RADIv1 was designed for deep-sea sediments and does not account for methane or solute transport by bottom-water hydrodynamics . In RADIv2, we added methane to address its role in sediment biogeochemistry, particularly in coastal regions where methane formation and release can play a larger role compared to the deep sea. In coastal and shelf environments, hydrodynamic properties are highly variable and can influence key transport mechanisms, such as the thickness of the DBL and enhanced porewater diffusivity through dispersion .

Figure 2a shows the fluxes simulated by RADIv2 alongside a comparison with values reported in the literature. For North Sea O2 fluxes, the model shows a positive bias (mean 13.8 vs 10.9 mmol m⁻² d⁻¹; Welch p=0.003): one third of the simulated values fall within the observational 25 %–75 % range, while all lie within the 5 %–95 % observational envelope. For DIC, the model substantially overestimates the observed fluxes (mean 16.8 vs 8.9 mmol m⁻² d⁻¹; Welch p=1.33×10-7), with ∼ 3 % of simulated values within the observational 25 %–75 % range and ∼ 23 % within the 5 %–95 % range, indicating that most model values exceed the bulk of observed variability. For North Sea TA fluxes, modelled and observed values overlap: all simulated fluxes fall within both the observational 25 %–75 % and 5 %–95 % ranges, and the Welch test (p=0.22) does not indicate a significant systematic bias, suggesting no model offset for TA.

The simulated DIC fluxes for the North Sea are higher than the literature range. This likely reflects a combination of too strong organic-matter supply and too reactive organic-matter parameters in our set-up. The imposed POM rain is based on annual-mean export estimates, which smooth over episodic blooms and short-lived pulses and can therefore misrepresent the supply at the time of the benthic flux measurements. Combined with highly reactive incoming OM, most degradation occurs near the sediment–water interface and drives larger DIC fluxes (and slightly elevated O2 uptake) than those inferred from the local measurements and modelling studies. Moreover, the model produces narrower flux ranges compared to in-situ measurements. This discrepancy can be attributed to the steady-state nature of the simulations, which do not account for transient variations such as temperature, tidal oscillations and internal waves. Internal waves, including their breaking counterparts, can induce sediment resuspension , which in turn affects benthic fluxes and influences the benthic boundary layer through turbulence . Tidal cycles can also introduce substantial short-term variability in fluxes . These dynamic processes, which are not captured by steady-state simulations, can lead to significant fluctuations in solute and particulate fluxes across the sediment-water interface, particularly in coastal regions where wave action and tidal forcing are more pronounced.

The modeled remineralization rates (Fig. 2b) are within the same order of magnitude than those observed in continental shelf environments, although they exceed observations from the North Sea. This positive bias is consistent with the combination of relatively high imposed organic-matter supply and reactive organic-matter parameters discussed above. The relative contributions of aerobic and anaerobic remineralization in RADIv2 align with both reported values for North Sea sediments and typical continental shelf environments (Fig. 2b). This agreement indicates that RADIv2 captures organic matter degradation pathways under varying redox conditions.

For Monterey Bay (Fig. 3a), O2 fluxes show good agreement between model and observations: the means are nearly identical (9.29 vs 9.11 mmol m⁻² d⁻¹), and no significant mean bias is detected (Welch p=0.85). Nearly all simulated values (97 %) fall within the observational 25 %–75 % range, and all lie within the 5 %–95 % envelope. For DIC, the predicts higher fluxes (mean 12.0 vs 9.95 mmol m⁻² d⁻¹), while the Welch test provides only marginal evidence for a mean difference (p=0.10), reflecting the small sample size (nobs=7). About one third of simulated DIC values fall within the observational 25 %–75 % range and ∼ 87 % within the 5 %–95 % envelope. TA fluxes agree closely (means 4.46 vs 4.35 mmol m⁻² d⁻¹; Welch p=0.90), with 63 % of model values inside the observational 25 %–75 % range and all within the 5 %–95 % envelope.

