Evidence for the Impact of Climate Change on Primary Producers in the Southern Ocean

Within the framework of the Marine Ecosystem Assessment for the Southern Ocean (MEASO), this paper brings together analyses of recent trends in phytoplankton biomass, primary production and irradiance at the base of the mixed layer in the Southern Ocean and summarises future projections. Satellite observations suggest that phytoplankton biomass in the mixed-layer has increased over the last 20 years in most (but not all) parts of the Southern Ocean, whereas primary production at the base of the mixed-layer has likely decreased over the same period. Different satellite models of primary production (Vertically Generalised versus Carbon Based Production Models) give different patterns and directions of recent change in net primary production (NPP). At present, the satellite record is not long enough to distinguish between trends and climate-related cycles in primary production. Over the next 100 years, Earth system models project increasing NPP in the water column in the MEASO northern and Antarctic zones but decreases in the Subantarctic zone. Low confidence in these projections arises from: (1) the difficulty in mapping supply mechanisms for key nutrients (silicate, iron); and (2) understanding the effects of multiple stressors (including irradiance, nutrients, temperature, pCO2, pH, grazing) on different species of Antarctic phytoplankton. Notwithstanding these uncertainties, there are likely to be changes to the seasonal patterns of production and the microbial community present over the next 50–100 years and these changes will have ecological consequences across Southern Ocean food-webs, especially on key species such as Antarctic krill and silverfish.

Increases of atmospheric carbon dioxide (CO 2 ) from ∼400 µatm today beyond 750 µatm by 2100 will likely lead to multifaceted environmental change in the Southern Ocean, including upper-ocean warming, ocean acidification (OA), changes to incident irradiance, increased vertical mixing in the water column, less sea-ice and changed patterns of nutrient input (including iron) (IPCC, 2019;Henley et al., 2020). These environmental and oceanographic changes will affect microbial community composition, patterns of primary production and ecological pathways in Southern Ocean marine ecosystems (Le Quéré et al., 2016;Schofield et al., 2017;Deppeler and Davidson, 2017;Freeman et al., 2019;Johnston et al., in review).
Despite the importance of projecting future changes to primary production and the microbial community composition of the Southern Ocean, current modelling methods have high uncertainty (Leung et al., 2015;IPCC, 2019). The response of the Southern Ocean microbial community to multiple environmental drivers is complex and poorly understood (Petrou et al., 2016;Deppeler and Davidson, 2017). Unlike most of the world's oceans, the vast majority of the Southern Ocean is considered to be replete with nitrate and instead, silicate and iron are limiting in some areas and seasons (Banse, 1996;Boyd et al., 2012Boyd et al., , 2001Hiscock et al., 2003;Doblin et al., 2011). Irradiance can also be crucial to primary production (Deppeler and Davidson, 2017;Kim et al., 2018) and microbial community composition (Arrigo et al., 2010;Kropuenske et al., 2010;Van de Poll et al., 2011;Trimborn et al., 2017) in the Southern Ocean, and irradiance, iron and nutrient availability interact in ways not fully understood at present Luxem et al., 2017;Trimborn et al., 2019). Further complexities arise from colimitation by iron and other micronutrients (e.g., Mn; Pausch et al., 2019). Microzooplankton grazing, which can reduce NPP by facilitating nutrient loss through the sinking of particulate detritus (e.g., Cadée et al., 1992;Perissinotto and Pakhomov, 1998;Vernet et al., 2011), will be affected by climate change through changes to grazing rates (Sarmento et al., 2010;Caron and Hutchins, 2013;Behrenfeld, 2014;Biermann et al., 2015;Cael and Follows, 2016) and phytoplankton nutrient density (Finkel et al., 2010;Hixson and Arts, 2016).
Within the framework of the Marine Ecosystem Assessment for the Southern Ocean (MEASO), we present new analyses of the spatial and seasonal patterns of near-surface chlorophylla concentration (chl-a) and satellite-based proxies of primary production in the Southern Ocean from Earth-observing satellites, and relate this to information from ships, shorestations, and autonomous instruments. We consider both primary production in the surface mixed-layer (from the surface to the depth of the seasonal pycnocline, at ∼50-150 m) and in the 'deep chlorophyll maximum' (DCM; Cullen, 2015;Carranza et al., 2018;Uchida et al., 2019). New approaches to track changes in primary production in the DCM based on satellite observations are proposed. Finally, information on future changes to primary producers from global Earth-system models are presented and discussed in the context of our present understanding of the role of multiple-drivers of changes to primary production and our ability to observe, combine and model these drivers.

