Imprint of Southern Ocean eddies on chlorophyll

Abstract. Although mesoscale ocean eddies are ubiquitous in the Southern Ocean, their spatial and seasonal association with phytoplankton has to date not been quantified in detail. To this end, we identify over 100,000 eddies in the Southern Ocean and determine the associated phytoplankton biomass anomalies using satellite-based chlorophyll-a (Chl) as a proxy. The mean eddy associated Chl anomalies (𝛿Chl) exceed p10 % over wide regions. The structure of these anomalies is largely zonal, with cyclonic, thermocline lifting, eddies having positive anomalies in the subtropical waters north of the Antarctic Circumpolar Current (ACC) and negative anomalies along the ACC. The pattern is similar, but reversed for anticyclonic, thermocline deepening eddies. The seasonality of 𝛿Chl is weak in subtropical waters, but pronounced along the ACC, featuring a seasonal sign switch. The spatial structure and seasonality of 𝛿Chl can be explained largely by lateral advection, especially eddy- stirring . A prominent exception is the ACC region in winter, where 𝛿Chl is consistent with a modulation of phytoplankton light exposure caused by an eddy-induced modification of the mixed layer depth. The clear impact of eddies on phytoplankton may implicate a downstream effect on Southern Ocean biogeochemical properties, such as mode water nutrient contents.

regions and at different times, and that the different polarity of the eddies tends to induce signals of opposite sign (Denman and Gargett, 1995;Lévy, 2008;McGillicuddy, 2016).
Lateral advection arising from stirring of eddies has been argued to be a major driver globally. The argument is based on the observed correlation of the magnitude of eddy-associated chlorophyll anomalies, δChl, and the larger-scale Chl gradient ambient to eddies (Doney, 2003;Uz and Yoder, 2004;Chelton et al., 2011a;O'Brien et al., 2013). Further, it has been suggested 5 that advection of Chl by eddies via trapping, i.e., the enclosing and dragging along of waters, causes δChl (Gaube et al., 2014), particularly in boundary current regions characterized by steep zonal Chl gradients. Numerous other potential mechanisms through which eddies affect phytoplankton have been identified (e.g., McGillicuddy et al., 2007;D'Ovidio et al., 2010;Siegel et al., 2011;Gaube et al., 2013Gaube et al., , 2014Dufois et al., 2016;Gruber et al., 2011), including modifications of mixed layer depth, vertical mixing, thermocline lifting, and providing of spatial niches. These mechanisms modulate the phytoplankton's light 10 exposure, their nutrient availability or their grazing pressure, i.e., they affect their net balance between growth and decay. Thus, in contrast to the physical mechanisms of stirring and trapping where phytoplankton is being advected merely passively, these mechanisms create eddy-associated phytoplankton biomass anomalies by altering their biogeochemical rates.
Here, we aim (i) to provide a reference estimate of the long-term mean chlorophyll anomalies associated with eddies in the different regions of the Southern Ocean, distinguishing anticyclones and cyclones, and (ii) to discuss the mechanisms likely 15 causing the observed imprint. The Southern Ocean is a region rich in eddies (e.g., Frenger et al., 2015) and important for setting the global distribution of biogeochemical tracers (Sarmiento et al., 2004). Previous studies used eddy kinetic energy as a proxy for eddy activity rather than sea level anomalies (SLA), which does not allow a distinction by polarity of eddies (Comiso et al., 1993;Doney, 2003), did not focus on the Southern Ocean (Chelton et al., 2011a;Gaube et al., 2014), or lacked a discussion of the seasonality of the relationship. 20 Our approach is to identify individual eddies based on satellite estimates of SLA and combine those with satellite estimates of Chl (Chelton et al., 2011a;Gaube et al., 2014). We discuss possible mechanisms playing a role based on large-scale Chl gradients (Doney, 2003;Chelton et al., 2011a;Gaube et al., 2014) and the local shape of the average imprint of eddies on Chl (Chelton et al., 2011a;Gaube et al., 2014;Siegel et al., 2011).
2 Methods and data 25 We first introduce our analysis framework before describing the methods and data sources. This permits us to explain the approaches we use to assess the potential mechanisms explaining the δChl associated with Southern Ocean eddies.

