Journal of Geophysical Research: Oceans

Physical drivers of interannual chlorophyll variability in the eastern subtropical North Atlantic

Authors


Abstract

[1] Interannual chlorophyll variability and its driving mechanisms are evaluated in the eastern subtropical North Atlantic, where elevated surface chlorophyll concentrations regularly extend more than 1500 km into the central subtropical North Atlantic and modulate the areal extent of the North Atlantic's lowest chlorophyll waters. We first characterize the considerable interannual variability in the size of the high chlorophyll region using SeaWiFS satellite data. We then evaluate the relationship between satellite chlorophyll and sea surface height (SSH), which are anticorrelated in the study region, most likely as a result of the inverse relationship between SSH and nutricline depth. To put these results in a longer temporal context, we study a hindcast simulation of a global ocean model with biogeochemistry (GFDL's MOM4.1 with TOPAZ biogeochemistry), after evaluating the model's skill at simulating chlorophyll and SSH relative to observations. In the simulation, the variability seen during the satellite era appears to be imbedded in a much larger multidecadal modulation. The drivers of such variability are assessed by evaluating all the terms in the nutrient budget of the euphotic zone. Because diffusive processes are not a dominant control on nutrient supply, stratification is not a good indicator of nutrient supply. Rather, vertical advection of nutrients, strongly tied to Ekman pumping, is the leading driver of variability in the size of the high chlorophyll region and the productivity within the study area.

1. Introduction

[2] The eastern subtropical North Atlantic accommodates one of the most important eastern boundary current ecosystems in terms of total annual primary production [Carr, 2002; Carr and Kearns, 2003]. The upwelling system is bounded in the west by the subtropical gyre, a low-productivity region characterized by downwelling. In a warming environment, conditions in the subtropical gyre and in the coastal upwelling region may evolve differently. Coastal upwelling rates have been hypothesized to increase due to an increased land-sea pressure gradient [Bakun, 1990; Bakun et al., 2010]. On the other hand, increased stratification in oligotrophic gyres (a possible consequence of warming) is thought to reduce upward mixing of nutrients and decrease photosynthesis [McClain et al., 2004; Gregg et al., 2005; Behrenfeld et al., 2006; Polovina et al., 2008]. However, the link between declines in surface chlorophyll and increased stratification has most often been inferred from correlations between chlorophyll and sea surface temperature, leaving the underlying physical mechanisms unexplored. Specifically, an expansion of the lowest chlorophyll regions has been seen in satellite chlorophyll data [Polovina et al., 2008], but it is not clear that the position of all of the oligotrophic boundaries are governed by vertical mixing. Furthermore, a debate exists about whether recently observed primary production decline is a climate-change induced trend [Polovina et al., 2008; Irwin and Oliver, 2009] or part of a multidecadal oscillation [Martinez et al., 2009]. A recent study of satellite ocean color data and output from three biogeochemical models suggests that distinguishing between climate-change driven trends and natural variability will require at least 40 years of continuous satellite chlorophyll measurements [Henson et al., 2010].

[3] Here, we focus on the physical mechanisms that give rise to variability in the size of the highly productive eastern boundary upwelling region in the subtropical North Atlantic, the most spatially and temporally variable of the four major eastern boundary current ecosystems [Carr, 2002]. Because the size of the high chlorophyll region governs the eastward extension of the neighboring oligotrophic region, our work also sheds light on the hypothesized link between the size of the oligotrophic region and vertical stratification [McClain et al., 2004; Gregg et al., 2005; Behrenfeld et al., 2006; Polovina et al., 2008]. The basin's eastern limb shows one of the strongest zonal gradients of chlorophyll in the world's ocean [e.g., McClain, 2009, Figure 2], where the boundary between upwelling and downwelling favorable wind stress curl is found. The large-scale wind-stress field drives downwelling in the subtropical gyre, which results in a deep pycnocline and nutricline, and consequent low surface chlorophyll (<0.07 mg m−3). Along the basin's eastern margin, the wind forcing causes divergence of the horizontal ocean currents as a result of both nearshore Ekman transport and offshore curl-driven Ekman pumping [McClain and Firestone, 1993]. The upward velocities induced by this coastal Ekman transport and Ekman pumping bring nutrient-rich waters to the euphotic layer, through a combination of along-isopycnal and diapycnal transfer, sustaining high chlorophyll concentrations at the basin's eastern margin [Pelegrí et al., 2006].

[4] Surface chlorophyll in the eastern subtropical North Atlantic has a strong seasonal cycle; yet, interannual variability can be as large as the amplitude of the seasonal changes. Figure 1 illustrates the range of variability by comparing the difference between the average summer and winter chlorophyll concentrations to the difference between February 2000, which saw maximum chlorophyll concentrations in the satellite record, and February 1998, which saw minimum concentrations. While recent advances have been made in understanding the differences among various eastern boundary upwelling systems [e.g., Demarcq, 2009; Lachkar and Gruber, 2011], the focus has been on the narrow near-shore regime where coastal upwelling is the principal source of nutrients. Our study region extends from the coast to the edge of the low chlorophyll subtropical gyre and its focus is on interannual to decadal-scale variability, which is at or beyond the limit of the time scale observed by satellites. We thus turn to a global ocean model coupled to a state of the art biogeochemistry model, which simulates 49 years of ocean variability using historical forcing from the CORE-reanalysis [Griffies et al., 2009; Large and Yeager, 2009]. The use of models allows in-depth study of several hypothesized physical drivers of chlorophyll variability such as changes in stratification and the large-scale wind field.

Figure 1.

Maps illustrating chlorophyll variability at three time scales. (a and b) Seasonal variability is shown as winter minus summer chlorophyll concentrations. (c and d) Interannual variability is illustrated by February 2000 (February with maximum satellite averaged chlorophyll) minus February 1998 (February with minimum satellite averaged chlorophyll). (e) Interdecadal variability is shown as chlorophyll averaged during the decade 1959 to 1968 minus the average from 1977 to 1986. (left) SeaWiFS data and (right) model output. Dark gray contour shows the 0.2 mg m−3 isoline of chlorophyll for winter, February 2000 and the period 1959 to 1968. Light gray contour shows the same isoline for summer, February 1998 and the period 1977–1986.

Figure 2.