Mixed-Layer Primary Production
Both satellite observations of phytoplankton biomass (proxy of chl-a) and satellite-based models were used to describe recent changes to Southern Ocean net primary production ( Table 1). There have been many comparisons between in situ and satellite estimates of chl-a in the Southern Ocean. In reprise, some studies conclude that the 'default' (i.e., globally tuned) satellite algorithm for chl-a should be adjusted to improve accuracy in the Southern Ocean (Johnson et al., 2013;Jena, 2017), whereas other studies have found that the default global chl-a algorithm was appropriate for the Southern Ocean (Arrigo et al., 2008;Haentjens et al., 2017;Moutier et al., 2019;Del Castillo et al., 2019). The reduction in absolute uncertainty in chl-a from adjusting the algorithm for the Southern Ocean tends to be small (e.g., Johnson et al., 2013) and the effect on trends will likely be even smaller, so we used the global default chl-a algorithm for SeaWiFS and MODIS-Aqua, and blended these as described in Pinkerton (2019).
For the same period, estimates of NPP are available from two widely used models: the Vertically Generalised Production Model, VGPM, Behrenfeld and Falkowski, 1997) and the Carbon Based Production Model (CBPM, Behrenfeld et al., 2005;Westberry et al., 2008). Alternative primary production  Reynolds et al. (2002) vgpm mgC m −2 d −1 Net primary production from the Vertically Generalized Production Model (VGPM). Based on blended SeaWiFS and MODIS-Aqua measurements. Behrenfeld and Falkowski (1997) and O'Reilly and Sherman (2016); www.science.oregonstate.edu/ocean.productivity/ models specifically developed for the Southern Ocean have been developed (e.g., Arrigo et al., 2008;Moreau et al., 2015) but data are not widely available and validation is scare. The accuracy of VGPM and CBPM data in the Southern Ocean is not known, so a (non-exhaustive) review of the existing NPP literature was carried out for preliminary comparison. A total of 573 measurements of (almost exclusively summer) primary production were sourced from 24 studies and each was assigned to a MEASO zone and sector based on geographical location ( Table 2) and compared to satellite-based NPP estimates.

Deep-Chlorophyll Maxima
To explore changes in the productivity of phytoplankton at the base of the mixed layer in deep chlorophyll maxima (DCM) we propose a novel metric: irradiance at the base of the mixed layer (E DCM , Equation 2).
Both broadband incident irradiance at the sea-surface (par) and diffuse downwelling attenuation (kpar) were obtained from satellite observations, and estimates of mixed layer depth (mld) were provided from a data-assimilating hydrographic model ( Table 1). Based on instrumented elephant seals in the Southern Ocean, Carranza et al. (2018) showed that DCM tends to occur close to the base of the mixed layer defined using a potential density threshold criterion of 0.03 kg m −3 . Hence, mixed layer depth (z m ) was obtained from GLBu0.08 hindcast results (sourced from orca.science.oregonstate.edu) using a potential density difference of 0.03 kg m −3 from the surface (Metzger et al., 2007;Chassignet et al., 2007;Wallcraft et al., 2009). A constant diffuse sea-surface reflectivity (ρ s ) of 0.07 was assumed (Campbell and Aarup, 1989).
The rationale behind this formulation is that high attenuation in the mixed layer makes it less likely that a DCM is present and vice versa. Essentially, either phytoplankton are distributed relatively evenly through the mixed layer (no DCM, higher mixed-layer attenuation), or are present in a narrow band of elevated concentration at the base of the mixed layer (typically co-located with the nutricline: Cullen, 2015) forming a DCM below a relatively oligotrophic mixed-layer. It is difficult to forecast which of these will occur without good knowledge of the factors involved (inter alia water column structure, nutrient supply/demand, incident irradiance, photoadaptive-capability of phytoplankton species present, loss terms including grazing and sinking; Parslow et al., 2001;Kemp et al., 2006;Cullen, 2015;Carranza et al., 2018;Uchida et al., 2019), but we hypothesize that satellite data may be able to tell us which of these situations has occurred after the event.

2015)
, productivity and biomass of phytoplankton at depth will be affected by other factors, such as phytoplankton composition, nutrient concentrations and temperature (Sallée et al., 2015). In addition, mixed layers can contain vertical structure in optical properties (Carranza et al., 2018). To evaluate the utility of E DCM as an indicator of DCM, a comparison was made between E DCM and the amount of phytoplankton biomass in the DCM from three Southern Ocean Carbon and Climate Observations and Modeling (SOCOMM project; Riser et al., 2018;Uchida et al., 2019) profiling drifters between 2013 and 2018. Phytoplankton biomass in the DCM was proxied from these float measurements as the depth-integrated phytoplankton carbon biomass (C p ) minus the surface phytoplankton carbon biomass multiplied by the mixed layer depth, C p (surf), following Uchida et al. (2019).

Spatial Summaries
Spatial variability in primary productivity was summarised using the MEASO spatial framework (Constable et al., in review; Supplementary Information). Briefly, MEASO defines five longitudinal sectors, three latitudinal zones (Northern between Subtropical and Subantarctic fronts; Subantarctic between Subantarctic and Southern Antarctic Circumpolar Current (ACC) fronts; and Antarctic south of Southern ACC front), and 15 areas from the sector-zone intersects.

Environmental Change in the Southern Ocean
Although autonomous profilers are delivering increasingly powerful datasets (e.g., Boyer et al., 2013;Buongiorno Nardelli et al., 2017;Uchida et al., 2019), large-area (Southern Ocean scale) and long-term (decadal) observations of environmental change in the Southern Ocean are predominantly from Earthobservation satellites (e.g., Cavalieri et al., 1990;Reynolds et al., 2002;Haentjens et al., 2017;Del Castillo et al., 2019), sometimes in conjunction with data-assimilating hydrodynamic models (e.g., Metzger et al., 2007;Chassignet et al., 2007;Wallcraft et al., 2009). To describe the oceanographic and environmental setting, this study focused on 4 key environmental data sets for which large-area, spatially resolved and longterm information is available ( Table 1): sea-surface temperature (sst); sea ice concentration (ice); mixed layer depth (mld); and photosynthetically active radiation at the sea surface (par). Together, these describe major environmental drivers of change in the Southern but clearly do not capture all factors relevant to primary production, including nutrient supply mechanisms, acidification and grazing.

