Journal of Geophysical Research: Oceans

On the drivers of phytoplankton blooms in the Antarctic marginal ice zone: A modeling approach

Authors


Corresponding author: M. H. Taylor, Alfred Wegener Institute for Polar and Marine Research, PO Box 120161, D-27515 Bremerhaven, Germany. (marchtaylor@yahoo.com)

Abstract

[1] The pelagic province of the Southern Ocean generally has low levels of primary production attributable to a short growing season in the higher latitudes, a deep mixed layer, and iron limitation. Exceptions include phytoplankton blooms in the marginal ice zone (MIZ) during spring and summer sea-ice retreat. The prevailing hypothesis as to the drivers of the blooms is that sea-ice retreat increases the vertical stability of the water column through the production of melt water and provides shelter from wind mixing in areas of partial sea-ice coverage. These conditions are favorable to phytoplankton growth by allowing them to maintain their position in the upper reaches of the water column. This work investigates the drivers of MIZ blooms using a biochemically coupled global circulation model. Results support the hypothesis in that physical conditions related to a shallow, vertically stable water column (e.g., mixed layer depth and available light) were the most significant predictors of bloom dynamics, while nutrient limitation was of lesser importance. We estimate that MIZ blooms account for 15% of yearly net primary production in the Southern Ocean and that the earlier phases of the MIZ bloom, occurring under partial ice coverage and invisible to remote sensing, account for about two thirds of this production. MIZ blooms were not found to enhance depth-integrated net primary production when compared to similar ecological provinces outside of the MIZ, although the elevated phytoplankton concentrations in surface waters are hypothesized to provide important feeding habitats for grazing organisms such as krill.

1 Introduction

1.1 Seasonal Ice Zone Biochemical Province

[2] The seasonal ice zone (SIZ) extends hundreds of kilometers from the Antarctic coast [Sakshaug et al., 1991] with sea ice typically being less than 1 m thick [Pfaffling et al., 2007] and growing from April and melting from October. The offshore water masses are characterized by latitudinal gradients in nutrients during the winter, which reflect the circumpolar frontal structure [Dafner et al., 2003]. The major fronts are found at 40°S (sub-tropical front), 45°S (sub-Antarctic front), 50°S (Antarctic polar front), and ca. 52° to >60°S, the southern boundary of the Antarctic Circumpolar Current (ACC) [Orsi et al., 1995]. Treguer and Jacques [1992] estimated the SIZ area at 16 × 106 km2, making it the largest of the Southern Ocean biogeochemical provinces identified in their review. The SIZ extends annually as far north as 55°S, while the southernmost areas of the SIZ lie directly over the Antarctic shelf, in the physical domain of the counterclockwise Coastal Current.

[3] The Southern Ocean (SO)has been described as the largest high-nutrient low-chlorophyll (HNLC) region in the world ocean [Martin et al., 1990; Minas and Minas, 1992]. Low levels of primary production have been attributed to the relatively short growing season in the higher latitudes, a deep mixed layer, and iron limitation [Martin et al., 1990; Mitchell and Holm-Hansen, 1991; Boyd et al., 2000]. It is hypothesized that, for open ocean areas of the SO, phytoplankton growth rates are enhanced by oceanographic fronts, whereby divergence of surface waters can bring iron-replete waters into the euphotic zone [Hense et al., 2000; Moore and Abbott, 2000; Osmund Holm-Hansen and Hewes, 2004; O. Holm-Hansen et al., 2005; Sokolov and Rintoul, 2007]. In open water, the timing and intensity of the spring bloom varies greatly, probably on account of water mass interactions [Dafner et al., 2003].

[4] The highest levels of Southern Ocean primary production have been associated with coastal polynyas [Arrigo and van Dijken, 2003; 2007], marginal ice zone (MIZ) blooms [Smith and Nelson, 1985; 1986], and the continental shelf [Smith and Gordon, 1997; Arrigo and van Dijken, 2004]. Again, iron supply has been suggested to be a factor, with important sources being the interaction between the ACC and bottom topography, upwelling, vertical diffusion, and melting of ice and icebergs [Moore et al., 1999; Fung et al., 2000; Law et al., 2003; Holm-Hansen et al., 2005].

1.2 Drivers of Marginal Ice Zone Blooms

[5] Of interest to this study are MIZ blooms occurring in the more pelagic extensions of the SIZ. From a physical standpoint, the MIZ is has been defined as “that part of the ice cover which is close enough to the open ocean boundary to be affected by its presence” [Wadhams et al., 1986]. In the context of associated phytoplankton blooms, a more bio-centric definition is usually applied to water characteristics; specifically, areas displaying vertical stability due to the production of meltwater during the seasonal ice retreat [Sullivan et al., 1988]. Although this includes conditions of partial ice coverage, remote-sensing studies are restricted to open water and thus usually define the MIZ based on the recency of sea-ice presence as a proxy for stratification and/or nutrient input.

[6] It is hypothesized that the stabilization of the surface layer by ice melt during the spring provides perfect growth conditions for phytoplankton by allowing for their concentration in the upper reaches of the euphotic zone [Smith and Nelson, 1985]. In concert with conditions of ice melt is the increased availability of light to the water column, which is ultimately needed for photosynthesis. Smith and Comiso [2008] find that the influence of the ice cover varies regionally and that only when the ice cover is thick and closed does it tend to control the light availability and hence the initiation of the bloom.