Analysis framework
Fundamentally, eddies can lead to local phytoplankton biomass anomalies through either advective processes, i.e., the spatial reshaping of existing gradients, or through biogeochemical fluxes and transformations that lead to anomalous growth or losses 30 of biomass. In the following, we present these potential mechanisms in more detail, and how we estimate their importance.

Causes of δChl by advective processes
Eddies may cause δChl as they laterally move waters, i.e., advect waters including their Chl characteristics. This mechanism may lead to δChl if (i) a lateral Chl gradient is present that is sufficiently steep at the spatial scale of the eddy-induced advection (Gaube et al., 2014), and (ii) the time scale of advection matches the time scale of the growth and loss of Chl (Abraham, 1998).
The time scale of Chl is order of days to weeks, possibly months, with the lower boundary representing roughly the reactivity 5 time scale of Chl governed largely by the growth rate of the phytoplankton, and the upper boundary the potential maintenance of Chl concentrations via recycling of nutrients within the mixed layer. Concerning the spatial scale of advection by eddies, we distinguish two effects, labeling them stirring and trapping.
With stirring, we refer to the distortion of a large-scale Chl gradient due to the rotation of an eddy, as illustrated in Figure   1a (left column, with black arrows indicating the eddy rotation and associated advection), see also Siegel et al. (2011). The 10 turnover time scale associated with the rotation of eddies is order of days to a few weeks which matches the time scales of Chl. The spatial scale of stirring is given by the spatial extent of an eddy and is somewhat larger than the eddy core, as defined based on the Okubo-Weiss parameter (Frenger et al., 2015), i.e., several tens to several hundred kilometers.
Next to stirring, eddies may advect material properties due to their intrinsic lateral propagation (Figure 1a, right column).
We refer to the ability of eddies to transport fluid along their propagation pathway in their core as trapping. The time scale 15 of trapping is given by the typical lifespan of Southern Ocean eddies which is weeks to months (Frenger et al., 2015), i.e. it may match the longer time scale of Chl. Propagation speeds are small (an order of magnitude smaller than rotational speeds) which implies that the majority of eddies tends to die before they can propagate far. The fraction of very long-lived eddies that propagate over distances exceeding a few hundred kilometers is small (Frenger et al., 2015).
A necessary condition for trapping to happen is that the eddies' swirl velocity is larger than their propagation speed (Flierl,20 1981), a condition generally met for mid-to high-latitude eddies (Chelton et al., 2011b). Indeed, observations of eddies carrying the signature of their origin in their cores support the trapping effect (Bernard et al., 2007;Ansorge et al., 2009;Lehahn et al., 2011), and so does the modeling study by Early et al. (2011). Yet, likely only few eddies are truly efficient in their trapping (Beron-Vera et al., 2013;Haller, 2015). They tend to continuously exchange some fluid with their surroundings, i.e., their trapping is limited by them being permeable. Nevertheless, we expect eddies to be able to drag along some entrained waters 25 for some time, hence displacing these waters for some distance as they propagate. This may be sufficient to displace waters from e.g., the south to the north of an ACC front along an intense Chl gradient, leading to δChl through (permeable) trapping.

Causes of δChl by biogeochemical processes
Eddies affect the biogeochemical/physical properties that control the rates of biogeochemical processes in their interior through many mechanisms. These include, e.g., the stimulation of phytoplankton growth through enhanced nutrient concentrations or 30 increased average light levels, or the modification of predator-prey encounter rates, affecting the mortality of phytoplankton cyclones (bottom row); a shows the effects of advection (lateral displacements) of Chl due to the eddies' rotational speed (stirring, left column) and lateral propagation (trapping, right column); trapping and stirring can cause δChl of either sign, depending on environmental Chl gradients; b shows multiple potential effects eddies may have on Chl by affecting biogeochemical processes, including modification of nutrient supply and light exposure through thermocline lifting or deepening (circle/cross) and modified vertical mixing (wiggly lines), and eddy current-wind interactions (black and thick white arrows), in turn causing thermocline displacements; the local shape of δChl is anticipated to look different depending on the mechanism active, i.e. a monopole δChl is expected for all eddy-effects except for stirring where an asymmetric dipole is excepted (Figure inspired by Siegel et al. 2011, Figure 1).
lifting cyclones (Falkowski et al., 1991, indicated as black circle in Figure 1b) has been challenged to be a major player (Oschlies, 2002), and multiple other mechanisms have been proposed how eddies may affect phytoplankton biomass. These include a modification of vertical mixing through changes of stratification (wiggly lines in Figure 1b) and eddy current-wind interactions causing thermocline displacements (eddy swirl currents and winds are indicated as black and thick white arrows in  Mahadevan et al., 2008;Siegel et al., 2011;Xiu et al., 2011;Lehahn et al., 2011;Boyd et al., 2012;Mahadevan et al., 2012;Dufois et al., 2016). The prevailing lack of data of sufficiently highly temporally resolved sub-surface observations hampers a systematic large-scale observationally-based assessment of the role of effects of eddies on biogeochemical processes.