Statistical Analyses
Linear trends in monthly anomalies (differences from climatological means) for each dataset (chl, sst, ice, mld, par, vgpm) at the pixel level (smallest sampling scale) were determined using the Sen slope (Sen, 1968). This value is the median slope of all pairs of points in the time series. The insensitivity of the Sen slope to outliers means that it is generally the preferred non-parametric method for estimating a linear trend (Hipel and McLeod, 1994). Seasonal trends were calculated using anomalies from three months (spring: September-November; summer: December-February; autumn: March-May; winter: June-August). The Sen slope was also used to describe trends at the scale of MEASO zones, sectors and areas. Satellite-based estimates of chl-a and NPP fail at low solar elevations, and we have not attempted to "fill in" the winter gaps in satellite data (e.g., as Park et al., 2019). The proportion of missing data increases with latitude and so to avoid any potential bias in area-averages, trends were only calculated when more than half of potential observations for an area were present. The statistical significance of trends was assessed using the non-parametric Mann-Kendall test (Mann, 1945;Kendall, 1975) which does not require a normal distribution assumption. We used the method of Yue and Wang (2004) to adjust the effective number of degrees of freedom for autocorrelation.
The influence of four environmental drivers (sst, ice, mld, par) on chl-a was considered by a multiple linear regression (Equation 1), where the a coefficients minimize the sum of squares of the error (ε). This analysis was not applied to satellite estimates of primary production (vgpm, cbpm or E DCM ) to avoid circularity: satellite measurements of chl-a are independent of observations of these environmental drivers but vgpm, cbpm and E DCM are not. Here, chl is the monthly anomaly in chl (and similarly for other variables).
An approximation to the contribution to overall trend (Sen slope) in chl-a from each candidate driver variable (sst, ice, mld, par) is given as the product of the respective linear coefficient (α) and the trend (Sen slope) of the variable. The analysis was carried out in IDL 8.5 (Research Systems Inc., Boulder, CO, United States).

Mixed-Layer Primary Production
Although our compilation of in situ measurements of depthintegrated NPP from research vessels and shore-stations was not exhaustive, it nevertheless shows a clear pattern of where primary production has been measured more often to date ( Table 2) and highlights the need to improve knowledge of primary productivity in the entire East Indian sector as well as the Subantarctic and Northern zones where data are scarce or non-existent. The paucity of the NPP data meant that no robust conclusions could be drawn as to the relative accuracies of VGPM and CBPM. Both NPP models look to be overestimating NPP in the Northern zone by a similar amount of ∼40% (Figure 1) and underestimating NPP in the Subantarctic and Antarctic zones by 55 ± 15 % (mean ± standard deviation).
Satellite observations (Figure 2A) show that chl-a was generally higher in Subantarctic areas, lower south of the Polar Front and that there were some areas of elevated productivity south of the southern limit of the ACC (especially in the Ross Sea and Bellingshausen Sea/western Antarctic Peninsula). Patterns of primary productivity estimated from satellite observations (Figures 2C,E) generally followed those of chl-a (higher values in Subantractic waters and over the Antarctic shelf) but there was a strong positive dependence of NPP on latitude. There were also significant differences in spatial variations of annual-average vgpm compared to cbpm, with vgpm higher to the north of the region, and cbpm higher to the south.
Increasing trends in chl-a between 1997-2019 were detected by satellites in most Northern and Subantarctic zones ( Figure 2B). In contrast, decreasing trends in chl-a were observed in most Antarctic continental shelf-sea waters, especially in the Ross Sea, Weddell Sea and Prydz Bay, except along the western Antarctic Peninsula. Trends in vgpm ( Figure 2D) closely followed those in chl-a. Seasonal trend analysis of chl-a and vgpm (see Supplementary Information) shows predominantly positive trends in autumn and negative trends (over the Antarctic shelf) in summer. Trends in cbpm were negative throughout the Southern Ocean ( Figure 2F), including in Northern and Subantarctic zones in contrast with vgpm. Seasonal analysis of trends in cbpm suggests that changes in productivity in the summer drive these overall trends.
In terms of environmental drivers of changes in chl-a, we found that only a small amount (mean 5.6%, 5th-95th percentile range 0.9-16.8%) of the monthly anomalies in chl-a over the last 20 years was explained by a linear combination of sst, ice, mld and par (Figure 3). The proportions of variance explained by the individual drivers were less than 2%, with sst explaining most and par the least. The linearized contribution to the trends in chl-a from these individual environmental drivers (Figure 4) showed little spatial structure and were typically small, less than 0.001 mg m −3 y −1 , whereas the trends in chl-a were up to ∼0.01 mg m −3 y −1 in some areas of the Southern Ocean.

