Biogeochemical‐Argo floats show that chlorophyll increases before carbon in the high‐latitude Southern Ocean spring bloom

In the Southern Ocean, phytoplankton blooms are an annually recurring prominent feature that play a significant role in ocean CO2 uptake. Understanding the timing of the phytoplankton bloom is necessary to provide insights into the underlying physiological drivers, for the study of ecosystem dynamics and consequent patterns in downward carbon export. Previous studies have used chlorophyll (chl) and particulate organic carbon, from either satellites or biogeochemical‐Argo (BGC‐Argo) floats, to investigate bloom phenology, but provide inconsistent findings regarding bloom timing. Here, we compare bloom dynamics based on three diagnostics from 7114 BGC‐Argo float profiles, south of 60°S. Bloom onset consistently occurs earlier when calculated using chl than when based on phytoplankton carbon or nitrate uptake, and the decoupling increases with latitude. This suggests that phytoplankton synthesize increased chl to acclimate to low‐light conditions, before increasing their biomass. These results highlight the importance of considering phytoplankton physiology when choosing proxies for phytoplankton growth.

profiles, south of 60 S. Bloom onset consistently occurs earlier when calculated using chl than when based on phytoplankton carbon or nitrate uptake, and the decoupling increases with latitude.This suggests that phytoplankton synthesize increased chl to acclimate to low-light conditions, before increasing their biomass.These results highlight the importance of considering phytoplankton physiology when choosing proxies for phytoplankton growth.
Phytoplankton photosynthesis is critical to the biological carbon pump, and annual phytoplankton blooms play a significant role in CO 2 uptake (Sabine et al. 2004;Deppeler and Davidson 2017;Gruber et al. 2019).For the North Atlantic, Sverdrup (1953) stated that the annual phytoplankton bloom is controlled by bottom-up processes including winter mixing of nutrients, followed by thermal stratification during spring, which facilitates phytoplankton growth in the sunlit epipelagic layer.Behrenfeld (2010) argued that bloom initiation occurs in the winter, when mixed layers are deep, with bloom onset resulting from the decoupling between phytoplankton division and loss rates (Behrenfeld andBoss 2014, 2018).It can be argued that bloom initiation sensu Sverdrup is more closely related to the climax of a bloom, that is, the phase of fastest increase in phytoplankton biomass, while Behrenfeld's definition refers to a bloom onset defined as growth terms being larger than losses.Here, we focus on the initiation of the bloom sensu Behrenfeld, which is a useful framework for investigating environmental factors that set the stage for bloom onset (Llort et al. 2015;Uchida et al. 2019a,b).The phytoplankton contribution to carbon sequestration will likely change as stratification, circulation, and nutrient supply respond to future warming (Hauck et al. 2015;Laufkötter et al. 2015;Rembauville et al. 2017;Henson et al. 2019).It is therefore important to understand the presentday dynamics and timing of phytoplankton blooms.
Satellite chlorophyll (chl) and particulate backscatter (b bp ) have been important for understanding the processes that govern bloom dynamics in the North Atlantic (Henson et al. 2009;Behrenfeld 2010), Pacific (Westberry et al. 2016;Graff and Behrenfeld 2018), and Southern Ocean (SO; Moore and Abbott 2000;Arrigo et al. 2008;Thomalla et al. 2011;Westberry et al. 2013;Sallée et al. 2015;Ardyna et al. 2017;Liniger et al. 2020;Kauko et al. 2021).However, phytoplankton can accumulate below the surface, in deep mixed layers, or under ice (Hague and Vichi 2021), and grow, as Behrenfeld (2010) suggested, at the end of the polar night.Satellites are therefore not best suited to observe high-latitude phytoplankton dynamics, because they have limited coverage there, and are compromised by even small amounts of sea-ice cover.Biogeochemical-Argo (BGC-Argo) floats now make it possible to gather information about the vertical structure of phytoplankton stocks, including under ice or at high latitudes in the dark (Mignot et al. 2018;Prend et al. 2019;von Berg et al. 2020;Hague and Vichi 2021).
Recently, bloom initiation studies have used BGC-Argo chl from fluorescence or phytoplankton carbon (C phyto ) from b bp to understand what triggers the bloom in the SO (Ardyna et al. 2019;Uchida et al. 2019a,b;Arteaga et al. 2020;von Berg et al. 2020;Hague and Vichi 2021;Schine et al. 2021).Interpretation of the results is dependent on the variables chosen.Using chl, von Berg et al. (2020) detected bloom initiation in June or July, but using C phyto , Uchida et al. 2019a,b identified bloom initiation in August and September, both in a similar area south of the sea-ice front.Although we acknowledge that the timing of the bloom onset varies across the SO (Sallée et al. 2015;Ardyna et al. 2017), our study aims to investigate systematic differences associated with the choice of variable that is being evaluated.
The mechanisms behind bloom initiation are still under debate (Rohr et al. 2017;Arteaga et al. 2020;Ferreira et al. 2021), and no studies have systematically investigated the differences that arise when chl, C phyto , and dissolved nitrate (NO 3 ) are interrogated concurrently.We hypothesize that (1) bloom phenology differs depending on the variable used and (2) that chl increases before C phyto at the beginning of the growing season.Here, we use high-latitude SO BGC-Argo floats to calculate phytoplankton phenology, based on chl, C phyto , and NO 3 , using the terminology onset (initiation of growth), climax (maximum growth rate), and apex (maximum biomass).We identify differences in bloom onset timing between all three variables and seek to understand the drivers of these differences.We suggest that early chl increases may be driven by light acclimation, while the carbon increase and NO 3 decrease reflect the initiation of phytoplankton growth after physiological priming.

