Reliability of the steric and mass components of Mediterranean sea level as estimated from hydrographic gridded products

Authors

  • Gabriel Jordà,

    Corresponding author
    1. Department of Ecology and Marine Resources, Institut Mediterrani d'Estudis Avançats (UIB-CSIC), Esporles, Spain
    • Corresponding author: G. Jordà, Department of Ecology and Marine Resources, Institut Mediterrani d'Estudis Avançats (UIB-CSIC), C/Miquel Marquès 21, ES-07190 Esporles, Spain. (gabriel.jorda@uib.es)

    Search for more papers by this author
  • Damià Gomis

    1. Department of Ecology and Marine Resources, Institut Mediterrani d'Estudis Avançats (UIB-CSIC), Esporles, Spain
    Search for more papers by this author

Abstract

[1] The steric component of Mediterranean sea level as obtained from different hydrographic gridded products is compared for the last decades. The widely used MEDATLAS2, ENACT/ENSEMBLES, ISHIIV6.7, and ISHIIV6.12 databases show large discrepancies, mainly derived from the halosteric component. The validation of the mass component against gravimetry data suggests that ENACT is slightly more accurate than the other data sets. Concerning long-term trends, the thermosteric component computed down to the bottom for the period 1960–2000 ranges from −0.06 ± 0.12 to 0.09 ± 0.16 mm/yr. Halosteric trends are negative for all products, but the magnitude and spatial patterns provided by the four products are statistically inconsistent among them. This prevents giving consistent figures for the overall steric term and mass term. Conversely, the trend of the freshwater content of the Mediterranean basin can be estimated more accurately, ranging from 1.0 to 1.3 mm/yr for the period 1960–2000.

1 Introduction

[2] The determination of the steric component of sea level variability is important as it provides information about the role of density on sea level changes. Steric changes are caused by changes in the temperature (thermosteric component) and salinity (halosteric component) of the water column and can therefore be estimated from hydrographic data. The mass component of sea level variability has not been directly observed until 2000, with the launching of the first satellite gravimetry (CHAMP) mission. Before that date, the mass component can only be estimated by subtracting the steric component from total sea level.

[3] In the Mediterranean Sea, the steric component has usually been characterized from gridded hydrographic products obtained from one of the databases covering the basin: MEDAR/MEDATLAS [Rixen et al., 2005], ENACT/ENSEMBLES [Ingleby and Huddleston, 2007], and ISHII [Ishii and Kimoto, 2009]. MEDAR/MEDATLAS has been used to explain sea level interannual variability [Tsimplis and Rixen, 2002; Tsimplis et al., 2008] and long-term trends [Marcos and Tsimplis, 2007; Tsimplis et al., 2008] at basin scale as well as at coastal stations. ENACT/ENSEMBLES has been used by García-García et al. [2010] to derive mass estimates and by Calafat and Marcos [2012] to compare steric sea level with altimetric observations. Finally, the ISHIIV6.7 product has been used by Calafat et al. [2010] to derive mass estimates and by Calafat et al. [2012] to validate regional models. The use of the most recent version of that product (ISHIIV6.12) is not yet in the literature.

[4] It is generally accepted that differences in the steric sea level inferred from the various databases covering the Mediterranean Sea are small at basin scale. In fact, a large amount of observations are common to all databases. However, the spatial and temporal coverage of historical profiles is so sparse that small differences in the amount or in the quality of the observations altogether with differences in the choices of the gridding procedure may result in significantly different products [Vargas-Yáñez et al., 2009; Tsimplis et al., 2011].

[5] The goal of this paper is to assess the reliability of steric sea level estimates in the Mediterranean Sea by comparing the results obtained from different hydrographic products. In particular, we compare the steric components computed from four gridded products (MEDAR/MEDATLAS, ENACT/ENSEMBLES, ISHIIV6.7, and ISHIIV6.12), focusing on the interannual variability and long-term trends. The reliability of the derived estimates of the mass component is also assessed.

