To simulate the coastal AO environment, we used the ED-SBS regional configuration, whose general numerical characteristics are fully detailed in Sect. S1 in the Supplement and in . ED-SBS explicitly simulated four plankton functional types (PFTs) representative of ecosystems in the AO (diatoms, large eukaryotes, and small and large zooplankton), along with phytoplankton Chlorophyll a (Chl a) concentration. Two marine dissolved organic carbon (DOC) pools are simulated with chemical properties representative of those found in the coastal AO: a semi-refractory pool (DOC_sr) characterizing the long-residence-time carbon loop with a lifetime of τ = 10 years , and a semi-labile pool (DOC_sl) characterizing the short-residence-time carbon loop with a lifetime of τ = 1 month (including DOC molecules characterized by turnover rates ranging from weeks to months; ). Land-to-sea forcing included daily discharge of freshwater and 6 biogeochemical tracers from the Mackenzie River, distributed over the three major Mackenzie Delta outlets: Shallow Bay (29.8 %), Beluga Bay (37.6 %), and Kugmallit Bay (32.6 %) . Freshwater discharge was driven by daily gauge measurements from the Arctic Great River Observatory ArcticGRO; and was linked to daily river temperature obtained from the dataset. Riverine concentrations of DOC, dissolved organic nitrogen (DON), dissolved organic phosphorus (DOP), dissolved silicate (DSi), dissolved inorganic carbon (DIC), and alkalinity (Alk) were forced as detailed in . As each export of dissolved organic constituents (DOC, DON & DOP) are estimated independently, the terrestrial dissolved organic matter pool is not constrained by a constant C:N:P ratio such as in , , .

In this study, we added an explicit “CDOM-like” tracer to ED-SBS, expressed as a carbon mass concentration (mmol C m⁻³), following the schematic shown in Fig. . Terrestrial CDOM, which is observed to be non-labile , was added to the long-residence-time carbon loop of the model using the same microbial turnover time as DOC_sr (τ = 10 years). The CDOM tracer also interacted with the short-residence-time carbon loop by photochemical alteration of CDOM into more-labile carbon . CDOM was photodegraded into DOC_sl with a maximum bleaching turnover time of 6 d , which was modulated by light intensity. Bleaching rate linearly increased from 0 when light intensity is 0 W m⁻² to a maximum value (0.167 d⁻¹) when light is above 13 W m⁻² . CDOM photodegradation rate corresponds to the bleaching rate modulated by a temperature function. When degraded into DOC_sl, CDOM products DON and DOP followed the molar ratio of 120:16:1, allowing to represent the additional nutrient input generated by organic matter consumption – Redfield ratio is used due to a lack of data regarding CDOM photodegradation products. A fraction fCDOM (=2 %) of mass fluxes received by DOC_sl through phytoplankton grazing/mortality and particulate organic carbon (POC) dissolution was also redistributed to CDOM.

In ED-SBS, the Mackenzie River terrestrial DOC (tDOC) mass flux was equally distributed (50 %) between semi-labile (DOC_sl) and semi-refractory (DOC_sr) DOC pools (based on recent estimates of the bioavailable tDOC fraction in the SBS, Fabien Joux, unpublished data from Nunataryuk field campaign; ). While 97 % of DOC concentration variance is explained by CDOM absorption , the mass concentration of riverine CDOM exported to SBS coastal waters remains unknown. As CDOM is part of the long-residence-time loop, we redistributed a percentage of tDOC mass flux from DOC_sr into the CDOM pool. After a sensitivity analysis (detailed in Appendix B), we set the ratio to 2 % – re-partitioning Mackenzie River tDOC mass flux into 50 %, 48 %, and 2 % DOC_sl, DOC_sr, and CDOM, respectively. Our ratio of total tDOC exported as CDOM falls in the lower range of estimates for the top 10 DOC exporting rivers (4 %–38 % of tDOC; ). Finally, we generated CDOM initial and boundary conditions following the methods detailed in Sect. S2.

