Seasonal and interannual variability of the pelagic ecosystem and of the organic carbon budget in the Rhodes Gyre (Eastern Mediterranean): influence of winter mixing

The Rhodes Gyre is a cyclonic persistent feature of the general circulation of the Levantine Basin in the eastern Mediterranean Sea. Although it is located in the most oligotrophic basin of the Mediterranean Sea, it is a relatively high primary production area due to strong winter nutrient supply associated with the formation of Levantine Intermediate Water. In this study, a 3D coupled hydrodynamic-biogeochemical model (SYMPHONIE/Eco3M-S) was used to characterize the seasonal and interannual variability of the Rhodes Gyre’s ecosystem and to estimate an annual organic carbon budget over the 2013-2020 period. Comparisons of model outputs with satellite data and compiled in situ data from cruises and BioGeoChemical-Argo floats revealed the ability of the model to reconstruct the main seasonal and spatial biogeochemical dynamics of the Levantine Basin. The model results indicated that during the winter mixing period, phytoplankton first progressively grow sustained by nutrient supply. Then, short episodes of convection driven by heat loss and wind events, favoring nutrient injections, organic carbon export and inducing light limitation on primary production, alternate with short episodes of phytoplankton growth. The estimate of the annual organic carbon budget indicated that the Rhodes Gyre is an autotrophic area with a positive net community production in the upper layer (0-150 m) amounting to 31.2 ± 6.9 g C m -2 year -1 . Net community production in the upper layer is almost balanced over the seven year period by physical transfers, (1) via downward export (16.8 ± 6.2 g C m -2 year -1 ) and (2) through lateral transport towards the surrounding regions (14.1 ± 2.1 g C m -2 year -1 ). The intermediate layer (150-400 m) also appears to be a source of organic carbon for the surrounding Levantine Sea (7.5 ± 2.8 g C m -2 year -1 ) mostly through the subduction of Levantine Intermediate Water following winter mixing. The Rhodes Gyre shows high interannual variability with enhanced primary production, net community production and exports during years marked by intense heat losses and deep mixed layers. However, annual

The statistics have been recalculated for the two distinct areas, the Rhodes Gyre and the surrounding Levantine Sea. We will modify the manuscript with the new statistics and to comment on the new figure: The model captures the seasonal dynamics of the observed satellite chlorophyll over the Levantine basin (Fig. 2a) and more particularly in the Rhodes Gyre (Fig. 2b). At the end of fall, the chlorophyll concentration begins to increase progressively and reaches its maximum in February/March, with higher maxima in the Rhodes Gyre compared to the surrounding Levantine Sea, in both the data and the model. The surface concentration is minimal in summer, for both the model and satellite. The model and satellite show differences in magnitude: in the model the winter maximum is generally higher, and the summer minimum values are lower, compared to the satellite data for both regions. The standard deviation (SD) of the model in the Levantine basin (0.04 mg Chl m -3 ) and the Rhodes Gyre (0.07 mg Chl m -3 ) is close to the mean chlorophyll concentration (0.05 mg Chl m -3 , 0.08 mg Chl m -3 respectively) which underlines the high variability of this oligotrophic system. The mean surface chlorophyll concentration in the satellite data for the Levantine basin and the Rhodes Gyre is 0.05 ± 0.02 and 0.06 ± 0.02 mg Chl m -3 . We obtain a highly significant correlation coefficient equal to 0.90 and 0.79 (p-value < 0.01), and low values for the NRMSD (an error of 23%, and 8%) and percent bias (-0.71, 12%), between model outputs and satellite data over the whole study period for the Levantine basin and the Rhodes Gyre, respectively.
Regarding the comparison with BGC-Argo float data in the Levantine Sea (Fig. 2c) and the Rhodes Gyre (Fig. 2d), the model reproduces correctly both the seasonal cycle and the magnitude of chlorophyll during the different periods of the year. Both model and float data show high variability in late winter/early spring in the Rhodes Gyre in agreement with previous studies (D'Ortenzio and Ribera d'Alcalà, 2009;Salgado-Hernanz et al., 2019;Kotta and Kitsiou, 2019) . The statistical metrics show a significant correlation equal to 0.73 0.66 (p-value <0.01) between the observed and modeled values in the Levantine Sea. The NRMSD is equal to 13% and the percent bias remains low (-136%). Similar statistical scores were obtained between the model outputs and the float data in the Rhodes Gyre, i.e. correlation (0.78, p-value <0.01) as well as low bias (-1923%) and NRMSD (15 8%). The difference between the comparisons of model results with satellite data and those with BGC-Argo float data could be attributed in part to an underestimation of satellite chlorophyll concentration during winter in the Levantine in winter of chlorophyll concentration in satellite data Sea as suggested by Vidussi et al. (2001) and reported by D' Ortenzio et al. (2021). ). This will be corrected in the revised manuscript.

