Climatological distribution of dissolved inorganic nutrients 1 in the Western Mediterranean Sea ( 1981-2017 ) 2 3

15 The Western MEDiterranean Sea BioGeochemical Climatology (BGC-WMED) presented here is a 16 product derived from quality controlled in situ observations. Annual mean gridded nutrient fields for the 17 period 1981-2017, and its sub-periods 1981-2004 and 2005-2017, on a horizontal 1/4° × 1/4° grid have 18 been produced. The biogeochemical climatology is built on 19 depth levels and for the dissolved 19 inorganic nutrients nitrate, phosphate and orthosilicate. To generate smooth and homogeneous 20 interpolated fields, the method of the Variational Inverse Model (VIM) was applied. A sensitivity 21 analysis was carried out to assess the comparability of the data product with the observational data. The 22 BGC-WMED has then been compared to other available data products, i.e. the medBFM 23 biogeochemical reanalysis of the Mediterranean Sea and the World Ocean Atlas 2018 (WOA18) (its 24 biogeochemical part). The new product reproduces common features with more detailed patterns and 25 agrees with previous records. This suggests a good reference to the region and to the scientific 26 community for the understanding of inorganic nutrient variability in the western Mediterranean Sea, in 27 space and in time, but our new climatology can also be used to validate numerical simulations making 28 it a reference data product. 29


31
Ocean life relies on the loads of marine macro-nutrients (nitrate, phosphate and orthosilicate) and other 32 micro-nutrients within the euphotic layer. They fuel phytoplankton growth, maintaining thus the 33 equilibrium of the food web. These nutrients may reach deeper levels through vertical mixing and 34 remineralization of sinking organic matter. Ocean circulation and physical processes continually drive 35 the large-scale distribution of chemicals (Williams and Follows, 2003) toward a homogeneous 36 distribution. Therefore, nutrient dynamics is important to understand the overall ecosystem productivity 37 and carbon cycles. In general, the surface layer is depleted in nutrients in low latitude regions (Sarmiento 38 and Toggweiler, 1984), but in some ocean regions, called high nutrient low chlorophyll (HNLC) regions, 39 nutrient concentrations tend to be anomalously high, particularly in areas of the North Atlantic and 40 Southern Ocean, as well as in the eastern equatorial Pacific, and in the North Pacific; see e.g. Pondaven The Mediterranean Sea is known to be a hotspot for climate change (Giorgi, 2006). During the early 63 5 The aim of this study is to give a synthetic view of the biogeochemical state of the WMED, to evaluate 118 the mean state of inorganic nutrients over 36 years of in situ observations and to investigate upon a 119 biogeochemical signature of the effect of the WMT . 120 The paper is organized as follows, section 2 describes the data sources used and the quality check; 121 section 3 is devoted to the methodology, section 4 presents the main results including a comparison of 122 the new climatology with other products. At the end, we address the change in biogeochemical 123 characteristics before and after WMT. 124 The climatological analysis depends on the temporal and spatial distribution of the available in situ data, 128 and the reliability of these observations. Due to the scarcity of biogeochemical observations in the 129 WMED, merging and compiling data from different sources was necessary. project or by regional institutions having as objectives the investigation of the deep water convection 144 and the biogeochemical properties of the of the WMED. Data were chosen to ensure high spatial 145 coverage (Fig. 3). 