The modest DIC bias likely reflects uncertainties in the imposed organic-matter supply and reactivity, while the comparatively narrow model spread in O2 is consistent with the steady-state nature of the simulations, which do not resolve short-term variability. Internal tides and associated internal-wave activity have been shown to significantly influence sediment dynamics in Monterey Bay , but are not represented in the present set-up. RADIv2 simulates an organic matter remineralization rate for Monterey Bay that is approximately 2–3 times higher than observed rates (Fig. 3b). However, the values in are lower than the global-mean Corg remineralization rate for continental shelf environments (∼ 12 mmol C m⁻² d⁻¹ ). Because did not measure POM rain to the seafloor directly, the POM flux range used in Table 2 was estimated by analogy with the North Sea calibration. This choice likely overestimates the actual organic-matter supply at Monterey Bay and contributes to the positive bias in the simulated remineralization rates. The relative contribution of aerobic and anaerobic organic matter pathways simulated by RADIv2 is consistent with values observed in other continental shelf studies (Fig. 3b) .

The Iberian Margin validation uses data from five transects spanning a wide range of depths, as reported by . Two input files were created: one for shallow locations (<400 m) and one for deeper locations (>400 m). For the simulations, the average organic carbon decay rate kavg=pf×kf+ps×kslow is constrained between 1.5 and 2.25 a⁻¹ (Table 2), based on high-resolution in-situ oxygen microprofiles .

For the Iberian Margin (Fig. 4a), RADIv2 overestimates O2 uptake at both stations. At the shallow site, the mean simulated flux is 4.38 vs 2.75 mmol m⁻² d⁻¹, and at the deep site 2.38 vs 1.07 mmol m⁻² d⁻¹, with both Welch tests indicating a significant positive bias at each location (p=1.45×10-5 and p=2.47×10-8, respectively). At neither site do simulated values fall within the observational 25 %–75 % range, and ∼ 37 % (shallow) and ∼ 43 % (deep) lie within the 5 %–95 % observational envelope. The model also overestimates the Corg remineralization rates (Fig. 4b). already noted that the observed remineralization rates on the Iberian Margin are unexpectedly low given the high primary production in this region (∼ 7–11 mmol C m⁻² d⁻¹) and that they fall well below predicted values based on empirical relationships described in . Shelf sediments off Peru exhibit rates of 5.3±2.9 mmol C m⁻² d⁻¹ , and margin sediments off Chile, up to depths of 2000 m, show rates approximately 15 times higher . In our set-up, we adopt the export flux estimates from without additional site-specific tuning, so uncertainties in the effective POM supply at these stations likely contribute to the positive bias in RADIv2 remineralization rates. A dedicated comparison of simulated and observed O2, NO3- and NH4+ porewater profiles at station 99–6 is provided in Sect. S1 of the Supplement, where we also assess the sensitivity of these profiles and the associated fluxes to vertical resolution.

Although DIC fluxes were not directly measured in this study, they can be inferred from the modeled O2 influx using the Respiratory Quotient (RQ), defined as the molar ratio of DIC production to O2 consumption during organic matter remineralization , RQ =ΔDICΔO2. This approach provides a practical means of evaluating DIC fluxes in the absence of direct in-situ measurements. Under idealized conditions, where organic matter is represented by redfield organic matter, CH₂O, the RQ equals 1.0, reflecting a 1:1 stoichiometric ratio between O2 consumption and DIC production. However, re-evaluated marine phytoplankton composition lead to slightly lower RQ values, around 0.88–0.90 . Some studies also account for additional oxygen-consuming processes, particularly nitrification, which further lowers RQ values . Additional variability in RQ can come from incomplete reoxidation due to losses of N₂ or burial of FeS₂. FeS₂ can later be reoxidized through bioturbation-driven exposure to oxygen, further influencing oxygen demand and introducing additional variability in RQ in both directions. An empirical approach to deriving RQ values involves using in-situ measurements of benthic DIC efflux and O2 influx. Flux-derived RQ values compiled by span a broad range, from 0.69 to 1.31. The variation from theoretical stoichiometry can be attributed to factors including carbonate dissolution, which enhances DIC fluxes without directly affecting O2 consumption , and the temporal decoupling of O2 and DIC fluxes during episodic organic matter deposition . RADIv2 simulates average flux-derived RQ values around 1.0 for both shallow and deep sites on the Iberian Margin. These values are consistent with the Redfield stoichiometry assumed in the model and fall well within the empirical range reported by .

