Interannual variability of the Azores Current strength and eddy energy in relation to atmospheric forcing

Authors


Abstract

[1] Spaceborne observations of sea surface topography have revealed a significant interannual variability of the Azores Current strength and eddy energy. The objective of this paper is to establish the relationship between these variations and atmospheric forcing over the subtropical North Atlantic. Based on satellite altimetry, hydrography, and atmospheric reanalysis products, it is demonstrated that the interannual variability of the Azores Current eastward velocity and eddy energy may be driven by the adjustment of the ocean to the strength of westerly and trade winds, modulated by the North Atlantic Oscillation. Surface intensification (frontogenesis), which is mainly due to the wind-driven meridional Ekman current convergence, is found significant, but not sufficient to explain the observed interannual variability of the Azores Current strength.

1. Introduction

[2] The Azores Current (AzC) is a prominent eastward jet-like flow in the subtropical North Atlantic between 32°N and 36°N, transporting about 10–12 Sv (1 Sv = 106 m3/s) toward the Gulf of Cadiz (Figure 1) [Käse and Siedler, 1982; Gould, 1985]. It represents a front (henceforth the Azores Front (AzF)) with significant temperature and salinity gradients [Käse et al., 1985]. East of the Mid-Atlantic Ridge (MAR), the AzC produces three main southward flowing branches with variable locations [Stramma and Siedler, 1988; Klein and Siedler, 1989]. Two branches recirculate into the North Equatorial Current flowing west [Maillard and Käse, 1989], while the easternmost branch joins the Canary Current [New et al., 2001]. Observations have shown an existence of a westward counterflow located to the north of the AzC that is usually referred to as the Azores Countercurrent (AzCC) [Onken, 1993; Cromwell et al., 1996]. It is believed that the formation of the AzC-AzCC system relies on the dynamical concept of β plumes [Jia, 2000; Özgökmen et al., 2001; Kida et al., 2008; Volkov and Fu, 2010].

Figure 1.

The December 1992 to December 2010 average surface EKE (cm2/s2, color) and geostrophic circulation (cm/s, arrows). Black dashed line indicates the time-mean location of the zero meridional Ekman transport. Bathymetry is shown at 2000 and 3000 m.

[3] The observed eddy energy of the AzC is associated with the meandering of the current, the generation of eddies, and the westward propagation of Rossby waves [Le Traon and De Mey, 1994]. The distribution and variability of eddy kinetic energy (EKE) in the North Atlantic including the Azores region have been described in several studies [e.g., Le Traon and De Mey, 1994; Ollitrault and Colin de Verdiere, 2002; Brachet et al., 2004; Volkov, 2005; Barbosa Aguiar et al., 2011]. Baroclinic instability is considered to be the major eddy generation mechanism in the AzC [Kielmann and Käse, 1987; Alves and Colin de Verdiere, 1999] and all over the ocean [Stammer, 1997]. Wind forcing can be another direct or indirect energy source for eddy variability in some regions. Significant direct impact of seasonally varying wind forcing on EKE has been found in the northern North Atlantic, characterized by a weak stratification and low background eddy energy [White and Heywood, 1995; Stammer and Wunsch, 1999; Stammer et al., 2001]. Some indirect effects of wind forcing have also been reported. For example, Qiu and Chen [2010] found that Ekman convergence could be responsible for the interannual variability of the Subtropical Countercurrent EKE in the North Pacific. Garnier and Schopp [1999] showed that changes of EKE in the Gulf Stream and the North Atlantic Current follow the alterations of the eastward Sverdrup transport.

[4] Using numerical experiments, Volkov and Fu [2010] demonstrated that wind affects the magnitude of the AzC transport and EKE. It has been reported that the interannual changes of EKE in the North Atlantic are probably related to the North Atlantic Oscillation (NAO) [Penduff et al., 2004; Volkov, 2004; Brachet et al., 2004]. Although the direct effect of wind stress on the AzC eddy energy is weak [Stammer et al., 2001], Brachet et al. [2004] noted that the AzC EKE reduced considerably following the switch of the NAO index from positive in 1995 to negative in 1996. This suggests that on the interannual time scale the AzC EKE is likely to be modulated by the adjustment of the ocean to the time-variable large-scale atmospheric forcing. The short duration of the observational records, however, has impeded the validation of this relationship.

[5] The objective of this paper is to establish the relationship between the interannual variability of the AzC eddy energy east of the Mid-Atlantic Ridge and atmospheric forcing. Nowadays, the multiyear observational records make it possible to establish a statistically significant relationship. In this study we estimate the AzC EKE using 18 years of sea surface height measurements by spaceborne altimeters and couple these estimates to hydrography and atmospheric reanalysis data. We also investigate the possibility of the surface intensification of the AzC eastward velocity and EKE by atmospheric forcing.

