Biogeochemical response of the Mediterranean Sea to the transient SRES–A2 climate change scenario

The Mediterranean region is a climate change hot-spot. Increasing greenhouse gas emissions are projected to lead to a substantial warming of the Mediterranean Sea as well as major changes in its circulation, but the subsequent effects of such changes on marine biogeochemistry are poorly understood. Here, our aim is to investigate how climate change will affect nutrient concentrations and biological productivity in the Mediterranean Sea. To do so, we perform transient simula5 tions with the coupled high resolution model NEMOMED8/PISCES using the high–emission IPCC SRES-A2 socio-economic scenario and corresponding Atlantic, Black Sea, and riverine nutrient inputs. Our results indicate that nitrate is accumulating in the Mediterranean Sea over the 21st century, while phosphorus shows no tendency. These contrasting changes result from an unbalanced nitrogen–to–phosphorus input from riverine discharge and fluxes via the Strait of Gibraltar, which lead 10 to an expansion of phosphorus–limited regions across the Mediterranean. In addition, phytoplankton net primary productivity is reduced by 10 % in the 2090s in comparison to the present state, with reductions of up to 50 % in some regions such as the Aegean Sea as a result of nutrient limitation and vertical stratification. We also perform sensitivity tests to separately study the effects of climate and biogeochemical input changes on the future state of the Mediterranean Sea. Our results show that 15 changes in nutrient supply from the Strait of Gibraltar and from rivers and circulation changes linked to climate change may have antagonistic or synergistic effects on nutrient concentrations and surface primary productivity. In some regions such as the Adriatic Sea, half of the biogeochemical changes simulated during the 21st century are linked with external changes in nutrient input while the other half are linked to climate change. This study is the first to simulate future transient climate change 20 effects on Mediterranean Sea biogeochemistry, but calls for further work to characterize effects from atmospheric deposition and to assess the various sources of uncertainty.


Introduction
The Mediterranean Sea is enclosed by three continents, and is surrounded by mountains, deserts, rivers, and industrialized cities. This evaporative basin is known as one of the most oligotrophic ma-25 rine environments in the world (Béthoux et al., 1998). Because of its high anthropogenic pressure and low biological productivity, this region is likely to be highly sensitive to future climate change impacts (Giorgi, 2006;Giorgi and Lionello, 2008).
Records of the past evolution of the Mediterranean circulation show that the Mediterranean has undergone abrupt changes in its circulation patterns over ancient times. In particular, high stratification 30 events, characterized by the preservation of organic matter in the sediments, known as sapropels, have been recorded several times through geological history. The most recent of such event occurred 10 000 years ago and lasted about 3 000 years. This accumulation of organic matter in the sediments is interpreted as the result of a strong stratification of the water column leading to suboxic deep layers (e.g. Rossignol-Strick et al., 1982;Rohling, 1991Rohling, , 1994Vadsaria et al., 2017). In more recent 35 times, abnormal winter conditions have led to changes in deep water formation, such as the Eastern Mediterranean Transient (EMT) event that occurred during the early nineties and had chemical impacts such as an increase in the Levantine basin salinity (see Theocharis et al., 1999;Lascaratos et al., 1999;Nittis et al., 2003;Velaoras and Lascaratos, 2010;Roether et al., 2014). Also, changes in the North Ionian Gyre circulation triggered the so-called Bimodal Oscillating System (BiOS) that 40 influenced phytoplankton bloom in the Ionian Sea by modifying water transport that led to modified nutrient distribution and altered local productivity (Civitarese et al., 2010). These events show that a semi-enclosed basin with short residence time of water (around 100 years, see Robinson et al., 2001) such as the Mediterranean Sea is highly sensitive to climate conditions and that perturbations of these conditions can modify the circulation, ultimately leading to changes in its biogeochemistry. 45 Future climate projections with high-emission scenarios for greenhouse gases simulate warming and reduced precipitation over the Mediterranean region (Giorgi, 2006;IPCC, 2012) leading to warmer and saltier seawater (Somot et al., 2006;Adloff et al., 2015). As a result of these changes, the Mediterranean thermohaline circulation (MTHC) may significantly change with a consistent weakening in the western basin and a less certain response in the eastern basin for such high emission 50 scenarios (Somot et al., 2006;Adloff et al., 2015). In all simulations under the A2 scenario, Adloff et al. (2015) find an increased stratification index in 2100. This increase will likely weaken the vertical mixing and may reduce nutrient supply to the upper layer of the Mediterranean, a supply that is essential for phytoplankton to bloom (d' Ortenzio and Ribera d'Alcalà, 2009;Herrmann et al., 2013;Auger et al., 2014). 55 Primary productivity in the ocean is influenced by its circulation and vertical mixing that brings available nutrients to phytoplankton (Harley et al., 2006). Changes in physical processes such as modification of vertical mixing can have dramatic effects on plankton community dynamics and ultimately on the productivity of the entire oceanic food web (Klein et al., 2003;Civitarese et al., in surface currents. The salinity increases by 0.5 (practical salinity units) on average across the basin.
These changes in hydrological characteristics generate substantial changes in the circulation and in particular the vertical mixing intensity. Under the A2 scenario, the Mediterranean basin is projected to become more stratified by 2100. Consequently, deep-water formation is generally reduced. 135 Here, the physical model NEMOMED8 is coupled to the biogeochemical model PISCES (Aumont and Bopp, 2006), already used for investigations in the Mediterranean basin (Richon et al., 2017(Richon et al., , 2018. This Monod-type model (Monod, 1958) has 24 biogeochemical compartments including 2 phytoplankton (nanophytoplankton and diatoms) and 2 zooplankton size classes (microzooplankton and mesozooplankton). Phytoplankton growth is limited by the external concentration of five 140 different nutrients: nitrate, ammonium, phosphate, silicic acid and iron. In this version of PISCES, elemental ratios of C:N:P in the organic matter are fixed to 122:16:1 following Takahashi et al. (1985). There is no explicit bacterial compartment but bacterial biomass is calculated using zooplankton biomass (see Aumont and Bopp, 2006, for details). Organic matter is divided in 2 forms: dissolved organic carbon (DOC) and particulate organic carbon. The biogeochemical model was run 145 in offline mode (see e.g. Palmieri et al., 2015): biogeochemical quantities are passive tracers, they are transported following an advection-diffusion equation using dynamical fields (velocities, mixing coefficients...) pre-calculated in a separate simulation with only the dynamical model NEMOMED8.