146   Hydrological and biogeochemical measurements have always been repeatedly collected along several 162 repeated transects, known as key regions as the Sicily Channel and the Algéro-Provençal subbasin; 163 likewise, the northern WMED is a well sampled area, as it is an area of DW formation. Observation 164 density is still scarce (less than 100 observations) in some areas like the northern Tyrrhenian Sea. 165 The total number of measurements at each depth range underlines similar remarks, an uneven 166 distribution that needs to be considered in the selection of the vertical resolution to estimate the 167 climatological fields. Though, the use of 36 years of nutrient measurements to generate the 168 climatological fields significantly reduces the error field. In our case and taking into account the irregular 169 distribution in seasons and different years. A climatological gridded field was computed by analyzing  (Table A1.), thus before 179 merging them, observations were first checked for duplicate (the number of profiles listed in Table 2  180 refers to all data after removing duplicate measurements). The criteria to detect and remove duplicates 181 is simple: observations collected during the same cruises extracted from the different sources were 182 removed. Since profiles were measured during specific cruise (identified with a unique identification 183 code) at specific time, data from duplicate cruises are removed. 184 Then, data was converted to a common format (similar to the csv CNR_DIN_WMED_20042017 data 185 product, Belgacem et al., 2019). This recently released product contains measurements covering the 186 WMED from 2004 to 2017. The data of the CNR_DIN_WMED_20042017 product have undergone a 187 rigorous quality control process that was focused on a primary quality check of the precision of the data 188 and a secondary quality control targeting the accuracy of the data, details about the adjustments and the 189 applied corrections are found in Belgacem et al.(2020). 190 As detailed in Table 2 10, 10-30, 30-60, 60-80, 80-160, 160-260, 260-360, 360-460, 460-560, 560-900, 900-197   1200, 1200-1400, 1400-1600, 1600-1800, 1800-2000, 2000-2200, 2200-2400, 2400-2600, >2600 dbar). 198 Any value that is more than three median absolute deviations from the median value is considered a 199 suspected measurement. The code is freely available at https://github.com/gher-ulg/DIVAnd.jl (last access: January, 2020). 219 DIVA is based on the variational inverse method (VIM) (Brasseur et al., 1996). It takes into account the 220 errors associated with the measurements and takes account of the topography/bathymetry of the study 221 area. The method is designed to estimate an approximated field close to the observations and find the 222 field that minimizes the cost function [ ]. 223 The cost function is defined as the misfit between the original data , an array of observations, the 224 analysis (observation constraint term) and a smoothness term. invariant (Brasseur and Haus, 1991). 0 minimize the anomalies of the field itself, 1 minimize the 237 spatial gradients, 2 penalizes the field variability (regularization). The reconstructed fields are 238 determined at the elements of a grid on each isobath using the cost function Eq. (1). 239 The grid is dependent on the correlation length and the topographic contours of the specified grid in the 240 considered region, so there is no need to divide the region before interpolating. 241 The method computes two-, three-to four-multi-dimensional analyses (longitude, latitude, depth, time). 242 For climatological studies, the four-dimensional extension was used on successive horizontal layers at 243 different depths for the whole time period. 244 Along with the gridded fields, DIVA yields error fields dependent on the data coverage and the noise in 245 the measurements (Brankart and Brasseur, 1998;Rixen et al., 2000). Full details about the approach are 246 provided extensively by Barth et al. (2014) and Troupin et al. (2018)

Interpolation parameters 249
DIVAnd is conditioned by topography, by the spatial correlation length (Lc) and by the signal-to-noise 250 ratio (SNR, λ) of the measurements, which are essential parameters to obtain meaningful results. They 251 are considered more in detail in the following sections.  For the BGC-WMED biogeochemical climatology, this parameter was optimized for the whole-time 265 span, and at each depth layer. The correlation length has been evaluated by fitting the empirical kernel 266 function to the correlation between data isotropy and homogeneity in correlations. The quality of the fit 267 is dependent on the number of observations (Troupin et al., 2018). The analytical covariance model used 268 in the fit is derived for an infinite domain (Barth et al, 2014). To assess the quality of the fit, the data 269 covariance and the fitted covariance are plotted against the distance between data points (Fig. 5). At 10 270 m, the correlation length was obtained with a high number of data points, indicating that the empirical 271 covariance used to estimate the covariance and the fitted covariance are in good agreement. 