2 Data and Methods

[6] The temperature and salinity gridded fields given by the databases have different spatial and temporal resolutions. MEDAR/MEDATLAS (hereafter MEDATLAS2; see http://www.ifremer.fr/medar/) provides yearly fields along the period 1945–2001 with a horizontal resolution of 1/5° × 1/5°; it covers the whole water column with 25 vertical levels. Here we use the most recent version of the database (M. Rixen, personal communication, 2012). ENACT/ENSEMBLES (hereafter EN3; http://www.metoffice.gov.uk/hadobs/en3/) provides monthly fields from 1950 to present with a spatial resolution of 1° × 1°; it also covers the whole water column, in this case with 42 vertical levels. The version EN3_v2a is used here. ISHIIV6.7 (hereafter ISHII; http://atm-phys.nies.go.jp/~ism/pub/ProjD/doc/) provides monthly fields for the period 1945–2006 with a spatial resolution of 1° × 1°; it only covers the upper 700 m of the water column, with 16 vertical levels. The most recent version ISHIIV6.12 (hereafter ISHIIUPD) spans the period 1945–2011 with the same horizontal resolution, but it covers down to 1500 m with 24 vertical levels. In all cases, the vertical resolution changes from 10 m in the upper levels to 300–500 m in the deepest levels. It is also worth noting here that only ISHII and ISHIIUPD include a time-varying correction for expendable bathythermograph (XBT)/mechanical bathythermograph (MBT) biases [Wijffels et al., 2008; Ishii and Kimoto, 2009]; the correction is not exactly the same for the two products: in ISHIIUPD it has been updated to better adjust the time-varying differences between conductivity-temperature-depth/Bottle/Argo and XBT data.

[7] Density can be derived at each grid cell from temperature and salinity data using the equation of state. Steric sea level anomalies (SLAs) can be then computed at each horizontal location by integrating the density anomaly from the surface down to a reference depth H that can be the bottom depth at that point or any other prescribed level:

display math(1)

where ρS(t) is the surface density and Δρ(z, t) is the density anomaly, defined as the difference between the actual density ρ(z, t) and any arbitrary reference density ρo(z). (Using a constant value for ρS and referring ρ(z, t) to another reference state would lead to practically the same steric sea level time series except for a constant). The thermosteric (halosteric) contribution of steric sea level is determined using the same procedure but keeping salinity (temperature) constant (38 psu and 15°C, respectively).

[8] Mediterranean total sea level from 1993 onward can be obtained from satellite altimetry. Gridded sea level anomaly (SLA) fields were collected from the merged Archiving, Validation, and Interpretation of Satellite Oceanographic data products (available at http://www.aviso.oceanobs.com); namely, we used the Delayed Time updated version v11.0.2, which has all the standard corrections applied. The atmospheric correction applied to low frequencies is simply the inverse barometer response, which implies that the altimetric signal will not account for the redistribution of mass due to atmospheric pressure but will account for redistributions forced by low-frequency winds. For the prealtimetric period, sea level is obtained from a regional reconstruction starting in 1950 [Calafat and Jordà, 2011]. The version used here does neither contain the effect of atmospheric pressure on sea level. Although the spatial structure of the reconstructed sea level may be uncertain, Calafat and Jordà [2011] have shown that the basin averaged sea level derived from the reconstruction is a robust estimate with an RMS error (RMSE) of 0.7 cm.