Boundary and initial physical and biogeochemical conditions
External nutrient supply for the biogeochemical model includes inputs from the Atlantic Ocean and 150 from Mediterranean rivers. We did not include submarine groundwater discharge and direct wastewater discharge as there is to date no climatology for these sources. Atlantic input is prescribed from water exchange through the Strait of Gibraltar in the NEMO circulation model along with the concentrations of biogeochemical tracers in the buffer zone. Nutrient concentrations in the buffer zone are prescribed from a global ocean climate projection using the A2 simulation values from IPSL-of the simulation in 2100. Seasonal variability coming from four of the largest rivers for Mediterranean and Black Sea (Rhône, Po, Ebro and Danube) is also included. According to Ludwig et al. (2010), the future trends in nutrient discharge from the major rivers of the Mediterranean stay within the interannual variability of the past 40 years. The "Order From Strength" scenario is based on hypotheses of very little efforts made towards mitigation of climate change. Moreover, Ludwig et al. 170 (2010) point out some substantial changes in the nutrient and water budget in specific regions. In particular, according to their scenario, the northern part of the Mediterranean has decreasing trends in nitrate discharge whereas it is increasing in the southeastern Levantine basin.
There is, to our knowledge, no transient scenario for the evolution of atmospheric deposition over the Mediterranean Sea. However, in order to evaluate the potential effects of aerosol deposition on 175 the future Mediterranean Sea, we used deposition fields of total nitrogen deposition (NO 3 + NH 4 ) from the global model LMDz-INCA (Hauglustaine et al., 2014) and phosphate deposition from natural dust modeled with the regional model ALADIN-Climat (Nabat et al., 2015a) respectively (see Richon et al., 2017, and references therein for the description and evaluation of the atmospheric models). The atmospheric deposition fields represent present-day aerosol deposition fluxes  2012 and 1980-2012 for total nitrogen and dust deposition respectively) that are repeated over the 1980-2099 simulation period.
Initial nutrient concentrations in the Mediterranean come from the SeaDataNet database (Schaap and Lowry, 2010) and initial nutrient concentrations in the buffer zone are prescribed from the World Ocean Atlas (WOA) (Locarnini et al., 2006). Salinity and temperature are initialized from 185 the MEDATLAS II climatology of Fichaut et al. (2003).
All simulations began from a restart of a historical run that started in January 1965 following a spin-up of more than 115 years made with a loop over the 1966-1981 period for the physical forcings and the river nutrient discharge. -up 190 All simulations were made for 120 years. The control run CTRL was made with forcing conditions corresponding to the 1966-1981 period looped over the simulation period. This period was chosen in order to avoid including in the CTRL the years with excessive warming such as the 1980s and 1990s (see Figure 12 in Adloff et al., 2015, for the surface temperature evolution from 1960). The scenario simulation is referred to as HIS/A2 as in Adloff et al. (2015). The HIS refers to the historical period 195 (in our case between 1980 and 1999), while A2 is the name of the 2000-2099 scenario simulation.

