Shutdown of Atlantic overturning circulation could cause persistent increase of primary production in the Pacific

A potential shutdown of the Atlantic meridional overturning circulation (AMOC) is commonly recognized to have a significant impact on the Northern hemispheric climate, notably in Northern Europe. The collapse of the northbound heat transport by the AMOC is supposed to cool down surface air temperatures at the Scandinavian coast by up to 6 K accompanied by a concomitant nutrient starvation of phytoplankton in Subarctic and Arctic regions. However, besides local and regional impacts, tipping the AMOC into a weaker state by anthropogenic carbon dioxide (CO 2 ) and associated freshwater forcing could also have surprising remote effects. In order to investigate possible long-term impacts of an AMOC shutdown on ocean biogeochemistry, we employ an Earth system model of intermediate complexity using idealized scenarios of century-scale atmospheric 2 × CO 2 and 4 × CO 2 pulses combined with North Atlantic freshwater forcing. The results show a continued increase in primary production, in particular in the Eastern equatorial Pacific, due to a decrease in iron limitation following the AMOC shutdown. Tracer simulations indicate that bioavailable dissolved iron brought by Aeolian


Introduction
Phytoplankton forms the base of the marine food web and is an integral component of the global carbon cycle.The current global distribution of pelagic biological production zones reveals a typical spatial pattern which can be characterized by four biomes (polar, westerlies, trade winds, tropical;Vichi et al 2011) and their subdivision into 56 ocean biogeochemical provinces (Longhurst et al 1995, Reygondeau et al 2013, 2020).In this context, the Southern Ocean (SO), the Western subartic Pacific and the equatorial Pacific are regarded as the most prominent high nutrient, low chlorophyll (HNLC) provinces in the global ocean.These areas are characterized by a comparatively low biological productivity despite a sufficiently high abundance of macro-nutrients such as nitrate (NO − 3 ) and phosphate (PO 3− 4 ).Furthermore, HNLC areas also exhibit low concentrations of micronutrients where the most important constituent is bioavailable iron.Already during the late eighties of the previous century, Martin and Fitzwater (1988) proposed iron limitation as the most probable cause of the existence of HNLC areas.
In their seminal papers, Kolber et al (1994) and Behrenfeld et al (1996) identified the Eastern equatorial Pacific (EEP) as a region of pronounced iron limitation with respect to phytoplankton photosynthesis and primary production.In situ iron enrichment experiments (IronEx II, Landry et al 2000) caused a relaxation of iron limitation in several spots in the EEP lasting over a couple of weeks by provoking a large increase in phytoplankton biomass and a shift in the community size structure towards larger cells, notably diatoms.Therefore, the results from IronEx II are regarded as a confirmation of the iron limitation hypothesis for the EEP HNLC province.
Current climatological estimates of annual net primary production (NPP) in the EEP range between about 4.5 and 6.5 gigatons of carbon per year (Rousseaux et al 2014; GtC yr −1 , 1Gt = 10 15 g), which is about 10% the total global ocean NPP based on a value of 50 GtC yr −1 derived from remote sensing data (Field et al 1998).The Southern Ocean (SO), a second important HNLC zone, contributes a comparable amount of NPP of the order of 5 GtC yr −1 .