We first developed a new method for simulating CDOM light attenuation across the shortwave spectrum, from 320–735 nm. This allowed us to resolve the physical effect of CDOM light attenuation occurring in the UV-A (320–400 nm) and in the visible (400–735 nm) bands; the latter is often associated with Photosynthetically Active Radiation (PAR; spanning from 400–700 nm). An analysis of 31 CDOM spectral absorption measurements taken during the 2009 Malina campaign for different CDOM conditions across the SBS (see sampling locations in Fig. S1 in the Supplement; ) revealed that 40 %±10 (min:26–max:55) of light is absorbed by CDOM in the UV-A spectrum. These observations highlight the need to include full-band CDOM representation in OBMs, as most models only include light attenuation effects across PAR wavelengths. Note that in this study, we focus on light attenuation driven by CDOM absorption and disregard any backscattering effect from particulate matter. We acknowledge that the backscattering effect could play an important role in the SBS as the Mackenzie River is the Arctic's greatest exporter of particulate matter, but we aim here to build a foundation for determining the contribution of each component of terrestrial organic matter.

In our ED-SBS configuration, we approximated the relationship between CDOM light attenuation and its mass concentration (mmol C m⁻³) in high CDOM environments such as Arctic river-influenced waters. In this regard, we empirically estimated the CDOM diffuse attenuation coefficient (kCDOM; m⁻¹) from 31 in-situ measurements of the CDOM spectral absorption (aCDOM[λ]; m⁻¹ nm⁻¹) across the SBS (Fig. a). The standard solar irradiance spectrum ASTM G-173; was used as the reference shortwave solar spectrum at the surface ocean (Qsw0; W m⁻² nm⁻¹) – terms are listed in Table A1. We first calculated the shortwave spectrum attenuated from the surface ocean to 1 m depth (Qsw1; W m⁻³ nm⁻¹) by multiplying Qsw0 with aCDOM[λ] (Fig. b). Then, kCDOM was retrieved by integrating Qsw0 and Qsw1 over the chosen wavelengths for each station using Eq. (). 1kCDOM=1-∫λQsw1dλ∫λQsw0dλ, where lambda is the discrete wavelength (nm). Then, CDOM concentrations were estimated from aCDOM[440 nm] (m⁻¹) using the relationship from (Eq. ). 2CDOM=aCDOM[440nm]+0.2409Mc×0.0478, where Mc is the carbon atomic mass (Mc = 12.0107 g mol⁻¹). Finally, we fitted a hyperbolic tangent function (Eq. ) to obtain the relationship linking kCDOM and CDOM concentrations across the range of conditions found in the SBS (Fig. c). 3kCDOM=a×tanh⁡(b×CDOM+c)+d.

As shortwave radiation and PAR were simulated independently in the physical and biogeochemical components of the model, we calculated two different sets of parameters for the kCDOM/CDOM concentration relationship for both components. Both relationships yielded R2 ≥ 0.98. Parameters fitted with the full shortwave spectrum (used in the physical component) were: a=-0.15, b=-1.31, c=1.04, and d=0.12. Parameters fitted with PAR (used in the biogeochemical component) were: a=-0.14, b=-1.18, c=1.04, and d=0.10.

We included the effect of CDOM on light attenuation in the biogeochemical component of the model (which already included light attenuation by water and Chl a). PAR intensity (I(z), W m⁻²) at depth z is calculated according to the following equation: 4I(z)=(1-fice)×0.4×Isw×e-[kw+kChl a×Chl a(z)+kCDOM(z)]dz, where Isw (W m⁻²) is the shortwave downwelling irradiance (input from the physical component of the model), for which 40 % is considered as PAR, fice is the ice-cover fraction, kw is the diffuse attenuation coefficient for pure seawater (kw = 0.04 m⁻¹), kChl a is the Chl a diffuse attenuation coefficient (kChl a = 0.04 m² mg Chl a-1), Chl a(z) (mg Chl a m⁻³) is the total concentration in Chl a at depth z, and kCDOM is the diffuse attenuation coefficient for CDOM at depth z.

We included the biophysical feedback of CDOM light attenuation ocean warming by including kCDOM, integrated over the entire shortwave spectra in the physical component of the model (see Sect. ). The physical component of the model already included the thermal effect of light attenuation by seawater, calculating a downwelling light decay profile (dksw; 1-D) based on Jerlov water types and decreasing from the ocean surface to seafloor starting with a value of 1 at the surface. We included the thermal effect of CDOM light attenuation by calculating a CDOM light decay profile (dkCDOM) based on kCDOM (Eq. ), also decreasing with depth starting from 1 at the surface. As CDOM concentrations are variable in space, the resulting light decay profile produces a 3-D field. 5dkCDOM(0)=1dkCDOM(z)=dkCDOM(z-1)×e-kCDOMd(z-1), where dkCDOM(0), the decay at the surface ocean (0 m depth) is set to 1, since simulated light has not yet been affected CDOM and z-1 is the depth of the vertical grid cell above z. The dkCDOM calculation is then propagated from the ocean surface to the seafloor, as its value at depth z depends on all the values above. We then multiplied both decay profiles to yield the total decay profiles (dktot; 3-D) as follows: 6dktot(z)=dksw(z)×dkCDOM(z).

