6.2. Long-Term Trends of Wind speed and Air-Sea CO2 Fluxes
 During the 1984 to 2005 period, the mean annual and seasonal wind speed has increased considerably. For example, synoptic meteorological data (i.e., BWS) and COADS data in the vicinity of the island of Bermuda exhibited an increase of wind speed by 1.00 and 1.10 m s−1, respectively (Tables 4a and 4b). The data assimilation models of ECMWF and NNR) also showed an increase in wind speed of 1.10 and 1.23 m s−1, respectively, over the last 22 years for the 2.5° box overlying Bermuda (Tables 4a and 4b). These changes represented a ∼10–17% change in wind speed over the period of this study.
 The long-term increase in wind speed was, however, not uniformly distributed seasonally. The smallest changes in wind speed occurred during the summertime (i.e., during air-sea CO2 efflux). The BWS, ECMWF and NNR data sets show modest increases in wind speed of 0.08, 0.60 and 0.46 m s−1, respectively, over the last 22 years (Table 6a and 6b; p values were not significant though). In contrast, the largest seasonal increases occurred during the fall and wintertime periods. For example, the BWS, ECMWF and NNR data sets exhibited statistically significant increases in wind speed of 0.93, 1.51 and 0.69 m s−1, respectively, during wintertime (Tables 5a and 5b), but also during the fall (Table 7a and 7b). This represents a 13–21% change in wintertime wind speed over the period of this study.
 The increase in wind speed observed in Bermuda has also observed elsewhere in the marginal seas of the North Atlantic [e.g., Pirazzoli and Tomasin, 2003; Pirazzoli, 2005]. Elsewhere, for example, the frequency of winter storms and wave heights occurring in the subpolar region of the North Atlantic, enhanced during the predominantly NAO positive period of 1980 to 1995 [Beersma et al., 1997; WASA group, 1998; Bijl et al., 1999; Alexandersson et al., 2000; Alexander et al., 2005] appears to be in decline since the mid-1990s [Weisse et al., 2005]. Large spatiotemporal changes in wind speed are evident between 1984 and 2005 are evident from both NNR and ECMWF data (Figure 7). For example, increases of +0.5−>1.0 m s−1 in wind speed have been observed near Bermuda in the western North Atlantic Ocean, but also in the marginal seas of Europe (e.g., North Sea, Mediterranean Sea, Baltic Sea, Bay of Biscaye), Greenland and Labrador Seas. In other regions, decreases in wind speed of up to 1 m s−1 have been observed in the tropical North Atlantic, off Nova Scotia and Newfoundland, and in the NE Atlantic Ocean. Such changes not only have potential influence for gas exchange rates, but, for heat and freshwater fluxes, mixing and stratification, circulation, biological production and ocean ecosystem structure, for example.
Figure 7. Spatiotemporal changes in wind speed (m s−1) over the North Atlantic Ocean over the 1984–2005 period. Although the plotted wind speed data are NNR data, ECMWF data (not shown) exhibit similar spatiotemporal patterns.
Download figure to PowerPoint
 Rates of air-sea CO2 flux (as with wind speed) also changed over the 1983 to 2005 period. Seasonally, the summertime oceanic CO2 source to the atmosphere in the subtropical gyre has substantially increased (Table 6a and 6b). For example, summertime air-sea CO2 effluxes computed with the BWS, ECMWF and NNR data sets exhibited an increase of ∼+11−14 mmol CO2 m−2 per year or +251−302 mmol CO2 m−2, respectively, over the 1983−2005 period (Table 6a and 6b). The trends in summertime air-sea CO2 fluxes were statistically significant for fluxes computed from BWS, ECMWF and NNR data sets (Table 6a and 6b; p values 0.02 to <0.01). The change in the summertime CO2 source status was however compensated for by a larger increase in the fall and wintertime air-sea CO2 influx. Wintertime air-sea CO2 influxes computed using BWS, ECMWF and NNR data sets increased by ∼−14−19 mmol CO2 m−2 per year or by −297 to −414 mmol CO2 m−2 period−1, respectively, from 1984 to 2005 (Tables 5a and 5b; p values <0.01). The trends in wintertime air-sea CO2 influxes were statistically significant for fluxes computed from BWS and NNR data sets (Tables 5a and 5b; p values <0.01), but not for ECMWF data sets (note: from 1984 to 2001 only). The increase in seasonal rates of air-sea CO2 flux appears primarily associated with long-term changes in wind speed rather than changes in ΔpCO2 conditions. As shown earlier, both surface seawater and atmospheric pCO2 have increased at similar rates over the period of this study (Table 1).
