by an ACE index above 103 × 104 kt2, combined with above-average numbers of at least two of the above three storm types. Near-normal seasons are defined by an ACE value in the range of 65–103 × 104 kt2, or by an ACE value slightly above this range but with average or below-average numbers in at least two of the above three storm types. Below-normal seasons have an ACE value below 65 × 104 kt2.
This classification yields an approximate tercile distribution, with 21 seasons being above normal, and the same number (17) of seasons being near normal and below normal. The MDR-based ACE index averaged over the above-normal seasons is 8.7 times higher than that averaged over the below-normal seasons. In contrast, the regional ACE indices for both the extratropics and Gulf of Mexico exhibit little systematic difference between the two season types.
3. Leading Tropics-wide modes during August–October
Interannual and multidecadal extremes in Atlantic hurricane activity are shown to result from a coherent and interrelated set of atmospheric and oceanic conditions associated with the three leading modes of climate variability in the Tropics (Figs. 3 and 4). Two of these are the leading tropical multidecadal modes, and the third is the leading tropical interannual mode (ENSO).
Following Chelliah and Bell (2004) the tropical multidecadal modes (interannual mode) represent the leading covariance-based, unrotated EOFs of 5-yr running-mean low-pass (LP)-filtered (high-pass-filtered) 200-hPa velocity potential anomalies in the analysis region 30°S–30°N. The EOF analysis is based on ASO seasonal 200-hPa velocity potential anomalies from the NCEP–NCAR reanalysis during the 56-yr period 1949–2004.
An identical approach was used by Chelliah and Bell (2004) to examine the first leading TMM and the ENSO mode for the June–August (JJA) and December–February seasons. Chelliah and Bell (2004) showed that using 200-hPa velocity potential anomalies to define these modes is very useful because it links them directly to the anomalous upper-level divergent circulation associated with fluctuations in tropical convective rainfall. These convective anomalies take on particular importance when interpreting the associated atmospheric anomalies at all levels, and when establishing consistency between these modes and the extensive published works identifying regional aspects of multidecadal climate variability. The use of velocity potential is also important because it implicitly includes the signal over land. Chelliah and Bell (2004) showed that the leading modes calculated in this manner are identical to the leading EOFs derived using surface temperature (land + ocean) anomalies.
In both cases, the leading multidecadal modes differ from those in which the EOFs are based on SST data alone, which capture only a subset of the total surface variance (Mestas-Nuñez and Enfield 1999). For instance, the leading SST-based EOFs do not directly incorporate land surface temperature anomalies associated with large multidecadal fluctuations in tropical convection in the West African monsoon region and the Amazon basin. This shortcoming limits the use of SST-based EOFs in the current study, because the atmospheric anomalies over large portions of the tropical and subtropical Atlantic are shown to be strongly related to fluctuation in the strength of the West African and Amazon basin convective regimes.
For the entire analysis region, 30°S–30°N, the two leading tropical multidecadal modes identified in this study account for almost 80% of the 5-yr LP-filtered variance in ASO velocity potential anomalies. The leading TMM accounts for 67% of the LP-filtered variance, and the second leading tropical multidecadal mode (TMM2) accounts for 13%. Because the TMM accounts for so much of the variance, a rotation of the EOFs to obtain TMM2 is not justified. The three leading tropical modes combined explain 77% of the unfiltered variance in ASO seasonal 200-hPa velocity potential anomalies. The TMM accounts for 43% of the unfiltered variance, ENSO accounts for 28%, and TMM2 accounts for 6%.
While the reanalysis eliminates climate discontinuities related to changes in the model and data assimilation systems, it does not completely resolve problems related to the uneven spatial and temporal distribution of the raw observational data over the 52-yr analysis period (Ebisuzaki et al. 1996; Kistler et al. 2001). Therefore, the following results focus on the regions where the loadings are deemed to be statistically significant (where their explained variance exceeds 15%), and where they can be substantiated by independent data sources and previous studies.
a. The leading TMM
The principal component (PC) time series (thick solid curve, Fig. 3) and loading pattern (Fig. 4a) of the TMM is similar for both the ASO and JJA seasons. Chelliah and Bell (2004) showed that this mode is largely distinct from ENSO and that it does not reflect a low-frequency imprint of ENSO onto the climate system. A similar finding for the leading multidecadal EOF of global SST anomalies is discussed by Mestas-Nuñez and Enfield (1999).