2. Data

[6] The maps of sea level anomaly (MSLA) from October 1992 to December 2010, produced from the merged geophysical data records of Topex/Poseidon, ERS-1/2, Jason-1, Jason-2/OSTM, Envisat, and Geosat Follow-On (GFO) altimetry missions, were used to compute geostrophic velocities and EKE. The satellite altimetry data were corrected for instrumental errors, geophysical effects, tidal influence, and atmospheric wind and pressure effects [e.g., Le Traon et al., 2003; Volkov et al., 2007], and interpolated to a 1/3° Mercator projection grid using a method based on a suboptimal space-time objective analysis that had taken into account along-track correlated errors [Le Traon et al., 1998]. Assuming geostrophy, the eddy velocity components (deviations from the time mean) are ug = −(g/f)(∂η/∂y) and vg = (g/f)(∂η/∂x), and thus instantaneous EKE = (ug2 + vg2)/2, where η is the sea level anomaly, g is gravity, and f is the Coriolis parameter. The absolute geostrophic velocities were computed from the absolute dynamic topography, obtained by adding the MSLA referenced to the 1993–1999 time period to the MDT_CNES-CLS09 mean dynamic topography (MDT) product for the 1993–1999 period, based on 4.5 years of Gravity Recovery and Climate Experiment (GRACE) data, and 15 years of altimetry and in situ (hydrography and drifter) measurements.

[7] Hydrography data consist of weekly gridded fields of temperature and salinity produced by objective analysis of temperature profiles and time series from profiling floats, XBTs, thermosalinographs, and drifting and moored buoys (more than 80% of profiles are acquired within the Argo program). These data (henceforth the Coriolis data) were collected and made freely available by the Coriolis project and programs that contribute to it (http://www.coriolis.eu.org). The Coriolis data are mapped on a 1/2° Mercator projection grid at 59 vertical levels from 5 to 1950 m depth for the time period from January 2002 to June 2009.

[8] Atmospheric data used in this study come from two sources. The first are the ERA-Interim monthly averages of daily mean wind stress and net surface heat flux at a 1.5° resolution from January 1992 to December 2010 provided by the European Centre for Medium Range Weather Forecast. The second is the winter (December through March) NAO index from 1992–1993 to 2010–2011 distributed by the Climate Analysis Section of the National Center for Atmospheric Research (Boulder, Colorado; J. Hurrell, NAO/NAM climate indicies, 1995, http://www.cgd.ucar.edu/cas/jhurrell/indices.html). The NAO index is based on the difference of normalized sea level pressure between Lisbon, Portugal and Stykkisholmur/Reykjavik, Iceland. The magnitude of this index is characteristic of the strength of westerly winds in the North Atlantic.

3. Results

3.1. Eddy Energy of the Azores Current

[9] The instability of the AzF converts its available potential energy into EKE. The December 1992–December 2010 mean surface EKE and geostrophic velocities in the subtropical North Atlantic Ocean are shown in Figure 1. Downstream from the MAR at about 38°W the EKE of the AzC significantly decreases. However, as the current proceeds toward the Gulf of Cadiz, it continues to manifest elevated EKE ranging from about 50 cm2 s−2 in the east to over 200 cm2 s−2 over the ridge. These estimates are similar to EKE produced with drifter data [Barbosa Aguiar et al., 2011] except a small offset (drifter EKE are typically larger than altimetry EKE) that is probably caused by the contribution of Ekman currents. The AzC axis, defined as the location of the maximum EKE, appears to have a slight SW-NE tilt. It crosses the MAR at about 34°N and approaches the Gulf of Cadiz above 35°N. Barbosa Aguiar et al. [2011] did not find the same tilt in drifter data and hypothesized that the tilt seen in the altimetry EKE maps may result from a high incidence of cyclonic eddies in the southwestern part of the AzC: “the number of cyclones (dominant in the south) tends to be higher than the number of anticyclones (dominant in the north), which translates into a band of EKE maxima shifted to the south.” We do not support this hypothesis, because the altimetry time-mean absolute eastward velocity (Figure 1) shows the same tilt. If we define the AzC axis as the local maximum of ug, then it is located at 33.5°N at 35°W (ug = 10 cm/s), at 34.3°N at 25°W (ug = 9 cm/s), and at about 35.1°N at 15°W (ug = 6 cm/s). The tilt was also documented by Ollitrault and Colin de Verdiere [2002] by analyzing the trajectories of floats drifting near 700 m depth. We hypothesize that the tilt may be related to topographic features that define the path of the AzC over the MAR and the tendency of the AzC to head toward the Gulf of Cadiz, where the entrainment of the North Atlantic Central Water by the underlying denser Mediterranean Outflow Water is believed to be responsible for the formation of the AzC [Jia, 2000; Özgökmen et al., 2001; Kida et al., 2008; Volkov and Fu, 2010]. Ekman convergence along the AzC can be another possible cause for the tilt, as will be illustrated in the next section.