Maps of SSH and wind stress curl, and the relationship of SSH to chlorophyll. Climatological maps of SSH (averaged from November 1997 to December 2006) from (a) satellite observations and (b) the model. The study region is enclosed in a black rectangle, and contour lines show wind stress curl (N m−3) from QuikSCAT data [Risien and Chelton, 2008] in Figure 2a and CORE2 winds in Figure 2b. Temporal correlation of SSH and chlorophyll monthly anomalies from (c) satellite data and (d) model output. Each data point in the scatterplots corresponds to the spatial average over the study region of monthly anomalies.

[5] The model's skill at simulating spatial and temporal variability of the size of the high chlorophyll region is first tested by comparing its output to 9 years of Sea-viewing Wide Field-of-view Sensor (SeaWiFS) data (section 'Satellite—Model Chlorophyll Comparison') and satellite measurements of sea surface height (SSH, section 'Sea Surface Height and Chlorophyll'). We examine links between chlorophyll and SSH to explore the premise that SSH gives an insight to possible mechanisms explaining chlorophyll variability. We then assess the drivers of chlorophyll variability in the model output during the second half of the twentieth century. Specifically, we test various hypotheses that might explain biomass variability in section 'Mechanisms of Chlorophyll Variability' through the assessment of each term in a nutrient budget. At the end of section 5, we consider the large-scale physical drivers influencing the delivery of nutrients, including shifts in the wind-driven boundary of the subtropical gyre and large-scale climate forcing. The conclusions are presented in section 'Summary and Conclusions'.

2. Data and Methods

2.1. Satellite Data

[6] To validate the model's ability to represent variability in our study region, we use satellite chlorophyll and altimetric observations. We use Level 3 SeaWiFS monthly chlorophyll downloaded from http://oceancolor.gsfc.nasa.gov at 9 km resolution for the period November 1997 to December 2006. What we call satellite SSH is the absolute sea surface dynamic topography produced by Ssalto/Duacs and distributed by Aviso, with support from CNES (downloaded from http://www.aviso.oceanobs.com/duacs), for the same time period. It corresponds to merged data from Topex/Poseidon and ERS satellites. We averaged the original weekly data to monthly resolution, and both data sets were regridded onto the model grid.

2.2. Model Output

[7] The global ocean general circulation model used is Version 4 of the Geophysical Fluid Dynamics Laboratory's Modular Ocean Model [Griffies et al., 2008], forced with the Common Ocean-Ice Reference Experiment (CORE) data set [Griffies et al., 2009; Large and Yeager, 2009]. We use version 2 of the CORE reanalysis effort (CORE2), which includes six hourly interannual varying meteorological fields for the period 1958–2006 (10 m air temperature, humidity, air density, zonal wind, meridional wind, and sea level pressure). The CORE2 winds are based on the 2° resolution NCEP reanalysis data set [Kalnay et al., 1996]. The speed and direction are adjusted using QuikSCAT satellite scatterometer wind vectors following Chin et al. [1998] (Figures 2a and 2b show the wind stress curl field from QuikSCAT data [Risien and Chelton, 2008] and from CORE2 winds). The original resolution of approximately 2° × 2° is interpolated to the model grid. Daily varying shortwave and longwave radiative fluxes are available from 1983 and monthly varying precipitation since 1979; prior to those years, these variables are specified from the climatological annual cycle. Continental runoff, imposed as a climatological annual mean, is part of the CORE2 configuration. It is based on flows at river mouths but highly smoothed at a scale of approximately 4° to give a salinity signature that is seen in climatologies. The ocean model has 50 levels in the vertical direction, a longitudinal resolution of 1° and a latitudinal resolution varying between 1° in the extratropics and 1/3° on the equator. Steric effects (i.e., effects of temperature, salinity, and pressure on seawater buoyancy) are not included in our estimates of sea level variability. The simulation was initialized from hydrographic mean properties taken from the World Ocean Atlas 2001 [Conkright et al., 2002]. It was spun-up for 348 years with forcing from a climatological year calculated using the mean CORE2 data during the years 1958–1977 before the final loop with the 49 years of CORE2 interannual variability was integrated.

[8] The biogeochemical component is given by the model Tracers for Ocean Phytoplankton with Allometric Zooplankton (TOPAZ), which simulates prognostically all major nutrient elements (N, P, Si, and Fe) [Dunne et al., 2010]. The ecosystem is based in three classes of phytoplankton. The small class dominates under nutrient limitation; this size class resists sinking. Large phytoplankton represent diatoms and other phytoplankton that bloom and sink quickly. Finally, diazotrophs fix dinitrogen gas directly. Phytoplankton growth rates are modeled as a function of variable chlorophyll to carbon ratios and colimited by nutrients and light. The model includes river run-off estimates of geographically variable NO3, NH4 and labile dissolved organic nitrogen concentrations from Green et al. [2004] and globally fixed river concentrations for iron, alkalinity and dissolved inorganic carbon. A more detailed description of the biogeochemical model structure can be found in Dunne et al. [2010] and Henson et al. [2009].

2.3. Methods

[9] We use the distance from the coast to the 0.2 mg m−3 isoline of surface chlorophyll (DCHL) to characterize the size of the high chlorophyll region between the oligotrophic subtropical gyre and the African coast. Larger distances indicate a larger area with chlorophyll higher than 0.2 mg m−3. Studies on chlorophyll variability in coastal upwelling systems generally have defined the areas of high productivity as those having >1 mg m−3 [e.g., Nixon and Thomas, 2001; Carr, 2002; Demarcq et al., 2007; Lathuilière et al., 2008], while studies on the oligotrophic subtropical gyres have focused on surface waters having less than 0.07 mg m−3 to limit the study region [e.g., McClain et al., 2004; Polovina et al., 2008]. By studying the area defined by waters exceeding 0.2 mg m−3, we cover an intermediate region between the oligotrophic subtropical gyre and the chlorophyll-rich coastal regime. We repeated the analysis using other chlorophyll isolines that fall within the same region (from 0.07 to 0.3 mg m−3), and found little change in the qualitative and quantitative results reported here. It is not possible to do the same analysis with the 1 mg m−3 isoline, as the model chlorophyll concentrations are lower than the satellite values near the coast, such that there are some latitudes that never see concentrations of 1 mg m−3 over the course of the year (see the supporting information for further details). We evaluate the time evolution of chlorophyll distances as a function of latitude as well as their meridional average.