Recent climate projections utilizing a variety of anthropogenic greenhouse gas emission scenarios by employing a suite of CMIP5/CMIP6-type models reveal an overall weakening of NPP on the global scale (Kwiatkowski et al 2020, Reygondeau et al 2020).As global mean sea surface temperatures (SSTs) rise, upper oceans stratification strengthens with the consequence of a steepening of the nutricline.As a result, less nutrients from the sub-surface layer will replenish the water masses in the euphotic zone.In this context the equatorial Pacific production zone is projected to shrink by the year 2100 under RCP-2.6 and RCP-8.5 emission scenarios (Reygondeau et al 2020).
In order to get deeper insights into the stability of the Earth's climate system under anthropogenic forcing, it is of greatest interest to identify and pinpoint potential tipping elements (Lenton et al 2008, Steffen et al 2021).Besides utilization of socio-economically based emission scenarios, it is often useful to artificially disturb an Earth system model by applying a pulse-like external forcing and to analyze the subsequent climate response.A classical example is freshwater hosing of the North Atlantic (NA) to provoke a shutdown of the Atlantic meridional overturning circulation (AMOC; Rahmstorf 1995, Liu et al 2017).This is particularly relevant in light of increased meltwater input from the Greenland ice sheet expected under future warming (Caesar et al 2018, Hofer et al 2020, Weijer et al 2020).
Starting from a preindustrial pCO 2 value of 280 ppmv, typical idealized experimental set-ups utilize prescribed exponentially growing atmospheric carbon dioxide (CO 2 ) levels up to a certain limit (e.g.2×CO 2 , 4×CO 2 ) followed by a subsequent transient decline to its initial level to trigger climate system responses capable to force the whole system into a new state (Jeltsch-Thoemmes et al 2020).This kind of experiments can be regarded as an attempt to probe the existence of potential tipping points (Lenton et al 2008) in an Earth system model.
Here we apply a comparable atmospheric CO 2 forcing to an Earth system model of intermediate complexity (EMIC), and combine it with a concomitant freshwater flux anomaly in the NA.As a result, in the aftermath of the perturbation when atmospheric CO 2 reaches its preindustrial level again, NPP in the EEP increases on century scale by up to 10 per cent.In this context, we are investigating whether a prolonged weakening or shutdown of the AMOC could cause an increase in NPP in the Pacific due to enhanced near-surface transport of bioavailable iron from the Atlantic.whereby an assumption of a universal vertical structure is made.Emission, atmospheric transport and deposition of Aeolian dust (which provides a source for bio-available iron to the ocean) is implemented according to Bauer and Ganopolski (2010).The spatio-temporal distribution of terrestrial vegetation types and inland ice is described according to climatological data sets (Montoya et al 2005).The horizontal resolution of the ocean component of 3.75 • ×3.75 • is rather coarse but finer than that of the atmosphere (7.5 • ×22.5 • ).Vertically, the ocean is divided into 24 layers with thickness increasing from 25 m for the uppermost layer to about 500 m for the lowest layer.