272 At some depth layers there are irregularities due to an insufficient amount of data points, making it 273 necessary to apply a smoothing filter/fit to minimize the effect of these irregularities. It has been tested 274 whether a randomly selected field analysis (nitrate data from 2006 and 2015) obtained with the fitted-275 vertical correlation profile is better than the analysis with zero-vertical correlation. A skill score relative 276 to analysis non-fitted-vertical correlation has been computed following Murphy (1988)  Eq. (2) 279 A large difference in the global RMS between the analysis with the fitted-vertical correlation and the 280 analysis with non-fitted-vertical correlation used for validation was found. The test shows whether the 281 use of the fit in the correlation profile is improving the overall analysis or not. We found that the RMS 282 error (nitrate analysis of 1981-2017) was reduced from 0.696 µmol kg -1 (analysis without fit) to 0.571 283 µmol kg -1 (analysis with fit) at 10 m depth, which means using the fitted vertical correlation profile in 284 the analysis improves the skill by 32 %, and the fit is improving the analysis fields. Based on the data, DIVA performs a least-square fit of the data covariance function with a theoretical 290 function. Then, a vertical filter is applied and an average profile over the whole period is used (Fig. 6). 291 This procedure is analogous to what has been used for the EMODnet climatology and the North Atlantic 292 climatology, except that in EMODnet climatology, seasonally averaged profiles were used (Buga et al., 293 2019) and a monthly averaged profiles were used in North Atlantic climatology (Troupin et al., 2010). 294 The filter is applied to discard aberration caused by outliers or scarce observations in some layers, as 295 described above. 296 Because of the horizontal and vertical inhomogeneity of the data coverage, the analysis was based on a 297 correlation length that varies both horizontally (Fig. 6a) and vertically (Fig. 6b). 298 As expected, Lc increases with depth ( Fig. 6), extending the influence area of the observation, a 299 consequence of the fact that variability at depth is lower and that observations in the deep layer are 300 scarcer (which on the other hand makes the Lc estimate more uncertain).

326
The signal-to-noise ratio (SNR) is related to the confidence in the measurements. It is the ratio between 327 the variance of the signal and the variance of the measurement noise/error. The SNR defines the 328 representativeness of the measurements relative to the climatological fields, in other words it is the 329 confidence in the data. 330 It not only depends on the instrumental error but also on the fact that observations are instantaneous 331 measurements, and since a climatology is a long-term mean, such observations do not represent exactly 332 the same. 333 Generally, small SNR values favor large deviations from the real measurements to give a smoother 334 climatological field. On the other hand, with a high SNR, DIVAnd keeps the existing observations and 335 interpolates between data points. The need is to find an approximation that does not deviate much from Following the same approach that many climatologies that used the DIVAnd method adopted, i.e. 338 EMODnet climatologies (available on the EMODnet chemistry portal), the Atlantic regional  (Table 1). 341 The analysis is performed with a predefined uniform default error variance of 0.5 for all parameters at 342 all depths, we presume that the data sources used to generate BGC-WMED climatology are consistent 343 products. Three iterations are done inside DIVAnd to estimate the optimal scale factor of error variance The automatic check measures how consistent the gridded field is with respect to the nearby 354 observations by estimating the difference between a measurement and its analysis scaled by the expected 355 error and based on that, a score is assigned to each observation. Data points with the highest scores were 356 considered as suspect and were removed from the analysis (Fig. A1, 2, 3). Overall, 0.031%, 0.014%, 357 0.004% data points, for nitrate, phosphate, and silicate, respectively, were considered inconsistent. 