Factors controlling the competition between Phaeocystis and diatoms in the Southern Ocean and implications for carbon export ﬂuxes

Abstract. The high-latitude Southern Ocean phytoplankton community is shaped by the competition between Phaeocystis and silicifying diatoms, with the relative abundance of these two groups controlling primary and export production, the production of dimethylsulfide, the ratio of silicic acid and nitrate available in the water column, and the structure of the food web. Here, we investigate this competition using a regional physical–biogeochemical–ecological model (ROMS-BEC) configured at eddy-permitting resolution for the Southern Ocean south of 35∘ S. We improved ROMS-BEC by adding an explicit parameterization of Phaeocystis colonies so that the model, together with the previous addition of an explicit coccolithophore type, now includes all biogeochemically relevant Southern Ocean phytoplankton types.
We find that Phaeocystis contribute 46±21 % (1σ in space) and 40±20 % to annual net primary production (NPP) and particulate organic carbon (POC) export south of 60∘ S, respectively, making them an important contributor to high-latitude carbon cycling.
In our simulation, the relative importance of Phaeocystis and diatoms is mainly controlled by spatiotemporal variability in temperature and iron availability. In addition, in more coastal areas, such as the Ross Sea, the higher light sensitivity of Phaeocystis at low irradiances promotes the succession from Phaeocystis to diatoms. Differences in the biomass loss rates, such as aggregation or grazing by zooplankton, need to be considered to explain the simulated seasonal biomass evolution and carbon export fluxes.



Supplementary material
The supporting information provides additional figures in section S1 with respect to the nutrient limitation of phytoplankton growth in ROMS-BEC (S1), the ecological niche analysis (S2-S3), the data coverage in a SO satellite derived chlorophyll product (S4), the model evaluation (S5-S8), the bloom timing (S9), the competition sensitivity simulations (S10), carbon cycling in the Ross Sea (S11), and the results when using a varying half-saturation constant of iron for Phaeocystis growth (S12). In section S2, results of the parameter sensitivity simulations are described (Table S1-S3, Fig. S13).   Overlain are the simulated area and biomass weighted ecological niche centers (median, triangle) and breadths (inter quartile ranges, dashed lines) for the three functional types. Figure S4: Assessment of the SO data coverage in the climatological (1998-2018, i.e. 21 years) daily Globcolor chlorophyll product (Fanton d'Andon et al., 2009;Maritorena et al., 2010): a)-f) Average number of years available for the calculation of the climatological chlorophyll concentration at each grid cell for each of the shown months (October-March), respectively. No minimum number of "days with data coverage" is required for a given month to be counted as "data available" (i.e. one day of data coverage in a month is enough for that month to be counted as "covered" in the respective year). g) Average number of years available for the calculation of the climatological chlorophyll concentration on each day for 10 • latitudinal bands across the SO. in ROMS-BEC (circles) and in observations (squares, Vogt et al., 2012) for each month between November-February and in the the upper 50 meters of the water column. For panels b)-d), the model output is colocated with observations in space and time, and observational data from all months and from above 1000 m are considered here (Balch et al., 2016;Saavedra-Pellitero et al., 2014;O'Brien et al., 2013;Vogt et al., 2012;Leblanc et al., 2012;Tyrrell and Charalampopoulou, 2009;Gravalosa et al., 2008;Cubillos et al., 2007). For more details on the biomass evaluation, see Nissen et al. (2018). The dotted line shows the perfect linear 1:1 fit, whereas the solid line is the actual fit of the data (linear regression). Pearson correlation coefficients of these regressions are given in the top right, those for Phaeocystis and coccolithophores are statistically significant (p<0.05). Points are color-coded according to the sampling latitude. Figure S6: a)-c) Relative contribution of the five phytoplankton PFTs to total chlorophyll biomass [mg chl m −3 ] for a) 30-90 • S, b) 60-90 • S, and c) the Ross Sea. The top pie charts denote the climatological mixed layer average community composition suggested by CHEMTAX analysis of HPLC pigments for spring, summer, and fall, respectively (the total number of available observations for a given region and season is given at the lower left side, Swan et al., 2016), and the lower pie charts denote the corresponding community structure in the top 50 m in ROMS-BEC in the 5-PFT setup (middle row, same as in Fig. 2 in the main text) and in the 4-PFT setup (lowest row, no Phaeocystis, Nissen et al., 2018), respectively. Note that the categories in the CHEMTAX analysis are not 100% equivalent to the model PFTs, and here, "Hapto-8 reassigned" corresponds to the contribution of Hapto-6 where the temperature is <2 • C (see also section 2.3.1 in the main text).      (Zoo) to the combined phytoplankton and zooplankton biomass (green) and total POC production (yellow) in the top 100 m, respectively. The arrows denote the relative contribution of the different POC production pathways associated with each PFT (black = grazing by zooplankton, grey = aggregation, blue = non-grazing mortality), given as % of total NPP in the top 100 m. Numbers are printed if ≥0.1% and rounded to the nearest integer if >1%. The sum of all arrows gives the POC production efficiency, i.e., the fraction of NPP which is converted into sinking POC upon biomass loss (p ratio). Note that diazotrophs are not included in this figure due to their minor contribution to NPP in the model domain. b)-d) Simulated vertically integrated production of particulate organic carbon (POC) b) as a function of time [mmol C m −2 d −1 ], c) cumulative over time (absolute production in Pg C yr −1 on the left axis and relative to annually integrated production on the right axis), and d) as a function of time via grazing and aggregation, respectively. The colors correspond to the different PFTs in ROMS-BEC, and the panels correspond to averages or integrals over the Ross Sea. and diatoms (dotted) used in the Baseline simulation of this study. c) Difference in days in the timing of the bloom peak of diatoms and Phaeocystis for each latitude, with negative values denoting a succession from Phaeocystis to diatoms throughout the season. d) Difference in day of bloom peak between Phaeocystis and diatoms. Stippling indicates locations where maximum chlorophyll concentrations never exceed 0.1 mg chl m −3 for Phaeocystis (orange) and diatoms (green), respectively. White areas correspond to areas where the peak total chlorophyll concentrations do not exceed 0.5 mg chl m −3 . Decrease γ PA a,0 by 50% grazing150