Marine carbon cycle and biogeochemistry
An interactively coupled marine carbon cycle and biogeochemistry model operates in CLIMBER-3α+C which exchanges fluxes and tracer concentrations with the OGCM MOM-3 and provides CO 2 levels to the long wave radiation module of POTSDAM-2.Its development branches off from the Hamburg ocean carbon cycle model version 3.1 (HAMOCC-3.1,Six and Maier-Reimer 1996) and includes several upgrades such as a simple parameterization of the marine iron cycle (Parekh et al 2005) and a mineral ballast driven vertical export of biogenic matter (Hofmann andSchellnhuber 2009, Hofmann et al 2019).The OGCM MOM-3 employs an interactive and circulation dependent parameterization of the transport of mesoscale eddies (Hofmann and Morales Maqueda 2011).
The terrestrial carbon cycle coupled to CLIMBER-3α+C was parameterized by employing a simple box model (Hofmann et al 2019) akin to MAGICC (Meinshausen et al 2011) comprising a plant, a litter, and a soil carbon box.However, the model does not account for land-use change.

Iron co-limitation
For the sake of simplicity and computational performance, CLIMBER-3α+C only simulates the dynamics of one single macro-nutrient: phosphate (PO 3− 4 ).Phosphate limitation of phytoplankton growth is parameterized along a Michaelis-Menten term where P 0 = 0.008 µmol l −1 is the half saturation constant and [PO −3 4 ] the phosphate ion concentration computed by CLIMBER-3α+C.
Besides the macro-nutrient phosphate, the model also accounts for the kinetics of the micro-nutrient iron.The iron limitation factor L Fe of phytoplankton growth is also parameterized according to a Michaelis-Menten function . (2) Here, [Fe T ] -the total dissolved iron concentration, which is calculated according to Parekh et al (2005)-is in units of µmol l −1 and the half saturation constant is assumed to be a function depending on SST in • C: which serves as an empirical allometric correction factor (EACF), where T Fe1 = 32 • C and T Fe2 = 60 • C are simple tuning parameters which provide almost realistic spatial Fe T distribution pattern in CLIMBER-3α+C (see figure S3).The EACF accounts for the different iron demand of different phytoplankton cell sizes.While smaller species with a higher surface to volume ratio and dwelling in warm tropical waters have a lower demand on bioavailable iron, larger cells living in cold polar regions need higher iron concentrations to grow (Sunda and Huntsman 1997).
Due to the general nutrient deficiency of the photic zone, a depth of 100 m was chosen for the geographical analysis, as this is where the assumed nutrient limitation can best be read.

External climate forcing and emission scenarios
Application of climate forcings have been started after a spin-up of several thousand simulation years at atmospheric CO 2 concentrations of 280 ppm (baseline), so that the simulated Earth system state could stabilize first under conditions similar to the pre-industrial era.Then six different scenarios were explored in separate runs, combining CO 2 and freshwater forcing (see figures 1(a) and (b), which are: (i) CONTROL: control run with CO 2 remaining at pre-industrial levels, i.e. baseline, with no freshwater addition.In this scenario, the model shows a drift per century of −1 GtC for total CO 2 , 0.05 GtC for NPP and 0.2 ppm for pCO 2 .
Maximum AMOC strength shows no drift but oscillates around 14.5 ± 0.5 Sv. (ii) 2×CO 2 : 1 % increase of atmospheric CO 2 concentrations until a doubling is reached, declining symmetrically afterwards back to baseline.(iii) 4×CO 2 : similar to 2×CO 2 but peaking at fourfold baseline CO 2 levels.(iv) 2×CO 2 HOSING: same as 2×CO 2 except for additional NA freshwater forcing of F FWT ⩽ 0.025 Sv uniformly distributed between 55 The idealized freshwater hosing scenarios emulate the effects of Greenland meltwater input into the NA in CLIMBER-3α+C, which cannot simulate dynamic melting of ice sheets.Values for freshwater forcing of F FWT ⩽ 0.05 Sv (1 Sv = 10 6 m 3 s −1 ) are in line with recent CMIP6 simulations under the SSP585 emission path revealing a surface mass balance anomaly of the Greenland ice sheet of −1332 Gt yr −1 (0.042 Sv) by the end of the 21st century (Hofer et al 2020).

AMOC weakening
As can be seen in figure 1(c), the AMOC stream function decreases in all simulation experiments, followed by a recovery above the baseline level, whereby the extent and duration depend largely on the respective scenario.Thus, the simulations without fresh water input (2×CO 2 and 4×CO 2 ) show a reduction of about 30% and 60%, respectively, after which a recovery occurs immediately and the AMOC returns to the baseline level at the end of the CO 2 forcing.In the scenarios with freshwater input into the NA (HOSING, 2×CO 2 HOSING, and 4×CO 2 HOSING, on the other hand, there is a decrease in AMOC strength of about 90%, which corresponds to a total shutdown.The subsequent recovery phase seems to depend exclusively on the amount of freshwater supplied: in the case of HOSING and 4×CO 2 HOSING, recovery to the base level occurs approximately at T = 500, more than 200 years after the end of the forcings and the recovery for scenario 2×CO 2 HOSING.