Chelliah and Bell (2004) also indicated that the spatial structure and total explained variance of the TMM and ENSO modes remained unaffected when the EOF calculations were performed on the unfiltered seasonal data, and when the unfiltered EOFs were subjected to Varimax rotation. The modes were also unchanged when defined using a 7-yr running-mean LP filter, when the EOF analysis was performed on smaller analysis domains (25°N–25°S and 20°N–20°S), and when the analysis period was reduced to 1961–90. In each case these two modes were well separated from each other and from the higher modes.
The PC time series of the TMM (Fig. 3) shows large positive values during the above-normal hurricane period of 1950–69 and negative values during the below-normal hurricane decades. Although the time series exhibits an upward trend since the mid-1990s, this trend alone cannot explain the sharp increase in hurricane activity since 1995.
As was also shown by Chelliah and Bell (2004), the TMM captures coherent fluctuations in 200-hPa velocity potential, divergence (Fig. 4a), and surface temperatures (Fig. 5a) in four core regions: the central Pacific (Morrissey and Graham 1996), the West African monsoon region, the Amazon basin (Chu et al. 1994; Kumar et al. 1999; Chen et al. 2001), and the Indian Ocean (Kawamura 1994; Hoerling et al. 2001). The positive phase of TMM coincides with an enhanced West African monsoon, suppressed convection over the Amazon basin, and the warm phase of the Atlantic multidecadal mode. The negative phase of TMM coincides with a suppressed West African monsoon, enhanced convection in the Amazon basin, and the cold phase of the Atlantic multidecadal mode (Thaiw et al. 1998; Goldenberg et al. 2001).
Overall, the TMM accounts for 15%–20% of the unfiltered ASO variance in sea surface temperature anomalies in the two core regions of the Atlantic multidecadal mode (Fig. 5a) and for 30%–40% of the 5-yr LP-filtered variance in these areas (Fig. 6a). From Mestas-Nuñez and Enfield (1999), the southern region is approximated by 9°–21.5°N, 20°–70°W, and the northern region is approximated by 45°–62.5°N, 20°–55°W (black boxes, Fig. 6).
The TMM also accounts for more than 80% of the LP-filtered surface temperature variance across northern Africa. Its positive phase indicates cooler temperatures in the West African monsoon region and warmer temperatures across the Sahara Desert, consistent with an enhanced West African monsoon during the 1950s–60s (Charney 1975).
b. The TMM2
The PC time series of TMM2 is particularly interesting because it captures the mid-1990s transition to the above-normal hurricane era (thin solid curve, Fig. 3). Conversely, this mode shows little association with the above-normal hurricane activity during the 1950s–60s. This result suggests important differences in the tropical climate between the two above-normal hurricane periods.
Because the PC time series shows TMM2 to exist mainly after 1975, its velocity potential loadings and explained variance are shown only for the period 1975–2002 (Fig. 4b). TMM2 is a more regional mode than TMM, with core loadings located over Africa and the Amazon basin where they generally explain 20%–40% of the unfiltered velocity potential variance during 1975–2002. The loadings are opposite in sign between the two regions, indicating an anomalous east–west seesaw in tropical convection and upper-level divergence.
The positive phase of TMM2 during 1995–2002 is associated with anomalous upper-level divergence over western Africa and compensating convergence over the Amazon basin and the subtropical South Atlantic. Its negative phase during 1975–94 coincides with a suppressed West African monsoon and anomalous upper-level divergence over the Amazon basin, which is consistent with the studies of Chu et al. (1994) and Chen et al. (2001). This inverse relationship is captured by both TMM and TMM2 and accounts for many of the large-scale and regional-scale circulation anomalies associated with Atlantic hurricane extremes (section 5).
In the central and western MDR, and both the central and high latitudes of the North Atlantic, TMM2 generally accounts for 20%–40% of the unfiltered SST variance during 1975–2002 (Fig. 5b) and for 40%–60% of the LP-filtered SST variance (Fig. 6b). For the entire 1951–2002 period, the combined TMM and TMM2 generally capture 40%–60% of the LP variance in the central MDR and Caribbean Sea, and in the northern core region of the Atlantic multidecadal mode (Fig. 6c).
The relationship of both multidecadal modes to North Atlantic SST fluctuations is further examined by regressing their PC time series onto the 5-yr running mean of area-averaged sea surface temperature anomalies in the core regions of the Atlantic multidecadal mode.