The setup described above represents a significant advancement over the previous model development by . We took the advantage of an extensive in-situ carbon dataset collected in 2009 to update the parameterization of CDOM mass fluxes as they transition between short and long-residence-time carbon loops, where it was previously represented using a single DOC pool . We also revisited the kCDOM/CDOM relationship, transitioning from a linear to a hyperbolic tangent relationship (see Fig. c) This is particularly relevant for river plume regions where CDOM concentration reaches high values. Finally, our developments included the heating contribution of CDOM UV-A absorption, which contributes to roughly 40 % of CDOM light absorption in the Mackenzie shelf region. The ED-SBS setup presented here is thus able to better represent the terrestrial browning effect on Arctic coastal regions.

We first describe the seasonal phenology of several important physical and biogeochemical variables in the simulated Mackenzie River plume. In the river plume, Run_full simulates an average surface CDOM concentration of 0.85 ± 0.08 mmol C m⁻³ from August to May, with a peak of 2.04 mmol C m⁻³ during the spring freshet, followed by declining concentrations in July (Fig. , black line). With regard to the sea ice phenology in the river plume, the model simulates an open-water period of ∼ 4 months (115 d), with SLIP and SGIP lasting 1 month (13 June to 9 July) and 1 week (2 to 10 November), respectively. From January to June, the SST is on average near the seawater freezing temperature (-1.93 °C) and slowly starts heating up in June with increasing shortwave downwelling irradiance at the ocean surface (Isw0; Fig. b) and accelerating freshwater discharge. In July, ocean-surface shortwave downwelling irradiance reaches a maximum, rapidly heating SST until it reaches a peak value of 10.3 °C on 8 August. Then, temperatures slowly cool until the end of SGIP. Phytoplankton rapidly bloom during the SLIP period, with a peak in surface NPP of 8.35 Gg C d⁻¹ occurring two days after DofR. The production period – defined as the duration when NPP exceeds half of its maximum – lasts 7 d (7 to 14 July) and coincides with the period when subsurface light is the most intense. Nitrate and phosphate are quickly consumed during the phytoplankton bloom until the nitrate stock is depleted. Nutrient stocks are replenished through vertical mixing, advective transport, and remineralization from October to June. The simulated silicate tracer is directly connected to DSi riverine mass flux and therefore increases with elevated runoff.

Within the Mackenzie River plume region, Run_full captures the mean SST amplitude and variability during the open-water period depicted by observations (Fig. D1). The model underestimates SST by 17 % from mid-July to mid-September. This is due to a later simulated SLIP, which delays surface-ocean heating and causes simulated SST to increase later in the season. Run_full also reasonably reproduces the amplitude of the phytoplankton bloom observed by remote sensing, as the simulated surface-ocean Chl a peaks at approximately the same concentration as reported by . However, the model underestimates the bloom's duration, simulating a bloom that lasts only half as long as observed by satellite. This discrepancy arises from the model's later simulated SLIP (similar to its SST behavior) and the rapid depletion of nitrates during the late open-water period. A more detailed and comprehensive model-data evaluation for 2012 is provided in Appendix D.

We next explore how the inclusion of riverine CDOM impacts light attenuation characteristics on the Mackenzie River shelf by comparing Run_full (presented above) to two similar set-ups: (1) excluding CDOM riverine forcing (Run_noriv; autochthonous CDOM only) and (2) using a linear CDOM light attenuation only in visible light (similar to ; Run_lin). We analyze the differences for the month of July, when shortwave downwelling irradiance (Isw) is maximum and terrestrial CDOM is more likely to affect the biophysical characteristics of the plume region. The simulation excluding river mass flux exhibits a space-time mean kCDOM of 0.02 m⁻¹ (Run_noriv) in the plume region (Figure a). Including riverine CDOM increases kCDOM to 0.13 and 0.16 m⁻¹ when using a linear (Run_lin) and hyperbolic tangent (Run_full) relationship with CDOM, respectively. In the vicinity of the river mouth, kCDOM reaches values 6.5 to 8 times higher than simulations without riverine CDOM forcing, highlighting the importance of including the riverine CDOM effect on light in the nearshore region. When using a linear relationship, kCDOM increases as CDOM concentration increases, triggering high values (>0.3 m⁻¹ with a maximum at 0.59 m⁻¹) in the direct vicinity of the river mouth, with a sharp transition to lower values further offshore (<0.2 m⁻¹) (Fig. b). When using the hyperbolic tangent relationship, kCDOM is capped to 0.26 m⁻¹, given the kCDOM/CDOM relationship fitted with in-situ observations (see Fig. c). As a result, CDOM attenuation is more evenly spread along the nearshore region (Fig. c).