Simulation set
In order to separately quantify the effects of climate and biogeochemical changes, we made 2 additional control simulations: (1) CTRL_R with climatic and Atlantic conditions corresponding to present-day conditions (no scenario for climate change or nutrient fluxes through the Strait of Gibraltar) and river nutrient discharge following the scenario evolution, and (2) CTRL_RG with 200 present-day climatic conditions, but river nutrient discharge and Atlantic buffer-zone concentrations following the scenario conditions. We made two supplementary simulations, one with total nitrogen deposition (HIS/A2_N) and an-210 other with total nitrogen and natural dust deposition (HIS/A2_NALADIN). These simulations include climate change and nutrient fluxes from rivers and via the Strait of Gibraltar that follow the scenario conditions. The results from these simulations should be considered as exploratory.
Nonetheless, they provide insight into the potential effects of future aerosol deposition.

215
3.1 Evaluation of the NEMOMED8/PISCES model NEMOMED8 has already been used in a number of regional Mediterranean Sea modeling studies either in hindcast mode Herrmann et al., 2010;Sevault et al., 2014;Soto-Navarro et al., 2015;Dunić et al., 2016) or scenario mode . It produces the main characteristics of the Mediterranean Sea circulation. Evaluation of the HIS simulation provided in 220 Adloff et al. (2015) shows that the main physical characteristics of the Mediterranean are produced, in spite of a too cold upper layer (1°C colder than observations) and too little stratification in comparison to observations. In particular, the HIS simulation matches closely the observed thermohaline circulation in the Adriatic and Ionian basins (see Adloff et al., 2015).
The regional NEMOMED physical model has already been coupled to the biogeochemical model 225 PISCES on a 1/12°grid horizontal resolution (Palmieri et al., 2015;Richon et al., 2017Richon et al., , 2018, but no future climate simulation has yet been performed. As a first study coupling NEMOMED8 with PISCES, we compared the main biogeochemical features of our simulations with available data. Figure 1 shows the surface average chlorophyll concentrations in the top 10 meters of the CTRL and HIS/A2 simulations, and from satellite estimations from MyOcean Dataset (http://marine. 230 copernicus.eu/services-portfolio/access-to-products/?option=com_csw&view=details&product_id= OCEANCOLOUR_MED_CHL_L4_NRT_OBSERVATIONS_009_041). Whenever we refer to chlorophyll, we always mean chlorophyll-a (hereafter noted chl-a). The model correctly reproduces the main high-chl-a regions such as the Gulf of Lion and coastal areas. However, Figure 1 shows an underestimation of about 50 % of the surface chl-a concentrations by the model in these productive 235 areas. The west-to-east gradient of productivity is also reproduced by the model with values that agree with satellite estimates (approximately 50 % decrease in average chl-a concentration between the western and the eastern basin in the satellite data and 30 to 50 % in the model). Moreover, this Figure shows that chl-a produced by the CTRL is stable over time. The model fails, however, to reproduce the observed chl-a-rich areas in the Gulf of Gabes and at the mouth of the Nile. This 240 discrepancy is probably linked with insufficient simulated nutrient discharge from coastal runoff in these regions. Moreover, several studies (see e.g. Claustre et al., 2002;Morel and Gentili, 2009) show that satellite estimates have a systematic positive bias in the coastal regions because of the presence of particulate matter (for instance, sediments). The general bias observed in the Mediterranean is linked with organic matter and the presence of dust particles in seawater which cause light 245 back scattering. Figure 2 provides an evaluation of the average normalized chl-a surface concentration evolution over the entire basin for the period 1997-2005. This Figure shows that the normalized chl-a surface concentration in the model is close to the estimates provided by the SeaWiFs satellite data (Bosc et al., 2004). Even though the interannual variability of the model is 50 % smaller than in

Evolution of temperature and salinity
Average surface temperature and salinity (SST and SSS) evolution in the entire basin during the 270 CTRL and HIS/A2 simulations are shown in Figure 3, which confirms results from Adloff et al. (2015) and shows that the CTRL simulation is stable over time. Beyond this basin-wide average variation in SST and SSS, a more detailed analysis reveals much greater variability depending on the region (Somot et al., 2006;Adloff et al., 2015). For instance, the Balearic Sea is more sensitive to warming than the rest of the western basin, and the eastern basin has a more intense warming than 275 the western basin (up to 3°C warming in the eastern basin and in the Balearic Sea). Also, the surface salinity in the Aegean Sea increases more than the other regions.

