The North Sea is a semi-enclosed marginal sea of the North Atlantic Ocean situated on the northwestern European Shelf . It covers an area of about 560 000 km2 (970 by 580 km). It opens to the Atlantic through the Norwegian Sea in the north and has a second much smaller connection through the English Channel in the south. Moreover, it is connected to the Baltic Sea in the east. Its mean depth is around 94 m, with shallow areas less than 10 m deep in the southern part and much deeper parts (up to 700 m) in the Norwegian Trench area and parts of the Skagerrak. Sea level dynamics in the North Sea are driven by various forcings, namely tides (mainly semi-diurnal with a tidal range of up to 8 m), wind and atmospheric pressure, and heat and water exchanges, as well as river runoff and forcing from open boundaries . The influence of wind on the sea level is large because the North Sea is very shallow . The general circulation pattern in the North Sea is mainly cyclonic. The water flows southwards along the coastal areas of the British Isles, continues eastwards and northwards along the coasts, and finally leaves the basin as the Norwegian Coastal Current .

This study makes use of observation data collected in an area between 4∘ W and 12.2∘ E longitude and 50 and 61∘ N latitude, except the regions of the Irish Sea and the Baltic Sea (Kattegat).

Satellite altimetry has been providing sea surface height information since the launch of the SeaSat mission in 1978. But only since 1992 have at least two simultaneously measuring satellites been in orbit, ensuring more reliable and precise height observations as well as improved data coverage and temporal resolution. This study includes almost all available missions since ERS-2 in 1995, namely TOPEX (TP), ERS-2, Jason-1 (J-1), Envisat, Jason-2 (J-2), CryoSat-2 (CS-2), SARAL, Jason-3 (J-3), and Sentinel-3A/B (S3-A/B) (ordered by launch dates). Data from the first phase of TP and from ERS-1 are not used because of doubts concerning their trend stability . North SEAL comprises the period between May 1995 and May 2019 (see Fig. ) and makes use of the latest data for all missions, updated by consistent external geophysical model corrections and ITRF2014-based orbits whenever available; see Sect.  for more details.

Monthly mean water level measurements of tide gauges are used for comparison and validation. The data are derived from the datum-controlled database of the Permanent Service of the Mean Sea Level (PSMSL) and from the German Wasserstraßen- und Schifffahrtsverwaltung des Bundes . The WSV provides TG data for the German Bight with measurements at semi-diurnal tidal maxima and minima, i.e., at about 6 h resolution, which are first smoothed with a 2 d running mean filter and then down-sampled to monthly mean observations. Values are retained which pass a 3-σ outlier test.

Among all available stations in the North Sea region, those are selected that contain at least 80 % valid measurements during the study period (1995–2019), resulting in a total of 54 stations. The same monthly averaged dynamic atmospheric correction (DAC) is applied as used for the altimetry data . To match the DAC with the tide gauge records, among the nine closest points of the DAC grid the one that results in the highest variance reduction is selected. TG data are not corrected for ocean tides, which are assumed to be removed by monthly averaging. Remaining influences from long-period tidal effects are assumed to have a smaller impact on the estimated trends than errors of ocean tide models would have directly at the coast.

In order to make trends determined from TG data comparable to absolute (i.e., geocentric with respect to an ellipsoid; ) sea level trends from satellite altimetry, the relative TG measurements need to be corrected for VLM.

VLM can be estimated from point measurements (e.g., from Global Positioning System – GPS – observations) or from regional or global models. Since these estimates differ significantly, data from different sources are used in this study. Beside GPS data and glacial isostatic adjustment (GIA) models, the nonlinear effect of contemporary mass redistribution (CMR) on VLM is also applied.

GPS trends are derived from the dataset of the Nevada Geodetic Laboratory (NGL) , which contains more than 17 000 vertical velocities (IGS08 reference frame). Only GPS VLM estimates from measurements with a minimum record length of 5 years between 1995 and 2019 are considered. To combine the VLM estimates with the TG trends, the nearest GPS station within a radius of 50 km around a TG is used.