For the tropical Atlantic region, the observed time series (black curve, Fig. 7a) shows negligible anomalies prior to 1970, negative departures of 0.2°–0.4°C during the early 1970s, and a rise to record positive anomalies of 0.4°C beginning in 1995. The observed time series for the high latitudes of the North Atlantic shows positive anomalies during the 1950s, negative anomalies during 1971–94, and a sharp increase during the mid-1990s to record positive values by 1998–2002 (black curve, Fig. 7b). The regression lines using TMM and TMM2 as predictors (gray curves) capture the overall character of both time series and account for 40+ % of their variance. For the more recent period 1971–2002, this explained variance increases to 73% for the tropical Atlantic region (Fig. 7c), and to 56% for the high latitude region (Fig. 7d). This link between Atlantic sea surface temperature variability and the leading tropical multidecadal modes is particularly notable during the mid-1990s, when the sharp transition to a positive phase of TMM2 accounts for the increase in SSTs in both areas.
The analysis shows that much of the coherent variability in the West African monsoon system, Amazon basin convection, the Atlantic multidecadal mode, and Atlantic hurricane activity, is linked to the larger-scale tropical climate variability. This result indicates that TMM and TMM2 provide a Tropics-wide perspective on the regional climate patterns associated with multidecadal fluctuations in Atlantic hurricane activity, which is similar to that provided by ENSO for interannual fluctuations.
The analysis also reveals important differences in the tropical climate between the 1950–69 and 1995–2004 periods of enhanced hurricane activity, indicating that the above-normal hurricane seasons since 1995 do not reflect an exact return to conditions seen during the 1950s–60s. These differences include a markedly enhanced West African monsoon (Thaiw et al. 1998) and negligible sea surface temperature anomalies across the central tropical Atlantic during 1950–69, compared with a modestly enhanced West African monsoon and exceptionally warm Atlantic SSTs during 1995–2004 (see also Bell et al. 2005).
c. The leading tropical interannual mode (ENSO)
The PC time series of the ENSO mode (dashed curve in Fig. 3) is shown with the positive (negative) phase corresponding to El Niño (La Niña). This mode exhibits a two-celled pattern of velocity potential loadings and accounts for 70%+ of the unfiltered variance over the eastern Pacific and Australasia (shading, Fig. 4c). Its anomalous divergent circulation reflects fluctuations in the equatorial Walker circulation (Rasmusson and Carpenter 1982; Wright et al. 1988) and in the Hadley circulations over both the central equatorial Pacific and Indonesia. ENSO differs from both tropical multidecadal modes, which exhibit an anomalous east–west divergent circulation between the West African monsoon region and the Amazon basin, and an anomalous Hadley circulation maximized between the mean subtropical ridge axes over the Atlantic Ocean in both hemispheres.
4. Regression of leading tropical modes onto the basinwide ACE index
Regression analyses are used to quantify the relationship between the three leading tropical modes and the basinwide ACE index. For the entire 1951–2002 period, the two multidecadal modes and ENSO modes capture 30% of the unfiltered variance in the seasonal ACE index (black curve Fig. 8a), while the multidecadal modes capture 45% of the variance in the 5-yr running-mean ACE index (Fig. 8b). For the more recent period, 1971–2002, the explained variance for all three modes increases to 58% for the unfiltered ACE index (Fig. 8c), and to an incredible 82% for the 5-yr running-mean ACE index (Fig. 8d).
These results establish a strong relationship between seasonal Atlantic activity and the Tropics-wide climate variability on both interannual and multidecadal time scales. When combined with the analysis in section 3, they also show that the strong relationship between seasonal hurricane activity, West African monsoon variability, and the Atlantic multidecadal mode primarily reflects their common association with TMM and TMM2.
