Exploring the impact of climate change on the global distribution of non‐spinose planktonic foraminifera using a trait‐based ecosystem model

Planktonic foraminifera are one of the primary calcifiers in the modern ocean, contributing 23%–56% of total global pelagic carbonate production. However, a mechanistic understanding of how physiology and environmental conditions control their abundance and distribution is lacking, hindering the projection of the impact of future climate change. This understanding is important, not only for ecosystem dynamics, but also for marine carbon cycling because of foraminifera's key role in carbonate production. Here we present and apply a global trait‐based ecosystem model of non‐spinose planktonic foraminifera (‘ForamEcoGEnIE’) to assess their ecology and global distribution under future climate change. ForamEcoGEnIE considers the traits of calcium carbonate production, shell size, and foraging. It captures the main characteristic of biogeographical patterns of non‐spinose species – with maximum biomass concentrations found in mid‐ to high‐latitude waters and upwelling areas. The model also reproduces the magnitude of global carbonate production relatively well, although the foraminifera standing stock is systematically overestimated. In response to future scenarios of rising atmospheric CO2 (RCP6 and RCP8.5), on a regional scale, the modelled foraminifera biomass and export flux increases in the subpolar regions of the North Atlantic and the Southern Ocean while it decreases everywhere else. In the absence of adaptation, the biomass decline in the low‐latitude South Pacific suggests extirpation. The model projects a global average loss in non‐spinose foraminifera biomass between 8% (RCP6) and 11% (RCP8.5) by 2050 and between 14% and 18% by 2100 as a response to ocean warming and associated changes in primary production and ecological dynamics. Global calcium carbonate flux associated with non‐spinose foraminifera declines by 13%–18% by 2100. That decline can slow down the ocean carbonate pump and create short‐term positive feedback on rising atmospheric pCO2.


