The 25-year period (1993–2017) involves 76 mission years and 12 different altimeters. The evolution of the altimeter constellation is shown in Fig. 1. The most notable change in the constellation compared to DT2014 concerns the availability of data from the Sentinel-3A and Hayaing-2A altimetry missions. For Sentinel-3, an additional 6 months of data (from June to December 2016) have been incorporated into the system. For Hayaing-2A, data from March 2016 to February 2017 have also been added.

DUACS takes level 2P (L2P) altimetry products as its input data. These data are disseminated by CNES and EUMETSAT. L2P products are supplied as are distributed by different agencies: NASA, NSOAS, ISRO, ESA, CNES, EUMETSAT. They include the altimetry standard, which includes the algorithms and parameters used to retrieve the sea level anomalies from the altimeter measurements (i.e., instrumental, geophysical and environmental corrections together with mean sea surface – MSS), and a validity flag that is used to remove spurious measurements.

Indeed, the altimeter measurement is affected by various disturbances (atmospheric, instrumental, etc.) that must be estimated to correct it. Specific corrections are also applied to remove high-frequency signals that cannot be taken into account in the DUACS processing (Escudier et al., 2017). Dynamic atmospheric correction (DAC) and ocean tide correction are the two main examples. The DUACS DT2018 global reprocessing was an opportunity to take into account new recommendations and new corrections from the altimetry community (Ocean Surface Topography Science Team, OSTST).

The altimetry standards were carefully selected in order to be as consistent and homogeneous as possible between the various missions, whatever their purpose (in particular the retrieval of mesoscale signals or climate applications). This selection was made possible between 2014 and 2017 in the framework of phase II of the ESA's Sea Level Climate Change Initiative (SL_cci) project. Part of the project activities included selecting a restricted number of altimetry standards (Quartly et al., 2017; Legeais et al., 2018a). Table 1 presents the altimetry standards used in DT2018 and the changes compared with the previous version (written in bold). The orbit standards from Jason-1, Jason-2, Cryosat-2, ALtiKa, Jason-3 and Sentinel-3A altimeter missions were upgraded from precise orbit estimation (POE-D) to a new POE-E. The new POE-E standards are of a very high quality (Ollivier et al., 2015; AVISO, 2017b). In this version, the main developments concern the evolution of the gravity field model that has a positive impact on regional MSL error and greatly reduces geographically correlated errors.

Various corrections were updated, of which the new MSS CNES-CLS-15 and ocean tide model (FES2014) have led to the greatest improvements in product quality. Valuable enhancements were made in the MSS to improve performance at short wavelengths (Pujol et al., 2018a). Furthermore, the sea level in coastal areas and the Arctic region is determined more accurately in the updated version, and errors were greatly reduced globally. Concerning the ocean tide correction, FES2014 is the latest version of the FES (finite-element solution) tide model developed between 2014 and 2016. This new release gives improved results in the deep ocean, at high latitudes and in shallow–coastal regions (Carrère et al., 2016).

DUACS processing involves an initial preprocessing step during which data from the various altimeters are acquired and homogenized. Next, along-track products (L3) and multi-mission gridded products (L4) can be estimated. Finally, the derived products are computed and disseminated to users. This section is not intended to describe the entire data processing system in detail, but rather to expose the major changes made for this DT2018 version. For a detailed description of DUACS processing, readers are advised to consult Pujol et al. (2016).

Two different types of sea level gridded altimetry products are available in DT2018 version. The first type, produced and distributed within the Copernicus Marine Service (CMEMS), is dedicated to mesoscale observation. The other type, produced and distributed within the Copernicus Climate Change Service (C3S), is dedicated to monitoring the long-term evolution of the sea level for use in climate applications and for analyzing ocean–climate indicators (such as global and regional MSL evolution). Two types of altimeter processing configurations are exploited to build these two products. The first difference of configuration is related to the number of altimeters used in the satellite constellation.