S2: Parameter sensitivity experiments
Increase γ PA g,max by 50% Param grazing grazing50 Decrease γ PA g,max by 50% thetaNmax50 Increase θ PA chl:N, max by 50% Param thetaNmax thetaNmax50 Decrease θ PA chl:N, max by 50% In order to more systematically quantify the sensitivity of simulated distributions of Phaeocystis and diatoms and integrated estimates of NPP and POC export in ROMS-BEC to Phaeocystis model parameter choices, we have performed a set of model parameter sensitivity experiments. To that aim, we have systematically increased/decreased all key Phaeocystis parameters by 50%, allowing for an objective ranking of model sensitivities. We varied the following seven parameters of Phaeocystis, resulting in a total of 14 simulations: the temperature optimum, the half-saturation constant of iron, α PI , the maximum chl:N ratio θ chl:N, max , the linear mortality rate, the quadratic mortality rate (aggregation), and the maximum grazing rate of zooplankton on Phaeocystis (see Table S1).
We then quantify the sensitivity S of any target variable A (here A being one of the following targets: total phytoplankton, Phaeocystis, and diatom chlorophyll concentrations, total NPP, and POC export across 100 m) to changes in the parameter X as follows, allowing for a ranking of the seven sets of simulations by the magnitude of the sensitivity (see Table S1): As expected (see also Nissen et al., 2018), we find that both total chlorophyll concentrations and chlorophyll levels of Phaeocystis and diatoms are highly sensitive to parameters describing the growth and loss of Phaeocystis biomass, with increases of up to 700% (grazing50) and declines of up to >90% (Topt50, thetaNmax50) in Phaeocystis biomass between 60-90 • S for a 50% change in the associated parameters (see Fig. S13). In general, any decline/increase in Phaeocystis chlorophyll biomass is associated with an increase/decline in diatom chlorophyll biomass, pointing to the direct competition for resources of these two phytoplankton types at high SO latitudes. Yet, the biomass compensation is not always complete due to non-linearities in the model system (e.g. food web feedbacks), resulting in changes of up to 70% (grazing150) in total chlorophyll levels upon changes in Phaeocystis parameters. The ranking of model sensitivities between 60-90 • S reveals the highest sensitivity of Phaeocystis and diatom chlorophyll concentrations to the maximum grazing rate γ PA g,max , the maximum chl:N ratio θ PA chl:N, max , the initial slope of the photosynthesis-irradiance curve (α PA PI ), and the temperature optimum T opt of Phaeocystis growth (Param grazing, Param thetaNmax, Param alphaPI, Param Topt Figure S13: Annual mean surface chlorophyll concentrations of all phytoplankton (total Chl), Phaeocystis (P A), and diatoms (D) in the parameter sensitivity simulations (see Table S1) relative to the Baseline simulation. The model output is averaged over a) 60-90 • S and b) the Ross Sea.
in Table S1 & S2). In comparison, the opposed changes in Phaeocystis and diatom chlorophyll levels (see Fig. S13) result in lower sensitivities of total chlorophyll levels to changes in Phaeocystis parameters in general and a lower ranking of the temperature optimum and thetaNmax experiments in particular (Param Topt and Param thetaNmax in Table S2).
In comparison to the ranking of model experiments for total chlorophyll, the model sensitivities for NPP and POC export across 100 m are similar in magnitude both between 60-90 • S and in the Ross Sea (20-90%, compare Table S2 & Table S3). Additionally, the ranking of model experiments for NPP and POC export reveals only small differences to the ranking of model sensitivities for total chlorophyll: While the experiments Param alphaPI and Param grazing consistently rank amongst the top two most sensitive experiments for NPP and POC export and between 60-90 • S for total chlorophyll concentrations, the experiments Param mortality/Param Topt are less/more important for NPP and POC than for total chlorophyll levels in ROMS-BEC (compare Table S2 & S3). In summary, this demonstrates the large model sensitivity of bulk biogeochemical quantities to parameter choices describing the temperature and light dependence of Phaeocystis growth and zooplankton grazing. Table S2: Ranking of the parameter sensitivity experiments by the absolute sensitivity of annual mean total surface chlorophyll (|S Chl X |), Phaeocystis chlorophyll (|S Chl PA X |), and diatom chlorophyll (|S Chl D X |) to a ±50% change in the model parameter X relative to the Baseline setup of ROMS-BEC between 60-90 • S and in the Ross Sea, respectively. The sensitivity S (%) is quantified using Eq. 1. See Table S1 for details on the experimental setup and Fig. S13 for details on the resulting chlorophyll fields in ROMS-BEC in each experiment. Note that the simulated changes in carbon biomass fields are qualitatively similar to those of chlorophyll (not shown) and that the ranking shown here is therefore insensitive to the choice of chlorophyll in the analysis.  Table S3: Ranking of the parameter sensitivity experiments by the absolute sensitivity of annually integrated NPP (|S NPP X |) and POC export across 100 m (|S POC 100m X |) to a ±50% change in the model parameter X relative to the Baseline setup of ROMS-BEC between 60-90 • S and in the Ross Sea, respectively. The sensitivity S (%) is quantified using Eq. 1. See Table S1 for the experimental setup.