In this study, we employed Community Earth System Model version 2 (CESM2; Danabasoglu et al., 2020) to investigate marine biogeochemical cycles, focusing on the dynamics of the nutrient and phytoplankton communities in the SPNA region. The marine ecosystem within CESM2 is represented by the Marine Biogeochemical Library (MARBL), which incorporates three explicitly modeled phytoplankton functional groups – diatoms, diazotrophs, pico/nano (small) phytoplankton – as well as one implicit group (calcifiers) and a single zooplankton group, each with distinct source-sink parameters. MARBL also simulates 32 different tracers, which include 17 abiotic constituents such as dissolved inorganic and organic carbon, alkalinity, nutrients, and oxygen, along with 15 biotic tracers associated with phytoplankton biomass (e.g., Carbon (C), Phosphorus (P), Nitrogen (N), Silica (Si), Iron (Fe), and Calcium Carbonate (CaCO₃)), as well as zooplankton carbon.

Within MARBL, phytoplankton growth rates depend on multiple environmental parameters, including temperature, nutrient limitation, and light availability, as detailed in Long et al. (2021). Nutrient limitation is determined by Liebig's law of the minimum, whereby the nutrient available in the smallest quantities limits phytoplankton growth. Changes in PFT concentrations are calculated by source (i.e., net primary productivity) and sink (grazing, linear mortality, and aggregation) terms as elaborated in Long et al. (2021).

Long et al. (2021) evaluated the model's performance in simulating biogeochemical variables such as macronutrient (Nitrate; NO₃, Phosphate; PO₄, Silicate; SiO₃) distribution and NPP and export production (EP). They showed that MARBL effectively simulates the geographical patterns of macronutrients in the SPNA region compared to World Ocean Atlas (WOA) observations. Moreover, comparisons with satellite observations revealed that the global distribution of NPP simulated by MARBL exhibits similar patterns, albeit with slight differences in magnitude.

We analyzed an idealized negative CO₂ emissions pathway that effectively reduces emissions incrementally, considering the development of Carbon Dioxide Removal (CDR) technology and global mitigation efforts. In our experiments, anthropogenic CO₂ emissions increase linearly from an initial CO₂ concentration of 383 ppm to the year 2050, based on the SSP5-8.5 scenario (Fig. S1 in the Supplement) as used in Park et al. (2025). Subsequently, emissions decrease at the same rate until the global mean atmospheric CO₂ concentration returns to its original value (in the year 2196). Immediately after reaching this level, net-zero emissions are maintained until the end of the simulation (year 2400). The atmospheric CO₂ level peaks in the year 2107 (725 ppm), just a few years before achieving net-zero emissions. For better readability, we define periods as follows: CO₂ increasing period (2001–2107) and CO₂ decreasing period (2108–2196), and restoring period as the duration of net-zero emissions (2197–2400), respectively. We also have defined the first 11 years (2001–2011) as the “Climatology period”, and the 11 years (2192–2202) at the end of the CO₂ decreasing period as the “CO₂ down period” to identify irreversible changes of climatic variables under the same CO₂ conditions. We have run a total of ten ensemble members, each with slightly different initial conditions to consider the variability.

We quantified shifts in phytoplankton community using the Bray-Curtis (BC) dissimilarity metric frequently utilized in ecology and biology to quantify differences in species composition between two sites (Bray et al., 1957; Herren and McMahon, 2017). For two time periods, t1 and t2, the BC index is defined as below: 1Bray-Curtis  (BC)  index=1-2∑i=1Rmin(Bt1i,Bt2i)∑i=1RBt1i+Bt2i where, Bti is the period-mean biomass of phytoplankton functional type (PFT) i, and R is the number of PFTs (here R= 3). BC ranges from 0 (identical composition) to 1 (no shared composition among PFTs). We computed the BC index at each grid point using period-mean PFT biomasses between the Climatology (2001–2011; t1), and CO₂ down (2192–2202; t2) periods.