Deep-Chlorophyll Maxima in the Southern Ocean
The regression between irradiance at the base of the mixed layer and the amount of phytoplankton biomass in the DCM from three SOCCOM floats (Figure 5) was highly significant (F 95 = 82.7, p < 0.001) with about half of the variance explained (R 2 = 0.47). Frontiers in Ecology and Evolution | www.frontiersin.org FIGURE 3 | Proportion of variance explained (R 2 ) in anomalies of chlorophyll-a concentration (chl) by a multiple linear regression of anomalies of environmental drivers: sea-surface temperature (sst), sea ice concentration (sea-ice), mixed layer depth (mld) and incident irradiance at the sea-surface (par). Other information as Figure 2.
Average E DCM values were low through most of the Southern Ocean except in a band close to the southern boundary of the ACC and especially in the Atlantic and Central Indian sectors ( Figure 6A). High values of mean E DCM were also found in parts of the Northern zone of the Pacific sector. Trends in E DCM ( Figure 6B) were negligible north of the northern limit of seasonal sea ice, and almost exclusively negative south of this. Decreasing trends in E DCM were greatest between longitude 0 • and 60 • (Atlantic and Central Indian sectors) and occurred almost exclusively during the summer months (December-February: see Supplementary Information).

Trends Summary by MEASO Areas
Significant increasing trends in chl-a and vgpm between 1997-2019 were found in all MEASO zones and sectors except for chl-a in the Antarctic zone (Table 3). Increases were highest in the Atlantic and West Pacific sectors. Significant increasing trends were also found for many (chl-a) or most (vgpm) MEASO areas. Positive trends in vgpm were almost exclusively greater in magnitude and more significant than trends in chl-a (with the exception of AON and WPA areas). In contrast, all significant trends in cbpm were negative, and were highest in Subantarctic and Antarctic zones. For the Southern Ocean as a whole, the mean trend and significance in vgpm was 0.8 % y −1 (p < 0.0001) compared to 0.5 % y −1 for chl (p = 0.002), whereas the overall Southern Ocean trend in cbpm was negative (−0.5 % y −1 ) but not significant (p = 0.17). Significant trends in irradiance at the base of the mixed-layer (E DCM ) over the same period were exclusively negative, and substantially larger than trends in chl-a and vgpm especially in the Subantarctic and Antarctic zones and in the Atlantic sector (trends > 5 % y −1 in magnitude). At the scale of the Southern Ocean, the Sen slope in E DCM was highly significant −3.2 % y −1 (p < 0.0001). Plots of time series by zones, sectors and area, and full statistics on trend analysis are given in Supplementary Information.

