Variable C : N : P stoichiometry of dissolved organic matter cycling in the Community Earth System Model

. Dissolved organic matter (DOM) plays an important role in the ocean’s biological carbon pump by providing an advective/mixing pathway for ∼ 20 % of export production. DOM is known to have a stoichiometry depleted in nitrogen (N) and phosphorus (P) compared to the particulate organic matter pool, a fact that is often omitted from biogeochemical ocean general circulation models. However the variable C : N : P stoichiometry of DOM becomes important when quantifying carbon export from the upper ocean and linking the nutrient cycles of N and P with that of carbon. Here we utilize recent advances in DOM observational data coverage and ofﬂine tracer-modeling techniques to objectively constrain the variable production and remineralization rates of the DOM C : N : P pools in a simple biogeochemical-ocean model of DOM cycling. The optimized DOM cycling parameters are then incorporated within the Biogeochemical Elemental Cycling (BEC) component of the Community Earth System Model (CESM) and validated against the compilation of marine DOM observations. The optimized BEC simulation including variable DOM C : N : P cycling was found to better reproduce the observed DOM spatial gradients than simulations that used the canonical Redﬁeld ratio. Global annual average export of dissolved organic C, N, and P below 100 m was found to be 2.28 Pg C yr − 1 (143 Tmol C yr − 1 ) , 16.4 Tmol N yr − 1 , and 1 Tmol P yr − 1 , re-spectively, with an average export C : N : P stoichiometry of 225 : 19 : 1 for the semilabile (degradable) DOM pool. Dissolved organic carbon (DOC) export contributed ∼ 25 % of the combined organic C export to depths greater than 100 m. in the surface layer (<10m). The prod-ucts of organic matter remineralization are dissolved inorganic carbon, nitrate, ammonium, and phosphate.


Introduction
Dissolved organic matter (DOM) is an important pool linking nutrient cycles of nitrogen (N) and phosphorus (P) to the ocean's carbon cycle. Following its net production in the surface ocean, DOM provides an advective pathway for removal of biologically fixed carbon (C) to the deep ocean, accounting for ∼ 20 % of the C exported within the ocean's biological pump (Hansell, 2013). Remineralization of DOM in the ocean's interior is carried out by microbial heterotrophs, respiring C while releasing inorganic N and P nutrients back into the water column. The concept of the Redfield ratio (Redfield, 1958;Redfield et al., 1963) has been a unifying paradigm in ocean biogeochemistry linking the stoichiometry of biological production and phytoplankton cellular material to that of the remineralization of detrital organic matter (OM) and inorganic nutrient ratios in the water column. At the global scale, production/decomposition of particulate OM (POM) in the ocean is thought to largely follow the canonical Redfield ratio of 106 : 16 : 1 for C : N : P, however some recent studies have suggested more variable C : N : P ratios (i.e., Martiny et al., 2013a, b) and only recently has variable C : N : P stoichiometry been introduced into Earth system models (e.g., Vichi et al., 2007;Dunne et al., 2013). Large deviations from the Redfield ratio have been documented for DOM (Aminot and Kérouel, 2004;Hopkinson and Vallino, 2005). Hopkinson and Vallino (2005) found DOM production and decomposition to follow a stoichiometry of 199 : 20 : 1, indicating the more efficient export of C within DOM per mole of N and P relative to sinking POM. This finding is significant in light of evidence that future perturbations to the ocean from global climate change may favor enhanced partitioning of production to DOM (Wohlers et al., 2009;Kim et al., 2011). Thus accounting for variable Published by Copernicus Publications on behalf of the European Geosciences Union. 210 R. T. Letscher et al.: Variable C : N : P stoichiometry stoichiometry within the DOM pool that deviates from the Redfield ratio requires a re-evaluation of the controls on C export and their response to future perturbations due to climate change.