Data acquisition
We used 7114 profiles from 56 floats (Fig. 1), spanning 2012-2021 (Argo 2000;Bittig et al., 2022).To focus on high-latitude processes, we selected floats that recorded data > 60 S for > 1 yr.Analyses were conducted on a year-by-year basis, for a minimum of 11 months from June to April, inclusive.Float data were downloaded (16 October 2021) from the Australian Antarctic Division (AAD) mirroring repository and can be obtained from ftp://ftp.ifremer.fr/ifremer/argo/dac/.Floats measured salinity (psu), temperature ( C), pressure (dbar), chl fluorescence (mg m À3 ), b bp (m À1 ), and NO 3 (μmol kg À1 ) every 10 d.Only adjusted and quality-controlled (QC-ed) data were included (Bittig et al. 2019).Data flagged as QC 4 and 3 were omitted.Floats had a vertical resolution of 5 m in the upper 100 m, 10 m below 100 m, 20 m below 360 m, and 50 m 400-2000 m.Data were interpolated from 4 to 1000 m with a resolution of 4 m in the upper 100 m, and 10 m below.

Float data processing
All chl and b bp data were visually QC-ed, and a running median was applied to both chl and b bp profiles to remove measurement noise and large spikes, or clusters of particles (Su et al. 2021).The adjusted chl data were dark-corrected, non-photochemical quenching-corrected, and divided by 2, reflecting the global factor between the factory calibration and in situ chl (Schmechtig et al. 2015;Roesler et al. 2017).We removed the latter by multiplying by 2, and applied a different slope factor, by dividing by 3.79, to correct for the known bias of phytoplankton iron limitation in the SO (Schallenberg et al. 2022).We calculated phytoplankton carbon biomass (C phyto ) using Eq. 1 (Graff et al. 2015) after converting float b bp (700 nm) to b bp (470 nm) using Eq. 2 (Boss et al. 2013;Boss and Haëntjens 2016): To exclude nonliving phytoplankton particulate organic matter, we applied a mask to the chl and b bp data based on Uchida et al. (2019a,b).Any part of a profile (chl or C phyto ) where chl concentrations were in the lowest 90% of data from below 200 m were ignored.Masked C phyto and chl were then vertically integrated over the upper 200 m and defined as bulk C phyto ([C phyto ]) and bulk chl ([chl]).Bulk NO 3 was defined as mean NO 3 in the MLD.We did not integrate NO 3 from 0 to 200 m because uniformly high values at depth would dominate the signal and reveal little temporal variability.We smoothed the time series with a 60-d running median filter (Chiswell et al. 2022), with 10-d increments (Uchida et al. 2019a,b).Most our floats were from south of the Polar Front and traveled limited distances over the course of a given year (Fig. 1a); so, we expect minimal interference from regional variability on the seasonal signal.However, any residual variability would have been smoothed with the temporal median filter.

