The Catalan Sea is the part of the Balearic Sea situated between the northeastern part of the Iberian Peninsula and the Balearic Archipelago (Fig. 1). It is bounded by the Gulf of Lion and the Ligurian Sea in the north and the Balearic Sea in the south. The Catalan Sea is a 400-km-long sub-basin that reaches a maximum depth of 2500 m. Its general circulation is cyclonic, with a central dense water mass surrounded by lighter continental and Atlantic surface waters (Font et al. 1988). Between the central and peripheral waters there are two permanent fronts, the Catalan and the Balearic fronts, which follow the continental shelf-break of the Iberian Peninsula and the Balearic Islands, respectively. The Catalan front is caused by differences in salinity (S) between continental and central waters, while the Balearic front is caused by differences in temperature (T) and S between Atlantic and central waters (Font et al. 1988). The Catalan Current, associated with the Catalan front, is an extension of the Liguro-Provençal Current, also known as the Northern Current. It occupies the top 300-400 m of the water column, flowing towards the southeast following the 1000 isobath. In front of the Gulf of Valencia, part of this current creates a cyclonic gyre, incorporates Atlantic waters and continues flowing towards the northwest, following the Balearic shelf-break, which generates the Balearic Current (Millot 1987, Font et al. 1988, Salat 1995). The Catalan Current, with average velocities of 20-30 cm s^–1, has a marked seasonality, its intensity being maximum during autumn and decreasing towards a minimum during summer (Font et al. 1988, Send et al. 1999).

The northernmost continental shelf of the Catalan Sea is narrow (less than 25 km wide) and irregular due to the presence of submarine canyons, most of which finish in submarine canals (Canals et al. 1982). Submarine canyons and other topographic features, such as the expansion of the continental shelf in front of the Ebre Delta (Fig. 1), lead to deviations of the Catalan Current, creating eddies and intrusions (filaments) of denser and saltier water onto the continental shelf (e.g. Font et al. 1988, Masó et al. 1990, Granata et al. 2004).

In the Catalan Sea four water masses can be found: Surface Water, Winter Intermediate Water (WIW), Levantine Intermediate Water (LIW) and WMDW. Three different surface water masses can be distinguished by their S and location (Salat and Cruzado 1981, Font et al. 1988; Salat 1995): (1) surface water with continental influence is located close to the peninsular coast, containing water from the Rhône and Ebre Rivers, which results in an S of between 35.5 and 37; (2) surface water with Atlantic influence is located close to the Balearic Islands and its S increases from 36.5 to around 37-37.3 as a result of progressive evaporation from its entrance to the Mediterranean Sea; and (3) resident surface waters (also known as “Mediterranean Waters” or “Modified Atlantic Waters”) are located the centre of the basin and have the highest S (37.8-38) as a result of evaporation during the time spent in the Mediterranean basin.

Below the surface waters, two types of intermediate waters can be found. WIW is formed during winter by convection of surface waters (from both the Gulf of Lion and the Catalan Sea) on the continental shelf, without mixing with deeper waters. It is characterized by a minimum potential temperature of 12 to 12.5°C and an S of 38.1 to 38.3. The LIW is located between 200 and 600 m, and is the result of the mixing of Levantine water formed in winter in the Aegean Sea and Cyprus with winter water formed in the Adriatic Sea. It is the saltiest water, with a maximum S of 38.55 and a maximum potential T of 13.5°C. In cases of coexistence, WIW is found above LIW (Salat and Font 1987). Finally, the WMDW (MEDOC Group 1970) is found below the LIW down to the bottom of the basin. It has a potential temperature of 12.65°C to 12.9°C, an S of 38.45 to 38.48 and a density of 29.08 to 29.11 kg m^–3.

The biogeochemical data used in this study included published historical data, biogeochemical datasets of cruises in which the authors have participated, and data extracted from MEDAR and MEDATLAS databases (0-4.5°E, 38.7-42.5°N) (Table S1). Apart from T, S and depth, each cruise included some or all of the following biogeochemical variables: dissolved inorganic nutrients (phosphate, nitrate+nitrite [hereinafter nitrate], nitrite, ammonium and silicate), dissolved oxygen and Chl a concentration. The list of biogeochemical variables available per cruise gathered, as well as the source of the data, is shown in Supplementary Material Table S1. The final dataset was carefully checked to avoid any duplicated data. The whole pool of data was quality controlled using flags: data within the ranges of concentration described in Segura-Noguera et al. (2011) were flagged “0”, data above the range were flagged “1”, and data below the range were flagged “2”. In the present study, only data with flag “0” (Table 1) were used to describe depth profiles and to characterize water masses. Sampling stations for each oceanographic variable are shown in Supplementary Material, Figure S1. The main features of the analyses of dissolved inorganic nutrient, dissolved oxygen and Chl a of the dataset, when available, are exhaustively described in Segura-Noguera et al. (2011) and summarized below.