L281 "NRMSD (23%)" I guess RMSD is normalized with data STD? Meaning that the error is 23% of the data STD. You could clarify this in the figure caption and/or text
L299 "The model reproduces the general features of the nitrate and phosphate concentration profiles with an increase from the surface to 500-1000 m and a gradual, low decrease below (Fig. 3)." Given that the model is initialized from observation profiles, isn't that expected? Perhaps it would be more meaningful if the comparison focuses on the surface layer, as with Chl. Also it is mentioned (section 2.1.3) that nutrients are initialized from summer CARIMED profiles, while summer nutrients are not included in Fig.3 Reply: First, we apologize for the lack of clarity concerning the initialisation of the biogeochemical simulation. For the initialization of the nutrients we used the observations gathered in the CARIMED database (Alvarez et al., 2019) and averaged them in each sub-region defined in Fig. S2. Generally we used the observations collected during summer periods before 2012, when the data were available, but in the Levantine Sea, where no summer data was available in the dataset we used spring observations before 2012. The manuscript will be modified as follows: "We initialized the biogeochemical model with observation profiles collected during stratified periods and averaged over 10 regions of the Mediterranean Sea (indicated in Fig. S2). For inorganic nutrients profiles, we used the CARIMED (CARbon, tracer and ancillary data In the MEDsea) database (Àlvarez et al., 2019; see Sect. 2.2.2), considering only summer data over the period 2011-2012, when data were available. Due to the lack of observations in summer for the Levantine region, we used spring observations. " We agree that it is expected that nutrient concentrations simulated in the deep layers over the period 2013-2020 are close to the nutrient concentrations imposed at the initialization. The assessment of the model results was performed with different in situ data than the ones used for initialization as recommended by Robson (2014) and Hispey et al. (2020). The comparisons with data over the period 2013-2020 show that the model is quite stable in the deep layers over the 9-year analysis period; the slight discrepancies between model results and observations could be explained by spatial variabilities.
In the revised manuscript, we will follow the suggestion of Reviewer 1, focus on the seasonal variations of the nutrient profiles in the surface and modify Figure 1.3 as follows: The model reproduces the general features of the nitrate and phosphate concentration profiles with an increase from the surface to 500-1000 m and a gradual, low decrease below, close to the profiles imposed at the initialization, showing a stability over the simulation period (Fig. 3) Reply: We agree with Reviewer 1, due to the high spatial variability for the nitrate in spring, the variable's concentrations between 500-1000 m seems to be located in the upper range values of observations. Therefore, we will remove "nitrate" from the sentence.
L306 "PERLE 1 and PERLE 2 phosphate observations show high variability, with a SD~0.065 and 0.062 mmol P m -3 respectively. " The Taylor diagram seems a bit confusing. Blue dots should normally be the data points (correlation=1), not the model.
Reply: Following Reviewer 1 suggestion, in the revised manuscript we will position the observations on the x axis with a correlation coefficient=1. We would like to point out that we will also change the centered RMSD into RMSD for simplicity. L370 "We display both nitrate and phosphate due to their role of limitation on primary production in this region (Moutin and Raimbault, 2002)." Not sure what you mean here. Isn't phosphate the ain limiting nutrient?
Reply: Most studies in the Levantine Sea advance a limitation of primary production by the availability in phosphate. A few studies (Zohary et al., 2005;Krom et al., 2005;Tanaka et al. 2011) have shown that the nutrient limitation in the Eastern Mediterranean is more complex and mention a co-limitation in nitrate and phosphate impacting the phytoplankton regime in this sea. It would be interesting to see what the model indicates on the nutrient limitation, but it appears for us outside the scope of this paper. We have chosen to show both nutrients to allow further studies and increase the possibility of comparisons to observations and (eventually) to discuss their seasonal evolution. Nevertheless, we will remove in the new version of the manuscript those few words that do not lead to a discussion in this paper.
L382 "Increases in plankton and DOC concentrations under the surface layer (0-150 m) are clearly visible during that period (Fig. 6).

"Not clear what you mean, as DOC follows a different pattern from phytoplankton (decreases during winter)
Reply: We agree with the comment of Reviewer 1, DOC and plankton follow different temporal evolution in the surface layer (0-100 m) in winter. The sentence here refers to the punctual increases of plankton and DOC concentrations below 150 m during winter as a consequence of vertical entrainment during the deepening of the mixed layer shown in Figure 6. We have realized that the sentence formulation is confusing and therefore we will be more precise in the description: The DOC concentration in the 0-100 m layer further decreases during the winter mixing period, from January to March ( Fig. 6e and S7f). One can also notice that the deepening of the mixed layer in winter is also responsible for the transfer of L386 "the surface nitrate concentration ranging between 0.3 and 1 mmol N m -3, in agreement with the observations of Yilmaz and Tugrul (1998)" Does this refer to winter period? After April this is <0.1mmol/m3. Please clarify in the text Reply: We agree with Reviewer 1, information on the season was missing in the manuscript. In the revised manuscript, we will specify that these values concern the winter period.
L422 "Lateral export is more important for POC, with values exceeding 10 mmol C m -2 d-1 over several months for some summer/fall, when the DOC lateral export shows little variation along the period (Fig. 7f). "Any explanation for this?" Reply: We would like to point out that we have found an error regarding the legend in Fig. 7f where POC and DOC were inverted. We apologize for this error that will be corrected in the revised manuscript in the text and the figure.
In the model, the lateral and vertical fluxes of DOC and water are significantly correlated (R=0.5 for vertical fluxes, R=0.86 for lateral fluxes). In particular, the DOC lateral export over several weeks in summer/fall periods, such as in 2016, is associated with water lateral inflow. Moreover the seasonal water mass budget in the surface layer of the Rhodes Gyre area shows an upward inflow at 150 m depth and a lateral outflow towards the surrounding areas when the cyclonic circulation intensifies in fall (Table 1.1). The higher lateral export of DOC related to POC in fall could be explained by the vertical distribution of current velocity, POC and DOC. Indeed the current velocity is generally higher near the surface where the DOC concentration is maximal, compared to 100-150 m where the POC is maximum. We will add this potential explanation in the revised manuscript as follows: Lateral export is more important for POC DOC, with values exceeding 10 mmol C m -2 d -1 over several months for some summer/fall, when the DOC POC lateral export shows little variation along the period (Fig. 7f). This could be explained by higher current velocities near the surface where the DOC concentration is maximal, compared to 100-150 m where the POC is maximum. The vertical export of total OC is reduced from spring onwards and becomes low (< 10 mmol C m -2 d -1 ) in summer and autumn, when DOC can be injected from the intermediate layer into the surface layer due to upwelling events.