We investigate the shifts in phytoplankton communities between the climatology period and the CO₂ down period. The statistical method known as BC dissimilarity was used to examine the similarity of the phytoplankton communities between the two periods following Cael et al (2021; Detail in Methods). There are pronounced positive values (indicating less similarity) of dissimilarity at high latitudes in the northern hemisphere (NH), including the Arctic and SPNA region. Notably, the most significant shifts in phytoplankton communities emerge in the SPNA region, indicating substantial variations in phytoplankton composition (Fig. 1a).

Figure 1b presents a time series of area-averaged PFT biomass in the SPNA region. As atmospheric CO₂ concentrations increase, diatoms and small phytoplankton exhibit a significant decrease and increase, respectively. These results are consistent with previous studies, which have reported shifts towards smaller PFT-dominant phytoplankton communities in a warmer world (Cael et al., 2021; Doney, 2006; Henson et al., 2021; Morán et al., 2010; Winder and Sommer, 2012). These variations in PFT biomass persist under negative CO₂ emissions. While small phytoplankton increase over time and have a peak in the middle of the CO₂ decreasing period, diatoms continue to decrease and have a minimum value at the same time. Similarly, the diazotroph concentrations decrease, except for a slight initial increase, eventually reaching a minimum at the end of the CO₂ decreasing period. During the restoring period, all PFTs – except diazotrophs, which overshoot – return to near their initial values with diatom concentrations showing a slight weakening. Comparing the two periods (CO₂ down – Climatology) with the same atmospheric CO₂ concentration reveals substantial shifts in the phytoplankton community. Specifically, there is a large decrease in larger cell-sized phytoplankton and an increase in smaller cell-sized phytoplankton, indicating significant downsizing of the phytoplankton community.

All PFTs exhibit strong irreversible shifts in response to CO₂ forcing (Fig. 1c–e). In particular, diatom concentrations decreased to 24 % of their initial value during the CO₂ increasing period, while small phytoplankton increased by about 283 % compared to the Climatology period. The phytoplankton biomass remained nearly constant during the CO₂ decreasing period when the atmospheric CO₂ concentration returned to its initial value of 383 ppm. These results indicate shifts in the phytoplankton community towards dominance by smaller phytoplankton in the SPNA region, despite the same atmospheric CO₂ concentration. In addition, although the diazotroph concentrations constitute a relatively smaller portion of the total PFTs, their pronounced decline during the CO₂ decreasing period leads to a weakening of biological N fixation to the ocean (Fig. S2a). The diazotrophs show different behavior compared to the other two species because of their temperature-limited biological properties (Breitbarth et al., 2007; Yi et al., 2020). Consequently, temporal changes in Sea Surface Temperature (SST), diazotroph concentrations, and N fixation rates in the SPNA region are all closely aligned (Fig. S2).

Smaller phytoplankton have greater competitiveness in oligotrophic and stratified marine environments in a warmer world due to their enhanced surface area-to-volume ratio (Morán et al., 2010; Winder and Sommer, 2012). Within MARBL – the biogeochemical component of the CESM2 Earth System Model – the advantage of smaller phytoplankton in low-nutrient environments is represented by prescribing different constant half-saturation coefficients for each PFT, reflecting the observed differences among phytoplankton size classes in the real ocean (Long et al., 2021). For instance, two PFTs (diatom and small phytoplankton) may possess the same maximum growth rate; however, the half-saturation coefficient is parameterized to be smaller for small phytoplankton. This implies that small phytoplankton can achieve their maximum growth rate with a lower nutrient concentration. Furthermore, the decrease in diatoms due to the shift towards oligotrophic conditions may lead to a reduction in zooplankton and a weakening of grazing by higher-level predators, thereby promoting the growth of phytoplankton. The time series of source and sink terms integrated to 100 m depth exhibits that grazing is the dominant sink contribution to loss for both PFTs, with other terms having relatively minor effects (Fig. S3).

