Southern Ocean regime shift: divergent trends in surface phytoplankton community composition using a diagnostic pigment framework

Abstract

Phytoplankton communities in the Southern Ocean (SO) play a central role in regulating global climate and ecosystems through the biological carbon pump (BCP), which is influenced by ocean warming and sea-ice variability. Here, we combine an expanded circumpolar surface pigment dataset with satellite ocean-colour observations to develop a SO-diagnostic pigment framework that resolves phytoplankton community composition for austral spring and summer between 1998 and 2024. We compare multi-year mean composition for 1998-2014 and 2015-2024 periods separated by a  ~ 7% step decrease in September sea-ice extent. Between these two periods, the mean diatom contribution to total chlorophyll increased by up to  ~ 10% during spring in sea-ice-influenced regions but decreased by up to  ~ 3% over the Antarctic shelf during summer. These findings suggest a seasonally and spatially variable reorganisation of phytoplankton community composition that may alter bloom dynamics, carbon-export efficiency, and the functioning of the SO BCP under continued climate-driven sea-ice variability.

The study reveals that Southern Ocean phytoplankton communities reorganised after the 2014 Antarctic sea-ice decline, with contrasting spring and summer responses that could reshape marine food webs and carbon storage.

Introduction

Phytoplankton are central to the Southern Ocean (SO) ecosystem by supporting Antarctic food webs and mediating carbon export through the biological carbon pump (BCP), which transfers organic carbon from the euphotic zone to the deep ocean^1–4. In addition to this biological pathway, the SO plays a key role in global carbon uptake through physically driven processes, particularly the solubility pump and large-scale ocean circulation. Together, these mechanisms enable the SO to account for a disproportionate share of global oceanic carbon uptake⁵.

The solubility pump governs the uptake and storage of atmospheric CO2 in dissolved form, whereas the BCP redistributes biologically fixed carbon into the ocean interior, contributing to long-term carbon storage². Consequently, changes in the strength and structure of the BCP can influence carbon sequestration efficiency and the ocean’s capacity to store carbon over long timescales⁶. Model projections further indicate that the SO carbon cycle responds heterogeneously to climate change with both increases and decreases in productivity and export depending on regional variations in sea-ice cover stratification and nutrient supply^6,7. For example, reduced sea ice cover can extend the growing season and enhance primary production in sea ice-influenced regions such as the seasonal sea ice zone (SSIZ)^3,8. Stronger winds and deeper mixed layers may increase nutrient supply and support enhanced productivity in regions such as the Subantarctic Zone (SAZ)⁹. In contrast, increased stratification and reduced vertical mixing can limit nutrient resupply to the euphotic zone, leading to declines in productivity and carbon export in other regions^6,7. These coupled physical and biogeochemical processes drive variability in phytoplankton biomass and productivity across SO regions and timescales^3,10.

Beyond total biomass, environmental variability also shapes phytoplankton functional types (PFTs) and size classes (PSCs), which differ in nutrient requirements, morphology, sinking behaviour and trophic roles^11–13. Large silica-rich diatoms typically support krill-based food webs and efficient carbon export, whereas smaller phytoplankton such as flagellates, including haptophytes and cryptophytes, more often sustain the microbial loop with lower export efficiency^2,14–17. Consequently, shifts in PFT and PSC composition can substantially alter ecosystem structure and modulate the strength of the SO BCP.

Recent evidence indicated that SO phytoplankton communities are undergoing restructuring in response to climate forcing. Pigment-based chemotaxonomic tools (e.g., Phytoclass and CHEMTAX) have enabled the reconstruction of multi-decadal summer trends across the Antarctic Shelf and SSIZ^18,19. Machine learning-based reconstructions using in situ pigment and environmental data reveal long-term shifts in summertime Antarctic shelf phytoplankton communities, including a decline in diatom chlorophyll (33% of the climatological mean), accompanied by 10% increases in haptophytes and cryptophytes¹⁵. These trends reversed around 2016 with a partial recovery of diatoms and a pronounced increase in cryptophytes coinciding with rapid sea-ice loss and indicating a climate-driven reorganisation of phytoplankton assemblages. Consistent evidence from satellite observations, in situ measurements, and Earth system models further demonstrates that SO phytoplankton communities are evolving over time with shifts in diatom dominance and seasonal timing linked to environmental changes such as warming, freshening, and variability in sea-ice conditions^20–25. Collectively, these studies highlight substantial changes in phytoplankton community structure during summer. However, the onset and progression of spring blooms, as well as the transition from spring to summer, remain poorly constrained, particularly at the circumpolar scale and across ocean basins. This limitation restricts our ability to detect early shifts in seasonal development and community composition that may precede or amplify observed summer changes. As a result, the basin-scale timing structure and seasonal evolution of phytoplankton regime shifts across both spring and summer remain unresolved.

Ocean-colour remote sensing is the only practical way to monitor SO phytoplankton at circumpolar scales and over decades. Although global PFT/PSC algorithms based on chlorophyll partitioning spectral anomalies or hyperspectral absorption^26–35 have improved our understanding of phytoplankton diversity, their performance in the SO remains uncertain due to limited validation and the region’s unique optical and ecological conditions^36,37. Earlier attempts to reconstruct PFTs/PSCs in the SO have provided valuable regional insights but were largely restricted to summer-focused on the seasonal sea-ice zone and centred primarily on diatoms²⁵. The circumpolar pigment dataset of Hayward et al.³⁸ has expanded this perspective by highlighting the importance of haptophytes and other flagellated groups under summer conditions¹⁹, yet seasonal transitions remain less well constrained. Extending PFT and PSC reconstructions across seasons, basin scales, and the full diversity of SO community types, therefore, remains a major priority. A key limitation is the scarcity of spatially and temporally extensive in situ datasets with matchup-quality pigment observations spanning shelf-sea-ice-influenced and open-ocean regimes south of 40^∘S.

Here we address this gap by integrating the Hayward et al.³⁸ surface pigment dataset, which spans continental shelf and seasonal sea-ice regions, with NASA SeaWiFS Bio-optical Archive and Storage System (SeaBASS) surface High-Performance Liquid Chromatography (HPLC) observations south of 40^∘S across the Atlantic, Indian and Pacific sectors (Fig. S1 and Table S1). To interpret spatial variability, we classified the SO into four ecological zones based on the September month of mean maximum and minimum sea-ice extent and bathymetric boundaries (Fig. 1a). The Shelf Zone (SZ) was defined using the Antarctic continental shelf break from Amblas et al.³⁹ derived from IBCSO v1.0 bathymetry⁴⁰. The shelf break was manually digitised at a spatial scale of 1:300,000 using 50 m bathymetric contours and slope-gradient maps with interpolation applied across permanently ice-covered regions. This zone represents shallow bathymetry-controlled shelf environments. The Winter Sea-Ice Zone (WIZ) was delineated using the mean maximum September sea-ice extent during 1998–2024. The Maximum Sea-Ice Extent Zone (MIEZ) was delineated as the region bounded by the mean minimum and maximum September sea-ice extents during 1998-2024, derived from the 15% sea-ice concentration contours provided by the National Snow and Ice Data Centre (NSIDC)⁴¹. This zone represents the interannual variability and seasonal retreat of sea ice and covers an area of approximately  ~ 8.83 × 10⁶ km². The Permanently Open Ocean Zone (POOZ) was defined as the permanently ice-free region north of the MIEZ extending to 40^∘S. September was selected because it represents the mean maximum Antarctic sea-ice extent, providing a consistent and physically meaningful reference for delineating ice-influenced regions (Fig. S2)^42–44. This framework enables the separation of the distinct physical processes influencing surface phytoplankton dynamics and allows consistent comparison of seasonal changes between austral spring and summer. We identified a pronounced post-2014 reorganisation of phytoplankton community structure between spring and summer. This shift was characterised by declining in situ total surface chlorophyll concentration (TChla) in both seasons, alongside contrasting seasonal changes in diatom dominance, with enhanced diatom contribution during spring but reduced diatom abundance during summer. These patterns were broadly consistent with concurrent changes in Antarctic sea-ice conditions.