[10] The temporal variability of SSH is also examined as a function of latitude and as an average over the rectangular study region. The differences in the mean SSH values between model and observations arises from the reference level used (Figures 2a and 2b). Remotely sensed absolute dynamic topography, here simply named satellite SSH, is obtained by adding the sea level anomaly to a mean dynamic topography, calculated using a combination of altimetry and in situ data, and subtracting a model geoid. In the numerical model, the surface of the ocean at rest is taken as the zero level and SSH at a given time step is given as a departure from this level.

[11] We evaluate three physical mechanisms that may cause interannual changes in the size of the high chlorophyll area through variable nutrient supply into the study region: coastal Ekman transport, Ekman pumping, and mixing of subsurface nutrients into the euphotic layer. Coastal Ekman transport and Ekman pumping are not calculated explicitly in ocean general circulation models. Wind stress components are introduced in the momentum horizontal equations, and the vertical velocity (either forced by Ekman processes or not) is a result of any ensuing divergence of the horizontal currents. Therefore, we diagnose the Ekman transports from the wind stress fields forcing the model and compare them to the modeled vertical transports. The vertical volume transport due to coastal Ekman transport (Vcoast, m3 s−1) is calculated as

display math(1)

where τs is the alongshore component of the near-shore wind stress (N m−2, positive northward), L (m) is the coastline length, f is the Coriolis parameter (s−1), and ρw is the water density (kg m−3), taken as a constant value of 1025 kg m−3. Vertical velocities driven by Ekman pumping (we, m s−1) are calculated as

display math(2)

where τ is the horizontal wind stress vector at the sea surface and the subscript z indicates the vertical component of the curl. The vertical transport (Vcurl, m3 s−1) related to this process is obtained by integrating the vertical velocities (we) over the area of the study region (10–24°N, 40°W to coast).

[12] We investigate the relative importance of nutrient supply to the top 80 m in the study region through lateral advection, vertical advection, and mixing. The 80 m depth level is close to the average euphotic layer depth, calculated as the depth of the 1 W m−2 irradiance. Phosphate (PO4) is used for the calculations, although it would be nearly equivalent to use nitrate as their temporal evolution is highly correlated in the model (R = 0.99, p << 0.01, for monthly mean concentrations in the top 80 m over the study region). The phosphorus cycle is simpler than the nitrogen cycle, as the latter includes N2 fixation at the surface and is strongly affected by denitrification. Therefore, fewer terms are required to close the phosphorous budget. Regarding silicic acid, this nutrient does not actually limit growth in the TOPAZ biogeochemical model; rather, it only increases the depth scale of penetration of exported material by increasing the amount of biogenic opal, which acts as a ballast for organic matter. The time rate of change of phosphate (mol m−2 s−1) is given by:

display math(3)

[13] The mixing term in equation (3) encompasses all subgrid-scale processes which result in either vertical or horizontal nutrient fluxes, including the advection due to the parameterization of submesoscale eddies according to the Gent and McWilliams [1990] scheme. Vertical mixing is parameterized in various schemes: a specified background diffusivity with a coefficient of 10−5 m2 s−1 diffuses tracers throughout the domain; the K-profile parameterization scheme from Large et al. [1994] is important in and near the surface boundary layer and represents convective mixing; and parameterizations of tidal mixing from Simmons et al. [2004] and Lee et al. [2006] are important near the sea floor. The last term in equation (3), jPO4, represents the biological sources minus the sinks of PO4, that is, the remineralization of particulate and dissolved phosphorus minus phosphate uptake by phytoplankton. The archived model data allows the calculation of a balanced nutrient budget everywhere except the surface grid cell. At the surface, a term used to filter SSH anomalies to reduce barotropic motions was accidentally excluded from the archived data. As a consequence, about 3% of the PO4 variance in the phosphate time series is not explained by the sum of the budget terms in the surface grid cell only. Also, note that the nutrient budget terms were archived from a shorter simulation (1959–2004) that was in all other ways identical to the simulation used for all other analyses, and therefore years 1958, 2005, and 2006 are missing.

[14] The study region is bounded by latitudes 10–24°N, from 40°W to the coast, in the top 80 m of the water column. The variables studied have different spatial dimensions, that is, chlorophyll distance varies only as a function of latitude, chlorophyll concentrations and SSH vary as a function of latitude and longitude, and PO4 concentrations as a function of latitude, longitude and depth. Throughout this work, all spatial averages and integrations of the variables are made within these boundaries, unless specified otherwise.

3. Satellite—Model Chlorophyll Comparison

[15] To test the model's ability to reproduce the observed spatial and temporal patterns of chlorophyll variability, we compare the distance to the 0.2 mg m−3 isoline of chlorophyll in the model to the satellite data for the period November 1997 to December 2006 (a more elaborate comparison between the model and observations is provided in the supporting information). The modeled climatological year (Figure 3a) shows same seasonality as satellite data (Figure 3d). The overall distances in the modeled climatological year are 400 km greater than those estimated from the satellite, but both model and observations have an averaged amplitude of the seasonal cycle of 780 km. In an average year, the size of the high chlorophyll region is largest in winter and spring and shrinks to its minimum in August and September. In both the model and observations in the vicinity of Cape Blanc (latitudes 18–23°N), the area with chlorophyll above 0.2 mg m−3 occupies an east-west coastal band that is greatest during the first quarter of the year but it remains wider than 400 km throughout the year. In summer, this elevated band near Cape Blanc is limited to a narrow meridional region in the SeaWiFS data, with a gap in the monthly climatology due to the presence of clouds in these months. In contrast, to the south of Cape Blanc, the region of high chlorophyll vanishes completely in summer. The correlation coefficient between the satellite and model chlorophyll distance monthly climatology varies between 0.94 at 13.7°N and 0.73 at 22.7°N, the meridional average correlation being R = 0.86.

Figure 3.

Hovmöller plots (latitude versus time) of the distance between the coast and the 0.2 mg m−3 isoline of chlorophyll (DCHL). (a and d) Monthly climatology, (b and e) monthly means, and (c and f) monthly anomalies, for the period of overlapping satellite and model output (November 1997 to December 2006). Figures 3a–3c show satellite data, while Figures 3d–3f show model output. Gray patches are due to missing data, principally caused by cloud cover. (g) The time series of meridionally averaged (10–24°N) monthly anomalies, after applying a 3 month low-pass filter.