358 Details about the quality check values and range are plotted in the appendix (Table A1). The quality of the climatology was checked against observations by estimating the mean residual and 361 the root mean squared (RMS) of the difference between the climatology and the observations. Averages 362 over the entire basin were calculated between depth surfaces (see section 2.3). Residuals are the 363 difference between the observations within the specific depth surface and the analysis (interpolated 364 linearly to the location of the observations) and are estimated by depth range (Fig. 7). The analysis fields 365 at each depth range (i.e. depth surfaces or domain on which the interpolation is performed) are the 366 interpolation on the specified grid. In Fig. 7, we present the vertical profile of the mean residuals and 367 RMS at different depth ranges for the three nutrients. 1.1 µmol kg -1 . This is justified by the inhomogeneity of the observations mainly in deep layers. 373 As for the average residual between phosphate observations and the gridded analysis (Fig.7b) was 374 around zero and varied between -0.0027 and 0.0026 µmol kg -1 . The RMS for phosphate was between 375 0.037 and 0.063 µmol kg -1 . 376 Silicate residuals (Fig. 7c), on the other hand, seemed more homogeneous at all depth levels. The highest 377 level of agreement was found below 20 m and at 600 m. Overall residuals varied between -0.057 and 378 0.063 µmol kg -1 , while the RMS ranged between 0.567 and 0.963 µmol kg -1 . 379 Over the entire water column, the mean residual was around zero (0.004 µmol kg -1 for nitrate, 0.0002 380 µmol kg -1 for phosphate and 0.003 µmol kg -1 for silicate) (Fig. 7); The RMS blue line fell within the 381 mean residual +/-standard deviation in the upper 25 th percentile at the different depth ranges and in all 382 parameters meaning that in general, the bias between the observations and the analysis is small and there 383 is a good agreement. 384  The resulting climatologies (Table 3)  Here is an example of the analysis output found in the netCDF. Figure 8 shows the unmasked 399 climatological field of the mean spatial variation of nitrate, relative error field distribution, the masked 400 climatological field using relative error with two threshold values (0.3 and 0.5) to assess the quality of 401 the resulting fields.   The distribution of phosphate concentration has striking similarities with that of nitrate (Fig. 9b). The Concerning the distribution of silicate concentration, the surface layer at 100 m (Fig. 9c) followed the 441 same pattern as nitrate and phosphate. Over this layer the mean silicate was about 2.7 ± 0.7 µmol kg -1 . 442 As for nitrate and phosphate, the highest values (3-4 µmol kg -1 ), were recorded in the Alboran Sea, At the basin scale, nitrate concentrations increase with depth (Fig. 10a), with the highest concentration 450 found at intermediate levels (250-500 m), ranging between 8.8 and 9.0 µmol kg -1 . In this 300 m layer 451 (Fig. 9d), nitrate concentration average is 7.2 ± 1.06 µmol kg -1 . High values (> 6.5 µmol kg -1 ) are found 452 in the westernmost regions (Alboran Sea, Algerian Sea, Gulf of Lion, Balearic Sea and the Liguro-453 Provençal Basin), while the easternmost regions (Tyrrhenian Sea, Sicily Channel), exhibit much lower 454 concentrations (between 4.5 and 6.5 µmol kg -1 ). 455 Similar features are observed in the deep layer, at 1500 m (Fig. 9a), with nitrate concentrations 456 increasing all over the basin, reaching on average 7.8 -7.9 µmol kg -1 between 1000 and 1500 m depth 457 (Fig. 10a). The average vertical profile over the entire region (Fig. 10b), reveals a maximum in phosphate 467 concentrations between 300 and 800 m depth, related to an increased remineralization process. 468 In the deep layer (see 1500 m, Fig. 9h), phosphate concentration average is 0.36 ± 0.02 µmol kg -1 . 469 Generally, the deep layer is homogeneous (Fig. 10b). The difference observed between westernmost 470 regions and the Tyrrhenian Sea remains, though the latter demonstrate higher phosphate concentrations 471 (~0.3 µmol kg -1 ). This variation could be due to the difference in the water masses. The IW inflow from 472 the EMED brings relatively young waters that are depleted in nutrients, while in the higher 473

