Sea level trends derived from North Atlantic and the Mediterranean Sea tide gauges have been re-evaluated with a common reference period (1960–2000) and with the atmospheric component of the observed sea level variability quantified and removed by means of regional barotropic ocean models forced by wind and atmospheric pressure. The atmospherically forced trends are important and have values of −0.2 ± 0.1 mm/yr in the North Atlantic (west coast), −0.2 ± 0.2 mm/yr in the NE Atlantic, 0.3 ± 0.4 mm/yr in North Sea and −0.7 ± 0.1 mm/yr in the Mediterranean. The residual sea level trends corrected for post-glacial rebound processes are 0.9 ± 0.4 mm/yr in the Mediterranean, 1.1 ± 0.6 mm/yr in the NW Atlantic, 1.3 ± 1.0 mm/yr in the NE Atlantic and 1.3 ± 0.8 mm/yr in the North Sea. Atmospheric forcing is partly responsible for the observed patterns of sea level rise and for part of the observed sea level acceleration during the 1990s. The residual trends have further been corrected for the influence of the steric effects. In the Mediterranean removing the steric component increases the trends by 40% and makes them consistent with the Atlantic trends. The remaining sea level rise rates are due to mass addition and their spatial pattern in the region can be related to Greenland ice-melting rates.
 The melting of polar ice sheets induces specific gradients along coastal regions with lower than the mean values of sea level closer to the melting location and higher in the far field [Conrad and Hager, 1997; Plag and Juttner, 2001; Mitrovica et al., 2001, hereinafter referred to as M01]. European sites exhibit lower sea level rise rates than the NE American coast [Woodworth et al., 1999] and this was suggested that is caused by the influence of melting of Greenland ice sheet (M01).
 In this paper we derive sea level trends in the North Atlantic and the Mediterranean Sea corrected for atmospheric pressure, wind and steric effects as well as vertical land movements due to post-glacial rebound. The remaining trends are attributed to mass addition and their spatial pattern to relative sea level produced by ice-melting rates of the Greenland ice-sheets.
2. Data and Methodology
 Mean monthly values of Revised Local Reference tide-gauge records from the Permanent Service for Mean Sea Level (PSMSL) database [Woodworth and Player, 2003] covering the period 1960–2000 for which barotropic modelling of the meteorological component is available are used. The stations cover the U.S. and Canadian East coasts (11 stations), the North Sea (16 stations), the European Atlantic coasts (10 stations) and the Mediterranean Sea (11 stations) (Figure 1). We estimate sea level trends for the observations and the meteorologically-corrected sea level. For 40 year long records inter-annual variability introduces errors of approximately 0.5 mm/yr [Tsimplis and Spencer, 1997].
 The meteorological effects are estimated by use of three 2D barotropic regional models. The NW Atlantic stations were corrected by a model with spatial resolution of 1/12° covering the area 38°N to 60°N and 72°W to 42°W [Bernier and Thompson, 2006] and forced by the AES40 data set [Swail and Cox, 2000] for the period 1958–1999. The NW European shelf stations were corrected by a model of spatial resolution of 1/3° latitude by 1/2° longitude (45°40′N to 62°20′N and from 15°W to 12°30′E) [Flather et al., 1998]. The forcing comes from the Norwegian Meteorological Institute analysis. The other stations have been corrected by a 1/4° × 1/6° model (30°N to 47°N and 12°W to 35°E) which includes the Mediterranean Sea and the NE Atlantic coast, forced by a dynamic downscaling of the NCEP/NCAR dataset (1958–2001) [García-Sotillo et al., 2005].
 Mean monthly values for each model were constructed at the grid point closest to each tide-gauge. These were in all cases highly correlated with the mean monthly values of the tide gauges (about 0.6 in the NW Atlantic and 0.7 in the other areas). Prior to the trend computation the mean value for each calendar month was removed in all the series. Linear trends have been obtained by performing a robust linear regression.