[16] Although the chlorophyll seasonal cycle in this area is very important [Lathuilière et al., 2008], especially in the coastal band, interannual changes represent a significant contribution to the total variability in the eastern subtropical North Atlantic (Figures 1 and 3). In early 1998, the first year of the satellite record (and the corresponding year in the model), there is essentially no area with chlorophyll in excess of 0.2 mg m−3. The biggest positive anomaly of the observational record occurs at the end of 1998 and beginning of 1999: at this time, the 0.2 mg m−3 chlorophyll isoline extends across almost the entire basin, an anomaly of 1000 km over the seasonal mean (Figure 3c). Although the precise mechanism for this event remains largely unknown, Pradhan et al. [2006] noted that it is correlated with the switch from an El Niño to a La Niña. The model reproduces the strong chlorophyll rise and sea level drop between February 1998 and February 1999, but misses the peak values when the chlorophyll was elevated across the width of the whole basin. We interpret the correspondence in the temporal pattern to suggest that the model simulates many of the dynamical responses to atmospheric variability. The mismatch in the peak values leads us to hypothesize a forcing mechanism not included in the model: a likely candidate is the input of iron-rich aerosol dust, as suggested by Pradhan et al. [2006]. As the study region is downstream of the world's largest dust source [Goudie and Middleton, 2006], it is thought that the delivery of dust can have an important impact on productivity and chlorophyll [Duarte et al., 2006] and dust is also known to influence the algorithms that retrieve chlorophyll from the satellite images [Moulin et al., 2001]. Using aerosol optical depth as a proxy for dust (from SeaWiFS data), we found that 1998 was the dustiest year during the SeaWiFS 1997–2007 period (not shown), which might have contributed to the 1998–1999 SeaWiFS satellite anomaly. Our simple analysis of SeaWiFS aerosol optical depth is consistent with a more sophisticated composite time series of tropical North Atlantic aerosols constructed from satellite, in situ, and proxy data by Evan and Mukhopadhyay [2010]. The relationship of aerosols to chlorophyll extends beyond the scope of this work, as our model prescribes wet and dry dust deposition from the monthly climatology of Ginoux et al. [2001] and thus no interannual variability in aerosols exists. The absence of this factor in the model allows us to focus on chlorophyll variability caused by ocean and atmosphere dynamics directly.

[17] Even excluding the very high positive anomaly mentioned above, the width of the high chlorophyll region displays large interannual variability, with the standard deviation of the monthly anomalies equal to 102 km in the satellite data and 144 km in the model output. This interannual variability is about half as large as the seasonal cycle, which has a standard deviation of 260 km for satellite data and 279 km for model data. The simulated distances show the high chlorophyll region being largest from 1999 to the end of 2003 (predominantly positive anomalies, Figures 3f and 3g) and smallest from 2004 to 2006. The SeaWiFS anomalies are dominated by the 1998/1999 event. The anomalies of the high chlorophyll region averaged from 10 to 24°N in the model and SeaWiFS are correlated beyond the 1% level (R = 0.42, Figure 3g). Though significant, this correlation coefficient shows that the model does not simulate a large part of the satellite-observed chlorophyll variability. However, both observed and simulated chlorophyll are linked to sea surface height (Figures 2c and 2d), suggesting that the model realistically simulates some of the physical drivers influencing biology, as explored in the next section.

4. Sea Surface Height and Chlorophyll

[18] Processes that influence SSH are thought to also impact chlorophyll variability. The link between SSH and chlorophyll has been hypothesized on the basis that SSH anomalies reflect anomalies of the thermocline depth [Stammer, 1997; Mayer et al., 2001] and the thermocline and nutricline are often coincident [Wilson and Coles, 2005; Signorini et al., 1999]. Therefore, SSH anomalies may reflect movement of the nutricline toward or away from the euphotic zone with an attendant influence on productivity. Because nutrient observations are sparse in space and time, we can evaluate the nutricline-SSH relationship on interannual time scales only in the model. In the simulation, there is a strong inverse relationship between monthly anomalies of SSH and nutricline depth (defined as the depth of the maximum vertical gradient in NO3) averaged over the study region (R = −0.71). As hypothesized, when SSH is depressed, the nutricline is closer to the surface in the simulation, and we would expect chlorophyll concentrations to be elevated (Figures 2c and 2d).

[19] Both satellite observations and model results substantiate the hypothesized link between SSH and chlorophyll (Figures 4 and 5). Years with negative SSH anomalies generally correspond with positive chlorophyll distance anomalies and vice versa, indicating an expansion of the high chlorophyll region in years of anomalously low SSH. The SSH-chlorophyll relationship is harder to discern in satellite observations because the 1998–1999 anomaly dominates the chlorophyll record more strongly than the SSH record; yet, the correlation coefficient between anomalies of the average chlorophyll distance and anomalies of the average SSH in the study region (R = −0.56) is significant beyond the 1% level. In the model, the relationship is stronger (R = −0.75, Table 1) using monthly anomalies over the 49 year simulation.

Figure 4.

Satellite-derived monthly anomalies of (a) distance from the coast to the 0.2 mg m−3 isoline of chlorophyll and (b) SSH averaged over 10–24°N, from 40°W to the coast, over the SeaWiFS data period (November 1997 to December 2006). (c) The time series of meridionally averaged (10–24°N) anomalies, after applying a 3 month low-pass filter. Note the inverse color axis in Figure 4b and flipped axis for SSH in Figure 4c.

Figure 5.

Modeled monthly anomalies of (a) distance from the coast to the 0.2 mg m−3 isoline of chlorophyll and (b) longitude-averaged (from 40°W to the coast) SSH, over the simulated period (1958–2006). (c) The time series of meridionally averaged (10–24° N) anomalies. Note the inverse color axis in Figure 5b and flipped axis for SSH in Figure 5c.

Table 1. Correlation Coefficients Between Anomalies of the Following Modeled Variables: Distance to the 0.2 mg m−3 Chlorophyll Isoline (DCHL), Chlorophyll Concentration (CHL), Biomass, SSH and Density Difference Between 80 m and the Sea Surface (Δρ); Vertical, Zonal, and Meridional PO4 Supply Into the Study Region (wPO4, uPO4, vPO4, Respectively); and Vertical Transport of Water Into the Study Region Due to Mixing, Ekman Pumping (Vcurl) and Coastal Ekman Transport (Vcoast), the Last Two Calculated Analytically From Wind Stress Data
 CHLDCHLSSHVcurlVcoastwPO4uPO4vPO4MixingΔρ
  1. a

    All correlations are significant at the 1% level.

