| 15 Jan 2026
Southern Ocean biological pump over the last glacial cycle from new diatom transfer functions
Abstract. We present new transfer functions to reconstruct deep ocean (~1000 m) particulate organic carbon (POC) flux and particulate inorganic to organic carbon export ratio (PIC : POC) from diatom assemblage in the Southern Ocean. The transfer functions were calibrated with modern sediment trap data covering the three ocean sectors of the Southern Ocean. They were then applied to ten sediment cores located in the Antarctic Zone (AZ) in the three Southern Ocean basins. The diatom community appears to catch efficiently the ecosystem structure that sets the magnitude and stoichiometry of the export fluxes with root mean square errors of the prediction ranging 17–19.6 % depending on the transfer function. A consistent climatic signal is observed in all sediment cores: the reconstructed deep-ocean POC export is higher during glacial than interglacial periods. The PIC : POC ratio is low during glacial periods and increases quickly after glacial maxima. These two signals suggest that both the increase in the biological carbon pump and the decrease in the carbonate counter-pump in the AZ during glacial periods could have contributed to the decrease in atmospheric pCO2. The reconstructed POC export is consistent with previously published diatom-bound δ15N and total organic carbon content but differs from elemental Ba/Fe ratio, hinting Ba potential preservation issues in Southern Ocean sediments. At the global Southern Ocean scale, the deep-ocean POC export flux decreases by 50 % and the PIC : POC export ratio increases by 17 % during the last deglaciation. While the glacial/interglacial POC flux change is comparable in the three SO sectors, the PIC : POC change is weaker in the Pacific, suggesting a distinctive response of the calcifying plankton community to glacial conditions in this sector. We suggest two mechanisms likely to increase the biological pump efficiency during glacial periods: 1) iron fertilization increasing primary production combined with diatom spore formation that increases export efficiency, and 2) a northward extension of sea ice edge supporting a greater zooplankton-mediated export that increases transfer efficiency. These new transfer functions quantitatively support a glacial iron fertilization effect in the AZ, contrasting with the view of a fertilization effect restricted to the Subantarctic Zone.
General comments
The manuscript by Rembauville and Pichat aims to reconstruct past deep-ocean particulate organic carbon (POC) flux and the particulate inorganic carbon-to-particulate organic carbon (PIC/POC) export ratio using diatom assemblages preserved in sediment cores. First, the authors identify statistical relationships between the relative abundances of diatom species and organic carbon fluxes measured in several sediment traps deployed at depths of 1000–2000 m. Second, they apply these statistical relationships to the relative abundances of diatoms in several sediment cores to estimate variations in POC flux and PIC/POC from 150 ka to the present, depending on the core examined. Third, they compare the new diatom-based reconstructions with established palaeoproductivity records (%TOC, %CaCO₃, δ¹⁵Ndb, Ba/Fe) previously published for the same cores. Overall, the new records broadly reproduce the patterns observed in the published palaeoproductivity records. One of the main findings of the manuscript is that this new approach reconstructs an increase in the biological carbon pump in the Antarctic Zone (AZ) during glacial periods, which contrasts with decades of research on this topic. The authors propose several hypotheses to explain these unexpected results, which is commendable. Such a potential paradigm shift would certainly stimulate further discussion.
However, the results and interpretations rely heavily on the robustness of the transfer function, which is based here on a purely statistical approach with limited ecological or biogeochemical justification and appears to be biased at several levels.
First, the transfer function is developed using the relative abundances of diatoms and POC fluxes. However, the metric used by the authors is inappropriate, as diatom relative abundances have little direct biogeochemical meaning. The more appropriate metric would be the flux of diatom species or, preferably, the flux of diatom biovolumes. These fluxes determine the amount of biogenic silica (BSi) and POC exported to the deep ocean, as demonstrated by the authors themselves in previous work (Rembauville et al., 2015) and by several other studies (Rigual-Hernandez et al., 2015, 2016). Indeed, the link between relative abundances—i.e. the proportion of each species within the total diatom assemblage—and POC flux (and even more so the PIC/POC ratio) remains unclear. Some studies have even reported the absence of a relationship between diatom relative abundances and POC flux (e.g. Grigorov et al., 2014; their Figure 5). For example, the same proportion of Fragilariopsis kerguelensis, the dominant diatom species in the open Southern Ocean, could occur in samples with very different absolute diatom abundances (millions of diatoms per litre of seawater) or export fluxes (millions of diatoms per litre per day). In such a scenario, a relative abundance of 50% F. kerguelensis in a flux of 10⁶ diatoms m⁻3 d⁻¹ would export far less BSi and POC than the same 50% proportion in a flux of 150 × 10⁶ diatoms m⁻3 d⁻¹. However, this fundamental difference is disregarded when relative abundances are used instead of fluxes. This issue is reflected in Figure 2 of the present study, where high POC values in the WSC correspond to a wide range of C. pennatum relative abundances, while low POC values from 47°S to P2 are associated with markedly different diatom assemblages.