Fig. 1. Southern Ocean (SO) matchup distribution and agreement between satellite-derived ocean-colour chlorophyll concentration (OCChla) and in situ total surface chlorophyll concentration (TChla).

a Spatial distribution of TChla-OCChla matchups (1998–2024) across the SO. The map was produced using MATLAB, with land areas represented from built-in coastline/land shapefiles and Antarctic sea-ice boundaries manually delineated from National Snow and Ice Data Centre (NSIDC) September sea-ice boundary shapefiles (1998–2024) using QGIS. Dot colours indicate matchup classification: black (Match,  ± 50%), red (Overestimation), and blue (Underestimation). Background shading shows the major SO zones, including the Shelf Zone (SZ), the Winter Sea-Ice Zone (WIZ; mean 1998–2024), the Maximum Sea-Ice Extent Zone (MIEZ; mean 1998–2024), and the Permanently Open Ocean Zone (POOZ). Basin sectors (Atlantic, Indian, and Pacific) are indicated for regional reference. b Log-scale comparison between TChla and OCChla for 5003 matchups. The dashed line represents the 1:1 relationship, while the shaded region denotes the  ± 50% agreement envelope. Correlation statistics, including the sample size (n), coefficient of determination (R²), and Pearson correlation coefficient (r), are shown for all observations and for each matchup class. c Distribution of matchup classifications as a function of latitude. d Distribution of matchup classifications as a function of longitude. OCChla was obtained from the Ocean Colour Climate Change Initiative (OC-CCI) Level-3 daily 4-km product⁶¹.

Results and discussion

PFTs/PSCs model performance

The diagnostic pigment framework developed here provides a robust and internally consistent reconstruction of phytoplankton community structure across the surface SO (see the “Methods” section for details). Starting from the non-negative least squares (NNLS) solution, the initial pigment mixtures produced a stable first-order approximation of TChla. Subsequent refinement using the gradient boosting machine (GBM) substantially improved model performance by reducing reconstruction error and capturing nonlinear relationships among diagnostic pigment groups. The final GBM model showed strong agreement between observed and reconstructed surface chlorophyll concentrations within the training dataset, with a high Pearson correlation (r = 0.995,  p < 10⁻²⁴) and a low root-mean-square error (RMSE  = 0.057 mg m⁻³). However, as expected, TChla reconstruction showed stronger agreement than the derived PSC and PFT classifications, which represent partitioned components of the bulk TChla signal. These results indicate that the GBM-derived pigment estimate (DP_total) provides a reliable basis for calculating diagnostic pigment fractions and deriving subsequent PSC and PFT classifications. Model robustness was further supported by validation against independent matchup datasets and by constrained fitting and weighted binning procedures designed to reduce overfitting and instability across the chlorophyll gradient.

The coefficients (Table S2) capture the relative diagnostic value of each pigment for distinguishing phytoplankton groups in the SO. High coefficients for DVChl-a and Zeaxanthin reflect their strong chemotaxonomic association with picophytoplankton, although in the SO these pigments covary with picoeukaryotes rather than representing true Prochlorococcus populations, which are physiologically excluded south of 45^∘S ⁴⁵. Pigments characteristic of nanophytoplankton, such as $19^{\prime }$-Hexanoyloxyfucoxanthin and $19^{\prime }$-Butanoyloxyfucoxanthin, and Alloxanthin, retained weights consistent with previous diagnostic pigment algorithms and reflect the widespread presence of Prymnesiophytes (Haptophytes) and cryptophytes in stratified summer waters^14,46,47. Fucoxanthin emerged as a robust and stable marker of diatoms, exhibiting the highest contributions in high-nutrient, bloom-favouring environments, consistent with its ecological role in SO microphytoplankton^25,47.

The models captured the nonlinear behaviour of PSC and PFT fractions along the chlorophyll gradient, reproducing the expected ecological transitions across different SO zones, from shelf and marginal-ice environments to the open ocean (Fig. 2a and Table S2). Correlations were strongest for groups with well-defined chemotaxonomic signatures, where Prymnesiophytes (Haptophytes) showed the strongest relationship (r = 0.881) (Fig. 2a column 5), followed by diatom-rich microphytoplankton (r = 0.780) (Fig. 2a column 1 and 4) and nanophytoplankton (r = 0.731) (Fig. 2a column 2). Picoeukaryotes and green algae showed weaker but significant relationships, consistent with their low abundances and their reduced spectral influence on ocean colour⁴⁸. The combined Prokaryotes +  Prochlorococcus group (Fig. 2a column 7) showed the weakest correlations, reflecting extremely low DVChl-a and Zeaxanthin concentrations typical of high-latitude regions, which are outside the thermal tolerances of Prochlorococcus⁴⁹. Cryptophytes exhibited higher variability and only a modest correlation with TChla (Fig. 2a column 6), consistent with their localised and episodic occurrence in meltwater-influenced coastal environments and their frequent dominance in the Western Antarctic Peninsula (WAP), where stratified, glacier-influenced surface waters favour flagellate proliferation⁵⁰.

Fig. 2. Performance of the pigment-based phytoplankton functional type (PFT) and phytoplankton size class (PSC) reconstruction algorithm using training and validation datasets.

PSC categories include Micro (fucoxanthin), Nano (19'-hexanoyloxyfucoxanthin + 19'-butanoyloxyfucoxanthin +  alloxanthin), and Pico (chlorophyll-b + zeaxanthin + divinyl chlorophyll-a). PFT groups include Diatoms (fucoxanthin), Prymnesiophytes (19'-hexanoyloxyfucoxanthin + 19'-butanoyloxyfucoxanthin), Cryptophytes (alloxanthin), Prokaryotes + Prochlorococcus (zeaxanthin + divinyl chlorophyll-a), and Picoeukaryotes + Green algae (remaining chlorophyll-b). a Relationship between in situ total surface chlorophyll concentration (TChla; mg m⁻³) and reconstructed PSC/PFT fractions derived from the training dataset. Black circles represent mean PSC/PFT fractions within 100 logarithmically spaced TChla bins, while grey error bars indicate  ± 1 standard deviation (SD) of observations within each bin. Red curves show the fitted fourth-degree polynomial functions describing the relationship between PSC/PFT fractions and log₁₀(TChla). Statistical metrics shown above each panel are the coefficient of determination (R²), Pearson correlation coefficient (r), and significance level (p). b Evaluation of algorithm performance for the training dataset, showing modelled versus measured TChla. The red line denotes the 1:1 relationship. Statistical metrics include sample size (N), r,  R², root mean square error (RMSE), and mean absolute error (MAE). c Independent validation against satellite-derived ocean-colour chlorophyll concentration (OCChla), showing the best 75% of observations retained after residual filtering. Reconstructed chlorophyll estimates derived from total diagnostic pigment concentration (DP_total) are compared with OCChla. The red line denotes the 1:1 relationship. Statistical metrics are as defined above. d Frequency distribution of validation residuals expressed as log₁₀(OCChla) subtract log₁₀(DP_total), where the vertical dashed line indicates zero misfit. Together, these panels demonstrate the robustness and consistency of the reconstruction algorithm across both training and independent validation datasets.