GIA VLM estimates are taken from two different models. The first one, ICE-6G D (VM5a) , was refined by geodetic constraints primarily by GPS observations from the Jet Propulsion Laboratory over 1994–2012 and from other complementary observations, such as very long baseline interferometry (VLBI) and Doppler Orbitography and Radiopositioning Integrated by Satellite (DORIS) . The model fit to those observations was improved by modifications of the glaciation history. The second GIA VLM estimate is from (hereinafter C18). The solution is based on an ensemble of 128 000 model runs. Among those, the highest likelihood of parameters describing the ice history and 1-D Earth structure was identified from an inversion of GPS and relative sea level data using Bayesian statistics. The GIA estimate represents the expectation of the most likely GIA signal of the ensemble, and formal uncertainty estimates are directly inferred from the Bayesian statistics.

Ongoing changes in terrestrial water storage and mass changes in glaciers and ice sheets cause elastic responses of the Earth, which can result in nonlinear vertical movements. These effects from CMR are not captured by GIA models and only partially detected by GPS observations due to the relative shortness of the record lengths. Using GRACE satellite gravimetry observations, showed that associated time-varying solid Earth deformations can lead to very different trends (of the order of mmyr-1), depending on the time period considered during the last 2 decades. Therefore, VLM estimates from GIA are supplemented with CMR-related land motions as used and distributed by . This estimate is based on a blend of GRACE and GRACE-FO observations during 2003–2018, as well as process model estimates, observations, and reconstructions for the period 1900–2003 (see , for a detailed description of the datasets and sources used).

Figure  gives an overview of the applied VLM corrections. There are some noticeable differences between the two GIA estimates. In particular, the dipolar feature spanning from Scotland to the German Bight in the ICE-6GD solution (Fig. a) is much less pronounced in C18's estimate (Fig. b). The linear CMR signal, which is computed over 1995–2019 (Fig. d), shows a non-negligible uplift signal of about 0.7 mmyr-1 in the entire region of the North Sea.

For validation purposes, North SEAL is compared with other gridded altimetry products available for the region. These products are provided by CMEMS and SL_cci . Both use a regular grid and have a coarser spatial resolution of 0.25∘. Given that the SL_cci dataset only lasts until 2015 (at the time of writing), all the following comparisons described in Sect.  are only performed over the period May 1995–December 2015.

In order to generate monthly sea level anomaly (SLA) grids, the altimetry along-track observations go through a chain of several preprocessing steps, including a retracking for conventional altimeters by the ALES retracker and for SAR altimetry by the ALES + SAR retracker , an empirical adaption of the physical ALES+ retracker to SAR waveforms. Moreover, a relative multi-mission cross-calibration referencing all altimetry missions used to the TOPEX (and later to the Jason) data is performed. Sea level anomalies are computed using the following equations. SSH=Horbit-(R+WT+DT+IONO+OT+SSB1+DAC+SET+PT+RC)2SLA=SSH-MSSH Here, Horbit, R, and MSSH mean the orbital height of the satellite referred to the TOPEX/Poseidon ellipsoid, the altimeter range, and the mean sea surface for reducing sea surface heights (SSHs) to SLAs. In this study, the mean sea surface DTU15MSS is used, which includes 20 years of data from the period 1992–2012. For correcting atmospheric loading and wind effects, the DAC based on operational atmospheric products is used. Even if corrections based on reanalysis data (i.e., DAC-ERA; ) may improve the results for the early years, DAC is the only product currently available for the full period under investigation. The other terms in Eq. () describe several geophysical and atmospheric effects, which are considered for SSH computation. They are listed in Table . The sources of the orbital heights Horbit are provided in the Appendix in Table .

In a post-processing step, the along-track sea surface heights are cleaned from possible outliers by applying the following four steps.

Distance to coast: this is the elimination of observations closer than 3 km to the coast (TOPEX: 5 km).

More details on this flagging are provided by .

All observations passing the outlier elimination are introduced into a least-squares gridding procedure (e.g., ). They are interpolated onto a triangular mesh (geodesic polyhedron) in order to generate monthly grids with a spatial resolution between 6 and 8 km. The gridding procedure mainly follows the process flow introduced by and . It is therefore only briefly described in the following text passages. Mean SLAs per grid node are computed by fitting an inclined plane (h) to the along-track observations. 3h(x,y)=c0+c1x+c2y A local Cartesian coordinate system (x,y) is defined around each grid node, which represents the origin. The grid node height is provided by the coefficient c0, and c1 and c2 are the slope coefficients. They are not used for the following processing. All along-track observations within a radius, the so-called cap size, of 150 km around each grid node are considered. They are spatially averaged, whereby a Gaussian weighting in consideration of their distance to the grid node is applied. The minimum weight at the cap-size edge is set to circa 10 %. Furthermore, uncertainty information is added to the least-squares approach based on the MAD of sea level anomalies per mission and month within a sub-area (0 to 6∘ E and 54 to 58∘ N). The chosen area is free of topographic features. The MAD provides a rough estimate of the SLA noise level of a certain mission in a certain month. More information is available in .