Here we aim to utilize recent advances in DOM data coverage to incorporate variable production and decomposition stoichiometry within the DOM tracers of the Biogeochemical Elemental Cycling (BEC) model in order to improve representation of this important carbon export flux and associated nutrient cycles. The BEC tracks the cycling of key biogeochemical tracers (e.g., C, N, P and Fe) and runs within the ocean general circulation component of the Community Earth System Model (CESM) (Moore et al., 2004). The current release of CESM v1.2.1 contains five DOM-related tracers: semilabile DOC (dissolved organic carbon), DON (dissolved organic nitrogen), and DOP (dissolved organic phosphorous) pools as well as refractory DON and DOP pools . Here we have added a sixth DOM tracer, refractory DOC. Our approach is to optimize the BEC DOM parameters using available observations, by applying a fast offline solver based on a direct-matrix inversion (DMI) of a linear model of DOM cycling; an approach similar to previous applications for marine radiocarbon (Khatiwala et al., 2005) and marine organic matter cycling (Kwon and Primeau, 2006;Hansell et al., 2009). The 3-D ocean circulation is obtained from the offline tracer-transport model for the ocean component of the CESM (Bardin et al., 2014). The DMI solver uses a parallel multifrontal sparse matrix inversion approach as implemented in the MUMPS (MUltifrontal Massively Parallel Sparse direct Solver) (Amestoy et al., 2001(Amestoy et al., , 2006 to quickly obtain the equilibrium solutions needed to objectively calibrate the biogeochemical parameters of the DOM cycling model by minimizing the misfit between the model and observations. The DOM cycling parameters from the equilibrium solution of the offline model are then incorporated within the BEC and optimized with only minor additional tuning. The remainder of this article is organized as follows. Section 2 describes (1) the current representation of DOM cycling in the BEC v1.2.1, (2) the global ocean data set of DOM observations utilized for the optimization, (3) the structure of the offline DOM cycling model and the DMI solver, and (4) the modified BEC model with improved DOM cycling parameters with the metrics employed for optimization. Section 3 details the results of (1) the offline DOM cycling model solution, (2) the reference CESM BEC v1.2.1 simulation, as well as (3) the BEC simulation with optimized DOM cycling, including a comparison of DOM cycling metrics. Sections 3.4 and 3.5 describe a comparison of multiple DOM cycling schema and an evaluation of direct uptake of DOP by phytoplankton in the BEC model, respectively. We conclude with a discussion and summary of our results in Sect. 4. Primary production is carried out by three phytoplankton functional types: small phytoplankton (which also contains a subgroup of calcifying phytoplankton), diatoms, and diazotrophs. Sources to DOM include direct losses from phytoplankton/zooplankton and from zooplankton grazing of phytoplankton. The major sink for DOM is microbial remineralization, parameterized with an assigned lifetime which differs between the euphotic zone and the mesopelagic ocean. A small fraction of phytoplankton production is converted to refractory DOM in the upper ocean with an additional source to DOMr from degradation of sinking POM in the mesopelagic. DOMr is also lost via UV photo-oxidation in the surface layer (< 10 m). The products of organic matter remineralization are dissolved inorganic carbon, nitrate, ammonium, and phosphate.

DOM cycling in the standard BEC v1.2.1
Model simulations with the optimized DOM parameters are compared against a reference simulation using the standard version of the CESM BEC v1.2.1, which we refer to as REF.
The BEC model runs within the ocean physics component of CESM1 (Gent et al., 2011), which is the Parallel Ocean Program v2 (POP2; Smith et al., 2010). Detailed description and evaluation of the ocean general circulation model is given by Danabasoglu et al. (2011). Additional documentation, model output, and model source code are available online (www2.cesm.ucar.edu). The REF simulation has a nominal horizontal resolution of 1 • with 60 vertical levels ranging in thickness from 10 m (in the upper 150 m) with increasing layer thickness increasing with depth below 150 m. Results are presented for the final 20 yr annual average from a 310 yr simulation.