RADIv2 was validated using data from two deep-sea stations located in the Arabian Sea, WAST (4050 m depth) and SAST (4420 m depth), as described in . These stations were selected for validation due to their contrasting organic matter export fluxes, benthic carbon remineralization rates, and benthic oxygen fluxes. Because the number of available observations is low (n≤3), we do not apply the statistical tests used for the other locations.

WAST is characterized by high benthic O2 uptake, one of the highest recorded in global deep-sea environments . This suggests that deep-sea sediments are not always stable, constant environments, but are influenced by variations in organic matter fluxes . The high benthic O2 uptake is primarily driven by high particulate organic carbon (POC) rain rates of 1.38 mol m⁻² a⁻¹, resulting from monsoonal upwelling and lateral advection from the productive Arabian shelves . These elevated POM fluxes lead to relatively high benthic remineralization rates compared to other deep-sea sites, although RADIv2 simulates higher values than observed (Fig. 5b). Due to the higher POM rain, RADIv2 predicts aerobic benthic remineralization rates that align more closely with those observed in more shallow marine environments (Fig. 5b). As a result, the simulated aerobic remineralization percentages are higher than those reported by .

For SAST, located further from the shelf, the significantly lower POM rain rates lead to lower O2 fluxes (Fig. 5a), more in line with typical deep-sea O2 uptake values . Here, RADIv2 overestimates both O2 fluxes and remineralization rates, though the model’s aerobic remineralization percentage are consistent with those reported by (Fig. 5b).

As no DIC measurements were available for these stations, we inferred DIC fluxes using the flux-derived RQ in RADIv2. The simulated RQ is 1.25 for WAST and 1.13 for SAST. Both are higher than those simulated for the Iberian Margin, due to more calcium carbonate dissolution. The values fall within the broader range of RQ values derived from in-situ flux measurements reported for marine sediments by .

The simulated ranges of solute fluxes in RADIv2 generally align with values found in the literature, although they are narrower. This narrowing likely results from the steady-state conditions employed in the simulations, which do not account for short-term variability, such as tidal and higher-frequency fluctuations. At some locations, RADIv2 shows a tendency to overestimate benthic fluxes, which likely reflect uncertainties in the prescribed POM supply and organic-matter degradation kinetics. For locations without observed DIC fluxes, the empirical RQ, inferred from benthic DIC and O2 fluxes, falls within the range of in-situ RQ values reported in the literature .

However, the model tends to systematically overestimate organic carbon remineralization rates compared to observations. For most sites, the POM rain rates are taken from literature estimates or export–production relationships and are not co-located and time-aligned with the individual in situ flux measurements, so uncertainties in the local POM supply and its degradation kinetics are likely the source of this positive bias. To quantify how much of the disagreement could in principle be absorbed by either the POM flux or the reactivity, we performed three sensitivity tests at the Iberian Margin shallow calibration site, holding all other parameters fixed. Scaling the imposed POM rain by factors 0.5–1.5, while keeping all other organic-matter parameters constant, changes the vertically integrated remineralization rate by roughly -40 % to +40 %, with an almost linear response. In contrast, globally scaling both kfast and kslow over a range of 0.5–2.0, while keeping the fast/slow partitioning and POM rain constant, produces smaller and more non-monotonic changes in the integrated rate because faster degradation near the surface reduces the amount of POM transported to depth. Varying the partitioning between the fast and slow pools (while keeping total POM rain and rate constants fixed) only weakly affects the depth-integrated remineralization, but does shift where in the sediment column carbon is oxidized and hence the balance between aerobic and anaerobic pathways. It should be noted that, FPOM, kfast/kslow and their partitioning are not independently constrained, and different combinations of these parameters can yield similar total remineralisation rates. These tests indicate that uncertainties in the local POM flux are sufficient to account for much of the overestimation of remineralization, and that global rescaling of the organic matter kinetics and their fast/slow partitioning also affects the integrated rate, but to a lesser degree and more non-linearly. At the same time, the sensitivity tests suggest that RADIv2 is not locked into a “too-reactive” regime. The systematic positive bias across case studies is consistent with uncertainties in the local POM supply, and/or to lesser a extent, the effective reactivity of the sedimentary organic matter.