Mesoscale observation requires the most accurate sea level estimation at each time step, along with the best spatial sampling of the ocean. All available altimeters are thus included in CMEMS products, and the sampling can vary with time depending on the constellation status. In contrast, the temporal stability of surface sampling is more important when monitoring the long-term sea level evolution. A steady number of altimeters (two) is thus used in C3S products. This corresponds to the minimum number of satellites required to retrieve mesoscale signals in delayed-time conditions (Pascual et al., 2006; Dibarboure et al., 2011). Within the production process, long-term stability and large-scale changes are established on the basis of records from the reference missions (TOPEX–Poseidon, Jason-1, Jason-2 and Jason-3) used in both CMEMS and C3S products. Any additional missions (e.g., as many as five additional missions in 2017) are then homogenized with respect to the reference missions and help to improve mesoscale process sampling, providing high-latitude coverage and increasing product accuracy. However, the total number of satellites has greatly varied over the altimetry era and biases may develop when a new satellite on a drifting orbit is introduced. Each addition may affect the stability of the global and regional MSL by several millimeters (data not shown here). Although spatial sampling is reduced when there are fewer satellites, the risk of introducing such anomalies is thus also reduced in C3S products, resulting in improved stability. In CMEMS products, stability is ensured by the calibration with the reference missions, and mesoscale errors are reduced due to the improved ocean surface sampling made possible by using all the satellites available in the constellation.

As a second difference of configuration, the reference used to compute sea level anomalies for C3S products was an MSS for all missions, whereas for CMEMS products, an MP was used along the theoretical track of satellites following a repetitive orbit (see Sect. 2.3.2). Considering the regional mean sea level temporal evolution, the combined use of MSS and MP for successive missions in the merged product gives rise to regional centimetric bias (data not shown here). Consequently, the systematic use of MSS for all missions has been privileged in the C3S products to ensure MSL stability, and the use of MP for repetitive missions has been selected in the CMEMS products to increase their accuracy.

The differences between CMEMS and C3S product quality are discussed on a climate scale in Sect. 3.4.

Optimizing the mapping process (Sect. 2.3.3) and incorporating the new altimetry corrections (Sect. 2.2) had a direct impact on the observation of ocean sea level and surface circulation dynamics in the gridded products. To characterize this impact, the difference between DT2014 and DT2018 temporal variability is shown in Fig. 3. An additional variance between 2 % and 5 % is observed for high-variability regions in DT2018 products. This increase is due to having changed the OI spatial and temporal scales used in the mapping process and decreased the suboptimal interpolation window size. The OI selection window is more focused on close observations (both spatial and temporal). In coastal areas, a substantial reduction in SLA variance is observed due to both the FES2014 tidal correction and, to a more limited extent, the new MSS. For the tidal correction, Carrere et al. (2016) have shown a reduction in SLA variance at nearshore crossovers. Pujol et al. (2018a) have emphasized that the new gridded MSS shows less SLA degradation near the coast. These improved standards contribute to a valuable local reduction in SLA variance (up to 50 % alongshore). At high latitudes, the difference of variance is significant (±100 to ±200 cm2) and is due to the new MSS correction. Indeed, Pujol et al. (2018a) have shown that the CNES_CLS 2015 MSS improves both coverage in the Arctic and the resolution of the shortest wavelengths at high latitudes.

Compared to DT2014, the new version reveals more intense western boundary currents (geostrophic part). This has a direct impact on the eddy kinetic energy (EKE) derived from these products. Figure 4 presents the spatial difference in the mean EKE over the global ocean between DT2018 and DT2014 products, along with their temporal evolution. As observed before for the differences of SLA variance, a higher energy is evident in high-variability areas. This corresponds to a 2 % increase in EKE in DT2018. However, in the equatorial belt (±20∘ N), the EKE in DT2018 is lower (-17 %). This is a direct consequence of the noise measurement that is taken into consideration in the mapping process for all satellites: observation errors prescribed during OI in the tropical belt have been increased, so the SLA signal is smoother and less energy is observed in this region. In coastal areas, the DT2018 version presents fewer spurious peaks of high EKE (Fig. 4b). As already stated, this is related to the improved altimetry correction and lower SLA variance. Considering the mean EKE time series, a global reduction of 26 cm2 (17 %) is observed for dataset DT2018. This is directly due to the lower tropical EKE. Another important point to note is that the standard deviation of EKE in these products is lower than in DT2014. This illustrates that EKE variations are less important, and there are fewer isolated anomalies (and these are mostly coastal) in the new DT2018 products.

