The coupling of FESOM2.1–REcoM3 is a state-of-the-art ocean biogeochemistry model. It performs reasonably well both on a global scale and in various regional applications . The ocean biogeochemistry model REcoM3 has also recently been included into the AWI Climate Model , an Earth System Model, to which computational performance is critical.

This study builds upon the FESOM2.1–REcoM3 setup of which is optimized in its parameterizations for the Arctic Ocean, for which the Arctic biogeochemistry has previously been successfully simulated with regard to primary production, light and nutrient availability . The current setup includes parameterizations for aeolian and riverine nitrogen input, and benthic denitrification . The photo-damage parameterization is turned off, which was developed for a global setup that was found to lead to a TChla concentration that was too high in the Arctic . This study contains the following changes compared to , which have already been presented in :

The cycle of organic carbon is specified in greater detail through the dissolved, particulate and intracellular carbon pools by adding the process description for PCHO and TEP, following and as presented in Sect. . The model now includes 30 different biogeochemical tracers.

This simulation is based on the fARC mesh, an irregular grid, resolving the Arctic region with a resolution of approximately 4.5 km (north of 60° N), and stepwise decreasing resolution (20–120 km) from the coasts and productive areas to the subtropical gyres . Simulations are carried out for the period 1958 to 2019, i.e., a total of 61 years, with atmospheric forcing from the atmospheric reanalysis datasets of JRA55-do v.1.4.0 . The ocean simulation is initialized from temperature and salinity fields of the Polar Science Center Hydrographic Climatology , the biogeochemical simulation from initial fields of dissolved inorganic nitrogen (DIN) and dissolved silicic acid concentration of the World Ocean Atlas climatology , and dissolved inorganic carbon (DIC) and alkalinity of the Global Ocean Data Analysis Project version 2 . Riverine inputs of inorganic and organic carbon and nitrogen are derived from .

The period 1958 to 1990 is considered as spin-up, which allows the biological processes in the surface ocean to reach quasi-equilibrium and with it the concentration of labile organic carbon compounds; thus, the period of 1990 to 2019 is analyzed. Monthly output is obtained for all years, and daily output for April to September 2017 for comparison to in situ observations of May to July 2017 from the ship-based Physical feedbacks of Arctic Boundary Layer Sea ice, Cloud And Aerosol (PASCAL) campaign PS106 on RV Polarstern. A campaign overview is presented in and . Additionally, a FESOM2.1–REcoM3 control run was conducted using the same setup as described above but without the additional organic carbon process descriptions for PCHO and TEP.

This section details the approach of PCHO and TEP implementation into the cycle of organic carbon through the ecosystem as simulated in REcoM3, while quickly recapitulating the relevant state equations and processes. The reader is referred to the overview publication on REcoM3 by for an extensive description of all the biogeochemical model processes.

and developed generalized equations and a parameterization for phytoplankton carbon exudation and for aggregation of organic carbon particles. A quantitative description of these processes and related parameters was obtained by fitting a zero-dimensional version of REcoM to data from a mesocosm experiment . Further, conducted a parameter optimization for REcoM. The PCHO and TEP pools were omitted in recent model releases on the global scale but re-integrated as part of this study.

In a first step, organic carbon is built up by photosynthesis in small phytoplankton and diatoms (Fig. , process A) and exuded in form of two different dissolved, labile pools: PCHO (B) and residual DOC (C, considered as consisting of labile organic compounds, but the chemical composition is not specified in greater detail). In a second step, multiple PCHO molecules may form aggregates or react with already existing TEP (D). Both DOC and TEP are remineralized to dissolved inorganic carbon (DIC, processes E and F). These processes are detailed out further below with respect to the simulated state equations. Following the organic carbon further, a part of the phytoplankton carbon is lost to detritus via aggregation (G) and grazing by the zooplankton (H). The small zooplankton is also grazed upon by macrozooplankton (I). Additionally, the two simulated zooplankton classes contribute to the build-up of detritus through sloppy feeding and mortality (J), but graze on these detritus particles as well (K). A surplus of organic carbon is excreted by zooplankton to DOC (L). Detritus particles can either degraded to DOC (M) or sink down into the water column (N), finally reaching the benthic layer.

The state equation for phytoplankton carbon (shown here for small phytoplankton, Cphy, as an example case) is given in Eq. () following the REcoM3 introduction by . It states that net photosynthesis contributes to the increase of Cphy, whereas aggregation of phytoplankton leads to formation of detritus (cf. Eq. ). Exudation of organic carbon and grazing by zooplankton reduces Cphy: 1S(Cphy)=(Pphy-rphy)⋅Cphy︸net photosynthesis-g⋅Cphy︸aggregation-εphyC⋅fphylim⋅Cphy︸exudation-Gphyzoo1-Gphyzoo2︸grazing by small & macrozoo., where the limitation terms and aggregation processes used in this state equation are presented below (Eqs. –). An explanatory list of state variables and parameters mentioned in this section can be found in Tables  and . Table  contains the terms leading to local sources and sinks of the biogeochemical state variables as presented in .

The DOC sources are represented by the release of organic carbon by two phytoplankton classes and two zooplankton classes, and degradation of organic carbon from both detritus classes. The remineralization to DIC is the only sink for DOC. The simulated source-minus-sink term is given as 2S(CDOC)=(1-fPCHO)⋅εphyC⋅fphylim⋅Cphy︸exudation by small phytoplankton+(1-fPCHO)⋅εdiaC⋅fdialim⋅Cdia︸exudation by diatoms+εzoo1C⋅Czoo1︸excretion by small zoo.+εzoo2C⋅Czoo2︸excretion by macrozoo.+ρDetC⋅fT⋅Cdet1+ρDetC⋅fT⋅Cdet2︸detritus degradation-ρDOC⋅fT⋅CDOC︸remineralization to DIC.

We implemented carbon-based source-minus-sink terms for PCHO and TEP into the current REcoM version 3 based on and with 3S(CPCHO)=fPCHO⋅εphyC⋅fphylim⋅Cphy︸exudation by small phytoplankton+fPCHO⋅εdiaC⋅fdialim⋅Cdia︸exudation by diatoms-αPCHO⋅βPCHO⋅CPCHO⋅CPCHO︸aggregation of PCHO with PCHO-αTEP⋅βTEP⋅CPCHO⋅CTEP︸aggregation of PCHO with TEP,4S(CTEP)=αPCHO⋅βPCHO⋅CPCHO⋅CPCHO︸aggregation of PCHO with PCHO+αTEP⋅βTEP⋅CPCHO⋅CTEP︸aggregation of PCHO with TEP-ρTEP⋅fT⋅CTEP︸remineralization to DIC.