L436 "A secondary peak of OC respiration is visible in fall when the maximum POC concentration is the deepest." Not totally clear. Please explain.
Reply: The time evolution of organic carbon content shows two peaks, the first one during spring followed by a second smaller one during fall. We divided the OC stock into DOC and POC stock and found the same secondary peak in autumn for both components as was previously simulated for the total OC stock. The secondary peak is explained by the general deepening of the ecosystem in summer/fall (Fig. 6 of the manuscript) due to the intensification of solar irradiance. The OC respiration follows the evolution of the OC stock. One can notice that since the depth of the DCM has been shown to be overestimated in the model, this peak might be also overestimated. We will modify the text as follow: A secondary peak of OC respiration is visible in fall when the maximum POC stock increases. This can be explained by the deepening of the ecosystem due to the increase in solar radiation at that period. However, the overestimation of the depth of the DCM shown in Section 3.1.2 suggests that it could be overestimated in the model results concentration is the deepest. In particular in the Levantine Sea, DOC concentration is minimum in the Rhodes Gyre. However, the surface layer DOC concentration is also influenced by the depth of the mixed layer that can be high in anticyclonic gyres, river inputs and/or inflow from the Ionian Sea. Therefore we will modify the sentence as follows: The surface values fall in the lower range of observations (41-100 mmol C m -3 ), which could be partly explained by the locations of the observations, mostly outside the Rhodes Gyre in more stratified and less productive regions where the concentrations are slightly lower compared to the surrounding Levantine Sea in the model outputs. L507 "Regarding the organic carbon biological fluxes, the seven year averaged annual NPP that amounts to 115 ± 15 g" Maybe you could add the mean NPP for the EM for comparison with other studies and also indicating the relatively higher magnitude in the RG area.
Reply: As the Eastern Mediterranean is composed of very various trophic regimes showing diverse characteristics in terms of river inputs and physical processes with highly productive regions such as the Adriatic Sea influenced by river inputs and ultra-oligotrophic regime such as the south Ionian and Levantine seas, we found more relevant in the revised manuscript to focus the comparisons of modeled NPP with previous estimates of NPP in the Levantine Sea.
However we specify here that we found a value of NPP of 119 ± 14 gC m -2 year -1 for the EM, that can be compared with estimates based on satellite data of 137 gC m -2 year -1 by Antoine et al. The model results show a higher magnitude of NPP in the RG area compared to the surrounding Levantine open sea: on an annual scale, 115 ± 15 versus 102 ± 17 gC m -2 year -1 , and in winter, 103 ± 20 versus 85 ± 17 gC m -2 year -1 . We will indicate it in the revised manuscript. Thus we will modify the paragraph as follows: Regarding the organic carbon biological fluxes, the seven year averaged annual NPP that amounts to 115 ± 15 g C m -2 year -1 falls in the range of the previous annual estimates for the northern Levantine Sea based on both the satellite ocean color data ( L515 "The mean annual POC export at 150 m depth is estimated in the model at 11.9 ± 3.4 g C m -2 year -1 ." This refers to the Rhodes gyre area? Please clarify in the text.
Reply: This has been added as suggested in the revised manuscript.
L540 "The intensification of the cyclonic circulation in fall favors the shallowing of the nutriclines." I think part of this shallowing might be also related to the solar radiation seasonal variability and the deepening of the DCM. Although this is partly illustrated based on the interannual variability, given that the study focuses on the Rhodes Gyre, it would be nice to show (maybe with a figure in the supplement) for comparison the difference from another area or an average over the Levantine where such shallowing of the nutriclines, related with the circulation does not occur.
Reply: We agree that it should be mentioned that this shallowing is first due to the solar radiation seasonal variability, and then that it could be reinforced by the circulation. To follow a recommendation of Reviewer 2 to lighten the discussion section we will remove this sentence from Sect. 4.1 but to take into account this comment we will change the text in Sect. 3.3: During fall, nutrient concentrations gradually increase in the surface layer with the weakening of the stratification and the gradual rise of the nutricline (defined here as isoline 1 mmol N m -3 for nitracline and 0.05 mmol P m -3 for phosphacline) up to the surface (Fig. 6a, 6b and S7) induced by the reduction of solar insolation and the shallowing of the DCM, possibly reinforced by the intensification of the cyclonic circulation.  L575 "On the other hand, the model PP relies at 30% on the uptake of nitrate, and at 70% on the uptake of ammonium (not shown). The former is significantly correlated with HL (R> 0.88, Fig. 10g) and MLD, whereas no correlation can be found between the latter and HL (R< 0.69, Fig. 10h) or winter mixing." This might be a bit misleading, as PP is generally well correlated with mixing/heat loss. I would suggest to rephrase Reply: We agree with this analysis. Contrary to what we wrote, Fig. 10 shows that the correlation of annual NPP with winter heat loss (HL) is not significant if 2013-14 is not considered. This result is in agreement with the lack of correlation of ammonium uptake (main contributor to NPP) with heat loss. Therefore, we will rewrite this section as follows: Although winter and spring NPP is higher under cold winters (not shown), annual NPP is not significantly correlated with heat loss if 2013-14 is not considered (Fig. 10e). The modeled NPP depends on the nitrate and ammonium uptake supporting, respectively, 30 and 70% of the NPP. Nitrate uptake is significantly correlated with HL (R= 0.88, p-value < 0.01, Fig. 10g) and MLD, whereas the ammonium uptake shows no significant correlation with HL (R= 0.69, p-value < 0.01, Fig. 10h).
Reply: We apologize for the lack of clarity in line 601. In the model results, we found that the date of the maximum chlorophyll concentrations was between early February and early March and, contrary to the NW Mediterranean deep convection region, doesn't depend on the winter severity. For example, it was found on 11 and 12 February during the mild winter 2015-16 and the severe winter 2016-17. This sentence, that will move in Sect. 3.3 following a suggestion of Reviewer 2, will be modified as follows: For example, it was found on 11 and 12 February during the mild winter 2015-16 and the severe winter 2016-17 (Table   S1).
Besides we will remove the line with the date of the MLD maximum since we will not present it in the revised manuscript.

