Seasonal Variation in the Correlation Between Anomalies of Sea Level and Chlorophyll in the Antarctic Circumpolar Current

The Antarctic Circumpolar Current has highly energetic mesoscale phenomena, but their impacts on phytoplankton biomass, productivity, and biogeochemical cycling are not understood well. We analyze satellite observations and an eddy‐rich ocean model to show that they drive chlorophyll anomalies of opposite sign in winter versus summer. In winter, deeper mixed layers in positive sea surface height (SSH) anomalies reduce light availability, leading to anomalously low chlorophyll concentrations. In summer with abundant light, however, positive SSH anomalies show elevated chlorophyll concentration due to higher iron level, and an iron budget analysis reveals that anomalously strong vertical mixing enhances iron supply to the mixed layer. Features with negative SSH anomalies exhibit the opposite tendencies: higher chlorophyll concentration in winter and lower in summer. Our results suggest that mesoscale modulation of iron supply, light availability, and vertical mixing plays an important role in causing systematic variations in primary productivity over the seasonal cycle.

described in Supplementary Information). In winter, the signal is not as strong nor as clear 88 as in the summer, which may be due to the lower density of wintertime observations and 89 generally lower net primary productivity. 90 To identify the mechanisms by which mesoscale processes in the ACC influence

127
In winter, light is the primary factor limiting productivity throughout the whole wa-128 ter column and iron limitation is of diminished importance (Fig. 3b). Since light is sup-129 plied at the surface and attenuates with depth, mixed-layer mean light (or photosyntheti-130 cally active radiation; PAR ) declines with increasing MLD. We find that PAR is nega-131 tively correlated with SSH throughout the year (Fig. 2d) and is approximately 30% lower 132 in positive SSH anomalies than negative ones in winter (Fig. 4c) and about 7% lower in 133 summer (Fig. 4a). Hence deeper mixing in positive SSH anomalies decreases PAR ex-134 perienced by phytoplankton in the mixed layer; it is this effect that has the potential to 135 explain lower wintertime CHL in positive versus negative anomalies in SSH (Figs 3a,d).

136
In contrast, productivity is iron limited in the summer (Fig. 3b). The model simu-137 lation shows that positive SSH anomalies in the ACC have 30% more Fe than negative 138 SSH anomalies in winter (Fig. 4c) and approximately 15% more in summer (Fig. 4a). A 139 budget analysis for Fe (Figure 4b,d; see Supplemental Information for details) quanti-140 fies the various mechanisms of iron supply and removal. Iron is supplied to the mixed 141 layer by lateral and vertical advection, vertical mixing, aeolian input of dust, and entrain-142 ment associated with changes in the MLD; it is consumed by phytoplankton uptake and 143 removed via scavenging on sinking particulates. Among these processes, we find that the 144 supply of iron by vertical mixing differs most between positive and negative SSH anoma-145 lies. Supply of iron by vertical mixing in positive SSH anomalies has a median value that 146 is roughly 10% higher than in negative SSH anomalies when normalized by the mean between positive and negative SSH anomalies are negligible. The biogeochemical sink is 150 the largest in summer with slightly more loss of iron in positive SSH anomalies due to 151 higher phytoplankton productivity.

152
The contribution from advection, including eddy-driven lateral advection via trap-   We report a higher level of iron in positive SSH anomalies along the ACC in both 205 seasons. A possible explanation for higher iron in positive SSH anomalies in summer is 206 preconditioning during winter. In winter when the MLD modulation is particularly intense, 207 deeper mixed layers in positive SSH anomalies have considerably higher iron concentra-208 tion than the shallower mixed layers of negative SSH anomalies. Anomalously high iron 209 in positive SSH anomalies in winter is not heavily used due to the lack of sunlight, and 210 subsequently may promote elevated primary productivity in summer. The mixed layer 211 shoals rapidly after winter, but vertical mixing at the base of the mixed layer continues 212 to entrain iron from the layer that was previously in the mixed layer during winter. Hence 213 positive SSH anomalies have relatively iron-rich water below the summertime mixed layer 214 compared to negative SSH anomalies, and this may contribute to differences in iron lim-215 itation among the two types of features. This explanation can be applied to well-formed 216 and long-lived SSH anomalies whose lifespans are of the order of months (eddies).

217
Our study defines eddies based on SSH anomalies (> 5 cm) instead of the closed        Figure 3. (a,b) The median vertical profiles of light limiting factor (marked by "V light " in orange) and iron limiting factors for positive SSH anomalies (anticyclones marked by "V Fe ae ") and negative SSH anomalies (cyclones marked by "V Fe ce ") in the biogeochemical model averaged in the ACC. In summer, iron is the limiting factor for the primary productivity within the mixed layer, however, light limits primary productivity more below the mixed layer (a). The magnitude of limiting factor is inversely related to its' affect on primary productivity, e.g., light is more important than iron concentration throughout the whole water column in win- shading represents the iron concentration and sunlight intensity, respectively. In summer, primary production is controlled by iron supply (blue arrows) and not light in the mixed layer (d). In winter, intensive vertical mixing enriches iron concentration near the surface, but low light availability limits primary production (e).