The reconstruction of the PIC/POC ratio from diatom assemblages is even more tentative, given that diatoms do not precipitate CaCO₃. As biogenic silica and CaCO₃ production and burial are currently concentrated south and north of the Polar Front Zone (PFZ), respectively (Ragueneau et al., 2000), and generally show an anti-correlation at glacial–interglacial timescales (Choi et al., 2025), estimating PIC/POC from diatom assemblages appears to rely on purely statistical relationships rather than on a clear mechanistic basis.
Second, the modern relationships between diatom assemblages and geochemical fluxes are derived from sediment traps deployed at depths of 1000–2000 m. The authors state that there is little flux attenuation and minimal modification of diatom assemblages below these depths; however, this assertion contradicts several previous studies. Indeed, a significant loss of small, lightly silicified diatoms (e.g. P. lineola, T. gracilis) and a concomitant enrichment of heavily silicified taxa (e.g. A. tabularis, T. lentiginosa, Chaetoceros resting spores) has been documented between deep sediment traps (1500 m in the PFZ or 3800 m in the SAZ) and surface sediments (Rigual-Hernandez et al., 2016). This selective preservation highlights the importance of species such as T. lentiginosa in the export of biogenic silica (BSi) (Shukla, 2016), and potentially of organic carbon (Corg_{org}org) if stoichiometric relationships are preserved through time.
Third, the POC flux and PIC/POC ratio reconstructed using the transfer function are compared with several palaeoproductivity records. However, none of the proxy records used represents the same metric, as the authors rely on %TOC, %CaCO₃, and δ¹⁵Ndb. Both %TOC and %CaCO₃ are strongly affected by sedimentary dilution, whereas δ¹⁵Ndb is generally considered a more conservative proxy. To allow a more meaningful comparison, the authors should present TOC fluxes from the same cores or from nearby cores. In addition, several earlier studies reporting Corg fluxes have been omitted (Shimmield et al., 1994; Bareille et al., 1998; Anderson et al., 2002). These studies consistently indicate lower glacial export and burial in the Antarctic Zone (AZ), which contrasts with the interpretation proposed in the present manuscript.
In conclusion, the manuscript presents a purely statistical approach that lacks a clear biogeochemical or ecological rationale. The methodology should employ more appropriate metrics—such as diatom fluxes or biovolume fluxes—which have a direct geochemical relevance for reconstructing POC export. In its current form, the approach relies on relative abundances, which do not adequately represent the magnitude of particle export. Furthermore, it is difficult to justify the reconstruction of the PIC/POC ratio from diatoms, particularly from diatom assemblages, given that diatoms do not produce CaCO₃. The authors should also better constrain the potential loss and selective preservation of diatoms between the deepest sediment traps available and the surface sediments. Only once these issues are addressed could the resulting palaeoceanographic interpretations—currently at odds with previous studies on palaeoproductivity and nutrient supply—be evaluated using metrics that are comparable to those being reconstructed.
Minor comments (chronologically presented here below)
Line 77: Please further detail the role of diatoms in controlling the intensity and stoichiometry of the biological pump at the global scale.
Line 80: Please clarify what is meant by the deep ocean.
Line 106: The water depth between 1000 and 2000 m corresponds to the upper Circumpolar Deep Water (CDW), which is recirculated within the upper cell of the global meridional overturning circulation (MOC) (Marshall, 2012). Its residence time is on the order of a few hundred years. The apparent age of waters within this depth range is approximately 500 years today and around 1000 years during the Last Glacial Maximum (LGM) (Rafter, 2022). For carbon to be sequestered on climatically relevant timescales, export must reach deeper waters, namely the lower CDW and the lower cell of the MOC.
Line 139: Please provide more detail on the approach. Do you mean that the 22 informative species were rescaled to 100% and that the proportion of each species within this restricted assemblage was used as input data?
Section 2.3: Studies that develop new transfer functions generally include a more detailed methodological section to ensure both the robustness of the approach and its reproducibility. Here, the methods—particularly the regression procedures—are not described in sufficient detail compared with similar studies (Esper et al., 2014; Ferry et al., 2015).
It is also worth noting that MLR (or IKM) and PLSR (or WA, WA-PLS) typically perform best with linear models. However, diatom distributions in the Southern Ocean are known to be non-linear (Armand et al., 2005; Crosta et al., 2005; Esper et al., 2010) as well as in northern seas (Oksman et al., 2019), as are the associated species–temperature relationships (Esper et al., 2014; Oksman et al., 2019). This may explain the relatively poor performance of MLR and PLSR in the present study (Fig. 4). Are these methods, therefore, the most appropriate choices here?
