Seasonal modulation of phytoplankton biomass in the Southern Ocean

Over the last ten years, satellite and geographically constrained in situ observations largely focused on the northern hemisphere have suggested that annual phytoplankton biomass cycles cannot be fully understood from environmental properties controlling phytoplankton division rates (e.g., nutrients and light), as they omit the role of ecological and environmental loss processes (e.g., grazing, viruses, sinking). Here, we use multi-year observations from a very large array of robotic drifting floats in the Southern Ocean to determine key factors governing phytoplankton biomass dynamics over the annual cycle. Our analysis reveals seasonal phytoplankton accumulation (‘blooming’) events occurring during periods of declining modeled division rates, an observation that highlights the importance of loss processes in dictating the evolution of the seasonal cycle in biomass. In the open Southern Ocean, the spring bloom magnitude is found to be greatest in areas with high dissolved iron concentrations, consistent with iron being a well-established primary limiting nutrient in this region. Under ice observations show that biomass starts increasing in early winter, well before sea ice begins to retreat. The average theoretical sensitivity of the Southern Ocean to potential changes in seasonal nutrient and light availability suggests that a 10% change in phytoplankton division rate may be associated with a 50% reduction in mean bloom magnitude and annual primary productivity, assuming simple changes in the seasonal magnitude of phytoplankton division rates. Overall, our results highlight the importance of quantifying and accounting for both division and loss processes when modeling future changes in phytoplankton biomass cycles.

For the present study, BGC-Argo float data deployed by the SOCCOM program between 2012 and 2019 was analyzed altogether and subdivided into environmental zone as explained in the Methods section. The Southern Ocean presents a clear spatial gradient in surface mixed layer biogeochemical properties (temperature, oxygen, and nitrate) across the four environmental zones defined in this study ( Figure S1). Float-based phytoplankton biomass and growth variables for the mixed layer are initially obtained for each individual float profiles ( Figure S2 and S4), and subsequently averaged to represent the integrated signal of the Southern Ocean and subregions (STZ, SAZ, PAZ, and SIZ) (Figure 1 and 2). The uncertainty in the seasonality of Southern Ocean division rate and phytoplankton net accumulation rate S1 is computed as the standard deviation of the multi-annual time series of µ and r ( Figure   3). Annual climatologies of float-sampled mean mixed layer nitrate, up-to-date compiled dissolved iron observations 1 (Figure S5), and satellite based mixed layer light estimates ( Figure S6) were produced and analyzed for each environmental zone in conjunction with temporal changes in phytoplankton biomass.
Individual float-based estimates of phytoplankton division rates (µ) and net accumulation rates based on changes in mixed layer biomass concentration (r mld ) and integrated inventory (r int ) were obtained as detailed in the Methods section ( Figure S7). The seasonality of r mld is similar to that of r int . However, clear differences exist during periods of mixed layer shoaling or deepening. Net accumulation rates based on the mixed layer integrated inventory of biomass (r int ) are higher than rates based on changes in the biomass concentration (r mld ) during periods of mixed layer deepening, and vice versa ( Figure S8). This pattern is consistent with the expected effect of dilution of the mixed layer phytoplankton concentration during increased surface vertical mixing on the computation of accumulation rates based on biomass concentration, and the expected effect from changes in the vertically integrated water layer on the computation of biomass accumulation based on the integrated phytoplankton carbon inventory in the seasonally varying mixed layer 2,3 . The smoothed time series of r int -r mld and the temporal derivative of the mixed layer (dMLD/dt) are computed as described in the Methods section.