Assessing mechanisms causing δChl
We employ two sets of approaches to assess the mechanisms causing δChl. In the first set of approaches, we diagnose whether the environmental conditions are actually met for supporting a major contribution of a particular set of mechanism. Namely, we assess if lateral Chl gradients sufficiently support advective effects of eddies to explain δChl.
In the second set of approaches we diagnose the spatial pattern associated with eddies as this spatial signature tends to differ 5 between the two major sets of processes, i.e., advective processes and biogeochemical rates (Siegel et al., 2011). Eddies that are associated with, for instance, stirring are anticipated to have a dipole shape (Figure 1a, left column), as they distort the underlying gradient field, with the rotation of the eddy determining the orientation of the dipole. In contrast, most mechanisms associated with modifications of the biogeochemical rates cause a monopole pattern, irrespective of polarity ( Figure 1b). This is a consequence of the δChl tending to be caused by anomalies in the nutrient supply or light availability, which are altered 10 inside eddies in a radially symmetric manner. Also the trapping mechanisms tend to cause a monopole pattern of δChl ( Figure   1a, right column), but they can be distinguished from the rate-based mechanisms by either their history, or their tendency to trap the anomalies very tightly in the inner domain of the eddy. The rate based mechanisms, in contrast, often have monopole patterns that extend more broadly over the eddy, or are, in certain case, even strongest at the edges (McGillicuddy, 2016). Here, we suggest effects on biogeochemical rates due to eddies to play a role in regions and seasons where the potential for advective 15 effects is insufficient to explain the observed eddy-induced δChl, i.e., we diagnose them largely as a residual.
Some complexity is added in the interpretation of the spatial pattern by the fact that the asymmetry of the dipole pattern arising from stirring causes also a monopole pattern ( Figure 1a). Such an asymmetry was suggested by Chelton et al. (2011a) to arise from the westward propagation of eddies and the leading (mostly western) side of an eddy affecting upstream unperturbed waters, resulting in a larger anomaly at the leading than the trailing side of an eddy, with the latter stirring already perturbed 20 waters. Also, the eddy may entrain some of the westward upstream waters into its core, labeled here lateral entrainment or permeable trapping (Hausmann and Czaja, 2012;Frenger et al., 2015). Indeed, averaged over an eddy's core, stirring will only cause a net anomaly if the dipole associated with stirring is asymmetric. It is not obvious how to quantify such an asymmetry.
Independent of its asymmetry, we will qualitatively discuss the potential maximum δChl induced by stirring.
We note that advection by an ambient larger-scale flow does not affect the stirring mechanism. For instance, the Antarctic 25 circumpolar flow in the Southern Ocean makes eddies propagate eastward in an Eulerian sense, nevertheless they propagate westward in a Lagrangian sense, relative to the ACC and ambient Chl.