[9] Observations of the mass component of sea level are obtained from monthly values of equivalent water thickness (http://grace.jpl.nasa.gov/data/mass/) based on the Release-05 gravity field from the Center for Space Research at the University of Texas, Austin, and cover the period from August 2002 to December 2011 [Chambers and Bonin, 2012]. Due to the coarse resolution of gravimetry data, only basin averaged estimates can be considered reliable. Because gravimetry data had been submitted to a spatial averaging smoother (smoothing radius of 500 km), we have applied a restoration factor (1.8) to compensate for the signal loss derived from the smoothing (more details are given in Tsimplis et al. [2013]). Additionally, because a change in the average of the atmospheric pressure over the global ocean will cause a change in the ocean bottom pressure measured by the Gravity Recovery and Climate Experiment (GRACE), we have subtracted the sea level pressure average obtained from the ERA-INTERIM reanalysis [Dee et al., 2011] from GRACE observations.

[10] Subtracting GRACE observations from SLA is assumed to be a good approach for the steric component [Fenoglio-Marc et al., 2006; García-García et al., 2010; Calafat et al., 2010]. These estimates will be used to validate the steric sea level computed from the different hydrographic products for the period 2002–2011. Conversely, subtracting the steric component computed from hydrographic products from total sea level (either SLA altimetry or the reconstruction for the prealtimetric period) will yield an estimate of the mass component. It is worth noting here that the latter accounts not only for changes in the amount of water but also for changes in the salt content [Jordà and Gomis, 2013] (see also the supporting information). Therefore, estimating the salt content of the basin (from the hydrographic databases) and subtracting it from the mass component allow inferring the variations in the freshwater content of the Mediterranean basin.

3 Results

3.1 Interannual Variability

[11] Time series of annual basin averages of the steric, thermosteric, and halosteric components computed down to 700 m (the maximum common depth to all databases) are shown in Figure 1. The four products show similarities in the decadal variability of the steric component (Figure 1a): They all show the pronounced decrease from the middle 1970s to the early 1990s, as well as the sudden increase afterward. However, there are also noticeable differences in the interannual variability, with several periods where products show inconsistent results (e.g., between 1960 and 1965 and between 1970 and 1975). The highest correlation is between EN3 and MEDATLAS2 (0.87), while the lowest correlation is between ISHII and the other products (0.55–0.60). The RMS differences range from 0.77 cm between EN3 and MEDATLAS2 to 1.41 cm between EN3 and ISHII.

Figure 1.

Comparison of Mediterranean average annual time series of (a) steric sea level and the (b) thermosteric and (c) halosteric components. They are all referenced to 700 m (the larger common depth to all products).

[12] Looking at the contributors to steric sea level, the thermosteric component is very consistent among products: They all show a similar decadal and interannual variability with the exception of ISHIIUPD during the period 1970–1980 (Figure 1b). The correlation between products is higher than 0.70 in all cases (0.90 if we do not consider ISHIIUPD). The RMS differences are smaller than 0.87 cm (0.53 cm if we do not consider ISHIIUPD), i.e., below the standard deviations of the EN3, ISHII, and ISHIIUPD series, which are 1.22, 0.90, and 1.21 cm, and higher than the standard deviation of MEDATLAS2 (0.80 cm).

[13] Conversely, the agreement in the halosteric component is lower: No clear decadal variability is observed, and at interannual scale, the products show RMS differences ranging from 0.59 to 1.32 cm (i.e., similar to the standard deviations of the series, which are 1.15, 1.07, 0.98, and 0.62 cm for EN3, ISHII, ISHIIUPD, and MEDATLAS2, respectively). The correlation between the halosteric components of EN3 and MEDATLAS2 is 0.79, between EN3 and ISHIIUPD is 0.72, and between ISHIIUPD and MEDATLAS2 is 0.66, while it is not statistically significant between ISHII and the other three products.