A flowchart of organic matter cycling in the BEC is shown in Fig. 1 and a list of DOM parameter values from REF are given in Table 1. Primary production is carried out amongst three phytoplankton groups, which take up available inorganic nutrients and have losses to zooplankton grazing, sinking POM, and semilabile DOM. Organic matter is Table 1. Optimized DOM parameters from the DMI-enabled linear DOM model (DMI-DOM solver) and the modified DMI model (MOD DMI-DOM solver) as well as the REF and DOM OPT simulations of the CESM BEC. Euphotic zone: 0-100 m for the DMI-DOM models and depths where PAR > 1 % for REF and DOM OPT. The "flux to DOM" represents the fraction of PP that accumulates as DOM while the "fraction of DOM flux" represents the portion of the DOM production flux that accumulates as semilabile (SL) or refractory (R) DOM. Parameters f i |i = 1· · ·2, κ i |i = 1· · ·4 are defined in Eqs. (1)-(4). Surf: surface layer (< 10 m), reminR: remineralization rate, sp: small phytoplankton, diat: diatoms, diaz: diazotrophs, k: half saturation constant for DOP uptake, and NA: not applicable. produced with a C : N : P stoichiometry set to the slightly modified Redfield ratio of Anderson and Sarmiento (1994), 117 : 16 : 1. Additional sources to semilabile DOM include grazing losses when phytoplankton are grazed by zooplankton as well as direct zooplankton losses. A variable fraction of DOM production is sent to the refractory DOM (DOMr) pool, with different fractions going to the dissolved organic N and P pools. Approximately 15 % of modeled primary production (PP) is sent to the DOM pool via these sources, with the remainder of PP cycling as POM. It is important to note that the BEC does not specifically track the total production/decomposition of DOM, which is estimated to be 30-50 % of net primary production (NPP) (Carlson, 2002; and references therein). Rather, BEC semilabile and refractory DOM tracers track the accumulated DOM pools that arise from the decoupling of DOM production and consumption in time and space and are thus subject to advection by the ocean circulation. These recalcitrant DOM fractions cycle on timescales of years to centuries and represent a smaller portion of NPP, i.e., ∼ 5-10 % (Hansell, 2013). The labile DOM pool, which cycles on timescales of minutes to days (Hansell, 2013) is not explicitly modeled and is instead rapidly converted to inorganic carbon and nutrients at each time step. Microbial remineralization is the dominant sink for both POM and DOM pools and is parameterized by assigned remineralization rates. POM is remineralized following a prescribed remineralization vs. depth curve, with a length scale that increases with depth . Semilabile DOM pools are assigned lifetimes (1 / remineralization rate) that depend on the light field with model grid cells where 212 R. T. Letscher et al.: Variable C : N : P stoichiometry photosynthetically active radiation (PAR) is > 1 % of surface irradiance being assigned a euphotic zone lifetime. Semilabile DOM in model grid cells with PAR < 1 % is assigned a mesopelagic zone lifetime. Remineralization is more rapid for semilabile DOM in the euphotic zone, with lifetimes on the order of 5 months for DON + DOP and ∼ 8 months for DOC. Longer lifetimes for semilabile DOM are assigned in the mesopelagic zone with the order of remineralization lifetimes following C > P > N. Remineralization of refractory DOM follows a similar light dependence with a faster remineralization rate given to DOMr in euphotic zone grid cells to parameterize a sink via UV oxidation (Carlson, 2002). DOMr below the euphotic zone is remineralized over centennial timescales.

Database of DOM ocean observations
We compiled publicly available and literature observations of DOM concentrations into a single database for use in both the DMI-enabled linear DOM model as well as to evaluate our BEC DOM optimization model runs. Briefly, the database contains over 34 000 observations of DOC, > 18 000 observations of DON, and > 2000 observations of DOP. Geographic coverage for the five ocean basins is moderately balanced for observations of DOC and DON; however, the Atlantic Ocean dominates available DOP observations with DOP data completely lacking for the Indian, Southern, and Arctic oceans. Semilabile DOM is defined as the total observed DOM concentration less the refractory concentration as determined from the asymptotic concentration of DOM depth profiles. Refractory DOC concentrations vary by ocean basin in the range of 37.7 (South Pacific)-45.0 µM (Arctic). Globally constant concentrations are used for refractory DON (1.8 µM) and refractory DOP (0.03 µM). Full details of this DOM database are given elsewhere .