[7] The duration of the bloom also depends upon nutrient availability and the maintenance of vertical stability, which is reflected by a shallow mixed layer. Vertical stability is likely to be disrupted once protection from partial sea-ice coverage has diminished and surface waters are subjected to wind mixing. Support for this scenario comes from remote-sensing observations indicating that bloom occurrence and intensity in the MIZ is correlated with wind speeds [Fitch and Moore, 2007]. Specifically, wind speeds above 5 m∙s−1 were negatively correlated with surface chlorophyll, whereas those below 5 m∙s−1 were positively correlated.

[8] Diatoms dominate in highly stratified waters of the MIZ, whereas Phaeocystis antarctica assemblages dominate where waters are more deeply mixed [Arrigo et al., 1999; Goffart et al., 2000]. The question of whether algae released from melting ice seed the pelagic spring phytoplankton bloom has been under debate for many years [e.g., Lancelot et al., 1993]. Phytoplankton populations observed within sea ice, themselves seeded from the pelagic population as phytoplankton are incorporated into the ice growth, have been found during some studies to resemble closely the pelagic populations in spring as well as in autumn [Krell et al., 2005], while other studies observe different communities as compared to the pelagic population [Mathot et al., 1991]. There are three possible scenarios for the origins of the spring MIZ bloom: (a) the sympagic population seeds the spring pelagic bloom upon ice melt, or (b) a background “ambient” population of cells which survives the winter beneath the sea ice (perhaps as cysts) seeds the spring pelagic bloom once conditions improve, or (c) cells to the north of the ice edge proceed southward with the retreating sea ice. Each of these scenarios is supported by documentation of the survival strategies of Antarctic phytoplankton.

1.3 Modeling Approach

[9] MIZ blooms have been reported on numerous occasions from both in situ data [Smith and Nelson, 1985; 1990; Lancelot et al., 1993; Bracher et al., 1999; Buesseler et al., 2003] and from remotely sensed data [Moore and Abbott, 2000; Fitch and Moore, 2007; Arrigo et al., 2008; Smith and Comiso, 2008]. MIZ blooms have been suggested to contribute a significant portion to overall Southern Ocean primary production due to the widespread occurrence of MIZ conditions during seasonal ice retreat. Smith and Nelson [1986] estimated that overall pelagic primary productivity for the entire Southern Ocean would be increased by at least 60% if ice-edge production were considered. However, recent remote-sensing estimates by Arrigo et al. [2008] show that MIZ zones contribute 4.4% of total Southern Ocean primary production and do not substantially increase productivity over non-MIZ conditions. Nevertheless, these estimates are likely to be conservative because the presence of sea ice prevents estimates of ocean color via remote sensing (Figure 1), even within the ice edge or low ice concentrations [Belanger et al., 2007]. It has been suggested that this is a likely source of underestimation in remote estimates given that areas of partial ice coverage may receive a substantial amount of irradiance into the water column to drive primary production [Smith and Comiso, 2008].

Figure 1.

Fraction of days with remote estimates of chlorophyll a during 1997–2007 (Globcolour GSM, 4 km product). Dashed line indicates the maximal extent of the seasonal ice zone (SIZ).

[10] The tradeoff between localized in situ sampling, which is limited to shipboard sampling, and remote-sensing estimates, which offers an incomplete view during MIZ conditions, illustrates the difficulty in the investigation of blooms. As an alternative, numerical ocean modeling allows for the examination of MIZ blooms over large areas and over the full growth period. In addition, modeling can help to elucidate the drivers behind MIZ blooms by allowing for the investigation of a range of parameters relating to phytoplankton dynamics.

[11] The objectives of this work focus on two main aspects of the MIZ: (1) to assess the drivers of MIZ blooms within the modeled ecosystem using a multivariate statistical approach and (2) characterize the importance of MIZ blooms to overall Southern Ocean primary production. Finally, we discuss the possible implications of MIZ bloom dynamics for the lifecycle of the trophically important krill.

2 Methods

2.1 Description of the Biophysically Coupled Global Circulation Model

[12] Simulations were conducted using the Massachusetts Institute of Technology General Circulation Model (MITgcm) [Marshall et al., 1997; MITgcm Group, 2012], which is integrated on a cubed-sphere grid, permitting relatively even grid spacing while avoiding polar singularities [Adcroft et al., 2004]. Each face of the cube is composed of a 510 × 510 grid (mean spacing = 18 km) and 50 vertical levels ranging in thickness from 10 m near the surface to approximately 450 m at a maximum model depth of 6150 m [Menemenlis et al., 2008]. Initial conditions, spin-up, physical forcing fields, sea ice, and further details of the global model are described in section 3 of Losch et al. [2010]. Of particular relevance to phytoplankton growth in the SIZ is the parameterization of light transmission through sea ice and into the euphotic zone; details on the parameterization of these processes can be found in Appendix Appendix. The simulation spanned the years 1992 through 2007.

[13] The MITgcm was coupled with a version of the biogeochemical model “Regulated Ecosystem Model” (REcoM) [Schartau et al., 2007]. REcoM uses phytoplankton growth parameterizations of Geider et al. [1998] in order to account for the effect of varying stoichiometry on phytoplankton growth and, subsequently, nutrient cycling and other biological processes. Loss of phytoplankton biomass is assumed to be due to grazing, particle aggregation, exudation, and leakage. Additional parameterization has been added to account for silica and iron limitation of phytoplankton growth [Hohn, 2009]. In this form, the phytoplankton component of the model mainly describes dynamics associated with a diatom dominated community. For additional details, see Losch et al. (Ocean State Estimation from Hydrography and Velocity Observations During EIFEX with a Regional Biogeochemical Ocean Circulation Model, submitted to Journal of Marine Systems).