Within the gridding procedure, the observations are filtered again in order to reject outliers from the estimation of the coefficients. This is done by performing a three-stage outlier detection.

Application of a standard 3-σ criterion to sea level anomalies within the cap size.

The monthly grids undergo a final check by removing sea level anomalies that exceed a threshold of ±2m. For each grid node, the monthly mean SLAs are provided together with an estimate of the uncertainty. In addition, a quality flag indicates if the node can be safely used or should be handled with care. This flag is allocated according to the standard deviation per node. If the standard deviation exceeds a specified threshold and the node has less than 280 months of valid data, it is labeled as “bad” quality (flag: 1). The threshold is set to the 90th percentile of all SLA standard deviations averaged over time.

The monthly SLA grids provide the basis for estimating a sea level trend per grid node as well as the amplitude and phase of the annual cycle. While a linear trend is fitted to the data, the annual cycle amplitudes are obtained from half of the differences of the months with the maximum and minimum multi-year monthly means. The uncertainties are based on the combined uncertainties of these months.

Trend uncertainties are derived while accounting for autocorrelated errors in the data using maximum likelihood estimation (MLE). To identify the most appropriate noise model required to accurately estimate the trend uncertainties, we investigate the fit of a variety of different stochastic noise model combinations as done in, e.g., . These are an autoregressive AR(1) noise model, a power law plus white noise, a generalized Gauss–Markov (GGM) plus white noise, a Flicker noise plus white noise, and an autoregressive fractionally integrated moving-average (ARFIMA(1,d,0)) model. For the considered domain, we find that on average (of all altimetry observations in the North Sea) the AR(1) has the lowest mean (or median) values of the Akaike information criterion (AIC; ) and the Bayesian information criterion (BIC; ). Thus, this model is selected to assess formal parameter uncertainties. Uncertainties of the linear trend and the annual amplitude are given at the 95 % confidence interval.

For a more detailed description of the parameter estimation please refer to .

To evaluate the performance of North SEAL, we compare absolute sea level trends and annual amplitudes derived from satellite altimetry and TGs. Moreover, correlations and rms between the two time series at TG locations are analyzed. It should be highlighted that taking the closest altimetry point to a TG is not always necessarily the highest correlated or most representative observation. Thus, to match the altimetry sea level data with the TG measurements, we follow the approach of , which only uses the most highly correlated data in the comparison instead of taking the altimetry observation closest to the TG. showed that this approach ensures increased consistency of along-track altimetry and TG observations and enhances the agreement of trends. Here, 20 % of the best-correlated gridded altimetry data within a radius of 200 km around a TG are selected. This region is hereinafter called the zone of influence (ZOI). Time series of sea level are computed from spatial averages of the altimetry data within the ZOI. Absolute sea level trend deviations between altimetry and TGs are subsequently derived by subtracting the TG measurements from the averaged altimetry data, whereby the correction for VLM is applied.

The uncertainties of absolute TG sea level trends u are based on the combined uncertainties of the TG and the VLM trends (u=uTG2+uVLM2). In order to compute uncertainties of trend differences (for significance tests), the altimetry uncertainties are also taken into account. Differences (and their uncertainties) in the annual amplitudes are computed by subtracting the amplitudes from the individual altimetry and TG time series and by computing the combined uncertainty of the annual amplitudes.

We note that the approach of using the ZOI significantly improves the comparability of sea level trends. Accordingly, the trend differences presented in Sect.  are on average about 20 % smaller when using the ZOI approach instead of taking the closest altimetry grid point.

In contrast to trends and the annual cycle, correlations and rms are computed at the closest valid (i.e., with at least 280 unflagged months of data available) altimetry node within a radius of 150 km. Correlations are derived from the detrended altimetry and TG time series. We apply the quality flag to the SLA dataset before computing correlations. This reduces the number of TG–altimetry pairs to 52. We also use the same number of TGs (52) for the comparison with other altimetry products.