 Although both fall/wintertime and summertime air-sea CO2 fluxes have increased, annual rates of air-sea CO2 influx exhibited relatively small increases of ∼−2–6 mmol CO2 m−2 per year, or 39–166 mmol CO2 m−2, respectively, over the 1984 to 2005 period (Tables 4a and 4b; computed from BWS, ECMWF and NNR data sets). If scaled to the entire subtropical gyre, these changes constitute a longer-term increase of ∼0.01–0.02 Pg C in the air-sea CO2 flux or oceanic CO2 sink in the midlatitude North Atlantic Ocean from 1984 to 2005. However, the long-term trends in annual air-sea CO2 fluxes were statistically insignificant for fluxes when computed from all wind speed data sets (Tables 4a and 4b; p values >0.05).
6.3. Potential Causes for Long-Term Trends of Wind Speed and Air-Sea CO2 Fluxes?
 What are the underlying factors that influence the large year-to-year variability and long-term trends of air-sea CO2 fluxes? The fall (i.e., OND) period is characterized by rapidly cooling and deepening mixed layers [Bates et al., 1996a], a transition to undersaturated seawater CO2 conditions and a period of net air-sea CO2 influx. The wintertime (i.e., JFMAM) period is typically characterized by higher wind speeds, deeper mixed layers (i.e., 100−>150 m [e.g., Steinberg et al., 2001]), seawater CO2 conditions undersaturated with respect to the atmosphere (i.e., negative ΔpCO2), and net air-sea CO2 influx. The following summertime June to September (i.e., JJAS) period is characterized by lower wind speeds, shallow mixed layers (i.e., ∼10–>50 m [e.g., Steinberg et al., 2001]), seawater CO2 conditions oversaturated with respect to the atmosphere, and net air-sea CO2 fluxes from the ocean. Since air-sea CO2 fluxes are directly related to ΔpCO2 and wind speed (equation (2)), air-sea CO2 fluxes each month were strongly correlated with both parameters. In addition, air-sea CO2 fluxes each month were correlated with surface temperature each season, and, with the exception of fall, weakly correlated with variability of other parameters such as salinity, DIC or total alkalinity.
 Previous studies have shown that interannual variability of hydrographic, and biogeochemical properties in the North Atlantic subtropical gyre are coupled to modes of low-frequency climate variability such as NAO, AO and ENSO [e.g., Oschlies, 2001; Bates, 2001; Gruber et al., 2002; Bates and Hansell, 2004]. The NAO/AO exerts influence throughout the year [Marshall et al., 2001], but is the dominant leading mode in the wintertime JFMAM period. Here low-frequency modes of climate variability can be viewed as the expression of physical changes in the North Atlantic Ocean that can in turn potentially influence air-sea CO2 fluxes by altering atmospheric forcing (e.g., wind speed distributions) and oceanic properties (e.g., surface temperature and salinity distributions, water-column stratification, and primary production) that in turn influences ΔpCO2 conditions.
 In the North Atlantic Ocean, the NAO is the dominant mode of low-frequency climate variability [Marshall et al., 2001]. The NAO can be viewed as the regional expression of the AO [e.g., Thompson and Wallace, 1998; Wallace, 2000] or Northern Annular Mode (NAM) [Thompson and Lorenz, 2004; Quadrelli and Wallace, 2004], with the AO being the leading wintertime Northern Hemispheric low-frequency mode of sea level pressure (SLP) variability. In the North Atlantic, the NAO manifests as a dipole SLP oscillation between the Icelandic low-pressure and Azores high atmospheric pressure centers [e.g., Rogers, 1990; Hurrell, 1995; Jones et al., 1997; Osborn et al., 1999; Hurrell and Van Loon, 1997; Hurrell et al., 2001, 2002], with variability of parameters influenced by the NAO exhibiting a tripole pattern in the North Atlantic Ocean [Marshall et al., 2001]. Maps of the AO and NAO modes of variability are nearly indistinguishable [Marshall et al., 2001].