Evolution of the nutrient budgets in the Mediterranean Sea
The nutrient budgets of the semi-enclosed Mediterranean basin are highly dependent on external 280 sources (e.g. Ludwig et al., 2009Ludwig et al., , 2010Huertas et al., 2012;Christodoulaki et al., 2013). We first input, river discharge and sedimentation. Nitrate can also accumulate in the Mediterranean waters through N 2 fixation by cyanobacteria, but this process accounts for less than 1 % of the total nitrate budget (Ibello et al., 2010;Bonnet et al., 2011;Yogev et al., 2011), and is neglected here.

290
In this Section, we refer to the period 1980-1999 as the beginning of the century, to the period 2030-2049 as the middle of the century and to the period 2080-2099 as the end of the century.

Fluxes of nutrients through the Strait of Gibraltar
The Mediterranean is connected to the global ocean by the narrow Strait of Gibraltar. Water masses transport through this strait contribute substantially to its water and nutrient budgets (e.g. Gómez,295 2003; Huertas et al., 2012). The Mediterranean is a remineralization basin that has net negative fluxes of inorganic nutrients (i.e. organic nutrients enter the basin through the Gibraltar Strait surface waters and inorganic nutrient leave the Mediterranean through the deep waters of the Gibraltar Strait Huertas et al., 2012). Figure 4 shows the evolution of incoming and outgoing nitrate and phosphate fluxes through the Strait of Gibraltar in the HIS/A2 and in the CTRL simulations. We observe similar In the HIS/A2 simulation, the incoming flux of nitrate decreases from 50 to 35 Gmol month −1 while 305 that for phosphate drops from 2.5 to 1.6 Gmol month −1 until the middle of the century despite a period of increased incoming fluxes of both these nutrients in the 1990s. After 2050, fluxes increase to reach values higher than the control in the last 25 years of simulations ( Figure 4). By 2100, incoming nutrient fluxes have increased in the A2 scenario simulation by 13 % (2080-2099 minus 1980-1999 periods). This increase is statistically significant, with a linear regression having a pos-

325
River discharge is the main external source of phosphate for the eastern part of the basin (Krom et al., 2004;Christodoulaki et al., 2013). Figure 5 shows the total discharge of phosphate and nitrate from rivers to the Mediterranean Sea.
Phosphate discharge decreases by 25 % between the beginning and the end of the simulation pe-330 riod. As suggested by Ludwig et al. (2010), phosphate discharge in the A2 period stays lower than in the HIS period, in spite of a small discharge enhancement between 2030 and 2049.

Sedimentation
Sedimentation removes nutrients from the Mediterranean Sea. In this version of PISCES, the loss of nitrogen and phosphorus to the sediment is calculated from the sinking of particulate organic carbon 340 (POC) to the sediment (linked through the Redfield ratio). Sediment fluxes of phosphorus and nitrogen during the simulations are shown in Figure 6.
These global nutrient budgets reveal that climate change and external nutrient fluxes to the Mediterranean can influence its nutrient content in different sometimes even in opposing directions. In particular, river inputs have large effects on nutrient content in the eastern basin, while input through 385 the Strait of Gibraltar has limited effects on the nutrient content even in the western basin.