First iteration -DOM source from BEC PP
The linear DOM cycling model cycles DOM with one source/sink and uses an idealized annual ocean circulation in offline mode from the CESM POP2 ocean circulation model (Bardin et al., 2014); nominal 1 • × 1 • horizontal resolution with 60 vertical levels, i.e., the same grid as the standard BEC v1.2.1. In this simple model of DOM cycling, two tracers of DOM are simulated for each element, C, N, and P: semilabile (SLDOM) and refractory (RDOM). The source for each DOM tracer is parameterized as some variable fraction, f , of primary production and is formed within the top model grid level with a thickness of 10 m. The sink for each DOM fraction is microbial remineralization parameterized with an assigned remineralization rate, κ, that differs for the euphotic zone and deep ocean layers in the case of SLDOM. The conservation equations for each DOM tracer are where T is the advection-diffusion transport operator (a sparse matrix constructed using output from the dynamical CESM POP2 model as described in Bardin et al., 2014) and PP is the annual average 3-D primary production field from the coupled ocean-atmosphere run of the CESM for the 1990s (Moore et al., 2013). We tested the sensitivity of the linear DOM model results to multiple production functions (CESM PP, DOM production flux from the BEC, satellite estimated PP); however, results suggest the differing source functions do not appreciably alter modeled DOM distributions or parameter values.
Our initial construction of the linear DOM model allowed the sum of f 1 + f 2 to vary continuously between 0 and 0.5 and κ i |i = 1· · ·3 to vary logarithmically between 0.25 and 20 000 yr −1 by 24 discrete values. The direct-solver technique makes it possible to objectively calibrate these parameters, f i κ i |i = 1· · ·3 {f 1 , f 2 , κ 1 , κ 2 , κ 3 }, by using a numerical optimization algorithm that rapidly tests each permutation of the discretized κi values, scaled by f i , in order to find the parameter set that minimizes the root mean square difference in the misfit between the model-predicted and observed DOM concentration. A separate linear DOM model (Eqs. 1, 2) is solved for the DOC, DON, and DOP cases. The DMI solver allows us to determine very efficiently the optimal lifetimes for the various DOM pools. It is not practical to determine these using multiple forward simulations of the full CESM BEC, which would require years to decades of computer time.

Optimized DOM parameter incorporation into the BEC model
The optimized parameter values obtained from the DMIenabled linear DOM model were incorporated within the BEC to improve its representation of DOM cycling. The BEC model has two tracers for each DOM pool, semilabile and refractory, with differing lifetimes for the euphotic vs. mesopelagic zones. Thus the SLDOM lifetimes, κ 1 and κ 2 , from the DMI-enabled DOM model were applied to the BEC model semilabile tracers for the euphotic zone and mesopelagic, respectively. The RDOM lifetime from the DMI-enabled DOM model was applied throughout the full water column of the BEC model. Further finetuning of DOM lifetimes was carried out to provide the best DOM optimized case, using the mean bias of the modeled concentrations versus the observations and the log-transformed regression correlation coefficient between simulated and observed DOM in the upper ocean, 0-500 m, as comparison metrics. The BEC simulation containing the set of improved DOM cycling parameters following the first iteration of the DMI-enabled linear DOM model is termed DOM DEV.