| INTRODUC TI ON
Planktonic foraminifera are calcifying marine protists. They occupy habitats generally at the top 100 m of the ocean (Berger, 1969;Field, 2004;Schiebel et al., 2001), although some species are found in sub-thermocline waters (200-2000m, Schiebel & Hemleben, 2005. Planktonic foraminifera are one of the least abundant zooplankton groups in the ocean, contributing less than 5% to the total microprotozooplankton abundance (Beers & Stewart, 1971) and 0.04% of the total plankton biomass in the upper 200 m (Buitenhuis et al., 2013).
Despite this, planktonic foraminifera play a major role in the calcium carbonate (CaCO 3 ) budget of the ocean and are thought to be responsible for 23%-56% of pelagic CaCO 3 flux globally (Buitenhuis et al., 2019;Schiebel, 2002).
CaCO 3 shells precipitated by foraminifera in the surface ocean ultimately sink through the water column. Some dissolve along the way and others at the sediment surface of the deep ocean. The associated removal of alkalinity and dissolved inorganic carbon (DIC) from surface seawater lowers pH. This favours dissolved carbon as CO 2(aq) over carbonate ions (CO 3 2 ), and on anthropogenic timescales (c. <10 3 years), increases CO 2 in the atmosphere (Frankignoulle et al., 1994). Any reduction in the rate of biogenic carbonate precipitation at the ocean surface would therefore have a (short term) positive impact on atmospheric CO 2 . On longer timescales (order 10 3 -10 4 years), lower pelagic CaCO 3 production and removal through burial in the sediments would act to increase CO 3 2− concentrations and hence lower atmospheric CO 2 (Ridgwell & Zeebe, 2005). Following burial in sediments, the carbonate shells of planktonic foraminifera also provide essential proxy records, widely used to reconstruct past environmental conditions in the ocean, such as temperature, salinity, circulation, oxygenation, and alkalinity (Henderson, 2002 and references within).
Despite planktonic foraminifera's importance to the marine carbon cycle, little is known about how their distribution and biomass may be impacted by climate change, and the strength and sign of the feedback with atmospheric CO 2 is uncertain (Ridgwell et al., 2009).
This lack of understanding is partly a consequence of the difficulty of sampling low standing stocks as well as due to a generally unsuccessful reproductive cycle in culture (Schiebel & Hemleben, 2017).
Prey availability has also been considered a key driver of planktonic foraminifera biogeography (Schiebel & Hemleben, 2017), although understanding of predator-prey interactions is equally limited. Planktonic foraminifera are immotile passive ambush feeders (Kiørboe, 2011), that use their rhizopodia's nets to sense, encounter, and control prey from their surroundings (e.g. Anderson & Be, 1976).
Field observations and culture experiments suggest that foraminifera are opportunistic omnivorous predators, with feeding preferences being linked to their external morphology (e.g. Anderson & Be, 1976;Anderson et al., 1979;Spindler et al., 1984). Roughly, half of foraminifera species develop calcium carbonate spines during their juvenile stages (called spinose). Spinose adult species (with the exception of Globigerina bulloides) are predominantly carnivorous and most abundant in oligotrophic regions (Schiebel & Hemleben, 2017).
In addition to controlling species biogeography, environmental conditions also impact the growth and calcification rates of individuals (e.g. Anderson et al., 1979;Davis et al., 2013;Schmidt, Renaud, et al., 2004;Weinkauf & Waniek, 2016). For example, starvation causes shell size reduction and death, while over-feeding leads to a shell size increase and a shorter lifespan (e.g. Anderson et al., 1979;Bé et al., 1981;Caron & Bé, 1984). Shell size is also correlated with temperature, with a large size under optimal conditions (Hecht, 1976;. In contrast, our understanding of F I G U R E 1 (a) Contribution of non-spinose foraminifera to foraminifera assemblages based on sediment traps (ForCenS database; Siccha & Kucera, 2017). Data meta-analysis by Kirsty Edgar). (b) Locations of the plankton net (circle) and sediment trap (square) observations of non-spinose species were used for the model-data comparison. The references are listed in Tables S1 and S2 [Colour figure can be viewed at wileyonlinelibrary.com] processes that drive shell density and weight and hence 'calcification' in general, is less clearly developed; species can grow in shell size but develop thinner shells and vice versa. For example, while some studies suggest that pCO 2 regulates calcification and shell morphology (e.g. Barker & Elderfield, 2002;Bijma et al., 1999;De Moel et al., 2009;Lombard et al., 2010;Moy et al., 2009;Russell et al., 2004;Spero et al., 1997), others suggest a combination of multiple environmental drivers (Davis et al.,2013;Weinkauf & Waniek, 2016;Zarkogiannis et al., 2019 and references within;Fox et al., 2020). Jointly, these studies indicate species-specific regional responses (e.g. Barker & Elderfield, 2002;Davis et al., 2013;Fox et al., 2020;Weinkauf & Waniek, 2016).
As a result of all these uncertainties surrounding the influences of various environmental drivers on planktonic foraminifera, it is difficult to quantify the impact of future climate change, particularly in terms of their biogeography, regional extirpation, biomass, and consequently calcium carbonate production. To better understand how climate influences future foraminifera distribution and biomass, we require a mechanistic way of connecting their physiology with the environment. A promising mechanistic approach for studying complex ecosystem behaviour is trait theory, that describes ecosystem dynamics based on individuals' key traits (e.g. size, feeding, motility, reproduction) and trade-offs (e.g. energetic cost, protection benefit; Flynn et al., 2015;Kiørboe et al., 2018b). Trait theory has been used to study the links between environmental conditions and plankton physiology, ecology, and populations, under both modern and future climate conditions (e.g. Brun et al., 2019;Dutkiewicz et al., 2013Dutkiewicz et al., , 2020Follows et al., 2007;Ward et al., 2012). Trait-based models also provide a framework for testing environmental impacts on species (such as foraminifera) where physiological information is limited, by exploiting knowledge from different organisms with similar traits.
In an initial investigation of applying this approach to foraminifera, Grigoratou et al. (2019) and    (Ward et al., 2018). For details on model code availability, see the Code and Data Availability section.
The food web in the model resolves a total of 16 plankton groups, which are split into eight different size classes of phytoplankton (autotrophs) and eight different size classes of zooplankton (heterotrophs). One of the 8 heterotroph size classes is reassigned to represent a calcifying zooplankton -here, non-spinose foraminifera (Table 1). In essence, the 0D model from Grigoratou (2019) is replicated across the model surface grid, which is now dynamically interlinked by the modelled climate, circulation, and biogeochemistry.
Phytoplankton are limited by light and nutrients (phosphorus, iron) while zooplankton growth depends on prey density and size (Ward et al., 2018). Zooplankton (excluding non-spinose foraminifera) are assumed to be omnivorous active predators. The grazing function of zooplankton follows a Holling type II response with a prey refuge term. The total zooplankton biomass varies as a function of grazing gains, predation, and mortality. Temperature effect on plankton growth is represented with an Arrhenius type equation (Data S1, equation 1). All groups include a background mortality to represent loss due to viral/bacterial infection or natural death. Life cycles and vertical migration are not considered in the model; the plankton biomass is restricted to the model's surface layer (0-80.8 m depth).