Continuous evolution of phosphate and nitrate concentrations
In order to observe the continuous evolution of nutrient concentrations in different layers over the 21st century, we plotted the evolution of phosphate and nitrate concentrations for the entire simu-  In the eastern basin, the impacts of river discharges of nitrate seem to have large influence on the 445 nitrate accumulation as shown by the similar evolution of HIS/A2 and CTRL_R simulations (Figures 8d, 8e and 8f). Figure 8d shows the contrasted effects of climate and biogeochemical changes.
The strong difference between CTRL_R and CTRL concentrations at the beginning of the simulation (almost 0.4 mmol m −3 ) indicates that riverine nutrient discharge has a strong influence on surface nitrate concentrations in the eastern basin and is responsible for an important part of the eastern Mediterranean nitrate budget (see also Table 3). But the strong difference between CTRL_R and HIS/A2 at the end of the century indicates that vertical stratification leads to a decrease in surface layer nitrate concentrations, probably linked both with lower winter mixing and nutrient consumption by phytoplankton. In the intermediate and deep layers, the evolution of physical conditions has a similarly large impact on the nitrate concentrations in the eastern basin as shown by the difference 455 between CTRL_R and HIS/A2 (see also Table 3  On the contrary, Figure 10 shows that the surface phosphate concentration is decreasing over most of the Mediterranean basin except near the mouth of the Nile, the Ionian, Algerian, Tyrrhenian, between Crete and Cyprus and in the Alboran Sea. The specific concentrations observed next to the 480 Nile mouth are linked to an inversion of the N:P ratio in this river in our scenario (i.e. an increase in P discharge and a decrease in N discharge). The distribution of surface phosphate concentration at the end of the century (2080)(2081)(2082)(2083)(2084)(2085)(2086)(2087)(2088)(2089)(2090)(2091)(2092)(2093)(2094)(2095)(2096)(2097)(2098)(2099) shows that all P-rich areas of the eastern basin at the beginning of our simulations are depleted by the end of the simulation. For instance, the P-rich areas around Crete and Cyprus no longer exist in the 2080-2099 period ( Figure 10). Moreover, Figure 11 shows that these areas match zones of high productivity. All the most productive zones of the beginning of the century are reduced in size and intensity by the end of the century. For instance, there is a 10 to 40 % decrease in primary production in the Gulf of Lion and around the Balearic Islands, more than 50 % reduction in the North Adriatic basin, in the Aegean Sea and in the eastern Levantine basin around Cyprus. There is also a reduction in primary productivity from 40-50 gC m −2 year −1 to 20-30 Gibraltar. The large scale reduction of surface primary productivity may be a cause for the observed reduction in sedimentation (see Figure 6). 500 Figure 12 presents the limiting nutrient calculated using PISCES half-saturation coefficients (see Aumont and Bopp, 2006). The limiting nutrient is derived from the minimal value of limitation factors. In the Monod-type model PISCES, nutrient-based growth rates follow a Michaëlis-Menten 505 evolution with nutrient concentrations. In the present period, most of the productive areas are N and P colimited in the simulation (Figure 12). This includes regions such as the Gulf of Lion, the South Adriatic Sea, the Aegean Sea and the northern Levantine basin. Future accumulation of nitrogen in the basin modifies the nutrient balance causing most eastern Mediterranean surface waters to become P-limited. The total balance of phosphate is more negative in the future than in the present 510 period whereas we observe an inverse situation for nitrate. Therefore, phosphate becomes the major limiting nutrient in most of the regions where productivity is reduced such as the Aegean Sea, the northern Levantine basin and the north eastern Ionian Sea. The DCM depth changes little during the simulation, even though salinity and temperature change.
The DCM deepens slightly in some regions such as the North Ionian and the South of Crete. Al-520 though the DCM depth changes little in the future, the intensity of subsurface productivity is reduced (see Figure 11). In the oligotrophic Mediterranean, the majority of the chl-a is produced within the DCM. There is an 8.9 % reduction in integrated chl-a production between the 1980-1999 and 2080-2099 due to circulation changes combined with changes in fluxes through the Strait of Gibraltar and riverine 540 inputs. Table 4 reports total chl-a budgets in the 1980-1999, 2030-2049 and 2080-2099 periods of all the simulations in all Mediterranean subbasins (Figure 2 from Adloff et al., 2015). It reveals that chl-a budget is stable over the CTRL simulation but decreases in all Mediterranean subbasins over the HIS/A2 simulation. The decrease in chl-a is larger in the eastern regions, in particular in the Adriatic and Aegean Seas (-17 and -19 % respectively). In the western basin, the decline in chl-a is 545 smaller (-5.1 %). The chl-a budget is probably maintained by the enhanced nutrient fluxes through the Strait of Gibraltar (chl-a in CTRL_RG does not significantly decrease in the western basin).
About 85 % of the future reduction in chl-a in HIS/A2 is explained by the effects of climate change (HIS/A2 minus CTRL_RG). However, the effects from increased nutrient inputs through the Strait 550 of Gibraltar, decreased riverine phosphate inputs and increased riverine inputs of nitrate seem to have opposing effects to climate and circulation changes on chl-a production. In particular, in the western basin, reductions in riverine discharge of nutrients reduce chl-a by 3.5 % (see CTRL_R values), whereas changes in fluxes through the Strait of Gibraltar enhance chl-a (only 1 % decrease in chlorophyll concentration in CTRL_RG in the western basin). Altogether, the analysis of plankton biomass evolution during the simulation period suggests that primary and secondary production in the eastern basin are more sensitive to climate change than in the western basin. The eastern basin is more isolated from the open Atlantic Ocean than western 590 basin as it receives less nutrients from the Atlantic and from coastal inputs. The eastern basin is also deeper and less productive than the western basin (Crispi et al., 2001). The eastern basin exhibits a decline in the phytoplankton biomass that is similar to the decline in the phosphate concentration.
Biological production is mainly P-limited in this basin (see also Figure 12). Therefore, the constant low concentrations of phosphate observed throughout this century limit biological production and 595 keep plankton biomass at low levels. waters with surface waters becoming more sensitive to external nutrient sources (Figures 7 and 8).