[14] Initial condition of all REcoM variables are derived from a spun-up simulation with a coarse version of the model (based on Hohn [2009]) by interpolation onto the fine grid. The first year (1992) of the coupled high-resolution run is not used in the analysis. With the initial conditions from the coarse resolution the model, simulated nutrient distributions are nonlimiting for nitrogen in the SO, nonlimiting for silica south of 45°S, and limiting for iron in the 40°S to 60°S latitude band. Further south, the surface is frequently replenished with iron from deeper layers via vertical mixing during ice formation, and iron is only marginally limiting [see Hohn, 2009, Figure 4.12]. Additional details regarding iron chemistry and the initial conditions for iron concentration can be found in Appendix Appendix.

2.2 Selection of Focus Areas

[15] The performance of the simulation was assessed through a comparison to remotely sensed data. Daily means from remotely sensed sea-ice coverage, sea surface temperature, and chlorophyll a were used for the comparison:

  • Sea-ice coverage—12.5 km resolution gridded product from IFREMER (http://cersat.ifremer.fr/), spanning the years 1993–2007. The product uses the ARTIST algorithm [Spreen et al., 2005; Spreen et al., 2008] on Special Sensor Microwave Imager data [Cavalieri et al., 2011].
  • Chlorophyll a—4.62 km resolution gridded product from the GlobColour Project (http://www.globcolour.info/), spanning the years 1997–2007. We used the GSM merged product (Garver, Siegel, Maritorena Model) [Garver and Siegel, 1997; Maritorena et al., 2002], which combines Medium-Resolution Imaging Spectrometer (MERIS), Moderate Resolution Imaging Spectroradiometer (MODIS), and Sea-viewing Wide Field-of-view Sensor (SeaWiFS) data. Due to the differences in satellite operation time, the period from 1997–2002 consists of SeaWiFS data only, while 2003 onward uses data from all three satellites.
  • Sea surface temperature—4 km resolution gridded product uses AVHRR Pathfinder Version 5 data, obtained from the US National Oceanographic Data Center and GHRSST (http://pathfinder.nodc.noaa.gov) [Casey et al., 2010], spanning the years 1993–2007.

[16] No additional model tuning was done beyond the aforementioned tuning of REcoM in the previously run coarse resolution model. Generally, simulated mean monthly chlorophyll a concentrations (log-transformed) correlated to remote-sensing data at R = 0.62 globally and R = 0.23 for the Southern Ocean. These global correlation values are higher than those presented by other coupled general circulation model (GCM) studies [Schneider et al., 2008; Doney et al., 2009]. Lower correlations appear to be common for polar regions. In a review of both GCM and remote-sensing algorithm models of primary production, the Southern Ocean was found to be an area of highest divergence of estimates [Carr et al., 2006]. In this work, we have chosen to focus our statistical analysis only on best performing sub-areas of the SIZ rather than specifically tune the MITgcm-REcoM model to SO conditions. The SIZ domain was first defined as the area covered by >15% sea-ice concentration during any point of the simulated period. Within the SIZ, criteria for best performing areas were based on the correlations (Spearman ρ) of daily averages between simulated and remotely observed estimates. Areas must have a correlation of >0.5 for sea surface temperature and sea-ice coverage to qualify for further analysis. Additionally, chlorophyll a must have a correlation of >0.1, and only areas where at least 10% of daily spring-summer remote-sensing estimates were considered. This final criterion limited our analysis to the more pelagic extension of the SIZ where a greater number of ice-free days allowed for a more robust comparison. All correlations must be significant at the p < 0.01 level.

2.3 Statistical Approach

[17] Several modeled parameters were assessed for their influence on surface phytoplankton concentrations (Table 1). These included four physical parameters [mixed layer depth (MLD), integrated photosythetically active radiation (PAR), sea surface temperature (SST), and sea surface salinity (SSS)], three limiting nutrient parameters [surface dissolved inorganic nitrogen (DIN), surface dissolved silicate (DSI), and surface dissolved iron (DFE)], and one biological parameter [surface zooplankton carbon (ZOOC), i.e., as a proxy for grazing losses].

Table 1. Model Parameter Descriptions
AbbreviationVariableUnits
CHLASurface chlorophyll αmg m−3
MLDMixed layer depthm
PARIntegrated photosythetically active radiation (<MLD)W m−1
SSTSea surface temperature°C
SSSSea surface salinitypsu
DINSurface dissolved inorganic nitrogenmmol m−3
DSISurface dissolved silicatemmol m−3
DFESurface dissolved ironμmol m−3
ZOOCSurface zooplankton carbonmmol m−3

[18] For each sub-area parameter field, an empirical orthogonal function analysis (EOF) was performed to reduce the highly dimensional spatio-temporal data to a single dominant mode of variability. EOF was applied to covariance matrices based on the centered (mean subtracted) time series of the grids in each sub-area. By using the leading EOF mode's coefficient (i.e., “principal component”), a single temporal signal was derived for each modeled parameter.

[19] EOF coefficients were used as covariates for our statistical model with surface chlorophyll a (CHLA) as the response variable. In order to reduce the influence of multicollinearity on fitted model terms, a pre-selection of predictor covariates was conducted based on their variance-inflation factor (VIF). We applied a commonly defined threshold for variable removal when VIF > 10. Furthermore, we defined the threshold for the mean VIF of included covariates as <6. When violations occurred, an iterative process was used to remove covariates until both criteria were satisfied.