The gridded SLA accuracy was estimated by comparison with independent along-track measurements. Maps produced by merging only two altimeters were compared with SLAs measured along-track from the tracks of another mission that was kept independent from the mapping process (see Pujol et al., 2016, for the full methodology). TOPEX–Poseidon interleaved was compared with gridded products that merged Jason-1 and Envisat over 2003–2004. It must be pointed out that these results are much more representative of gridded products combining two altimetry missions. Products combining all available missions can usually benefit from improved sampling when three to six altimeters are used. The errors described here should thus be considered the upper limit. Table 3 summarizes the results of comparisons over different areas. Figure 5 shows the percentage of the difference in variance between gridded products and TP independent along-track measurements for DT2018 and DT2014 products. The gridded product error for mesoscale wavelengths ranges between 1.4 cm2 (for a low-variability area) and 37.7 cm2 (for a high-variability region). The improvements in DT2018 compared with DT2014 affect all areas. Offshore, the improvement is fairly low (around 3 %) and is associated with the enhanced version of the OI mapping parameter. In coastal areas, the improvements are more significant (around 10 %) and caused by the new tidal correction (FES2014) and, to a lesser extent, the MSS and MPs. In the tropical belt, improvements are also significant (around 9 %) and related to the observation errors that were increased in this area for the OI processing.

Absolute geostrophic currents for DT2018 were assessed using drifter data for the 1993–2017 time period. The AOML (Atlantic Oceanographic & Meteorological Laboratory) database was used for the comparison (Lumpkin et al., 2013). These in situ data were corrected for Ekman drift (Rio et al., 2011) and wind if a drifter's drogue had been lost (Rio, 2012) so as to be comparable with the altimetry absolute geostrophic currents. Drifter positions and velocities were interpolated using a 3 d low-pass filter in order to remove high-frequency motions (Rio et al., 2011). The absolute geostrophic currents derived from altimetry products were then interpolated onto drifter positions for comparison.

The distribution of the current's intensity shows an overall underestimation of magnitude in altimetry products compared to drifter observations (data not shown). Figure 6 shows the RMS difference between the DT2018 geostrophic current and that of drifters. The mean RMS is nearly 10 cm s-1 and the main errors are located nearshore and in a high-variability region with peaks higher than 20 cm s-1. Taylor skill scores (Taylor, 2001) were computed for the zonal and meridional components of the current in DT2018. This assessment took into consideration both the signal's correlation and its standard deviation. Results are quite robust: 0.89 for the zonal and 0.87 for the meridional component.

Table 4 summarizes the mean RMS of the differences between geostrophic current maps and drifter measurements over different areas for versions DT2018 and DT2014. DT2018 products are more consistent with drifter measurements than DT2014 version products. The improvement is clearly visible in the intra-tropical belt. The variance of the differences with drifters is reduced around 20 % to 40 % in this area. Additional noise-like signals present in the DT2014 version had reduced consistency with drifter measurement (Pujol et al., 2016). This degradation was corrected for by the change in mapping parameters used for this updated version (Sect. 2.3.3). A significant improvement can also be observed in coastal areas, where the variance of differences with drifter measurements is reduced by nearly 15 % (Table 4). Elsewhere, this reduction in the variance of difference ranges from 4 % to 7 %.

As described in Sect. 2.3.1 and 2.3.2 the new DUACS DT2018 processing has a key impact on coastal areas, and overall, all missions have more measurements available in DT2018 compared to DT2014.