Four different autoanalysers were used for nutrient analysis (Technicon, Skalar, Evolution-II of Alliance and AA3 of Bran+Luebbe) following similar methodologies (Hansen and Koroleff 1999). A salicylate-hypochlorite method was used for ammonium analysis between 2001 and 2004 (Bower and Holm-Hansen 1980). Until 1999, samples were usually analysed on board, and after that year they were preserved by freezing and analysed on land. Nutrient standards were always prepared at room temperature (20-25°C). The analysis precision of each biogeochemical variable was estimated for several cruises using different methods (Segura-Noguera et al. 2011). The detection limit for the whole dataset was estimated as three times the average of the analysis precision calculated after removing the outliers with the interquartile method. The resulting detection limits were 0.02 µM for phosphate, 0.01 µM for nitrite, 0.08 µM for nitrate, 0.07 µM for ammonium and 0.12 µM for silicate. Detection limits of the 2001-2004 cruises “Canyons” I to IV and “Caco” 1 to 4 were calculated during the present study, and were 0.006 µM for phosphate, 0.004 µM for nitrite, 0.015 µM for nitrate, 0.020 µM for ammonium and 0.017 µM for silicate. Nutrient concentrations are given in µM instead of µmol kg^–1 because data of salinity and temperature were not available for all the corresponding nutrient concentrations. Consequently, the conversion to µmol kg^–1 would represent a data loss of around 6% of inorganic nutrient and up to 11.2% of ammonium, a very scarce nutrient in our dataset. Except for very few stations close to the Ebre Delta with a freshwater signature, the σ-T range in our dataset is 13.7 to 29.4 kg m^–3. This difference means that the values in µM are 2.9% larger, on average, than those in µmol kg^–1, and that the maximum difference between the two units (corresponding to the largest concentrations of these variables in the Catalan Sea) are around 0.27 of nitrate, 0.30 of silicate, 0.07 of ammonium, 0.02 of nitrite, 0.01 of phosphate and 8.9 of dissolved oxygen.

Throughout the years, dissolved oxygen was consistently measured following Winkler (1988). The difference between cruises was the method for determining the final titration point, which was visual or automatic. Dissolved oxygen data were converted to µΜ using the molar volume of oxygen gas (22.3916 L mol^–1) and the density of water at zero pressure calculated with SEABIRD software from CTD temperature (°C, converted to potential temperature with the same software) and salinity (psu). The detection limit of dissolved oxygen of the whole dataset was 7.7 µM. All our dissolved oxygen values were above detection limit.

The apparent oxygen utilization (AOU, µM) was calculated as the difference between dissolved oxygen in saturation ([O₂]*) calculated from in situ T and S following Owens and Millard (1985), and the measured dissolved oxygen in the water ([O₂]_meas).

Chl a data were determined fluorometrically after filtering with glass fibre filters and extracting with acetone (Yentsch and Menzel 1963). The average precision of Chl a analysis of all cruises was 0.01 µg l^–1 (Segura-Noguera et al. 2011).

Depth profiles have been built up for dissolved nutrients, Chl a and dissolved oxygen by calculating the median value at each standard level depth (SLD). SLDs are the depths that have been historically most sampled, and include data comprised between 25% of the distance with the upper SLD and 75% distance with the lower SLD. We used the SLDs proposed for the Mediterranean Sea by Fichaut et al. (1997): 0, 5, 10, 20, 30, 40, 50, 60, 80, 100, 120, 160, 200, 250, 300, 400, 500, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000 and 2500 m. Continental shelf and open sea profiles were differentiated because the area closest to the coast (bottom depth up to 200 m, as in Fichaut et al. 1997) is theoretically influenced by coastal and anthropogenic processes that could presumably increase the concentration of the different oceanographic variables. The data were grouped according to the stratification index estimated for each station and profile depth, as described in Segura-Noguera et al. (2011). The most representative seasons for each stratification index are winter for “1”, spring for “2”, summer for “3” and autumn for “4”.

Depth median concentration profiles of each variable, as well as stoichiometry profiles, were created for each stratification index. Depth-integrated (0–100 m) T, S, density, as well as nutrients, Chl a and dissolved oxygen concentrations, were calculated for surface waters. Seasonal differences in depth-integrated concentrations in surface waters were studied with the PAST software v. 3.06 (Hammer et al. 2001). A univariate non-parametric Kruskal-Wallis test was used to find differences, followed by a post-hoc Mann-Whitney pairwise comparison with Bonferroni corrected p values. Statistical significance was accepted when p≤0.05.

The molar stoichiometry between nutrients, as well as between nutrients and dissolved oxygen, were calculated for each available pair of data. Data below the estimated detection limits were not used to calculate the stoichiometry. Element:element correlations were studied using Standard Major Axis (SMA), a model II regression that takes into account differences in the scale of the axes (e.g. N:P relationship). The SMATR software v. 2.0 (Falster et al. 2006) was used to calculate SMA slopes and intercepts, p values and confidence intervals, and seasonal differences in slopes and intercepts.

Figures 2, 3 and 4 show the seasonal depth profiles of T, S, density anomaly and biogeochemical variables. The concentration of each variable at each SLD and season can be obtained from Supplemetary Material, Table S2 to S11, and depth-integrated (0-100 m) concentrations are shown in Table 2. A clear seasonality in temperature is found in surface waters (Fig. 2, Table 2). Warming of surface waters in spring outsets the stratification of the water column and the development of a mixed layer. In summer, the mixed layer depth is minimum and a strong thermocline can be observed. The stratification weakens in autumn and the thermocline disappears in winter (Fig. 2). Surface nutrients and Chl a concentration follow the general pattern of thermal stratification, with larger concentrations in winter and minimum surface concentrations during maximal stratification of the water column (Figs 3 and 4).