Data
To assess the relationship between ocean eddies and Chl anomalies, we use the data set of Southern Ocean eddies and their characteristics as derived and described in detail in Frenger et al. (2015). The data set contains about 1,000,000 snapshots of For Chl we use the merged ESA GlobColour Project product (http://www.globcolour.info, case-1 waters, merged according to Maritorena and Siegel (2005) with a spatial and temporal resolution of 0.25 • and one day, respectively. We choose a merged product for Chl as the merging doubles the spatial coverage of the daily data in the Southern Ocean, on average (Maritorena et al., 2010). Of the data of the up to three available sensors, i.e., SeaWIFS (SeaStar), MODIS (Aqua) and MERIS (Envisat),

5
SeaWIFS generally features the best spatio-temporal coverage, but its contribution drops below 40% in high latitudes and partly in the western ocean basins of the Southern Hemisphere. For these areas after 2002, SeaWIFS data were complemented with MODIS as well as MERIS data. We average the Chl data to weekly fields to match the temporal resolution of the eddy dataset.
To examine δChl of eddies, we compare the Chl averaged over their core to background fields of Chl. For the latter, a monthly 10 climatology of Chl proved not to be appropriate due to high spatio-temporal variability of Chl unrelated to eddies. Hence, we obtain the background fields the following way: we apply a moving spatio-temporal Gaussian filter (Weierstrass transform, spatial filter similar to e.g., Siegel et al. 2008, with 2σ of 10 boxes/∼200 km at 45 • S, 8 boxes/∼200 km and 1 week in longitudinal, latitudinal and temporal dimensions, respectively) to each of the weekly Chl fields. We then subtract the resulting from the original fields to produce δChl fields. The δChl fields are not sensitive to the selected σ. The choice of a rather small 15 filter makes δChl amplitudes smaller compared to if a larger filter is chosen, producing a conservative estimate of δChl. In order to generate spatial maps of δChl, we averaged all eddy associated anomalies of the respective eddy polarity into 5 • × 3 • longitude/latitude boxes.
Prior to all analysis we log-transform Chl, due to Chl being lognormally distributed (Campbell, 1995). δChl is frequently given in percentage difference relative to the background Chl as 20 δChl = exp log(Chl e )-log(Chl bg ) − 1 × 100 = Chl e Chl bg − 1 × 100 with subscripts e and bg denoting eddy and background, respectively. Where we show absolute δChl on a logarithmic scale, we use the base 10 logarithm.
Regarding the spatial, i.e., geographical analysis, we use on the one hand the positions of the main ACC fronts (Polar Front, PF, and Subantarctic Front, SAF) as determined by Sallée et al. (2008). On the other hand, we make use of a climatology of 25 sea surface height (SSH) contours (Maximenko et al., 2009), which are representative for the long-term geostrophic flow in the area. The mean positions of the PF and SAF align approximately with the mean SSH contours of about -40 cm and -80 cm, respectively. We select the -20 cm SSH contour to separate waters of the southern subtropical gyres to the north of the ACC, referred to as subtropical waters from waters in the "ACC influence area", referred to as ACC waters. This choice is based on both, a tendency for net eastward propagation of eddies south of this contour (Frenger et al., 2015) indicating advection 30 by the ACC mean flow, and a seasonal sign switch of δChl, which will be addressed later in the paper. Waters south of the PF/-80 cm SSH we refer to as Antarctic waters. Finally, we use mixed layer depths derived from Argo floats, available at http://www.locean-ipsl.upmc.fr.

Analysis of environmental Chl conditions
Using the data presented in the previous section, we calculate a monthly Chl climatology. Based on this climatology we derive potential δChl (δChl) eddies may induce due to lateral advection ( Figure 1a): in order to assess theδChl emerging from stirring in the Southern Ocean, we compute the climatological meridional Chl gradient at the spatial scale of individual eddies, here taken as two eddy radii (δChl stir ). To assessδChl emerging from trapping, we estimate the Chl variation along individual 5 eddies' pathways by computing the difference of the climatological Chl at the location of an eddy origin and the climatological Chl at the present location of the eddy at the present month (δChl trap ).