5. August–October conditions associated with seasonal hurricane extremes
The local factors influencing tropical cyclogenesis in the MDR have been described in numerous studies (e.g., Gray 1984; Goldenberg and Shapiro 1996; DeMaria 1996; Landsea et al. 1998; Chelliah and Bell 1998, 1999; Bell and Chelliah 1999; Bell et al. 1999, 2000, 2004). These factors are summarized using composite analyses calculated from the 14 most active and 14 most inactive Atlantic hurricane seasons during 1950–2004 (bold type, Table 1). Because these composites are biased heavily toward the above-normal and below-normal hurricane decades, they capture many of the mean circulation features during those decades.
a. Upper-level circulation
The ASO composite 200-hpa streamfunction anomalies for both the above-normal (Fig. 9a) and below-normal (Fig. 9b) seasons indicate a pronounced interhemispheric symmetry in the subtropics of both hemispheres from the Americas eastward to Australasia, with the anomalies aligned along the axes of the climatological mean subtropical ridges (Fig. 9c). This interhemispheric symmetry signifies a response of the upper-level circulation to anomalous tropical convection. The associated 200-hPa zonal wind anomalies extend from the eastern equatorial Pacific to central tropical Africa along the axis of the climatological mean tropical easterly jet (Fig. 9d).
The above-normal hurricane seasons feature anticyclonic streamfunction anomalies and easterly zonal wind anomalies in these regions, indicating above-average strengths of the mean subtropical ridges, and both a strengthening and westward extension of the tropical easterly jet. Conversely, the below-normal hurricane seasons feature anomalously weak subtropical ridges and a below-average strength of the tropical easterly jet.
For these key regions, the three leading tropical modes account for more than 70% of the unfiltered seasonal variance in 200-hPa streamfunction and zonal wind anomalies (Fig. 10). The multidecadal modes account for most of the signal over the central and eastern Atlantic Ocean and Africa (Figs. 10a–d), with TMM accounting for 40% to 60%+ of the variance in these regions. During the period 1975–2002, TMM2 accounts for 20% to 40% of the unfiltered streamfunction and zonal wind variance across the central and eastern tropical Atlantic (Figs. 10c,d). In contrast, the ENSO contribution dominates the explained variance over the Americas and the western Caribbean and is well separated from the multidecadal signals (Figs. 10e,f).
The regressed 200-hPa streamfunction anomalies associated with TMM and ENSO exhibit an approximate zonal wave-1 pattern in the subtropics of both hemispheres (Figs. 11a,c), whereas the main anomalies associated with TMM2 are centered over the subtropical Atlantic Ocean and Africa in both hemispheres. Consistent with the composite analyses, the dominant streamfunction and zonal wind anomalies for all three modes are found in the Tropics and subtropics. All three modes feature a pronounced interhemispheric symmetry to the streamfunction anomalies, along with an associated anomalous tropical easterly jet.
Across the central and eastern Atlantic Ocean and Africa, the positive phases of both multidecadal modes (Figs. 11a,b) are consistent with the above-normal hurricane season composites, and the negative phases of both modes (opposite anomalies to those shown) are consistent with the below-normal seasonal composites. These results link key aspects of the large-scale circulation associated with seasonal and multidecadal hurricane extremes to TMM and TMM2, and in particular to the east–west seesaw in anomalous convection between the West African monsoon region and the Amazon basin (see also Chelliah and Bell 2004).
Because of the strong relationship between TMM2 and the recent increase in Atlantic hurricane activity (section 4), we examined the extent to which this mode captures the change in the mean upper-level circulation anomalies across the Atlantic sector between 1995–2002 and the below-normal period of 1975–94. The observed differences in both the streamfunction and zonal wind anomalies between the two periods (Fig. 12a) are consistent with the seasonal composites. The above-normal period features a stronger subtropical ridge across the Atlantic Ocean in both hemispheres and an amplified tropical easterly jet from Africa to the Caribbean Sea. These differences are almost entirely captured by TMM2 (Fig. 12b).
The above-normal period also features 1) more negative velocity potential and stronger upper-level divergence over western Africa, 2) more positive velocity potential and upper-level convergence over the Amazon basin, and 3) a more east-to-west orientation of the divergent wind vector consistent with an enhanced overturning between western Africa and tropical South America (Fig. 12c). These differences are also captured by TMM2 (Fig. 12d).
Focusing now on the interannual variability, the ENSO teleconnections during individual hurricane seasons (Fig. 11c) are consistent with classical interpretations of the atmospheric response to anomalous equatorial Pacific convection (Mo and Kousky 1993). These teleconnections are out of phase with both multidecadal modes over the Atlantic Ocean and Africa.
For individual hurricane seasons, the ENSO signal can be significantly masked or accentuated by the multidecadal signal (Fig. 13). In fact, the multidecadal signal is so large that it offers a substantially more complete view of the climate control over Atlantic hurricane activity, even during individual seasons, than is afforded by ENSO alone. This finding has implications for understanding differences in the apparent ENSO teleconnections between the above- and below-normal hurricane decades, and between the two above-normal hurricane periods (1950–69 and 1995–2004). This result is also consistent with the relatively low correlations between Atlantic hurricane activity and ENSO (Bove et al. 1998).