Phosphate, nitrate and silicate show the typical nutrient depth profile, with low surface concentrations that increase with depth (Fig. 3). At open sea stations, the largest surface (0-50 m) phosphate, nitrate and silicate concentration is found in winter (0.11, 1.07 and 1.98 µM, respectively), and the concentrations decrease throughout spring to reach minimum values in summer (0.04, 0.11 and 0.63 µM, respectively). In autumn, nitrate and phosphate concentrations increased, while silicate concentration remained low. Depth profiles from continental shelf stations show a similar pattern, except that summer phosphate concentration is similar to the spring concentration (0.06 µM) and nitrate concentration in winter is higher than in the open sea (1.70 µM). In winter mixed water columns, nutrient concentration increases linearly from the surface to 400 m depth at different rates (phosphate 0.7 nM m^–1, r²=0.92; nitrate 19.4 nM m^–1, r²=0.98; silicate 12.7 nM m^–1, r²=0.95). In spring, a nitracline and a phosphocline start to develop in the open sea at 60 and 100 m depth, respectively. The nitracline is deeper in summer and autumn (80 m, Fig. 3). On the continental shelf, the phosphocline is smoother than in the open sea. Phosphate and nitrate concentrations continue to increase with depth until 500 m. In spring, a maximum concentration of phosphate (0.44 µM) can be observed at this depth, but not of nitrate, whose concentration is around 8.6 µM from 500 m to the bottom of the basin. Silicate concentration continues to increase below 500 m depth, and concentrations around 8.8 to 9.2 µM are found below 1200 to 1500 m depth. The average (± standard deviation) nutrients concentration in deep waters (WMDW with depth ≥800 m) are 0.40±0.05 µM of phosphate, 8.68±0.04 µM of nitrate and 8.72±0.91 µM of silicate.

Ammonium concentration in the open sea (0-200 m depth) is higher in winter, decreasing throughout spring to a minimum in summer (around 0.6, 0.2 and 0.1 µM, respectively) (Fig. 4). In autumn, ammonium concentration increases again and is higher than that in spring (around 0.4 µM). At continental shelf stations this seasonality cannot be clearly seen, and the largest concentrations are found in winter below 30 m depth. It is difficult to detect clear features below 200 m depth because of the scarcity of data. The average concentration below 250 m depth is around 0.64±0.34 µM.

Surface Chl a, nitrite and dissolved oxygen concentrations show a similar pattern: highest concentrations in winter, lowest in summer, and a presence of a maximum (Fig. 4). In winter, a high Chl a concentration is found between 0-60 m (around 0.75 µg l^–1 in the open sea and 0.61 µg l^–1 on the continental shelf), and two maximums can be observed, one above 20 m (0.86 and 0.75 µg l^–1 in the open sea and on the continental shelf, respectively) and another at 50 m (0.83 and 0.68 µg l^–1, respectively). The maximum at 50 m is coincident with a maximum of ammonium and of nitrite, although only in the open sea (1.21 and 0.23 µM, respectively). Surface Chl a, nitrite and dissolved oxygen concentrations decrease in spring. Chl a and nitrite accumulate between 40 and 100 m depth, forming the DCM around 50 to 60 m depth (0.58 and 0.49 µg l^–1 in the open sea and on the continental shelf, respectively) and PNM around 60 to 80 m depth (0.11 and 0.16 µM, in the open sea and on the continental shelf, respectively), while dissolved oxygen forms a shallower maximum around 30 to 40 m depth (261.7 and 262.6 µM in the open sea and on the continental shelf, respectively). In summer, the DCM is larger and deeper than in spring (0.61 µg l^–1 at 80 m in the open sea, and 0.62 µg l^–1 at 60 m on the continental shelf), as is the dissolved oxygen maximum (272.9 µM at 40 m in the open sea, and 268.4 µM at 50 m on the continental shelf). By contrast, the PNM is lower (0.11 and 0.08 µM in the open sea and on the continental shelf, respectively) and only deeper in the open sea (80 m). In autumn, surface Chl a and nitrite increase and are similar to spring values, while dissolved oxygen is lower than in spring. The DCM is shallower (around 50 m) and lower than during the rest of the year (0.38 and around 0.3 µg l^–1 in the open sea and on the continental shelf, respectively), while the PNM is higher on the continental shelf than in the open sea (0.21 and 0.07-0.09 µM, respectively). The dissolved oxygen maximum is at the same depth as in summer, but its intensity is much lower (246.5 µM in the open sea, and 242.9 µM on the continental shelf).

Below the DCM, the Chl a concentration decreases until it falls below the detection limit around 200 m, except in winter, when concentrations of 0.03 to 0.06 µg l^–1 have been measured between 300 and 600 m depth. Below the PNM, the nitrite concentration also decreases, but its concentration remains detectable throughout the water column (up to 0.08 µM). Dissolved oxygen concentration decreases below the surface maximum to reach minimum values at 500 to 800 m depth, although its intensity is variable (189.8-201.0 µM), and below 1400 m to the bottom of the basin it increases again up to around 205.9±3.64 µM.