3.2 Long-Term Trends

[14] The basin averaged trends of the steric component have been computed for the period 1960–2000 using different reference levels (Table 1). Focusing first on the estimates obtained for a 700 m reference (those shown in Figure 1), the steric sea level trends are −0.90 ± 0.17, −0.31 ± 0.17, −0.33 ± 0.16, and −0.59 ± 0.12 mm/yr, for EN3, ISHII, ISHIIUPD, and MEDATLAS2, respectively. It is worth noting here that the quoted uncertainties are strictly statistical (i.e., derived only from the length and variability of the series) and therefore they do not account for inaccuracies in the observations or the mapping technique. The relative differences are of the order of 30–50% and are mainly induced by the halosteric component, more particularly by the salinity of the EN3 product. Halosteric trends are −0.55 ± 0.05, −0.14 ± 0.14, −0.21 ± 0.03, and −0.25 ± 0.06 mm/yr for EN3, ISHII, ISHIIUPD, and MEDATLAS2, respectively. That is, the low agreement in the interannual variability of the halosteric component does not translate in this case to the trends, which are statistically consistent except for EN3 (it is worth noting that this consistency only stands when using the 700 m reference depth; see below and Table 1). Thermosteric trends are all consistent, with values between −0.17 ± 0.16 and −0.35 ± 0.17 mm/yr. It is worth mentioning that large discrepancies can be also found when looking at the spatial structure of the trends (see the supporting information).

Table 1. Trends for the Sea Level Components (1960–2000) Obtained When Referring Steric Sea Level to Different Reference Levels and Computed for the Different Productsa
Trends (mm/yr) 1960–2000EN3ISHIIISHIIUPDMEDATLAS2
  1. a

    The total sea level trend estimated from the reconstruction is 1.01 ± 0.21 mm/yr. ND, no data.

Steric Referenced to 300 m
Steric−0.50 ± 0.16−0.09 ± 0.14−0.16 ± 0.16−0.44 ± 0.11
Thermosteric−0.16 ± 0.17−0.12 ± 0.16−0.04 ± 0.16−0.25 ± 0.08
Halosteric−0.32 ± 0.04−0.01 ± 0.04−0.12 ± 0.02−0.20 ± 0.04
Mass1.48 ± 0.251.08 ± 0.151.15 ± 0.151.44 ± 0.21
Salt mass contribution0.36 ± 0.090.01 ± 0.150.12 ± 0.150.22 ± 0.06
Freshwater mass contribution1.18 ± 0.111.14 ± 0.211.05 ± 0.141.21 ± 0.21
Steric Referenced to 700 m
Trends (mm/yr) 1960–2000EN3ISHIIISHIIUPDMEDATLAS2
Steric−0.90 ± 0.17−0.31 ± 0.17−0.33 ± 0.16−0.59 ± 0.12
Thermosteric−0.35 ± 0.17−0.22 ± 0.12−0.17 ± 0.16−0.34 ± 0.11
Halosteric−0.55 ± 0.05−0.14 ± 0.14−0.21 ± 0.03−0.25 ± 0.06
Mass1.88 ± 0.151.33 ± 0.251.32 ± 0.151.58 ± 0.22
Salt mass contribution0.56 ± 0.120.12 ± 0.130.23 ± 0.150.24 ± 0.07
Freshwater mass contribution1.36 ± 0.281.27 ± 0.221.14 ± 0.141.30 ± 0.23
Steric Referenced to Bottom
Trends (mm/yr) 1960–2000EN3ISHIIISHIIUPDMEDATLAS2
Steric−1.29 ± 0.17ND−0.35 ± 0.16−0.56 ± 0.13
Thermosteric−0.03 ± 0.17ND0.09 ± 0.16−0.06 ± 0.12
Halosteric−1.23 ± 0.08ND−0.44 ± 0.04−0.54 ± 0.12
Mass2.23 ± 0.33ND1.35 ± 0.161.55 ± 0.24
Salt mass contribution1.17 ± 0.18ND0.42 ± 0.160.53 ± 0.15
Freshwater mass contribution1.05 ± 0.27ND0.92 ± 0.141.00 ± 0.23

[15] An important result already pointed out by Tsimplis et al. [2011] is that the choice of the reference level has a large impact on the steric trend estimates. Namely, the negative steric trends increase significantly (in absolute value) with the depth of the reference level. Thus, the EN3 steric trends are −0.50 ± 0.16, −0.90 ± 0.17, and −1.29 ± 0.17 mm/yr when referenced to 300 m, 700 m, and the bottom depth, respectively (see Table 1). The other products show similar changes with the exception of MEDATLAS2, which shows similar trends when referred to 700 m and to the bottom.