Effects of aerosol deposition on surface primary productivity
On the other hand, Macias et al. (2015) found that primary productivity slightly increased as a result of decreased stratification in the climate change scenarios RCP 8.5 and RCP 4.5. The A2 scenario that we used was the only one available with 3-D daily forcings, as necessary for coupling with the PISCES biogeochemical model. However, Adloff et al. (2015) showed that other SRES scenarios such as the A1B or B1 may lead to a future decline in the vertical stratification with probably different consequences on the Mediterranean Sea biogeochemistry. Our study is thus only a first step

Uncertainties from the PISCES model
The evaluation of the CTRL simulation showed that NEMOMED8/PISCES is stable over time in spite of a slight drift in nitrate concentrations (see Figure 8). Nutrient concentrations in the intermediate and deep layers were underestimated in comparison to measurements (see Appendix). Nutrient In the version of PISCES used in this study, variations in nitrate and phosphate are linked by the Redfield ratio (Redfield et al., 1963). The Redfield hypothesis of a fixed nutrient ratio used for plankton growth and excretion holds true for most parts of the global ocean, but may not be true 665 for oligotrophic regions such as the Mediterranean Sea (e.g. Béthoux and Copin-Montégut, 1986).
Moreover, changes in nutrient balance influence the nutrient limitations as shown by Figure 12

External nutrient sources
Climate change may impact all drivers of biogeochemical cycles in the ocean. In the case of semienclosed seas like the Mediterranean, the biogeochemistry is heavily influenced by external sources of nutrients (namely rivers, Atlantic and atmospheric inputs, see Ludwig et al., 2009;Krom et al., 2010) and modification of the physical ocean (e.g. vertical mixing, horizontal advection, see San- Therefore, there is no incompatibility issue between for the forcing and model.

Riverine nutrient fluxes
Additionally, our CTRL_R simulation shows that the increase in riverine nitrate fluxes leads to the 690 accumulation of nitrate in the surface Mediterranean, in particular in the eastern basin and in the Adriatic. For the riverine nutrient inputs, scenarios from the MEA report are based on different assumptions from the IPCC SRES scenarios used to compute freshwater runoff in the HIS/A2 simulation. Freshwater discharge from Ludwig et al. (2010) is based on the SESAME model reconstruction and differs from freshwater runoff in the ARPEGE-Climate model used to force our physical model. This may lead to incoherences between water and nutrient discharges, but the nutrient discharges from Ludwig et al. (2010) are the only ones that are available. Furthermore, the SESAME model is not coupled with NEMO/PISCES. Associated discrepancies and the uncertainties linked with the use of inconsistent scenarios in our simulation should be addressed by developing a more integrated modelling framework to study the impacts of climate change on the Mediterranean Sea 700 biogeochemistry. As there is no consensus nor validated scenario for nutrient fluxes from riverine runoff in the Mediterranean, we chose to use one scenario from Ludwig et al. (2010). This scenario has the advantage of being derived from a coherent modeling framework. However, the Ludwig et al.
(2010) nutrient discharge transient scenario does not represent the interannual variability of nutrient runoff from rivers. Moreover, according to these authors, the socio-economic decisions made in the 705 21st century will influence nitrate and phosphate discharge over the Mediterranean. It is difficult to forecast these decisions and the resulting changes in nutrient fluxes are uncertain.