Surface water (0-50 m) dissolved oxygen concentrations and saturation concentrations plotted vs T are shown in Figure 5. Open sea stratified surface waters (0-50 m) of the Catalan Sea are oversaturated in dissolved oxygen (104.9% in spring, 108.5% in summer and 101.8% in autumn). In the open sea mixed water columns are undersaturated in winter (mean value 98.0%). This winter value is highly influenced by the low values (<230 µM) at 13°C, which correspond to stations sampled during WMDW formation. On the continental shelf, the oversaturation is found throughout the year in surface waters between 0 and 40 m in spring (106.8%) and between 0 and 50 m in summer and autumn (107.0% and 102.3%, respectively), but the oversaturation in the continental shelf mixed water columns is restricted to the upper 20 m (102.5% in winter). Both in stratified open sea waters and continental shelf waters, the depth at which the consumption of oxygen exceeds its production, i.e. at which median AOU values turn positive, is just above the DCM. The absolute minimum dissolved oxygen concentration found at 500 to 600 m depth (189.8-201.0 µM), the same depth at which maximum phosphate and nitrate concentrations are found, is corresponded by an absolute maximum of AOU (58.4-69.5 µM), which represents a consumption of around 23.7% to 25.2% of the saturating oxygen.

Surface water ratios are shown in Table 3. In the open sea, the N:P ratio under stratified conditions (spring to autumn) from 0 to 50 m depth is lower than the Redfield ratio (median±median absolute deviation, 2.6±1.3, n=717) (Fig. 6). On the continental shelf, the N:P ratio is significantly larger in autumn than in spring and summer. The N:P ratio increases significantly when ammonium is added to the calculations, but it is still below the Redfield ratio. In the open sea, the dissolved inorganic nitrogen (DIN):P ratio is higher in summer than in spring and autumn, while on the continental shelf the DIN:P ratio in summer and autumn is larger than in spring. In mixed water winter conditions, the N:P ratio is higher and more variable. Below the photic layer (around 200 m depth to the bottom of the basin) the average±standard deviation N:P ratio is 22.6±5.7 (n=1203), and DIN:P is 21.9±5.8 (n=194). This values are similar to the slopes obtained with regression lines for all N:P (22.6, r²=0.646, n=6560) and DIN:P data (22.2, r²=0.578, n=1605), which are significantly different from 16 (p≤0.001) (Table 4). Lower DIN:P compared with N:P ratios are a consequence of using different datasets, since not all the variables are analysed in all the cruises (See Table S1).

Median Si:N ratio shows a clear seasonality in surface open sea waters (0–50 m), with a significantly lower value in winter and a higher value in summer, in terms of both average and standard deviation (Fig. 6, Table 3). The median Si:N is not different from spring to autumn (pooled Si:N=4.4±2.7, n=1825). As in the open sea, the lowest Si:N ratio on the continental shelf is found in winter, and there are no significant differences between the ratio from spring to autumn (pooled ratio: 5.3±2.6, n=1108). Below 1200 m depth average, the Si:N concentration is 1.0±0.1 (n=144). In surface waters (0-50 m) median Si:P ratios are highly variable (Fig. 6, Table 3) and, as for the Si:N ratio, the largest values are in winter and the lowest are in autumn in the open sea, and in summer on the continental shelf (Table 3). In the open sea, Si:P ratios in spring and summer are the same (pooled ratio: 16.3±9.1, n=854). Below 1200 m depth, the average Si:P ratio is 23.2±3.4 (n=69).

The SMA regression slopes of AOU:P and AOU:N are 150.2 (r²=0.353, n=1420) and 8.9 (r²=0.603, n=1864) (Table 4). The scatter plots in Figure 7 show that Si:N, Si:P and AOU:Si relationships are not linear, since silicate concentration still increases below the depth at which phosphate and nitrate have been completely remineralized, corresponding to the WMDW and the deepest LIW (from 500-600 m depth). Therefore, we have not calculated the linear relationship for these ratios. The average AOU:Si calculated from pairs of data in deep waters (>1200 m depth) is 6.06±0.7 (n=58).

Depth distribution of phosphate, nitrate and silicate described in this study (Fig. 3) show the typical nutrient profile, with low surface concentrations due to their assimilation by phytoplankton and other osmotrophic organisms, followed by an increase in concentration with depth, as described in other parts of the NW Mediterranean Sea (e.g. Marty et al. 2002 at the neighbouring DYFAMED site). Deep water phosphate, nitrate and silicate concentrations are in agreement with published concentrations from the western Mediterranean Sea (e.g. Ballester et al. 1967, Cruzado 1985, Béthoux et al. 1998).

Maximum concentrations of phosphate and nitrate in intermediate waters have been observed in other areas of the Mediterranean Sea (around 400-700 m, Béthoux et al. 1998, Moutin and Raimbault 2002). Although a clear maximum cannot always be distinguished in our seasonal median profiles, the observation of individual profiles confirms the presence of higher nitrate and phosphate concentrations between 500 and 800 m depth that follow the pattern of the depth profiles of salinity. In addition, we have also observed in individual profiles the occasional presence of a nutrient peak between 150 and 300 m depth, associated again with a peak in salinity. Although the intensity of this peak is variable, it can still be observed in the median profiles of nitrate and phosphate (Fig. 3).