| The symbiont barren non-spinose foraminifera functional type
Non-spinose foraminifera are represented by a spherical equivalent shell diameter of 190 µm (Table 1). This shell diameter (190 μm) characterizes adult stages of non-spinose planktonic foraminifera from all species, considering small species such as Neogloboquadrina pachyderma, and Neogloboquadrina incompta (Schmidt, Renaud, et al., 2004). We define non-spinose foraminifera in the model as passive herbivorous feeders, following the observations showing that due to the lack of spines, non-spinose diet mostly consists of phytoplankton (Schiebel & Hemleben, 2017).
As the environmental factors affecting calcification are complex and incompletely understood, we do not explicitly represent the calcification process in the model. Instead, our approach is to simulate the existence of this trait based on the trade-offs of energetic costs versus predation benefits (Grigoratou et al., 2019, more details in Data S1, section 1.2 'ECOGEM'). Following the 'defence' theory, we account for energy loss to be the main cost of creating a shell (in this case, out of calcium carbonate) and protect the main benefit (Ehrlich & Gaedke, 2018;Harvell, 1990

| Empirical observations of non-spinose foraminifera
For the model-data comparison, we selected empirical observations from plankton tows and sediment traps across 26 representative locations of foraminifera's key geographical zones (Figures 1-4, Tables S1 and S2). We compile observations from sediment traps to investigate the modelled flux. We do not include observations from the Mediterranean Sea and the Arctic Ocean as those regions are poorly spatially resolved in the model and thus excluded from our results.
The majority of the plankton tow observations are from the upper F I G U R E 2 (a) Global biomass concentration (log10 mmol C m −3 ) of non-spinose planktonic foraminifera in response to pre-industrial pCO 2 (278 ppm  (Table S1) where most of the non-spinose species can be found (Bé & Tolderlund, 1971;Berger, 1969;Rebotim et al., 2017).
In the model, foraminifera are simulated in terms of organic biomass concentration (mmol C m −3 ). We also consider the organic matter export flux arising from foraminifera (mmol C m −3 day −1 ), which comprises component fluxes from (1) background mortality and (2) the foraminifera biomass which has not been assimilated by predators ('sloppy' feeding; for more details see Ward et al., 2018 to 30 μg CaCO 3 ind −1 (e.g. Globorotalia truncatulinoides), assuming the arithmetic mean is unlikely to be representative. We hence convert organic carbon flux to that of CaCO 3 for end-member weights of 5 and 30 μg CaCO 3 ind −1 to create a maximum range of 'uncertainty'. For simplicity, we also report the results of assuming a typical species shell weight of 10 μg CaCO 3 ind −1 (Table S4). We compare our global export flux with Schiebel (2002) estimations for the first 100 m.

| Model set up
We spun up the model for 10,000 years under an invariant pre- To evaluate the link between foraminifera's distribution and environmental conditions, we applied a pairwise Spearman correlation between foraminifera's biomass, temperature, salinity, nutrient concentration, and with the plankton groups for the following biozones: subpolar, Indian Ocean, mid, and low latitudes. We applied the correlation for the pre-industrial and RCP8.5 scenario (2050, 2100). We ran the statistical analysis in Python 3.9 with the Pingouin0.3.8 software (Vallat, 2018). Further details of the statistical analysis can be found in the Data S1.