Mixed-Layer Primary Production
The present study found that chl-a and NPP were higher in the Atlantic sector than in other sectors of the Southern Ocean, likely as a result of higher iron availability due to land proximity (de Baar et al., 1995;Banse, 1996), and productivity was also elevated around the Kerguelen Plateau and Balleny Islands, consistent with enhanced supply of iron and major nutrients to surface waters (Blain et al., 2007). Annual-average values of chl-a and NPP were also high in some parts of the Antarctic zone, especially Prydz Bay (CIA), Ross Sea (WPA), and Bellingshausen Sea/western Antarctic Peninsula (EPA), likely because of persistent polynyas (Arrigo et al., 2015).
In terms of trends in chl-a, our results agreed with Del Castillo et al. (2019) who found statistically significant increases in chl-a in all sectors of the Southern Ocean, with an especially strong increase in the Northern and Subantarctic zones. Primary production in these zones are seasonally limited by the availability of silicate availability, a constraint which favours phytoplankton communities made up of small flagellates coccolithophores, cyanobacteria, and dinoflagellates (Wright et al., 2010;Balch et al., 2011;Freeman et al., 2019). In these areas, upper-ocean warming in conjunction with higher pCO 2 was anticipated to increase phytoplankton primary production (Steinacher et al., 2010;Boyd, 2019) which agrees with positive trends observed in chl-a (Figure 2).
Further south, silicate-rich waters in the southern Subantarctic and Antarctic zones tend to favour diatom-dominated communities (Petrou et al., 2016;Balch et al., 2016;Rembauville et al., 2017;Nissen et al., 2018;Trull et al., 2018), and here we found the trends in chl-a to be more mixed, with both increases and decreases over the last few decades. The Ross Sea was the main Antarctic area with negative trends in chl-a, but we are not aware of in situ measurements to confirm this.
To date, the most consistent long-term observations of phytoplankton and factors affecting primary production have been made from coastal research stations, notably along the coastal West Antarctic Peninsula as part of the Long Term Ecological Research Network (LTER; Montes-Hugo et al., 2009;Moreau et al., 2015;Kim et al., 2018;Brown et al., 2019). Over the 20-year LTER record, Kim et al. (2018) found significant positive trends in chl-a at some field stations on the Antarctic Peninsula but decreasing phytoplankton biomass at others. Our study shows increases in chl-a along the West Antarctic Peninsula (Figure 2) but the spatial scale of the satellite data is coarser and unreliable within a few km of the shore. Brown et al. (2019) found that increasing upper ocean stability along the West Antarctic Peninsula between 1993-2017 due to a combination of wind, sea ice and meltwater dynamics, led to enhanced primary production, especially of diatoms. It appears that local-scale forcing (e.g., changes to sea ice, glacier melting, changes to coastal current patterns), and large-scale climate cycles like El Niño Southern Oscillation (ENSO) and SAM both affect long-term change in Antarctic coastal productivity (Montes-Hugo et al., 2009;Schloss et al., 2014;Kim et al., 2018;Brown et al., 2019;Höfer et al., 2019).
It is notable that trends from vgpm were predominantly positive throughout the Southern Ocean whereas trends in cbpm were almost exclusively negative (compare Figures 2D,F). The vgpm data are based on chl-a, so it is expected that trends in chl and vgpm agree, whereas cbpm is based on satellite estimates of the C:Chl ratio and does not use chl-a per se. The paucity of in situ measurements of NPP in the Southern Ocean (see Table 2) means that we cannot empirically compare the accuracy of vgpm versus cbpm, or carry out independent trend analysis, but we note that the assumption of nitrate-limited phytoplankton production implicit in vgpm is unlikely to be valid for the Southern Ocean, so cbpm may simply be more reliable than vgpm. Alternatively, the different patterns of change in chl-a (and vgpm) compared to cbpm may be associated with a change in community composition. Laboratory and shipboard experiments show that the responses of phytoplankton to environmental changes such as these will be species-specific (Hoppe et al., 2013(Hoppe et al., , 2017Alvain and d'Ovidio, 2014;Trimborn et al., 2017;Andrew et al., 2019;Strzepek et al., 2019). We speculate here that whereas vgpm is essentially showing trends in phytoplankton biomass (via the proxy of chl-a), cbpm could also be responding to changes in the microbial community, at least in terms of size classes. Decreases in cbpm compared to vgpm in the Southern Ocean would be consistent with a shift towards smaller phytoplankton species at high latitudes at the expense of larger species, in line with some predictions (e.g., Rousseaux and Gregg, 2015;Petrou et al., 2016;Deppeler and Davidson, 2017;Trimborn et al., 2017) and observations (e.g., Moline et al., 2004), although changes in size structure vary spatially and with climatic variability such as SAM and ENSO (Montes-Hugo et al., 2008). Higher scattering efficiencies of smaller species would likely lead to lower satelliteestimates of Chl:C ratios and tend to give lower estimates of growth rates (µ) by the cbpm (Behrenfeld et al., 2005). This suggestion is unproven and warrants further research.
It is increasingly clear that the response of phytoplankton to multiple stressors acting concurrently cannot be obtained by superimposing their separate responses (Boyd and Brown, 2015;Zhu et al., 2016;Boyd et al., 2016;Luxem et al., 2017;Andrew et al., 2019;Strzepek et al., 2019;Trimborn et al., 2019;Boyd, 2019). Satellite observations in the Southern Ocean show complex patterns of environmental change; surface warming in much of the Northern zone contrasts with slight cooling trends over the last 40 years further south (Maheshwari et al., 2013;Kostov et al., 2016;Sallée, 2018); average sea ice concentration has decreased in the Amundsen Sea but increased in parts of the Weddell, Bellingshausen and Ross Seas (Vaughan et al., 2013;Zhang et al., 2018;Pinkerton, 2019;IPCC, 2019); irradiance at the sea-surface has increased north of the Subantarctic Front and generally reduced to the south over the last 20 years; over the same period, mixed-layer depths have likely shallowed in the Northern zone, and both deepened and shallowed in different parts of the Subantarctic and Antarctic zones (Leung et al., 2015;Pinkerton, 2019).
Given the multifaceted and non-linear response of phytoplankton to these environmental changes, it is perhaps unsurprising that the linear model with 4 forcing factors (sst, ice, mld and par) explained only a small fraction (5.6%) of the variance in chl-a over the Southern Ocean (Figure 3), and the individual environmental contributions to chl-a trends were negligible (Figure 4). Also, we recognize that satellite observations were not available for important factors including ocean pH, pCO 2 , and grazing, so these factors were not included as drivers in our empirical analyses of recent changes to chl-a.
Based on long-term sampling from research stations in the Antarctic peninsula region, Kim et al. (2018) showed that variability in chl-a is strongly linked to large-scale climate cycles (ENSO and SAM). Analyses of biogeochemical models are also providing similar insights into these climatebiology relationships acting over large areas (e.g., Hauck et al., 2015). Given the important effects of multi-decadal climate variability such as ENSO and SAM on patterns of primary production, Henson et al. (2010) estimated that ∼40 years of continuous satellite ocean−color data was needed to reliably ascribe any trends in chl-a and production to climate change rather than variability. Del Castillo et al. (2019) repeated the analysis for the Atlantic sector of the Southern Ocean and found that ∼34 years of continuous data were needed. We caution therefore that the satellite record is still not long enough to separate long-term trends from climate variability.