A maximum of nutrients and AOU associated with a minimum of dissolved oxygen (100-1000 m) is a feature present in most oceans in the mesopelagic zone (Sarmiento and Gruber 2006), and between 400 and 800 m depth in other areas of the NW Mediterranean Sea (e.g. Marty et al. 2002, Moutin and Raimbault 2002). It has been associated with both animal respiration and active microbial remineralization of organic material, mostly dead phytoplankton, faecal pellets and other detrital particles that sink from surface waters (e.g. Sarmiento and Gruber 2006). Large zooplankton organisms and micronekton also play an important role in transporting dissolved organic carbon and nutrients from surface waters to the mesopelagic zone through diel vertical migrations (e.g. Longhurst and Harrison 1988, in the NW Mediterranean Sea: Isla et al. 2015). A higher abundance of animals and microbes at a certain depth of the mesopelagic zone (for example because of the accumulation of sinking organic matter in the interface of two water masses of different density such as WIW and LIW or LIW and WMDW) could lead to the accumulation of nutrients, and hence to the formation of a maximum. Also, the association of the nutrients maximum and dissolved oxygen minimum with salinity indicates that their origin is the remineralization of surface organic matter that has been trapped within the LIW (or WIW) during its formation and sinking. It is unlikely that the decrease in nitrate from this maximum to the bottom of the basin (Fig. 3) is due to denitrification, since this process requires lower oxygen concentrations than those measured at deep Catalan Sea waters (Riley 1971, Millero 1996).

Phosphate and nitrate concentrations remain approximately constant below the maximum in intermediate waters to the bottom of the basin, indicating that all the labile phosphate and nitrate contained in sinking organic matter have already been remineralized. On the other hand, silicate concentration steadily increases from the surface to 1400 m depth, from where it remains constant down to the bottom of the basin. This difference in the nutrient profiles is a consequence of the slow chemical dissolution of siliceous diatom frustules throughout the water column, as opposed to the active remineralization of phosphate and nitrate, carried out mainly by bacteria and other protozoa (Millero 1996). The different remineralization rates between these nutrients results in the non-linearity of the Si:N, Si:P and AOU:Si ratios, as opposed to AOU:P, AOU:N and N:P, as can be observed in Figure 7.

A statistically significant relationship between the depths of the DCM, the PNM and the nitracline has been described in the Catalan Sea (Estrada 1999) using a small set of stations from cruises included in the present study (Fronts 90, 91, 92 and Varimed 93). Estrada (1999) compared the position of the maximums and nitracline at 58 stations, showing that the PNM was almost coincident with the nitracline, the DCM was slightly above them, and finally the maximum of dissolved oxygen was the shallowest. This pattern is almost the same as that resulting from comparing our median profiles, except that in our profiles the DCM is found at the same depth as both the PNM and the nitracline in winter and summer, and immediately below them in spring and autumn (Fig. 4). However, the observation of individual profiles shows that these maximums are not always coincident, even on consecutive days of the same cruise. This differences could be related to changes in the nitrite release by phytoplankton at the DCM driven by light or nutrient availability, which could also affect the abundance and/or activity of nitrifiers. Unfortunately, there are no data about light or organisms abundances in our dataset that could help us to understand the relative changes in the peaks position.

In other oligotrophic areas (North Pacific, Bering Sea, Greenland Sea, Sargasso Sea and Gulf Stream) a maximum ammonium concentration has been detected close to the thermocline (i.e. under stratified conditions) and to the DCM (e.g. Saino et al. 1983, Brzezinski 1988), indicating the importance of recycled production in oligotrophic seas. The examination of individual profiles also confirms the accumulation of ammonium around the DCM and PNM in stratified surface waters. However, this pattern is not reflected in the median profiles because a clear peak is not always observed (Fig. 4). On the other hand, under mixed water conditions in the Catalan Sea, an absolute maximum of the median ammonium concentration (1.21 μM) is commonly observed at 50 m, overlapping with a relative maximum of Chl a and an absolute maximum of nitrite.

The analysis of ammonium is very sensitive to environmental contaminants and processes both in situ and on land (e.g. cleaning products, smoke, unclean glassware or freezing as a preservation method, see Holmes et al. 1999). Thus, large ammonium concentrations are usually discarded under the assumption of possible sampling or analysis contamination. Taking into account the ease of contamination, the lack of sampling details and the scarcity of the ammonium dataset, the clear seasonality found is striking. Ammonium concentrations in open sea surface waters follow the same seasonal pattern as phosphate, nitrate and silicate, with the highest concentrations in winter and the lowest in summer. The accumulation of ammonium in surface waters demonstrates an imbalance in the production-consumption processes, especially in winter, when the highest primary production takes place. In the North Atlantic Ocean an increase in ammonium concentration has been described after a silicate-limited diatom bloom, which reversed the direction of the air-sea flux of this nutrient (Johnson et al. 2007). The molecules of atmospheric ammonium suffer a series of chemical reactions and can act as cloud condensation nuclei (Johnson et al. 2007). The winter-spring phytoplankton maximum in the Catalan Sea is dominated by diatoms (Estrada et al. 1999) and, although it is most probably not silicate-limited (Si:N and Si:P ratios in winter surface waters are larger than in deep waters, Fig. 6), the accumulation of ammonium (median ~0.6 µM, similar to that in the study of Johnson et al. 2007) indicates that a release of ammonium to the atmosphere could also be happening in the NW Mediterranean Sea during winter, a hypothesis that remains to be quantified.