| Global pre-industrial conditions
3.1.1 | Biogeographical patterns, distribution, and calcium carbonate flux of Southern Ocean in the model (Figures 2a, 3 and 4, Data S1). The modelled biogeographical patterns generally agree with the main large-scale features of the relative prevalence of non-spinose species that shows an abundance increase form low to high latitudes (Figures 1a and 2a). For a more in-depth model-data comparison, the reader is referred to Data S1 section 2.1. Comparison with observations.
Considering foraminifera distribution and seasonality, the model overall exhibits a lower variance as compared to field observations and often differs in the timing of the seasonal changes (Figures 3 and   4, Data S1: section 2.1). For instance, although modelled seasonality generally follows the observed patterns in subpolar regions (Figure 4, and reduce short snapshots which cannot reflect intra-annual variability. We also note (as do the authors) that the global foraminifera biomass estimates of Schiebel and Movellan (2012) are somewhat conservative as they focus on size fractions greater than 125 µm.
Considering size fractions >100 µm allows for global planktic foraminifera biomass production to be two times higher, and including all size fractions (including juveniles), to be more than three times higher. Although this would bring our modelled estimate of 7  Depending on the region, we find that the abiotic parameters temperature and phosphate (the limiting nutrient for foraminifera's prey) exhibit the highest correlation coefficient with foraminifera abundance (r; Table 2). Phosphate is the main abiotic driver of foraminifera's concentration for the Indian Ocean and mid-latitudes, followed by temperature. Temperature is the major abiotic driver of foraminifera distribution in the subpolar regions. Alkalinity and pH have a weak correlation with foraminiferal biomass, except for the Indian Ocean. However, we do not consider the role of these two factors any further as we do not explicitly model the calcification process that should have the strongest effect.

| Future projections of foraminifera biogeography, biomass, and calcification
The climate response to imposed increasing atmospheric CO 2 concentrations in ForamEcoGEnIE is characterized by a global and annual mean sea surface average (SST) warming of 1.2°C (RCP6) or 1.4°C (RCP8.5) by 2050, and 2.1°C (RCP6) or 2.8°C (RCP8.5) at 2100 ( Figure S3). This climate response lies within the approximated ±1°C spread of higher spatial resolution and fully coupled climate models, that project a model ensemble mean SST rise relative to year 1901 of 3.2°C (RCP8.5) by 2100; table 1 in Gattuso et al., 2015).
Globally, and for both scenarios, the projected changes in foraminifera abundance are strongly related to the overall dynamics of the simulated plankton community, showing that resource availability and competition remain the main drivers of distribution changes.

2100.
Based on the projected changes in foraminifera organic carbon export and applying our preferred (10 µg) conversion factor between organic and carbonate carbon described earlier, we project a global decrease in foraminifera CaCO 3 flux of 8% (RCP6) to 10% (RCP8.5) by 2050 and 13% (RCP6)-18% (RCP8.5) by 2100 ( Figure 5b). Again, there is a regional heterogeneity and the projected CaCO 3 flux increases in subpolar North Atlantic and Southern Ocean. It is important to remember that these estimates do not consider any potential changes in foraminifera's biomineralization due to warming and acidification and are therefore likely conservative values (Fox et al., 2020;Lombard et al., 2010).