[10] The AzC EKE, averaged between 15°W and 35°W, exhibits significant interannual variability (Figure 2a). The level of EKE is in part related to the eastward geostrophic velocity (contoured in Figure 2a), an integral quantity for the meridional density gradient at the AzF. Stronger (weaker) eastward flow becomes less (more) stable and thus its EKE increases (reduces). The periods with the largest EKEs (1994–1995 and 2007–2008) immediately follow the periods when the eastward flow of the AzC was the strongest. The eastward geostrophic velocity in 1996–2000 was lower than in 1993–1995, coinciding with a strong reduction of the AzC EKE. The meridional position of the AzC jet in 1993–1999, defined as the location of the maximum eastward velocity, was just below 34°N. In 2000–2003 the eastward geostrophic velocity and EKE increased and the AzC jet shifted northward by about 0.5°. In 2004–2005 the geostrophic velocity and EKE decreased and the AzC jet started to shift southward. In 2006–2007 the AzC jet accelerated to 10 cm/s, returned to its 1993–1995 location at 34°N, and its EKE started to rise, reaching a maximum in winter 2007–2008. The meridional migrations of the AzC jet described here in general agree with those documented by Barbosa Aguiar et al. [2011]. The differences can be attributed to averaging EKE over smaller zonal sectors and defining the meridional position of the jet as the location of the maximum EKE.

Figure 2.

(a) Time-latitude diagram of EKE (cm2/s2, color) and eastward surface geostrophic velocity (cm/s, contour) averaged over 15°W–35°W. (b) EKE averaged over 15°W–35°W and 33°N–36°N (black curve) and winter (December through March) NAO index (bars). The EKE and geostrophic velocity time series are low-pass filtered with a yearly running mean.

3.2. Relation to Atmospheric Forcing

[11] The anticyclonic pattern of atmospheric circulation in the subtropical North Atlantic is composed of westerly winds in the north and trade winds in the south. These winds drive the convergent meridional Ekman transport, bringing cold water from the north and warm water from the south toward the AzF (Figure 3). The AzC essentially flows along the zero line of the meridional Ekman transport (Figure 1). It is interesting to note that this line has a SW-NE tilt similar to the observed tilt of the AzC axis, suggesting that it is likely influenced by Ekman convergence. The near-surface Ekman convergence in the subtropical North Atlantic induces Ekman pumping that causes a predominantly southward Sverdrup transport that affects the entire water column below the mixed layer. The impact of wind forcing on the meridional density gradient, eastward transport, and EKE of the AzC was demonstrated by the authors using numerical experiments in an earlier study [for details see Volkov and Fu, 2010]. In an experiment driven only by the lateral boundary conditions and buoyancy fluxes without wind forcing, the time-mean transport of the AzC was about 2 Sv smaller than in an experiment with wind forcing. Figure 4 shows the difference between the meridional density gradient in the experiment with wind and in the experiment without wind and corresponding isopycnals across a quasi-meridional curvilinear section (which is due to the model grid geometry) centered at approximately 25°W. The meridional density gradient below the upper mixed layer (Figure 4) and surface EKE [Volkov and Fu, 2010, Figure 17d] at the AzF in the experiment with wind are considerably larger than in the experiment without wind. Within the upper mixed layer the absence of wind greatly reduces mixing, so that the meridional density gradient is mainly determined by buoyancy fluxes and appears to be larger than that in the experiment with wind. The numerical experiments thus explicitly show that wind forcing is capable of modifying the AzC strength.

Figure 3.

The January 1992 to December 2010 average ECMWF-Interim wind stress (N/m2, arrows) and meridional Ekman transport (m2/s, color). Black dashed line indicates the time-mean location of the zero meridional Ekman transport. Bathymetry is shown at 2000 and 3000 m depths.

Figure 4.

Potential density (kg/m3) for the numerical experiments forced by realistic wind (solid contour) and without wind forcing (dashed contour), and the difference of the meridional density gradients (color, kg/m4) between the two experiments across a quasi-meridional curvilinear section centered at approximately 25°W.