I am less familiar with decision-tree approaches, but the method does not appear to be sufficiently described or constrained (see Grenier et al., 2010, Water Quality Research Journal of Canada, 45(4), 413–425; Salonen et al., 2016, The Holocene, 26(7), 1040–1048). Overall, the transfer function is based on 11 sites and four seasons (i.e. 44 data points), which represents a rather limited dataset for calibration.
Table 3 and associated text: Most of the species with high loadings appear to be extremely rare in the sediment traps (<1%), based on the mean values presented in Figures 2 and 3. First, I would recommend presenting the ranges of relative abundances. Second, this suggests that the transfer function is largely driven by very rare taxa, which may have limited ecological and biogeochemical significance.
In addition, the ecology of several species is not accurately described. In Factor 1 (SAZ), S. trifultus is not a typical SAZ species but rather a ubiquitous taxon, possibly because several varieties have been grouped. Thalassiothrix antarctica is a northern POOZ species (Grigorov et al., 2014) that has been used to infer past shifts in the Antarctic Polar Front (APF) position (Kemp et al., 2010; Duffy et al., 2025). In Factor 2 (sea ice), F. separanda is not strictly a sea-ice-associated species and can occur at temperatures up to 10 °C (Esper et al., 2010). In Factor 3 (POOZ), the bloom-forming species mentioned are typically abundant around the Antarctic coast and around the SAZ islands rather than in the open ocean (Armand et al., 2005; Buffen et al., 2007). Fragilariopsis obliquecostata is not a POOZ species but a sea-ice-associated taxon commonly used to infer the presence of summer sea ice (Gersonde and Zielinski, 2000).
There is also a significant issue related to the lumping together of varieties with distinct ecological signals (e.g. E. antarctica var. recta and var. antarctica, T. antarctica var. T1 and var. T2, Hyalochaete spp.).
Line 278: It is not clear why deep-ocean POC export should be associated with small, sea-ice-related diatoms that are readily dissolved, thereby exposing their organic content to remineralisation. These small species rarely reach the sediments in high proportions, except on the Antarctic shelf (Burckle et al., 1987). Moreover, the slope between Corg flux and diatom flux in the AAZ MS4 trap—where F. cylindrus and F. curta are abundant (both in flux and relative abundance)—is much lower than in the SAZ MS2 trap, where F. kerguelensis and Thalassiothrix dominate. This suggests limited export associated with these sea-ice-related taxa.
Line 283: The quantitative relationship was established between diatom fluxes—not relative abundances—and organic carbon fluxes.
Line 285: Only the cold-water varieties of Hyalochaete and E. antarctica are associated with sea ice. It is also worth noting that an inverse relationship is often observed between CRS and F. curta on the continental shelf (Leventer et al., 1993; Campagne et al., 2016; Torricella et al., 2024).
Line 290: The correlation coefficients with Factor 1 (warm water; 0.033) and Factor 2 (sea ice; −0.149) are not significant (Table 4). How, then, can the PIC/POC ratio be reconstructed from diatom relative abundances?
Line 292: The terms northward and southward are not sufficiently informative; please specify the regions more precisely. To my knowledge, coccolithophores are nearly absent south of the Antarctic Polar Front (APF) (Eynaud et al., 1999; Saavedra-Pellitero et al., 2014), whereas planktic foraminifera (mainly Neogloboquadrina pachyderma sinistral) remain present (Niebler and Gersonde, 1998; Haddam et al., 2016).
Line 295: This interpretation appears to rely purely on statistical relationships without an underlying ecological or biogeochemical rationale.
Line 357: While this is correct, models tend to overestimate the geographical extent of the LGM high-export zone (Tagliabue et al., 2009) compared with observational data (Kohfeld et al., 2005).
Line 362: This statement is incorrect. These studies indicate an increase in the relative utilisation of nutrients (reflected by higher δ¹⁵Ndb) in the Antarctic Zone due to reduced nutrient supply, as you also state in lines 376–378.
Lines 379–383: I recommend presenting the dominant diatom species in these cores to support this interpretation. For example, in core PS2606-6, CRS increases from 0% during the Holocene to ~10% during the glacial period; E. antarctica increases from 0 to ~10%; F. curta from 0 to ~6%; F. kerguelensis decreases from ~85% to ~50%; and T. lentiginosa increases from ~5% to ~15%. Based on these changes, it is not clear that there is a shift from “silica sinkers” to “carbon sinkers”, as silica-exporting taxa remain dominant during the glacial interval.
Section 4.4: The interpretations rely heavily on the robustness of the transfer function, which currently appears to be based on an inappropriate metric. It would be important to determine whether the results—and therefore the palaeoceanographic interpretations—remain valid when diatom fluxes rather than relative abundances are used.