IW. 478
Silicate concentration distribution at intermediate (300 m, Fig. 9f) and deep layers (1500 m, Fig. 9i), 479 were as expected, showing a notable increase, compared to the surface. Here the silicate average 480 concentration is 5.83 ± 0.66 µmol kg -1 . The maximum values were observed below 800 m, > 8.034 µmol 481 kg -1 (Fig. 10c) Comparing the three nutrients at the same depth levels, at the surface (100 m), it appears that they all 491 show local surface maximum, depending on local events such as strong winds, local river discharge and 492 vertical mixing (Ludwig et al., 2010). Overall, in surface layer, circulation, physical processes, and vertical mixing increase nutrient input 509 while the biological pump controls the decrease. 510 In the deep layer, the variability is lower (standard deviation is reduced toward the bottom for all three 511 nutrients, see Fig.10), the deep layer accumulates dissolved organic nutrients. In the WMED, the deep 512 layer constitutes a reservoir of inorganic nutrients. 513  provides a qualitative distribution of the error estimate. This estimate is used to generate a mask over 531 the analysis fields. Two error thresholds were applied (0.3 (L1) and 0.5 (L2)). Fig.8b., show the main 532 error that occurs in regions void from measurements. An example of the analysis masked with the error 533 thresholds output is shown in Fig.8c (L1) and Fig.8d (L2). The associated error fields with the analysis 534 fields are integrated in the data product. 535 23

Comparison with other biogeochemical data products 536
In this section a comparison of the BGC-WMED product with the most known global and/or regional 537 climatologies, that are frequently used as reference products for initializing numerical models, is made. 538 Specifically, the analyzed fields are compared to the reference data products WOA18 (Garcia et al.,  Since the products used for inter-comparison were not originated from the same interpolation method, 544 not for the same time period and with different spatial resolution, here the comparison is mostly targeted 545 on the general patterns of nutrients in the region. 546 Comparisons are carried out between horizontal maps (Fig.11-12-13), as well as along a vertical 547 longitudinal transect (Fig.15-16). In addition, following Reale et al. (2020), the first 150 m have been 548 evaluated (Fig.14), since this is a depth level with a representative amount of in situ observations in all 549 three products. The evaluation is based on the estimation of horizontal average, on BGC-WMED 550 climatology, the medBFM biogeochemical reanalysis and the WOA18 climatology by subregion. i.e. a 551 spatial subdivision made according to Manca et al. (2004). We then calculated spatial maps of the mean difference at 150 m between the new climatology and the 562 reference products and then an average across subregions was performed. (1981-2017) product reveals detailed aspects of the general features of nitrate (Fig. 11.a), phosphate 566 ( Fig. 12a) and silicate (Fig.13a). 567 For the three nutrients, the new product reproduces patterns similar to the WOA18 all over the region. 568 It shows well-defined fields and higher values of nitrate and phosphate concentrations. In the new 569 product, nitrate concentrations varied between 2.31 -7.3 µmol kg -1 the WOA18 values were 2.19 -5.99 570 µmol kg -1 .Phosphate ranges were similar between the two products between (0.092-0.35 µmol kg -1 571 (BGC-WMED) and 0.095 -0.35 µmol kg -1 (WOA18)). Likewise, Silicate range values at 150 m were 572 not different (2.07 -4.99 (BGC-WMED) and 1.57 -5.75 µmol kg -1 (WOA18)). 573 The average RMS difference (RMSD) calculated from the difference between the WOA18 and BGC-574 WMED all over the region at 150 m is about 1.14 µmol kg -1 nitrate (Fig. 11c), 0.055 µmol kg -1 for 575 phosphate (Fig. 12c) and 0.91 µmol kg -1 for silicate (Fig. 13c). Overall, the RMS error values were low 576 indicating limited disparity between the two products. 577 The difference field for every grid point reflects this discrepancy and shows areas with limited 578 agreement between the two products that can have a difference >2 µmol kg -1 for nitrate (Fig. 11c), >0.1 579 µmol kg -1 for phosphate (Fig. 12c), >1.5 µmol kg -1 for silicate (Fig. 13c). This dissimilarity is also noted 580 with the low r 2 (Fig. 14) (0.34, 0.20, 0.095 for nitrate, phosphate, and silicate respectively) 581 The distribution of the surface nitrate concentrations (at 150 m) (Fig. 11a)