[20] Using the R statistical package “mgcv,” we applied a generalized additive model (GAM) [Wood, 2004, 2006]. GAM models allow for nonlinear relationships among covariates through the fitting of spline functions to model terms. Cubic regression splines were fit with the number of basis dimensions left open to a penalized fitting. Under these settings, the addition of regression spline “knots” is penalized by the associated increase in degrees of freedom. As a consequence, cases where nonlinear regression splines do not improve the fitting will be fit by a simple linear regression.

[21] Model fitting was done according to recommendations of Zuur et al. [2009]. A full model, which included all terms, was fit using “maximum likelihood” (ML) as a criteria for the estimation of smoothing parameters. Removal of terms was done in a stepwise fashion by comparing the fits of a “bigger” model, which included the term under consideration, versus a “smaller” model, where the term was dropped. Best models were assessed through the minimization of the Akaike information criterion (AIC) and significance via a likelihood ratio (L) test. The final model was refitted using “restricted maximum likelihood” (REML) as the fitting criteria. An example of a fitted GAM model applied to a single geographic point of the SIZ can be seen in Figure 2. Each predictor variable is fit with a spline function describing its effect on CHLA. For example, mixed layer depth contributes positively only when less than 45 m. The example uses time signals with their actual units for easy interpretation. In the actual application using EOF coefficients, the signals are normalized (mean = 0, SD = 1) and of arbitrary sign, which does not affect the significance or shape of the fitted spline function in the model.

Figure 2.

Example of generalized additive model (GAM) model fit to time series from a single grid location in the SIZ. Fitted smooth terms are to the right of each covariate's time series with surface chlorophyll α (CHLA) as the response variable. Prediction values are shown as blue dots in CHLA time series.

2.4 Estimation of Primary Production in the Marginal Ice Zone

[22] In order to quantify the impact of MIZ conditions on depth-integrated net primary production (NPP), we followed the protocol outlined by Arrigo et al. [2008], whereby they distinguished four ecological provinces for the Southern Ocean (<50°S) based on depth and sea-ice presence: (1) pelagic (>1000 m, 0% ice, and 0% ice for >14 days), (2) shelf (<1000 m, 0% ice, and 0% ice for >14 days), (3) pelagic MIZ (>1000 m, 0% ice, and >0% ice at some time in the last 14 days), and (4) shelf MIZ (<1000 m, 0% ice, and >0% ice at some time in the last 14 days). The MIZ threshold of 14 days is based on in situ measurements of low salinity water and phytoplankton bloom persistence following sea-ice retreat [Smith and Nelson, 1986; Lancelot et al., 1991]. Arrigo et al. [2008] estimated NPP with an algorithm based on remotely sensed ocean color and other parameters and thus were restricted to open water conditions (i.e., 0% sea-ice coverage). Due to our ability to observe and quantify the simulated NPP even under conditions of sea-ice coverage, we were able to define an alternative MIZ criterion that better captured the beginning of the bloom period. In the simulation, blooms usually began soon after the initial breakup and retreat of sea ice, when concentrations fell below 90% coverage (Figure 3). As a consequence, the spatial development of the bloom largely follows the southward retreat of sea ice. Therefore, we also compare NPP using an alternative definition for the MIZ conditions: (5) pelagic MIZ (>1000 m, <90% ice, and >0% ice at some time in the last 14 days), and (6) shelf MIZ (<1000 m, <90% ice, and >0% ice at some time in the last 14 days). The two MIZ definitions will later referred to as “MIZ-0” for the Arrigo et al. [2008] definition and “MIZ-90” for our alternate definition.

Figure 3.

Simulated daily average chlorophyll a concentration (mg m−3; color gradient) and sea-ice coverage (%, white isolines). (a–d) Two week snapshots from 15 October to 1 December 2004 show the development of a phytoplankton bloom in areas where sea-ice coverage has been reduced to below about 90%.

3 Results

3.1 Correlation to Observed Estimates and Focus Area Selection

[23] Figure 4 shows the spatial correlations of the simulation versus remote-sensing data. Nine sub-areas, with longitudinal extension of <30°, were identified that passed all of the aforementioned selection criteria (Figure 4, bottom right). Sea surface temperature and sea-ice coverage were, generally, very well correlated with observed values throughout the SIZ. Areas of lower correlation occurred inshore near larger ice shelves (i.e., polynyas), which are not explicitly modeled, and, in the case of sea ice, near the more variable outer extension of the SIZ. The correlation of chlorophyll a was generally lower and patchier than the other two fields, although large areas of significantly positive correlations can be seen encircling the Antarctic continent. The fulfillment of criteria for chlorophyll a was the most restrictive of the comparisons to remote-sensing data in defining the sub-areas for further statistical analysis.

Figure 4.

Correlation of simulated vs. remote-sensing estimates for chlorophyll a, sea surface temperature, and sea-ice coverage. Isolines indicate areas of strong correlation among all three fields. Bottom right map shows the nine sub-areas identified for further statistical analysis. Black dashed isoline shows the maximum extent of the SIZ over the study period.

3.2 Statistical Exploration

[24] The leading EOF mode explained a large percentage of each field's variance, usually >75% (Figure 5). In order to reduce the impact of multicollinearity between the model predictors, several covariates were excluded from the GAM analysis due to their VIF. ZOOC was removed from all nine sub-area models; DIN was removed from seven sub-area models; and SST, SSS, and DSI were each removed from one sub-area model (Table 2).