Analysis of the spatial shape of δChl
We compute the composite spatial shape of Chl and δChl associated with eddies the same way as done in Frenger et al. (2015) for sea surface temperatures: we extract a squared subregion of side lenghts of 10 eddy radii for each individual eddy from 10 the weekly maps of SLA and Chl, centered at the eddy center. We rotate Chl snapshots according to the ambient Chl gradient and average over all eddies to produce the eddy composite. Note that the estimate of the magnitudes of the dipole and the ambient Chl gradient (see below) tend to be slightly weaker without rotation. Nevertheless, as averages are taken over regimes of largely similar orientation of the ambient Chl gradient (see Discussion section 4), our conclusions do not depend on whether we rotate snapshots or not. 15 Further, we decompose the eddy-induced spatial average δChl pattern into a monopole (MP) and dipole (DP) pattern by first constructing the monopole by computing radial averages of δChl around the eddy center, i.e., δChl(r) MP = δChl(r), where r is the distance from the eddy center. In the second step, we calculate δChl DP as a residual, i.e., by differencing the monopole pattern from the total signal. Even though this residual approach captures in the dipole structure any non-monopole pattern, experience has shown that the δChl DP typically feature dipoles (Frenger et al., 2015). In the final step, we quantify the 20 amplitudes of the monopoles and the dipoles, assess the contribution of the two components to the spatial variance of the total signal based on the sum of variances (var), i.e. var(δChl) = var(δChl MP ) + var(δChl DP ), and compute the local Chl gradient at the scale of two eddy radii, as an estimate of the potential maximum contribution of stirring to δChl.

Handling of measurement error and data gaps
As the error of the satellite retrieved Chl for each individual eddy can easily be as large as the anomaly, an individual eddy 25 signal may be undetectable with in-situ measurements (Siegel et al., 2011). The significance of our results arises from the large number of analyzed eddies. A reduction of the sample size due to missing Chl data arising from frequent cloud cover in the SO may affect the significance: On average for 45% of the eddies' Chl data was entirely missing. Missing values due to cloud cover-only (leaving aside missing data due to the polar night) increase from 20% at 30 • S to 60% at 65 • S. Anticyclones exhibit a higher percentage of data gaps than cyclones (47% versus 42%), which can be explained by the impact of their sea surface 30 temperature anomalies on cloud cover (Park et al., 2006;Small et al., 2008;Frenger et al., 2013). We account for the issue  Figure S1) for both anticyclonic (warm-core, SLA lifting and thermocline deepening) and cyclonic (cold-core, SLA deepening and thermocline lifting) eddies. The overall mean δChl associated with anticyclones is -4 %, while that for cyclones is of even smaller magnitude, i.e., +1 %. The distributions around these small means are very broad, however, with many anticyclones and cyclones having both, positive or negative δChl, depending on the region and time of the year. 10 The long tails of the distributions, corroborated by visual inspection of the individual δChl of eddies suggest anomalies that are substantially larger than the mean. Thus, it appears that by averaging the signals in time and space, a substantial amount of information is lost. As a consequence, it is more insightful to disentangle the signals and to examine the regional and seasonal variation of δChl.
3.1.2 Spatial variability of imprint 15 The maps of the annual mean imprint of cyclonic and anticyclonic eddies on Chl clearly supports this hypothesis of a strong regional cancellation effect ( Figure 2). First, the regional mean signal associated with eddies is indeed much larger than sug- west of New Zealand, and more subtly east of Kerguelen Islands and the Drake Passage (see also Sokolov and Rintoul 2007), where δChl tends to be positive for both, anticyclones and cyclones.