A fraction of the exuded organic carbon by small phytoplankton and diatoms builds up the PCHO pool, while the aggregation of PCHO with other PCHO molecules or with already existing TEP removes PCHO (Eq. ). These aggregation processes are described via a two-size class model for PCHO and TEP. The equations were derived from the Smoluchowski equations for the cluster-to-cluster aggregation of the individual polymer molecules by summation of the carbon content in the different molecule size classes contributing to the PCHO and TEP pool, respectively. Mathematically, this is done by introducing a Heaviside step function which defines the size class for the distinction of PCHO and TEP . The resulting equations depend on the concentrations of PCHO and TEP, on the stickiness of the substrates, and on the carbon specific collision kernels for PCHO–PCHO and PCHO–TEP interaction (Eq. , cf. ). The carbon-specific collision kernels describe the probability of encounter of molecules via molecular diffusion. In this approach, the values of these kernels are parametrized as the collision rate parameters βPCHO and βTEP . Consequently, the TEP pool is increased by aggregation products of PCHO–PCHO and PCHO–TEP. So far, only the bacterial remineralization to DIC (described for simplicity as a temperature-dependent linear loss rate) is considered as sink for TEP (Eq. ).

The limiter functions fphylim and fdialim are given exemplarily for small phytoplankton as 5fphylim=1-exp⁡(-θmax⁡N⋅(∣Δq∣-Δq)2),6Δq=qphyN:Cmax-qphyN:C and regulate the phytoplankton metabolic processes via a non-linear function based on the intracellular nitrogen to carbon ratio (N : C ratio) following and modified for REcoM (Sect. A6.1 in ). The functions for diatoms have a similar structure; parameter values are given in Tables  and . The limiter function is zero for N : C ratios above the value of N : C = 21.2 : 106 = 0.2. No nitrogen uptake and no carbon exudation take place in this case. When the N : C ratio becomes lower, the limiter function increases, hence both nitrogen uptake and carbon exudation of phytoplankton begin. When the current N : C ratio is lower than the Redfield ratio (N : C = 16 : 106 = 0.151), the limiter function is one, not limiting uptake nor exudation. The thresholds have been proposed by based on phytoplankton growth experiments and are not modified further in our model setup. These values are not spatially varying in our setup.

The aggregation rate determines the amount of small phytoplankton and diatom carbon transferred to the detritus pool in the phytoplankton state equation (Eq. ). The aggregation rate g in Eq. () was calculated based on REcoM3 (Eq. A42 in ) with a new term describing the particle stickiness depending on TEP concentration (Eq. B13 in ). The aggregation loss for phytoplankton and detritus classes is based on their corresponding nitrogen concentration. Additionally, the aggregation loss for diatoms depends on nutrient limitation mimicking diatom mucus production . High TEP concentrations (CTEP) increase the particle stickiness based on a Michaelis-Menten-type term and, thus, increase the aggregation rate g: 7g=(ϕPP⋅Nphy+1-qdialim⋅Ndia+ϕPD⋅(Ndet1+Ndet2))⋅CTEPkTEP+CTEP

Analysis is carried out for the Arctic Ocean and contextualized within a global perspective. Boundaries for the Arctic Ocean basins and seas (Fig. ) are set according to and .

The upper 30 m of the ocean are considered for volume-weighted mean concentration of model variables, a depth range which roughly corresponds to the summer mixed layer depth in the Arctic Ocean following the definition of . The surface microlayer, where organic particles are expected to get enriched in a natural ecosystem, is neglected in the current model setup. The volume-weighted mean concentration c‾vw for each grid cell is calculated as 8c‾vw=∑iA⋅di⋅ciVtot over the depth range depending on the total volume Vtot 9Vtot=∑iAi⋅di with area Ai and thickness di of layer with index i, and concentration ci corresponding to that layer. An overall spatial mean and the 0.25, 0.5, and 0.75 quantile concentrations are computed using these monthly volume-weighted mean data. A standard deviation and coefficient of variation are calculated from the volume-weighted mean over the monthly time steps. Differing from this depth range, volume-weighted mean concentrations of simulated TEP are calculated over the same depth range as stated for observational datasets when comparing to these. The standard deviation is calculated over the grid points inside the area of concern and over the specific month corresponding to the observation data used. Volume-weighted mean TEP concentrations of the time frame 1990 to 2019 are provided as binned product according to Longhurst provinces in Table  in Appendix A to enable a comparison to upcoming studies.

To evaluate the performance of the simulation, the model is compared to various in situ or remote-sensing datasets, which are obtained directly from the cited literature and repositories. The Copernicus Marine Environment Monitoring Service level 4 monthly reprocessed Arctic Ocean Color product (Product ID: OCEANCOLOUR_ARC_BGC_L4_MY_009_124) – CMEMS, in the following – provides estimates for TChla in the upper ocean with a resolution of 1 km and is available at the Copernicus Marine Services website (https://marine.copernicus.eu, last access: 1 February 2023).). The CMEMS TChla error is provided by Copernicus and is defined as the standard deviation derived from daily TChla datapoints of the corresponding month. Matching the last two decades of the simulation, monthly CMEMS data of 2000 to 2019 is downloaded and processed by averaging to a summer (May to September) mean concentration for TChla and its standard deviation. Both CMEMS and FESOM2.1–REcoM3 data are interpolated to a regular grid of 0.2 degrees for comparison. Likewise, the Arctic Ocean TChla product based on the algorithm AOreg.emp from is retrieved as a supplementary comparison for TChla, composed of the years 2003 to 2019, of which the Modis/Aqua TChla only is processed.

Since satellite TChla products have quite large uncertainties of 25 % to 150 % and standard deviations (Fig. d on CMEMS and Fig. d in Appendix A on ) in the Arctic Ocean, we also compare the FESOM2.1–REcoM3 simulation to in situ observations. A valuable dataset for the model evaluation is the long-term compilation of in situ observation data of TChla in the Fram Strait and adjacent Arctic seas covering mainly the summer months May to September 1991 to 2015, presented in . Following this study, the results of the simulation are split into box plots for different Arctic regions, summarizing the summer months May to September of 2000 to 2019 (to match the time frame of remote-sensing comparison consistently) and evaluated for the upper 100 m depth-integrated data.