Model Simulations
The mechanisms generating observed ρ SS H ,C H L were examined using the Biogeochemical Elemental Cycling (BEC) model [Moore et al., 2002[Moore et al., , 2004[Moore et al., , 2013 coupled to the ocean circulation component of the Community Earth System Model (CESM) with a resolution of 0.1 • (less than 10 km in zonal direction in the Drake Passage). The total chlorophyll concentration is computed as the sum of chlorophyll of three phytoplankton functional groups whose biomasses are affected by the uptake of varied nutrients including iron, and grazing by zooplankton. The vertical mixing is estimated by the K-Profile Parameterization (KPP) mixing scheme [Large et al., 1994] and we treated the depth of planetary boundary layer as MLD in the analysis.
The model was integrated for 5 years archiving 5-day means. The simulated SSH anomalies were computed by removing the spatial mean of 4×4 degree grid boxes. The procedure for the chlorophyll anomaly computation follows that used in the satellite data but from simulated 5-day mean total chlorophyll concentrations. For Figures 1 and 2, the solutions were mapped to the same grid as the satellite SSH anomalies before computing the correlation.

Nutrient limitation
Phytoplankton growth rates in the ocean biogeochemistry component of the CESM are computed as where µ i is the C-specific growth rate (d −1 ) for phytoplankton functional type (PFT) i, µ i,r e f is the maximum growth rate (referenced to 30 • C), and T f is the temperature limitation ("Q10") function; V i and L i are the nutrient and light response functions, respectively [Geider et al., 1998]. For diatoms (diat) and "small" phytoplankton (sp), the nutrient response function follows Liebig's law of the minimum, such that the ultimate limitation term used to compute growth is that of the most limiting nutrient:

Evaluation of the Model Simulation
The eddy-rich 0.1 • CESM has the sea surface height (SSH) variability that agrees well with the one estimated from space. Figure S1(a) shows the standard deviation of SSH in the observation from Collecte Localis Satellites (CLS/AVISO) that we used in this study. The regions of elevated SSH variability include the Antarctic Circumpolar Current (ACC), Brazil-Malvinas Confluence, Agulhas Current retroflection and East Australian Current. The model simulation captures not only the spatial pattern of elevated SSH variability but also the magnitude of it ( Fig. S1(b)). The simulated chlorophyll (CHL) at the surface agrees with SeaWiFS observation to a somewhat lesser degree than SSH variability. Although it underestimates the CHL concentration, the model generally shows similar patterns of high CHL as satellite observations. The simulated iron also captures a large scale iron distribution in observations (Fig. S2).
Iron is the limiting nutrient in the south of ACC for all phytoplankton types in the model, hence the information of the iron meridional gradient is important to understand the chlorophyll variability associated with the mesoscale. In summer (January-March), the Indian-Pacific sector has little meridional gradient as iron is depleted by active primary production, which makes it unlikely that lateral advection such as through stirring and trapping are driving the iron anomaly in Fig 4a. In the Atlantic sector, however, iron concentration generally increases equatorward (Fig. S3(top)). Consistent with in situ observations [Bowie et al., 2002], iron is supplied from the Patagonian Shelf to the northern ACC creating a meridional gradient in the model. Also, the dust input from the atmosphere is the greatest in the Atlantic sector [Luo et al., 2008]. With increasing iron concentrations toward equator, the eddy-driven advection, as well as vertical mixing, can create iron perturbations associated with eddies.
In winter (July-September), the surface ocean features a higher iron concentration than in summer. The primary productivity is more regulated by the light availability, nevertheless the wintertime iron distribution shows how closely iron is linked to the vertical mixing, especially in the Indian-Pacific sector. There, deeper vertical mixing enriches the surface ocean with iron as indicated by the fact that higher concentration of iron collocates with the region of relatively deep mixed layers (> 50 m) (Fig. S3(bottom)). The mixed layer depth in the Indian-Pacific sector is spatially inhomogeneous, so there are larger horizontal iron gradients than in summer. As a result, it is more likely that the eddy-driven advection sets the perturbations in iron, but vertical mixing modulation becomes larger than in summer at the same time. The Atlantic sector shows a meridional gradient of iron that is not different from summer.
The budget of the iron averaged over mixed layer We analyzed the budget of iron ( f e(z,t)) averaged over the mixed layer where η(t) is the sea surface height and h(t) is the time-varying depth of the boundary layer determined by the KPP mixing scheme. The temporal evolution of the iron averaged over the mixed layer ( Fe ≡ 1 h(t ) f e(z,t)dz) can be written as The first and second terms represent the mean tendency of iron in the mixed layer and the contribution by entrainment/detrainment, respectively.
The f e(z,t) tendency in the model is computed as follows.
where A h is the horizontal advection, w is the vertical velocity, κ(z,t) is the vertical diffusivity, F (z,t) is the surface iron flux (nonzero only at the surface) and B(z,t) is the biological source/sink term. Using (5), (4) can be written as with the surface value of the diffusivity, κ(η(t),t) = 0.
We constructed distributions of Fe , PAR and the terms in (6) at the locations of anticyclones and cyclones in summer and winter along the ACC. Since not all distributions are normally distributed (e.g. Fe ), the median is used as a representative measure of the distributions. The medians of Fe and PAR distributions for anticyclones and cyclones in the ACC are first obtained. Then the differences in medians between anticyclones and cyclones are normalized by that appropriate to the entire ACC. In Figures 4a,c, we plot the percent value of the median differences.
The terms in (6) are normalized by the median of Fe along the entire ACC after being multiplied by 10 days, the time interval used in the tendency equation in the model. We then compare the medians of each term's distribution to quantify the systematic differences between anticyclones and cyclones in Figures 4b,d in the main article. The 95% confidence intervals are estimated using bootstrapping and are very close to the median itself due to a large sample size, hence they are not plotted in Figure 4. It is noted that the advection term includes all advective processes in the region whose absolute SSH anomaly exceeds 5 cm. Hence, it does not solely represent the advection by coherent eddy structures through stirring or trapping.