Transient and long-term NPP changes
For all scenarios, a long-term increase of global NPP was found, following a transient decrease during the period of elevated CO 2 (see figure 1(d)).The strongest NPP decrease always occurs a few years after reaching the CO 2 maximum of the respective scenario.A maximum NPP decrease of 15%-20% is shown for 4×CO 2 and 4×CO 2 HOSING, while the NPP for the 'weaker' scenarios 2×CO 2 and 2×CO 2 HOSING decrease by less than 10%.Subsequently, the global NPP recovers rapidly, whereby the minimum AMOC strength and the maximum recovery rate for NPP coincide for 4×CO 2 HOSING, and rises above the baseline value to reach its maximum about 100 years (depending on the scenario) after the return to the pre-industrial CO 2 level.As with the preceding decrease, the maximum NPP increase is by far the strongest in scenarios 4×CO 2 (ca.10%) and 4×CO 2 HOSING (ca.8%).In addition, global NPP remains elevated (above the range of variation) for several centuries in all scenarios, with a full return to pre-forcing values not having occurred for 4×CO 2 by the end of the simulation (740 years after the end of the CO 2 increase).
In comparison to CO 2 , the pure effect of additional freshwater hosing only leads to a comparatively small decrease of NPP during AMOC shutdown in the HOSING scenario.Comparing 4×CO 2 HOSING and 4×CO 2 , the freshwater effect is reflected in a weaker deflection of the NPP curve.In other words, the NPP reaches higher minimum and lower maximum values in 4×CO 2 HOSING and also returns to its initial baseline state earlier.At the exemplarily chosen time of 500 years after the start of the simulation in figure 2 (corresponding to 240 years after the end of forcing), scenario 4×CO 2 HOSING (figure 2(c)) shows a reduction of the NPP in the NA and an increase in the South Atlantic, in contrast to an increase for the NA and the Indian Ocean for 4×CO 2 (figure 2(a)).Both scenarios 4×CO 2 and 4×CO 2 HOSING have a distinctive NPP increase in the EEP in common.For 2×CO 2 and 2×CO 2 HOSING (figures 2(b) and (d), consistent geographical patterns of a minimal NPP increase are recognizable in the EEP and SO.As can also be seen in figure 1(d), the NPP has almost completely returned to the baseline for these scenarios at T = 500.Therefore, we focus on the 'stronger' scenarios 4×CO 2 and 4×CO 2 HOSING in the following sections.
In addition to the global NPP, we will also focus on the SO and the EEP, which are known as HNLC zones.The EEP (figure 1(e)) shows similarities to the global mean values, as the NPP for the 4xCO2 and 4xCO2HOSING scenarios drops significantly during the forcing phase and afterwards exceeds the baseline (CONTROL) level by far.In comparison, the NPP for the other scenarios swings relatively close to the baseline.For the SO (figure 1(f)), the NPP is strongly influenced by freshwater input in all the HOSING scenarios (2×CO 2 HOSING, 4×CO 2 HOSING, HOSING), showing a sharp increase at the beginning of the forcing period.After the end of the forcing, this is followed by a rapid decline in the NPP, especially for 2×CO 2 HOSING and HOSING, which have already fallen back to the baseline at the end of the forcing at T = 280.In contrast, 4×CO 2 HOSING shows a rather asymptotic decline, similar to 4×CO 2 , whereby the NPP moves close to the baseline after T = 500.

Long-term change in nutrient limitation
On global average, the limitation of marine NPP by Fe is much more pronounced than by PO 4 , as indicated by significantly lower values of the limiting factor L Fe (see figures 1(e) and (f) over the entire simulation period.Concurrently, Fe availability (i.e.L Fe ) increases strongly in the course of the CO 2 elevation for all scenarios, rising to a maximum of +14% for 4×CO 2 and +22% for 4×CO 2 HOSING shortly after the CO 2 peak is reached.If we compare this with the fact that a moderate L Fe increase of +8% is also observed for the HOSING scenario, it can be deduced that the effects of CO 2 increase and freshwater forcing are essentially additive.As can be seen figure 3(a), L Fe has increased practically worldwide in 4×CO 2 HOSING and for the majority of the globe in 4×CO 2 , especially in the South Atlantic and the equatorial Pacific.This implies that 500 years after the start of the simulation, Fe limitation of NPP in these regions is still reduced over large areas.
In comparison, L PO4 (figures 3(b) and (d) shows a temporary decrease, which, however, subsides more quickly than L Fe and remains visible only in some places after 500 years.Geographically, this creates a rather mixed picture of decreases and increases, especially in the Atlantic and the Indian Ocean.