[16] The impact of the reference level is apparent in both the thermosteric and halosteric components. For the four data sets, the thermosteric trend is more negative when it is referenced to 700 m than when it is referenced to 300 m, indicating that all data sets capture a cooling of intermediate waters in the 300–700 m range. When the deeper layers are included in the computation, the thermosteric trend is reduced in magnitude, indicating that EN3, ISHIIUPD, and MEDATLAS2 capture a warming on the deeper layers. When referenced to the bottom, the thermosteric trends range from −0.06 ± 0.12 mm/yr of MEDATLAS2 to 0.09 ± 0.16 mm/yr of ISHIIUPD. A worth noting feature is that thermosteric trends are statistically consistent among data sets for all reference levels.

[17] The halosteric trends are significantly more negative for all products when using deeper reference levels, which suggests a salinization of the upper, intermediate, and deeper layers. However, trends are not consistent among data sets, so that the magnitude of the salinization is uncertain. This has a direct impact when trying to determine the dominant component of steric trends: From EN3, it could be concluded that salinity dominates the long-term changes of Mediterranean steric sea level, while ISHII suggests that steric sea level is dominated by temperature changes. An interesting point to highlight is that the differences between the products do not increase when changing the reference level.

3.3 The Mass Component

[18] When the mass component of sea level variability is inferred by subtracting the steric component from total sea level, the estimate will obviously be affected by the large uncertainties assessed for the steric component. However, if the steric uncertainty comes mainly from the halosteric component, the freshwater contribution of the mass component could still be determined with an acceptable degree of accuracy.

[19] Trends for the mass component and for the salt and freshwater contributions for the period 1960–2000 are shown in Table 1. The total sea level trend computed from the sea level reconstruction (without the atmospheric pressure effect) is 1.01 ± 0.21 mm/yr. When the steric trend is referenced to 300 m, the trends of the mass component are 1.48 ± 0.25, 1.08 ± 0.15, 1.15 ± 0.15, and 1.44 ± 0.21 mm/yr for EN3, ISHII, ISHIIUPD, and MEDATLAS2, respectively. Using deeper reference layers, the trends in the mass component increase, especially for the EN3 data set, for which it reaches 2.23 ± 0.33 mm/yr if the steric trend is referenced to the bottom.

[20] Looking at the contributors to the mass component (salt and freshwater), the largest discrepancy in the salt trend (expressed in terms of equivalent sea level) comes again from EN3, which gives 1.17 ± 0.18 mm/yr in front of the 0.42 ± 0.16 mm/yr given by ISHIIUPD or the 0.53 ± 0.15 mm/yr given by MEDATLAS2 (all of them referenced to the bottom). Conversely, the trends for the freshwater content are more robust: The differences among data sets are small and always within the range of the statistical uncertainty. Using different reference levels, the trends change, but discrepancies are still statistically consistent (the mean values are 1.14 ± 0.17, 1.27 ± 0.22, and 0.99 ± 0.21 mm/yr when using 300 m, 700 m, and the bottom as the reference depth).

4 Validation Against Satellite Data

[21] Once the significant differences between the different products have been made apparent, a relevant question is which of the products is more reliable. To investigate this issue, the spatial mean of the steric component of Mediterranean sea level derived from satellite observations (i.e., altimetric SLA minus the mass component inferred from GRACE) is compared with the estimates derived from the hydrographic databases for the common period 2002–2011 (Table 2). MEDATLAS2 cannot be included in the comparison because it ends before the onset of the GRACE mission. We use monthly steric sea level referred to 700 m because it is the deepest common level for all the data sets, but the statistics and the main conclusions of the comparison are practically unaltered when using another reference level.