[12] On the basis of 8 years of altimetry observations, Brachet et al. [2004] reported that the reduction of EKE in 1995–1996 shown in Figure 2b followed a switch of the NAO index from positive to negative. The positive (negative) phase of the NAO is characterized by an increased (reduced) pressure gradient between the Azores high- and Icelandic low-pressure centers that leads to stronger (weaker) westerly winds in the North Atlantic. Updated time series (Figure 2b) confirm the relevance of the NAO. The correlation coefficient between the winter (December through March) NAO index and yearly averages of EKE averaged over 15°W–35°W and 33°N–36°N is 0.56, which is above the 95% significance level for 18 degrees of freedom. Because the zero crossing of the autocorrelation function of the winter NAO index is close to the time lag of 1 year, it is reasonable to assume that each year out of 18 provides an independent sample. The removal of a quadratic fit (a second-degree polynomial that fits the data best in a least squares sense) from both time series, which is appropriate in order to filter out the long-term (interdecadal) tendencies and to focus on the interannual variability only, increases the correlation to 0.72. The relevance of the NAO implies that at the interannual time scale the AzC eddy energy is influenced by the adjustment of the ocean to the time-variable large-scale atmospheric forcing.

[13] Although the time series of the AzC EKE and the winter NAO (Figure 2b) are significantly correlated, the relationship between these quantities appears to be rather complex. After negative values in 1995–1996 and 1996–1997 the NAO phase returned to positive in 1998–2000. Although EKE experienced an increase through 2001, it remained low compared with that of the 1993–1995 time interval. On the other hand, a switch from the 2005–2006 negative NAO to the short-term positive phase in 2007–2008, with the same magnitude as in 1999–2000, was followed by a stronger increase of EKE.

[14] Figure 5 shows the 2 year running means of EKE averaged over 15°W–35°W and 33°N–36°N and the larger-scale ERA-Interim wind stress curl averaged over 15°W–35°W and 25°N–45°N. A quadratic fit was removed from both time series to filter out long-term tendencies. The maximum correlation between the time series is −0.77 (significant at 95% confidence for 9 degrees of freedom) and it is obtained when the wind stress curl leads the EKE time series by about 7 months (Figure 6). We suggest that this lag may represent the time required for the AzC to adjust to the time-varying wind forcing plus the time for instabilities to grow. It should be noted that the time lag in the correlation arises mainly from the period before 2008.

Figure 5.

The time series of wind stress curl averaged over 15°W–35°W and 25°N–45°N (dashed curve) and EKE averaged over 15°W–35°W and 33°N–36°N (solid curve) and the winter (December through March) NAO index (black bars). The time series of wind stress and EKE are low-pass filtered with a 2 year running mean, and a quadratic fit is removed.

Figure 6.

Cross correlation between the interannually varying wind stress curl in the eastern subtropical North Atlantic and the AzC EKE plotted in Figure 5.

3.3. Surface Intensification of the Azores Front

[15] The process by which weak horizontal gradients are intensified into strong fronts is referred to as frontogenesis. Ekman convergence is one of the most important mechanisms for mixed layer frontogenesis [Gill, 1992]. EKE is released deeper at depths where the thermocline is located. However, because the AzC is surface intensified [e.g., Tychensky et al., 1998], it is reasonable to expect that the strength of the AzC and its EKE are sensitive to the near-surface Ekman forcing. The NAO-modulated interannual variabilities of westerly and trade winds alter the Ekman currents that bring colder water from the north and warmer water from the south toward the AzC and modify the near-surface meridional temperature gradient at the AzF. In this section we use a simple dynamical framework and investigate to what degree the frontogenesis can influence the interannual variability of the AzC strength.

[16] By the virtue of the thermal wind balance the vertical shear of the zonal geostrophic velocity (ug) is proportional to the meridional temperature gradient:

equation image

where θ is temperature and α is the thermal expansion coefficient. The time-depth diagram of −∂θ/∂y averaged over 15°W–35°W and 33°N–36°N and computed from the Coriolis data is presented in Figure 7. The meridional temperature gradient exhibits a local maximum in 2003, a depression in 2005, and intensification in 2006–2008, being in a good agreement with the altimetry zonal geostrophic velocity (Figure 2a). The maximum −∂θ/∂y in the upper 100 m is observed in winter months. As one can see in Figure 7, the surface signal is likely to propagate into deeper layers. The sign of −∂θ/∂y then reverses at a depth of about 800 m, probably owing to the intrusion of the Mediterranean water.