Determining bloom phenology based on chl, C phyto , and NO 3
We define the bloom cycle as the period from bloom onset, or initiation, to the end of the growing season (Sallée et al. 2015;Uchida et al. 2019a,b).Bloom onset, climax, and apex were calculated using [chl], [C phyto ], and mean NO 3 (Fig. 2).The rate of change (r) was defined as: r = (1/P)(dP/dt) based on Llort et al. (2015), where P is [chl], [C phyto ], or mean NO 3 .The onset corresponds to the minimum of the bulk variable before it increases, when r = 0.The climax is when the maximum value of r occurs, and the apex is when the maximum value of the bulk variable co-occurs with r = 0 (Behrenfeld and Boss 2018; Uchida et al. 2019a,b).For NO 3 , because it is consumed rather than produced, the onset was at  the maximum value, the apex was at the minimum, and the climax was at the minimum of r.Because for some floats the minimum/maximum did not coincide with r at zero, each variable had a margin applied to the minimum and maximum (AE 0.01 mg m À2 for chl and C phyto , AE 0.001 μmol kg À1 for NO 3 ), and to r (AE 0.05 d À1 for chl and C phyto , and AE 0.008 d À1 for NO 3 ).Then, the earliest point in time where the minimum/ maximum AE margin coincided with r = 0 AE margin was chosen as the onset/apex.
To avoid including years with big data gaps, bloom periods with > 1 missing profile were not included.Onset, climax, and apex were tested for normality using the Kolmogorov-Smirnov test (n > 50).As the data were not normally distributed, a Kruskal-Wallis analysis was performed to test the differences between chl, C phyto , and NO 3 for each bloom stage (Heumann et al. 2016).A Dunn's test was used to confirm which variable occurred earlier than others at significance p < 0.05.A correlation analysis was performed to test whether the observed differences in onsets between chl and C phyto could be explained by temperature and latitude.All analyses were done with MATLAB ver.R2018b.

Bloom onset differs between variables
We compare chl-, C phyto -, and NO 3 -based bloom phenology as measured by BGC-Argo floats in the high-latitude SO (> 60 S), including the sea-ice zone.Although the timings of the climax and apex mostly agree among the variables (Supporting Information Figs.S1, S2), we find a decoupling between chl, C phyto , and NO 3 at the bloom onset (Kruskal-Wallis p < 0.05; Supporting Information Table S1).Bloom onset from chl occurs, on average, 34 AE 8 d (SD) earlier than C phyto onset (Dunn p < 0.05; Figs. 3, 4) and 63 AE 15 d earlier than NO 3 onset (Dunn p < 0.05; Fig. 4).Although the decoupling between onset from C phyto and NO 3 is much smaller, there is still a difference of 29 AE 7 d between them (Fig. 4e,f; Dunn p < 0.05).The larger differences between the onset detected by chl and C phyto are found at higher latitudes, clustered around temperatures < À1.6 C, indicating proximity to sea-ice (Fig. 4a,b).The difference between onsets from chl and from C phyto increases as temperature decreases and latitude increases (Supporting Information Fig. S3).The latitudinal trend indicates that light may be the driver of early increases in chl, and the temperature pattern suggests that light limitation of phytoplankton is exacerbated by sea-ice (r = À0.48 and r = À0.45,respectively).