Next to correlations, we also analyze the rms difference of TG and altimetry time series in order to study the spread between the data. To account for datum differences between TGs and altimetry, offsets are removed from the difference time series. Three different solutions are computed: monthly detrended, monthly detrended and deseasoned, and annual detrended. In order not to distort the annual values by incomplete years, only the period from January 1996 to December 2016 is used for all comparisons. To deseason the data, we subtract the multi-year monthly averages. Annual averages are studied to compare the variability on interannual timescales and to also assess the correctness of the representation of lower-frequency processes.

Figure  shows the mean SLA averaged over the observation period of 24 years. In addition, three selected time series at different locations in the North Sea are displayed. As described in Sect. , the SLAs are referenced to the DTU15 mean sea surface. Since the input data and time period of DTU15MSS (1992–2012) and North SEAL (1995–2019) are different, the SLAs do not average to zero everywhere. A geographical pattern of negative offsets around 2 cm is visible in large parts of the coastal areas of the North Sea, especially along southern Norway and the coasts in the south. This effect is probably due to the use of a global MSS product that uses a different set of retracked data and geophysical corrections. It has no influence on the time series analysis of SLA, in particular not on derived trends and annual amplitudes derived in this study. In fact, most of the affected regions are edited when taking the quality flag (see Sect. ) into account.

The three SLA time series on the right-hand side of Fig.  show the temporal evolution of sea level for different grid nodes. A distinct annual oscillation and interannual changes are clearly visible for the three locations. Moreover, all three curves show a rise of sea level.

The monthly SLA time series displayed in Fig. b–d suggest that the sea level variability differs strongly over the domain. In agreement with, e.g., and , the variance of the time series towards the German Bight (Fig. c) strongly exceeds the variance observed at the Norwegian and British coastlines (Fig. b and d). Using long TG records, demonstrated that the sea level variability at the southern and eastern coastlines of the North Sea is particularly dominated by westerly winds, which explains up to 80 % of the observed variability in these regions. The regional differences in variability also influence the estimated sea level trends and their associated uncertainties, as with a length of 24 years the time span is still relatively short.

Besides the mean sea surface, the temporal evolution of sea level is of great interest. In view of global change, the long-term change in particular is highly relevant for predicting future sea levels and their impact for society and environment. Thus, as part of North SEAL, sea level trends are also provided. While the mean sea level trend between May 1995 and May 2019 in the North Sea amounts to 2.61±0.95 mmyr-1, the trend varies between 1.5 and 3.5 mmyr-1 over the region as illustrated in Fig. . The highest trends are observed in the German Bight and around Denmark, whereas significantly lower trends are observed around the southern part of Great Britain. Figure a and b include a black contour line. This line confines the coastal areas flagged as decreased quality in the SLA dataset in the framework of the gridding procedure (see Sect. ). This flag is defined quite conservatively and tuned to provide optimal SLA time series. For trend computations, a rejection of the flagged areas is not necessary when the trend uncertainties are taken into account in the course of the data analysis. When the flagged coastal regions are excluded, the overall trend for the North Sea changes only marginally to 2.60±0.95 mmyr-1.

Even with almost 2.5 decades of observation data, the trend uncertainties are still of the same order of magnitude as the trend itself. As visible from Fig. b, the trend uncertainties vary between 0.5 and 2.5 mmyr-1, with the smallest values in the northern part of the North Sea, especially close to the Norwegian coast. The highest uncertainties can be seen in the German Bight and at some smaller bays along the coasts of Great Britain and France. The particularly large uncertainties in the German Bight coincide with the aforementioned larger sea level variability described by and may also be affected by poorer ocean tide corrections in these areas.

The average trend of 2.61±0.95 mmyr-1 differs from the MSL trend of 4.00±1.53 mmyr-1 observed by over 1993–2009. The difference is, however, still within the limits of the confidence bounds. Deviations between the trends may, on the one hand, be due to the different periods of observation. On the other hand, the trend of is based on TG observations and thus may not unequivocally be compared to the sea level trend from gridded altimetry data that are spatially distributed over the North Sea. In addition, the region was not exactly identical, as it extended further into the English Channel where they found much smaller trends (1.32±1.11 mmyr-1) than in the inner North Sea (4.59±1.82 mmyr-1). Compared to the sea level trend of the North Sea over the last century of 1.53±0.16 mm , our study reveals a significantly increased sea level trend over the past 2.5 decades. Again, however, the value from is based on TGs and not entirely comparable. In general, our observed average trend is of the order of the global sea level trend of 3.1±0.1 mmyr-1 (from 1995 to 2018) reported by .