Figure 5.

Variance explained by the leading empirical orthogonal function (EOF) for each variable field.

Table 2. Fitted GAM Model Statistics by SIZ Area
AreaNR-sq. (adj.) Terms
MLDPARSSTSSSDINDSIDFE
  • a

    d.f., degrees of freedom; vif, variance-inflation factor; L, log likelihood ratio.

  • *

    All terms are significant at the p < 0.001 level.

154780.83d.f.a7.917.338.388.12 8.338.51
vifa4.52.74.96 4.57.1
La*21912017907323 140415
254780.87d.f.a8.756.948.898.888.58 8.58
vifa4.13.45.98.26 2.9
La*214527221326380176 123
354780.87d.f.a8.446.757.42  8.288.6
vifa3.93.94.7  5.62.1
La*25812194797  701307
454780.87d.f.a7.87.358.018.55 8.568.77
vifa4.93.948.1 6.13.1
La*133839651324630 594495
554780.82d.f.a8.087.097.068.18 8.468.7
vifa4.73.36.34.9 3.43.4
La*13353531424666 213350
654780.85d.f.a5.637.668.815.96 8.446.31
vifa5.32.67.28.6 5.95.6
La*28967181232163 114124
754780.83d.f.a7.938.11 8.358.198.818.48
vifa3.51.5 3.674.39.2
La*10315403 72760395237
854780.86d.f.a8.227.688.297.38 7.687.49
vifa4.22.56.34.8 3.72.6
La*256619191101904 113365
954780.89d.f.a8.787.297.918.62 7.898.47
vifa4.42.39.33.8 44.1
La*94549462409229 419382

[25] For all sub-area models, stepwise removals of remaining terms did not improve the model, and all included terms were deemed highly significant. Fitted spline functions varied in their complexity, as revealed by their associated degrees of freedom, but in no case was there an indication that linear relationships were more appropriate. The squared correlation coefficients (R2) were >0.8 for all models, indicating a relatively good predictive power (Table 2). Generally, covariates associated with physical conditions were the most significant predictors of CHLA. In particular, MLD and PAR were consistently among the most significant terms. Nutrient concentrations (DIN, DSI, and DFE) were of much lower significance (Figure 6).

Figure 6.

Log likelihood ratios of GAM model term inclusion. All terms are significant at the p < 0.001 level. Asterisks indicate terms not included in the sub-area model.

3.3 Primary Production of the Marginal Ice Zone

[26] Figure 7 shows yearly cycles in area, NPP and NPP/area by ecological province. The pelagic province dominates in terms of area and NPP, although the pelagic MIZ is a significant contributor to NPP from November to January, especially when using the MIZ-90 definition. On a per area basis, shelf provinces are the most productive, with the maximum NPP values around 250 mg C m−2 d−1 during November and December. Both the shelf MIZ and the pelagic MIZ had NPP/area values similar to that of the pelagic province when using the MIZ-0 definition. When applying the MIZ-90 definition, the MIZ NPP/area is much lower than its respective non-MIZ province.

Figure 7.

Southern Ocean ecological province yearly cycles in (a) area, (b) net primary production (NPP), and (c) net primary production per area (NPP/area). Year-day mean (solid line) and standard deviation (shaded area) are shown.

[27] Figure 8 shows yearly means in area, NPP, and NPP/area by ecological province as compared to Arrigo et al. [2008]. Both studies show similar values in ecological province area. NPP/area estimates of Arrigo et al. [2008] are consistently ~2× higher, which translates to higher values for total NPP by province, although the relative contributions were similar. Our MIZ-90 definition resulted in about a fivefold increase in area and threefold increase in NPP over values calculated using the MIZ-0 definition. This increased the MIZ's contribution from 5% to 15% of Southern Ocean NPP.

Figure 8.

Comparison of Southern Ocean ecological province yearly means in (a) area, (b) net primary production (NPP), and (c) net primary production per area (NPP/area) to remote sensing based on estimates of Arrigo et al. [2008]. Denoted by asterisks, not applicable to the Arrigo et al. [2008] study.

4 Discussion

4.1 Drivers of Marginal Ice Zone Blooms

[28] Our findings support the hypothesis that the stability of a shallow pycnocline, associated with melting sea ice, is most responsible for the development of phytoplankton blooms in the MIZ [Smith and Nelson, 1985; Sullivan et al., 1988]. In particular, light availability (PAR) and MLD are the most significant predictors of surface chlorophyll a dynamics in explored sub-areas of the SIZ (Figure 6). PAR is determined primarily by season and sea-ice coverage in the SIZ. MLD, as determined by gradients in water density, is mainly affected by the processes of melting sea ice and mixing of the water column by wind. From the example GAM model (Figure 2), we can see that the contribution of PAR is largely flat after a minimum threshold is reached, and similar spline function forms were also observed in each of the sub-area models. This threshold is met once the sea ice begins to break up in spring, and it is this melting that also results in a shallow MLD due to the strong density gradient created by the fresh water lens of less dense, lower salinity waters at the surface. Additionally, the partial ice coverage hinders mixing by surface winds. The combination of these factors allows phytoplankton to maintain their position in the high light conditions of the upper layer of the water column, which results in enhanced growth and maintains bloom conditions. The termination of the bloom coincides with sea-ice concentrations reducing to levels near zero, and a deepening of the MLD in response to wind-forced mixing. This mechanism has also been observed in the MIZ through remote sensing, whereby bloom occurrence was inversely related with wind speed [Fitch and Moore, 2007]. Specifically, wind speeds <5 m∙s−1 were most associated with bloom conditions. This speed corresponds to the threshold for turbulent mixing and a deepening of the MLD in coastal waters [Kullenberg, 1971; 1972; 1976], with higher speeds shown to be related to decreases in phytoplankton patchiness [Therriault and Platt, 1981; Demers et al., 1987]. The GAM results also indicate that SST and SSS are relatively important predictors of chlorophyll a dynamics for some sub-areas. These variables are closely related to the dynamics of PAR and MLD, and thus their significance is likely mainly due to their association with the formation of the freshwater lens during ice retreat.