We now examine how riverine CDOM influenced the physical conditions of the SBS during 2012, introducing a control simulation (Run_ctrl) that differs from Run_full by turning off both CDOM light attenuation (Sect. ) and its effect on seawater heating (Sect. ) (see Table ).

In the river plume, Run_full simulates a peak of surface CDOM concentration during the spring freshet, which coincides with the SLIP and the increase in surface-ocean shortwave downwelling irradiance (Fig. ). As a result, the subsurface shortwave irradiance (Isw3) – defined as the shortwave irradiance (W m⁻²) below the model surface layer (3 m depth) – decreases by 13.4 W m⁻² (40 %) on average during the SLIP (Fig. ) compared to the simulation without CDOM effects (Run_ctrl). CDOM light attenuation in the plume region then triggers an additional SST increase (ΔSST up to 1 °C), driving a decrease in sea ice cover by up to 5 % (Fig. ). We note a delay of 1 d in the DofR in Run_full compared to Run_ctrl (not shown), demonstrating the limited influence of riverine CDOM on sea ice phenology. Terrestrial CDOM has a maximum impact on the physical condition of the plume one week after the DofR, with a 45 % decrease in subsurface shortwave downwelling irradiance and an increase of by up to 1.68 °C (Fig. ). Finally, the impact of riverine CDOM gradually diminishes as the tracer becomes diluted in the open ocean during the IIFP.

Following the approach in Sect. , we analyze the influence of the kCDOM parameterization on the river plume's temperature by comparing the changes in SST simulated by Run_noriv, Run_lin, and Run_full, relative to Run_ctrl. We focus on the month of July, when CDOM has the greatest impact on SST in the Mackenzie River plume (Fig. ). In Run_noriv, the change in CDOM heating relative to Run_ctrl is solely attributed to marine CDOM produced by phytoplankton grazing and mortality. The spatially-averaged change in SST due to phytoplankton-generated CDOM, based on the improved CDOM-carbon loop connection (see section ), is 0.45 ± 0.09 °C. The specific contribution of riverine CDOM leads to increases of 84 % (0.83 ± 0.24 °C, Run_lin) and 144 % (1.10 ± 0.28 °C, Run_full) using the linear and hyperbolic tangent kCDOM/CDOM relationships, respectively. We note that the kCDOM relationship in Run_lin only considers a classic linear CDOM warming effect resulting from PAR attenuation, emphasizing the dominant role of UV-A in SST warming.

In the remainder of the study, we explore the specific effects of CDOM light attenuation and ocean heating on the coastal primary producers and the carbon cycle, focusing on the biological and solubility pump. From here, we only focus on three simulations: Run_full, Run_light, and Run_ctrl. The later two simulations deviate from Run_full by turning off aspects of the CDOM light absorption (see Table ): In Run_ctrl, we turn off both CDOM light attenuation (Sect. ) and its effect on seawater heating (Sect. ). In Run_light, we turn off only the CDOM heating effect (Sect. ) but include its effect on light attenuation. We then disentangle the individual impacts of light attenuation and their influence on ocean temperature over seasonal timescales.

Annual surface-ocean NPP integrated in the river plume region remains similar across simulations, whether including the influence of CDOM on light and temperature (Run_full) or not (Run_ctrl), yielding 0.10 and 0.13 Tg C yr⁻¹, respectively. However, a mean delay of 15 ± 3 (min: 9–max: 23) days occurs in the seasonal phytoplankton bloom, defined here as the day when Chl a reaches its peak value. The surface-ocean NPP maximum, initially occurring in the middle of SLIP, is delayed to DofR by the end of the sea ice melt season due to CDOM (Fig. a). Introducing both CDOM light and biophysical parameterizations (Run_full) results in a 85 % increase in the peak of NPP, with 78 % attributed to the change in CDOM/light interactions and 7 % to increasing SST. However, the production period – defined as the duration when NPP exceeds half of its maximum – decreases from 12 to 5 d, thereby explaining the similar annual NPP.