Table 2. Comparison of Mediterranean Averaged Steric Sea Level Computed From the Different Databases With the Steric Sea Level Inferred From Satellite Observations (See Text for Details)a
 Nm (months)Total SignalDeseasoned Signal
  1. a

    The standard deviation (STD) of each time series is also included. We use monthly data, with Nm being the number of months used in the computation. All correlations are significant at 95% level. The steric component is computed using the 700 m reference level.

  CorrelationRMSE (cm)STD (cm)CorrelationRMSE (cm)STD (cm)
EN3950.902.984.250.412.251.70
ISHII590.803.664.170.302.662.13
ISHIIUPD960.863.364.060.372.191.27

[22] The comparison shows that when the seasonal cycle is not subtracted from the time series, the correlation with the satellite-derived steric sea level is high in all cases (>0.80). The reasons are that the signal is dominated by the seasonal cycle, that this is, in turn, dominated by the thermosteric component, and that all data sets correctly reproduce the phase of the cycle. Note however that despite the high correlation, the RMS error is larger than 2.9 cm, which represents around 70–80% of the standard deviation of each time series (~ 4 cm). The best agreement is for EN3 (RMS error of 2.98 cm), followed by ISHIIUPD (RMS error of 3.36 cm) and ISHII (RMS error of 3.66 cm).

[23] When the seasonal cycle is removed, the correlations strongly decrease, being 0.41 for EN3, 0.37 for ISHIIUPD, and 0.30 for ISHII. The RMS errors are 2.19, 2.25, and 2.66 cm for ISHIIUPD, EN3, and ISHII, respectively. They are all larger than the standard deviation of each time series denoting the little representativity of steric estimates from hydrographic databases at intra-annual scales. Unfortunately, the satellite-derived estimates are only available for a 9 year period, which is too short to produce meaningful statistics for interannual variability.

5 Discussion and Conclusions

[24] The estimates of the steric component of Mediterranean sea level variability have usually been given without paying attention to the eventual differences between the available hydrographic gridded products. Here we have shown that the agreement is reasonably good for the basin averaged thermosteric component, especially for the representation of the decadal variability and long-term trends, but for certain periods, discrepancies can be as large as the interannual variability. Most products show a similar spatial distribution of thermosteric trends (positive in the western basin and negative in the eastern one), although the magnitude can differ by up to 0.5 mm/yr among products. Conversely, the halosteric component has been shown to be dependent on the product, which results in significant differences in the interannual variability of the steric component and also in the derived mass component. Although all halosteric trends point toward a salinization of the basin, the magnitude of the change is not statistically consistent among products. Nevertheless, we recall that we use one standard error for the consistency analysis; using 95% confidence intervals and submitting the data to further processing (e.g., a correction for serial correlation), the conclusions regarding the consistency of the halosteric term might be a bit more optimistic.

[25] The validation of the different steric sea level estimates against the steric component inferred from satellite observations for the period 2002–2011 suggests that EN3 is the product performing better, although in all cases, the agreement with observations is low once the seasonal cycle is removed (RMS error larger than the standard deviation). Conversely, the former ISHII product, which has been widely used in Mediterranean studies, is the product performing the worst (see Table 2).

[26] Identifying the origin of the discrepancies among products is anything but straightforward. They can be due to differences in the mapping procedure and/or to differences in the set of profiles used to generate the gridded maps. All products are generated with similar objective analysis techniques: The gridded values are obtained as a linear combination of the observed anomalies with respect to a prescribed background field (or climatology). However, the weights given to the anomalies depend on parameters (correlation length scale and noise-to-signal ratio) that are prescribed differently in each product. Depending on these parameters, for instance, the interpolated field can depart more or less from the climatology, which, in turn, is also different among products. A comparison of the spatial correlation of the temperature and salinity gridded values (not shown) did not yield any clue for the different performances of the products. Moreover, the fact that the largest differences are between ISHII and ISHIIUPD, which, in principle, share the same mapping parameters, suggests that the mapping procedure alone cannot explain the discrepancies among data sets.