The simulated PCHO represents the acidic dissolved combined carbohydrates, which are reported to form a major part of dissolved carbohydrates in the ocean , and are complemented by a small fraction of dissolved free carbohydrates (various monosaccharides). Since explicit in situ measurements of PCHO are not available, simulated PCHO concentrations were compared to in situ observations of dissolved combined carbohydrates (DCCHO). DCCHO are typically measured as individual monosaccharides released from oligo- and polysaccharides, including PCHO, after acid hydrolysis. A small DCCHO mass contribution originates from monosaccharides with acidic moieties, such as carboxyl groups in uronic acids e.g.,. However, PCHO may also contain acidity from sulfate and phosphate side groups, which are not detected using the DCCHO chromatographic protocol after acid hydrolysis with hydrochloric acid . The exact combination of monosaccharide units within a combined carbohydrate often remains unclear . Therefore, it can be assumed that a significantly higher proportion of DCCHO could be attributed to PCHO than is directly apparent from the measurements. As such, a comparison of PCHO and DCCHO appears valid. PCHO and DCCHO concentrations are expected to be within the same order of magnitude.

The dataset of in situ DCCHO presented here is published as and is available at 10.1594/PANGAEA.961004. It is derived from field bulk water samples collected at one meter depth during the PASCAL campaign on RV Polarstern in 2017, following the protocol of using high-performance anion exchange chromatography coupled with pulsed amperometric detection (HPAEC-PAD). For the validation, the in situ DCCHO concentrations are differentiated into three groups: (a) ice-free ocean, (b) marginal ice zone (MIZ), and (c) leads/polynyas within the ice pack. For the model validation, only the mean and standard deviations of DCCHO concentrations from each of the three groups are considered.

TEP can be determined in water samples observationally by microscopy of the abundances of particles [# L⁻¹], by the determination of particle surface area [cm² L⁻¹] , or colorimetrically by staining with Alcian Blue solution resulting in Xanthan gum equivalents [µg Xeq L⁻¹] . As these methods are not fully comparable to one another – even considered unreliable – colorimetrically determined in situ TEP is prioritized in this study, as variations in TEP measurement methods can lead to inconsistent comparability . Data on in situ TEP concentration are converted to [µg C L⁻¹], as explained below, as REcoM3 biogeochemical processes are described in units of carbon. Namely, the datasets of , , , , and are used. Furthermore, the TEP concentration given in is converted from Xanthan gum equivalents [µg Xeq L⁻¹] to carbon [µg C L⁻¹] by a factor of fconv=0.05 mol C (g Xeq)⁻¹ ⋅ 12.01 g mol⁻¹ following the review of .

The Arctic setup of FESOM2.1–REcoM3 simulates the phytoplankton seasonal cycle, where phytoplankton depends on nutrient and light availability and is controlled by zooplankton grazing, aggregation, and degradation losses. This is later discussed in comparison with remote-sensing products and in situ observations of the Arctic Ocean.

The phytoplankton biomass in the surface Arctic Ocean in terms of TChla is depicted in Fig. , where the FESOM2.1–REcoM3 simulation is compared to the CMEMS Arctic satellite re-analysis product and a control run without TEP. The data is averaged over the years 2000 to 2019 (as a mean model state) for the months May to September when satellite data are available. The CMEMS TChla product shows no coverage of the central Arctic Ocean due to the satellite sensors' configuration not enabling observations at these high latitudes (Fig. a). In the Barents, Kara, and Laptev Seas, TChla ranges from 0.5 to 5 mg m⁻³, with peaks of up to 15 mg m⁻³ close to the coastline. In the East Siberian, Chukchi, and Beaufort Seas, the concentration of TChla is lower, ranging from 0.1 to 1 mg m⁻³, also with a very high concentration close to the coast. The standard deviation of TChla from the CMEMS TChla product is mostly lower than 0.5 mg m⁻³. However, it increases to more than 2 mg m⁻³ along the coastline (Fig. d). In the control simulation of FESOM2.1–REcoM3 without TEP, TChla concentration range from below 0.2 mg m⁻³ in the central Arctic Ocean to approximately 2 mg m⁻³ in the Fram Strait and Chukchi Sea, reaching even values of 3 mgm-3 close to the coast of the Russian shelves. The FESOM2.1–REcoM3 including TEP simulates highest TChla concentrations of up to 6 mg m⁻³ along the Russian coast and in other shelf seas between 0.5 and 3 mg m⁻³ (Fig. c). In the Fram Strait, TChla range from 1 to 3 mg m⁻³ with highest concentration in its central part, while in the central basins of the Arctic Ocean, TChla is generally low, with concentrations of up to 0.8 mg m⁻³. Compared to the control run without the TEP implementation, TChla concentrations are slightly elevated by 0.2 to 1 mg m⁻³ in the Fram Strait, in the Beaufort Gyre, and in the Chukchi Sea (Fig. f). In the coastal areas of the Russian shelves, the simulated TChla is predominantly lower compared to the remote sensing product.

Compiling the TChla concentrations as in , they range from close to zero to 69.2 mgm-2 regarding first to third quartile, with the highest median of 40.4 mgm-2 in the eastern Fram Strait (Fig. ). In the western Fram Strait and in the Siberian seas, TChla median concentration ranges from 24.5 to 32.9 mgm-2. In the Eurasian and Amerasian Basins, median concentration is close to 0 mgm-2 with first to third quartile spanning the range 0.002–14.2 mgm-2. FESOM2.1–REcoM3 produces some outliers with up to 140 mgm-2 in the eastern Fram Strait. Complementing the regional summary, monthly climatological maps of the simulated TChla concentration and of CMEMS are also included as Figs.  and .

Broadening the perspective from phytoplankton biomass to POC, the sum of the two phytoplankton, detritus and TEP (in the model run) pools is shown in Fig. . The simulated concentration reaches up to 400 µg C L⁻¹ in the eastern Fram Strait and along the coastline of the shelves averaged over May to September of the upper 30 m (Fig. b). In the central Arctic Ocean, POC is mostly simulated with concentrations below 100 µg C L⁻¹. Thus, the model run results in POC concentrations of 100–200 µg C L⁻¹ higher compared to the control, except for the central Arctic Ocean, where no large differences are found (Fig. c).

POC is assessed in regional box plots similar to in Fig. b, where POC represents the sum of the particulate carbon pools in the model, i.e., phytoplankton carbon, detrital carbon, and carbon contained in TEP. Computed POC integrals for the upper 100 m ocean depth range from 0.2 to 18.6 g C m⁻² (first to third quartile), with highest integrals in the Barents Sea between 6.0 and 15.1 g C m⁻², and in the eastern Fram Strait with 4.7 to 18.6 g C m⁻². As for TChla, intermediate levels of integrated POC (0.2–14.3 g C m⁻²) are simulated for the western Fram Strait and on the shelves. In the central Arctic basins, the amount of integrated POC is low with 0.3 to 2.7 g C m⁻².