of the new product is similar 582
to that shown in WOA18 (Fig. 11b). The largest difference between the two products occurs in northwest 583 areas and in the Alboran Sea (Fig. 11c), areas of higher concentrations, a more nutrient rich surface  Phosphate surface concentrations (Fig. 12) show similar differences as nitrate. The largest difference 591 with the surface phosphate of the WOA18 is found in the Alboran Sea, Northern WMED and Sicily 592 region (Fig. 12c). 593 As for silicate, the surface distribution shows large differences (Fig. 13c). The highest values are 594 observed in the northwest area of the new product, and in the Alboran Sea in the WOA18 climatology , 595 this again accounts for the data coverage difference.    Table 4 and Fig. 15). 601 Results show a general agreement between BGC-WMED and the other two products in some 602 subregions, nonetheless, there are some differences as shown in section 4.3.1. 603 Upper layer nitrate average concentrations (Fig. 15a)  In the eastern regions, the lowest average concentrations of the WMED are found. Here, the difference 613 between products is smaller, with medBFM reanalysis being lower than the new product and the 614

WOA18. 615
As for phosphate (Fig. 15b), known to be the limiting nutrient of the WMED, because it is rapidly The BGC-WMED climatology shows reasonable agreement in the upper average concentrations of 626 nitrate and phosphate that are similar in order of magnitude to the other products (Fig. 15). The On the other hand, the average silicate (Fig. 15c) of the new product and the WOA18 varied between 631 regions. Significant difference is found between the two products in DS2, DS4, DF1, DF2, DT1, DT3, 632 DI1 and DI3, while in DS1, DS3 and DF4 mean silicate is consistent between the two products. 633 Overall, the three products show strongly similar features between regions (similar curve shape).  Table 4. Nutrient average concentrations and standard deviation in the upper 150 m. All products were 647 interpolated on 1° grid resolution (see Figure S2 (Belgacem et al., 2020)). 648

Subregion/ Coverage
Data product Nitrate Phosphate Silicate DS1-Alboran Sea (35°N-37.3°N, -6°E--1°E   We extracted data values along a longitudinal transect across the Algerian basin in the west-east 653 direction (Fig. 16) showing markedly features, a transect across the Tyrrhenian Sea is extracted as well (Fig. 16). Silicate 656 is not included as it was not represented in the medBFM model.  The vertical section along the Tyrrhenian Sea (Fig. 16) also shows a decrease from west to east in nitrate 671 concentrations. The same gradient is found also in phosphate in agreement with nutrient distribution 672 shown from the WOA18. From the section of the medBFM reanalysis, it is not easy to identify the west-673 east gradient that we mentioned before. It could be suggested that the model under-estimate the vertical 674 features in the Eastern (Tyrrhenian Sea: 100-300 m, nitrate vary between 1.4 and 4.2 µmol kg -1 , 675 phosphate between 0.13 and 0.20 µmol kg -1 ) and western part (Algerian basin: 100-300 m, nitrate vary 676 between 2.1 and 5.4 µmol kg -1 , phosphate between 0.15 and 0.255 µmol kg -1 ). These values are lower 677 than the ones found in the BGC-WMED (Tyrrhenian Sea: 100-300 m, nitrate range between 3 to 6 µmol 678 kg -1 , as for phosphate values oscillate between 0.10-0.27 µmol kg -1 ;Algerian basin: 100-300 m, nitrate 679 range between 3.6 to 8 µmol kg -1 , as for phosphate values oscillate between 0.18-0.36 µmol kg -1 ). 680 While the WOA18 reproduce similar patterns as the new climatology (Tyrrhenian Sea: 100-300 m, 681 nitrate vary between 1.8 and 5.7 µmol kg -1 , phosphate between 0.33 and 0.20 µmol kg -1 ) and western 682 part (Algerian basin: 100-300 m, nitrate vary between 2.8 and 6.8 µmol kg -1 , phosphate between 0.16 683 and 0.34 µmol kg -1 ). 684 The products illustrate the nutrient-poor water in the eastern side (Tyrrhenian Sea) and the relatively 685 nutrient-rich water found in the western transect (Algerian basin). 