 The NAO/AO has a pronounced effect on the Northern Hemisphere [e.g., Visbeck et al., 2001; Hurrell et al., 2002], exhibiting considerable seasonal and interannual variability. In the subpolar region, positive (negative) states of the NAO index tend to result in enhanced (reduced) precipitation, storminess and wave heights [e.g., Beersma et al., 1997; WASA group, 1998; Bijl et al., 1999; Alexandersson et al., 2000; Alexander et al., 2005]. During a NAO positive state, the mean position of Gulf Stream tends to shift northward [Taylor et al., 1998; Taylor and Stephens, 1998; Weisse et al., 2005], and baroclinic mass transport of the Gulf Stream increases [Curry and McCartney, 2001; Molinari, 2004].
 In the midlatitudes of the North Atlantic Ocean, positive and (negative) states of the NAO index result in reduced (enhanced) midlatitude westerlies, reduced (enhanced) wind stress and heat exchange [Bjerknes, 1964; Cayan, 1992a, 1992b], enhanced (reduced) sea surface temperature [e.g., Davies et al., 1997; Kapala et al., 1998; Rodwell et al., 1999] recorded in ocean temperatures [Bates, 2001] and corals from Bermuda [e.g., Kuhnert et al., 2005]. Interannual anomalies of mixed layer depths and temperature, integrated primary and new production, phytoplankton community structure, and seawater CO2 properties in the North Atlantic Ocean are correlated to the NAO [Oschlies, 2001; Bates, 2001; Bates and Hansell, 2004; Lomas and Bates, 2004].
 The NAO/AO may also be related to the Tropical Atlantic Variability (TAV; a covarying fluctuation of SST and trade winds straddling the Intertropical Convergence Zone, ITCZ), with both the TAV and NAO/AO exerting influence on air-sea interaction in the tropics of the North Atlantic [Marshall et al., 2001]. The Pacific Ocean ENSO, however, exerts its influence on the tropical Caribbean Sea and western Atlantic Ocean with a typical lag time of 6–9 months [e.g., Zhang et al., 1996; Bojariu, 1997; Penland and Matrosova, 1998]. For example, sea surface salinity and temperature anomalies observed at BATS and Hydrostation S in the 1990s have been shown to be correlated to ENSO [Bates, 2001]; a pattern also observed in the Caribbean Sea and tropical Atlantic Ocean [e.g., Zhang et al., 1996; Bojariu, 1997; Penland and Matrosova, 1998].
 Do air-sea CO2 fluxes in the subtropical gyre of the western North Atlantic Ocean correlate with the low-frequency modes? On monthly timescales, monthly averaged air-sea CO2 fluxes, ΔpCO2, and wind speed were not significantly correlated with monthly values of climate indices such as NAO, AO and ENSO (i.e., Southern Oscillation Index, SOI, with and without a 6-month lag). This is perhaps not surprising as the low-frequency modes in the Northern Hemisphere exhibit considerable month-to-month variability. In addition, the Northern Hemisphere modes including the NAO and other patterns such as the East Atlantic (EA) [e.g., Wallace and Gutzler, 1981; Quadrelli et al., 2001; Ferreira and Frankignoul, 2005] and Pacific/North American (PNA) [e.g., Barnston and Livezey, 1987; Wallace et al., 1990; Honda and Nakamura, 2001; Wallace and Thompson, 2002] typically explains ∼50−70% (and often much less) of the SLP variance (http://www.cpc.ncep.noaa.gov). On seasonal timescales, air-sea CO2 fluxes were correlated with NAO, but not generally with the AO and SOI (or other Northern Hemisphere modes).