In this section, the production of PCHO is examined in relation to the phytoplankton bloom development. First, we analyze the regional long-term FESOM2.1–REcoM3 PCHO data as we did above for TChla and POC. Subsequently, the focus is directed particularly on the Fram Strait to assess the time evolution of phytoplankton blooming in comparison to PCHO production in a spatial context (Fig. ). Finally, we present a seasonal cycle of phytoplankton TChla and carbon to show the organic carbon flux from photosynthesis towards particle formation in Fig. .

Regarding a regional overview in Fig. , PCHO concentration as volume-weighted mean over the upper 100 m is highest in the Barents Sea with a median of 13.15 µg C L⁻¹, while on the shelf seas, values of 1.8 to 7.0 µg C L⁻¹ are computed (Fig. c). The eastern Fram Strait PCHO median concentration of 9.5 µg C L⁻¹ is higher than 5.3 µg C L⁻¹ in the western Fram Strait. In contrast, much lower concentrations are obtained for the Eurasian and Amerasian Basin, where the median ranges from 1.3 to 1.4 µg C L⁻¹. Climatological maps for 0–30 m volume-weighted mean concentration are presented in Fig.  to complement the regional overview.

Comparing the dissolved part of organic carbon in FESOM2.1–REcoM3 between control run (without PCHO) and model run (with PCHO), both simulations result in highest DOC concentrations in the eastern Fram Strait and Barents Sea, with peaks along the coastline of the Russian shelves (Fig. d and e). The DOC concentration in the control run spans approximately 100–150 µg C L⁻¹, where the DOC concentration in the model run ranges from 75 to 200 µg C L⁻¹.

To analyze the link between phytoplankton bloom evolution and PCHO production, the time period of May to July 2017 is chosen corresponding to the time of the PASCAL campaign and evaluated for sea ice concentration (SIC), DIN, TChla, and PCHO of the model's surface layer (0–5 m) in Fig. . In the northern Atlantic and Arctic Ocean, phytoplankton growth was mostly limited by light. Early in the season, sea ice covered most parts of the Arctic Ocean and Fram Strait. In contrast, the eastern Fram Strait was ice-free in May 2017 (Fig. a). A phytoplankton bloom formed in the northern Atlantic and in the MIZ in the Fram Strait, most prominently along the ice edge (Fig. g). This bloom shifted further north into the whole eastern Fram Strait in June, using up DIN. In the MIZ, a high TChla concentration of up to 6 mg m⁻³ is simulated. In July, SIC decreased further and TChla concentrations of 2–7 mg m⁻³ were found in the whole Fram Strait, with highest concentrations in the western part, and additionally in the northern Barents Sea of 2–5 mg m⁻³ (Fig. c and i, respectively). Simulated PCHO follows elevated TChla first northward and then westward in the Fram Strait area and reaches a surface concentration of up to 120 µg C L⁻¹ (Fig. j–l), particularly when DIN is depleted (Fig. d–f). In May, PCHO concentrations are most elevated close to the ice edge. In June, areas of elevated PCHO concentration are found towards the eastern Fram Strait and Barents Sea. In July, PCHO concentrations peak in the MIZ of the western Fram Strait and the northern Barents Sea alongside the phytoplankton blooms with 90 to 120 µg C L⁻¹.

The PASCAL campaign data of the same months in 2017 provides in situ measured concentrations of DCCHO in the open ocean, MIZ, and ice-covered regions around Svalbard. In situ concentrations were found to be lowest in the sea-ice leads within ice-covered regions (11.3 ± 10.1 µg C L⁻¹) and highest in the MIZ of 51.1 ± 37.0 µg C L⁻¹. In the ice-free ocean, the mean measured DCCHO concentration was 17.2 ± 5.2 µg C L⁻¹.

To assess the mean seasonal cycle over the years 2000 to 2019, phytoplankton carbon, TChla, PCHO, and TEP (the latter presented in detail in Sect. ) concentrations are compiled as a volume-weighted mean of the upper 30 m of the ocean for the eastern Fram Strait (Fig. ). The seasonal patterns of TChla and PCHO are evident in the climatology and similar to the results for 2017 shown in Fig. . The phytoplankton carbon concentration increases to peak at approximately 230 µg C L⁻¹ in July in line with the peak in TChla, with highest concentrations of 1.4 mg TChla L⁻¹. DIN concentration decreases from a winter average of about 3.5 mmol m⁻³ with the start of the phytoplankton bloom in May to 0.5 mmol m⁻³ in August. Following the bloom closely in time, PCHO is exuded by phytoplankton and its concentration rises to a maximum of approximately 50 µg C L⁻¹ in June. In the following months, PCHO concentration declines whereas TEP concentration rises quickly from May to July to 200 µg C L⁻¹, transferring the organic carbon from PCHO to TEP, and lagging approximately one month behind the peak of the phytoplankton bloom.

In this section, we first give an overview of the TEP simulation in the Arctic Ocean before describing the seasonal cycle with a focus on the Fram Strait, then broadening the perspective to an Arctic-wide climatology.

As an overview, the TEP volume-weighted median concentrations of the upper 100 m of the water column are compiled as for PCHO in Fig. d. Regional variations of TEP are high. The highest median of 44.9 µg C L⁻¹ is reached in the Barents Sea. In the eastern Fram Strait, the median lies at 31.8 µg C L⁻¹, whereas in the western Fram Strait, Kara and Laptev Sea, median concentrations of 15.3–19.8 µg C L⁻¹ are simulated. In the East Siberian and Chukchi Sea, as well as in the Eurasian and Amerasian Basin, median TEP concentration are below 3 µg C L⁻¹. Maximum simulated TEP concentrations reach nearly 190 µg C L⁻¹ in the eastern Fram Strait.

The TEP concentration exhibits a phase-shifted seasonal cycle following the phytoplankton bloom in the eastern Fram Strait area (Fig. ). It is close to zero during winter months and starts to rise in early May, lagging approximately one month behind the TChla and phytoplankton carbon concentrations. It peaks at approximately 200 µg C L⁻¹ in July and quickly declines thereafter.