Fe transport
Given that the most plausible explanation for the NPP increase, especially in the EEP, is an increase in Fe availability, a possible transport pathway for Fe into the EEP was traced here.Based on dust input from the Sahara into the NA as a source of iron, figure 4 shows the results of the tracer experiments.In the process, the tracer that was concentrated on the West African coast at T = 5 has spread across the entire Atlantic and the Arctic Ocean by T = 140.The near-surface concentrations for CONTROL and 4×CO 2 remain very similar until the end of the experiment and show only a moderate increase in the SO apart from the Atlantic.In contrast, there is a much stronger dispersion in 4×CO 2 HOSING, as the area of high tracer concentrations first extends eastward along the SO and then northwards into the Indian Ocean and the EEP.It can be seen in figure 5 that for 4×CO 2 HOSING, different from CONTROL and 4×CO 2 , there is no longer any enriched, southward deep water convection in the NA, which results in the near-surface concentration of the tracer.

Discussion
Decreasing NPP and nutrient concentrations as a result of the CO 2 increase corresponds to the results that have been found both for socio-economically based emission scenarios RCP and SSP (Kwiatkowski et al 2020, Reygondeau et al 2020) as well as idealized scenarios (Boucher et al 2012), partly explained by diminishing vertical nutrient supply caused by the increasing stratification of the uppermost water column under rising temperatures.This largely corresponds to the simulations evaluated here, since in the EEP, SO and Indian Ocean regions, a decrease or increase in mixed layer depth is followed by a similar change in NPP (see figure S2).Apart from this, however, no overarching temporal coupling can be discerned, whereby geographically the NPP increase coincides with an increase in Fe concentrations in the EEP, the SO and the South Atlantic regions.The EEP can serve here as an exemplary HNLC region, since the Fe limitation of algal growth there has already been established and an increase in NPP through iron addition has been experimentally proven (IronEx II;Landry et al 2000).
Complementary to the vertical mixing already mentioned, the tracer simulations of the 4×CO 2 HOSING scenario show that a lateral redistribution of iron from the NA to the EEP is possible.The NA deep water formation usually removes Fe from the near-surface layers, which is regularly brought in by dust from the Sahara (here Saharan dust being the only external source of marine iron).A possibly extensive weakening of the AMOC due to warming and/or freshwater input would interrupt the removal.This would retain iron in or near the photic zone (see figure 5), which would normally be bound in deep water for many centuries, releasing an increasing amount of Fe over time into the SO and further into the Pacific.If the input is maintained, a long-term NPP increase can develop, provided that there are no other substantial limiting factors.It seems plausible that this could also explain the increase in NPP in the SO, another Fe-limited HNLC region.The fact that the NPP increase in the SO is significantly lower than in the EEP is probably partly due to a higher phosphate and light limitation, but partly also due to the fact that the SO has a shorter retention time as a "transit station" for iron.In the NA, a sharp decline in PO 3− 4 appears to be the limiting factor for the NPP, but this does not seem to play a significant role for long-term effects in the NPPglobally due to the generally high availability of phosphate as characterized by L PO4 .