| DISCUSS ION
Our study is the first trait-based approach to assess the global distribution and climate change sensitivity of non-spinose foraminifera.  (Jonkers et al., 2021).
The absence of non-spinose forms in most oligotrophic regions in the model suggests that herbivory might not be an efficient-enough diet due to the limited food availability and strong resource competition with other zooplankton groups. This is supported by observations which show that in the oligotrophic (sub)tropics, non-spinose species complement their diet with detritus or small zooplankton (Bé & Tolderlund, 1971;Schiebel & Hemleben, 2017) and are symbiont facultative (Takagi et al., 2019). Laboratory studies have shown that symbiosis with autotrophic algae positively impacts planktonic foraminifera growth (e.g. Bé, 1982;Spero & De Niro, 1987). However, symbiotic relationships might increase the vulnerability of foraminifera to future warming as bleaching due to symbiont loss has been documented in the geological record (Edgar et al., 2013) akin to coral bleaching.
Our model does not capture foraminifera's standing stocks in the modern polar Southern Ocean. We speculate that this is due to the lack of an omnivorous/detritus diet of polar non-spinose species (Greco et al., 2021) or other polar traits such as dormancy (Dieckmann et al., 1991;Spindler & Dieckmann, 1986) and thermal tolerance. It is thus crucial that future studies include different foraging strategies and the symbiotic relationship with algae to study the foraminifera distribution and their potential vulnerability in the future.
Temperature has previously been considered the main driver of foraminifera distribution, followed by food availability, stratification, and carbonate saturation (Bé & Tolderlund, 1971;Ottens & Nederbragt, 1992;Schiebel et al., 2001Schiebel et al., , 2002Schmidt, Renaud, et al., 2004;Storz et al., 2009). Sediment data have recorded community shifts in foraminifera related to temperature increases occurring since pre-industrial times (Jonkers et al., 2019). It is important to note, though, that temperature is the environmental factor with the best data availability and proxy understanding (CLIMAP, 1976;Kucera et al., 2005;MARGO Project, 2009 This poleward shift has also been documented for many plankton groups over the last decades (Beaugrand et al., 2012;Hastings et al., 2020;Poloczanska et al., 2013). Only one other study has modelled planktonic foraminifera's distribution under future climate scenarios (RCP8.5), using the FORAMCLIM model (Roy et al., 2015).
FORAMCLIM is calibrated on species' specific growth rates and non-spinose taxa are represented as herbivorous feeders. The foraminifera community depends on temperature and prey availability without resource competition or top-down control from other zooplankton. Despite the differences in modelling approaches between FORAMCLIM and ForamEcoGEnIE (see Data S1 for more details), both models project similar changes in the distribution of nonspinose foraminifera (i.e. higher loss in the tropics and increase in (sub)polar regions). This increases our confidence in our understanding and shows that our selected traits and trade-offs allow future projections without being calibrated on taxon-specific physiological rates.
Changes in foraminifera distribution can impact the ocean inorganic carbon cycle via changes in ocean carbonate production. At low latitudes, non-spinose contribute approximately 1/3 to the foraminifera assemblage (Schmuker & Schiebel, 2002) and the calcium carbonate production is mainly driven by changes in the biomass of spinose species. The loss of non-spinose biomass and extirpation will remove this contribution to foraminifera calcium carbonate flux, despite those regions staying supersaturated with regards to calcite under both CO 2 emissions scenarios. At the same time, calcium carbonate export is projected to increase in the high-latitude regions, specifically the subpolar North Atlantic and the Southern Ocean.
It is important to note that these regions are highly susceptible to changes in carbonate chemistry due to ice melt and they are naturally lower in carbonate saturation due to higher CO 2 gas solubility in colder waters. This dissolution would exert a small but negative feedback on CO 2 . Should foraminifera reduce calcification rates in response to lower ambient carbonate saturation (Bijma et al., 1999;Lombard et al., 2010;Russell et al., 2004;Spero et al., 1997), this would further increase a negative stabilizing feedback on CO 2 and saturation. Still, foraminifera calcification process cannot be separated from their ecophysiology. Laboratory studies have shown that the prey type and amount influence calcification rates (Anderson et al., 1979;Spindler et al., 1984). Starvation leads to slower chamber formation and death, while overfeeding causes higher growth rates of cytoplasm, shell formation, and gametogenesis resulting in shorter life cycles (e.g. Anderson et al., 1979;Bé et al., 1981;Caron & Bé, 1984;Spindler et al., 1984). Therefore, a lack of prey could also influence carbonate production even under fully oversaturated conditions.

| CON CLUS IONS
We developed the first Earth System model which includes a traitbased ecosystem model of planktonic foraminifera. With this model, we assessed the environmental controls on modern non-spinose foraminifera and found that their distribution can be explained by the dynamics of the plankton community, followed by temperature and phosphate. In response to future climate change, the model suggests that the biomass of non-spinose foraminifera decreases by 8%-11% by 2050 and 14%-18% by 2100 (RCP6.5 and RCP8.5).
Losses are not global though; shifts in the primary productivity to higher latitudes result in foraminifera biomass increasing in the subpolar North Atlantic and Southern Ocean. The model projects that the non-spinose abundance will strongly reduce in low latitudes with an extirpation in the South Pacific by 2100. As a consequence of this biomass shift, we estimate that the calcium carbonate flux of non-spinose foraminifera will drop by 13%-18% by 2100 globally but with an increased flux at higher latitudes. A much deeper understanding of the physiological process of foraminifera calcification and relationship to environmental conditions via new field, laboratory and modelling studies, is clearly going to be necessary for refining projections of calcium carbonate production by foraminifera and hence future feedbacks on atmospheric CO 2 . sharing the data analysis for the Figure 1a. We also thank the Editor and the two anonymous reviewers for their thoughtful and constructive feedback which improved an earlier version of the manuscript and the supporting information.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The ForamEcoGEnIE code can be found here https://github.com/ derpy code/cgenie.muffin.
The data and code for the statistical analysis and figures are publicly available and can be found online at https://zenodo.org/recor d/5564573 (Grigoratou, Monteiro, Wilson, et al., 2021).