Ammonium and nitrite concentrations are still measurable below the euphotic layer, throughout the water column (below 400 m depth: 0.59±0.36 µM ammonium and 0.04±0.02 µM nitrite, Fig. 4). This result is unexpected in an oxic environment like the Catalan Sea, and the origin of such high values remains unclear. Relatively high ammonium concentrations (0.4 µM at 800 m depth) in deep waters have also been reported in another oligotrophic area, the North Pacific Subtropical Gyre (Karl and Knauer 1984). The authors suggested that its origin was the chemolithotrophic activity associated with sinking particles, and their accumulation was the result of the inefficiency of the aphotic microbial community to use them (Karl and Knauer 1984). Assuming that the amount of sinking particles is proportional to surface water primary production, our average ammonium concentration is larger than that measured in the North Pacific Subtropical Gyre, both systems having similar primary production estimates: 300 to 400 mg C m^–2 day^–1 in the western Mediterranean (Estrada 1996) and 463 mg C m^–2 day^–1 in the North Pacific Subtropical Gyre (Estrada 1996) and 463 mg C m^–2 day^–1 in the North Pacific Subtropical Gyre (Karl et al. 1996). However, in the Catalan Sea, additional particles from the shelf and slope can also reach the water column, e.g. through nepheloid layers, trawling (e.g. Palanques et al. 2006) or DSWC (see below). Also, several studies point out that surface particulate organic matter in the NW Mediterranean Sea is N-rich (e.g. Copin-Montégut and Copin-Montégut 1983, Mostajir et al. 1998). Therefore, chemolithotrophic activity in sinking particles appears to be a realistic pathway for the accumulation of ammonium in the mesopelagic zone of the Catalan Sea.

In this section we inspect data that have failed the quality control, i.e. they have been excluded from the establishment of the reference depth profiles described above. We have identified three processes that affect the concentration of biogeochemical variables in the Catalan Sea: coastal eutrophication, DSWC and deep water formation.

High nutrient concentrations in surface coastal waters due to river discharges or anthropogenic influence can be easily identified in our dataset, though coastal conditions are poorly represented in this study: only 5.3% of the stations have a bottom depth lower than 50 m, the lowest bottom depth being 20 m. The Catalan coast shows a high variability in nutrient concentrations in response to the characteristics of the coast (i.e. beaches, rocky coasts and harbours), and the existence of continental or anthropogenic outflows (Flo and Camp 2005). This variability is also influenced by seasonality. For example, before the installation of a wastewater treatment plant, in summer months nutrient and Chl a concentration in surface waters of the Blanes Bay reached maximum values related to the increase in population (>Duarte et al. 1999, Guadayol et al. 2009). However, the present study shows that on the continental shelf the lowest nutrient concentrations are found in summer, suggesting that occasional coastal fertilization processes have local effects, and cannot be distinguished with the dataset used. Still, data outside the quality control ranges of concentration in surface waters occur mainly near the Ebre River and near large cities such as Barcelona, Tarragona and Palma (see Fig. 1), indicating that these high values are related to anthropogenic activity and river discharges.

In the open sea, anomalous depth profiles are found in the northern part of the Catalan Sea in winter, and are related to WMDW formation and to DSWC. WMDW is formed in winter in the Gulf of Lion and the northern part of the Catalan Sea (MEDOC Group 1970, Salat 1983, Font et al. 1988). During WMDW formation, T, S, and density characteristics are homogeneous throughout the water column (0-2800 m) (Salat 1996). Nutrients, density anomaly and dissolved oxygen profiles of two stations of the cruise Hivern 2000, sampled during the process of WMDW formation, are shown in Figure 8. It can be observed that the biogeochemical depth profiles resulting from the mixing during the WMDW formation are also homogeneous throughout the water column, with concentrations equivalent to deep waters. For comparison, the median winter concentration profiles of the same variables are also plotted in Figure 8, together with the limits of the winter quality control ranges. These profiles “fail” the quality control, which is constructed on the basis of depth. However, the concentrations are in agreement with those measured in deep waters, which can be easily identified by their S, θ, and σ-T characteristics.

DSWC also occurs in winter, when cold and dry northerly winds cause heat loss and evaporation of surface Gulf of Lion shelf waters. The waters therefore become cooler and mix with off-shelf waters, becoming denser than the surrounding waters and rapidly sinking, overflowing the shelf edge and cascading downslope through submarine canyons. Depending on the resulting density, the waters can reach the bottom of the basin, or detach from the slope when they reach their equilibrium with surrounding waters (Canals et al. 2006). In the Cap de Creus Canyon (Fig. 1) four major cascading events were identified in 1980, 1988, 1999 and 2005 (Canals et al. 2009). In our dataset, we have identified relatively large (~12 µM) nitrate and silicate concentrations in La Fonera Canyon, and very large concentrations below 500 m depth in the winter of 1999 (cruise Hivern 1999), up to 41.5 µM nitrate and 24.5 µM silicate (Fig. 9), which are more than three times the corresponding concentrations in deep waters. The corresponding phosphate values are also high (between 0.37 and 0.71 µM), but because the range of concentration of phosphate in the Catalan Sea is narrower, they cannot be so easily identified. In shelf surface waters of the Gulf of Lion (0-5 m depth) in winter, high nutrient concentrations have been measured on several cruises (up to 41.0 µM nitrate, 42.7 µM silicate and 2.00 µM phosphate), including years of intense cascading (Pelagolion 2 in 1986, Discovery88 and Pelagolion 3 in 1988, and Tyro in 1991). These high concentrations are related to nutrient discharges from the Rhône River (Cruzado and Velásquez 1990). We suggest that deep shelf cascading of winter surface waters, enriched in nutrients through the discharge of the Rhône River, could be the process that causes the occasional presence of high nutrients in intermediate waters in the Catalan Sea.