Nevertheless, the mechanism of a lateral redistribution of near-surface iron from the NA appears to necessitate freshwater addition causing a (almost) complete shutdown of the AMOC.The reason is that there is no visible enhancement of tracer concentrations in the same experiment for 4×CO 2 (relative to CONTROL).That means that even a considerable reduction of the AMOC strength by 60% (as it occurs for 4×CO 2 ) does not suffice to trigger this event.Considering the AMOC in the context of Greenland meltwater input supports the view of the AMOC as a mediator rather than the initiator of a tipping point cascade (Wunderling et al 2021).
Overall, NPP remains increased both globally (figure 1(c)) and in the EEP (figure 2(a)) in the 4×CO 2 scenario, as does the Fe availability L Fe .Preliminary results from another tracer experiment (figure S4) show that surplus iron could be transported from the Arabian Sea (that also receives high loads of dust input from the Sahara).But currently it can only be speculated what could be the cause for iron accumulation in the Arabian Sea (figure S3(a)), especially as there is no increase of Fe availability (figure 3(a)).One possible explanation lies in the enlargement of the oligotrophic (PO 4 poor) region in the Indian Ocean, either by an expansion of Indian subtropical gyre (represented in the model) or due to the intensification of the Indian Monsoon (Katzenberger et al 2021, not represented in the model) which may results in an dilution of surface waters by increasing rainfall and river runoff into the Indian Ocean (represented in the model).Thereby, the resulting increase in PO 4 limitation would reduce algal growth (primary production) and the concomitant Fe consumption.The temporary enlargement of the subtropical gyre might also be a relevant factor in the decrease of the global mean of L PO4 , but was not yet further investigated here.If it can be assumed that near-surface transport of iron towards the EEP takes place under 4×CO 2 , this would probably be supported by stratification in the Indian Ocean and in the SO (red curves in figure S2).
Another possible source of nutrients for scenarios 2×CO 2 HOSING and 4×CO 2 HOSING may also have been created by the mixing from deeper layers in the SO.Comparing HOSING and 4×CO 2 in the SO, it becomes clear how the freshwater input counteracts the stratifying effect of warming (i.e. the CO 2 effect).The massive increase in mixed-layer depth during the forcing phase suggests that if iron is released in the SO, it would also have contributed to the increase in NPP in the EEP.This aspect is all the more important because scenarios 2×CO 2 HOSING and 4×CO 2 HOSING may be regarded more realistic than 2×CO 2 and 4×CO 2 as the absence of Greenland melt water input can be assumed to be unlikely in the long term.
In addition to the above-mentioned, possible effects of reduced iron scavenging (i.e.reduced Fe binding to CaCO 3 ballast due to acidification and decalcification) due to increased CaCO 3 decomposition remain to be considered, in our simulations this has turned out to be insignificant for NPP (see figure S1).Grazing by zooplankton can be a significant factor, but due to its complexity it cannot be sufficiently illuminated here.The associated change in food webs and ecosystems both during and after CO 2 rise (Reygondeau et al 2020, Schwinger et al 2022) probably needs to be considered in a much more regional, ecological context.
The potential effects of the marine nutrient redistribution presented here are difficult to assess in the long term and will in any case first need to be confirmed.Nevertheless, the consequences for the ecosystems of the EEP alone could be massive, because in addition to all the known consequences of climate change, a century lasting iron fertilization pulse could fundamentally change the food web in the pelagic zone.In terms of societal impacts, this would not only pose a major uncertainty for the global food supply, but also for the structures in the neighboring nations of the affected areas that have grown out of the use of seafood.