Figure 7.

Time-depth diagram of the meridional temperature gradient averaged over 15°W–35°W and 33°N–36°N (°C/m).

[17] The depth of the surface mixed layer and its seasonal variability in the AzC region can be inferred from the vertical profiles of temperature and its vertical gradient (Figure 8). The maximum depth of the bottom of the mixed layer exceeding 100 m is observed in the winter-spring period. It shoals to a minimum depth of about 35 m in summer and 55 m in autumn. The relative proximity of the bottom of the mixed layer to the surface means that it is strongly influenced by near-surface Ekman currents. It also suggests a possibility for the AzC EKE to be modulated by Ekman convergence.

Figure 8.

Vertical profiles of potential temperature (°C, left) and vertical temperature gradient (°C/m, right) averaged over 15°W–35°W and 33°N–36°N. The profiles were averaged over the seasons: December through February (DJF, black), March through May (MAM, green), June through August (JJA, red), and September through November (SON, blue).

[18] The vertical shear of the zonal geostrophic velocity in equation (1) over a layer k with a thickness Δzk can be approximated as follows:

equation image

where Gk = −∂Tk/∂y and Tk = equation image θdz. If we assume the level of no motion to be at the 800 m depth, which is reasonable given Figure 7, equation (2) for the upper 800 m gives an approximation of the zonal geostrophic velocity at the surface. In order to estimate the relative contribution of the surface and deep layers to the interannual variability of ug at the surface, we also compute Δugk over four depth intervals: 0–100 m (surface layer in agreement with Figures 7 and 8), 100–200 m (subsurface layer), 200–500 m (intermediate layer), and 500–800 m (deep layer). The thermal expansion coefficient αk is computed for each layer as a function of salinity, temperature, and pressure.

[19] Figure 9 shows the anomalies (deviations from the 2002–2009 mean) of ug estimated from altimetry data versus Δugk estimated from the Coriolis temperature data for the entire 800 m and for each layer. The time series are averaged over 15°W–35°W and 33°N–36°N and smoothed with a yearly running mean filter. Despite some differences, the time series of ug and Δug(0–800 m), in general, agree with each other. The correlation coefficient between them is 0.74, which is significant at 95% confidence for 7 degrees of freedom. Both time series show an intensified surface eastward flow in 2003, a depression in 2004–2005, and an intensification in 2006–2007. However, the time series disagree in 2008 and the maximum cross correlation of about 0.79 between them is observed when Δug(0–800 m) lags behind ug by 140 days. This discrepancy can be attributed to different spatial and temporal scales resolved by altimetry and the Coriolis data, the choice of the reference level for the computation of Δug, the neglect of halosteric effects, and measurement errors. The Coriolis data are mainly based on temperature and salinity profiles from a sparse (average 3° × 3° spacing) array of Argo floats. XBT and CTD measurements are also rather sparse in the region. The actual resolution of the Coriolis data is, therefore, several times lower than the resolution of the merged multisatellite altimetry data.

Figure 9.

The altimetry zonal geostrophic velocity anomaly (solid black) versus the Coriolis zonal vertical shear anomalies over 0–800 m (dashed black), 0–100 m (solid red), 100–200 m (dashed red), 200–520 m (solid blue), and 520–800 m (dashed blue) computed using equation (2). The time series are averaged over 15°W–35°W and 33°N–36°N and smoothed with a yearly running mean filter.

[20] Because here we discuss the surface intensification of the AzC, our attention is focused on the mixed layer above a 100 m depth. As illustrated in Figure 9, the contribution of this layer (marked with a solid red curve) to the interannual variability of the eastward flow obtained from the altimetry data and that integrated over the upper 800 m using the Coriolis data is significant. To quantify the relative importance of each layer, Δugk, we calculate explained variances for the Coriolis (VexpC) and altimeter (VexpA) data: VexpC = 1 − equation image and VexpA = 1 − equation image. The results are summarized in Table 1. It appears that the Coriolis Δug(0–100 m) explains 20% of the Coriolis Δug(0–800 m) and 26% of the altimetry ug variance. It is interesting to note that while the range of Δug(0–800 m) and ug variability in 2002–2004 is smaller than in 2005–2008, the contribution of Δug(0–100 m) to this variability in the former time interval is apparently larger than in the latter time interval. In 2002–2004 the Coriolis Δug(0–100 m) explains 40% of the Coriolis Δug(0–800 m) and 46% of the altimetry ug variance, while in 2005–2008 it explains 24% and 27% of the variance, respectively. This means that the relative importance of the upper mixed layer to the variability of the AzC is not constant in time.