Seasonal dynamics and spatial trends
In all regions, MLDs are deepest early in the growing season and shoal toward summer.Deepest MLDs are observed at lower latitudes; MLDs are generally shallower in the sea-ice zone (compare Fig. 4b,d).The largest differences in onset from NO 3 and C phyto also occur at lower latitudes and coincide with deep MLDs (Fig. 4f).At the start of the growing season, chl concentrations increase earlier than C phyto (Fig. 5a,b), leading to elevated chl : C phyto ratios early in the season, ranging from 0.01 to 0.04 g chl g C À1 (Fig. 5c).The difference between onset from chl and onset from C phyto increases as the season progresses (Fig. 4a,b,d).That is, the later the onset from chl, the larger the difference between the onset from chl and C phyto .Many initiation events based on chl occur as early as late-June/early-July, when light is low (Fig. 5e) and mixed layers are deep.Such chlbased early onsets can even be observed in the sea-ice zone, while C phyto -based onsets in that region appear later in the season, in October (Fig. 5f).

Discussion
In prior SO BGC-Argo studies, there was a consistent difference of $ 1 month between the bloom onset detected by chl (Prend et al. 2019;von Berg et al. 2020;Hague and Vichi 2021) and C phyto (Uchida et al. 2019a,b;Arteaga et al. 2020;Bisson and Cael 2021).We confirm this lag between chl and C phyto (Fig. 3) for the first time in a systematic review and find that the decoupling is also observed, but larger, between chl and NO 3 .Furthermore, we find that the discrepancy between chl and C phyto increases with latitude (Fig. 4a,c).Our results suggest that the early chl increase is due to phytoplankton synthesizing chl before growing, to acclimate to the high-latitude, low-light environment (Strzepek et al. 2012(Strzepek et al. , 2019;;Graff and Behrenfeld 2018).