Deep Chlorophyll Maxima (DCM)
Deep chlorophyll maxima (DCM) can form and persist over several months in the Southern Ocean depending on factors including the state of the seasonal pycnocline and nutricline, rates of nutrient supply, incident irradiance, photoadaptation, grazing and sinking (Parslow et al., 2001;Kemp et al., 2006;Cullen, 2015;Carranza et al., 2018;Uchida et al., 2019). Phytoplankton in the DCM have elevated pigment concentrations, likely indicative of production being lightlimited, though iron and silicic acid limitation may also be present (Parslow et al., 2001). The factors affecting when, where and how a deep phytoplankton bloom develops have been extensively studied (review by Cullen, 2015), and although observations generally agree with established hypotheses, the lack of accurate information on key drivers mean that forecasts of DCMs are unreliable (Uchida et al., 2019). For this reason, we presented a new approach to tracking changes in DCM in the Southern Ocean. Based on data from three SOCCOM floats in the Southern Ocean, about half the   for: phytoplankton biomass (chl-a concentration); net primary production (NPP) by the vertically-generalised production model (VGPM) and carbon-based production model (CBPM); and irradiance at the base of the mixed-layer (E DCM ) proxy of primary production in the deep chlorophyll maximum (DCM).  variability in the amount of phytoplankton biomass in the DCM was explained by our simple metric of irradiance at the base of the mixed layer (E DCM ). The comparison shown in Figure 5 provides preliminary support for E DCM being a useful metric for tracking changes in DCMs in the Southern Ocean.
The new E DCM metric revealed a significant decrease in NPP at depth in the Antarctic sector of the Southern Ocean (Figure 6), but there were not co-located hotspots of trends in chl-a, diffuse irradiance attenuation, or mixed layer depth. The interpretation is that relatively small changes over time to attenuation and mixed-layer depth acting together can lead much more significant trends in E DCM . Further analysis of this preliminary result is recommended as variations in DCMs could have important consequences for ecosystems and biogeochemistry because of higher efficiencies of organic matter export from sub-surface primary production (Tilstone et al., 2017;Henley et al., 2020).

Mixed-Layer Primary Productivity
Based on Earth-systems models, the 'Special Report on the Ocean and Cryosphere in a Changing Climate' (IPCC, 2019; Meredith et al., 2019) included a summary of observed changes in NPP and drivers, and projected future changes. Future projections were based largely on the CMIP5 (Coupled Model Intercomparison Project) which used two Representative Concentration Pathways (RCPs): RCP2.6 (low greenhouse gas emission, high mitigation future) and RCP8.5 (high greenhouse gas emission scenario, 'business as usual' , in the absence of policies to reduce climate change). Key conclusions from CMIP5 model projections of NPP (Leung et al., 2015;Meredith et al., 2019) were as follows: • In the Northern MEASO zone, higher mean underwater irradiance (from reduced mixed-layer depths) and higher iron supply were projected. Overall, this projection points to increased primary production in the mixed-layer and increased phytoplankton biomass (Leung et al., 2015;Meredith et al., 2019). These model projections are consistent with recent observations provided in the current study (chl-a, vgpm) and elsewhere (e.g., Le Quéré et al., 2005;Doney, 2006;Del Castillo et al., 2019). • In the Subantarctic zone, deeper summertime mixedlayer depth together with increased cloud albedo were projected to lead to lower average irradiance in summer, leading to lower chl-a and NPP (Leung et al., 2015). This result agreed with a modelling study by Moore et al. (2018) which found that changes to sea ice and circulation patterns under climate warming scenarios would lead to a global reorganization of nutrient distributions and a steady decline in global-scale marine biological production. However, recent satellite observations show recent increasing rather than decreasing trends in chl-a in this zone (Del Castillo et al., 2019;present study). • In the Antarctic zone, CMIP5 models generally suggested that less seasonal sea ice and a warming ocean will lead to greater iron supply, higher underwater irradiance and thence to increases in chl-a and NPP (Leung et al., 2015;Rickard and Behrens, 2016). These projected increases agree with recent trends in chl-a and vgpm (present study), except in the Ross Sea where chl-a, vgpm and cbpm all show negative trends.
Confidence in future projections of chl-a and NPP remains low (IPCC, 2019). The strongest drivers of future changes in NPP in CMPI5 models were iron availability and irradiance; acidification and temperature per se were less important (Leung et al., 2015). The balance between different iron-supply mechanisms (i.e., vertical mixing versus aeolian versus ice-mediated supply) are hence crucial to our ability to forecast future changes to the productivity of the Southern Ocean Boyd et al., 2014;Leung et al., 2015;Hutchins and Boyd, 2016;Hopwood et al., 2019). High spatial and seasonal variability in the relative importance of various iron-supply mechanisms, coupled with a lack of long-term observations, the simultaneous change of a number of environmental conditions, and the complexities of phytoplankton response to different environmental drivers hence severely limit our ability to anticipate these changes (Hutchins and Boyd, 2016;Mongwe et al., 2018;Freeman et al., 2019;Smith et al., 2019;Hopwood et al., 2019). In the longer term, longer-time series of satellite observations, increasingly sophisticated laboratory experiments, and more co-ordinated and extensive Antarctic observations such as the Southern Ocean Observing System (SOOS) are expected to help provide the information required to improve these future projections (Newman et al., 2019).

Deep Chlorophyll Maxima (DCM)
Prognoses of future changes in production in the DCM in the Southern Ocean are not reliable and Earth system models focus instead on depth-integrated estimates of NPP (e.g., Leung et al., 2015;IPCC, 2019). Better satellite-based observation of the occurrence of DCMs in the Southern Ocean using the simple metric described here (E DCM ) and better in situ observations using autonomous technology could improve this situation in the future. In particular, the advent of biogeochemical Argo floats (Rembauville et al., 2017;Briggs et al., 2018;Carranza et al., 2018), the SOCOMM project (Riser et al., 2018;Uchida et al., 2019) and Southern Ocean and Climate (SOCLIM) floats 1 represent a major step forward.