For example, the regressed 200-hPa anomalies have largest amplitude in the MDR when the time series of the dominant tropical multidecadal mode and ENSO are out of phase. The positive phase of TMM (TMM2) combined with a moderate-strength La Niña captures the large-scale and regional-scale conditions associated with the above-normal hurricane seasons during the 1950s–60s (1995–2004) (Figs. 13a,c). Combining the negative phases of both tropical multidecadal modes, as seen during 1975–94, with a moderate-strength El Niño captures conditions associated with below-normal hurricane seasons and decades (Fig. 13f).
When the time series of the multidecadal and ENSO modes have the same sign, the regressed anomalies are notably weaker and do not support seasonal extremes in activity (Figs. 13b,d,e). These results are consistent with the observation that a moderate El Niño is more likely to be associated with well-below-normal activity during a below-normal decade, while a moderate La Niña is more likely to be associated with well-above-normal activity during an above-normal decade.
b. Low-level circulation
The seasonal hurricane composites show that the 200- (shading, Figs. 14a and 15a) and 850-hPa (Figs. 14b and 15b) zonal wind anomalies have opposite sign across the tropical North Atlantic and North Pacific. This vertical wind structure is consistent with the baroclinic response of the tropical atmosphere to anomalous convection. At 200 hPa, the above-normal seasons feature a near absence of zonal winds over the central and eastern MDR, while the below-normal seasons feature mean westerlies averaging 3–5 m s−1 (contours).
For the above-normal seasons, this wind pattern produces anomalous easterly vertical wind shear (U200 hPa minus U850 hPa) from western Africa to the eastern tropical Pacific (light shading, Fig. 14b). The resulting magnitude of the total vertical shear (contours) across the central and western MDR is well below the |8 m s−1| threshold for tropical cyclone formation (Gadgil et al. 1984; Shapiro and Goldenberg 1998; Goldenberg and Shapiro 1996).
During ASO, tropical cyclogenesis in the MDR is generally associated with amplifying African easterly wave disturbances moving within the region of high cyclonic vorticity along the equatorward flank of the 700-hPa AEJ (Burpee 1972; Reed et al. 1977). During above-normal seasons, the reduced low-level easterlies are concentrated south of the AEJ core, thus contributing to a well-defined AEJ with enhanced cyclonic vorticity extending well into the MDR (Fig. 14d; also see Bell and Chelliah 1999). Developing disturbances remain in this extended region of increased cyclonic vorticity while moving westward over anomalously warm SSTs (Fig. 14f) into the low-shear environment of the central and western MDR. Combined with anomalously low sea level pressure over the tropical Atlantic (Fig. 14e; see also Knaff 1997), these conditions are extremely conducive to tropical cyclone formation in the MDR, as was observed during the above-normal 1998–99 (Bell et al. 1999, 2000) and 2003–04 hurricane seasons (Bell et al. 2004, 2005).
During below-normal hurricane seasons, anomalous westerly vertical shear throughout the MDR results in total shear values that are often too high for tropical storm formation (Fig. 15c). In addition, the AEJ features a more uniform distribution of easterly winds in response to enhanced easterly trades along its equatorward flank (Fig. 15b), which also act to shift the axis of cyclonic shear equatorward to near 10°N. As a result, anticyclonic relative vorticity now overspreads most of the high-shear environment of the MDR (Fig. 15d), resulting in exceptionally unfavorable conditions for tropical cyclone formation.
The below-normal seasons also feature an extensive area of anomalously high sea level pressures across the MDR (Fig. 15e), and anomalously cool sea surface temperatures at high latitudes of the North Atlantic (Fig. 15f). These anomalies are opposite to those seen in the above-normal seasons and are consistent with the cold phase of the Atlantic multidecadal mode. In both the above-normal and below-normal seasonal composites, the main sea level pressure anomalies are found across the tropical and subtropical North Atlantic, well south of the core loading regions of the North Atlantic Oscillation (NAO), which are centered over Greenland and the Azores Islands.