Seasonality of imprint
The pronounced zonal bands of δChl for anticyclones and cyclones persist over the year, but tend to migrate meridionally This becomes even more evident when inspecting the zonally averaged Chl and δChl as a function of season and SSH, i.e., plotted in the form of a Hovmoeller diagram (Figure 4). Along the ACC, anticyclones exhibit negative δChl in winter to spring concurrent with deep mixed layers, followed by positive δChl in summer to autumn (Figure 4b). Cyclonic δChl patterns are opposite, featuring negative δChl in spring to autumn, with close to zero to positive δChl in winter to spring (Figure 4c). This 5 implies that SLA and δChl are positively correlated summer to autumn, followed by a negative correlation in winter to spring.
This sign switch of the correlations shows a seasonal lag towards Antarctic waters, with positive correlations prevailing autumn to winter, and negative correlations prevailing spring to summer, resulting in the aforementioned apparent southward migration of the sign switch of the seasonality of δChl over the course of the year. In the northern domain, i.e., in subtropical waters (here SSH larger -20 cm) the sign ofδChl stir tends to agree with δChl throughout the year for both anticyclones and cyclones (Figures 4b-e). So does the seasonal variation of the magnitude of δChl stir , with the largest magnitudes found in summer to autumn. Also the regional variations match, such as a weakerδChl stir 20 and δChl in the Pacific sector compared to the Atlantic and Indian Ocean sectors (Figure 3, middle/right columns and Supplementary Figure S2, left column).
Also, along the ACC and its northern flank in summer to autumn,δChl stir and δChl agree in sign, and are of the same order of magnitude. Finally, along the southern ACC and in Antarctic waters,δChl stir mirrors the seasonal sign switch of δChl, and the apparent seasonal southward migration of the zonal bands of δChl (Figures 3 and 4b,c). Thus, it appears that stirring can 25 already explain a good fraction of the observed δChl (i) in subtropical waters outside of those characterized by winter deep mixed layers, (ii) along the ACC and its northern flank in summer to autumn, and (iii) south of the ACC.
That stirring has the potential to produce the observed δChl in large parts due to a sufficiently large average ambient gradient of Chl, is corroborated by the actual observed local shape of δChl (Method section 2.4). For instance: averaged over eddies in subtropical waters in the northern domain in winter to spring, the average absolute gradient at scales of two eddy radii is  that stirring alone may largely explain observed δChl (Figure 6a; anticyclones: gradient of 9 % and maximum δChl of 5 %; cyclones: gradient of 9 % and maximum δChl of -11 %; and Figure 5b; anticyclones: gradient of 5 % and maximum δChl of -6 %; cyclones: gradient of 5 % and maximum δChl of 5 %).
The advective potential for the other lateral advective mechanism, i.e., trapping,δChl trap , partly opposes and partly enhanceŝ δChl stir (Figures 4d-g). For instance, for cyclones along the ACC in summer to autumn, trapping possibly contributes to a δChl 5 (11 %) signal that is slightly larger than the Chl gradient at two eddy radii (9 %), and the contribution of the variance of the monopole is increased compared to anticyclones (Figure 6a, 96 % versus 87 %). Yet, overall the trapping potentialδChl trap is weak compared to δChl (Figure 4b,c,f,g), and outweighed byδChl stir .

Biogeochemical rates
Even though advective processes and particularly stirring appear to be the dominant driver for the eddy-associated chlorophyll       The zonal pattern of the eddy induced Chl anomalies, δChl, identified here for the Southern Ocean is similar to that seen in global ocean-based analyses (Gaube et al., 2014). Also the magnitude of δChl is similar north of the ACC to what Gaube et al. (2014) reported on a global basis. Yet, along the ACC we find more widespread and more intense δChl than the global study, especially in summer and autumn. Regional variations in the sign of δChl associated with either cyclonic and anticyclonic 5 eddies have been reported previously (Gaube et al., 2014). Such regional variations are considerable also in the Southern Ocean. In contrast, seasonal variations have been reported to be relatively weak globally (Gaube et al., 2014), except for the eastern Indian Ocean and the South China Sea (Gaube et al., 2013;Guo et al., 2017). And seasonal changes in the sign of δChl in a particular region have to our knowledge not been reported before. Hence, the strong seasonality with a seasonal change in the sign of δChl along the ACC and south of the ACC appears to be rather specific to the Southern Ocean.