[27] Regarding the set of profiles used to generate the gridded maps, MEDATLAS2 is generated from the MEDAR database [MEDAR Group, 2002] while EN3 and ISHII are both generated from the World Ocean Database (WOD05) [Boyer et al., 2006] but with a different quality control [Ishii et al., 2003; Ingleby and Huddleston, 2007; Ishii and Kimoto, 2009]. It is not clear whether all MEDAR data are included in the WOD05 database, but the high correlation between MEDATLAS2 and EN3 products suggests that MEDAR and WOD05 observations must be highly coincident in the Mediterranean. Conversely, EN3 and the two ISHII products are generated from the same database, but they show large discrepancies. This suggests that the quality control procedure that determines which profiles are to be used in the interpolation must be playing a role, but there is not enough information in the literature to confirm this extent.

[28] Another issue that merits attention is the potential impact of time-varying XBT/MBT bias corrections, which are applied in the ISHII products but not in the others. The differences found in the 1970s are probably due to the improvement of the correction implemented in ISHIIUPD, because this is the most sensitive period to XBT/MBTs. However, although the impact of the correction is definitely important for the 1970s, it is difficult to conclude that this can explain by itself the better performance of ISHIIUPD. A main reason is that the thermosteric component (the one affected by the correction) is rather consistent among products, in comparison with the total steric component. Moreover, the fact that ISHII and ISHIIUPD also differ apparently in the last decades (when the importance of XBT/MBTs is very minor) suggests that other updates of ISHIIUPD could also play a significant role.

[29] The worse agreement in the halosteric component with respect to the thermosteric one is very likely due to the lack of salinity observations, especially at intermediate and deep layers. When there are few observations, the impact of the mapping procedure and of the number and quality of observations onto the final gridded product is enhanced. In the Mediterranean, the number of historical observations rapidly decreases with depth [Vargas-Yáñez et al., 2009], with the number of salinity profiles being less than a half of the number of temperature profiles. The lack of deep salinity observations is particularly harmful when noting moreover that in the Mediterranean Sea, the relative impact of intermediate and deep layers on vertical averages is higher for salinity than for temperature [see, e.g., Calafat et al., 2012].

[30] Finally, it is important to note that the thermosteric and halosteric estimates are integral measurements of the heat and salt contents of the basin. While the first is better determined, our results suggest that the interannual variability of the Mediterranean salt content and its long-term trend are rather uncertain so far. To be more precise, the uncertainties linked to the data set choice are as large as the observed changes. Although the uncertainties on the salt content affect the derived estimates of the long-term mass component of sea level, the estimation of the freshwater contribution to the mass component is more robust with respect to the product and reference depth choice.

Acknowledgments

[31] This work has been carried out in the framework of the projects VANIMEDAT-2 (CTM2009-10163-C02-01, funded by the Spanish Marine Science and Technology Program and the E-Plan of the Spanish Government) and ESCENARIOS (funded by the Agencia Estatal de Meteorología). G. Jordà acknowledges a JAE-Doc contract funded by the Spanish Research Council (CSIC) and the European Science Foundation. We thank M. Rixen and Météo-France for supplying the updated MEDATLAS2 database and the HyMeX database teams (ESPRI/IPSL and SEDOO/Observatoire Midi-Pyrénées) for their help in accessing the data. We thank the Met Office Hadley Centre observations for providing the ENACT/ENSEMBLES data set and M. Ishii for kindly providing the ISHII data sets. GRACE ocean data were processed by Don P. Chambers, supported by the NASA MEASURES Program. Altimetry data have been provided by AVISO (http://www.aviso.oceanobs.com/).

[32] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.

Ancillary