highlighted interannual variability" the interannual variability
Reply: This will be corrected as suggested in the revised manuscript.
Reply: This sentence will be corrected in the revised manuscript to take into account this suggestion, as well as a comment of Reviewer 2, as follow: For that, we analyzed a simulation of a 3-D hydrodynamic-biogeochemical coupled model implemented over the Mediterranean Sea over the period from December 2013 -April 2021, and we focused on the Rhodes Gyre.
L146 "As for the Gibraltar Strait, a narrowing was conducted with a 1.3 km grid for a better representation of the exchange area between the Mediterranean Sea and Atlantic Ocean." Better rephrase e.g. ..Strait, the model resolution was further increased… Reply: This will be corrected as suggested in the revised manuscript.
L148 "and closer levels ranging near the surface." better rephrase e.g. and increased resolution near the surface Reply: This will be corrected as suggested in the revised manuscript.
Reply: This will be corrected as suggested in the revised manuscript.
L154 "monthly discharges were based on the study of Poulos et al. (1997)," you mean monthly climatology?
Reply: Indeed we meant monthly climatology. This will be corrected as suggested in the revised manuscript.
L156 "We used the daily 3D current velocity, temperature, salinity and vertical diffusivity outputs of the hydrodynamic simulations as forcing fields for the biogeochemical model run". Is salinity somehow involved in biogeochemical processes?
Reply: Salinity is involved in the calculations of air-sea oxygen fluxes through the oxygen solubility (Ulses et al., 2021) and in the carbonate system dynamics (Ulses et al., Biogeosciences Discussion, in revision). However these processes are not presented and discussed in this article.

L175 "concentrations of nutrients were imposed at subbasin scale" Not sure what you mean by subbasin
Reply: Ludwig et al. (2010) estimated the river nutrient inputs for the main rivers and 10 sub-regions (Alboran, South-Western, North-Western, Tyrrhenian, Adriatic, Ionian, Central, Aegean, North-Levantine, South-Levantine) of the Mediterranean Sea. In our simulation, we imposed the concentration of nutrients in the rivers of the same 10 sub-regions based on this study. We will change the sentence to clarify this point: At the river mouths, concentrations of nutrients were imposed at subbasin scale using the results dataset of Ludwig et al.
L211 "The internal variation of organic carbon inventory, biological term and lateral physical term were calculated online," Please rephrase "online" e.g. calculated from model output Reply: "online" meant that we computed the terms of the budget during the simulation at each time step and not after the simulation had run using instantaneous or mean model outputs. The online calculation allows an exact calculation of the various fluxes. To clarify this we will slightly modify the sentence: The internal variation of organic carbon inventory, biological term and lateral physical term were calculated online, i.e.
during the simulation, while the vertical term was calculated as the residual based on values of all other terms.
L259 "The hydrodynamical model was evaluated and validated ""evaluated" and "validated" appear quite similar Reply: We will remove "and validated" from the sentence in the revised version of the manuscript.
L292 "could be attributed in part to an underestimation in winter of chlorophyll concentration in satellite data in the Levantine Sea" rephrase to be more concise e.g. an underestimation of satellite chl concentration during winter in the Levantine...
Reply: This will be corrected in the revised manuscript.
L318 " In winter, the surface oxygen concentration is maximal coinciding with the peak of surface chlorophyll." Rephrase e.g As with Chl-a, the surface oxygen concentration is maximum during winter..
Reply: This will be corrected in the revised manuscript.

L386 "Phytoplankton accumulation" Not sure if accumulation is the wright word. Maybe phytoplankton growth?
Reply: This will be corrected in the revised manuscript.
L409 "the time series of the variation of the organic carbon inventory, of biogeochemical fluxes and of vertical and horizontal exchanges at the limits of the two boxes." Rephrase e.g. the variability of the organic carbon inventory, the biogeochemical fluxes and the vertical and horizontal exchanges at the limits of the two boxes.
Reply: This will be corrected in the revised manuscript.

L488 "We notice however an underestimation in the magnitude of the modeled maximum chlorophyll and dissolved oxygen concentration when comparing with both the BGC-Argo float and cruise data." You refer to the sub-surface maximum? Please clarify in the text
Reply: In the revised manuscript, we will add "subsurface" maximum for better clarification.

L561 "We found a significant correlation between nutrient injection and mean winter HL (heat loss)
or mean winter MLD (higher than 0.85)" repeated above. Please rephrase or merge.
Reply: To avoid repetitions we will remove the previous sentence

L700 "High interannual variability of annual" The high…
Reply: This will be corrected in the revised manuscript.

FigS2 "Red dots represent the river mouths." is repeated in the Fig.caption
Reply: This will be corrected in the revised manuscript. Reply: We appreciate this overall positive assessment.

The introduction gives rich background information, but the key research questions addressed by this study are not explicitly stated. I think the last paragraph of the intro (lines 112-116) could be reformulated to highlight the research gap that is addressed here and to make a short outline of the paper in a couple of sentences.
Reply: We will add two sentences on the gaps in the studies cited above (the detailed description of the ecosystem that biogeochemical floats do not allow, and the interannual variability that the intermittent bloom trophic status of the area suggests).
However, this study is limited to parameters measured by biogeochemical floats, which does not allow for a more detailed exploration of biogeochemical and ecosystem dynamics. On the other hand, the intermittent trophic status of the area suggests significant interannual variability that remains poorly understood. In order to fill these gaps, Tthe present study aims to gain insight into carbon dynamics through the examination of seasonal and interannual variabilities of the biogeochemical and physical fluxes of organic carbon, under particulate and dissolved forms, in the Rhodes Gyre and the estimate of an annual budget of organic carbon in the area over a multi-annual period.
Finally, we will add a short outline of the paper: For that, we analyzed a simulation of a 3-D hydrodynamic-biogeochemical coupled model implemented over the Mediterranean Sea, over the period from December 2013 to April 2021, and we focused on the Rhodes Gyre. The paper is organized as follows: first we describe in Sect. 2 the numerical models and the various data sets used to evaluate the model. Then, in Sect. 3 we present the assessment of the coupled model, the seasonal and interannual physical and biogeochemical variability and an annual budget of organic carbon. Results are discussed in Sect. 4 and conclusions are given in Sect. 5.
Methods are described in detail, yet, a few key information are missing. In particular, the time resolution of the model and the outputs are not described.
Reply: The time step is 20 minutes for advection and diffusion of biogeochemical variables and 2 hours for biogeochemical reactions. The model outputs are stored every 24 hours for 2D variables of fluxes and every 5 days for 3D variables. The outputs of the budget terms are stored every 3 hours and 20 min. As suggested by Reviewer 2, we will complete the paragraph with the time step of the biogeochemical model: The time step is 20 min for advection and diffusion of biogeochemical variables and 2 h for biogeochemical reactions.