Biomass0.940.91−0.810.550.600.82−0.570.730.200.18
CHL0.94−0.780.550.600.79−0.620.700.310.13
DCHL0.94−0.730.480.590.77−0.610.720.250.17
SSH−0.79−0.75 −0.67−0.28     

[20] The SSH-chlorophyll relationship is weaker between 1999 and 2006 (Figure 5c). During this time, the anomalously low SSH appears in the northern part of the domain (north of 14°N, Figure 5b). In this northern region, the SSH-nutricline correlation is weaker because the nutricline is deeper and thus a divergence of the surface waters does not easily mine nutrient-rich subsurface waters, a result shown in at least one other study [Wilson and Coles, 2005].

[21] The correlation between SSH and chlorophyll distance is understood to be due to a shared physical mechanism: the divergence or convergence of mass at a given locale, which influences both the temporal evolution of SSH [Griffies and Greatbatch, 2012] and the rate of upwelling. A divergence (convergence) of mass above the nutricline causes upwelling (downwelling) and the vertical heaving of the nutricline toward (away from) the euphotic zone. The wind-driven Ekman currents are a major driver of these upwelling velocities: averaged over the study region, a correlation coefficient of 0.84 between simulated vertical velocities at 60 m and Ekman pumping, calculated from the wind stress reanalysis field via equation (2), confirms a strong wind-driven control on upwelling in the model. In reality, temperature and salinity changes also impact steric SSH, but no straightforward conceptual model exists linking steric sea level anomalies to thermocline/nutricline depth anomalies or chlorophyll. Steric effects are not included in the model diagnostic of sea level, so all variability in model SSH is due to the convergence of the vertically integrated mass transport. The absence of the steric sea level anomalies in the simulated SSH may remove a source of sea level variability that does not systematically influence chlorophyll and be one cause of the stronger SSH-chlorophyll relationship in the model than the observations.

[22] There is no clear trend in the width of the high chlorophyll region over the 9 year satellite chlorophyll record (Figure 4), and previous work noting an apparent expansion of the lowest-chlorophyll North Atlantic waters concurrent with rising sea surface temperatures [Polovina et al., 2008], was likely influenced by the strong 1998/1999 anomaly at the start of the record. In any case, trends observed in the global satellite record have been shown to be small relative to simulated decadal variability [Henson et al., 2010]. The study region is no exception: the subtle downward trend in the size of the high chlorophyll region from 2002 to the end of the simulation is well within the size of the interannual variability and cannot be interpreted as a part of a long term trend (Figure 5). The dominant temporal signal appears to be a multidecadal oscillation, which is strongly linked to SSH. We next quantify the terms giving rise to the biological temporal variability before looking for a cause of this low-frequency modulation in section 'Nutrient Supply in Relation to Climate Variability'.

5. Mechanisms of Chlorophyll Variability

[23] The size of the high chlorophyll region, as defined by the distance between the coast and the 0.2 mg m−3 chlorophyll isoline, is correlated to the mean chlorophyll concentration in the study region (R = 0.97 for model output, R = 0.73 for satellite data, both p << 0.01). Chlorophyll concentrations depend both on biomass and on the chlorophyll to carbon ratio (Chl:C) in the biomass. Here, we choose to focus on the biomass variability (i.e., the concentration of carbon rather than chlorophyll) to exclude any variability caused by changes in intracellular Chl:C. Therefore, all of the model analysis that follows is based on carbon-biomass. In practice, biomass and chlorophyll are tightly related because higher growth rates (more biomass) and higher Chl:C (more chlorophyll) vary in concert in response to increases in iron and light.

[24] Growth rates and biomass are modeled as a function of irradiance, nutrient availability, and temperature. In our study region, irradiance and temperature are not dominant controls on biomass. Prior to 1983, the model irradiance is prescribed as a climatological annual cycle. From 1983 onward, the model uses variable incoming radiation, but we find no correlation to biomass over the latter time period. The temperature dependence of growth in the model would mean that an increase in temperature increases growth rates, all else being equal. However, in our study region, the average temperature in the top 80 m is negatively correlated to biomass; this inverse relationship likely results from the role of upwelling in decreasing temperature at the same time it increases nutrient supply. Because variability in the nutrient supply is primarily responsible for changes in biomass concentration and the size of the high chlorophyll region, we focus our investigation on the different mechanisms of nutrient supply into this region and their relative importance.

5.1. Nutrient Supply

[25] In order to address interannual variability in nutrient supply, we consider each term in the PO4 budget (equation (3)) for the study region, above 80 m, the average depth of the euphotic layer. Studying these terms provides a mechanistic view of the physical controls on phytoplankton biomass. Because a number of previous studies have linked stratification variability with chlorophyll and biomass variability [McClain et al., 2004; Gregg et al., 2005; Behrenfeld et al., 2006; Polovina et al., 2008], we also compare our biomass time series with a common measure of stratification, the density difference between 200 m and the surface. However, as density variations at 200 m are small, this measure of stratification reflects only density variability at the surface, predominantly driven by temperature changes. Higher temperatures at the surface are indeed correlated with lower biomass, but this gives little indication of the physical mechanisms responsible for the decrease. The density difference between the surface and the base of the euphotic zone (80 m) may be a more suitable measure of the stratification impacting the sunlit layer. The anomalies of this measure of stratification are weakly and actually positively correlated with anomalies of biomass (R = 0.18, p << 0.01, Table 1) and surface chlorophyll concentrations (R = 0.13, p <0.01, Table 1), opposite to the expected correlation if stratification were exerting a leading control on the nutrient supply. Instead, this slight positive correlation suggests a distinct mechanism controlling phytoplankton variability that is not suppressed under increasing stratification.

[26] The PO4 budget equation (3) includes lateral advection, vertical advection and mixing. Anomalies of the PO4 mixing term only play a relatively important role in the months of February and March (Figure 6d), when convection and vertical diffusion inject PO4 into the northern part of this region. On an interannual time scale, anomalies of PO4 diffusive mixing averaged over the study region explain less than 4% of the biomass variability (R = 0.2). Conversely, the advective PO4 flux seems to be a key factor (correlation of 0.86 with biomass, explaining 74% of the variability). Among vertical, meridional, and zonal PO4 fluxes, we find that interannual variability of biomass is most highly correlated to the vertical supply of PO4 to the euphotic zone (R = 0.82, Table 1), although correlation to the meridional supply is also high (R = 0.73). Most of the meridional transport of PO4 into the study region takes place through the southern border (95% on a yearly average). This is consistent with waters south of the study region having higher nutrient concentrations than those to the north [Pastor et al., 2012]. Nevertheless, the vertical supply is much larger than the meridional supply both in their anomaly (Figures 6b and 6c) and their monthly mean. The mean peak value over the annual cycle for the vertical upwelling term is 6.34 × 10−10 mol m−2 s−1, and is at least a factor of 2.6 and 10 bigger than the meridional and zonal phosphate supply, respectively. The vertical fluxes are thus the most important supplier of nutrients into this region and the dominant control on both phytoplankton biomass variability and the areal extent of the high chlorophyll region. Interannual variability of these vertical PO4 fluxes are caused by variability in vertical velocities rather than changes in the subsurface nutrient reservoir, as measured by the PO4 concentration at 80 m (Figure 7).