[29] Nutrient dynamics were less important predictors of surface chlorophyll a dynamics. When these variables were included in the model, they were also significant yet to a much smaller degree than the above-mentioned physical parameters. The Southern Ocean is the largest HNLC region in the world ocean [Martin et al., 1990; Minas and Minas, 1992]. Despite concentrations of macronutrients sufficient to support a greater phytoplankton community, there is evidence that phytoplankton growth is limited by iron and occasionally silica [Martin et al., 1990; Treguer and Jacques, 1992; Boyd et al., 2000; Osmund Holm-Hansen and Hewes, 2004; O. Holm-Hansen et al., 2005]. This is particularly apparent in the pelagic waters north of the SIZ, where productivity is relatively low. The SIZ has been shown to be less limited by silica than the larger Southern Ocean [Sarmiento et al., 2007; Hohn, 2009], yet iron input to the upper ocean may also be limited outside the shallower shelf regions. Iron supply to the ocean surface is dominated globally by atmospheric deposition [Fung et al., 2000; Mahowald et al., 2005]. However, in the Southern Ocean this component is small, so that interactions between the Antarctic Circumpolar Current and bottom topography, upwelling, vertical diffusion, and melting of ice and icebergs provide comparatively important sources of iron [Moore et al., 1999; Fung et al., 2000; Law et al., 2003; O. Holm-Hansen et al., 2005]. While our model does consider some of these processes in the cycling of iron, their dynamics were not found to have as much importance to the overall phytoplankton dynamics in the selected sub-areas of the pelagic province (Figure 6). In situ estimates of iron concentration are scarce, especially at depth, and thus our initial fields likely include a higher degree of error than for the other limiting nutrients. Nevertheless, a review of Southern Ocean dissolved iron measurements did not support the link between iron cycles and uptake by phytoplankton and suggested that other processes might be more important drivers in its variability (e.g., recycling, exogenous inputs, and/or mixed layer dynamics) [Tagliabue et al., 2012]. The influence of iron limitation on phytoplankton dynamics may be more pronounced in the inshore shelf areas where vertical diffusion plays a more important role in its resuspension to the euphotic zone. Consistent with in situ observations, the highest values in simulated NPP were also found within the shelf province (Figure 7). Uncertainty in initial iron concentrations will likely be improved in future simulations due to the growing amount of observed data being generated in recent years (e.g., GEOTRACES program, http://www.bodc.ac.uk/geotraces/).

[30] Due to constraints of our statistical approach, we were unable to assess the impact of zooplankton grazing on phytoplankton dynamics due to problems of multicollinearity with other covariates. We believe that these losses will likely be small at the onset of the bloom, when zooplankton development is likely to lag that of the phytoplankton, but may increase later in the growing season.

4.2 Importance of the Marginal Ice Zone to Southern Ocean Primary Production

[31] The highest rates of primary production in the Southern Ocean are generally associated with coastal polynyas [Arrigo and van Dijken, 2003; 2007], the MIZ [Smith and Nelson, 1986], and the continental shelf [Smith and Gordon, 1997; Arrigo and van Dijken, 2004], whereas the pelagic province is usually associated with lower productive waters. One exception is along the Antarctic Polar Front due to upwelling of nutrient-rich waters to the euphotic zone via divergence of surface waters [Bracher et al., 1999; Hense et al., 2000; Moore and Abbott, 2000; Tremblay et al., 2002].

[32] Both our results and more recent remote-sensing estimates, which use an algorithm developed especially for the Southern Ocean [Arrigo et al., 2008], suggest that MIZ conditions in the pelagic province do not enhance NPP over non-MIZ conditions. Given that surface chlorophyll a levels in the MIZ are often much higher than in the open waters and clearly show bloom conditions, lower associated NPP would appear to contradict the longstanding view that MIZ blooms are among the areas of highest primary production in the Southern Ocean [Smith and Nelson, 1986]. Arrigo et al. [2008] argued that the most likely reason for this result over larger scales of the Southern Ocean was that the conditions necessary to create highly productive blooms are not often met. In particular, they note that conditions leading to a well-stratified mixed layer may often fail to develop or are destroyed by wind driven turbulent mixing before a phytoplankton bloom can form [Fitch and Moore, 2007]. In other words, using the criterion of recency of sea-ice presence as a proxy for the MIZ may falsely identify locations where conditions of vertical stability have been prematurely destroyed by wind mixing. Our results suggest that blooms are much more ubiquitous and stable in conditions of partial sea-ice coverage (Figure 3), possibly due to protection from wind mixing. Furthermore, through the inclusion of this area in the MIZ-90 definition, we find that NPP/area in the MIZ is even more dramatically reduced over estimates that use the MIZ-0 definition from Arrigo et al. [2008] (Figures 7 and 8), supporting the finding that MIZ blooms are not associated with enhanced NPP. Figure 9 shows the relationship between surface chlorophyll a and NPP throughout the Southern Ocean, indicating lowest NPP rates associated with the higher surface chlorophyll a levels of the MIZ. This negative relationship is also apparent in the cross-section views, where the strongest and most concentrated surface blooms are associated with the lowest NPP. To the contrary, the non-MIZ areas of the ice-free pelagic province show the more typical positive relationship between surface chlorophyll a and NPP. The reduced NPP rates in the MIZ are mainly due to the lower integrated PAR caused by partial sea-ice coverage, although self-shading is also likely given the elevated phytoplankton concentrations in the upper levels of the water column.