 During the wintertime period (i.e., JFMAM), the correlations between air-sea CO2 influxes, and NAO, AO and SIO were generally poor (r2 < 0.2; p values >0.05) for fluxes calculated over the 1984–2005 period from the different types of wind speed data (i.e., BWS, NNR, and ECMWF) and using quadratic and cubic wind speed–flux relationships (Table 9). Weak correlations (r2 values of ∼0.17 to 0.40; p values of 0.44 to 0.06) existed between air-sea CO2 fluxes calculated from NNR wind speed data, and NAO, AO and SOI (with a 6-month lag). However, some generalizations can be made about wintertime air-sea CO2 fluxes. The highest air-sea CO2 influxes (calculated with NNR wind speed data and cubic wind speed–flux relationships) tended to occur during years with strongly negative wintertime AO values (r2 value of ∼0.40; p value of 0.06; Figure 8). Furthermore, higher air-sea CO2 fluxes (calculated with NNR wind speed data and cubic wind speed–flux relationships) also tended to occur during years with negative wintertime SOI values (r2 value of ∼0.35; p value of 0.10). In winters with strongly negative NAO and AO values (i.e., 1985, 1996, 2001, 2005), air-sea CO2 fluxes were 9–43% higher in El Niño years (i.e., winters with negative SIO with 6-month lag). In winters with negative NAO and positive AO values (i.e., 1984, 1986–1988, 1998, 1999, 2004), air-sea CO2 fluxes were also higher by 15–28% in El Niño years compared to La Niña years. Last, in winters with strongly positive NAO and AO values (i.e., 1989–1995, 1997, 2000, 2002, 2003), air-sea CO2 fluxes were higher by 11–22% in El Niño years.
Figure 8. Seasonal relationships between air-sea CO2 fluxes and low-frequency climate modes, and wind speed and ΔpCO2. Regression statistics are detailed in Table 9. (a) wintertime (JFMAM) cubic air-sea CO2 fluxes (NNR data, solid diamonds; BWS data, grey circle) and mean JFMAM Arctic Oscillation (AO) index. (b) Mean wintertime (JFMAM) ΔpCO2 and mean wintertime (JFMAM) wind speed (m s−1) (NNR data, solid diamonds; BWS data, grey circle). (c) Summertime (JJAS) cubic air-sea CO2 fluxes (NNR data, solid diamonds; BWS data, grey circle) and mean JJAS North Atlantic Oscillation (NAO) index. (d) Fall cubic (OND) air-sea CO2 fluxes (NNR data, solid diamonds; BWS data, grey circle) and mean OND North Atlantic Oscillation (NAO) index.
Download figure to PowerPoint
Table 9. Correlations of r2 and p-Values Between Seasonal Mean Air-Sea CO2 Fluxes and Mean Seasonal Low-Frequency Climate Modesa
| ||NAO||AO||SOI||SOI With 6 Month Lag|
|Quadratic Wind Speed–k Relationship|
|JFMAM air-sea CO2 fluxes|
|JJAS air-sea CO2 fluxes|
|OND air-sea CO2 fluxes|
|Cubic Wind Speed–k Relationship|
|JFMAM air-sea CO2 fluxes|
|JJAS air-sea CO2 fluxes|
|OND air-sea CO2 fluxes|
 Wintertime air-sea CO2 influxes were higher in the 2000s (i.e., 2000–2005) compared to the 1980s (i.e., 1984–1989) and 1990s (Table 10). During the 2000s, there were proportionately more winters with higher air-sea CO2 influx cooccurring with winters of strongly negative AO and/or negative SIO values (i.e., El Niño years). Thus the increase in wintertime air-sea CO2 influx described earlier (in section 6.2) presumably reflects this interannual variability.