 Stations have been grouped in four geographical regions and trends for 1960–2000 were calculated. In the NW Atlantic region the stations are ordered with increasing latitude. In this region sea level trends are consistent, varying between 1.2 mm/yr and more than 3 mm/yr (Figure 2a). Error bars in Figure 2 for each record represent the standard errors (that is the standard deviation of each series divided by the squared root of their length). They are in all cases less than 0.2 mm/yr that is, smaller than the estimated 0.5 mm/yr for a 40 year long time series. The mean sea level rise value computed with sea level observations is 2.3 ± 0.6 mm/yr. In each averaged region the error bar corresponds hereinafter to the standard deviation of the individual records within the same region. The trends in the model time series of this region are fairly constant with a mean value of −0.2 ± 0.1 mm/yr and stronger negative trends towards the north (Figure 2b). As a result, the trends of the meteorologically-corrected time series (Figure 2a, red) are higher than the observed ones with a mean value of 2.4 ± 0.6 mm/yr ranging between 1.2 and 3.2 mm/yr.
 In the shallow North Sea the observed sea level rise values range between −2 and 5 mm/yr. The large scatter is due to post glacial rebound effects which causes large vertical land motions to a number of stations [Milne et al., 2001]. Because of the large scatter of the observed trends no attempt has been made to compute a mean value for the region. The meteorologically-induced trends vary between −0.3 and 1 mm/yr, being larger at the easternmost stations. The average meteorological trend is 0.3 ± 0.4 mm/yr.
 The European Atlantic stations are ordered with decreasing latitude. Observed trends range from around −0.5 mm/yr in Malin Head and Dublin in the Irish coast to 1.6 mm/yr in Vigo. Larger trends of about 2.5 mm/yr are also observed in Devonport and Le Havre. Devonport, has datum shifts during the 1970s and 1980s [Woodworth et al., 1999] and shows a higher interannual variability than the surroundings. Le Havre is located at the Seine river mouth, in an area frequently dredged thus it is unsuitable for trend computation (G. Woppelmann, personal communication, 2007). When these stations are omitted the mean trend in the area is 1.0 ± 0.9 mm/yr with values increasing towards the south. The meteorologically-induced sea level trends in this region continue the decreasing pattern observed in the southernmost stations of the North Sea. The mean value of the atmospherically corrected records is 1.2 ± 0.9 mm/yr.
 The Mediterranean Sea shows the smallest observed trends (−0.7 to 0.5 mm/yr) with a mean trend of 0.0 ± 0.4 mm/yr. The atmospherically-induced trends in the Mediterranean Sea are very homogeneous, varying between −0.5 and −0.8 mm/yr, mainly due to an increase of the atmospheric pressure over this region during the last decades of the 20th century [Tsimplis et al., 2005]. Thus the meteorologically-corrected series show significantly increased trends with mean value of 0.7 ± 0.4 mm/yr.
 The GIA correction (Figure 2c, Table 1) in the NW Atlantic decreases the average trend from 2.3 ± 0.6 to 1.0 ± 0.6 mm/yr for the observations and from 2.4 ± 0.6 to 1.1 ± 0.6 mm/yr for the meteorologically-corrected series. On the east coast of the British Isles the GIA corrected trends become more coherent, mainly due to the correction of the Aberdeen station. The tide gauges in Cuxhaven, located in mud at a river mouth, and Borkum located on a small island, show significantly different trends. This part of the North Sea is very shallow and the resolution of the model may not be able to adequately resolve the effects of bathymetry. Excluding these two stations the sea level trend ranges between 2.4 and −0.8 mm/yr for observations and between 2.5 and −1.1 mm/yr for meteorologically-corrected series. On the European Atlantic coast the mean value after the GIA correction is 1.0 ± 0.7 mm/yr and 1.3 ± 0.8 mm/yr for observations and meteorologically-corrected series, respectively. Finally, in the Mediterranean the GIA corrections are very small except in the Adriatic Sea, where values reach −0.3 mm/yr. The mean values are 0.2 ± 0.4 and 0.9 ± 0.4 mm/yr for observations and meteorologically-corrected series, that is, an average increase of 0.2 mm/yr due to GIA corrections.