Low-light acclimation and iron limitation
The effects of iron limitation on phytoplankton growth in the SO are well studied (Martin 1990;Boyd et al. 2005Boyd et al. , 2007)).Iron stress can decrease bulk chl and decrease chl : C phyto ratios (Greene et al. 1991;Sunda and Huntaman 1997).Analysis of a decade of satellite data found that the seasonal cycle in chl was muted in the iron-limited North Pacific, compared to the ironreplete North Atlantic (Westberry et al. 2016).This was attributed to lower chl per cell (chl : C phyto ) due to iron stress (Marchetti and Harrison 2007;Behrenfeld and Milligan 2013;Westberry et al. 2013).In situ and laboratory studies have similarly shown that chl : C phyto increases under iron-replete conditions (Marchetti and Harrison 2007;Weis et al. 2022).Under iron stress, increases in fluorescence : chl are expected (Schallenberg et al. 2022), which would appear as high float-derived chl in our study.However, early in spring, when chl increases, nutrients, including iron, have not yet been depleted.MLDs are still deep (Fig. 4d,f) and mesoscale eddies are likely supplying nutrients (Uchida et al. 2019a(Uchida et al. ,b, 2020)).Iron conditions are therefore expected to be nonlimiting early in the season (Boyd and Ellwood 2010;Tagliabue et al. 2014Tagliabue et al. , 2017) ) and cannot explain the observed increase in chl.
We suggest that the early bloom onsets detected by chl relative to C phyto and NO 3 are caused by phytoplankton light acclimation and increased cellular chl, a common manifestation of which is an increase in chl : C phyto (Behrenfeld and Boss 2003;Graff et al. 2016;Burt et al. 2018).When phytoplankton are subject to light limitation, they increase cellular chl before growing (Geider 1987;Geider et al. 1998), leading to high chl : C phyto ratios at the beginning of the growing season (Macintyre et al. 2000).Photoacclimation tends to occur when nutrients are high, because iron and nitrogen are needed in the photosynthetic machinery of cells (Kolber et al. 1998).Thus, high chl : C phyto ratios due to light acclimation are expected in high-nutrient-low-light conditions (Arteaga et al. 2016).The high chl : C phyto ratios that we observe in the high-nutrient early spring in the SO (0.01-0.04 g chl g C À1 ; Fig. 5c) are similar to those observed by others (Pollard et al. 2006;Venables and Moore 2010), who also attributed these high ratios to light acclimation (Arteaga et al. 2016;Thomalla et al. 2017;Schallenberg et al. 2019).
The observed increasing latitudinal trend in bloom onset decoupling (Fig. 4a; Supporting Information Fig. S3b), especially when accounting for the presence of sea-ice, further suggests that early increases in chl reflect light acclimation.The discrepancy between chl-and C phyto -based bloom onsets is most pronounced at sea surface temperatures that suggest the presence of sea-ice (Fig. 4b; Supporting Information Fig. S3a).Many chl-based onsets are detected early in the season, around June and July even in the sea-ice zone, where light should be extremely limiting at this time of year (Fig. 5e).The C phyto -based bloom onsets, however, occur later in the season, as sea-ice retreats and is sufficiently thin or patchy to permit light transmission (Bisson and Cael 2021).Given our previous statements about replete nutrients, specifically iron, at this time of year, light acclimation is the most likely driver of the early chl onsets observed.In the sea-ice zone, ice cover (apparent as temperatures < À1.6 C; Fig. 4b) is an important driver of light limitation that is not captured in satellite or reanalysis PAR products.
Limitations of indicators for phytoplankton growth in the high-latitude SO chl and C phyto , derived from fluorescence and b bp , respectively, are proxies of phytoplankton abundance or biomass.Both proxies have some limitations.The conversion from fluorescence to chl is defined by a slope factor, which varies considerably in the SO (Roesler et al. 2017).Its magnitude is affected by species composition, photoacclimation, temperature, and nutrients (Cullen 1982;Schrader et al. 2011;Roesler and Barnard 2013), and it has recently been shown to increase under iron limitation (Schallenberg et al. 2022).We used the slope factor of 3.79, suggested by Schallenberg et al. (2022) for the SO, but we acknowledge potential regional and temporal variability in the fluorescence : chl relationship due to iron limitation.C phyto is estimated from b bp and also has potential biases.Phytoplankton constitute 10-49% of particulate organic carbon in polar waters (Behrenfeld et al. 2005) but the conversion factor proposed by Graff et al. (2015) assumes that C phyto is a constant proportion of b bp .Because most of the backscattering signal comes from particles > 1 μm (Organelli et al. 2018), large phytoplankton can be an important contributor to b bp (Dall'Olmo et al. 2009).Detritus and small zooplankton can also be of similar particle size.We sought to minimize the interference from nonphytoplankton particles by applying a mask, based on chl, to our C phyto estimates (see "Methods" section; Uchida et al. 2019a,b).However, this mask cannot distinguish particles that co-vary with phytoplankton.
In our study, NO 3 data were averaged over the MLD.Ephemeral deepening of MLDs during spring due to storms would entrain NO 3 from depth, dampening or resetting the NO 3 decrease due to uptake by phytoplankton (Gardner et al. 1993).Indeed, the discrepancy between bloom onset from NO 3 and C phyto is largest at lower latitudes, where bloom onsets based on C phyto occur relatively early in the season, at a time when MLDs are still deepening (Fig. 4e,f; Supporting Information Fig. S4).During this deepening, NO 3 uptake cannot be detected, so the bloom onset calculated based on NO 3 is later than that based on chl and C phyto .
Each variable also has an associated sensor error that may affect its sensitivity for detecting change.Adjusted NO 3 data have an accuracy of 2 μmol kg À1 (Johnson et al. 2017a,b), translating to an error of 5-8% of NO 3 concentrations in our study.Uncertainties in b bp are 2 Â 10 À4 m À1 (Bittig et al. 2019) and 2.71 mg m À3 in C phyto (Graff et al. 2015), translating to an error of < 1% for C phyto .chl accuracy is 0.08 mg m À3 (Schmechtig et al. 2014), 2.6-4% of the concentrations in our dataset.We tested the sensitivity of our results to sensor biases by applying randomized errors based on the maximum percentage error for each sensor.Then, we repeated the analysis using the original values with the error added.The observed trends and statistically significant differences in bloom onset, climax, and apex remained unchanged among chl, C phyto , and NO 3 .