686 The BGC-WMED product capture details in Fig. 16 about the longitudinal gradient in nitrate and 687 phosphate, along the water column where nutrient sink deeper from west to east as previously seen in 688 Pujo-Pay et al. (2011) and Krom et al. (2014), an increased oligotrophy from west to east with higher 689 concentrations in the two nutrients in the western side of the section and a more oligotrophic character 690 toward east. 691 The differences between products could be explained by the difference in the data coverage, time span 692 and the difference in methods used to construct the climatological fields. 693 The variability in nitrate and phosphate fields along the transect extracted from the BGC-WMED reflects 694 the high resolution of the product allowing the screening of vertical structure controlling nutrient 695 contents. Based on a visual comparison, the new product is able to reproduce similar patterns as to the 696 WOA18 and to a lesser extent the medBFM reanalysis. 697 with the medBFM reanalysis along the Algerian basin (Fig.17a, nitrate; Fig.17b, phosphate) and 699 WOA18 (Fig.17c, nitrate; Fig.17d, phosphate). 700 The vertical section shows a strong agreement at the surface for nitrate between the BGC-WMED and 701 the medBFM reanalysis (Fig. 17a), while the vertical difference with WOA18 demonstrates that nitrate 702 values in the new product are lower than the WOA18 at 50-75 m (Fig. 17c). 703 The difference increases with depth, below 100 m, the BGC-WMED nitrate climatology is higher than 704 the medBFM with a difference ranging between 0.6 and 2.4 µmol kg -1 , similar observation is noted in 705 the WOA18 (Fig. 17c). In Fig.17a and Fig.17c, we identify patterns in the vertical structure of nitrate 706 in the eaten portion of the transect. 707 Regarding phosphate, differences between the new climatology and the medBFM reanalysis are noted 708 (Fig. 17b) where the BGC-WMED shows high concentrations in the first 100 m and between 150 m and 709 300 m (differences of 0.02 -0.08 µmol kg -1 ), this difference decreases at 100-150 m. At the eastern 710 portion of the transect (6°E to 7.5°E), we find an agreement between the two products. 711 Conversely, the vertical sections of the differences between BGC-WMED and WOA18 in phosphate 712 (Fig.17 d) show similarities, with the new product being lower than the WOA18 in the first 50 m. Large 713 difference is found on both sides of the transect below 100 m, while in the center of the transect, the 714 difference in phosphate is reduced to 0-0.02 µmol kg -1 . 715 The study also provides an examination of the nitrate and phosphate distributions along a longitudinal 727 transect across the Algerian Basin (Western WMED) and across the Tyrrhenian Sea (Eastern WMED). 728 We have shown that the western basin is relatively high in nutrients compared to the Eastern basin. The   The result of this climatological event was that a newly generated DW, denser, saltier, and warmer than  (Li and Tanhua, 2020). In this section, we investigate the 757 possible impact of WMT on biogeochemical characteristics at different depth levels (with a focus on 758 nitrate, phosphate and silicate regional distribution and patterns). 759 We considered depth levels that represent the usual three layers: the surface (100 m; Fig.19a-20a-21a), 760 intermediate (300 m; Fig.19b-20b-21b) and deep layers (1500 m; Fig.19c-20c-21c). 761 The WMED surface layer is dominated by the AW coming through the Alboran Sea, a permanent area 762 of upwelling (García-Martínez et al., 2019), where there is a continuous input of elements from the layer 763 below to the surface (Fig. 19a-20a-21a). Nitrate increased after WMT (Fig. 19d-20d-21d) by +0.4137 764 µmol kg -1 (Fig. A4a). The largest difference between the two periods reached >+2 µmol kg -1 in Sardinia 765 Channel and the Alboran Sea that was explained by the favorable conditions for nitrogen fixation as 766 discussed in Rahav et al. (2013), revealing also that nitrogen fixation rate increased from east-to-west. 767 Phosphate and silicate on the other hand described a decrease at 100 m ( Fig. A4a)   The surface layer exhibits an irregular distribution since it is subjected to seasonal variability. We found 771 an increase in all nutrients at 300 and 1500 m with a maximum identified