Potential effects of aerosol deposition
The biogeochemistry of the Mediterranean is significantly influenced by aerosol deposition (e.g. Krom et al., 2010;Dulac et al., 1989;Richon et al., 2018Richon et al., , 2017Guieu et al., 2014). The future 710 evolution of the multiple aerosol sources surrounding the Mediterranean will likely influence the response of the Mediterranean to climate change.
Results from the HIS/A2_NALADIN simulation show that enhanced phosphate fluxes from aerosols may limit the surface decrease of phosphate concentrations and limit phosphorus limitation. However, in the HIS/A2_NALADIN simulation, the surface Mediterranean is still P-limited in most of 715 the Mediterranean because the atmospheric nutrient fluxes are low in comparison to riverine nutrient fluxes from rivers and the nutrient flux through the Strait of Gibraltar (see Richon et al., 2017).
Therefore, it appears unlikely that changes in aerosol deposition from natural dust would greatly influence future Mediterranean biogeochemistry. However, there are multiple sources of aerosols that are not included in atmospheric models, e.g., anthropogenic, volcanic and volatile organic com-720 pounds (e.g. Wang et al., 2014;Kanakidou et al., 2016). Their combined influence could perhaps constitute an important nutrient flux to the Mediterranean, thus altering the evolution of its biogeochemistry. Moreover, aerosols affect radiative forcing over the Mediterranean and may impact the climate conditions (Nabat et al., 2015b). Thus, efforts should be made to accurately represent this nutrient source in Mediterranean models to assess the effect on Mediterranean Sea biogeochemistry 725 with regards to climate change.
Our results show that the state of the Mediterranean biogeochemistry at the end of the 21st century is the result of the combined evolutions of both climate and external nutrient fluxes. Therefore, it is very difficult to predict the future evolution of the Mediterranean based on the evolution of one of these components only. This is why it is important, in the case of semi-enclosed basins, to produce 730 reliable estimates of the evolution of all the components influencing the biogeochemistry.  In some parts of the eastern basin, the effects from riverine nutrient fluxes on chl-a appear more important than those from climate change (see Table 4). In the Adriatic Sea, Table 3 shows that 765 riverine nitrate discharge is responsible for 41 % increase in nitrate concentration over the simulation period (2080-2099 minus 1980-1999). In the CTRL_RG simulation, nitrate concentrations are similar to those in the CTRL_R simulation, indicating no influence of fluxes through the Strait of Gibraltar in the Adriatic Sea. Finally, nitrate concentrations in the HIS/A2 simulation are close to the CTRL_R values showing that most of the nitrate evolution in the Adriatic Sea is linked with river-770 ine discharge. Lazzari et al. (2014) also conclude that the river mouth regions are highly sensitive because the Mediterranean Sea is influenced by external nutrient inputs. The choice of river runoff scenario will likely influence the evolution of nutrient concentrations and the biogeochemistry in many coastal regions such as the Adriatic Sea (see also Spillman et al., 2007).

Climate versus biogeochemical changes effects
To our knowledge, this is the first attempt to study the basin-scale biogeochemical evolution using ranean. Both studies found that chlorophyll concentration and plankton biomass increase slightly due to changes in vertical stratification. In our simulations, average phytoplankton biomass decreases by 785 2 to 30 % (see Figure 15) and average zooplankton biomass decreases by 8 to 12 % (see Figure 16). However, our transient simulations revealed non linear trends in plankton biomass evolution as a results of the influence of external nutrient fluxes. Chust et al. (2014) have shown that regional seas and in particular the Aegean and Adriatic were sensitive to trophic amplification. Our results appear to agree, showing signs of trophic amplification (see Figures 15 and 16). Assessing the sensitivity of 790 the Mediterranean to trophic amplification would require more simulations focused on the evolution of Mediterranean planktonic biomass under different climate change scenarios.
The modifications of chl-a production and plankton biomass are linked to changes in nutrient limitation ( Figure 12). Finding no clear definition of nutrient co-limitation, we consider that N and P are co-limiting when the difference in limitation factors is less than 1 %. This definition of nutrient 795 co-limitation applies well to the Mediterranean case because of its very low nutrient concentrations.
Our results are confirmed by some studies (Thingstad et al., 2005;Tanaka et al., 2011). However, our nutrient limitations are calculated from 20-years average nutrient concentrations and nutrient limitation may vary greatly during the seasonal cycle (Marty et al., 2002;Diaz et al., 2001). It has also been hypothesized by Luna et al. (2012) that the warm temperature of the deep Mediterranean en-800 hance nutrient recycling via prokaryotic metabolism. Therefore, a part of the nutrient accumulation we observed may be linked with the increase in temperature.