Second iteration of DMI-enabled linear DOM model -DOM source from BEC DOM production flux
Initial improvements to DOM cycling metrics within the BEC model were large upon incorporation of the DMIenabled linear DOM model parameter values; however, because of differences between the offline model and the full BEC model, further improvements to the DOM tracer lifetimes were possible. To achieve this the DMI-enabled linear DOM model was modified such that the production for each tracer was held constant allowing only the remineralization rate, κ i , to be optimized from a choice of 48 discrete tracer lifetimes spanning the range 0.7-20 000 yr −1 . Rather than using PP to get the production flux of each DOM tracer, the semilabile and refractory DOM production fluxes (SLDOMprod, RDOMprod) were extracted from the DOM DEV simulation and prescribed in the modified DMI-enabled DOM model. The fraction of SLDOMprod and RDOMprod to be applied each κ i |i = 1· · ·4 was diagnosed from the relative proportions of each tracer residing in the euphotic or deep layers at the end of the DOM DEV simulation of the BEC (see Fig. 2). At this step it was also desired to solve for the remineralization rate associated for a secondary sink for DOMr due to photo-oxidation in the surface layer. Thus Eqs. (1) and (2) were modified to become Eqs. (3) and (4) as follows: The results obtained from the modified DMI-enabled linear DOM model were incorporated into the final DOM optimized ( DOM OPT) simulation of the BEC following minor tuning of the κ parameter values. Note the only minor changes to tracer lifetimes, κ −1 i , between the modified DMI-DOM model and the DOM OPT simulation. The value "%remin" represents the percentage of the DOM production flux that is remineralized within each depth horizon on an annual basis and is common to both models.

Second iteration of DMI-enabled linear DOM model -DOM source from BEC DOM production flux
Results    Table 1. Approximately 7 % of PP is routed to production of DOM, which is divided amongst semilabile (SLDOM) and refractory pools (RDOM). Remineralization lifetimes (κ −1 i ) differ for SLDOM depending on location in the water column with longer lifetimes for the euphotic zone (depths where PAR > 1 %) than for the mesopelagic zone. A faster rate of RDOM remineralization is assigned in the surface layer (< 10 m) to parameterize a sink due to photooxidation. The parameter, %remin, represents the percentage of the DOM production flux that is remineralized within each depth horizon on an annual basis with the sum equal to 100 % and is diagnosed from the DOM DEV simulation. The relative magnitude of SLDOM remineralization that occurs within the EZ vs. the MZ was found to be ∼ 1.8 : 1 (Fig. 2). Only a small percentage of RDOM remineralization occurs in the surface layer, i.e., 0.01-0.03 % (Fig. 2). The optimal tracer lifetimes from the modified DMI-DOM model were 15 yr for SLDOC in the EZ, 5 yr for SLDOC in MZ, 15 000 yr for RDOC, and 15 yr for RDOC whilst in the surface layer (< 10 m). DON tracer lifetimes were: 15 yr for EZ SLDON, 5 yr for MZ SLDON, 8000 yr for RDON, and 15 yr for RDON at the surface. DOP tracer lifetimes were 62 yr for SLDOP in the EZ, 4.5 yr for MZ SLDOP, 6000 yr for RDOP, and 15 yr for photo-oxidation removal.

Modeled DOM in the standard CESM BEC v1.2.1 (REF simulation)
A set of metrics were used to assess the performance and improvements to DOM cycling for the CESM BEC simulations including the global integrals of DOM production, export, and C : N : P stoichiometry (Table 2) as well as the mean bias and correlation coefficient (r) of the simulated DOM concentrations against the observational data set in the upper 500 m (Table 3). Results and comparison of DOM cycling metrics from the REF simulation are presented in Tables 2 and 3.  (Table 2). About 85 % of this DOC production is remineralized within the euphotic zone, yielding DOC export from the euphotic zone of ∼ 1.7 Pg C yr −1 . Modeled semilabile DOC concentrations from the REF simulation are shown for the surface (Fig. 3a)   DON. Total euphotic zone DON production is 120 Tmol N yr −1 with a > 100 m depth export of 25.0 Tmol N yr −1 (Table 2). Modeled total DON concentrations (semilabile + refractory) at the surface are similarly overestimated in the REF simulation (Table 3) by up to 100 % within the subtropical gyres of the Pacific and the eastern South Atlantic oceans (Fig. 4a). Modelobservation misfit is better at 200 m (Fig. 4b); however, biases of 15-25 % remain (Table 3) in a number of regions (e.g., central equatorial Pacific and southern Indian oceans).