The assessment of gridded products in coastal areas included a comparison with tide gauge (TG) measurements. We used mean monthly TG measurements from the PSMSL network (Permanent Service for Mean Sea Level; PSMSL, 2016) from 1993 to 2017. We used only long-term monitoring stations with a lifetime of more than 2 years. Sea surface height measured by TG was compared with gridded SLA by considering the maximum correlation with the nearest neighboring pixel (Valladeau et al., 2012; AVISO, 2017a). In Fig. 7 the variance of the difference between DT2018 altimetry products and TG measurements is compared with that obtained from the differences using DT2014 altimetry products. The results show a global reduction in the variance (0.6 %) when DT2018 data are used. There is a clear improvement along the Indian coast, Oceania and northern Europe. Local degradation can be observed along the coast of Spain and along the US west coast. These degradations, which are not observed in other diagnoses such as independent along-track measurements, still need to be further investigated.

The global mean sea level (GMSL) is a key indicator of climate change since it reflects both the amount of heat added in the ocean and the land ice melt coming mainly from Antarctic and Greenland ice sheets and glaciers. Three different altimeter products can be used to compute three GMSL estimates: the time series of the box-averaged along-track measurements of the reference missions only (Ablain et al., 2017) and L4 merged gridded sea level products from CMEMS and C3S (e.g., Fig. 8a). For the same product versions and computation periods, these three GMSL estimates are considered to be equivalent since almost the same altimetry standards are used to compute sea level anomalies, and the long-term stability for all products is ensured by using the same reference missions. The remaining GMSL differences observed (∼0.17 mm yr-1) are not significant given the uncertainty on different scales (the uncertainty in the GMSL trend is 0.4 mm yr-1 at the 90 % confidence level given by Ablain et al., 2019). Note that as previously mentioned (Sect. 2.4), differences can be found between the two different Copernicus gridded products (CMEMS–C3S) when computing regionally averaged MSL.

When computing area-averaged MSL time series, users are advised that DUACS products are not corrected for the effect of glacial isostatic adjustment (GIA) due to post-glacial rebound. A GIA model should be used to estimate the associated sea level trends.

In addition, between 1993 and 1998, GMSL is known to have been affected by instrumental drift in the TOPEX-A measurement, as quantified by several studies (Watson et al., 2015; Beckley et al., 2017; Dieng et al., 2017). The sea level altimetry community agrees that it is necessary to correct the TOPEX-A record for instrumental drift to improve accuracy and reduce uncertainty in the total sea level record. However, there is no consensus so far on the best approach to estimate drift correction at global and regional scales. DUACS sea level altimetry products are not corrected for TOPEX-A drift, pending ongoing TOPEX reprocessing by CNES and NASA/JPL, but users can apply their own correction. Adjusting for this TOPEX-A anomaly creates a GMSL acceleration of 0.10 mm yr-2 for the 1993–2017 time span that does not otherwise appear (WCRP Global Sea Level Budget Group, 2018).

Figure 8a shows the GMSL temporal evolution and associated trend computed with the new DT2018 and former DT2014 versions of DUACS C3S products. In the latest version, the global mean sea level trend is 3.3 mm yr-1 (including GIA correction of -0.3 mm yr-1). The origin of the associated uncertainty is discussed by Legeais et al. (2018b). The map of the differences for the local MSL trend derived from the latest and previous product versions (Fig. 8b) displays a pattern predominantly associated with the different orbit standards used in the two product versions (GDR-E versus GDR-D; see Table 1). Such a result is confirmed by comparing altimetry products with the independent dynamic height measurements derived from in situ Argo profiles (Valladeau et al., 2012; Legeais et al., 2016).

As previously discussed for global ocean products, the quality of regional gridded SLA products is estimated through comparison with independent altimeter along-track and tide gauge measurements.

Figure 9 shows the spatial distribution of the RMS of the differences between regional DT2018 SLA gridded products and independent along-track measurements (TOPEX–Poseidon interleaved along-track measurements over the 2003–2004 period). The main statistics for these comparisons, as well as a comparison with the previous DT2014 version, are also given in Table 5. In contrast with the global products assessment, the evaluation of regional products cannot include the mesoscale signal analysis: the short length of the tracks segments available over the regional seas does not allow for the accurate filtering of the signal in order to focus specifically on the mesoscale. The results obtained show that for the DT2018 Mediterranean product, the main errors are located in coastal areas and in the Adriatic and Aegean seas, with RMS values ranging from 6 to 9 cm. The Black Sea products also show higher errors in coastal areas (results not shown here). The mean variance of the differences between gridded products and along-track measurements is nearly 17 and 23 cm2 over the Mediterranean Sea and the Black Sea. This value is higher than the mean error observed over low-variability areas in the global ocean (Table 3), mainly due to the different wavelengths addressed in these comparisons. Compared to the previous regional DT2014 version, the error is reduced by 4.2 % for the Mediterranean Sea and 3.5 % for the Black Sea. It is important to note that these results are representative of gridded product quality when only two altimeters are available. These products can be considered to be degraded products for mesoscale mapping since they use minimal altimeter sampling.