Figure 6.

Time series of modeled monthly anomalies averaged in the study region (latitudes 10–24° N, from 40°W to the coast, top 80 m depth). (a) Distance to the 0.2 mg m−3 chlorophyll isoline, and phytoplankton biomass as the black line, (b) vertical PO4 supply term, (c) meridional PO4 supply and (d) PO4 input due to mixing. Anomalies of the zonal PO4 supply are not shown as their magnitude is much smaller than meridional and vertical supply anomalies.

Figure 7.

Scatterplots of the anomalies of modeled (a) vertical PO4 advection at 80 m (wPO4) and PO4 concentration at 80 m, and (b) vertical PO4 advection at 80 m (wPO4) and vertical velocity at 80 m (w). Each data point corresponds to a monthly anomaly in each grid cell of the study region.

5.2. Coastal Upwelling Versus Offshore Upwelling

[27] Having shown that vertical advection is the leading driver of biomass variability (Figure 6), we next consider two ways in which nutrients are brought to the euphotic layer through vertical advection. One is coastal upwelling, determined in part by the alongshore component of the wind stress. In our study region, the equatorward trade winds generate an offshore Ekman transport of coastal surface waters, leading to nearshore upward velocities that bring subsurface nutrient-rich waters to the surface. The upwelling occurs in a narrow meridional band extending less than 100 km from the shelf break [Barton et al., 1998], with seasonally variable meridional extension. The coastal upwelling further influences the offshore domain where filaments form that transport coastal waters rapidly offshore, as happens when the southward flowing Canary Upwelling Current and poleward Mauritania Current converge [Mittelstaedt, 1991; Pelegrí et al., 2005]. The second vertical nutrient supply mechanism is due to Ekman pumping, driven by wind stress curl that causes divergence of the horizontal flow, and induces vertical velocities at the base of the Ekman layer [Bakun and Nelson, 1991]. The northeast quadrant of the study region, occupied by the subtropical gyre, is characterized by downward curl-driven velocities, whereas the southern part of the domain is characterized by cyclonic wind stress curl and thus upward Ekman suction.

[28] In the study region, the interannual variability of the modeled vertical transports at 60 m in the first cell adjacent to the shelf break is well correlated (R = 0.87) to the variability of the Ekman coastal upwelling calculated using equation (1), confirming that the along-coast wind stress exerts a leading control on variability in coastal upwelling. However, the mean value of upwelling is nearly a factor of 4 smaller in the model (0.4 Sv) than from the theoretical Ekman upwelling (1.6 Sv). The theoretical upwelling values inferred from the steady, two-dimensional, offshore Ekman transport need not match the model's vertical velocities, which result from the divergence of the total horizontal velocity field (the sum of geostrophic, wind-induced, and other ageostrophic nonsteady contributions). A diagnosis of the surface geostrophic velocities from the simulated sea surface height gradients shows convergence of the horizontal geostrophic velocity field (the sum of the along-shore and cross-shore transport) of 0.4 Sv in the first cell along the coast. Such convergence reconciles some of the difference between the simulated vertical velocities along the coast and the Ekman calculation. As a further check that the Ekman transport calculations are reasonable, we also compared the time mean total modeled offshore transport to the sum of the offshore Ekman and geostrophic velocities as in Colas et al. [2008], and found that these agreed to within a tenth of a Sverdrup. The upwelling is also surface intensified in this region, with a mean vertical transport averaged over the first cell adjacent to the shelf break of 0.85 Sv at 20 m. Thus, the mismatch between Ekman and model vertical flux is partly dependent on our chosen depth of analysis. We speculate that the remaining mismatch between modeled and Ekman vertical transport likely arises from convergence of the along-shore transport that is not captured by the geostrophic calculation from the monthly mean sea surface height field, perhaps due to its transient nature as was found by Mason et al. [2012].

[29] Unlike the results for the coastal Ekman upwelling, both the strength of the curl-driven Ekman transport and its variability agree well with the modeled vertical transport averaged at 60 m over the study region. The simulated and theoretical upwelling follow a similar seasonal cycle, peaking in August with values of 3 Sv for model vertical transport and 3.7 Sv for Ekman pumping, and their monthly anomalies are strongly correlated (R = 0.84). This vertical transport supplies nutrients to the euphotic zone. In our simulations, the vertical transport of nutrients to the euphotic zone in the offshore domain (i.e., everywhere in the study region offshore of the first cell adjacent to the shelf break) is more than double the offshore transport of nutrients from the first cell adjacent to the shelf break.

[30] Importantly, coastal upwelling in the model might be weak relative to true upwelling in nature, because of both coarseness in the ocean model grid [Marchesiello and Estrade, 2010; Capet et al., 2004] and the atmospheric reanalysis driving it. Thus, although the simulated coastal upwelling of nutrients influencing the offshore domain is about half the curl-driven pumping of nutrients to the euphotic zone, the relative strength of these terms might be sensitive to model resolution. Because of uncertainty in the model's ability to quantify coastal upwelling, we cannot rule out the possibility that, in nature, the influence of coastal upwelling on the offshore domain might be a more equal partner with the curl-driven Ekman suction in setting the variability in the size of the high chlorophyll region, and we leave this open question to future study with higher-resolution models.

5.3. Size of Upwelling Region Versus Strength of Upwelling Velocities

[31] The importance of offshore upwelling to nutrient budgets over our study area is primarily due to the large area over which the wind stress curl is favorable to upwelling. We next turn our attention to the hypothesis of whether chlorophyll variability is controlled by shifts in the boundary between upwelling and downwelling favorable winds, by changes in the magnitude of the vertical velocities, or a combination of both.