DOP. Total euphotic zone DOP production is 7.43 Tmol P yr −1 with export out of the euphotic zone valued at 1.30 Tmol P yr −1 (Table 2). Modeled DOP distributions are shown in Fig. 5a (surface), and 5b (200 m), with observations mostly limited to the Atlantic Ocean. The region of elevated simulated DOP (> 0.25 µM) in the eastern South Atlantic surface waters is located further to the east than is observed (Fig. 5a), possibly owing to the snapshot nature of the observations (collected in January-February) compared to the annually averaged simulation. Modeled DOP in the subsurface agrees reasonably well with the Atlantic observations, except for a ∼ 70 % model overestimate in the South Atlantic subtropical gyre (Fig. 5b).

Modeled DOM in the DOM OPT simulation
Results and comparison of DOM cycling metrics from the DOM OPT simulation against the observational data set and REF simulation are presented in Tables 2 and 3. For a comparison of the set of DOM cycling parameter values between the REF and DOM OPT simulations, see Table 1. DOC. Total DOC production in the euphotic zone (upper 100 m) for the DOM OPT simulation is 4.16 Pg C yr −1 ( Table 2). About 45 % of this DOC is remineralized within the euphotic zone, yielding DOC export from the EZ of 2.28 Pg C yr −1 , which is ∼ 20 % larger than the result from a separate DOC data assimilative modeling study (Table 2; Hansell et al., 2009). Combined with the particulate organic carbon export from 100 m in the DOM OPT simulation of 7.01 Pg C yr −1 , DOC contributes ∼ 25 % to the total 9.29 Pg C yr −1 of export production in the CESM BEC. Modeled total DOC concentrations (semilabile + refractory) from the DOM OPT simulation are shown for the surface (Fig. 3c) and at 200 m (Fig. 3d). There is generally good agreement between the simulated fields and observations (colored dots) with the mean bias being < 20 % for the upper ocean (0-500 m; Table 3). Slightly larger model overestimations (up to ∼ 30 %) exist at the surface for certain low-latitude ocean basins (e.g., tropical Atlantic, and Indian oceans).
DON. Total euphotic zone DON production is 30.7 Tmol N yr −1 with a > 100 m depth export of 16.4 Tmol N yr −1 (Table 2). Modeled total DON concentrations are improved over the REF simulation at 200 m (Fig. 4d); however, overestimations of DON at the surface remain in DOM OPT (Fig. 4c). Simulated surface DON overestimation is largest in the low to mid latitudes, reaching ∼ 30 %. Opposite the pattern obtained for the low latitudes, high-latitude simulated DON is underestimated at the surface in the Southern Ocean (Fig. 4c) by up to ∼ 35 %. However, overall, DON mean biases are small in the DOM OPT simulation, i.e., < 10 % (Table 3).  DOP. Total euphotic zone DOP production is 2.94 Tmol P yr −1 with export out of the euphotic zone of ∼ 1 Tmol P yr −1 (Table 2). Modeled DOP distributions are shown in Fig. 5c (surface), and 5d (200 m). The DOM OPT simulation captures the low DOP concentrations observed in the North Atlantic, largely due to enhanced phytoplankton direct uptake of DOP (see Sect. 3.5). The region of elevated simulated DOP (> 0.25 µM) in the eastern South Atlantic surface waters continues to be located further to the east than is observed (Fig. 5c) in the DOM OPT simulation as was also the case in the REF simulation. Modeled DOP in the subsurface agrees reasonably well with the Atlantic observations, reducing the large overestimates in the REF simulation (Fig. 5d, b). Overall, mean DOP biases are similarly < 10 % for both the total and semilabile pools (Table 3).