Table 1. The Portion of Explained Variance (%) by the Interannual Variability of the Vertical Shear of the Zonal Velocity Over Different Depth Intervals (Δugk) Relative to the Interannual Variability of the Coriolis Surface Zonal Velocity Referenced to 800 m Depth and the Altimetry Zonal Velocity
Δugk0–100 m100–200 m0–200 m200–520 m520–800 m200–800 m
Δug(0–800 m)204053791786
ug26285041336

[21] The contribution of the subsurface layer confined to 100–200 m depth interval (dashed red curve in Figure 9) is somewhat different from the contribution of the surface layer. The Coriolis Δug(100–200 m) explains 40% of the Coriolis Δug(0–800 m) and 28% of the altimetry ug variance. In 2002–2004 it explains 38% of the Coriolis Δug(0–800 m) and 18% of the altimetry ug variance, and in 2005–2008 it explains 40% and 27%, respectively. The maximum correlation of 0.67 between the surface Δug(0–100 m) and subsurface Δug(100–200 m) velocity anomalies is observed when the latter lags behind the former by approximately 70 days. This suggests that signals generated in the surface layer may propagate into deeper layers. Such a propagation reaching nearly 200 m depth can also be inferred from Figure 7. Therefore, it seems likely that effects of the surface Ekman convergence at the AzF can be translated deeper, beyond the mixed layer, and influence the generation of the AzC EKE. The combined time series of the surface and subsurface vertical shears of the eastward velocity, Δug(0–100 m) + Δug(100–200 m), explains 53% of the Coriolis Δug(0–800 m) and 50% of the altimetry ug variance.

[22] Nevertheless, as one can see in Figure 9 and Table 1, the most important contribution to the interannual variability of the AzC strength apparently comes from the intermediate layer confined to the 200–520 m depth interval. It alone explains 79% of the Coriolis Δug(0–800 m) and 41% of the altimetry ug variance. The interannual variability in the intermediate layer is in phase with that in the subsurface layer, supported by the maximum correlation of 0.95 between Δug(100–200 m) and Δug(200–520 m) observed at zero time lag. In 2002–2004 Δug(200–520 m) explains 44% while in 2005–2008 it explains 79% of the Δug(0–800 m) variance. The deep layer (520–800 m) has a smaller effect than each of the upper layers, but the combination of the intermediate and deep layers explains 86% of the Coriolis Δug(0–800 m) and 36% of the altimetry ug variance.

[23] To summarize this section, we can conclude that although the contribution from deeper layers is obviously more important, the interannual variability of the meridional temperature gradient in the upper 100 m makes a sizable contribution to the interannual variability of the AzC eastward velocity and, therefore, EKE. It should be noted, however, that while the contribution from the upper 100 m in 2006–2007 was smaller than in 2003, the maximum acceleration of the AzC and increase of its EKE was largest in 2006–2007 and in 2007–2008, respectively.

3.4. Mechanisms of Near-Surface Frontogenesis

[24] In this section we investigate the forcing mechanisms that drive the interannual variability of the meridional temperature gradient in the upper 100 m layer. Let this layer be characterized by a uniform temperature T and its meridional gradient G = −∂T/∂y. According to a slab mixed layer approach [de Ruijter, 1983; Kazmin and Rienecker, 1996], the zonally averaged temperature balance within the mixed layer of thickness H can be approximated as follows:

equation image

where VEK = −τx/(ρfH) is the vertically averaged meridional Ekman velocity; WEK = equation image × (equation image/ρf) is the vertical velocity at the base of the mixed layer (Ekman pumping); equation image (τx, τy) is the wind stress; ΔT is the temperature difference between the mixed layer and the water immediately below; QNET is the net surface heat flux (positive into the ocean); CP is the seawater specific heat capacity; ρ is the water density; and D represents all the terms that cannot be easily estimated with available data or that are assumed negligible (mainly diffusion, geostrophic advection, entrainment at the bottom of the mixed layer, etc.).