TMM accounts for generally 50%–70% of the unfiltered ASO variance in these key composite features (Fig. 16), while during 1975–2002 TMM2 accounts for generally 20%–40% of their variance (Fig. 17). Consistent with this result, the circulation anomalies for the positive phases of both modes are consistent with the above-normal hurricane season composites (cf. contours in Figs. 16 and 17 with Figs. 14b–e). The anomalies for the negative phases of both modes (opposite to those shown) are consistent with the below-normal hurricane season composites.
Both TMM and TMM2 are associated with anomalous easterly trade winds at 850-hPa from the eastern North Pacific to Africa (Figs. 16a and 17a). These anomalies have opposite sign to their 200-hPa counterparts, which is consistent with their relationship to anomalous tropical convection over both western Africa and the Amazon basin. The main vertical wind shear anomalies associated with both modes are located over the central and eastern tropical Atlantic, including a large portion of the MDR (Figs. 16b and 17b). A +2.0 standard deviation of TMM2 seen during the late 1990s is associated with a 3.0 m s−1 decrease in vertical wind shear in the central MDR, while the −1.0 standard deviation of TMM seen during the 1980s is associated with a 4.0 m s−1 increase in vertical shear in that region. This result is consistent with the large multide-cadal variability in vertical wind shear seen across the central and southern MDR by Shapiro and Goldenberg (1998) and Goldenberg et al. (2001).
Consistent with the pattern of anomalous easterly winds at 850 hPa, both multidecadal modes are also associated with relative vorticity anomalies along the equatorward flank of the AEJ (Figs. 16c and 17c) similar to that seen in the composite analyses. Both modes are also associated with a consistent pattern of sea level pressure anomalies across the heart of the MDR and over the western North Atlantic (Figs. 16d and 17d). The strong relationship between the TMM and sea level pressure anomalies over western Africa is consistent with previously described surface temperature departures associated with fluctuations in the strength of the West African monsoon.
Because ENSO is the dominant contributor to the upper-level variance in 200-hPa zonal winds across the extreme western MDR, it also dominates the vertical shear variance in this region (Fig. 18). However, over the eastern Caribbean Sea and western tropical Atlantic, which is sometimes a critical subregion within the MDR during individual seasons, the amount of explained vertical shear variance for all three tropical modes is comparably low.
Consistent with previous results, the large and spatially extensive vertical wind shear anomalies seen in the above-normal hurricane composite (Fig. 14c) are recovered when the time series of the dominant multidecadal mode and ENSO are out of phase (Figs. 19a,c,f). Examples of this situation are seen when a moderate strength La Niña occurs during the above-normal hurricane decades (Figs. 19a,c) or when a moderate strength El Niño occurs during the below-normal hurricane decades (Fig. 19f). In contrast, the vertical shear anomalies are notably weaker and tend to change sign within the MDR, when the time series of the dominant multidecadal mode and ENSO are in phase (Figs. 19b,d,e). These results reinforce the finding that the multidecadal signal significantly masks or accentuates the ENSO teleconnections across the MDR and western Africa. They also highlight the fundamental point that the observed circulation anomalies in the MDR during a given season often reflect the combined influences of both the multidecadal signal and ENSO.
6. Conclusions and discussion
Seasonal and multidecadal extremes in Atlantic hurricane activity result from changes in the number and intensity of hurricanes that originate as tropical storms in the main development region (MDR) during August– October. These extremes are not random but instead are found to result from a coherent and interrelated set of atmospheric and oceanic conditions associated with the three leading modes of tropical convective rainfall variability. Two of these are the leading multidecadal modes (referred to as TMM and TMM2), and the third is the leading tropical interannual mode (ENSO). A concurrent examination of these modes on both the interannual and multidecadal time scales leads to a more comprehensive and larger-scale view of the climate fluctuations influencing Atlantic hurricane activity than has appeared previously in the literature.
Because the circulation in the Tropics and subtropics is strongly influenced by the distribution of convection, it is not surprising that all three tropical modes are linked to fluctuations in tropical convection. Characteristic convective signatures of these modes include 1) anticyclonic (cyclonic) circulation anomalies in the subtropics of both hemispheres flanking the regions of enhanced (suppressed) tropical convection, and 2) a reversal in sign of the zonal wind anomalies between 200 and 850 hPa from the eastern Pacific to Africa, consistent with the known baroclinic response of the tropical atmosphere to anomalous convection. These anomalies can be interpreted as teleconnections occurring on both interannual and multidecadal time scales between the Pacific and Atlantic basins, and between the Northern and Southern Hemisphere subtropics.