For the Arctic-wide context, monthly climatological maps of TEP are provided for April to November as a volume-weighted mean of 0–30 m (Fig. ). In May, TEP concentrations rise first in regions with phytoplankton blooms (cf. monthly climatology of TChla and PCHO in Fig. , i.e., in the Fram Strait and on the continental shelves, 50–150 µg C L⁻¹). The climatology further shows that from June to August, TEP increase in the Fram Strait, on the Siberian shelves, in the Chukchi and Beaufort Seas, reaching approx. 200–400 µg C L⁻¹. The mean concentration in the central Arctic Ocean reaches 10–50 µg C L⁻¹ in the upper 30 m of the water column. There is a gradient of TEP from the shelves towards the Arctic Ocean basins during summer in the simulation. Beginning in September, TEP declines in the whole Arctic region. TEP is then remineralized or sinks to deeper ocean layers, and surface concentrations decline.

As detailed in Sect. , the FESOM2.1–REcoM3 setup used in this study is optimized in its parameterization and irregular grid to enable reliable Arctic eddy-permitting simulations of PCHO and TEP. However, since FESOM2.1–REcoM3 is a global model in principle, we additionally evaluate the TEP concentration relative to its global pattern. Because in situ observations are available mostly for surface waters (reviews by ), the model results are presented here as volume-weighted mean of the upper 30 m of the ocean averaged over 2000 to 2019. For completeness, global model results of TChla and PCHO can be found as Fig. . Furthermore, mean TEP concentrations are also summarized by binning for each Longhurst Provinces , presented as Table .

In large areas of the global ocean, simulated TEP concentration in FESOM2.1–REcoM3 ranges from 5 to 50 µg C L⁻¹ (Fig. a). TEP concentrations are especially high in the Nordic Seas and the Arctic Ocean with approx. 75 to 150 µg C L⁻¹. In the region of the Antarctic Polar Front and in the Peruvian upwelling, TEP concentrations reach a minimum below 5 µg C L⁻¹. The coefficient of variation of monthly, volume-weighted mean TEP concentration is highest in the Arctic region north of 75° N with up to 800 %, while between 30° S to 30° N, the coefficient of variation is lowest with less than 30 % (Fig. ). In the Southern Ocean, the coefficient of variation ranges from 100 % to 200 %.

In terms of Arctic-wide TChla, the FESOM2.1–REcoM3 output is evaluated with the remote-sensing product provided by CMEMS and with various in situ datasets. Generally, the model run aligns with the compiled remote-sensing data of CMEMS TChla (Fig. ). The control run without the TEP implementation underestimates the TChla concentration in large parts of the Barents Sea and on the Arctic shelves, which was reported in as well. Overall, the model run including TEP performs in better agreement with CMEMS. Still in the Chukchi Sea and the Beaufort Gyre, as well as in the Fram Strait, the model run overestimates TChla compared to both the control run and CMEMS. Climatological monthly maps of TChla are included as Fig. .

As such, this pattern of the FESOM2.1–REcoM3 model run compares well to the results from and also to other observations. The generally higher TChla concentrations in the model run compared to the control might reflect the higher turnover of phytoplankton biomass: TEP increases the aggregation of phytoplankton to detritus, which partly gets remineralized, replenishes nutrients to the seawater, and enables new build-up of phytoplankton biomass. In accordance with this hypothesis, we observe higher nutrient concentrations in the central Arctic Ocean, in the Beaufort Sea, and in the Chukchi Sea compared to the control run (Fig. ). However, we have not yet investigated the influence of TEP on the nutrient dynamics in the Arctic Ocean any further.

Evaluating the TChla patterns in more regional detail, we start with the eastern Fram Strait. Both CMEMS and the model run present the beginning of the phytoplankton bloom in May which prevails with TChla of approximately 1–3 mg m⁻³ through the summer months (Figs. a and ). In the eastern Fram Strait, in situ measurements of result in a median vertically integrated concentration of 44 mgm-2 (0–100 m depth), and in the Barents Sea of 42 mgm-2 which agree with simulated median concentrations of 40.4 and 33.8 mgm-2, respectively (Fig. a). The results are also in line with the two-year round mooring observations at the long-term ecological research observatory HAUSGARTEN in the eastern Fram Strait with simulated TChla concentration reaching up to 5 mgm-2 in the upper 30 m compared to 7 mgm-2 in measurements . In the MIZ, especially in the area of Fram Strait, phytoplankton growth is expected to be highest, as the sea-ice breaks up, light availability is increased and the water column is stratified . Likewise in the western part of Fram Strait and in the Siberian seas, the lower amount of simulated TChla matches in situ data for these regions spanning 13 to 26 mgm-2 .

In the Chukchi and Beaufort Seas, the simulated TChla concentration itself (and the difference between FESOM2.1–REcoM3 and CMEMS) is generally low (less than 1.5 mg m⁻³, Fig. c and f). However, the timing of the phytoplankton bloom differs from May in CMEMS to June–July in the model run (Figs. a and ). Other observations from in situ data and satellite-derived products draw a diverse picture in these seas with massive under-ice blooms in the Chukchi Sea with high TChla concentration of up to 30 mg m⁻³ , and low (0.02–0.25 mg m⁻³) TChla concentration in the northern Chukchi and Beaufort Seas . In the central Arctic Ocean, vertically integrated TChla is low in both simulation (integrated over 0–100 m water depth, median 0.01 mgm-2) and measurements (7–8 mgm-2). No coverage by the remote-sensing product by CMEMS is available.

As such, the higher TChla concentration of the FESOM2.1–REcoM3 model run might be a better fit to observation data than CMEMS. One explanation for the observed difference may be that CMEMS sampled far less days than are simulated by FESOM2.1–REcoM3. explain lower TChla in the simulation by full spatial and temporal coverage whereas only open-water productive regions are accessible from remote-sensing measurements. However, suggest that remote-sensing products don't capture the phytoplankton bloom forming under thin ice or melt ponds, therefore often result in too low TChla estimates. A likely explanation for the higher simulated median TChla concentration in several regions and higher variability in comparison to might be the consideration of up to 62k data points in the regional subsets (Fig. a) compared to only a few hundred in , where the in situ sampling is mostly limited to one campaign each spring-summer season. Additionally, the evaluated regions of the model run contain the whole continental shelf grid points, whereas the in situ measurements are located mostly in the northern parts of the shelf seas .

However, in the coastal areas of the shelf areas (the Barents, Kara, and Laptev Seas and close to the Canadian coast) the FESOM2.1–REcoM3 model run highly underestimates TChla concentrations provided by CMEMS (Fig. c). There, CMEMS is error-prone (high standard deviation, Fig. d), most likely due to the very high colored dissolved organic matter and total suspended matter concentrations, which have not been sufficiently accounted for in the retrieval process. This results in a significant overestimation of TChla . Also other remote-sensing products overestimate TChla on the Arctic shelves .