Also, there is no mention of a model spin-up or of model drift (or absence of drift) in the biogeochemical variables.
Reply: As was mentioned in the manuscript, the biogeochemical model runs for the period from 15 August 2011 to 2 May 2021. In this study, we started analyzing the model's outputs from the end of 2013 (this has been added in Section 2.1.3) in order for it to reach a relatively stable nominal state. We considered the first two years as two years of spin-up for the model. We also checked for possible drifts in the biogeochemical variables. We show in Figure 2.1, the time evolution of two biogeochemical variables, the nitrate and the phytoplankton (the one of total organic carbon is discussed in a following reply), in the surface layer (0-150 m), the blue line represents the trend calculated with the first year 2013-14 whereas the red line is the trend excluding the first year. As was mentioned in the manuscript, winter 2013-14 is considered as an exceptional warm year and therefore might influence the trends. The trend (blue line) shows a gradual increase in the nitrate (+0.0061 mol N m -2 yr -1 ) and phytoplankton (+0.0005 C mol m -2 yr -1 ) for all the period, however when removing the first year (red line), the increased trend is not clearly visible. The red line slopes show values 2 and 5 times lower than what was previously observed (+0.002, +0.0001 mol m -2 yr -1 respectively). It should be noted that very little data exist and are available for the period 2013-2021 to compare our results and draw a firm behavior of the biogeochemical variables with time.

The paragraph in lines 141-150 relates to the model description and could go in 2.1.1.
We have chosen for the clarity of the article, to have two sections (2.1.1 and 2.1.2) that describe the hydrodynamic model and the biogeochemical model, respectively, (its state variables as well as the basic references that allow to go into more details of the equations) and then a third section (2.1.3) that gives the characteristics of the simulation carried out with this coupled model (the region, the grid). Lines 141-150 are specific to the simulation and are therefore in Section 2.1.3.

Lines 156-158: did you do a spin-up of the model? Did you check for drift in the biogeochemical variables?
Reply : We answered above to these questions. Gyre (Napolitano et al. 2000), while most 3D modeling studies investigated the whole Mediterranean Sea (Lazzari et al., 2012;Macias et al., 2014;Guyennon et al., 2016;Richon et al., 2017Richon et al., , 2018Karaloni et al., 2020;Cossarini et al., 2021) or eastern Mediterranean Sea (Petihakis et al., 2009) without focusing on the LIW formation region of the Rhodes Gyre." These sentences seem to imply that you are about to use a dedicated model of the Rhodes gyre. It's fine to use a model of the global basin, but I think it would be less misleading if the introduction and methods section mentioned explicitly your model domain. Maybe you could add a sentence at the end of the introduction saying that you are using a model for the global basin, as done in previous studies, but the originality is that you are using precise criteria for describing and analyzing the Rhodes gyre?

If I understand correctly, you used a model for the global Med basin and zoomed on the Rhodes gyres which you identified using the criteria you describe in 2.1.4? In the introduction, the need for dedicated models of the Rhodes Gyres is highlighted (lines 99-103): "On the other hand, only one 1-D coupled hydrodynamic-biogeochemical model has been carried out in the Rhodes
Reply: The originality of our study is the focus on the Rhodes Gyre whereas all the mentioned modeling studies described all the Mediterranean Sea and briefly mentioned our region of interest. As mentioned before, we will clarify this point, on one hand at the end of the introduction, and on the other hand in Section 2.1.3 of the methods. We will also add a sentence at the beginning of Section 2.1.4 "Definition of the study area…".
In this paper, the analysis of the simulation of the whole Mediterranean basin is restricted to the Rhodes Gyre.
The results are overall well described.

On Figure 7a, it looks like there is a drift in the OC inventory? Is that so? If yes, please discuss the reasons for it.
Reply: The annual budget of organic carbon shows positive and negative variations in the surface layer (Table 1), that are not linked to the magnitude of winter heat loss or vertical mixing. As Reviewer 2 has noticed, we find a global increase trend in the OC inventory (with and without considering the warm year 2013-14) equal to 0.44 mol C m -2 year -1 in the surface layer over the period 2013-2020 (Table 1), as mentioned in Sect. 3.5 (L463-464 of the submitted manuscript). This is in general agreement with the observations by Ozer et al. (2022). In their study, these authors found a general long positive trend for the depth-integrated chlorophyll a measured offshore Haifa, to the east of the Levantine Basin, between 2002 and 2021, superimposed by interannual variations. They suggested that the long-term warming and salinification result in an increased buoyancy and a shallowing of the LIW (up to 110 m) enabling a higher level of nutrients to become available to the photic zone from below, supporting the observed rise of the integrated chlorophyll a.
We want to emphasize here that the increase in organic matter content (Fig. 7a) is not so clear for winter values, which remain low for all cold winters. The possible trend would therefore rather concern the stratified period and would be independent of the winter conditions that are the focus of our paper. This was discussed in Section 4.2 when we observed that NPP and ammonium uptake are poorly correlated with winter heat loss and could be influenced by trends in temperature or nutricline depth. Furthermore, we believe that our time series is too short to determine a long-term trend. We will add a discussion on that point at the end of Section 4.1. Although figure 7 is described in great details, I think the opposite trends seen between 7c and 7d could be mentioned and discussed. I find interesting that the NCP and total transport seem to compensate for each other during the cold winters (see the peaks at~+60mmol/m2/d for NCP in 2015 that are outbalanced by the ~-60mmol/m2/d for total transport).
Reply: We agree with Reviewer 2. In the revised manuscript, we will mention this in sections 3.4, 3.5, 4.3 and 4.4 to take into account this suggestion.