By early June, surface-ocean nutrient stocks are replenished through vertical mixing, advective transport, and remineralization that primarily occurred during winter – Note that the differences in May surface nitrate concentrations observed in Fig. a are related to changes in stock replenishment over the spin up period due to CDOM inclusion. High sea ice concentrations during most of the year result in light availability being primary limiting factor for phytoplankton growth, with temperature as a background limitation (See Appendix E). As the season progresses into SLIP the sea ice concentration decreases, leading to higher light penetration into upper-ocean waters. In Run_ctrl, this allows phytoplankton to utilize nutrients and initiates a bloom (Fig. S2a) that persists until the nitrate stock is entirely consumed and thus limits further phytoplankton growth. However, by early June, riverine CDOM (Run_light) drives additional light attenuation, counterbalancing the increased light penetration resulting from sea ice loss (see Fig. ), hence slowing down the bloom initiation and delaying it by roughly 2 weeks (see Figs.  and S2b). Consequently, phytoplankton bloom latter in the season until the nitrate stock is exhausted and again limits further growth. We find an east-west gradient in the maximum bloom day (Fig. c), correlated with the DofR (Fig. d). This supports our hypothesis that light attenuation from riverine CDOM export complements light attenuation from sea ice during the melting period and delays the seasonal phytoplankton bloom until the open-water period.

In the absence of CDOM effects (light and heating effect), Run_ctrl results in a net annual CO₂ sink of -11.40 Gg C yr⁻¹ within the plume region. Over seasonal timescales, the air-sea CO₂ exchange occurs from DofO to DofC, with four distinct phases (Fig. ). The following figures show the net air-sea CO₂ flux, integrated within the river plume region over the time period considered. The initial phase, starting from DofO and extending to one week after DofR, exhibits a substantial net CO₂ sink of -53.2 Gg C, which is attributed to phytoplankton growth (see Fig. ). Following this, the second phase, which spans two months at the onset of the inner ice-free period (IIFP), is marked by a significant net CO₂ outgassing of 137.9 Gg C – this results from the decline in phytoplankton abundance and heightened local concentrations of DIC/DOC from river discharge . Subsequently, a less-variable, one-month long phase follows, characterized by a delicate balance (air-sea CO₂ flux near 0 Gg C d⁻¹) that results in a moderate net uptake of -10.3 Gg C. The third phase, starting in early October and extending to one week after DofC, exhibits a strong net CO₂ sink of -99.2 Gg C. During this last phase, phytoplankton declines due to depleted nitrate levels and DIC/DOC concentrations return to background levels as river discharge diminishes.

Over seasonal timescales, substantial changes in the timing and patterns of air-sea CO₂ flux occur during the two initial phases due to the inclusion of CDOM effects. As a result of CDOM light attenuation, we observe a delay in phytoplankton activity from the first phase (prior to DofR) to the subsequent phase (Fig. ), leading to a 79 % reduction in simulated CO₂ uptake during phase 1 (+42.0 Gg C; Fig. ). Furthermore, the increase in SST due to CDOM is minimal during this period (Fig. ), resulting in a negligible impact on the net air-sea CO₂ flux (0.4 Gg C).

As the phytoplankton bloom simulated by Run_full peaks at the onset of the second phase, CDOM light attenuation reduces net CO₂ outgassing by 47.0 Gg C. However, the warming effect of SST counteracts the reduced CO₂ outgassing (caused by phytoplankton growth), driving a CO₂ outgassing of 19.8 Gg C during this period. Consequently, the net CO₂ outgassing for this period is reduced by 27.2 Gg C. Comparing the loss in CO₂ uptake on the first period (42.0 Gg C) and the gain in CO₂ uptake (-27.2 Gg C), the reduction in the CO₂ sink during the first period is 14.8 Gg C higher than the gain in the second period. Thus, changes in CO₂ fluxes during these two periods represent 80 % of the annual net loss in CO₂ sink. As a consequence, when including the CDOM bio-physical feedback (Run_full), the plume switches to a net annual CO₂ outgassing of 7.35 Gg C yr⁻¹. We show here that, despite the greater effect of light attenuation on the magnitude and sign of air-sea CO₂ flux, the temperature effect is the dominant contributor in the transition of the plume from a sink to a source of CO₂, as it dampens the increased CO₂ uptake due to phytoplankton growth in early summer.