[25] The upper layer frontogenesis equation is the meridional gradient of (3), which after neglecting D becomes

equation image

[26] This approximation states that the tendency of the local meridional temperature gradient is governed by (1) the near-surface Ekman forcing that consists of an Ekman current convergence component G(∂VEK/∂y) and a meridional Ekman advection component VEK(∂G/∂y), (2) Ekman pumping that can modify the depth of the pycnocline, and (3) the meridional variation of the net surface heat flux. It should be noted that equation (4) represents an idealized balance, and the contribution of the neglected terms is probably not small. For example, turbulent mixing can be significant in the eddy-rich AzC region [e.g., Ollitrault and Colin de Verdiere, 2002]. The effect of the meridional temperature gradient advection by non-Ekman currents can also be important and govern the variability at depth. But because in this section we only investigate the role of the upper mixed layer directly influenced by wind, we leave the consideration of the terms contained in D for a future study. The components of equation (4) were estimated using the ERA-Interim wind stress, net surface heat flux, and Coriolis temperature data. ΔT in equations (3) and (4) is space and time dependent and varies predominantly with seasonal periodicity from about 0.2°C in March to nearly 4°C in August.

[27] Figure 10 shows the low-pass filtered (with a yearly running mean) time series of the observed (in Coriolis temperature data, ∂G/∂t) and predicted (the sum of forcing terms in equation (4)) frontogenesis. The correlation coefficient between the time series is 0.81, which is significant at 95% confidence. The frontogenesis predicted by the forcing terms in equation (4) appears to make a substantial contribution to the observed interannual variability of the meridional temperature gradient. The sum of the forcing terms explains 65% of the observed frontogenesis. The predicted frontogenesis is always positive, meaning that the sum of the forcing terms acts to intensify the AzF. The offset between the observed and predicted frontogenesis is due to the neglected terms in equation (4), mainly diffusion, and also due to the difference in spatial and temporal scales resolved in the altimetry, Coriolis, and ERA-Interim data sets.

Figure 10.

Tendency in the meridional temperature gradient averaged over the upper 100 m (frontogenesis) observed in the Coriolis data (solid curve) and predicted by the forcing terms (dashed curve) in equation (4). The time series are averaged over 15°W–35°W and 33°N–36°N and smoothed with a yearly running mean filter.

[28] The forcing terms of equation (4) are shown separately in Figure 11. The Ekman forcing term ∂(VEKG)/∂y is split into the Ekman current convergence (G(∂VEK/∂y)) and the meridional Ekman advection (VEK(∂G/∂y)). The primary contribution to the AzC frontogenesis is provided by the Ekman current convergence, with a time mean of ∼2.4 × 10−14 °C m−1 s−1, which always enhances the AzF and explains 71% of the variance of the total estimated forcing. The contribution from other terms is smaller and generally not frontogenetic. Ekman advection provides a weak frontolysis (frontal decay), except for 2002, with a time mean of ∼−0.25 × 10−14 °C m−1 s−1, and explains 38% of the forcing variance. The second important contribution to the interannual variability of the AzC frontogenesis comes from the Ekman pumping effect that explains 47% of the forcing variance. Although the time mean of the Ekman pumping term is rather small (∼−0.4 × 10−14 °C m−1 s−1) and provides frontolysis, its interannual variability is correlated (correlation coefficient = 0.75) with the sum of the forcing terms in equation (4). The meridional gradient in the net surface heat flux with a time mean of about −0.19 × 10−14 °C m−1 s−1 appears to provide the weakest forcing to the interannual variability of the AzC frontogenesis. Most of the time it causes frontolysis because on average the ocean loses slightly more heat to the south of the AzC than to the north.

Figure 11.

The terms of equation (4): tendency in the meridional temperature gradient predicted by the sum of the forcing terms (black curve), the Ekman current convergence (blue curve), the meridional Ekman advection (dashed blue curve), the Ekman pumping component (red curve), and the meridional variation of the net surface heat flux (green curve). The time series are averaged over 15°W–35°W and 33°N–36°N and smoothed with a yearly running mean filter.

4. Summary and Discussion

[29] Multiyear monitoring of sea surface heights in the North Atlantic has revealed that the eastward transport and eddy energy of the AzC are subject to a significant interannual variability. In this paper, using observational data and an atmospheric reanalysis product from 1992 to 2011, we have analyzed the relationship between the interannual variability of atmospheric forcing and the AzC strength and EKE. We have investigated one possible mechanism of the AzC adjustment to atmospheric forcing: surface intensification that is due to frontogenesis processes in the upper mixed layer.

[30] Eddy energy of the AzC mainly results from baroclinic instability [Kielmann and Käse, 1987; Alves and Colin de Verdiere, 1999]. We have demonstrated that on the interannual time scales the intensifications of the AzC EKE follow the periods of stronger eastward flow, because a stronger (weaker) AzC becomes less (more) stable and generates more (less) eddies. We have compared the interannual variability of the AzC EKE with the winter (December through March) NAO index and found that the two time series are significantly correlated. Positive (negative) winter NAO indices correspond to increased (reduced) EKE. This result confirms the suggestion, made by Brachet et al. [2004] and based on a shorter record, that large-scale atmospheric forcing can modulate the interannual variability of the AzC EKE.