Seasonal differences in the biogeochemistry of surface waters have been studied by the statistical comparison of integrated concentrations (0-100 m). As expected, a clear seasonality in T is found in surface waters (Salat 1996, Duarte et al. 1999), driving the stratification of the water column (Segura-Noguera et al. 2011). The stratification is broken by convective processes (vertical mixing) related to an increase in the wind velocity and storms in autumn and winter (Salat 1996, Send et al. 1999, Guadayol and Peters, 2006), resulting in an enrichment of surface waters. Surface S does not show such strong seasonality, but even so, higher values are found in autumn and winter due to wind-driven evaporation and winter mixing with deeper and saltier waters (Salat 1996), which results in a larger surface density anomaly in this season. Seasonal biogeochemical differences are not always statistically significant (Table 2). For example, phosphate concentration in winter is not statistically different from that in autumn, possibly because of its fast uptake by phytoplankton as soon as nutrients become available.

Chl a concentration is higher in mixing conditions (winter) than during the rest of the year, in contrast to the most usual spring bloom (Longhurst 1988). This is possible because of the higher irradiance (around 20%) as well as warmer temperatures (3 to 5°C higher) in the Mediterranean Sea than in the Atlantic Ocean at the same latitude (Duarte et al. 1999), as well as because deep mixing occurs during winter (e.g. Salat 1996), which result in nutrient enriched surface waters. Chl a concentration decreases from winter to summer following a decrease in autotrophic organisms, both because the nutrients are being consumed (and hence the DCM moves towards deeper waters), and because grazing pressure is increased (Bahamón and Cruzado 2003).

In autumn, after the stratification is broken and the nutrient concentration increases in surface waters, the DCM is found at shallower depths, as already described (Estrada 1985a,b, Bahamón and Cruzado 2003). However its intensity, as well as that of the PNM, is smaller than in summer and spring. As a consequence, unlike nutrient concentration, integrated Chl a concentration is lower in autumn than in summer (Table 2). Assuming that there are no changes in irradiance compared with the summer months, the smaller DCM in autumn could partly result from a decrease in pigment concentration per cell due to photoacclimation. In addition, the lower %O₂ saturation in autumn suggests that the decrease in Chl a in this season could also be related to a decrease in the abundance of autotrophic organisms. Finally, the sudden increase in the Catalan Current intensity in autumn driven by wind and storms (Astraldi and Gasparini 1992, Font et al. 1995) could result in an increase in Chl a exported from the Catalan Sea. However, additional data such as particulate organic carbon and plankton composition are needed to determine whether the observed decrease in average Chl a concentration corresponds to a decrease in pigment cell content, a decreased in total biomass, or a change of populations with a different pigment composition.

The nitrate to phosphate ratio described by Redfield in deep waters (N:P=16:1, Redfield 1963) is typically used to determine whether a system is N- or P-limited. Our N:P ratio in deep waters, calculated with values above detection limit, is larger than the Redfield ratio (N:P>22), as is usually found in the western Mediterranean Sea (e.g. Copin-Montégut and Copin-Montégut 1983, Ribera d’Alcalà et al. 2003, Pujo-Pay et al. 2011). Similarly, in our deep waters (>1200 m) Si:N and Si:P are similar to other deep values in the western Mediterranean Sea (Si:N=0.9-1.1, Ribera d’Alcalà et al. 2003; N:Si:P=22:19.5:1, Béthoux et al. 2002) and in the Gibraltar Strait (N:Si:P=31.5:26.5:1, Dafner et al. 2003), and the differences can be explained by the ratios being calculated at different depths. Moreover, as observed in the eastern Mediterranean Sea (Krom et al. 1991, Kress and Herut 2001), the N:P and DIN:P scatterplots show that the relationship is non-linear at low and high N and P concentrations (Fig. 7). These results suggest that P is the element that generally limits the production in the Catalan Sea, as found in the rest of the Mediterranean Sea (e.g. Margalef 1963, Marty et al. 2002, Krom et al. 2010). Several hypotheses explain the high N:P in the Mediterranean Sea: atmospheric deposition, unbalanced river discharges, phosphate adsorption onto Saharan dust, and nitrogen fixation by cyanophyceae and marine phanerogams (Béthoux and Copin-Montégut 1986, Krom et al. 1991, Ribera d’Alcalà et al. 2003). Indeed, in the Mediterranean Sea, P has long been known to limit phytoplankton production in surface waters (Margalef 1963), as has been confirmed in numerous experimental and descriptive studies (e.g. Thingstad et al. 1998, Moutin and Raimbault 2002).