Model training for the PSC and PFT reconstruction model based on TChla partitioning (Fig. 2b) showed excellent agreement (r ≈ 0.93 − 0.98,  RMSE ≈ 0.04 − 0.2 mg m⁻³) while independent validation using satellite-derived ocean-colour chlorophyll concentration (OCChla) (Fig. 2c) reproduced the same distributional patterns with the training data. The similarity between the training and validation datasets indicated strong model transferability and a low potential for overfitting (Fig. 2d). This was reflected in the distribution of root mean square error (RMSE), which was approximately normal for the top 75% best-fit cases, with values ranging from 0.1 to 0.3 mg m⁻³, where the highest RMSE was observed for cryptophytes (Fig. 2d column 6). Here we observed that Micro (diatoms) dominate in high-chlorophyll conditions (${\log }_{10}\mathrm{TChla}\approx -0.5$ to  + 1, corresponding to TChla ≈ 0.3–10 mg m⁻³) (Fig. 2a, b, and c, column 1 and 4), consistent with their strong Fucoxanthin signature and competitive advantage under high-nutrient, bloom-favourable environments⁵¹. Nano groups and Prymnesiophytes (Haptophytes) occupy intermediate phytoplankton levels (${\log }_{10}\mathrm{TChla}\approx -1.5$ to 0, corresponding to TChla ≈ 0.03–1 mg m⁻³) (Fig. 2a, b, and c, column 2 and 5), reflecting their prevalence in moderately stratified waters and their characteristic higher concentrations before or after diatoms blooms⁵¹. The pico domain, defined here as PicoEukaryotes + Green algae together with the Prokaryotes +  Prochlorococcus, becomes most abundant in low-phytoplankton waters (${\log }_{10}\mathrm{TChla}\approx -2.5$ to  − 0.5, corresponding to TChla ≈ 0.003-0.3 mg m⁻³) (Fig. 2a, b, and c, column 3, 7, and 8). The model’s robust performance and cross-validation confirm its suitability for basin-scale, multi-decadal reconstructions of phytoplankton diversity, seasonal phenology, and ecosystem regime shifts.

Community composition and spatial distribution of PFTs/PSCs in the Southern Ocean

Consistent with other ecological studies^1,4,11,52, our results showed that low-chlorophyll waters were dominated by small cells, whereas higher-chlorophyll regions favoured large diatoms. Because PFTs/PSCs were interpreted as fractions of TChla, microphytoplankton (diatoms) exhibited a higher relative contribution in high-TChla waters (TChla  > 1-3 mg m⁻³) (Fig. 3a). Spatial interpolation of the full in situ dataset (1998–2024) revealed that microphytoplankton (diatoms) had the highest relative contribution within sea-ice-influenced regions, including both the WIZ and MIEZ (Fig. 4a rows 1 and 2). This pattern was further supported by the latitudinal structure, with regions south of  ~ 65^∘S characterised by higher diatom contributions (Fig. S3). When averaged by zone, microphytoplankton accounted for approximately 49–60% and 48–57% of the TChla fraction in the WIZ and MIEZ, respectively, across basins (Fig. S4 column 2 and 3), compared to only 25-31% in open-ocean waters (Fig. S4 column 4). In the SZ, micro- and nanophytoplankton contributed comparably across basins (approximately 43-48%; Fig. S4 column 1 row 1).

Fig. 3. Log-binned relationships between in situ total surface chlorophyll concentration (TChla; mg m⁻³) and phytoplankton size classes (PSCs) and phytoplankton functional types (PFTs) derived from Southern Ocean (SO) surface High-Performance Liquid Chromatography (HPLC) pigment measurements collected between 1998 and 2024.

PSC categories include Micro (fucoxanthin), Nano (19'-hexanoyloxyfucoxanthin + 19'-butanoyloxyfucoxanthin +  alloxanthin), and Pico (chlorophyll-b + zeaxanthin + divinyl chlorophyll-a). a Relationships between TChla (log₁₀ scale) and PSC contributions (% of total surface chlorophyll). Solid lines represent bin-mean PSC fractions, and shaded envelopes indicate  ± 1 standard deviation (SD) within each logarithmically spaced TChla bin. b Fractional composition of the Nano PSC, showing the relative contributions of Prymnesiophytes (19'-hexanoyloxyfucoxanthin + 19'-butanoyloxyfucoxanthin) and Cryptophytes (alloxanthin) across the TChla gradient. c Fractional composition of the Pico PSC, showing the relative contributions of Picoeukaryotes + Green algae (remaining chlorophyll-b fraction) and Prokaryotes + Prochlorococcus (zeaxanthin +  divinyl chlorophyll-a). Together, these panels illustrate biomass-dependent transitions in phytoplankton community structure across Southern Ocean chlorophyll regimes.

Fig. 4. Spatial distribution of phytoplankton size classes (PSCs) and phytoplankton functional types (PFTs) across the Southern Ocean (SO), mapped at 1^∘ × 1^∘ spatial resolution using Gradient Boosting Machine (GBM)-derived diagnostic pigment fractions reconstructed from surface in situ High-Performance Liquid Chromatography (HPLC) pigment observations.

PSC categories include Micro (fucoxanthin), Nano (19'-hexanoyloxyfucoxanthin + 19'-butanoyloxyfucoxanthin +  alloxanthin), and Pico (chlorophyll-b + zeaxanthin + divinyl chlorophyll-a). a PSC Micro (%). b PSC Nano (%). c PSC Pico (%). Row 1 shows the spatial distributions of the three PSCs: Micro in column 1, Nano in column 2, and Pico in column 3. Rows 2 and 3 show the corresponding PFT components within each PSC column: Diatoms (fucoxanthin) under Micro; Prymnesiophytes (19'-hexanoyloxyfucoxanthin + 19'-butanoyloxyfucoxanthin) and Cryptophytes (alloxanthin) under Nano; and Picoeukaryotes + Green algae (remaining chlorophyll-b fraction) and Prokaryotes + Prochlorococcus (zeaxanthin + divinyl chlorophyll-a) under Pico. Contours indicate the major SO oceanographic zones: the grey line denotes the Shelf Zone (SZ), the solid black line denotes the Winter Sea-Ice Zone (WIZ; mean 1998–2024), and the dotted black line denotes the Maximum Ice Extent Zone (MIEZ; mean 1998–2024). Waters north of the MIEZ are classified as the Permanently Open Ocean Zone (POOZ), extending to 40^∘S. Basin boundaries (Atlantic, Indian, and Pacific sectors) are shown for regional interpretation. Colours indicate the percentage contribution of each PSC or PFT to in situ total surface chlorophyll concentration (TChla). Spatial fields were generated in MATLAB using a two-step interpolation approach, in which observations were first interpolated onto a regular latitude-longitude grid using natural-neighbour interpolation and remaining gaps were filled using nearest-neighbour interpolation. The interpolated fields were then smoothed using a Gaussian filter, with periodic longitude wrapping at  ± 180^∘ applied to minimise edge artefacts across the dateline.