Figure 9.

Comparison of December 2004 monthly means of surface chlorophyll a concentration (top left) and integrated net primary production (NPP) (top right). White isolines indicate the mean sea-ice concentrations. Dashed black lines along 30°E and 150°W indicate the locations of the cross-section views of chlorophyll a concentrations (lower plots). In the cross-section views, dashed white lines indicate NPP, and solid white lines indicate the integrated photosythetically active radiation (PAR) down to the depth of the mixed layer.

[33] Our results suggest that MIZ conditions account for about 15% of Southern Ocean NPP. In particular, our modeling approach has allowed for the quantification of the entire development of the bloom, including the earlier stages during the initial breakup and retreat of sea ice. We estimate that these earlier phases during partial ice coverage account for about two thirds of MIZ NPP. Without this additional component (i.e., using the MIZ-0 definition), the contribution of MIZ to total Southern Ocean NPP would be 4.7%, which is close to the estimate of 4.4% by Arrigo et al. [2008]. Additionally, the inclusion of partially ice covered waters allows for a much more complete view of total yearly NPP in the Southern Ocean; by using the MIZ-0 definition (i.e., open water conditions), only 88% of the total Southern Ocean NPP was accounted for within the four ecological provinces, whereas the MIZ-90 definition accounts for 99%. The inability to estimate NPP in partially ice covered areas with remote sensing has been previously highlighted as a possible source of underestimation of NPP in the Southern Ocean [Smith and Comiso, 2008], and thus our approach has helped to shed light on the importance of the MIZ to overall Southern Ocean NPP budgets.

[34] Given that our model largely describes the dynamics of the diatom component of the phytoplankton community, the results do not fully describe all NPP in the SO. Generally, diatoms have been shown to dominate the highly stratified waters associated with the MIZ, whereas Phaeocystis antarctica assemblages dominate where waters are more deeply mixed [Arrigo et al., 1998; Arrigo et al., 1999]; however, the phytoplankton community contains many species across a spectrum of life history strategies, including within the diatoms themselves. Nevertheless, the simplification of the modeled phytoplankton should not diminish the importance of our results regarding MIZ processes, although NPP estimates of the Southern Ocean as a whole are likely incomplete and lower than reality. For example, a newer version of the REcoM model, which incorporates nanophytoplankton, estimates that this smaller fraction accounts for 40% of Southern Ocean NPP (Hauck et al., Inter-annual variability of Southern Ocean organic and inorganic carbon fluxes, submitted to Global Biogeochemical Cycles).

4.3 Importance of Seasonal Ice Zone Blooms for the Larger Ecosystem

[35] Despite the finding that MIZ processes do not enhance NPP in the pelagic province, their associated blooms may still have an important role for functioning of the larger ecosystem. MIZ blooms provide a concentrated food source in the upper layers of the water column that are likely to improve the feeding efficiency of grazing organisms. Antarctic krill (Euphausia superba) have long been of particular interest due to their central role in the Southern Ocean food web, both as important grazers of plankton and as prey to a variety of higher predators. Krill are found in highest abundances on the shelf but can be generally described as a pelagic species, with 87% of the total krill stock living in deep ocean water (>2000 m) [Atkinson et al., 2008].

[36] Krill distribution largely overlaps with that of the SIZ, and sea-ice habitats are required during parts of its lifecycle. Whereas adult krill are known to have a more benthic feeding mode during the winter period [Kawaguchi et al., 1986], krill larvae actively feed on the sea-ice biota living within and below the sea ice [Daly and Macaulay, 1991]. Several studies also link sea-ice extent and duration with recruitment success and abundance changes [Siegel and Loeb, 1995; Loeb et al., 1997; Atkinson et al., 2004], highlighting the long-term trend toward reduced krill populations in response to global warming induced reductions in sea-ice extent. This scenario is most probable for the west Antarctic Peninsula region, which has shown the largest increases in temperature and subsequent reductions in sea ice.

[37] Food availability has been identified as the most important factor in the krill's lifecycle [Siegel, 2005; Atkinson et al., 2008; Meyer et al., 2009; Meyer et al., 2010], including several critical stages for recruitment success [Meyer, 2012]. Our results show a maximum in the MIZ province's area and NPP during spring, a period when phytoplankton blooms have been shown to be important for successful recruitment, allowing for early ovarian development, early spawning, and multiple egg batches [Quetin and Ross, 1991; Schmidt et al., 2012]. MIZ bloom dynamics may also be of particular relevance to the first larval feeding stage (calyptopsis 1, “C1”), which reach surface waters after their developmental ascent and must find food within the first 10 days in order to survive [Ross and Quetin, 1989]. An interesting analog from the tropics is that of the Peruvian anchovy (Engraulis ringens), which has been found to have highest recruitment within an “optimal environmental window” of conditions conducive to the formation of phytoplankton blooms. In particular, the relationship between wind speed and recruitment is bell shaped with a maximum near the threshold of turbulent mixing (5 m∙s−1) [Cury and Roy, 1989]. The authors hypothesized that under these optimal conditions, upwelling of nutrient-rich waters can fuel phytoplankton growth without destroying the blooms through turbulent mixing. As mentioned before, the same threshold has been associated with the diminishment of blooms in the MIZ [Fitch and Moore, 2007].