In Section 3.4:
Thus the increase in total OC transport during cold winters seems to be counterbalanced by an increase in NCP. For instance in winter 2014-15 peaks reaching 60 mmol C m -2 d -1 are visible for both NCP and OC total transport.
A the end of Section 3.5: Finally, the excess of biological production during cold winters is almost entirely compensated by an excess in total OC export.

In Section 4.3:
This could explain the compensation between the excess in NCP and OC export during cold winters.
In Section 4.4: The high interannual variability of annual NCP (SD of 22%) in the Rhodes Gyre appears to be primarily linked to the intensity of winter atmospheric HL and vertical mixing (significant correlation > 0.88 between annual NCP and winter HL, Fig. 10f), which indicates an enhanced autotrophic metabolism during cold years, that is almost counterbalanced by an enhanced OC export.

The paragraph on lines 437-445 could go after line 426 in order to group the results regarding the surface.
Reply: We agree with Reviewer 2, therefore in the revised version we will group the results regarding the surface as suggested.
The discussion is overall well written and detailed, but some minor rearrangement could help making it easier to read. The introductory paragraph on lines 476-489 can probably be discarded. This would help streamline the article.
Reply: We will remove the introductory paragraph of the Section Discussion in the revised manuscript.
Lines 515-522: You explain that your modelled estimates of POC export are different to those measured and those from other models. Can you give a short explanation of what may cause these discrepancies and the potential implication for your results?
Reply: The observations to which we compare the model are, apart from that of Moutin and Raimbault (2002), all located in other regions of the Mediterranean. Some are stronger than ours and others are weaker. We believe that it is not possible with these too rare and short-lived observations to conclude that the model is biased and give implications for our results. We will modify this paragraph in the revised manuscript as follows: The mean annual POC export at 150 m depth in the Rhodes Gyre is estimated in the model at 11.9 ± 3.4 g C m -2 year -1 . The Lines 575-580: Do you think the NH4 uptake could be linked with atmospheric deposition? Could the influence of atmospheric deposition of NH4 explain the absence of correlation between NH4 uptake and HL? (i.e. maybe NH4 uptake actually correlates with deposition).
Reply: The ammonium atmospheric deposition was applied as a constant value over the whole simulation period. We will add this missing information in Section 2.1.3 of the revised manuscript.
To answer this question, we have run a new simulation in which we removed the ammonium atmospheric deposition. The uptake of ammonium is reduced by 1% and the correlation between winter heat loss and ammonium uptake is decreased by 0.2%, if ammonium deposition is neglected and thus remains insignificant.
There is a lot of information on Figure 11, but I'm wondering if all (or any) is necessary to the article. This figure focuses on 3 case studies over you time series and the long text associated is actually very descriptive. Maybe you could put this figure and the associated text (lines 582-607) in supplement and only keep in the main article a few sentences highlighting the key informations brought by those case studies.
Reply: We realize that Figure 11 shows a lot of information that might not be commented in the text, therefore we will remove the fluxes (OC export and NPP) and we will modify Figure 11 to only show 2 case studies representing the general behavior of cold and mild winter years, 2013-14 and 2014-15 (see Fig. 2.2 with the modifications in Figure 11 of the manuscript). In the revised manuscript, the description of Figure 11 (moved in result section 3.3 to follow a following comment) will be reduced: The interplay between vertical mixing, deep nutrient injection and increase in surface phytoplankton shows interannual variability as illustrated in Fig. 11. When the mixed layer punctually reaches the nutriclines in early winter or throughout a mild winter as in 2013-2014 (Fig. 11a), surface nutrients and chlorophyll increase gradually and nearly synchronously.
When the winter is severe as in 2014-2015 (Fig. 11b), gales expand the area in which MLD breaks through the nutriclines.
The surface nutrient response is each time a rapid increase (<1 day see for example early January and early February 2015) while the chlorophyll response depends on the depth of the MLD. When it is shallower than the euphotic layer, chlorophyll increases gradually (~12 days in January 2015). When the MLD exceeds the euphotic layer as in February 2015, chlorophyll development is delayed due to dilution of phytoplankton cells in the deep ML and light limitation for phytoplankton growth. Lines 609-629: this long text describing deep convection in other regions than the Rhodes gyre feels of place and can probably be discarded since they do not bring additional information.
Reply: We agree that this text is too theoretical for this paper and so we will remove it until Line 633. Some ideas from lines 625-633 will be introduced after the description of the export in the Rhodes Gyre in order to compare the processes in the deep and intermediate convection regions.
Similarly on lines 639-646: this paragraph feels like a description of the figures 7 and 10 and should therefore go in the results.
Reply: We have chosen to separate the results and the discussion. Figure 7, which presents the time series produced by the model, is clearly a result. The discussion concerns the interpretation of these results and as such, Figure 10, which illustrates the dependence of different biogeochemical fluxes on winter heat loss, has been introduced into the discussion to analyze the curves of Figure 7. We referred to Figure 7 to discuss the particularity of the year 2013-14 and how its consideration can alter the analysis. It seems to us therefore that this structure should not be modified. Figure 12 focuses on a specific event over the time series and introduces some confusion. I think this example and the text associated could go in supplement.
Reply: We realize that the description of these physical dispersion processes is outside the scope of this paper. They could be the subject of a specific article. We have preferred to delete this section and Figure 12 in the revised version of the manuscript. We will add some words to indicate that the lateral export corresponds to the subduction that follows the formation of the intermediate water: The annual lateral OC flux from the Rhodes Gyre to the Levantine Basin in the intermediate layer is clearly related to winter severity (Fig. 10c). It shows a correlation of 0.97 (p-value < 0.01) with OC vertical export at 150 m allowing to identify the responsibility of physical processes of LIW formation followed by subduction from the Rhodes Gyre to the Levantine basin.
Overall, maybe the discussion sections 4.3 and 4.4 can be merged and significantly streamlined. As it is presented, the text feels long and some paragraphs are very descriptive. Plus, the case studies of Figures 11 and 12 make it difficult to read the messages of the authors clearly. I think all text related with description of the figures should go either in the results section, or in supplement (for the text associated to Figs 11 and 12). The discussion should be limited to the key messages of the authors placing their results in the context of their research questions and the current state of knowledge regarding the carbon cycle in the Rhodes gyres.