However, in the Atlantic sector SZ, cryptophytes showed a relatively higher contribution (up to  ~ 18%; Fig. S4 column 1 row 2). Latitudinal binning indicated that these contributions were concentrated around  ~ 65^∘S (Fig. 4 row 3 and S3 column 1 row 2), coinciding with regions adjacent to the Antarctic Peninsula. This pattern was consistent with previous in situ observations from the Western Antarctic Peninsula and East Antarctic shelf regions, where cryptophytes are associated with meltwater-influenced, stratified coastal environments near the ice edge and continental shelf^51,53, and was also supported by the findings of Hayward et al.¹⁴.

As TChla decreased from 0.3 to 1.0 mg m⁻³ (Fig. 3a, b), community structure transitioned toward nanophytoplankton (40–59%; Fig. 3a), with prymnesiophytes dominating the nano fraction (87–96%; Fig. 3b). Spatial patterns (Fig. 4b row 1 and 2) and latitudinal bins (Fig. S3 row 1) indicated a gradual shift from micro- to nano-dominated assemblages northward from  ~ 60^∘S to  ~ 50^∘S, corresponding broadly to the transition from ice-influenced to open-ocean conditions. In these regions, nanophytoplankton contributed approximately 45–54% of TChla, with prymnesiophytes as the primary component (Fig. S4 row 2). At the lowest TChla levels (0.07–0.10 mg m⁻³), nanophytoplankton remained dominant (44-72%), while microphytoplankton contributed less than 30% in most cases (Fig. 3a). Prymnesiophytes consistently accounted for 60–100% of the nano fraction (Fig. 3b), whereas cryptophytes remained low (0–7%) except within intermediate chlorophyll ranges where they accounted for about 40-50% of the nano fraction (Fig. 3b).

Picophytoplankton showed relatively stable contributions across regions, accounting for approximately 5–20% of TChla (Fig. 3a). Within this size class, picoeukaryotes and green algae were the dominant contributors compared to the Prokaryotes + Prochlorococcus group (Fig. 3c). However, latitudinal patterns indicated that Prokaryotes + Prochlorococcus were more prevalent in open-ocean regions north of  ~ 50^∘S, with increased contributions particularly north of  ~ 45^∘S (Fig. 4c row 3, Fig. S3 row 2, and Fig. S4 row 2 column 4). This distribution is consistent with previous studies showing that Prochlorococcus is primarily confined to low- to mid-latitude, stratified subtropical waters and is largely absent from high-latitude SO environments⁴⁹.

Overall, microphytoplankton are primarily represented by diatoms, nanophytoplankton by prymnesiophytes, and picophytoplankton by a mixture of prokaryotic and eukaryotic groups. However, picophytoplankton contribute less than  ~ 20% of TChla, indicating a relatively minor contribution to overall surface biomass. Given these relationships, we focus subsequent analyses on PSCs as the primary quantitative framework, with dominant PFTs interpreted as associated ecological groupings inferred from PSC structure.

The spatial patterns of PSCs and associated PFTs further indicate that sea-ice influence, stratification, and basin-scale structure play key roles in shaping phytoplankton community composition in surface waters across the SO. These derived fields capture the dominant ecological gradients observed in the dataset, including a transition from pico- and nano-dominated assemblages in low-chlorophyll open-ocean regions to increased diatom contributions in higher-chlorophyll, ice-influenced, and shelf environments.

The Southern Ocean regime shift and impact on ecological changes

We define the onset of the SO regime shift as the earliest detectable departure from the preceding multi-decadal sea-ice state. In marine climate systems, a regime shift refers to a persistent reorganisation of physical and ecological system structure, variability, and feedback processes that results in a transition toward a new dominant state rather than short-term interannual fluctuations⁵⁴. Figure S2 shows an increasing trend in Antarctic sea-ice extent during 1998–2014 (slope = 0.0505 × 10⁶ km² yr⁻¹), followed by a reversal after 2014 (2015–2024; slope = − 0.106 × 10⁶ km² yr⁻¹). A pronounced step change occurred between 2014 and 2015, corresponding to a decrease of approximately 6.68%. This abrupt reversal from a prolonged period of expansion marks a clear departure from prior conditions and coincides with the documented Antarctic sea-ice maximum in 2014⁴³.

Thus, we identify the austral spring 2014–2015 as the initiation phase of this transition because it represents the earliest large-scale and persistent deviation from the previous sea-ice trajectory. Whereas Hayward et al.¹⁵ identified a similar shift primarily from austral summer sea-ice concentration patterns around 2016–2017, likely reflecting both methodological and seasonal differences between the studies. Their analysis emphasised summertime sea-ice concentration declines that became increasingly persistent over the Antarctic Shelf, whereas our analysis uses seasonal maximum sea-ice extent and therefore captures an earlier springtime onset across broader regions of the SO. Importantly, the transition after 2014 was not limited to a reduction in sea-ice extent alone, but was also accompanied by changes in sea-ice persistence, regional variability, seasonality, and atmosphere-ocean coupling, all of which indicate a broader reorganisation of SO climate dynamics rather than a transient anomaly^54,55. This interpretation is consistent with recent studies suggesting that Antarctic sea ice has shifted away from its previous expansionary state following the 2014 maximum and may now be transitioning toward a new climatic regime characterised by enhanced instability and reduced resilience⁵⁵. Together, these lines of evidence support 2014–2015 as the early initiation phase of a broader transition in Antarctic sea-ice dynamics and SO system structure.

In parallel, a basin-scale, physically driven reorganisation of OCChla and PSCs is observed across the surface SO between the pre-2014 (1998–2014) and post-2014 (2015–2024) periods (Figs. 5 and 6). This shift exhibits a clear latitudinal structure, extending from SZ through WIZ and MIEZ, while POOZ remains relatively stable. This reorganisation shows a pronounced seasonal contrast. In the WIZ and MIEZ, OCChla declines consistently across all basins in both spring and summer (Figs. 5a, e and 6). However, during austral spring, this decline is accompanied by a reorganisation of community structure, with microphytoplankton (diatoms) increasing in relative contribution across all basins (Figs. 5b and 6a–c; S5 column 1 row 1 and 2), indicating a shift toward larger cells despite reduced surface biomass. In contrast, during austral summer, microphytoplankton contributions decrease across basins, reflecting an opposing seasonal response and a shift toward smaller PSC (Figs. 5f and 6d–f; and S5 column 2 row 2).

Fig. 5. Spatial patterns of decadal changes in satellite-derived ocean-colour chlorophyll concentration (OCChla) and phytoplankton size classes (PSCs) across the Southern Ocean (SO), expressed as the difference between the mean post-2014 (2015–2024) and mean pre-2014 (1998–2014) periods.

PSC categories include Micro (fucoxanthin), Nano (19'-hexanoyloxyfucoxanthin  + 19'-butanoyloxyfucoxanthin +  alloxanthin), and Pico (chlorophyll-b + zeaxanthin + divinyl chlorophyll-a). PSC fractions were reconstructed by applying the fourth-degree polynomial regression model to OCChla from the Ocean Colour Climate Change Initiative (OC-CCI) dataset. The model was applied directly to the OCChla concentration, and all maps were generated in MATLAB using the model outputs together with binned OCChla data. The top row shows spring conditions: a ΔOCChla (mg m⁻³), b ΔPSC Micro (%), c ΔPSC Nano (%), and d ΔPSC Pico (%). The bottom row shows summer conditions: e ΔOCChla, f ΔPSC Micro, g ΔPSC Nano, and h ΔPSC Pico. Positive (red) and negative (blue) values indicate increases and decreases, respectively, during the mean 2015–2024 relative to the mean 1998–2014. Contours indicate the major Southern Ocean oceanographic zones: the grey line denotes the Shelf Zone (SZ), the solid black line denotes the Winter Sea-Ice Zone (WIZ; mean 1998-2024), and the dotted black line denotes the Maximum Ice Extent Zone (MIEZ; mean 1998-2024). Waters north of the MIEZ are classified as the Permanently Open Ocean Zone (POOZ), extending to 40^∘S. Basin sectors (Atlantic, Indian, and Pacific) are shown for regional interpretation.