[32] To test this hypothesis, we compare months with anomalously large regions of high chlorophyll (i.e., monthly DCHL more than one standard deviation above the corresponding monthly mean) with those of anomalously small regions (i.e., monthly DCHL more than one standard deviation below the corresponding monthly mean). Figure 8 shows the advective and diffusive terms of the PO4 budget for the mean of the months with anomalously large and small extension of the high chlorophyll region. In both situations, vertical advection of PO4 (wPO4) provides the largest nutrient source. In the high chlorophyll months, as compared to the low chlorophyll months, wPO4 is 2.3 times larger, mixing remains more or less constant and northward transport at 10°N doubles. Upward PO4 transport and northward transport at 10°N are linked through the development of the Guinea Dome, a cyclonic circulation in the southern part of the study region [Mazeika, 1967]. As the dome strengthens, upward velocities increase, as does the northward flow at the eastern flank of the dome, located at about 20°W. The zero wind-stress contour, which sets the boundary between the downwelling-favorable winds of the subtropical gyre and the upwelling-favorable winds of the coastal upwelling and tropical gyre, remains close to coast north of 16°N; south of this latitude, the zero curl contour shifts to the north in the high chlorophyll years.

Figure 8.

Maps showing relevant modeled properties averaged over those months of the time series that have DCHL one standard deviation above their monthly mean (high chlorophyll, left. Sample size n = 85) and those that have DCHL one standard deviation below their monthly mean (low chlorophyll, right. Sample size n = 82). Chlorophyll concentrations (mg m−3) are shown by color, black line is the 0.2 mg m−3 chlorophyll isoline, white line shows the zero wind-stress curl isoline, bold arrows show the advective components of the PO4 budget and curly arrows the mixing term (all components in 10−10 mol m−2 s−1).

[33] While in summer months the boundary between upwelling and downwelling favorable winds is relatively static (Figure 9a), in winter months the contour of zero wind stress curl can shift dramatically: when the extension of the high chlorophyll region is anomalously large during winter months, the area of positive wind stress curl increases by 9 × 105 km2, and this newly positive area adds a total of 60 mol s−1 of PO4 to the photic layer by vertical transport (Figure 9d). On the contrary, when the extension of the high chlorophyll region is anomalously large during summer months, the area of positive wind stress curl increases by only 1.6 × 105 km2 and the newly positive area adds a total of 14 mol s−1 of PO4 to the photic layer by vertical transport (Figure 9c). In summer, the shift in the zero wind stress curl line is small but the overall strengthening of the upward velocities is very important. Thus, the size of the high chlorophyll region is influenced both by the strength of the upwelling and the area over which it occurs.

Figure 9.

Maps showing differences in (a and b) modeled vertical velocities, w, and (c and d) modeled vertical PO4 advection at 80 m, wPO4, between months with high DCHL (months that have DCHL one standard deviation above their monthly mean) and months with low DCHL (months that have DCHL one standard deviation below their monthly mean). Maps in Figures 9a and 9c show the differences during summer months (July to September), maps in Figures 9b and 9d show the differences during winter months (January to March). Dark gray contour marks the zero wind-stress curl isoline for the high months, light gray for the low months in all panels. The study region is enclosed by a black rectangle.

5.4. Nutrient Supply in Relation to Climate Variability

[34] Having demonstrated the dependence of the variability in the size of the high chlorophyll region on the strength of the upwelling of nutrients, it is interesting to investigate whether upwelling is related to a known mode of climate variability. El Niño Southern Oscillation (ENSO) has been suggested to be a remote driver of chlorophyll variability in the eastern tropical North Atlantic. Pradhan et al. [2006] showed a moderate anticorrelation between the Multivariate el Niño Index (MEI) and the satellite chlorophyll anomaly between 10 and 25°N, and 30° to the coast over a 6 year time period during which the satellite data were available (R = −0.52, p <0.01). They hypothesized that the correlation was driven by the influence of ENSO on tropical Atlantic meridional wind stress and coastal upwelling. In qualitative agreement, we find a weak, but statistically significant, anticorrelation between the MEI and modeled chlorophyll variability in our study region (R = −0.39, p <0.01 for the 12 month running mean of the deseasonalized chlorophyll distance anomaly). This correlation is computed from the full time series of the modeled chlorophyll and the MEI produced by NOAA's Earth System Research Laboratory from reanalysis data of the atmosphere (such as pressure, temperature, winds, cloudiness) and ocean (sea surface temperature) [Wolter, 1987; Wolter and Timlin, 1993]. In addition to explaining very little of the variance in our study region, ENSO variability does not have a strong decadal signal, and therefore cannot explain the multidecadal modulation of simulated chlorophyll distances, with the area of high chlorophyll being large in the 1950s–1960s, shrinking in the 1970s–1990s, and recovering to the average after 2000.

[35] The dominant statistical mode of tropical Atlantic variability on an interdecadal time scale is the Atlantic Meridional Mode (AMM). The AMM is characterized by the anomalies of the meridional sea surface temperature (SST) gradient across the mean intertropical convergence zone; these SST gradients are intimately linked to the surface winds over the ocean, which flow toward the anomalously warmer hemisphere [Nobre and Shukla, 1996]. Given that the AMM describes a mode in the surface wind field over our study region, it is natural to hypothesize a connection between AMM and the variability in biological productivity, which we have shown is principally wind-driven. Indeed, the positive AMM has been associated with stronger Ekman upwelling in our study region in a coupled model [Doi et al., 2009]. The AMM is calculated by applying maximum covariance analysis to the zonal and meridional components of the 10 m wind field over the tropical Atlantic (21°S–32°N, 74° W–15° E) after removing ENSO-related variability from the wind field. The updated AMM time series, first described in Chiang and Vimont [2004], has been made available at www.esrl.noaa.gov/psd/data/timeseries/monthly/AMM/, where it is calculated with the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis wind field. A map of the wind pattern is available at this website. Because the CORE2 winds driving our model are derived from the NCEP/NCAR reanalysis with the major modification being a time-invariant speed offset and directional rotation [Large and Yeager, 2009], the modes of variability should be the same for both wind fields. We compare the AMM time series (Figure 10b) to the model vertical velocities at 60 m and to the Ekman pumping from the CORE2 wind stress (Figure 10a). Visually, the time series appear to be at most weakly related at interannual time scales, and this weak relationship is substantiated by a very low correlation coefficient (R = 0.1), which is significant at the 1% level only if we ignore autocorrelation in the individual time series. However, the correspondence between the AMM index and the vertical velocities (and wind-stress curl) appears stronger on longer time scales, as the AMM, like the upwelling velocities and chlorophyll distance, is highest in the 1960s, diminished in the 1970s–1990s, and recovers to the mean in 2000. After filtering the AMM and the model vertical velocities with a 36 month zero-phase filter, the correlation coefficient rises to 0.56 (which is significant at the 5% level, even assuming only 10 degrees of freedom, given that the filtered time series are autocorrelated for 60 months). Because it is the decadal-scale component of the AMM that is correlated with the upwelling velocity and chlorophyll in our study region, our 48 year simulation is barely long enough to capture one cycle of this low-frequency modulation. It would be intriguing to examine the relationship of this mode of variability with biological productivity using a longer integration of a coupled ocean-atmosphere model.