In RADIv2, diffusive mass transport across the SWI occurs via molecular diffusion in low-permeability sediments, with an additional dispersion term included for highly permeable sediments. This is particularly relevant in coastal sediments, where permeability is typically higher and bottom-water currents are stronger compared to deeper regions, resulting in sediment fluxes that exceed diffusive fluxes alone . In cases where sediments are sufficiently permeable, a third transport regime, turbulent transport, can contribute to mass transfer across the SWI . The transitions between these three regimes (molecular diffusion, dispersive diffusion, and turbulent transport), are characterized by the permeability Reynolds number, Rek=Ku*/ν, where K is the sediment permeability, u* is the shear velocity and ν is the kinematic viscosity. Rek parameterizes the turbulent regime at the seafloor, with the permeability of the sediments determining whether the turbulence can penetrate through the SWI. A low Rek indicates a regime dominated by molecular diffusion, while a high Rek suggests that turbulent transport dominates mass transfer across the SWI. These different regimes can be parameterized by effective diffusion coefficient and can be integrated into future updates of RADIv2. However, the main limitation of this method is its reliance on sediment permeability, a variable property that is challenging to measure in situ. While efforts are underway to map seabed permeability in coastal regions , a global dataset for linking permeability to RADIv2 is not yet available. A transient test case with tidal currents (Supplement Sect. S2, Fig. S2) illustrates how the present formulation in RADIv2 translates such hydrodynamic forcing into time-varying benthic O₂ fluxes.

Future efforts will focus on improving the representation of organic carbon degradation kinetics, a key driver of sediment fluxes. Currently, RADIv2 divides organic carbon into two reactive classes (kfast and kslow), each following first-order reaction kinetics. This approach to organic matter degradation produces a vertical structure in bulk (apparent) reactivity: the more labile pool is consumed near the sediment–water interface, so deeper layers are dominated by less reactive carbon. This is sufficient for well-mixed sediments, where organic carbon reactivity is relatively constant within the mixed layer . However, experimental studies , field observations , and recent theoretical advances have shown that organic carbon reactivity often declines with degradation. In RADIv2 each organic-matter pool has a fixed degradation rate constant, and degradation does do not weaken with depth or time. The work of summarizes the conditions under which different degradation kinetics are suitable, and highlights the need to model organic carbon reactivity as a continuum that declines with age and burial depth in regions that receive highly reactive OM or poorly-mixed environments, rather than a small number of fixed-k pools. Incorporating such an approach into RADIv2 would enhance its ability to simulate organic matter degradation under a broader range of marine sediments. Future work could focus on dynamically adjusting organic matter degradation kinetics based on the sediment mixing regime (quantified by the Péclet number) and the reactivity of incoming organic carbon. This would also help address the current overestimation of organic matter degradation rates by better capturing the declining reactivity of organic carbon over time, providing more accurate remineralization estimates.

Fluxes across the sediment–water interface are controlled by processes that are strongly concentrated in the upper centimetres of the sediment, so the vertical grid spacing must be sufficiently fine to resolve the associated reaction and concentration gradients. Future applications of RADIv2 could benefit from a non-uniform vertical grid, with finer spacing near the sediment–water interface and coarser spacing at depth, particularly in coastal sediments or simulations that require a deeper sediment column. RADIv2 is a living model, and ongoing developments focus on supporting non-uniform grids (for example, with finer dz near the SWI and coarser spacing at depth) to further improve the efficiency of simulating of near-surface processes.

Both RADIv1 and RADIv2 are optimised for simulating carbon and oxygen cycling. In the present configuration, RADIv2 includes dissolved Fe²⁺ and Mn²⁺ and their oxidised solid phases, and represents their role as electron acceptors in organic-matter degradation. However, it does not resolve a full Fe–Mn–S mineral network. Sulfide produced by sulfate reduction is treated with simplified oxidation and precipitation terms (see Table 1), and processes such as explicit FeS/pyrite formation, burial and re-oxidation are not included. These missing pathways can be important for the alkalinity budget in iron-rich, bioturbated coastal sediments, because they control whether sulfide is buried as pyrite or repeatedly re-oxidised and recycled back to sulfate through Fe–S redox cycling, and these alternative pathways have different net TA yields . Future extensions of RADIv2 should focus on implementing more comprehensive iron and manganese cycles with explicit sulfide–mineral pathways, following formulations such as and , to improve applications in such settings.