However, our data also show that, in stratified surface waters above the DCM (0-50 m), N:P and DIN:P ratios are below the Redfield ratio, while in mixed surface waters they are above it, both in the open sea and on the continental shelf. This suggests a shift from N-limitation, in which the water column is stratified to some degree, to P limitation, in which the water column is mixed (Fig. 6). Using a model II regression method (our data are not normally distributed), the intercepts of both N:P and DIN:P regressions are negative and significantly different from 0 (Table 4), indicating an excess of P when N is 0. However, if we use ordinary least squares model I (e.g. as in Kress and Herut 2001), the intercepts are positive and not significantly different from 0 (p>0.05). An examination of paired data from 0-50 m depth shows that it is more frequent to find phosphate data below detection limit when there are still measurable concentrations (i.e. above detection limit) of nitrate (59% of the cases) or DIN (67% of the cases), than vice versa. Low N:P ratios in surface waters, as well as median phosphate concentrations above detection limits, are unexpected results compared with the observed deep water N:P ratio and with experimental studies about nutrient limitation in the Mediterranean Sea. Low N:P ratios could be the result of an overestimation of phosphate, as found when chemical and radiochemical methods are compared (Hudson et al. 2000), and would indicate the need to use more sensitive methods, which provide equivalent detection limits for both nutrients, to measure phosphate in surface waters of the Catalan Sea.

In addition, as observed in the western Mediterranean Sea (Raimbault and Coste 1990, Estrada 1999, Moutin and Raimbault 2002) and the Levantine Basin (e.g. Krom et al. 1991), we found a shift from N-limitation to P limitation within the DCM, where the nitracline is found above the phosphocline (Pujo-Pay et al. 2011). The uncoupling of nutriclines causes the N:P ratio to reach maximum values at the depth of the DCM (Fig. 10). It is suggested that the lack of phosphorus prevents the utilization of nitrogen, so nitrate (and silicate) is accumulated in the water column at a shallower depth than phosphate (Raimbault and Coste 1990).

The N:P ratio below the DCM (160-400 m depth) is around 23, which corresponds to the remineralization of particulate matter—mostly dead phytoplankton, faecal pellets and other detrital particles (Sarmiento and Gruber 2006)—from waters immediately above. Consequently, in the Catalan Sea this sinking organic material would contain more N than P relative to the Redfield ratio. The rapid turnover rate of phosphorus in the euphotic zone (Benitez-Nelson and Buesseler 1999) could result in a larger N:P ratio of non-living sinking particles. Indeed, Mostajir et al. (1998) found higher than Redfield N:P ratios in detrital matter from the Ligurian Sea (N:P=32). In addition, some analyses of particulate matter from the NW Mediterranean Sea indicate that the N:P ratio of the whole phytoplankton community is also larger than Redfield. Copin-Montégut and Copin-Montégut (1983) found a particulate N:P ratio between 20 and 23 in the NW Mediterranean Sea, and suggested that the high N:P ratio in deep waters was the result of remineralization of microphytoplankton, which has an N:P ~30 and higher than smaller organisms. Later studies in living cells found that the N:P ratios of cultured picoplankton cells (Synechococcus and Prochlorococcus, including one strain from the NW Mediterranean Sea) were higher than Redfield in cells grown without nutrient limitation, and up to N:P=109 under P limitation (Bertilsson et al. 2003). Single-cell analysis of diatoms and dinoflagellates sampled directly from the Catalan Sea also indicates that the cells are N-rich (slope N:P quota=24.6 in dinoflagellates and 38.7 in diatoms, Segura-Noguera et al. in press).

Redfield noted that the N:P obtained from nutrients in deep oceanic waters should be an average of plankton stoichiometry, which is able to adapt to oceanic nutrient availability (Redfield 1963). Several studies confirm the plasticity of plankton stoichiometry as a result of changes in the metabolic cellular machinery in response to environmental nutrient availability (e.g. Margalef 1998, Geider and LaRoche 2002), with a theoretical optimal N:P stoichiometry between 8.2 and 45.0 (Klausmeier et al. 2004). For example, diatoms, which dominate the phytoplankton community in the NW Mediterranean Sea in winter (Estrada 1999, Arin et al. 2013), are able to store nitrate, as well as phosphate, inside intracellular vacuoles, allowing them to continue growing for several generations after the nutrients are depleted (Rosen and Lowe 1984, Margalef 1998). Also, diatoms could store nitrate as pigments (Rosen and Lowe 1984), and since each chlorophyll molecule contain four atoms of N, the resulting N:P ratio would be higher. A larger chlorophyll concentration in diatoms compared with cultured dinoflagellates and other autotrophic organisms in natural samples has actually been suggested (e.g. Ríos et al. 1998). Another adaptation that would lead to higher cellular N:P ratios is the preferential synthesis of N-rich substitute lipids in response to P limitation, as found in the Adriatic Sea (Ivančić et al. 2012). Since the cellular stoichiometry is the result of the species-specific genetic stoichiometry and the adaptation to the nutrient availability, the coincidence in the Catalan Sea of the N:P ratio in particulate matter and in dissolved nutrients below the euphotic zone, both of them higher than the canonical Redfield ratio, suggests an adaptation of the phytoplankton to the low P availability of the Mediterranean Sea (Klausmeier et al. 2004), and/or an evolutive selection of high N:P stoichiometry plankton cells (Kay et al. 2005). To confirm this, more measurements of the elemental composition of plankton are needed in the NW Mediterranean Sea, including other abundant groups (e.g. Primnesiophyceae), in addition to the community composition and participation in the downward flux of particulate organic matter.