Compared to the previous version, consistency with monthly TG measurements (Fig. 10) is improved locally in the regional DT2018 Mediterranean gridded product in the western part of the Mediterranean basin. Degradation is observed, however, in some other coastal areas, especially in the center of the basin and along the Turkish coast. For the Black Sea gridded product, only nine tide gauges were available for the comparison. With the exception of a tide gauge at the eastern end of the Black Sea on the Georgian coast, these DT2018 regional products are improved of the order of 1 %.

DT2018 regional absolute geostrophic current in the Mediterranean basin was assessed using drifter data for the 1993–2017 period. The data were collected from drifters released in the Mediterranean Sea as part of the AlborEx (Pascual et al., 2017) and MEDESS-GIB (EU MED Program; http://www.medess4ms.eu/, last access: 9 September 2019; Sotillo et al., 2016) multi-platform experiments as well as other experiments incorporated into CMEMS In Situ Thematic Centre (INS TAC) products. These data are processed similarly to the global product (Sect. 3.2).

Table 6 summarizes the main statistical results for the whole basin. The DT2018 regional product presents a correlation coefficient with drifter data 4 % greater than that obtained when using the DT2014 regional product. Moreover, the errors in the later version are slightly lower at 1 %, whilst its improvement in explained variance is as high as 14 %.

The analysis was then repeated for the different dynamical subregions of the basin (see Fig. 11a) reported by Manca et al. (2004). This differentiation is based on the typical permanent features in the upper 200 m of the water column. Overall, comparisons between geostrophic velocities derived from the DT2018 regional product and absolute surface velocities retrieved by the drifters (Fig. 11b–e) reveal a correlation coefficient greater than 0.40 in most of the boxes. Correlations greater than 0.50 are mainly located in the southernmost part of the basin where stronger mesoscale activity occurs, namely the Alboran Sea (DS1), the Algerian Basin (DS3 and DS4), the Sardinian Channel (DI1), the Strait of Sicily (DI3), the Ionian Sea (boxes DJ7, DJ8 and DJ5) and the Cretan passage (DH3). The overall RMS difference between the two datasets ranges between 8 and 11 cm s-1, although it reaches 20 cm s-1 in DS1 due this area's strong dynamics. Slightly larger errors are obtained when comparing the DT2014 product with drifter observations (not shown here). Furthermore, drifter data collected in boxes DS1, DS3 and DS4 have the largest variability due to the aforementioned mesoscale activity. This fact is also reflected in the two altimetry products, which have the largest variance values in the Mediterranean basin.

Overall, the correlation coefficient between the DT2018 regional product and in situ drifter data is improved by 5 %–10 % with respect to that obtained when using the DT2014 product (Fig. 11g). Here, positive values denote an improvement in DT2018 over DT2014. This fact is mainly observed in areas of strong mesoscale activity. Moreover, the errors (Fig. 11f) are reduced by 2 % in the northernmost part of the western Mediterranean basin and Adriatic Sea. However, negative values lower than 2 % (slightly larger errors when using DT2018) are observed in the Algerian Basin and most of the eastern part of the Mediterranean basin. The main improvement in DT2018 with respect to DT2014 lies in the variance explained (Fig. 11h), which presents values nearly 20 % higher in the later product in some areas of the western part of the basin and nearly 10 % higher in the eastern part. This is due to the better capturing of mesoscale activity. This improvement is not observed in the northernmost part of the basin, where less mesoscale activity occurs.