Comparison of multiple DOM cycling schemes in the CESM BEC
We have also tested other hypotheses for DOM cycling formulations such as non-variable C : N : P cycling stoichiometry (i.e., DOM cycling occurs at the Redfield ratio) as well as more rapid turnover of DOM in the EZ compared to the MZ (the DOM OPT simulation contains more rapid turnover of DOM in the MZ, following the work of Carlson et al., 2004;Letscher et al., 2013a). To test these hypotheses, we performed two additional BEC simulations termed RED-FIELD and EZRAPID using a coarser-resolution version of the BEC model with a nominal 3 • horizontal resolution. The optimized cycling parameter values obtained for DOC from the DOM OPT simulation were assigned to the DON and DOP pools for the REDFIELD simulation to allow all DOM (C : N : P) to cycle at the same rate and in the same proportions. The ability for phytoplankton to directly utilize DOP is also turned off in the REDFIELD simulation. The optimized EZ and MZ lifetimes for each DOM tracer from the DOM OPT simulation were reversed for the EZRAPID simulation such that the shorter lifetime (more rapid remineralization rate) was assigned to SLDOM in the EZ. Results from 310 yr simulations of these are compared against ∼ 3 • simulations of REF and DOM OPT in Table 4. Results are similar for DOC when comparing the DOM OPT and REDFIELD simulations, which is to be expected as the REDFIELD simulation used the same DOC cycling parameters as the DOM OPT simulation. Faster turnover of DOC in the EZ (EZRAPID simulations) had a detrimental effect on DOC mean biases, resulting in large overestimations in the upper 500 m (Table 4) when compared with faster turnover in the MZ (DOM OPT). Large positive mean biases were also found for DON within the REDFIELD and EZRAPID simulations when compared to the DOM OPT (Table 4). Similar positive biases were found for DOP within the RED-FIELD and especially for the EZRAPID simulations, i.e., up to ∼ 135 % (Table 4).

Direct DOP uptake by phytoplankton
The longer lifetimes for semilabile DOP in the DOM OPT simulation (on the order of years) allow for significant horizontal advection of DOP from the more productive gyre margins (e.g., the NW African upwelling region) towards the Sargasso Sea, providing an additional phosphorus source to the western North Atlantic. Each phytoplankton group within the BEC model can directly utilize DOP to satisfy their phosphorus requirements when phosphate concentrations are low . Literature reports of this phenomenon are numerous (e.g., Bjorkman and Karl, 2003;Casey et al., 2009;Lomas et al., 2010;Orchard et al., 2010) whereby phytoplankton make use of extracellular alkaline phosphatases to cleave phosphate groups from DOP moieties such as phosphate mono-and diesters (Dyhrman and Ruttenberg, 2006;Sato et al., 2013) for subsequent uptake of the liberated phosphorus. Sohm and Capone (2006) provide half-saturation constants for DOP uptake by Trichodesmium spp. (a diazotroph) and bulk phytoplankton (dominated by nano-and picophytoplankton) from the subtropical North Atlantic, and suggested Trichodesmium species obtained much of their required phosphorus from DOP in this region. Based partly on this study, the diazotrophs have been given a lower half-saturation constant for DOP uptake than the other phytoplankton . Diatoms also exhibit alkaline phosphatase activity albeit at lower rates than other plankton groups (Dyhrman and Ruttenberg, 2006;Nicholson et al., 2006), and were thus assigned a greater half-saturation for DOP uptake than the other phytoplankton groups in the BEC (consistent with their reduced efficiency in taking up dissolved inorganic phosphorus in the model).
The fraction of total phosphorus uptake that is sustained by DOP uptake for each phytoplankton group in the DOM OPT simulation is shown in Fig. 6. DOP uptake is largest by diazotrophs (Fig. 6b), with generally ∼ 20 % of P uptake from DOP in the subtropical gyres, increasing to ∼ 30-50 % in the subtropical North Atlantic, western side of the subtropical South Atlantic, and the eastern Mediterranean Sea. DOP uptake represents a small fraction (< 5 %) of P uptake by the small phytoplankton and diatoms (Fig. 6a, c) over much of the ocean, increasing to ∼ 10 % in the subtropical ocean gyres.