[38] Light has been hypothesized to be an important cue in the triggering of metabolic changes in krill following winter (see review by Meyer [2012]). From an evolutionary perspective, such mechanisms would only develop if light is a consistent and accurate cue of improved feeding conditions. Our results support this view through the finding that PAR is a highly significant predictor of surface chlorophyll a concentrations in the SIZ. In addition, the increase in PAR following sea-ice melting will be more abrupt than the more gradual seasonal increase in daylight experienced outside the SIZ, possibly providing a more obvious cue signaling increased food availability.

5 Conclusions

[39] This study sheds light on the drivers of MIZ blooms and their importance to overall primary production in the Southern Ocean. Our results support the prevailing hypothesis that MIZ blooms are driven mainly by physical processes; i.e., the formation of a shallow, vertically stable water column during sea-ice retreat allows for the development and maintenance of the phytoplankton bloom in the upper reaches of the water column. Although nutrient concentrations are significantly related to phytoplankton concentrations, bloom diminishment is more related to the deepening of the mixed layer depth following wind mixing.

[40] We estimate that MIZ blooms account for about 15% of total NPP in the Southern Ocean, of which two thirds occurs under partial ice coverage. This finding indicates that remote-sensing estimates may substantially underestimate their contribution. MIZ blooms occurring under partial ice coverage are not associated with enhanced NPP over comparable open ocean areas, likely due to lower light availability caused by partial sea-ice coverage. Nevertheless, the high concentration of phytoplankton within the shallow upper reaches of the water column likely provides conditions of enhanced grazing for zooplankton, such as krill, in the postwinter period of enhanced productivity. The finding that light availability is a highly significant predictor of elevated surface phytoplankton concentrations and that these blooms are ubiquitous within the partially ice covered regions of the SIZ supports the hypothesis that krill may use light in the triggering of metabolic changes following winter in preparation for improved feeding conditions.

Appendix A

[41] The coupled numerical biogeochemical ocean general circulation model contains many parameterizations that determine its performance in a global simulation. Here we present those parameterizations that are particularly relevant to our study: iron chemistry, iron cycling, and light transmission through sea ice.

A1 Iron Chemistry

[42] All biogeochemical tracers of REcoM are advected as physically passive tracers with individual local source terms. For the micronutrient iron (Fe), which limits growth via a Michaelis-Menten law with a half-saturation of kFe = 0.12 μmol m−3, the source term consists of iron scavenging following Parekh et al. [2004], and the source term of the macronutrient dissolved inorganic carbon (DIC) scaled by a fixed iron-to-carbon ratio of 0.005. For DIC, the source term consists of respiration of phytoplankton and zooplankton, a sink due to photosynthesis, degradation of extracellular organic carbon. The source term for iron then reads

display math

with a scavenging rate kSC of 0.001 d−1. The free iron Fe′ is computed following Parekh et al. [2004]

display math

where FeL is complexed iron associated with an organic ligand, LT is the total ligand, assumed constant (1), L′ is free ligand, and inline image is the conditional stability constant (100) when the system is in equilibrium.

[43] Iron concentrations are initialized with output of the PISCES model [Aumont et al., 2003]. These concentrations were determined to be too high in the Southern Ocean as compared to iron distribution fields based on in situ measurements by de Baar et al. [1999], with highest overestimation of Fe concentrations in deep waters. The PISCES data is therefore corrected toward lower concentrations in the Southern Ocean. Following this correction, the mean (±standard deviation) initial Fe concentrations (µmol m−3) of the SIZ domain were 0.022 (±0.01) at the surface and 0.326 (±0.124) at about 1000 m.

A2 Light Transmission

[44] The sea-ice model use two “ice” classes: open water and ice. All net fluxes, including shortwave radiation (i.e., light), are computed from the open water fraction and the ice fraction separately and then averaged to give the net flux. In the case of light, the net light that reaches the surface layer of the ocean model is

display math

where c is the fractional ice cover ([0,1]), and QSW,ice/water/net are the ice, water and net downward shortwave heat fluxes. Light can penetrate ice but is attenuated with an exponential law:

display math

with the ice thickness hice measured in meters. The albedo of sea ice αice is a function of temperature. QSW is the incoming shortwave radiation. In the case of a snow cover ice, there is no penetration.

Acknowledgments

[45] The authors are grateful to the German Research Foundation for funding of the project (ID: LO-1143/6). A.B. would also like to acknowledge funding from the Helmholtz-Gemeinschaft Deutscher Forschungszentren e.V. (HGF) Innovative Fund for support of the “Phytooptics” project. The authors thank the following people for their assistance during the study: Bank Beszteri, Tilman Dinter, Stephan Frickenhaus, Judith Hauck, Frank Kauker, Bettina Meyer, Cyril Piou, and Christoph Völker. Special thanks to Jill Schwarz for initiation of the project idea.

Ancillary