at intermediate depth in both 772 nitrate and phosphate which is explained by the remineralization of organic matter along the path of the 773 IW. The latter flows westward (from the Levantine to the Atlantic Ocean). Its content in nutrients 774 increases (relatively to the conditions in the EMED) with age (Schroeder et al., 2020). It arrives at the 775 Tyrrhenian Sea, where in Fig.19b-20b-21b (at 300 m depth), we identify a nutrient-depleted intermediate 776 layer. At this depth level, we observe a gain in the three nutrients after WMT (Fig.19e-20e-21e). On 777 average, the difference between the two periods (pre/post-WMT) for nitrate, phosphate, and silicate, is 778 around +0.8648, +0.0068 and +0.2072 µmol kg -1 (Fig. A4b), respectively. 779 A similar increase after WMT in the deep layer (1500 m), is also found for nutrient concentrations (Fig.  780 19f, 20f, 21f) in the magnitude of +0.753 for nitrate, +0.025 for phosphate, and +0.867 for silicate (Fig.  781 A4c), which highlights an increase in the downward flow of organic matter remineralization that is 782 supplying the existing pool. 783

784
This increase is also illustrated in the climatological mean vertical profile of Fig. 22 in the three 785 nutrients. Nitrate displays a notable vertical difference to the pre-WMT period below 200 m (Fig. 22a). 786 Phosphate difference between the two-time period is larger below 400 m (Fig. 22b). Silicate was 787 different from nitrate and phosphate. It increases progressively with depth (Fig.22c) and demonstrates 788 an enrichment of the DW compared to the 1981-2004 period (Fig. 21c). The maximum values are found 789 in the deep layer, due to the low remineralization rate. With the warming climate, biogenic silica tends 790 to dissolve faster which explains the high concentrations all over the basin even the Tyrrhenian Sea after 791 the WMT. In this study, we investigated spatial variability of the inorganic nutrients in the WMED and presented 833 a climatological field reconstruction of nitrate, phosphate, and silicate, using an important collection 834 dataset spanning 1981 and 2017. The BGC-WMED new product is generated on 19 vertical levels on a 835 1/4° spatial resolution grid. 836 The new product represents the spatial patterns about nutrient distribution very well because of its higher 837 spatial and temporal data coverage compared to the existing climatological products (see Table 1), it is 838 contributing to the understanding of the spatial variability of nutrients in the WMED. 839 The novelty of the present work is the use of the variational analysis that takes into consideration 840 physical, geographical boundaries and topography, the resulting estimate of the associated error field. 841 Comparison with previously reported studies gives that the BGC-WMED reproduces common features 842 and agrees with previous records. The reference products WOA18 and medBFM biogeochemical 843 reanalysis tend to underestimate nutrient distribution in the region with respect to the new product. 844 The new product captures the strong east-west gradient of and vertical features. The results obtained do 845 not include seasonal or annual analysis fields. However, the aggregated dataset here does show 846 improvements in describing the spatial distribution of inorganic nutrients in the WMED. We The results support the tendency to a relative increasing load of inorganic nutrients to the WMED and 852 possibly relate the change in general circulation patterns, changes in deep stratification and warming 853 trends, however, this remains to be evidenced. 854 The BGC-WMED is a regional climatology that has allowed the identification of a substantial 855 enrichment of the waters, except for the Tyrrhenian Sea where the water column is depleted in nutrients 856 with respect to the western areas of the WMED. The climatology gave information about the spreading 857 of inorganic nutrients inside the WMED at surface, intermediate and deep layers. 858 A future work will suggest a better understanding of the change in nutrients related to water masses 859 associated with ventilation rate, a climatological field along isopycnal surfaces instead of depths and the 860 correlation between potential temperature and nutrients. 861 Appendix A: Additional information about cruise metadata 862