modeled (a) phosphate (mmol P m -3 ), (b) nitrate (mmol N m -3 ), (c) phytoplankton (mmol C m -3 ), (d) zooplankton (mmol C m -3 ) and (d) dissolved organic carbon (mmol C m -3 ) concentrations averaged over the Rhodes Gyre, from December 2013 to January 2021. The black dotted line in (a) and (b) indicates the mixed layer depth. The red line represents the depth of the nitracline in (b), and the black one of the phosphacline in (a). The green dotted line in (c) indicates the deep chlorophyll maximum. C refers to cold winter years and M to mild winter years.
-Line 356 replace "smaller" with "lower".
Reply: This will be corrected in the revised manuscript.
-Line 377: "when the mixed layer depth increases" instead of "intensifies" Reply: This will be corrected in the revised manuscript.
-Lines 635-636: I am not sure I understand this sentence. Please rephrase.
Reply: In the revised manuscript, we will remove the introductory part of Section 4.3 and therefore remove the sentence in question.
-Line 714-715: Is there other references for DIC budgets than this unpublished work?* Reply: We will remove the reference to this unpublished work. We kept the other references.

Responses to the comments of Maurizio Ribera d'Alcalà
First we would like to warmly thank Maurizio Ribera d'Alacalà, the Reviewer 3 for his relevant and constructive comments which helped to improve the manuscript.
The paper presents the results of a modelistic study in the Easterm Mediterranean sea with a specific focus on the Rodhes gyre, a permanent cyclonic structure playing an important role in the Mediterranean intermediate water formation (LIW) and displaying higher level of phytoplankton biomass during late winter-early spring in respect to the rest of the basin, which is well known for its extreme oligotrophy. The main aim of the study is to quantify the contribution, in terms of carbon fluxes in different forms, of the gyre activity to the Eastern Mediterranean basin biogeochemistry, and to connect their variability to physical forcing and the linked nutrient fluxes from different sources, with a key role played by the vertical transport. Most of the relevant processes and fluxes are included in the model whose simulations are build on the GCM Symphonie, already calibrated for the whole basin, whose daily averaged outputs are used for assessing the transport of tracers which react according to the formulations of Eco3-M model. The latter is a classical biogeochemical flux model validated in many previous studies (see refs in the text). Results are compared with different data sets, both from in situ measurements and from satellite observations. The match between data and model outputs is definitely good (see below for further comments) especially considering the differences in time and space resolutions among the compared data.
As the authors rightly comment (l.491) the resolution of in situ data does not provide a convincing test for a model performance, unless some macropatterns are missed. To some extent one would be tempted to rely more on the calibrated model outputs than on the interpolations/extrapolations of in situ data to assess global or regional fluxes. This is, indeed, the main contribution of the study, which flanks other recent modelistic studies carried out on the basin (cited in the text), though with the specific focus on the Rhodes gyre. The main conclusions of the study are: 1. that there is a net lateral export of fixed carbon from the gyre toward to the Eastern basin, in other words the gyre 'feeds' the basin, with the export being more significant than the utilization within the structure; 2. that the heat fluxes that drive the mixed layer dynamics correlate well with the nutrient fluxes and the main biological responses, suggesting that they are the dominant driver of biotic response.
While the latter is quite expected for the bottom-up structure of the model and for what is already well established, the former is the first quantitative assessment of the contribution of the gyre activity to the biogeochemistry of the basin and is an interesting result considering the closeness of gyres in respect to their boundaries.
The paper is quite long, likewise the discussion which is more focused on the comparison of the model outputs with other estimates than on the few observed discrepancies between the model and the observations. From a conceptual point of view there is a sort of circularity in these modelistic studies. The results of model simulations are consistent with the basic oceanographic processes, that are known since decades, which are at the base of their formulations. e.g., the sequence nutrient transport to the photic zone-mixed layer dynamics-phytoplankton growth. Since they do not include all the possible processes, and with the caveat of the difficulty to compare the results with observations carried out often at drastically different scales, e.g., Bio-ARGO data with model cells, what would deserve further analyses should be the mismatches. Instead, the focus is on the ability of the model to get as close as possible to the reality, which is certainly useful for operational purposes but not for a better understanding of the role of the many processes which are not included in the formulations. I am aware that the conclusion could be that they play a minor role but this is not tested. I consider this paper worth to be published because in all sections there is a comprehensive discussion of the existing information on the processes that the model simulates and because it provides useful results on the biogeochemistry of the Eastern Mediterranean.
Reply: We appreciate this positive general assessment.
My suggestion to the authors for this study or for its follow-ups is to analyze and discuss the following aspects: -The delimitation of the Rhodes gyre domain is based on hydrographical and geometric properties. Did they consider to delimit the domain using dynamic data, i.e., velocity fields?
Reply: We did not think of this interesting possibility. The velocity field which is rather related to the sea surface height gradient is indeed commonly used to detect eddies and could certainly be applied to the Rhodes Gyre. We retain this possibility for other applications.
-How do they interpret two evident mismatches, the vertical winter chl.a profile ( fig.3 top left plot) and the systematic difference in the DCM depth ( Fig.3 upper plots), and the overestimate of the model of the Chl.a maxima (Fig.2a).