In addition to long-term changes in sea level, North SEAL allows for the analysis of seasonal variations. Figure  shows the estimated amplitudes of the annual cycle as derived from multi-year monthly means (see Sect. ). The highest amplitudes of more than 10 cm are visible in the German Bight and close to the Danish coasts. The signal is much smaller in the north of the region around Norway. Estimates of the annual amplitudes are much more accurate than the estimates of the trends. Uncertainties vary between 1 and 4 cm and amount to approximately one-third of the signal itself.

In order to assess how good North SEAL represents sea level variability on different timescales in comparison with TGs, correlations of the time series and their rms differences are analyzed. The median and mean correlation between North SEAL SLA and 52 TGs amounts to 0.86 and 0.85, respectively (Fig. a). The highest correlations (up to 0.93) are found for the TGs at the Shetland Islands and along the northern coast of Denmark. The lowest correlations appear in small bays and fjords, e.g., the Firth of Forth (0.69) and the Oslofjord (0.72). Low correlations are also observed along the coast of Belgium and the Netherlands at the mouth of the English Channel (Strait of Dover).

This distribution is also visible in the correlations between the CMEMS/SL_cci products and TGs (Fig. b and c). Very likely, some of those TGs in river mouths or smaller bays do not provide data that are representative for the sea level variations of the coastal or open ocean in their vicinity. Dividing all TG stations into different categories according to their locations (Fig. ) demonstrates that the majority of optimally located stations show correlations of around 0.8 or better, whereas stations located at fjords, rivers, or close to floodgates show generally lower correlations. For stations located in river mouths the spread of correlation values is largest. On the other hand, stations sheltered by islands show high correlations with a small spread. For TGs at rivers and near floodgates North SEAL clearly outperforms the other two products. Obviously, here, the quality flag successfully prevents the use of inappropriate data. However, in fjords, especially in the Oslofjord, North SEAL shows lower performance. This may be caused by a distance between the respective grid node and the TG that is too large.

Overall, with median and mean correlations of 0.86 and 0.85, respectively, North SEAL matches the TG measurements better than SL_cci (0.82/0.78) and CMEMS (0.79/0.78). Only for a minority of TGs are the correlations lower than for SL_cci (23.1 %) and CMEMS (11.5 %). The average difference in correlation is about 0.06 for both datasets (0.0651 for SL_cci; 0.0622 for CMEMS). This indicates an average improvement in correlation between 8.4 % (CMEMS) and 10.5 % (SL_cci).

The better performance close to the coast can be attributed in large part to the consideration of quality flags in the North SEAL SLA grids. They enable the selection of the closest grid node with reliable SLA information and improve the correlations with TG measurements by about 12 % on average. Moreover, the flag led to the exclusion of two TG stations for which no valid grid node could be found within a distance of 150 km (Oslo in Norway and Newhaven at the south coast of Great Britain). Figure a in the Appendix shows the correlations between altimetry and TGs if the quality flags are not considered, i.e., if the SLA series at the grid node closest to the TG is applied.

In order to evaluate and quantify the consistency in sea level variability, next to the correlations, the rms differences between altimetry and TG time series are computed and analyzed. This is done for all three altimetry datasets for monthly and for annual time series, both reduced by potential offsets and trends. In addition, a monthly deseasoned time series is included in the investigation. Table  shows the results from these comparisons. The median rms values are all between 3 and 8 cm. North SEAL shows values of 4.9 cm on a monthly scale (without annual signal) and 2.2 cm on an annual scale. With this, it clearly outperforms the other two products on a monthly scale. However, for the representation of lower-frequency processes, especially interannual variability, the CMEMS product slightly outperforms the other datasets.

As described in Sect. , TG trends are corrected for VLM in order to make them comparable with the absolute sea level trends from satellite altimetry. Figure  shows the absolute sea level trends from three altimetry datasets and from TG measurements (top row), their standard deviations (middle row), and the trend differences at the TG locations (bottom row). Since GPS information is not available for all TG locations (see Sect. ), the figure only contains 27 TGs.

The trends from the three altimetry datasets show clear regional differences. While all products show higher trends around Denmark, discrepancies are visible in the central North Sea. For large areas south of 56∘ N, the trends from SL_cci and (to a smaller extent) CMEMS are about 1 mmyr-1 higher than the North SEAL trends. Likewise, both products show higher trends close to the coasts of Denmark and Norway. These differences are up to about 1.5 mmyr-1. However, none of these differences are significant in view of the trend uncertainties.