Fig. 6. Basin- and zone-resolved changes in phytoplankton size class (PSC) fractions and satellite-derived ocean-colour chlorophyll concentration (OCChla) across the Southern Ocean (SO), expressed as the difference between the mean post-2014 (2015–2024) and mean pre-2014 (1998–2014) periods.

Panels show spring conditions for the a Atlantic, b Indian, and c Pacific basins, and summer conditions for the d Atlantic, e Indian, and f Pacific basins. The left y-axis shows changes in PSC fractions (ΔPSC, %), while the right y-axis shows changes in surface chlorophyll (ΔOCChla; mg m⁻³). Stacked bars represent changes in the relative contributions of the three PSCs: Micro (red; fucoxanthin), Nano (blue; 19'-hexanoyloxyfucoxanthin + 19'-butanoyloxyfucoxanthin + alloxanthin), and Pico (orange; chlorophyll-b + zeaxanthin + divinyl chlorophyll-a). Values printed within each coloured bar indicate the change in PSC fraction (ΔPSC, %), calculated as the difference between the mean post-2014 and pre-2014 periods. Where space within a bar is insufficient, the corresponding values are displayed in coloured boxes connected to the bars by arrows for clarity (these boxed values do not represent additional data). The black line with circular markers shows the corresponding change in mean ΔOCChla (mg m⁻³), calculated as the difference between the post-2014 and pre-2014 means, with numerical labels indicating the exact chlorophyll change. Positive values indicate increases, whereas negative values indicate decreases relative to the pre-2014 mean. Zonal categories include the Shelf Zone (SZ), Winter Sea-Ice Zone (WIZ), Maximum Ice Extent Zone (MIEZ), and Permanently Open Ocean Zone (POOZ). This zonal framework illustrates basin-specific restructuring of phytoplankton community composition and associated changes in surface chlorophyll across the SO.

Similar but more spatially variable patterns are observed in the WIZ and MIEZ. During spring, OCChla declines across basins, while diatom contributions increase and nano- and picophytoplankton decline, suggesting a restructuring of community composition under reduced surface biomass conditions (Figs. 5b–d and 6a–c; S5 row 1 column 2–4). This pattern coincides with the post-2014 decline in maximum Antarctic sea-ice extent (Fig. S2), indicating that changes in sea-ice dynamics may influence the timing and structure of phytoplankton communities. In contrast, summer responses differ across basins. In the Indian sector, OCChla increases within the WIZ and MIEZ, accompanied by enhanced diatom contributions (Fig. 5e, f and 6e; and S5 column 2 row 2), whereas in the Atlantic and Pacific sectors, OCChla declines and diatom contributions decrease (Figs. 5e, f and 6d, f; S5 column 2 row 2). This basin-specific divergence highlights the role of regional physical conditions in modulating phytoplankton responses. Meanwhile, the open ocean exhibits relatively limited seasonal changes in both chlorophyll and community composition, indicating a more stable regime compared to ice-influenced regions (Figs. 5 and 6; S5).

Over longer timescales, both satellite-derived PSCs and HPLC observations indicate a reorganisation of phytoplankton community structure in the SO, rather than a uniform shift toward smaller size classes. While previous studies have suggested increasing contributions of nano- and picophytoplankton under changing environmental conditions^{15,22,23,53,55}, our results highlight strong seasonal and basin-scale variability in this response, with contrasting patterns between spring and summer and across different regions.

Overall, these results reveal a decoupling between surface biomass and community structure, where declining chlorophyll is accompanied by seasonally dependent and regionally variable shifts in phytoplankton composition. The spatial patterns of PSCs and associated PFTs indicate that sea-ice influence, stratification, and basin-scale structure likely play key roles in shaping phytoplankton community composition across the SO. These derived fields capture the dominant ecological gradients in the dataset, including a transition from pico- and nano-dominated assemblages in low-chlorophyll open-ocean waters to enhanced diatom contributions in higher-chlorophyll, ice-influenced, and shelf environments.

Spring regime shift

OCChla declined across most ice-influenced regions of the SO during austral spring, but the magnitude of change was spatially heterogeneous rather than uniformly negative across all latitudes (Figs. 5a and 6a–c; S5 column 1 row 1). At the zonal scale, all three basins showed negative OCChla anomalies in the SZ, WIZ, and MIEZ, whereas the POOZ only changed weakly. The strongest spring decline occurred in the Pacific SZ ( − 0.35 mg m⁻³), while smaller but coherent declines were observed in the Atlantic and Indian SZ ( − 0.09 and  − 0.05 mg m⁻³, respectively) (Fig. 6a–c and S5 column 1 row 1). In the WIZ, OCChla decreased in all basins ( − 0.11 in the Atlantic,  − 0.14 in the Indian, and  − 0.08 mg m⁻³ in the Pacific), and the MIEZ showed similarly consistent declines ( − 0.06,   − 0.06, and  − 0.07 mg m⁻³, respectively) (Fig. 6a–c). By contrast, POOZ changes were small ( − 0.003 to  − 0.02 mg m⁻³), indicating relative stability in open-ocean surface chlorophyll after 2014 (Fig. 6a–c).

PSC changes indicate that community structure reorganised alongside declines in surface biomass. However, because PSCs are derived from OCChla using a log-transformed, fourth-degree polynomial relationship, they are not fully independent of chlorophyll variability and primarily reflect a redistribution of chlorophyll among size classes. Here, the clearest spring signal occurred in the WIZ, where microphytoplankton increased in all three basins, while nano- and picophytoplankton decreased (Fig. 6a–c). This pattern was strongest and statistically significant in the Atlantic WIZ, where microphytoplankton increased by +10.3 % (p = 0.0126), accompanied by declines in nanophytoplankton ( − 5.2 %, p = 0.0166) and picophytoplankton ( − 3.7 %, p = 0.0076) (Fig. 6a and S6 column 2 row 1). The Indian WIZ showed a similar but weaker pattern, with a non-significant increase in microphytoplankton (+4.9 %) and a significant decline in picophytoplankton ( − 2.3%, p = 0.0273) (Fig. 6b and S6 column 2 row 3), whereas the Pacific WIZ also shifted toward higher microphytoplankton contribution (+3.3%), although these changes were not statistically significant (Fig. 6c and S6 column 2 row 2). Thus, despite lower OCChla, the WIZ became relatively more microphytoplankton-rich after 2014.