10
The spatial and seasonal variability of δChl may not be that surprising in hindsight, given that the same mechanism, e.g., advection can lead to either positive or negative signs for the same polarity depending on the sign of the lateral gradient. In addition, several mechanisms may be involved simultaneously, so that small differences in their relative importance can lead to substantial differences in the net sign of the response (Siegel et al., 2011;Gaube et al., 2014;McGillicuddy, 2016). In the end, we demonstrated that most of the eddy induced signatures of δChl in the Southern Ocean are likely due to stirring, a 15 mechanism that has been shown to control δChl in the low to mid-latitude ocean as well (Chelton et al., 2011a). But we showed also that there are several other regions/seasons where other processes, namely trapping and changes in biogeochemical rates appear dominant. To illustrate this, we took the Hovmoeller diagram of Figure 4 and identified the dominant mechanism based on the results of the analyses of the two advective potentials (Figure 7). This synthesis figure (Figure 7) reveals that the dominance of stirring as the sole mechanism is limited to the subtropical 20 waters outside of deep mixed layers, and for anticyclones along the northern ACC in summer to autumn (Figure 7, yellow). Our results suggest that trapping contributes to δChl for anticyclones along the southern ACC in summer to autumn and in Antarctic waters in autumn and spring, and to δChl of cyclones in most regions and seasons, except for subtropical waters in winter to spring (see also Figure 8a, south and southwest of Australia). Yet, the magnitude of the potential of trapping is generally weak, with the exception, perhaps, of a few specific regions, such as by eddies originating from eastern boundary currents 25 (Supplementary Figure S3). They tend to move westward across transitional regions of coastal to off-shore waters, i.e., down intense Chl gradients (Supplementary Figure S2, right column), with a resulting positive δChl, e.g., in the southeast Pacific or southeast of Kerguelen Islands and Drake Passage. In these regions, δChl is positive year round for both anticyclones and cyclones ( Figure 3). A possible explanation next to advection of Chl is the offshore advection of iron trapped in the nearshore region by eddies that fuels extra growth in the offshore waters, as suggested e.g., for Haida eddies in the North Pacific (Xiu 30 et al., 2011). A substantial effect of trapping to cause δChl in boundary currents corroborates previous results (Gaube et al., 2014).
The nevertheless overall weaker role of trapping relative to stirring is consistent with (i) a propagation distance of eddies over their lifetime that is on average smaller than two eddy radii, meaning that the scale of impact due to eddy propagation tends to be smaller than the one due to eddy rotation, and (ii) an inherently westward propagation of eddies, meaning a propagation largely along Chl isolines, as zonal Chl gradients typically are much smaller than meridional Chl gradients.
Moreover, a weak trapping signal also is anticipated as trapped waters from the eddies' origins will be diluted along the eddies' pathways. Eddies tend to not trap perfectly but continuously leak and entrain waters ambient to their cores (Beron-Vera et al., 2013;Wang et al., 2015;Haller, 2015). The importance of such lateral entrainment or permeable trapping is supported 5 by the pronounced monopole contributions to the shapes of the local δChl imprint (Figures 5,6). The dipole contributions are relatively weak despite the comparatively steep ambient Chl gradients favoring stirring. In summary we hypothesize that the relative weakness of the dipoles stems from lateral entrainment of Chl into the eddies' cores, resulting in a pronounced asymmetry of the dipoles and a large monopole part in the δChl decomposition (see illustration in Figure 1a).
The clearest case for a substantial contribution of changes in biogeochemical rates on δChl was found for the northern ACC The associated negative δChl of anticyclones is consistent with the mechanism of an amplification of large-scale prevailing light limitation in deep mixed layers (Boyd, 2002;Moore and Abbott, 2002;Venables and Meredith, 2009;Fauchereau et al., 2011): Anticyclones tend to deepen isopycnals, causing deeper mixed layers of several tens of meters and weaker mixed layer stratification, especially in winter (Song et al., 2015;Hausmann et al., 2017;Dufois et al., 2016). Hence, phytoplankton within 15 the mixed layer will be exposed to reduced mean radiation in anticyclones as compared to ambient waters, and vice versa for cyclones.
Our result of a pronounced monopole-shape of δChl despite the weak trapping potential suggests that also in summer to autumn, positive correlations of SLA and δChl are at least partly caused by effects of eddies on biogeochemical rates. Here, prevailing iron limitation could be modulated by eddies, with an abatement of the iron limitation and associated positive δChl 20 caused by weakly stratified anticyclones in high wind conditions and associated intensified vertical mixing, and vice versa for cyclones (Boyd et al., 2012;Dufois et al., 2016;Song et al., in revision). Moreover, alleviation of grazing pressure due to reduced predator-prey encounter rates in deepened mixed layers in anticyclones could favor positive δChl, again vice versa for cyclones. Thus, we argue that along the northern ACC, the seasonal sign switch of δChl could be explained by varying degrees of light and iron limitation and grazing pressure over the course of the year (Boyd, 2002;Venables and Meredith, 25 2009; Carranza and Gille, 2015;Le Quéré et al., 2016).
Finally, along the southern ACC and in Antarctic waters in autumn to spring, the potentials of stirring and trapping oftentimes are of the same sign. However, δChl associated with eddies is insignificant (dark gray regions, Figure 7). Presumably, these situations when δChl are insignificant arise because eddy effects on biogeochemical rates oppose advective effects.
We note that our analysis is constrained to the surface ocean, hence three aspects need to be kept in mind: (i) one potential 30 issue are non-homogeneous vertical Chl profiles, e.g., the presence of unrecognized subsurface Chl maxima. As in previous studies (Sallée et al., 2015), we assume that in our focus region, at the core latitudes of the Southern Ocean across the ACC, likely not the major factor responsible for δChl. Finally , (iii) potential effects of eddies on phytoplankton growth presumably occur in the lower euphotic zone and may be expressed more weakly at the surface where they are detected by satellite sensors (McGillicuddy et al., 2007;Siegel et al., 2011). We therefore note that effects of eddies on biogeochemical rates may be underestimated in this surface based study.