The trend uncertainties show similar geographic patterns in all datasets. The lowest accuracies are found in the area of the German Bight. They improve towards the central North Sea and are highest in the region of the Shetland Islands and west of Norway. The main differences between the three datasets can be seen in small bays (e.g., along the British coast), where North SEAL is characterized by higher uncertainties than the other two products. Based on the existing data, it cannot be decided whether this behavior is caused by less correct trends or by more realistic uncertainties.

The comparison of altimetry and TG trends (the latter corrected for VLM) reveals large discrepancies, while the uncertainties for both data types agree quite well. Trend differences reach up to 3.8 mmyr-1 with a median of -0.13 mmyr-1 and an rms of 1.40 mmyr-1. Table  indicates that the rms values for North SEAL are slightly smaller than for the other two altimetry datasets (CMEMS: 1.42 mmyr-1, SL_cci: 1.49 mmyr-1). However, since the time period of about 20 years is still quite short for a reliable trend estimation and since uncertainties of three different data types are involved (altimetry, TGs, and VLM), the trend differences for almost all locations are not statistically significant.

Moreover, the values are dependent on the applied VLM correction of the relative TG trends. Table  provides the rms of trend differences and the median bias between the trends from altimetry and TGs. For consistency, these results only refer to the 27 TGs that are co-located with a GPS station. Interestingly, implementing the GIA VLM correction C18 results in a lower deviation of the trends than using local GPS information. For North SEAL, for example, the improvement is about 18 % in terms of rms. This VLM correction also outperforms the second GIA-based estimate from ICE-6G. The better performance of GIA-based VLM corrections compared to GPS corrections could be caused by the relatively large maximum allowed distance of 50 km between a TG and GPS station. Locally unequal VLM might introduce different signals at a GPS and TG location even over such distances. For example, two GPS stations located on the island of Sylt (RANT and HOE2, in the vicinity of TG Hörnum) indicate VLM estimates differing by more than 1 mmyr-1, even though they are within a distance of only about 10 km. This could make a smooth long-term GIA model make a better-suited correction. As can be seen in Fig. g–i, the largest scatter of absolute sea level trends is observed in the German Bight, more precisely at the offshore-located islands. Such areas could be more strongly affected by local VLM than, for instance, the TG locations along the British coastlines.

Adding the effect of CMR to GIA estimates influences the agreement of the trends in different ways. We observe the strongest improvement for SL_cci (for both rms and bias) and moderate improvement for the rms of CMEMS but with an increase in the bias. Likewise, the bias becomes larger for North SEAL when the effect of CMR is added, and in this case, the rms also increases (for either GIA-solution). The applied CMR correction generates a large-scale uplift signal of about 0.7 mmyr-1 over the domain (Fig. ), which leads to an increased absolute sea level trend at the TG. This effect projects into the negative biases for North SEAL and CMEMS. The highest consistency between altimetry and TGs is obtained for North SEAL when the TG measurements are corrected using C18 VLM. Nevertheless, further investigations are worthwhile to understand why the CMR correction introduces pronounced biases for North SEAL and CMEMS, while it improves the consistency for SL_cci. For such studies, however, longer time series would be desirable in order to reduce the trend uncertainties.

A comparison of the annual cycles among the altimetry products and TG data reveals much better consistency than in the case of sea level trends. Figure  shows the amplitudes (top row), their uncertainties (middle row), and the discrepancies between altimetry and TGs (bottom row). Overall, we find qualitatively similar patterns of the amplitudes and of their uncertainties.

The difference between altimetry and TG amplitude is significant for only very few stations (three stations for North SEAL and SL_cci, seven stations for CMEMS). The largest deviation is found for CMEMS at Dagebüll in the German Bight (-6.0 cm). This value is just slightly larger than the combined uncertainty of 5.8 cm (95 % confidence level). The German Bight is generally characterized by the largest annual amplitudes and largest uncertainties. Note that a negative deviation means an underestimation of the amplitude in the altimetry dataset. For the whole domain, North SEAL shows the lowest absolute mean deviations of the annual cycle of 1.3 cm (CMEMS: 1.8 cm; SL_cci: 1.5 cm) and the lowest rms difference of 1.6 cm compared to 2.3 and 1.9 cm (see Table ). This consolidates its coastal performance compared to the other altimetry products.