Figure 10.

Time series of (a) anomalies of curl-driven Ekman upwelling (Vcurl, Sv) and model vertical transport at 60 m (Vmodel, Sv), and (b) AMM index from Chiang and Vimont (2004).

6. Summary and Conclusions

[36] The high chlorophyll region along the eastern boundary of the subtropical North Atlantic varies strongly on seasonal and interannual time scales. Such variability is highly correlated to the SSH variability averaged over our study region, supporting the hypothesis that SSH variability in this region is caused in large part by divergent surface currents that lead to vertical upwelling and a shallower thermocline and nutricline.

[37] Using an ocean model forced by reanalyzed atmospheric data, we have investigated possible causes of chlorophyll variability in this region and placed the short satellite record in a longer temporal context. Our major findings are threefold. First, the dominant mechanism controlling interannual biomass variability in this eastern boundary region is the advection of nutrients, mainly vertical advection. Because mixing plays a relatively minor role in the nutrient budget, stratification is not a good indicator of chlorophyll concentration or the size of the high chlorophyll region. Relatedly, mixing plays a relatively minor role in the eastward extension of the oligotrophic region, though a frequent assumption is that the strength of vertical mixing drives variability in the size of the oligotrophic gyre [e.g., McClain et al., 2004; Polovina et al., 2008]. Second, variability in the upwelling of nutrients on interannual to interdecadal time scales is governed by changes in vertical velocities rather than variability in the subsurface nutrient reservoir. Third, the upwelling of nutrients brought about by the offshore wind stress curl is critical in determining the size of the high chlorophyll region. There is a tantalizing, yet difficult to substantiate, relationship between this curl driven upwelling and the large-scale climate forcing of the AMM.

[38] Given that the source of nutrients to this region is their vertical advective supply and that nutrient-containing particles sink to depths of 1000 m or more (see results of sediment trap studies in the nearby Canary Islands [e.g., Sprengel et al., 2000]), the pool of nutrients in the underlying layers would require a long-term input in order to maintain this vertical supply. In these subsurface layers, where the wind-induced vertical transport decreases and vertical diffusion is small, horizontal subsurface currents must transport nutrient-replete waters toward the upwelling region to close the budget [Pelegrí et al., 2006].

[39] The importance of both curl-driven and coastal upwelling has been noted in other eastern boundary regimes. Messié et al. [2009] characterize the curl-driven and coastal upwelling in four of the world's major eastern boundary systems, but restrict their analysis to a region within 150 km of the coast. Even in this narrow coastal band, curl-driven upwelling may supply as much as 30% of the upward volume transport. Here we have shown that the size of the curl-driven upwelling region can be highly variable, but typically extends much beyond 150 km from the shore in the North Atlantic. Both the area over which the wind stress curl favors upwelling and the strength of the resultant vertical velocities critically influence the size of the high chlorophyll region, using a definition of high chlorophyll ranging from >0.07 to >0.3 mg m−3. In agreement with our analysis, Rykaczewski and Checkley [2008] note that the curl-driven upwelling offshore of the California Current System, though weaker than the coastal upwelling, gives rise to a greater overall upward volume transport because it operates over a much greater spatial scale. Moreover, the average size of the cyclonic wind stress and related upwelling region offshore of California is smaller than the average upwelling region in our study region where it can extend across the entire basin (Figures 8 and 9). The wide extension of the curl-driven upwelling domain is a likely reason that the high chlorophyll region of the eastern subtropical North Atlantic extends further offshore than its counterparts in other basins [Carr, 2002]. Understanding the quantitative impact of the coastal upwelling requires observations and simulations that resolve the mesoscale [Lachkar and Gruber, 2011; Gruber et al., 2011], which we leave for future exploration. Evidence that the trade-off between the strength of the coastal and curl-driven upwelling can manifest as variability in the success of different ecosystems [Rykaczewski and Checkley, 2008] provides additional motivation to further explore variability in the strength and spatial extent of both processes.

[40] In conclusion, recent studies have attributed a leading role to stratification in determining the size of the oligotrophic subtropical gyres [e.g., McClain et al., 2004; Polovina et al., 2008], with the view that increased stratification reduces vertical mixing and thus decreases the entrainment of deeper, nutrient-rich waters into the surface layer. While the stratification variability hypothesis has been shown to modulate the poleward extent of the North Pacific oligotrophic gyre [Polovina et al., 2011], in the upwelling region of the eastern subtropical North Atlantic, the correlation between stratification and surface chlorophyll or integrated primary production in the euphotic layer is very weak, because the dominant nutrient supply terms are not due to mixing. Instead, variability in strength of upwelling and the position of the boundary between the upwelling and downwelling domains exert the dominant control on nutrient availability, and stratification variability does not strongly influence the strength of upwelling. Rather, it is the wind that is driving variability in the size of the high chlorophyll domain and, equivalently, the position of the oligotrophic region's eastern boundary. Neither the reanalysis wind stress curl nor the simulated productivity that responds sensitively to this wind shows a clear trend over the second half of the twentieth century.

Acknowledgments

[41] The authors thank three anonymous reviewers and Editor Frank Bryan for insightful comments. The authors also gratefully acknowledge Jorge L. Sarmiento for his useful comments and for hosting M.V.P. at Princeton University. J.B.P. is grateful for funding from NSF's International Research Fellowship Program, NOAA, McGill University, and NSERC Discovery. M.V.P. and J.L.P. acknowledge funding from the Spanish government through projects MOC2 (ref. CTM2008-06438-C02-01) and TIC-MOC (ref. CTM2011-28867). M.V.P. would like to acknowledge Spanish Consejo Superior de Investigaciones Científicas for funding through an I3P grant.

Ancillary