[31] The atmospheric circulation in the subtropical North Atlantic is composed of westerly winds in the north and easterly trade winds in the south. We have established a statistically significant relationship between the interannual variability of wind stress curl in the eastern part of the subtropical North Atlantic and the interannual variability of the AzC EKE after filtering out longer-term tendencies. Consistent with the NAO, it appears that elevated EKE follows the periods of intensified anticyclonic atmospheric circulation with a time lag of about 7 months. The correlation with both the NAO and wind stress curl indicates that the AzC adjusts to a large-scale atmospheric forcing. The mechanisms of this adjustment were explored by Spall [1997], who showed that the presence of a large-scale deformation field, such as the convergent Ekman transport, could maintain a strong baroclinic jet. The resulting frontogenesis is balanced by the frontolysis of baroclinic instability.

[32] In this paper we have attempted to quantify the contribution of the surface intensification of the AzC related to atmospheric forcing. A combined analysis of altimetry and hydrography data available from 2002 to 2009 has shown that a significant part of the interannual variability (∼1/5) of surface eastward flow can be explained by variations in the meridional temperature gradient in the upper 100 m. We have demonstrated that the surface signal is likely to propagate deeper and the variability in the 100–200 m layer lags behind the 0–100 m layer by about 2 months. The variability in the upper 200 m appears to explain over 50% of the eastward flow variance. Thus, the AzC is indeed surface intensified, in agreement with earlier studies [e.g., Tychensky et al., 1998]. Although EKE is released at the thermocline, the depth of the top of the thermocline is rather shallow, varying from less than 40 m in summer to about 100 m in winter, meaning that the thermocline is subject to atmospheric influence.

[33] The observed interannual variability of the near-surface meridional temperature gradient from 2002 to 2009 has been analyzed with respect to three major forcing mechanisms: (i) surface Ekman forcing consisting of an Ekman current convergence and a meridional Ekman advection components, (ii) the meridional variations in Ekman pumping, and (iii) the meridional variations in the net surface heat flux. The sum of these forcing mechanisms appeared to explain 65% of the observed near-surface frontogenesis variance. It is demonstrated that the primary contribution to the frontogenesis at the AzF is provided by the meridional Ekman current convergence that always acts to tilt the near-surface isopycnals enhancing the front and explains 71% of the variance of the total estimated forcing. This result agrees with the studies of the subtropical frontal zones in the North Pacific [Kazmin and Rienecker, 1996; Qiu and Chen, 2010]. Although the time-mean forcing related to the meridional variations in Ekman pumping is rather small and provides frontolysis, this forcing term appears to explain 47% of the total forcing variance. The contributions from the Ekman advection component and the meridional variations in the net surface heat flux are small.

[34] In summary, this study provides three major conclusions:

[35] 1. The AzC EKE is significantly correlated with the winter NAO index and with the wind stress curl over the eastern part of the subtropical North Atlantic, suggesting the adjustment of the current strength to large-scale atmospheric forcing;

[36] 2. Near-surface intensification of the AzF (frontogenesis) can be responsible for a significant portion of the interannual variability of the AzC strength, but the major contribution apparently comes from other processes not considered in this study;

[37] 3. Frontogenesis at the AzF is mainly due to the wind-driven meridional Ekman current convergence, which always acts to bring isopycnals closer, thus increasing the meridional temperature gradient and influencing the strength and the variability of the AzC.

[38] The largest contribution to the interannual variability of the AzC strength is apparently due to deeper processes and/or the processes neglected in the approximation used in this study (diffusion and geostrophic advection). In an earlier study it has been experimentally shown that the strength and variability of the AzC is sensitive to the water mass exchange through the Strait of Gibraltar [Volkov and Fu, 2010]. Although the authors found no significant impact of the seasonal variability of the exchange transport on the AzC EKE, it is possible to expect that long-term variability of the exchange transport and/or water mass properties may also affect the AzC EKE. Additional research using an ocean general circulation model is required to address all these issues.

Acknowledgments

[39] This research was carried out at Jet Propulsion Laboratory, California Institute of Technology, and sponsored by the NASA Physical Oceanography program. MSLA/MDT_CNES-CLS09 are produced by Ssalto-Duacs/C.L.S. Space Oceanography Division and distributed by Aviso (www.aviso.oceanobs.com) with support from CNES (Centre National d'Etudes Spatiales, Toulouse, France). Government sponsorship is acknowledged.