Assessment of division rates estimated by the CbPM
Division rate (µ) outputs from the CbPM were compared against a productivity algorithm parameterized specifically for Southern Ocean waters (Arrigo2008) 4 . The Arrigo2008 algo-S2 rithm computes division rates as: (1) where, following the Arrigo2008 (Equation 8 and 9) notation 4 , µ at a given time (t) and depth (z) depends on G max (t), the temperature (T ) dependent upper limit to net phytoplankton growth rate (i.e., division rate) and an irradiance limitation term (L). G 0 is the phytoplankton net growth rate at 0 • C (0.59 d −1 ) and r is a rate constant (0.0633 • C −1 ) that determines the sensitivity of G max (t) to temperature. The light limitation term, L(z, t), is calculated for each depth and each time step as: where PUR is the photosynthetically usable radiation (assumed equivalent to PAR) and E k is the spectral photoacclimation parameter (see Equations 10-14 of the Arrigo2008 algorithm description for details on the computation of these parameters) 4 .
The temporal evolution of division rate estimated in our study by the CbPM agrees well with that estimated from the Southern Ocean-aimed formulation of Arrigo2008 ( Figure S9a).
In particular, the 'timing' (increase/decrease) of both estimates of µ follows a similar seasonal cycle over the interannual time series of profiling floats observations, providing confidence in the in situ-based observation of a temporal lag between division rates and the net biomass rate of change (r) (Figure 1).
The CbPM allows for a decomposition of the nutrient and light controlling effects on µ.
Nutrient limitation (low nutrient index) is diagnosed to occur during summer months, in opposite fashion to the annual cycle of the light index ( Figure S9b). The impact of nutrient S3 stress on µ is relatively low (i.e., the nutrient index only decreases to about 0.6) which might be due to not explicitly accounting for iron limitation in the model. However, the timing of low nutrient index in summer and high in winter is consistent with the seasonal expectation of micronutrient availability in the Southern Ocean 5 and therefore provides confidence in the ability of the model to detect the correct seasonality of nutrient limitation in this region.
As observed above, a productivity algorithm parameterized specifically for Southern Ocean waters (Arrigo2008) 4 presents a very similar seasonality in µ.
We compare float-based estimates of division rates obtained from the CbPM and Ar-rigo2008 with division rates estimated from a data base of in situ carbon-14 ( 14 C) net primary productivity measurements 6 . In order to infer division rates from in situ-based measurements of vertically integrated NPP, we computed µ = NPP int C phyto ·Zeu where NPP int (mg C m 2 d −1 ) is vertically integrated 14 C-based net primary production over the euphotic depth, A strong correlation between float-based and in situ-based estimates of division rates was not initially expected, since 14 C NPP measurements were combined with satellite biomass data to infer µ, float and in situ data are not spatially coincident, and both data sets S4 were monthly matched but do not coincide perfectly in time (they represent different years).
Despite these sources of discrepancy between data sets, we find that float-based division rates estimated from the CbPM compare well and more favorably with in situ-based µ (R 2 = 0.25, root-mean-square error of the scatter around the least-squares linear fit (RMSE fit ) = 0.24 d −1 , root-mean-square error between linear fit and one-to-one line (RMSE model ) = 0.35 d −1 ) ( Figure S10a) than estimates from Arrigo2008 (R 2 = 0.13, RMSE fit = 0.19 d −1 , RMSE model = 0.58 d −1 ) ( Figure S10b). These results indicate that the CbPM is able to reasonably predict phytoplankton division rates (µ) in the Southern Ocean, thereby, providing a robust validation constraint for the results presented in this study.

Sensitivity of phytoplankton seasonal bloom in the Southern Ocean
The analysis of the sensitivity of the Southern Ocean phytoplankton seasonal bloom magnitude and net primary production is based on the reconstruction of the climatological phytoplankton loss rate for the Southern Ocean based on a 2-days temporal lag in µ 8 ( Figure S11).
The reconstruction of the mean loss rate (l) in the Southern Ocean permitted the assessment of the sensitivity of vertically integrated net primary productivity to induced changes in the climatological seasonal cycle of phytoplankton division rate (µ) ( Figure S12).         Variations in the annual cycle of vertically integrated net primary production (NPP) in the Southern Ocean resulting from relative changes in µ.