The three tropical modes account for much of the observed variance in seasonal Atlantic hurricane activity, as measured by the basinwide Accumulated Cyclone Energy (ACE) index. For example, during 1971–2002 they account for 58% of the unfiltered seasonal ACE variance, while TMM and TMM2 account for 82% of the variance in the 5-yr running-mean ACE index. These large amounts of explained variance are a consequence of the modes capturing much of the August–October variance in all key regional and large-scale circulation features associated with seasonal and multidecadal extremes in Atlantic hurricane activity.
Over the tropical Atlantic, these conditions include anomalies in features such as the upper-level subtropical ridges and tropical easterly jet, the low-level tropical easterlies, the vertical wind shear, the 700-hPa African Easterly Jet (AEJ), sea level pressure, and sea surface temperatures. Because of their association with circulation anomalies in the MDR, the multidecadal modes are found to provide a global-scale perspective on the climate factors associated with low-frequency fluctuations in Atlantic hurricane activity. This perspective is analogous to that provided by ENSO for interannual fluctuations in activity (Gray 1984).
The TMM and TMM2 link coherent low-frequency fluctuations in Atlantic hurricane activity, the West African monsoon, and tropical Atlantic SSTs to the Tropics-wide climate variability. It is also shown that the known link between Atlantic hurricane activity and SST anomalies in the core regions of the Atlantic multidecadal mode primarily reflects their common association to the leading tropical multidecadal modes.
Both TMM and TMM2 also capture an east–west seesaw in anomalous convection between the West African monsoon region and the Amazon basin. This relationship helps to account for the interhemispheric symmetry of the 200-hPa streamfunction and divergence wind anomalies over the Atlantic basin and Africa, as well as the spatial scale of the low-level tropical wind anomalies, associated with Atlantic hurricane extremes.
The analysis also indicates differences between the above-normal hurricane decades of the 1950s–60s and 1995–2004. The period 1950–69 shows a strong link to TMM, whereas the transition in the tropical climate from the below-normal 1975–94 era to the above-normal era seen since 1995 is mainly associated with TMM2. These differences between the two periods include a very strong West African monsoon and near-average sea surface temperatures in the MDR during 1950–69, compared with a modestly enhanced West African monsoon and exceptionally warm Atlantic SSTs during 1995–2004.
For individual hurricane seasons, ENSO teleconnections and ENSO impacts on Atlantic hurricane activity can be significantly masked or accentuated by the multidecadal signal. This result has implications for understanding apparent differences in the ENSO teleconnections between the above- and below-normal hurricane decades, and between the two above-normal hurricane periods (1950–69 and 1995–2004). It also highlights the important finding that the tropical multidecadal modes provide a substantially more complete view of the climate factors associated with Atlantic hurricane extremes than can be gleaned from ENSO alone.
In particular, the large-scale and regional-scale atmospheric anomalies and levels of activity associated with seasonal hurricane extremes are recovered when the dominant tropical multidecadal mode and ENSO are out of phase. It is shown that a moderate El Niño is more likely to be associated with well-below-normal activity during a below-normal decade, while a moderate La Niña is more likely to be associated with well-above-normal activity during an above-normal decade. For instance, it was shown that during above-normal hurricane decades the dominant tropical multidecadal mode captures a well-defined African easterly jet with increased cyclonic vorticity along its equatorward flank. When La Niña is also present, this jet structure means that African easterly disturbances can remain in the extended region of enhanced cyclonic vorticity while moving westward over anomalous warm SSTs into the low-shear environment of the central and western MDR.
The results suggest that useful dynamically based seasonal hurricane outlooks will require accurate simulations not only of ENSO and its associated teleconnections during the August–October months of historically low ENSO predictive skill, but also of the underlying climate conditions captured by the leading tropical multidecadal modes.
We thank Vernon Kousky, Kingtse Mo, Huug Van den Dool, Chris Landsea, Stanley Goldenberg, Richard Pasch, Eric Blake, and Lixion Avila for their helpful comments and insight during the course of this work. We also thank the anonymous reviewers for helping to clarify key portions of the manuscript.
Corresponding author address: Dr. Gerald Bell, Analysis Branch, Climate Prediction Center, Room 605, 5200 Auth Road, Camp Springs, MD 20746. Email: email@example.com