RADIv2 was designed with extensibility in mind, providing a flexible and adaptable framework that makes the addition of new components or species straightforward. Moreover, RADIv2 can be updated to represent specific environments, such as mangrove forests, estuaries, or nearshore intertidal zones. This versatility enables RADIv2 to serve as a robust foundation for advancing future diagenetic research across a wide range of applications.

Significant discrepancies exist between estimates of the ocean carbon sink derived from pCO₂ products and those produced by ocean biogeochemical models , which are further complicated by inter-model variability . One major uncertainty comes from the poor representation of particulate deposition at the seafloor and the resulting benthic fluxes from remineralization, dissolution, and exchange at the SWI. The exclusion of sediment-seawater exchange processes in current models limits their ability to accurately simulate coastal-ocean productivity and air-sea gas exchange as benthic processes play an important role in nutrient recycling and carbon dynamics , particularly in shallow regions where the seafloor and bottom waters directly interact with the surface layer. An analysis by of the marine biogeochemical components in CMIP5 and CMIP6 Earth System Models (ESMs) reveals that most CMIP5 models lacked any representation of sediments. While CMIP6 models have made significant progress incorporating sediment processes, their treatment remains highly simplified and, in some cases, is still omitted. When sediments are not explicitly represented, particulates that reach the seafloor are either instantaneously remineralized or permanently buried upon deposition (numerically lost) . Most ESMs that include sediments rely on a box-model approach, where organic matter reaching the seafloor is either remineralized or buried based on simplified relationships with downward organic fluxes. Some ESMs estimate organic matter remineralization through denitrification following the formulation of , with additional pathways for aerobic respiration or sulfate reduction . More advanced sediment models, such as those incorporated in HAMOCC and PISCES-v2 , adopt a layered sediment module based on . While this represents a significant step forwards, these models still rely on highly simplified treatments of key diagenetic processes. Important mechanisms, such as bioturbation (irrigation), dispersion, and anaerobic organic matter degradation are either absent or not fully accounted for.

Since RADIv2 is developed in Julia, while ESMs are typically written in FORTRAN, direct coupling is not straightforward without significant code translation. Additionally, fully kinetic diagenetic models like RADIv2 are computationally intensive, making their real-time integration into Earth System Models currently unfeasible. This study introduces a process-based metamodel that captures key diagenetic processes, often oversimplified or absent in existing parameterizations, without requiring full integration.

The metamodel is validated for diverse marine environments, accurately reproducing SWI exchanges simulated by RADIv2. By accounting for the heterogeneity and variability in sediment properties and environmental conditions, the metamodel is designed for broad applicability. However, the present training and validation set primarily samples temperate shelves, margins, and deep-sea settings, and does not explicitly include environments such as deltaic systems, very shallow nearshore regions, or carbonate-rich tropical shelves. Expanding the training ensemble to a wider, yet physically consistent, range of bottom-water conditions and particulate fluxes, that can also be validated with in-situ observations, represents an important future improvement, especially for more localized applications. The flexibility of RADIv2 allows for customization to specific marine environments, such as bays and estuaries, when region-specific data are available, enabling either the expansion of the present metamodel or the generation of specialized metamodels for regional applications.

Future work will focus on expanding the metamodel to include additional particulate fluxes, such as particulate metallic fluxes, and on parameterizing the corresponding SWI exchange. Ongoing developments will also expand the range of solute fluxes covered by the metamodel, including phosphate, nitrate, ammonium, methane, and particulate organic and inorganic carbon fluxes. This expansion is expected to improve biogeochemical budget estimates, as the lack of sediment regeneration of nutrients from remineralized organic matter can lead to underestimation of primary production in biogeochemical models . The adaptability of the RADIv2 framework allows the approach presented here to be extended to develop metamodels for any solute or particulate flux included in RADIv2, facilitating its application across diverse marine environments and geochemical processes.