Comparing FESOM2.1–REcoM3 to the Arctic TChla product presented in , the agreement is similar as with the CMEMS product, but the spatial patterns are very different (Fig. ). This satellite product shows even lower TChla concentration in large parts of the Arctic but agrees better with the simulation in the southern Barents Sea and Laptev Sea shelf break where TChla is still higher with reaching maximum values of 7 mg m⁻³. derive TChla estimates by applying different retrieval parameters in different regions, as developed by . This leads to a good fit to in situ measurements but produces abrupt changes in TChla estimates across region boundaries. The choice of methods (CMEMS compared to ) profoundly impacts the remote-sensing products. As such, priority is given to the CMEMS Arctic TChla product as it is more consistent across the whole Arctic Ocean, despite overestimation on the inner shelves .

The implementation of the additional processes for carbon exudation and aggregation impacts the amount and distribution of POC (Fig. ). As such, the general agreement of simulated POC concentrations to observations confirms the improvement made in the model run. Similar to TChla, simulated POC is assessed in regional overviews (Fig. b). These are generally in agreement with the observations of , who also measured the highest vertically integrated median POC concentration of 13 g C m⁻² in the eastern Fram Strait (integrated over 0–100 m water depth). Also, found approx. 16 g C m⁻² POC in the eastern Fram Strait over 2009 to 2017, comparable to a median POC concentration of 9.5 g C m⁻² in FESOM2.1–REcoM3. Regarding the western Fram Strait and the Siberian shelves, the simulation yields higher vertically integrated median concentrations (6.2–10.2 g C m⁻²) at spatio-temporal measurement locations of , i.e., 5–8 g C m⁻². The higher simulated POC on the shelves might be caused by including also southern shelf regions with higher POC concentration into the median calculation, whereas the measurements do not cover these areas. With respect to the central Arctic, estimated the interquartile range of vertically integrated POC concentration to 1 to 8 g C m⁻² from their long-term observations in the Eurasian Basin, while provide data on POC of approx. 3 to 5 g C m⁻² in the Eurasian Basin in 2012, which are on the same order of magnitude as the simulated vertically integrated POC concentration of 0.4 to 2.7 g C m⁻² interquartile range in the present analysis. Additionally, report 240 µg C L⁻¹ POC in upper 30 m of the eastern Fram Strait in summer, which are comparable to the FESOM2.1–REcoM3 simulation of 200–350 µg C L⁻¹ POC in the same area (Fig. b). Further to the north of the Fram Strait, measured subsurface POC concentrations between 72 and 204 µg C L⁻¹, where the simulated POC concentration spans 50–100 µg C L⁻¹. The control run without the TEP implementation underestimates the POC concentration in all cases.

Based on a reasonable agreement of the simulation in terms of TChla and POC laid out, the focus is now directed towards the evaluation of the PCHO implementation into FESOM2.1–REcoM3. There is only a limited amount of observational data available to discuss the simulation of PCHO in the Arctic. As such, the model results in the Fram Strait are singled out along with the DCCHO dataset of the PASCAL campaign in 2017, before the perspective is broadened to the few studies in the other Arctic regions. Further, the impact of the model implementation on the DOC pool is discussed in comparison to the control run and additional observation data.

Focusing on the Fram Strait area in the single year 2017, there is a reasonable agreement with the PASCAL campaign data of the same summer months, assuming that the simulated PCHO reflects most of the DCCHO pool (Methods Sect. ). The MIZ exhibits the highest concentrations in both measurement and simulation. In general, in FESOM2.1–REcoM3, the regions of elevated PCHO follow the phytoplankton bloom patterns. Interestingly, there is a case in the simulation, where high TChla does not result in high PCHO concentration: in June in the south-western Fram Strait (Fig. k). The carbon overflow hypothesis probably offers an explanation : DIN has not been used up completely in this area; hence, the phytoplankton are not experiencing nutrient depletion. The implemented limiter function for phytoplankton carbon exudation (depending on the intracellular N : C ratio, Eq. ) is apparently still low, and as such, the phytoplankton cells are not exuding much excess carbon.

In the eastern Fram Strait, observe DCCHO in concentrations of 55.8 µg C L⁻¹ in July and 22.5 µg C L⁻¹ in September (0–100 m), which are slightly higher than the FESOM2.1–REcoM3 summer PCHO median (Fig. c). Comparable to the simulation, the DCCHO concentration is significantly correlated to TChla in their study. Regarding other regions in the Arctic, the model results are in the same order of magnitude as observations by in the Eurasian Basin, the northern Barents Sea, and northern Laptev Sea. measured 25.5–30.8 µg C L⁻¹ DCCHO in the upper 40 m in August to October 2012 compared to 10–80 µg C L⁻¹ of PCHO in this analysis (Fig. ). In the Eurasian Basin, report 88–502 µg C L⁻¹ DCCHO in August for the upper ocean (0.5 m) during a drift campaign in a re-freezing lead. These observations show higher concentrations than results presented here, but it is difficult to compare the FESOM2.1–REcoM3 simulation with bulk measurements for the upper five meter of the water column. Additionally, the authors attribute the high DCCHO to the exudation of DCCHO due to the freezing conditions and not to nutrient stress.

At greater depths, there is a steep decline in simulated PCHO concentration. also report a decreasing DCCHO concentration with depth for the Eurasian Basin similar to, though less pronounced than in, the simulation. The volume-weighted median of 0–100 m depth of PCHO in FESOM2.1–REcoM3 drops to 1.3 µg C L⁻¹ (Fig. c), while approx. 19–23 µg C L⁻¹ DCCHO in the depth range 40 to 150 m are reported by . This discrepancy might be caused by the specific timing of the campaign during a year with an extraordinarily low sea-ice concentration, which led to high TChla and a strong nutrient depletion, to which relate the high amounts of exuded DCCHO. Overall, the simulated PCHO concentration clearly underlines the link of high PCHO to high TChla concentrations coinciding with nutrient depletion in the open ocean and in the MIZ (e.g., July 2017 in western Fram Strait, Fig. ). Furthermore, the simulation results show that PCHO largely does not accumulate in the upper ocean, but quickly aggregates to TEP (Fig. ).