10
The prevalent and strong correlations between anomalies in surface Chlorophyll and mesoscale variability have triggered substantial research, but many unresolved issues remain, particularly associated with the question of their causes (Lévy, 2008;Gaube et al., 2014;McGillicuddy, 2016). With this study, we aim to provide an observational reference for the seasonal climatological δChl associated with mesoscale eddies across the Southern Ocean, a region where the detailed regional and seasonal relationship of eddies and Chl previously had not been considered. We have obtained the estimate by combining 15 satellite estimates of Chl with ocean eddies identified based on satellite estimates of SLA. A large number of collocations of eddies and Chl allowed us to retrieve statistically robust results despite frequent data gaps and high spatio-temporal variability of Chl.
We found δChl associated with eddies of >10% over wide areas in the Southern Ocean. The large-scale patterns are positive and negative anomalies for cyclones in subtropical waters and along the ACC, respectively; anticyclones show a similar pattern, 20 but of opposite sign. A pronounced seasonality of the imprint is apparent especially along the ACC and in Antarctic waters, featuring a sign switch of the anomaly over the course of the year.
While multiple mechanisms may be at play at the same time to cause δChl (Gaube et al., 2014;McGillicuddy, 2016), our analyses based on climatological Chl gradients, eddy rotation and propagation pathways, and the local shape of δChl of eddies lead us to conclude that lateral advection due to stirring by eddies and associated lateral entrainment and permeable trapping 25 have the potential to explain a large fraction of Southern Ocean eddy-induced δChl.
A prominent region and season where eddy-induced advection is insufficient to explain δChl are the northern ACC characterized by deep mixed layers in winter to spring and the seasonal sign switch of δChl in the same region: Here, winter to spring negative and positive δChl of anticyclones and cyclones, respectively, are consistent with an enhancement and reduction of deep mixed layer light limitation. The opposite signs of δChl in summer to autumn are consistent with an abatement of grazing by effects of eddies on biogeochemical rates, even though our results suggest that lateral advection has the potential to be the dominant mechanism.
Future work may include to further investigate the extent of lateral entrainment and permeable trapping of eddies versus non-local trapping, and to test where and when Southern Ocean eddies substantially affect biogeochemical rates, such as through modulation of alternating roles of iron and light limitation and grazing pressure along the ACC. The growing number 5 of sub-surface biogeochemical measurements across eddies may be of help, collected by biogeochemical floats, gliders and seals. In addition, targeted experiments with numerical ocean-biogeochemical models with the option to alternately switch on and off Chl sources-sinks would be useful to shed light on the questions, of what the role of eddy-effects is on Chl sources-sinks relative to advection, for higher trophic levels (Xiu et al., 2011;Nel et al., 2001;Godø et al., 2012), or for the intensity of export (Waite et al., 2016). Furthermore, numerical models would allow us to assess if these effects of eddies on Chl substantially 10 affect Southern Ocean biogeochemistry, such as of mode and intermediate waters which form from winter deep mixed layers, supply low latitude ocean with nutrients and sequester anthropogenic carbon (Sarmiento et al., 2004;Sallée et al., 2012).
Data availability. The identified eddies we use in this study including their Chl characteristics are publicly available (http://dx.doi.org/10.3929/ethzb-000238826). Other presented data are available from the corresponding author upon request.