Primary Production by Sea Ice Algae
Although not a focus of the present study, we note that tracking and anticipating changes to primary production in the Southern Ocean should include that within sea ice (Arrigo, 2014;Saenz and Arrigo, 2014;van Leeuwe et al., 2018) and beneath it (Arteaga et al., 2020). Although sea ice algae only contribute ∼1% of total Southern Ocean primary production, and 12-50% of total primary production in the sea ice zone Grossi et al., 1987;Saenz and Arrigo, 2014;van Leeuwe et al., 2018), sea ice algae are of disproportionate ecological importance because their production occurs in locations and at times when production in the water column is low (Quetin et al., 1996;Saenz and Arrigo, 2014;McCormack et al., in review). As such, sea ice algae are a crucial bridge between low and high productivity periods for many mid-trophic level species, such as Antarctic krill (Euphausia superba; Daly, 1990;Smetacek et al., 1990;Loeb et al., 1997;Kohlbach et al., 2017;Meyer et al., 2017), and Antarctic silverfish (Pleuragramma antarctica) (Guglielmo et al., 1998;Vacchi et al., 2004).
Drivers of primary production by sea ice algae include iceextent and thickness, incident light availability, and nutrient supply (Kottmeier and Sullivan, 1990;Arrigo et al., 1998;Arrigo, 2014;Hobbs et al., 2016;Tedesco and Vichi, 2014), though temperature and salinity can be important (Tedesco and Vichi, 2014;Saenz and Arrigo, 2014). Variations in snow depth will influence both light penetration into sea ice, and nutrient input (Saenz and Arrigo, 2014). Both laboratory manipulations and in situ experiments indicate that sea ice algae are little affected by changes to pH (McMinn, 2017).
Complex models have been developed to simulate the largescale primary production by sea ice algae (Arrigo et al., 1991(Arrigo et al., , 1997Arrigo and Sullivan, 1994;Saenz and Arrigo, 2012;Arrigo, 2014;Tedesco and Vichi, 2014) but these have neither been well validated to date (Meiners et al., 2012) nor used to investigate long-term changes in sea ice algae primary production. Furthermore, no future projections of changes to sea ice algae production are available in CMIP5 models (IPCC, 2019).
In the longer term (∼100 years hence) it is likely that sea ice algae primary production will decrease in line with reducing sea ice extent and concentrations throughout the Southern Ocean (IPCC, 2019), but these changes are likely to be spatially heterogenous and non-linear (van Leeuwe et al., 2018). We also note that the ecosystem consequences of a reduction of primary production in sea ice could be significant and far reaching (Atkinson et al., 2004;Meyer et al., 2017;Saenz and Arrigo, 2012;Arrigo, 2014;Tedesco and Vichi, 2014;McCormack et al., in review) Figure 7) shows that primary production in the Southern Ocean will likely increase over the next 100 years, with more primary production in the water column and less in sea ice. The distinctive nature of the microbial community of the Southern Ocean (i.e., the dominant species of phytoplankton, phenology (seasonality) and spatial distribution of primary production) will likely reduce, with shifts towards communities more dominated by small and flagellated species, and with endemic Southern Ocean phytoplankton species losing out to temperate species (Deppeler and Davidson, 2017). Year-to-year variability in primary production in the Southern Ocean will likely increase with increasing prevalence of marine heatwaves (Oliver et al., 2019). Recent increases in sea ice concentration (and potentially associated sea ice algae production) will likely reverse in the next few decades and decrease consistent with less ice in a warming Southern Ocean (IPCC, 2019). The importance of local-scale forcing means that forecasting the effects of climate change on coastal primary production, crucial to the reproductive success of many endemic Antarctic species, is especially challenging (Montes-Hugo et al., 2009;Kim et al., 2018). The effects of future changes to primary production are likely to be greatest for those Southern Ocean species with life-cycles intimately linked to seasonal sea ice growth and retreat, including 'keystone' species such as Antarctic krill and silverfish (Smetacek et al., 1990;Quetin et al., 1996;Loeb et al., 1997), and their predators (McCormack et al., in review).

KM1
Satellite observations show mainly increases in phytoplankton biomass in the Southern Ocean over the last 20 years (1997-2019) (high confidence), but the direction of recent changes to primary production are generally not known except to say that it is likely that primary production has decreased in the Ross Sea over the recent past (medium confidence).

KM2
At present the satellite record is not long enough to separate longterm trends from climate variability, so we cannot say whether these observed trends will continue in the future (Henson et al., 2010;Del Castillo et al., 2019) (high confidence).