Conclusion
In this study, data from different idealized model scenarios were analyzed to determine relationships between AMOC strength, NPP, Fe and PO 4 during and after extended periods of atmospheric CO 2 rise and NA freshwater input.Depending on the strength and combination of the applied forcings, a decrease or shutdown of the AMOC is observed.This is accompanied by a transient decrease and postforcing increase in marine NPP, with a particularly pronounced change in the EEP.Tracer experiments demonstrate that this is due to a global Fe redistribution and fertilization in HNLC areas caused by the AMOC shutdown.
The large-scale redistribution of nutrients, especially Fe, and the accompanying NPP increase can be directly related to the change in ocean currents caused by the collapse of the AMOC.At the same time, similar long-term effects can occur alongside a strong weakening (rather than shutdown) of the AMOC, without a connection between these phenomena being discernible.It remains unclear whether the strength of the AMOC stream function must fall below a critical value to cause an accumulation of near-surface nutrients in the NA, or what mechanisms can trigger this in the Arabian Sea, for example.Due to model limitations, the necessary investigations into climate change-related changes in precipitation, wind flow patterns and the monsoon system cannot be carried out here.Likewise, the tracer simulations applied here can only reflect the passive transport of Fe, but not the biogeochemical changes, for example through scavenging or uptake by phytoplankton.More advanced EMICs or higher resolution Earth system models can help determine whether the effects found here can be reproduced under more realistic conditions.If this proves to be robust, further studies on the impacts on regional marine ecosystems will be necessary.

Figure 1 .
Figure 1.Forcings (a)-(b) and Earth-system response (c)-(h) for the simulation experiments: (a) Global mean atmospheric CO2 concentrations, (b) total North Atlantic freshwater input, (c) maximum of the AMOC stream function, (d) global marine net primary production, (e) net primary production in the Eastern equatorial Pacific, (f) net primary production in the Southern Ocean, (g) mean limitation factor for PO4, and (h) mean limitation factor for Fe.Vertical lines mark the beginning, peak, and end of the CO2 and freshwater forcings, as shown in (a) and (b).

Figure 5 .
Figure 5. Vertical cut at 30 • W of Fe concentration for the tracer experiment shown in figure 4(color shading).The Atlantic overturning streamfunction is shown as contours.
Petoukhov V, Ganopolski A, Brovkin V, Claussen M, Eliseev A, Kubatzki C and Rahmstorf S 2000 CLIMBER-2: a climate system model of intermediate complexity.Part I: model description and performance for present climate Clim.Dyn.16 1-17 Rahmstorf S 1995 Bifurcations of the Atlantic thermohaline circulation in response to changes in the hydrological cycle Nature 378 145-49 Reygondeau G, Cheung W W L, Wabnitz C C C, Lam V W Y, Froelicher T and Maury O 2020 Climate change-induced emergence of novel biogeochemical provinces Front.Mar.Sci. 7 657 Reygondeau G, Longhurst A, Martinez E, Beaugrand G and Maury D O 2013 Dynamic biogeochemical provinces in the global ocean Glob.Biogeochem.Cycles 27 1046-58 Rousseaux C S and Gregg W W 2014 Interannual variation in phytoplankton primary production at a global scale Remote Sens. 6 1-19 Schwinger J, Asaadi A, Goris N and Lee H 2022 Possibility for strong northern hemisphere high-latitude cooling under negative emissions Nat.Commun.13 1095 Séférian R et al 2020 Tracking improvement in simulated marine biogeochemistry between CMIP5 and CMIP6 Curr.Clim.Change Rep. 6 166-177 Six K D and Maier-Reimer E 1996 Effects of plankton dynamics on seasonal carbon fluxes in an ocean general circulation model Glob.Biogeochem.Cycles 10 559-83 Steffen W et al 2021 Trajectories of the earth system in the anthropocene Proc.Natl Acad.Sci.USA 115 8252-59 Sunda W G and Huntsman S A 1997 Interrelated influence of iron, light and cell size on marine phytoplankton growth Nature 390 389-92 Vichi M, Allen J I, Masina S and Hardman Mountford N J 2011 The emergence of ocean biogeochemical provinces: a quantitative assessment and a diagnostic for model evaluation Glob.Biogeochem.Cycles 25 Weijer W, Cheng W, Garuba O A, Hu A and Nadiga B T 2020 CMIP6 models predict significant 21st century decline of the Atlantic meridional overturning circulation Geophys.Res.Lett.47 1-14 Wunderling N, Donges J F, Kurths J and Winkelmann R 2021 Interacting tipping elements increase risk of climate domino effects under global warming Earth Sys.Dyn. 12 601-19