The SZ exhibited a more basin-dependent response. In spring, microphytoplankton increased in the Atlantic and Indian sectors (+1.7 and +4.0 %, respectively), while the Pacific SZ showed only a weak positive change (+0.6%) and substantial spatial variability (Fig. 6a–c). Only the Indian SZ showed a significant spring decline in nanophytoplankton ( − 2.2 %, p = 0.0162), whereas most other SZ PSC shifts were not significant (Fig. 6d and S6 column 1 row 3). In the MIEZ, microphytoplankton changes were small and not significant in all basins, but nanophytoplankton increased significantly in the Atlantic, Indian, and Pacific sectors (+1.4, +1.3, and +1.2%, respectively), while pico fractions generally declined (Fig. 6a–c). This indicates that the spring response in the MIEZ differed from the WIZ, with weaker microphytoplankton enhancement and a modest but consistent increase in nano contribution.

In the POOZ, chlorophyll changed little, but PSC composition shifted slightly and consistently toward lower microphytoplankton and higher nanophytoplankton fractions. These spring POOZ changes were statistically significant for micro- and nanophytoplankton in all three basins, although their magnitudes were small (generally <1%; Fig. 6a–c and S6 column 4), indicating compositional adjustment without strong biomass change.

Taken together, the post-2014 spring response is best described as a surface chlorophyll decline across SZ, WIZ, and MIEZ combined with a zone-dependent reorganisation of PSC structure, rather than a uniform shift toward smaller cells. The strongest restructuring occurred in the WIZ, where reduced OCChla coincided with an increased relative contribution of microphytoplankton and decreased nano- and picophytoplankton fractions. Because these strongest and most coherent spring changes are concentrated in the sea-ice-defined zones, particularly the WIZ, and occur following the post-2014 reduction in Antarctic sea-ice extent (Fig. S2), they are consistent with an association between changing sea-ice conditions and the reorganisation of surface phytoplankton communities. However, this relationship is based on spatial and temporal correspondence rather than direct attribution. The contrasting responses observed among the SZ, WIZ, MIEZ, and POOZ further indicate that sea ice is not the sole driver, and that regional hydrographic structure and basin-scale variability likely modulate the biological response.

Summer intensification

By the austral summer, changes in OCChla and phytoplankton community structure became more spatially heterogeneous compared to spring, particularly within ice-influenced regions (Figs. 5e, f and 6d–f and S5 row 2). At the zonal scale, OCChla declines persisted in the SZ across all basins, with the strongest reduction observed in the Atlantic sector ( − 0.39 mg m⁻³), followed by the Pacific ( − 0.34  mg m⁻³) and Indian sectors ( − 0.06 mg m⁻³) (Figs. 5e and 6d–f; S5 column 1 row 2). In contrast, responses in the WIZ and MIEZ varied by basin. The Atlantic and Pacific sectors generally showed negative OCChla anomalies in the WIZ and MIEZ (Fig. 6d, f), whereas the Indian sector exhibited weak positive changes in both zones (+0.04 and +0.03 mg m⁻³, respectively) (Fig. 6e). The POOZ remained relatively stable, with only minor changes across basins (generally within  ± 0.01 mg m⁻³) (Fig. 6d–f).

In the SZ, microphytoplankton decreased in all basins (e.g.,  − 1.4% per year in the Atlantic and  − 2.8 % per year in the Pacific; Fig. 6d, e and S5 column 2 row 2), accompanied by increases in nanophytoplankton (e.g., +2.5% in the Atlantic and +2.4 % in the Pacific) (Fig. 6d, f and S5 column 3 row 2). These changes were statistically significant in some cases (Fig. S6), particularly for nanophytoplankton increases in the Atlantic SZ (p = 0.017) (Fig. S6 column 1 row 1) and for both nano and pico increases in the Pacific SZ (p < 0.01) (Fig. S6 column 1 row 2), indicating a consistent shift toward smaller size classes under reduced chlorophyll conditions.

In the WIZ and MIEZ, PSC responses were more variable and basin-dependent. In the Atlantic sector, changes were generally small and not statistically significant (Fig. S6 column 2 and 3), indicating limited reorganisation of community structure despite reduced OCChla. In the Pacific sector, microphytoplankton decreased in the SZ but showed only weak changes in the WIZ and MIEZ, while nanophytoplankton increased modestly (Fig. 6f). In contrast, the Indian sector showed slight increases in microphytoplankton contribution in both WIZ and MIEZ (+2.0 and +1.3%, respectively) (Fig. 6e), although these changes were not statistically significant (Fig. S6 column 2 and 3 row 3). Overall, these results suggest that summer community responses in ice-influenced zones were not uniform across basins, but instead reflect regionally variable adjustments in size structure.

In the POOZ, both OCChla and PSC contributions exhibited minimal changes across all basins (Figs. 5e–h and 6d–f; S5 row 2), with most differences being small and not statistically significant (Fig. S6 column 4). This indicates that, unlike the ice-influenced zones, the open ocean maintained a relatively stable phytoplankton regime during summer.

Collectively, the summer response differs from spring in that changes in chlorophyll and PSC structure are less spatially coherent and more basin-specific. While reductions in OCChla persist in the SZ and parts of the WIZ and MIEZ, associated shifts in PSC composition generally reflect a tendency toward increased nanophytoplankton contribution and reduced microphytoplankton in several regions, particularly in the Atlantic and Pacific SZ. These patterns indicate a decoupling between surface biomass and community structure, where declining surface chlorophyll is accompanied by shifts in the relative partitioning of phytoplankton size classes rather than uniform changes in surface biomass.

Synthesis of seasonal changes and basin asymmetry

Taken together, the results indicate that post-2014 changes in the SO primarily reflect a reorganisation of surface phytoplankton community structure rather than a uniform or persistent shift in surface biomass. Across both spring and summer, the most consistent signal is a decoupling between OCChla and surface phytoplankton size structure, with PSC contributions varying by zone, basin, and season (Figs. 5 and 6; S5).

At the zonal scale, the largest changes occur in ice-influenced regions (SZ, WIZ, and MIEZ), whereas the POOZ remains comparatively stable in both chlorophyll and community composition. In spring, the WIZ exhibits the most coherent response across basins, characterised by reduced OCChla and increased relative contributions of microphytoplankton, accompanied by declines in nano- and picophytoplankton (Figs. 5a–d and 6a–c; S5 row 1). In contrast, summer responses are more spatially variable, with reductions in microphytoplankton and increases in nanophytoplankton observed in several regions (Figs. 5e–g and 6d–f; S5 row 2), particularly in the SZ of the Atlantic and Pacific basins (Fig. 6d, f), while other zones show weaker or non-significant changes (Fig. S6).

Basin-scale differences are evident but not uniform. The Atlantic and Pacific sectors generally exhibit stronger summer reductions in OCChla in the SZ and parts of the WIZ, accompanied by shifts toward increased contributions from nanophytoplankton (Fig. 6d, f). The Indian sector shows more variable responses, including localised increases in OCChla and modest changes in PSC composition, particularly within the WIZ and MIEZ (Fig. 6e). However, many of these differences are small or not statistically significant, indicating that basin asymmetry is present but not consistent across all zones and seasons (Fig. S6 row 3).

Overall, these results highlight that post-2014 changes are best interpreted as spatially and seasonally variable adjustments in surface phytoplankton community composition, rather than a uniform transition toward a specific size class or a consistent basin-wide trend. The greatest and most coherent changes occur within sea-ice-defined zones, while the open ocean remains comparatively stable.