PCHO is considered a dissolved pool of organic carbon in FESOM2.1–REcoM3, which is generally part of the DOC pool when it comes to in situ methodology. However, we introduce a distinction between the DOC and the PCHO pool in the model setup, as the exuded phytoplankton organic carbon is split between DOC and PCHO. Furthermore, FESOM2.1–REcoM3 simulates only the (semi-)labile (fresh) DOC pool, which is considered as consisting of labile organic compounds and not refractory (old) ones. As such, a direct comparison of simulated DOC to observations results per se in the underestimation of DOC in the model. As such, in the Fram Strait, DOC concentrations of 720–2040 µg C L⁻¹ have been reported for the upper 200 m of the water column . estimate the fraction of semi-labile DOC to 3.5 %–8.4 ± 4.7 % in their campaign, resulting in a concentration of 30–74 µg C L⁻¹ semi-labile DOC. In the East Siberian Sea, DOC concentration reached 648 ± 96 µg C L⁻¹ during a bloom phytoplankton bloom . The FESOM2.1–REcoM3 model run yields higher DOC concentrations (sum of DOC and PCHO) compared to the control run (DOC, Fig. f), with values still falling at the lower end of the in situ measurements cited above. Thus, we assume that the implementation of PCHO and TEP might result in a better fit to in situ measurements.

At the beginning of the year, TEP concentrations rise with the beginning of summer and decrease quickly in autumn (Fig. ). The highest concentrations are simulated along the coastline and with a decrease toward the central Arctic Ocean. highlight this TEP concentration gradient in their study on the Chukchi shelf and Canadian basin (shelf 138.9 ± 64.7 µg C L⁻¹ to slope/basin: 79.3 ± 15.5 µg C L⁻¹, 0–50 m) and attribute it to TChla co-occurrence, transport from the shelf towards the basin, and additional TEP production by prokaryotes. Further, the TEP seasonality is in accordance with TEP observations of revealing a strong decline from summer to autumn (observation of 21.4 ± 14.5 µg C L⁻¹ in July, 7.1 ± 5.2 µg C L⁻¹ in September, 0–100 m, compared to simulated 76.10 ± 50.68 µg C L⁻¹ in July, 27.81 ± 14.51 µg C L⁻¹ in September), which the authors link to remineralization and sinking as larger particles out of the water column. In the compiled climatology for the eastern Fram Strait, TEP concentration lags approximately one month behind the TChla increase and decrease.

From 2017 data, it is concluded that TChla alone is not a good predictor for TEP concentration. The release of PCHO as TEP-predecessor depends also on the nutrient stress of phytoplankton (Fig. ), and as such, the model setup is in agreement with the carbon overflow hypothesis . Furthermore, reported a strong link between TEP and Phaeocystis spp. occurrence patterns in the Fram Strait region. With respect to this phytoplankton species, also found TEP concentration positively correlated to phytoplankton biomass along a wide transect in the Atlantic Ocean. The setup of FESOM2.1–REcoM3 includes only diatoms and small phytoplankton as distinct groups with – so far – similar implementation of carbon exudation. Here, the recent developments of REcoM3 towards including additional phytoplankton functional groups could help to distinguish contributions of different phytoplankton functional groups to carbon exudation. Regarding the Arctic-wide occurrence of TEP, the available observations are very limited in space and time. Each measurement depends on a very local set of ecological conditions shaped by sea ice concentration, nutrient availability, and phytoplankton blooms. These in situ observations are compared to monthly integrated simulation results for a small area enclosing the corresponding observation location in Table , complementary to the maps of simulated TEP concentration in Fig. . Overall, there is a fairly good agreement of the simulated TEP concentration to in situ observations (correlation coefficient of 0.71; p=0.11, see also Fig. ) when the simulation also successfully models the phytoplankton blooms. However, the modeled TEP concentrations being often slightly higher and differing particularly in two cases which have been excluded from the correlation: Firstly, the Catlin Ice Base measurement in the Canadian Arctic Archipelago, where obtained their measurements from an under-ice phytoplankton bloom, which resulted in a very high production of TEP. Such an event is not reproducible by FESOM2.1–REcoM3 because ice-algae are not explicitly modeled, and light-through-ice transmission is not adequately represented. Secondly and similarly in the Central Arctic Ocean, conducted a drift experiment measuring TEP concentration in the pack ice. The authors found a uniform depth distribution of TEP along the drift track, which could be supplied by the continuous primary production in the upper water column in open leads and melt ponds. However, there is nearly no TEP simulated in this area by FESOM2.1–REcoM3. The absence could originate from the missing under-ice or ice-algae production in the model, which have been part of other modeling approaches e.g.,.

Overall, it has been shown that FESOM2.1–REcoM3 is capable of simulating the TEP distributions in the Arctic Ocean well. As this is a first step to simulate TEP in the Arctic Ocean, the full complexity of TEP cycling is not yet represented in the model parameterization. provide a conceptual model of TEP dynamics, of which only the exudation of organic carbon, its aggregation and TEP formation, and the remineralization are implemented in this simulation. Other processes of the life cycle of TEP, such as TEP ballasting by other particles, and sinking in the water column, or microbial/zooplankton grazing on TEP, have not yet been implemented . Additionally, describe high concentrations of TEP and its precursors in melt ponds and in leads in the central Arctic Ocean and argue that TEP could also be released from melting sea ice. This would require a more sophisticated implementation of the sea-ice compartment in FESOM2.1–REcoM3. have already coupled FESOM2 with the model Icepack, providing more detailed sea-ice physics but also increased model complexity and cost. Another categorical light-through-ice transmission approach was suggested by , both could offer valuable perspectives for improved representation of under-ice algae blooms and therefore, also of PCHO and TEP.

The peak concentration of TEP in FESOM2.1–REcoM3, especially regarding the ocean surface, might be too high. One reason might be that in the current implementation, TEP aids in the aggregation of other particles (the TEP dependency of the aggregation rate, Eq. ), but is not itself transferred in the process. Another reason could be the missing transfer of particles into the surface microlayer and an enrichment there. The surface microlayer is the very thin transition film at the ocean–atmosphere interface, which is not part of the model structure. TEP was reported to be positively buoyant when not ballasted by other minerals or particles, thus, TEP would ascend through the water column and become enriched in this layer . In a methodological study, demonstrate the enrichment of TEP via bubble scavenging, a process by which rising air bubbles transport organic matter. Observational studies report contrasting results on surface microlayer enrichment: report only a minor enrichment in the surface microlayer in the open water of the central Arctic, whereas calculated an enrichment factor of 1.7 ± 0.5 in the Canadian Arctic. Two obstacles make it difficult to simulate the surface microlayer. Firstly, incorporating such a thin layer in a global ocean model is challenging due to numerical constraints. In FESOM2.1–REcoM3, the upper ocean is only resolved as layers spanning five meter depth. Secondly, the complexity of processes in the surface microlayer is difficult to capture. It would be necessary to consider the physical interactions at the air–sea interface and a different balance of the biogeochemical processes (microorganisms, gel phases, aerosol formation; ) compared to the underlying bulk water.