Discussion and summary
This study utilized a rapid solver of a simple linear biogeochemical cycling ocean model, constrained by our compilation of marine DOM observations, to efficiently optimize DOM biogeochemistry in the larger complexity CESM BEC model. This approach allows for a quicker and more quantitatively robust method for optimizing biogeochemical ocean model parameters over the traditional "hand"-tuning approach. Model parameters determined with the modified DMI-enabled linear DOM model carried over well when implemented in the full CESM BEC (see Fig. 2). The DOM OPT simulation contains reduced mean biases, improved correlation coefficients, and is more consistent with the DOM observational constraints when compared to the REF simulation (Figs. 3,4,5; Tables 2, 3).
Our results demonstrate that allowing for non-Redfield stoichiometry in the DOM pools significantly improves the match to observed DOM distributions. The order of lability follows P > N > C, diagnosed from the calculated effective tracer lifetimes in DOM OPT which include the net result of the sum of tracer sinks (∼ 3.2 vs. 6.3 vs. 6.8 yr for semilabile P, N, C; ∼ 4300 vs. 6360 vs. 13 900 yr for refractory P, N, and C). The exact values of the DOM lifetimes determined in our study are dependent on the underlying ocean circulation model used; however, Hansell et al. (2012) determined similar values for refractory DOC (16 000 yr) and combined semilabile and semirefractory DOC (∼ 7 yr; estimated from their Fig. 5) while using a distinct ocean circulation model than the one employed in the current study. In addition, the DOM lifetimes from the DOM OPT simulation are in general agreement with available estimates from the literature. Semilabile DOC lifetime has been estimated at ∼ 1-13 yr in the mesopelagic zone of the Sargasso Sea (Hansell and Carlson, 2001), and ∼ 7-22 yr in the mesopelagic zone of the North Pacific subtropical gyre (Abell et al., 2000). Semilabile DON lifetimes have been estimated at ∼ 3-12 yr (Letscher et al., 2013a) or ∼ 11-20 yr (Abell et al., 2000) for marine DON and ∼ 4-14 yr for terrigenous-derived DON in the Arctic Ocean (Letscher et al., 2013b).
Our DOM OPT simulation-estimated export C : N : P ratio of 225 : 19 : 1 for the semilabile DOM is in excellent agreement with the estimate of 199 : 20 : 1 by Hopkinson and Vallino (2005) and strongly supports the idea that DOC is exported efficiently relative to DOP compared with the canonical Redfield ratio. The calculated export efficiencies, that is the fraction of euphotic zone DOM production that is exported below 100 m, are 55, 53, and 17.5 % for DOC, DON, and DOP, respectively.
We found that best agreement with observed DOM distributions required a more rapid degradation of semilabile DOM in the mesopelagic than in the euphotic zone. This result is consistent with some incubation studies of DOM degradation (Carlson et al., 2004;Letscher et al., 2013a). Possible hypotheses for this depth dependence on DOM lifetimes in the real ocean are numerous, including differences in DOM composition/quality (Skoog and Benner, 1997;Aluwihare et al., 2005;Goldberg et al., 2011), microbial community structure (Giovannoni et al., 1996;Delong et al., 2006;Treusch et al., 2009;Carlson et al., 2004Carlson et al., , 2009Morris et al., 2012), availability of inorganic nutrients for heterotrophic utilization (Cotner et al., 1997;Rivkin and Anderson, 1997;Caron et al., 2000), abundance of bacterial grazers (Caron et al., 2000), and the presence or specific affinity of microbial cell membrane nutrient transporters (Azam and Malfatti, 2007;Morris et al., 2010). However, the relative importance of each of these mechanisms is not well constrained, nor are any considered in the BEC model formulation, and thus require further investigation.
Direct uptake of DOP by phytoplankton seemed necessary in our simulations to capture the observed very low surface DOP concentrations in the Sargasso Sea. Yet there are large uncertainties in the preference and uptake efficiencies for dissolved inorganic phosphorus versus dissolved organic phosphorus by different phytoplankton groups. Future field and lab studies are needed to reduce these uncertainties and to better quantify the role of DOP in determining spatial patterns of nitrogen fixation. There is also a great need for additional DOP measurements in every basin except the North Atlantic, along with improved quality control and the development of a DOP standard reference material.