Conclusion
We find that chl consistently increases early in the season, before C phyto and NO 3 in the high-latitude SO, and that this systematic lag is driven by light acclimation.chl is therefore not a good growth indicator when investigating the onset of the spring bloom.Phytoplankton growth is defined by cell division, rather than an increase in chl, so it is readily detected in C phyto , making this backscatter-based variable a sensible choice for investigations into the spring bloom.Similarly, NO 3 is essential for phytoplankton growth, to produce cellular proteins (Johnson et al. 2017a,b), but the signal of NO 3 uptake can be masked by deepening mixed layers, making NO 3 a less sensitive choice for detecting bloom onset, especially at the lower latitudes.Overall, in the first systematic analysis of the differences in bloom phenology between three variables, this study highlights the importance of carefully choosing and interpreting indicators for the study of phytoplankton springtime phenology in the high latitudes of the SO, where light is limiting and can drive changes in phytoplankton physiology and chlorophyll cell content.

Fig. 1 .
Fig. 1.Trajectories for all floats used in this study (a), including Float WMO#5905379 indicated by a cyan star at the start of its trajectory, which is used as an example of bloom phenology in Fig. 2. The beginning of each float trajectory is represented by a yellow circle, and the progression of each trajectory is included in the color bar, where each annual cycle starts in June.Southern Ocean frontal zones (subtropical front, polar front, and seasonal ice front) are presented in black and are taken from Arteaga et al. (2019).The maximum winter sea-ice extent, based on satellite data, is presented in a black dash-dot line.Histograms of the latitude of all profiles and for onsets from C phyto are presented in (b) and (c).Profiles and onsets under ice are in brown.

Fig. 2 .
Fig. 2. Phenology calculated with integrated chl (plotted in log 10 space), C phyto (plotted in log 10 space), and mean NO 3 using Float WMO#5905379, spanning June 2018 to May 2019.The black line represents the bulk or integrated variable and the red line represents the rate of change (r) for each variable.The full trajectory of this float is shown in Fig. 1.The onset is shown with a black dot, the climax with a circle and the apex with a star.

Fig. 3 .
Fig. 3. Time of the occurrence for bloom onset (a) from chl and (b) from C phyto .The polar front and seasonal ice front are shown in black and are taken from Arteaga et al. (2019).The maximum winter sea-ice extent, based on satellite data, is in a black dash-dot line.

Fig. 4 .
Fig. 4. Difference in bloom onset time from chl and C phyto , colored by (a) latitude, (b) temperature, and (d) Mixed Layer Depth (MLD).The color bar chosen in panel (b) is used to highlight onsets with a temperature below À1.6 C, indicative of sea-ice, in red.Panels (c), (e), and (f) compare the onset timing from chl, C phyto , and NO 3 colored by (c,e) latitude and (f) MLD.The black line represents the 1 : 1 line between the two variables.Coloring corresponds to the value occurring at the point in time of the y-axis.

Fig. 5 .
Fig. 5. Changes in (a) chl, (b) C phyto .and (c) chl : C phyto ratios for BGC-Argo float WMO#5904768.The red line in the time series represents the Mixed Layer Depth.The trajectory of the float is presented in (d).Monthly climatology (2012-2022) of satellite PAR for (e) June and (f) October, including onsets from chl (yellow circles) and C phyto (red stars) occurring during that month.There are only two chl onsets in October.The red line in panels (d-f) represents the maximum winter sea-ice extent.