The spatial distribution of PFT biomass differences between the two periods reveals a significant shift across the SPNA (Fig. 2a–c). Across the entire Atlantic and Arctic regions, and even globally (not shown), the most dramatic differences are observed in the SPNA region. For diazotrophs, whose growth rates are temperature-dependent, high latitudes above 50° N exhibit scant biomass climatologically and no variations (Figs. S4 and 2c). Except for diazotrophs, changes in phytoplankton concentrations are generally driven by nutrient availability. While NO₃ is commonly considered the most limiting nutrient globally, observation-based studies have identified silicate as the primary limiting component for phytoplankton growth in the SPNA region (Allen et al., 2005). The CESM2 biogeochemical model, MARBL, also simulates well the Si-limited growth of diatoms in the SPNA region, consistent with observations (Long et al., 2021). The spatial distribution of differences between the two periods for both macronutrients (NO₃, SiO₃) shows similar patterns to PFT biomass changes.

The North Atlantic, including the SPNA region, is characterized by deep convection associated with the AMOC. Changes in the AMOC lead to significant physical alterations in the SPNA region. In particular, the AMOC plays a pivotal role in transporting water volumes from low to high latitudes, and through this process, nutrient-rich water transport can trigger biological changes (Boot et al., 2023). Under global warming, AMOC is expected to progressively weaken due to increased surface temperature and the influx of freshwater. Additionally, it has been reported that under negative emissions, AMOC strength will not recover immediately but will instead decrease further with a delay (An et al., 2021; Oh et al., 2022). This implies that changes in AMOC strength may influence shifts in the phytoplankton community in the SPNA region under a climate mitigation scenario.

Thus, we investigate the changes in AMOC strength and nutrient concentrations in our simulations (Fig. 3a). In our experiments, the intensity of the AMOC continues to decrease with increasing atmospheric CO₂ concentration, persisting until the middle of the CO₂ mitigation pathway and reaching its weakest state around the year 2142 (Fig. 3a). Subsequently, the AMOC gradually strengthens again, returning to its initial condition with overshooting during the restoring period (An et al., 2021; Jackson et al., 2014; Wu et al., 2011). Consistent with the weakening of AMOC strength over time, both nutrient concentrations also continue to decline (Fig. 3a). Interestingly, while both nutrients decline over time, the reduction in SiO₃ concentration halts around the year 2075 and remains steady until approximately the year 2250 before recovering. Conversely, NO₃ decreases until around the year 2170, past the point of minimal AMOC strength, and then recovers with the AMOC turnabout. The distinct responses of SiO₃ and NO₃ to CO₂ forcing arise largely due to the diverging responses of PFTs from the climatology period to the CO₂ down period. While decreases in diatoms-driven NPP reduce the consumption rates of both SiO₃ and NO₃, increases in small-phytoplankton-driven NPP counterbalance the decreasing NO₃ consumption rates, resulting in a more persistent decline in NO₃ (Fig. S7). Reduced diazotrophs-driven NPP and associated N-fixation also contribute to further decline in surface NO₃ concentrations. Changes in the limiting nutrient components may affect the nutrient dynamics of phytoplankton, particularly diatoms, which are composed of larger cells in the SPNA region.

We identified changes in the limiting nutrients for the growth of two PFTs (diatom, small phytoplankton). The spatial patterns of limiting nutrients are shown in Fig. 3c–d for diatoms and in Fig. S5b-c for small phytoplankton, with the SPNA-averaged time series are shown in Figs. 3b and S5a. When spatially averaged, small phytoplankton, unaffected by SiO₃ availability, experience N limitation throughout all timeframes owing to continuously decreasing NO₃ levels (Fig. S5). For diatoms, however, the predominant limiting nutrient shifts from Silicon (Si) to N around the year 2076, coinciding with the stabilization of SiO₃ levels (Fig. 3a–b). During the Climatology period, Si limitation for diatom dominates across most of the SPNA region (Fig. 3c). However, as NO₃ decreases more rapidly than SiO₃, N limitation spreads across the SPNA region, eventually overtaking Si limitation. By the CO₂ down period, Si limitation becomes confined only to small regions, with most areas transitioning to N limitation (Fig. 3c–d).