Biogeochemical and ecosystem implications

Changes in surface phytoplankton community structure across the SO have important implications for both biogeochemical cycling and ecosystem functioning. Rather than indicating a uniform decline in diatom abundance, our results reveal a seasonally and spatially variable reorganisation of surface phytoplankton community composition and structure, with increases in microphytoplankton contribution during spring in ice-influenced regions and more variable responses during summer (Figs. 5 and 6; S5). These patterns suggest that changes in carbon export potential are unlikely to be uniform across regions or seasons.

This interpretation is consistent with previous in situ and machine-learning reconstructions that reported a 25-year decline in summer diatom chlorophyll across Antarctic shelf regions¹⁵. Here, we extend that analysis by incorporating broader spatial and seasonal coverage across the full SO region. While Hayward et al.¹⁵ focused primarily on austral summer conditions and absolute PFT chlorophyll within shelf and sea-ice environments, our analysis evaluates relative surface chlorophyll contributions of PSCs and PFTs across all seasons and SO ecological zones. This framework, therefore, captures shifts in community composition across broader spatio-temporal scales. However, despite these methodological differences, both studies show strong agreement in austral summertime trends, particularly when assessed using relative PFT proportions (see supplementary material of Hayward et al.¹⁵).

Phytoplankton size structure plays a central role in regulating carbon export efficiency, as larger cells such as diatoms are generally associated with faster sinking rates and enhanced export, whereas smaller cells tend to remain in the upper ocean and support recycling pathways^2,4,23. The observed increase in microphytoplankton contribution during spring, particularly within the WIZ, may therefore sustain greater export potential despite overall declines in surface chlorophyll. In contrast, summer shifts toward increased nanophytoplankton contribution in some regions, particularly in the SZ of the Atlantic and Pacific sectors (Fig. 6d, f), may favour more efficient recycling of organic matter in the upper ocean. Together, these contrasting seasonal responses indicate a decoupling between surface biomass and export potential, with implications for the timing and efficiency of carbon transfer to depth.

Beyond biogeochemical fluxes, shifts in phytoplankton community composition also have broader ecosystem implications. Changes in size structure influence trophic transfer efficiency, grazing dynamics, and the structure of pelagic food webs¹⁷. For example, increased contributions of smaller phytoplankton may enhance microbial loop activity and reduce energy transfer to higher trophic levels, whereas periods of increased microphytoplankton contribution may support more efficient transfer to zooplankton and higher consumers². The spatial and seasonal heterogeneity observed in this study, therefore, suggests that ecosystem responses are likely to be regionally variable rather than uniform across the SO.

Although the observed patterns are consistent with the influence of post-2014 sea-ice changes, phytoplankton responses are governed by multiple coupled physical and biogeochemical processes, including stratification, mixed-layer dynamics, freshwater input, and nutrient supply. In this study, sea ice is therefore interpreted as an integrative proxy for these interacting drivers rather than a direct causal mechanism. A formal attribution analysis quantifying the relative contributions of environmental variables (e.g., nutrients, salinity, and mixing) was beyond the scope of this work, given the limited availability of consistent, basin-scale observational datasets. Future studies integrating satellite observations with reanalysis products and biogeochemical profiling data (e.g., BGC-Argo) will be required to disentangle these controls and assess their relative importance.

Overall, these results indicate that post-2014 changes in the SO are best interpreted as a reorganisation of phytoplankton community structure, with implications for both carbon cycling and ecosystem dynamics, rather than a uniform, basin-wide decline in export efficiency.

Methods

Data

We compiled 9,986 surface pigment observations (0–20 m) south of 40^∘S from 1998-2024, combining NASA SeaWiFS Bio-optical Archive and Storage System (SeaBASS) records with an expanded SO dataset provided by Hayward ¹⁵ (Table S1). To justify the definition of this surface layer, we evaluated the light optical depth using the diffuse attenuation coefficient at 490 nm (K_d(490)) derived from the Ocean Colour Climate Change Initiative (OC-CCI) Level-3 daily 4 km product, matched to all available in situ observations across the SO. Using these data, we calculated the first optical depth (1/K_d) for each observation^56,57. Across regions spanning the continental shelf to the open ocean, the mean first optical depth was approximately 20 m, consistent with values commonly reported for the global ocean^56,57. These results support the use of the upper 20 m as the optically resolved near-surface layer effectively detectable by satellite ocean-colour sensors.

Additionally, to assess how representative this layer is of phytoplankton biomass within the surface layer, we analysed depth-resolved chlorophyll profiles from 19,321 in situ samples spanning the SO (south of 40^∘S). We compared chlorophyll integrated over 0–20 m with that integrated over the full water column (0–150 m), averaged by zone and season (spring and summer). The results show that the upper 20 m contains 11–27% of total biomass, contributing  ~ 25–27% in the SZ (spring: 25%, summer: 27%),  ~ 18–19% in the sea-ice zone (WIZ+MIEZ) (spring: 18%, summer: 19%), and  ~ 11–16% in the POOZ (spring: 11%, summer: 16%) (Fig. S7).

However, importantly, although the upper 20 m captures a measurable fraction of phytoplankton biomass and represents the optically accessible surface layer, satellite-derived observations remain restricted to ice-free or low sea-ice concentration conditions (typically <15%), consistent with the sea-ice extent definition applied in this study. Consequently, phytoplankton production occurring beneath or within sea ice is not resolved by satellite ocean-colour observations. As a result, early-season blooms and under-ice production, often associated with large phytoplankton such as diatoms, are likely underrepresented within this framework. Therefore, the surface layer analysed here should be interpreted as representing the post-ice-retreat, optically detectable expression of phytoplankton communities rather than the complete annual cycle of production. This limitation further implies that the contribution of diatoms may be underestimated in regions where substantial biomass develops under partial ice cover, and that our results primarily reflect changes in surface community structure and chlorophyll partitioning rather than total water-column biomass or net primary production.

After defining the optically resolved surface layer, in situ total surface chlorophyll concentration (TChla) was defined following SeaBASS conventions as the sum of chlorophyll a, chlorophyll b, and chlorophyll c. Eight diagnostic pigments were retained for analysis: fucoxanthin (Fuco), peridinin (Per), 19$\prime$-hexanoyloxyfucoxanthin (Hex-fuco), 19$\prime$-butanoyloxyfucoxanthin (But-fuco), alloxanthin (Allo), chlorophyll b (Chl b), zeaxanthin (Zea), and divinyl chlorophyll a (DVChl-a). Quality control followed^{13,27,58–60}, where pigment concentrations  < 0.001 mg m⁻³ were set to zero, samples with TChla ≥15 mg m⁻³ were removed, and only profiles with at least four measurable pigments were retained.

Satellite-derived ocean-colour chlorophyll concentration (OCChla) was obtained from the OC-CCI Level-3 daily 4 km product⁶¹. In situ samples were collocated to the nearest OCChla pixel within 8 km in the same day time window, yielding 3,160 matchups (Fig. S1.b). A relative difference filter (absolute relative bias  < 0.5) produced 1,499 high-quality matchups with strong agreement (Pearson r = 0.93, 95% confidence) (Fig. 1b and S1.c). We further assessed the spatial distribution of the matchups, confirming broad coverage across longitude and latitude (Fig. 1c, d), and evaluated the temporal distribution, which is well represented across months and years (Fig. S1.b). A 70/30 split was applied to form the training (1050 samples) and validation (449 samples) datasets (Table S4), ensuring an approximately even representation across all sampling years. Model robustness was further evaluated using K-fold cross-validation.