Table 10. Mean Seasonal Air-Sea CO2 Fluxes and Mean Seasonal Low-Frequency Climate Modes for the 1980s, 1990s and 2000sa
| ||Mean Air-Sea CO2 Flux||Mean Climate Index|
|JFMAM Air-Sea CO2Fluxes|
|JJAS Air-Sea CO2Fluxes|
|OND Air-Sea CO2Fluxes|
 Previous studies [Gruber et al., 2002; McKinley et al., 2004] have suggested that wintertime air-sea CO2 influxes in the subtropical gyre of the North Atlantic are coupled to the wintertime state of the NAO. However, the lack of good correlations between wintertime air-sea CO2 influxes and low-frequency climate modes observed at the BATS site (1984–2005) suggests that climate patterns such as the NAO only partially explain the variance observed in SLP (and other factors) that in turn can influence gas exchange. In addition, wintertime wind speed and ΔpCO2 tend to be anticorrelated (Figure 8b) potentially suppressing large interannual changes in wintertime air-sea CO2 fluxes due to atmospheric forcing associated with low-frequency modes or other factors. For example, the enhancement of westerlies in the midlatitudes of the North Atlantic Ocean during a negative NAO (or AO) state [Bjerknes, 1964; Cayan, 1992a, 1992b; Davies et al., 1997; Kapala et al., 1998; Rodwell et al., 1999] should presumably increase wind speed and enhance air-sea CO2 flux. During a negative NAO (or AO) state, sea surface temperatures cool, the mixed layer deepens and integrated primary and new production increases due to enhanced vertical supply of nutrients [e.g., Oschlies, 2001; Bates, 2001; Lomas and Bates, 2004]. However, deeper vertical mixing tends to bring excess DIC relative to Redfield stoichiometric ratios [e.g., Redfield et al., 1963] of DIC and nitrate resulting in positive surface DIC anomalies [Bates, 2001] that in turn reduce ΔpCO2 values (and reduce air-sea CO2 flux). Thus interannual changes in atmospheric forcing (as reflected by changes in the NAO, AO or SIO indices) may not result in large year-to-year variability in wintertime air-sea CO2 fluxes.
 In contrast to the wintertime, summertime (JJAS) and fall (OND) air-sea CO2 effluxes were correlated with NAO, but not significantly correlated with AO or SIO (with or without 6-month lag) (Table 9). In summertime, air-sea CO2 effluxes were correlated with summertime NAO values (r2 values of ∼0.30–0.41; p values of 0.16–0.06). The highest air-sea CO2 effluxes tended to occur during years with negative summertime NAO values (Figure 8c). In the fall, air-sea CO2 influxes were correlated with fall NAO values (r2 values of ∼0.48–0.56; p values of 0.02–<0.01). Higher air-sea CO2 fluxes tended to occur during years with negative NAO values for the fall. In both seasons, irrespective of the direction of gas exchange (and ΔpCO2 values), the highest air-sea CO2 fluxes cooccurred with periods of negative NAO values (Figure 8d). Although NAO cannot directly influence gas exchange, NAO variability appears to characterize changes in atmospheric forcing that influence air-sea CO2 fluxes. Previous studies have also shown that summertime SST and SLP anomalies in the midlatitudes of the North Atlantic Ocean exert a preconditioning influence on the following October to December fall period [e.g., Czaja et al., 2002; Czaja and Frankignoul, 2002; Cassou et al., 2004; Frankignoul and Kestenare, 2005; Wu and Liu, 2005].
 Summertime and fall air-sea CO2 fluxes were higher in the 1990s and 2000s (i.e., 2000–2005) compared to the 1980s (i.e., 1984–1989). For example, the average air-sea CO2 fluxes were lower during the 1980s when summertime and fall NAO values had mean positive values (+0.15 and +0.12). In contrast, in the 1990s and 2000s average air-sea CO2 fluxes were higher when summertime and fall NAO values had mean negative values (Table 10). Thus the increase in summertime and fall air-sea CO2 flux described earlier (in section 6.2) presumably reflects interannual changes in atmospheric forcing between the 1980s and 1990s. Indeed, the NAO (and AO) transitioned to a period of positive states in 1989 [e.g., Curry and McCartney, 2001], cooccurring with an increase in sea surface salinity in the subtropical gyre of the North Atlantic Ocean [Hakkinen, 2001] and BATS site (Figure 1a) [Bates, 2001], presumably as a consequence of weakening of overturning circulation [e.g., Wu and Liu, 2005], and changing heat and freshwater fluxes. Since 1989, the NAO has remained generally positive, though it has trended toward a negative state, with fluctuations between strongly negative years (e.g., 1996 and 1998) and positive years (e.g., 1999, 2000, 2002). Coincident with this change is a reduction in baroclinic mass transport of the Gulf Stream [Curry and McCartney, 2001; Molinari, 2004] and freshening (particularly during the summer) of the surface layer (Figure 1a). Beginning in 1995, there has also been a stepwise increase in the number of major hurricanes occurring in the North Atlantic basin each year (0–2 pre-1995; 2–6 post-1995 [Molinari and Mestas-Nuñez, 2003; Elsner et al., 2004]), which may have also contributed to the enhancement of summertime air-sea CO2 fluxes in the midlatitudes of the North Atlantic Ocean.