From the Irminger Sea, located beneath Greenland, to the Nordic Sea, the Si-limited environment persists even during the CO₂ down period. This region, characterized by intense deep convection in the present climate (Heuzé, 2017; Renssen et al., 2005), exhibits the most pronounced shallowing of the mixed layer depth (MLD) than elsewhere, indicating that it is particularly affected by the weakening of the AMOC (Fig. S6). Consequently, SiO₃ concentrations are significantly reduced compared to other regions, maintaining Si limitation. Furthermore, even as the AMOC enters a recovery phase, nutrient levels do not rebound immediately and exhibit a delayed response. During the weakened AMOC phase, MLD in the SPNA region remains in a shutdown state, and its recovery requires sufficient salt-advection accompanying AMOC strengthening (Oh et al., 2022). Therefore, the MLD can begin to recover after the salt advection has sufficiently accumulated. In addition, global warming strengthens Southern Ocean (SO) productivity, which weakens global nutrient redistribution – intensifying deep-ocean nutrient trapping and reducing surface nutrients that might persist even under climate mitigation (Moore et al., 2018; Laufkötter and Gruber, 2018). Therefore, even when the AMOC starts to recover, the amount of nutrients transported from the Southern Hemisphere and low latitudes remains reduced compared to the climatology period. As a result, the recovered nutrient levels stay below the climatological state, giving rise to the delayed recovery. Thus, lagged nutrient retrieval causes a corresponding lagged response in phytoplankton community recovery. As the AMOC strength and nutrient transport are reinstated, nutrient concentrations increase sufficiently by around the year 2250, causing a return to Si limitation. This suggests that while the AMOC plays a critical role in nutrient distribution and the marine ecosystem in the SPNA region, the phytoplankton community may experience stronger irreversible changes than the recovery of AMOC strength under a climate mitigation scenario.

The shift towards smaller phytoplankton communities can have significant ecological, biogeochemical, and climatic impacts. Phytoplankton, composed of larger cells possess a greater capacity for sequestering CO₂ through photosynthesis and gravitational sinking into the ocean. In contrast, organic carbon exported by smaller phytoplankton is more readily degraded by higher-level predators such as bacteria and zooplankton, and can be re-released back to the water column as CO₂ quickly compared to diatoms (Leblanc et al., 2018; Tréguer et al., 2018). Therefore, the collapse of diatom populations during the CO₂ down periods could lead to a loss of biological carbon sequestration competence (biological pump). This attenuation is particularly evident in the SPNA region, where the phytoplankton community shifts are most pronounced (Fig. 4a). The weakening of the biological pump is evident throughout the SPNA region and extends beyond in parts of the Arctic region. Consistent with the strong biological irreversible responses observed in the above results, there is a significant irreversible decline in the organisms' ability to export carbon (POC flux). The abundance of diatoms, which typically dominate the biological carbon pump, has decreased to about a fourth, while small phytoplankton have increased by a factor of 2.7. As a result, when CO₂ concentrations return to initial levels, the POC flux is less than half of the initial condition, indicating that the decrease in diatoms plays a dominant role in POC export changes. Specifically, the SPNA region, currently a key area for CO₂ sequestration (Sabine et al., 2004), is projected to be less effective at exporting carbon than the global average after CO₂ mitigation (Fig. 4b–c). Consequently, while the North Atlantic Ocean is presently known for its strong biological pump, it may lose this status under CO₂ mitigation scenarios (Fig. 4). Therefore, the longer climate mitigation is delayed, the more the miniaturization of phytoplankton communities will accelerate, further slowing down the marine biological pump.

Therefore, even if atmospheric CO₂ concentrations return to their original levels through negative emissions, marine ecosystems, especially the composition of PFTs will exhibit strong irreversible shifts. This disrupted phytoplankton community has the potential to impact regional and global trophic dynamics and food webs, as well as commercial fisheries and more. Moreover, if global warming accelerates again in the context of a downsizing of the phytoplankton community, it is possible that we will experience even more abrupt climate change than at present, as the Earth's capacity for biological carbon export is diminished.