KM3
Over the next 100 years, Earth system models project increasing primary production in the northern and Antarctic MEASO zones but decreases in the Subantarctic zone (Leung et al., 2015;Rickard and Behrens, 2016) (low confidence). Low confidence in these projections arise from the difficulty in mapping supply mechanisms for the key nutrients of iron and silicate in the Southern Ocean and understanding the effects of multiple stressors on different species of Antarctic phytoplankton Arrows show recent trends, in order: phytoplankton biomass in the mixed-layer (chl-a concentration, 1997-2019, chl); net primary production (NPP, 1997-2019) by Vertically Generalised Production Model (vgpm) and by Carbon Based Production Model (cbpm); irradiance at base of the mixed-layer (1997-2019, E DCM ). The arrows show: ↑ indicates a significant increasing trend > 1% y −1 ; significant increasing trend < 1% y −1 ; -no significant trend over the period analysed; significant decreasing trend < 1% y −1 ; ↓ significant decreasing trend > 1% y −1 . So, for example, "−↑ −" indicates no trend in chl, strongly increasing trend in vgpm, decreasing trend in cbpm, and no trend in E DCM . 'Future projections' are based on CMIP5 models (Leung et al., 2015; Figure 1) just for NPP: ↑ increases projected for the future;no changes (or a mixture of increases and decreases) projected; ↓ decreases projected for the future.
FIGURE 7 | There are different types of primary production in the Southern Ocean -in sea ice, in the surface mixed-layer of the ocean, and in the deep chlorophyll maximum (DCM). Only some of these types of primary production can be observed by satellite over the last 20-30 years. Earth-system models can project future changes to primary production in the next 50-100 years. Changes vary between the Northern, Subantarctic and Antarctic zones. (Hutchins and Boyd, 2016;Boyd et al., 2016Boyd et al., , 2019Deppeler and Davidson, 2017;Freeman et al., 2019;Meredith et al., 2019;Trimborn et al., 2019).

KM4
As well as affecting the total amount of primary production in the Southern Ocean, climate change will likely alter seasonal patterns in production (phenology) and the relative abundances of different types of phytoplankton (Wright et al., 2010;Petrou et al., 2016;Balch et al., 2016;Kaufman et al., 2017;Nissen et al., 2018;Trull et al., 2018) (high confidence). These changes will affect zooplankton, have ecological consequences across Southern Ocean food-webs including on keystone species and top predators (Atkinson et al., 2004;Moline et al., 2004;Pinkerton and Bradford-Grieve, 2014;Johnston et al., in review;McCormack et al., in review) (high confidence).

KM5
Primary production by sea ice algae will likely decline in the future as sea ice extent shrinks (medium confidence). Because of the ecological importance of sea ice to a range of key Southern Ocean species, the reduction in Southern Ocean sea ice and its associated primary production could lead to a critical tipping point in Southern Ocean ecosystems (medium confidence).

KM6
Better methods of monitoring change to phytoplankton and sea ice algae are needed to improve confidence in future projections of primary production in the Southern Ocean (high confidence). Efforts to maintain long-term coastal observations (e.g., Palmer Long Term Ecological Research Network LTER, e.g., Kim et al., 2018), improve satellite observations, co-ordinate field sampling, and develop and deploy autonomous instrumentation in the Southern Ocean are essential to assess long-term trends (Newman et al., 2019) (high confidence). Long-term, multi-trophic level monitoring is required to understand the effects of changes to primary production on middle and upper level predators such as krill, salps, fish, seabirds and marine mammals (high confidence).

FREQUENTLY-ASKED QUESTIONS (FAQ)
What is primary production and why is it important?
Primary production is the formation of organic matter by photosynthesis. It is the process by which energy enters the marine food-web. All marine animals ultimately rely on organic matter created by primary production. Why can't satellites see ice algae?
Ice algae typically grow within or below sea ice, so that the sea ice stops satellites seeing the colour of the algae directly.
How does climate change affect primary production?
Primary production depends primarily on the amount of light available and on the concentration of nutrients in sea-water. If the amount of cloud cover, sea ice or snow changes, this affects the amount of light entering the water column. The depth of mixing in the ocean also affects the light availability. Many processes affect nutrient supply, including those associated with water column mixing, dust input, circulation patterns, icebergs, snow and sea ice. There are lesser effects on phytoplankton due to warming, acidification, carbon dioxide concentration and indirect effects from changes to zooplankton grazers. Why is it so hard to predict what will happen to phytoplankton?
There are many different species of phytoplankton in the Southern Ocean and all respond differently to environmental changes. The effects of multiple stressors acting on species at the same time also makes it difficult to anticipate phytoplankton responses to global change. Why do the predictions vary with location?
Ocean physics differ drastically among regions of the Southern Ocean, which means that environmental changes in the Southern Ocean are not the same everywhere. For example, some areas are warming and other cooling; some areas have more ice and some less. Also, phytoplankton communities are not the same everywhere, and different species respond to environmental change in different ways.

AUTHOR CONTRIBUTIONS
All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

FUNDING
Funding for MP, SD, and AH was provided by New Zealand MBIE Endeavour Fund programme Ross-RAMP (C01X1710) and NZ MBIE Strategic Science Investment Fund: "Coasts and Oceans Programme 4". JH was supported by ANID (FONDAP-IDEAL 15150003) and FONDECYT (Postdoctorado 3180152). ICED partially supported the assistance of JH to the MEASO workshop. PB was supported by the Australian Department of Industry funded ACE-CRC and AAPP programmes.

ACKNOWLEDGMENTS
SeaWiFS and MODIS data were used courtesy of NASA. We acknowledge the HyCOM project for access to mixed-layer depth data. We thank the University of Oregon (United States) for primary production data. Sea ice data were accessed courtesy of the United States National Snow and Ice Data Centre. We also thank Dr. Stacey McCormack (Institute for Marine and Antarctic Studies, University of Tasmania) for the Infographic (Figure 7). We are grateful to two referees and the editors of the Special Issue who provided valuable feedback and suggestions that helped improve this manuscript.