Regarding the carbon export to the deep ocean in FESOM2.1–REcoM3, there are two detritus classes sinking with either increasing speed with depth (following ) or constant, fast sinking mimicking zooplankton fecal pellets . Regarding TEP, no explicit sinking of TEP themselves has been implemented in the model. Still, the FESOM2.1–REcoM3 simulation contains a substantial decrease of TEP concentration with depth alongside the phytoplankton biomass decrease with depth (Fig. ). In the Arctic Ocean, the TEP concentration rapidly decrease to zero in the upper 40 m, whereas the averaged TEP concentration in the Northern and Southern Hemisphere decrease more slowly to zero over the upper 100 m. Generally, TEP themselves are less dense than seawater and often remain suspended near the surface or even become buoyant. Therefore, the direct contribution of TEP to net carbon export is limited and is not taken into account in our model run. However, TEP interact with denser particles forming TEP-rich aggregates, which can significantly contribute to the vertical transport of POC out of surface waters. Our model describes the primary mechanism driving this aggregation by increasing the particle stickiness depending on TEP concentration (Eq. ). Hence, the role of TEP in driving sinking is included in our model run.

There are several aspects to consider for the addition of a TEP sinking parametrization: Firstly, report that there is often no correlation between particle size and sinking speed. Similarly, have recently discovered a matrix of organic carbon forming around TEP, which slows down the sinking considerably in the water column. Secondly, there is a degradation of particular and dissolved organic matter which is regarded as ecosystem-dependent, especially concerning the specific microbial community engaged in turnover processes . Additionally, the explicit representation of the heterotrophic microbial community in a global ocean model suggests that environmental conditions impact the degradation of dissolved organic matter, predominantly by nutrient limitation of heterotrophs in the upper ocean and carbon limitation in the deep ocean . These aspects represent ongoing research on the link between the upper ocean and the export of organic carbon to greater depths, also called the biological carbon pump. The cited studies provide insight into the diverse and complex biogeochemical processes that warrant consideration, underscore the prevailing analytical uncertainties, and point out the need for reassessment of particle sinking speed and organic matter degradation. As a result, there is a need to revisit the carbon sequestration estimates for the global ocean . Therefore, the development of FESOM2.1–REcoM3 to explicitly resolve the sinking of TEP through the water column and to provide a detailed account of the degradation processes could lead to a more realistic representation of TEP in the Arctic and the global ocean. To date, most Earth System Models have not fully incorporated these processes, resulting in significant uncertainty regarding estimates of the biological carbon pump in response to climate change . It is anticipated that further advances in the observational understanding of particle sinking and the biological carbon pump, along with their incorporation into model parameterization, may lead to an updated quantification of carbon export.

The observational caveats of this analysis include the scarcity of sampling campaigns, the broad spatial and temporal gaps inbetween studies, and their differences in sampling depths and methodologies (listed in Table ). These make it difficult to construct consistent time series for biogeochemical parameters from observational studies, which therefore limit the model evaluations. At least for the Fram Strait, there is a valuable in situ time series on TChla and POC , which is complemented by DCCHO and TEP measurements in the recent years . Several studies provide campaign data repeatedly from areas within the pack ice in the high Arctic of marine organic carbon particles and their transfer from the ocean into the atmosphere . Ideally, the measurement campaigns at these locations – as examples for regions with varying sea-ice conditions and year-round ice coverage – should be extended by DCCHO and TEP sampling in the upper ocean, and as such, be continued to infer long-term trends. This can also help to distinguish between the ice-algae or pelagic phytoplankton release of organic carbon. Additionally, further studies should provide nutrient measurements alongside DCCHO determination in order to foster understanding of phytoplankton carbon exudation as postulated by the carbon overflow hypothesis in different Arctic environments. This could also enable a model refinement with phytoplankton group-specific PCHO exudation ratios or their time-dependencies during different phases of the bloom.

propose three different ocean regimes of TEP occurrence, to which the model results are compared: an oligotrophic regime, a polar regime, and an eastern boundary upwelling regime. The first, oligotrophic regime, is characterized by low phytoplankton productivity, which also results in low TEP concentrations. In large parts of the Atlantic, Pacific, and Indian Ocean, the simulated annual mean TEP concentration ranges between 5 and 25 µg C L⁻¹ (Fig. ). This agrees with in situ data: Data reviewed in show the same concentration range of 5 to 36 µg C L⁻¹. Furthermore, in the tropical Atlantic, the observed TEP concentration is also low at approximately 5–10 µg C L⁻¹ .

According to , within the second regime represented by the polar regions, TEP accumulates in the upper ocean layer. The authors attribute this mainly to the occurrence of the phytoplankton Phaeocystis spp. and microbial degradation focusing primarily on proteinaceous compounds rather than on TEP, while observed diatom bloom dominance at their Arctic stations. For the Arctic Ocean, the simulated strong seasonality of TEP is discussed above (Sect. ). In the Southern Ocean during the blooming period between December and March, simulated TEP concentrations increase to approximately 10–150 µg C L⁻¹. In the open waters of the Southern Ocean, observational studies report 0–39 µg C L⁻¹ with a large standard deviation of ±117 µg C L⁻¹ , whereas in the coastal waters, even higher TEP concentrations are measured of approximately 200 µg C L⁻¹ in Bransfield Strait ( reviewed in ). The exact timing of measurements in relation to the state of phytoplankton blooms seems to be critical, as the standard deviation of the measured TEP concentration can be rather large and the authors link the TEP production primarily to phytoplankton production.

As a third regime, the eastern boundary upwelling systems are described as highly productive regimes, which also results in high TEP production and fast particle sinking because of ballasting . The authors report TEP concentration of up to 85 µg C L⁻¹ in the Mauritanian, but only 9 µg C L⁻¹ in the Peruvian upwelling system, which they attributed to high grazing rates by zooplankton and heterotrophic consumption. High TEP amounts of up to 1200 µg C L⁻¹ are reported for the California Current . In the Arctic configuration of FESOM2.1–REcoM3 used here, these eastern upwelling regions are not sufficiently resolved, and as a consequence, have a primary production that is too low and thus, lack the phytoplankton carbon exudation and the resulting high TEP concentrations.

Overall, the simulated global TEP patterns are in good agreement with observational studies when the phytoplankton blooms are simulated successfully (Fig.  on TEP and Fig.  on TChla). This study is a first step of modeling TEP in a large-scale ocean biogeochemistry model.