Reply: We agree with Reviewer 3 that the model clearly shows some discrepancies in chlorophyll magnitude and vertical distribution that have not been discussed. Regarding the winter, the model produces on average a mixed chlorophyll profile while the ARGO floats show a DCM around 50 m. To better understand this difference, we take the example of float 6901764 (Fig. 4) for which we see during the winters of 2015-16 and 2016-17 an alternation of mixed profiles and typical DCM-like profiles. If we consider the model in Fig. 4, we see that these alternations also exist but are less marked than in the observations (in particular when the Chl maximum deepens, the surface concentration does not decrease as much as in the observations) which probably explains the difference in the average profiles. Different hypotheses can be formulated to explain these alternations, which may correspond to small-scale spatial variability (of the order of a few kilometers for the dominant sub-mesoscale in winter) and/or temporal variability (wind intermittency). The study of this variability is clearly outside the scope of this paper. Nevertheless, it would be very interesting to study them in particular in order to understand if the model suffers rather from a lack of spatial resolution to reproduce the submesoscale or if the defect concerns rather the representation of the too slow phytoplankton response to the variability of physical forcing.
For the other seasons the systematic overestimation of the modeled DCM depth compared to float data, especially during summer and fall, could also be related to rates of biogeochemical processes in the model such as the remineralization, decomposition of particulate organic matter into dissolved organic matter, bacterial processes or settling velocity of detritus or micro-phytoplankton. This default could also be linked to the optical module that calculates the photosynthetically active radiation (PAR) available to support photosynthesis. Currently, water and phytoplankton concentrations are taken into account in the calculation, but not the concentrations of CDOM and detritus. Thus, sensitivity tests to biogeochemical rates and PAR calculation have to be performed to further improve the modeling of the chlorophyll profiles in our model.
As for the modeled surface chlorophyll a, Fig. 2 shows that during the mixing period, the model overestimates the maxima compared to the satellite data whereas it represents their magnitudes well when compared to the float data. We mentioned in the submitted manuscript, a potential explanation: the underestimation of satellite chlorophyll concentration during winter as suggested by Vidussi et al. Nevertheless, our validation shows that the main biogeochemical characteristics of the gyre such as the occurrence, shape and variability of the deep chlorophyll maximum (DCM), low nutrient concentration at the surface are well simulated. Thus, we believe that our modeling tool is at the state of the art and is fairly good to be used for the analysis of organic carbon in the Rhodes Gyre.
Concerning the question about the misrepresented DCM we add a comment in Section 4.1 (Model skill assessment): The model also produces chlorophyll maxima that are too deep in summer and profiles that are too mixed in winter.
Concerning the first point, further studies will be necessary to improve the model parameterizations (optical model, POC degradation processes, particle sinking). Concerning the second point, in winter, mixed chlorophyll and DCM-like profiles alternate indicating small scale (few kilometers) or temporal variability related to meteorological conditions. The study of the physical processes driving this variability and their impact on phytoplankton deserves a dedicated study of physical and biogeochemical Argo profiles and probably a higher spatial resolution modeling.
-The model produces grazers maxima at the level of DCM, which is not what is generally observed, especially considering that DCM phyto belong to the small size classes. Could they discuss this results analyzing which components of phyto and zoo are in those maxima?
Reply: In the model results, the zooplankton concentration and grazing rates are maximum~15 m above the DCM. Regarding the phytoplankton, we found that, in summer, the maximum of pico-and nano-phytoplankton is, respectively, 25 m and 15 m above the depth of the DCM, while the micro-phytoplankton maximum is 15 m below the DCM. While the nano-and micro-planktons growths are concentrated in the lower part of the surface layer, the pico-phytoplankton develops over the whole surface layer and shows a less pronounced maximum than the larger phytoplanktons. Over the surface layer (0-200 m) of the Rhodes Gyre, the pico-, nano-and micro-phytoplanktons represent, respectively, 46%, 33% and 21% of the total phytoplankton biomass in average over the year. This distribution varies little throughout the year.
Regarding the zooplanktons, their maximum concentrations in summer are located between 15 m and 50 m above the DCM (15 m, 30 m and 50 m for micro-, meso-and nano-zooplankton, respectively). The three zooplankton size-classes develop over the whole surface layer.
In previous modeling studies on the NW Mediterranean Sea based on the same coupled model, comparisons of modeled phytoplankton composition and zooplankton concentration with the data at the DYFAMED site were performed and showed that the model satisfactorily reproduced the main features in magnitude and variability, except the seasonal variation of meso-plankton that appears too smooth in the model (Auger et al., 2014;Ulses et al., 2016). In the open Levantine Sea and in particular in the Rhodes Gyre region, there are very few observation and modeling studies on consistent composition and vertical distribution of phytoplankton and zooplankton. For the present simulation the composition of phytoplankton over the Levantine Sea shows a domination of small phytoplanktons on average over the year and in winter (58%, 30% and 12% of the total phytoplankton for, respectively, pico-, nano-and micro-phytoplanktons). Based on in situ measurements, Vidussi et al. (2001) found that pico-, nano-and micro-phytoplanktons contribute in winter, respectively, to 27%, 62% and 12% of the total biomass. Besides, these authors showed a spatial variability of phytoplankton composition linked to meso-scale structures, with a higher contribution of micro-phytoplankton (26%) and lower contribution of pico and nano-planktons (15% and 59%) in cyclonic gyres. The spatial variability in the composition of phytoplankton is well reproduced in the model (Fig. 3.1) and we obtain a similar contribution for micro-phytoplankton but we found a higher contribution for pico-phytoplankton and the opposite for nano-plankton. On the other hand, the modeled contribution of pico-phytoplankton in the South Aegean Sea in March and September (between 44% and 52%) is in the lower range of the observations of Ignatiades et al. (2002)