We then reconstructed surface SO PFTs and inferred associated PSCs using an empirical abundance-based diagnostic model, in which pigment-derived chlorophyll contributions are used to estimate the fractional composition of micro-, nano-, and picophytoplankton, following established approaches²⁹. We then extend this framework to satellite ocean-colour observations through a bin-based empirical mapping approach. We used this approach to quantify basin-scale patterns in phytoplankton community structure (i.e., PFT and PSC composition) and their seasonal variability during austral spring and summer over 27 years (1998–2024) across the three major SO basins (Pacific, Atlantic, and Indian).

Environmental context was provided using National Snow and Ice Data Centre (NSIDC) September sea-ice boundary shapefiles (1998–2024)⁴¹: (1) The WIZ was delineated using the mean maximum September sea-ice extent during 1998–2024, representing the late-winter distribution of persistent sea ice. (2) The MIEZ was delineated as the region bounded by the mean minimum and maximum September sea-ice extents during 1998–2024, derived from the 15% sea-ice concentration contours provided by NSIDC, capturing the interannual variability in sea-ice extent. (3) The POOZ was defined as the open-ocean region north of the MIEZ extending to 40^∘S (Fig. 1a). The SZ was defined using continental shelf boundaries following Amblas et al.³⁹, representing shallow, bathymetry-controlled regions. Basin boundaries followed standard oceanographic definitions: Atlantic (20^∘W-80^∘E), Indian (80^∘E-160^∘E), and Pacific (160^∘E-20^∘W).

Multi-stage diagnostic pigment framework

We implemented a multi-stage diagnostic pigment framework to reconstruct PFTs and PSCs from in situ High-Performance Liquid Chromatography (HPLC) pigments and OCChla. The workflow consists of five sequential steps: (1) pigment reconstruction using non-negative least squares (NNLS), (2) statistical refinement using Gradient Boosting Machine (GBM), (3) PSC/PFT classification, (4) bin-based characterisation along the chlorophyll gradient, and (5) polynomial modelling for satellite application.

Step 1: Initial pigment reconstruction using NNLS

We first estimated pigment-specific chlorophyll contributions from the eight diagnostic pigments using a NNLS approach⁶². This method shares the same mathematical foundation as established pigment-based algorithms such as CHEMTAX and PhytoClass^18,19, and constrains all solutions to be non-negative, ensuring physically meaningful pigment contributions. This step provides an initial linear estimate of total diagnostic pigment concentration (DP_lin), obtained by solving Eq. (1), which minimises the difference between reconstructed and observed TChla under non-negativity constraints.

$${\min }_{C\ge 0}{∥PC-\mathrm{TChla}∥}^2$$ 1

where P is the matrix of diagnostic pigment concentrations, and C is the vector of pigment contribution coefficients. The product PC represents the reconstructed chlorophyll from pigment contributions.

The NNLS estimate of pigment-derived chlorophyll was computed as:

$${\mathrm{DP}}_{\mathrm{lin}}={\sum }_{i=1}^8{P}_i{C}_i$$ 2

Following this step Eq. (2), dinoflagellates were excluded from the analysis because a stable and positive peridinin signal could not be reliably resolved from the available pigment dataset. Consequently, the microplankton fraction consisted only of diatoms (Table S2).

Step 2: Statistical refinement using Gradient Boosting Machine

To account for non-linear relationships and reduce bias in the linear NNLS estimate, we applied a GBM to refine pigment-derived chlorophyll (DP_total). This step improves predictive accuracy by capturing complex interactions among pigments that are not resolved by linear methods alone.

A Gradient Boosting Machine (GBM; LSBoost, 1600 trees, learning rate  = 0.02, subsampling  = 0.6) was trained using raw pigment concentrations, $\log \left(\mathrm{pigment}+10^{-6}\right),{\mathrm{DP}}_{\mathrm{lin}}$, and $\log \left({\mathrm{DP}}_{\mathrm{lin}}\right)$. The final GBM model achieved high predictive skill (R² = 0.995, RMSE  = 0.057 mg m⁻³). The resulting estimate of pigment-scaled chlorophyll, DP_total, was then used to compute pigment fractions according to Eq. (3):

$$f_i=\frac{P_i{C}_i}{{\mathrm{DP}}_{\mathrm{total}}}$$ 3

where f_i represents the fractional contribution of each pigment to surface total chlorophyll.

Fractions were normalised to 0–100%.

Step 3: PSC and PFT classification

Pigment fractions were then grouped into PSCs and PFTs based on established DP relationships.

Pigment fractions were regrouped into:

• PSC: Micro (Fuco), Nano (Hex-fuco + But-fuco + Allo), Pico (Chl b + Zea + DVChl-a);

• PFT: Diatoms (Fuco), Prymnesiophytes (Hex-fuco +  But-fuco), Cryptophytes (Allo), Prokaryotes (Zea), Prochlorococcus (DVChl-a), and PicoEukaryotes +  Green algae (remaining Chl b component).

Negative values were set to zero and all groups were renormalised to 100%.

Step 4: Log-binned PSC/PFT-TChla relationships

To quantify how PSC/PFT structure varies along the chlorophyll gradient, TChla was binned into 100 logarithmically spaced intervals. For each bin, mean fraction, standard deviation, and sample count were computed; bins with fewer than three samples were discarded (Fig. 2a and Fig. S7). This step provides a smoothed empirical relationship between biomass and community composition.

Step 5: Polynomial modelling and satellite application

For each PSC and PFT, a weighted fourth-degree polynomial was fitted relating fractional contribution to ${\log }_{10}\left(\mathrm{TChla}\right)$. Bin sample counts were used as weights. These functions enable continuous prediction of PSC/PFT composition from satellite-derived chlorophyll fields. Model performance was evaluated using weighted coefficient of determination (R²), weighted Pearson correlation (r), and significance level (p-values) derived from Kish’s effective sample size⁶³. Predicted fractions were constrained between 0 and 100%. Final PSC/PFT fields were applied to both TChla and OCChla and validated against the independent matchup dataset (Fig. 2c, d and Table S3). This framework allows consistent reconstruction of phytoplankton community structure from both in situ and satellite observations, enabling basin-scale and decadal analysis.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

N.A. led the study’s conceptualisation, performed the data curation, formal analysis, investigation, and methodology development, generated all visualisations, and carried out the original writing. A.H. contributed to data curation and methodology development, and in writing review and editing. P.S. contributed to conceptualisation and methodology development, provided supervision, and participated in writing review and editing. M.E. contributed to conceptualisation, provided overall supervision, and was involved in writing review and editing.

Peer review

Peer review information

Nature Communications thanks Thomas Mock, Walker Smith and Sandy Thomalla for their contribution to the peer review of this work. A peer review file is available.

Funding

This research was funded by the Australian Research Council Special Research Initiative, Australian Centre for Excellence in Antarctic Science (Project Number SR200100008).

Data availability

The matchup dataset, including the eight diagnostic pigments, in situ total surface chlorophyll concentration (TChla), satellite-derived ocean-colour chlorophyll concentration (OCChla), and monthly phytoplankton size class (PSC) and phytoplankton functional type (PFT) fields (September-February, 1998–2024), has been deposited in Zenodo and is publicly available at 10.5281/zenodo.17875100⁶⁴. This publicly available Zenodo dataset contains the minimum dataset required to reproduce and verify the analyses presented in this study.

Code availability

All code used for data processing, analysis and statistical modelling in this study is openly available on GitHub and is publicly available at 10.5281/zenodo.20839101⁶⁵.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-75303-2.