Conclusion
This study aims at assessing the transient effects of climate and biogeochemical changes on the Mediterranean Sea biogeogeochemistry under the high-emission SRES A2 scenario. The NEMOMED8/PISCES 805 model adequately reproduces the main characteristics of the Mediterranean Sea: the west-to-east gradient in productivity, the main productive zones, and the presence of a DCM. Hence, it appears reasonable to use it to study the future evolution of the biogeochemistry of the Mediterranean basin in response to increasing atmospheric CO 2 and resulting climate change. Our study is the first to offer a continuous simulation over the entire period of the future IPCC scenario (A2), between 2000 810 and 2099.
Its results illustrate how future changes in physical and biogeochemical conditions, including warming, increased stratification, and changes in Atlantic and river inputs, can lead to a significant accumulation of nitrate and a decrease in biological productivity in the surface, thus affecting the entire Mediterranean ecosystem.

815
Our results also illustrate how climate change and nutrient inputs from riverine sources and fluxes through the Strait of Gibraltar have contrasting influences on the Mediterranean Sea productivity. In particular, the biogeochemistry in the western basin displays similar trends as that for nutrient input across the Strait of Gibraltar. Therefore, it appears critical to correctly represent the future variations of external biogeochemical forcings of the Mediterranean Sea as they may have an equally 820 important influence on surface biogeochemical cycles as does climate. The biogeochemistry of the eastern basin is more sensitive to vertical mixing and river inputs than is the western basin, which is regulated by input through the Strait of Gibraltar. Increased future stratification also reduces surface productivity in the eastern basin.
Although this study does account for the changes in fluxes through the Strait of Gibraltar and river-825 ine inputs, some potentially important sources are missing such as direct wastewater discharge, submarine groundwater input and atmospheric deposition. These additional nutrient sources are poorly known, with a general lack of both measurements and models as needed to build comprehensive datasets for past and future evolution of these nutrient sources. The HIS/A2_N and HIS/A2_NALADIN simulations presented in this study include continued present-day nitrogen and 830 phosphate deposition. Although these atmospheric fluxes have been evaluated previously and were shown to represent correctly the deposition fluxes, there is no guarantee that these fluxes will remain constant over the next century. Results indicate that the future sensitivity of the Mediterranean to atmospheric deposition depends on the surface nutrient limitation, which may in part be influenced by aerosol deposition. However, there is to our knowledge no available transient scenario for the 21st 835 century evolution of atmospheric deposition and no ensemble simulations to assess the future evolution of the Mediterranean Sea under different climate change scenarios. A new generation of fully coupled regional models have been developed and used to study aerosols climatic impacts (Nabat et al., 2015b). These models include a representation of the ocean, atmosphere, aerosols and rivers and could eventually be used to make consistent future climate projections at the regional scale of   rine Systems, 109-110, 78-93, doi:10.1016/j.jmarsys.2012.07.007, http://linkinghub.elsevier.com/retrieve/ pii/S0924796312001583, 2013. afficheN&cpsidt=4250659, 1994: Future projections of the surface heat and water budgets of the Mediterranean Sea in an ensemble of coupled atmosphere-ocean regional climate models, Climate Dynamics, 39, 1859-1884, doi:10.1007/s00382-011-1261-4, http://adsabs. 940 harvard.edu/abs/2012ClDy...39.1859D, 2012 Climate change projections using the IPSL-CM5 Earth System Model: from CMIP3 to CMIP5, Climate Dynamics, 40, 2123-2165, doi:10.1007/s00382-012-1636-1, https://link.springer.com/article/10.1007/s00382-012-1636-1, 2013: Atmospheric input of trace metals to the western Mediterranean: uncertainties in modelling dry deposition from cascade impactor data, Tellus B, 41, 362-378, http://onlinelibrary.wiley.com/doi/10.1111/j.1600-0889.1989.tb00315.x/full, 1989 Dense water formation and BiOS-induced vari-955 ability in the Adriatic Sea simulated using an ocean regional circulation model, Climate Dynamics, doi:10.1007Dynamics, doi:10. /s00382-016-3310-5, http://link.springer.com/10.1007Dynamics, doi:10. /s00382-016-3310-5, 2016.   1966-19811966-19811966-1981No No CTRL_R 1966-19811966-19811980-2099No No CTRL_RG 1966-19811980-20991980-2099No No HIS/A2 1980-20991980-20991980-2099No No HIS/A2_N 1980-20991980-20991980-20991997No HIS/A2_NALADIN 1980-20991980-20991980-209919971980