Sea level trends in mm/yr for the four regions in Figures 2 and 3 and for the period 1960–2000. Errors correspond to the standard deviations of each area.
2.3 ± 0.6
−0.2 ± 0.1
2.4 ± 0.6
1.0 ± 0.6
1.1 ± 0.6
0.3 ± 0.4
1.6 ± 0.9
1.3 ± 1.0
1.0 ± 0.9
−0.2 ± 0.2
1.2 ± 1.0
1.0 ± 0.7
1.3 ± 0.8
0.0 ± 0.4
−0.7 ± 0.1
0.7 ± 0.4
0.2 ± 0.4
0.9 ± 0.4
 The same analysis performed for the period 1960–1990 gives observed sea level trends in the NW Atlantic region and in the Mediterranean 0.5 and 0.4 mm/yr smaller. This was an expected result since the last decade has been identified as a period of enhanced warming and sea level rise [Holgate and Woodworth, 2004]. We also note that the atmospheric sea level trends in this period are slightly smaller (larger in amplitude but negative) than for the longer period, except for the North Sea. Thus the observed sea level rise during the last decade is partly due to the atmospheric forcing. However, the spatial patterns are not significantly affected by these increases.
 Except in the NW Atlantic, where there is no difference between summer and winter, winter trends in the observations dominate the signal (Figure 3). In the North Sea, winter trends are up to 4 mm/yr larger than the annual trends. In the Irish coast the same pattern is present although differences are smaller. Along the southern British coast, the European coasts of the eastern Atlantic and in the Mediterranean, winter trends are again stronger (more negative) than the annual trends. Most of the seasonality in the trends is removed in the meteorologically-corrected series indicating that it is almost exclusively generated by the atmospheric effects. However some differences remain in the North Sea.
 The atmospheric correction for the last four decades has been shown to be important in obtaining realistic spatial patterns. After the corrections the trends in the North Atlantic and the Mediterranean Sea are regionally more coherent but remain significantly smaller than the recently stated global average values of 1.8 mm/yr (Table 1). Significant differences in the seasonal trends are found in the North Sea where the meteorological forcing introduces trends of up to 4 mm/yr. We now turn into the exploitation of the signal contained in spatial patterns.
 There is no east-west gradient detectable in the GIA-corrected values for the North Atlantic region examined (Table 1). In an attempt to estimate errors associated to GIA corrections, when the ICE-5G (VM4) values are used instead of the VM2 version, the NW Atlantic stations display an average increase of 0.5 mm/yr, while the European region do not change significantly. Even in this case, the east-west differences are less than 0.3 mm/yr. The Mediterranean sites, corrected for GIA, exhibit lower sea level rise than the Atlantic stations.
 When we remove from the meteorologically and GIA corrected Mediterranean stations the steric effect, estimated from the 1-year Medar climatology [Rixen et al., 2005], the mean trend is around 1.2 mm/yr, that is, 40% higher than the mean uncorrected for the steric effect value. Therefore differences between sea level trends at Mediterranean stations and at nearby European Atlantic sites are partly explained in terms of the atmospheric contribution (which increases the average trend in the Mediterranean by 0.7 mm/yr, see Table 1) and partly due to different steric effects. Thus, observed Mediterranean sea level rise rates cannot be directly compared with other European sites unless the differences in the atmospheric and steric forcings are considered.
 The steric contribution to sea level changes for the Atlantic sites has been estimated from both Ishii and Levitus data sets [Ishii et al., 2003; Levitus et al., 2000]. After correcting for atmospheric pressure, wind, steric and post-glacial rebound effects, the remaining sea level change rates must be due to local vertical land movements not attributed to GIA processes and to mass addition. According to previous studies [Conrad and Hager, 1997; Plag and Juttner, 2001; M01] ice melting from Greenland ice sheet would induce a spatial gradient of relative sea level rise in the northern hemisphere. Melting rates in Antarctica and continental glaciers do not have a spatially varying signature in the region we are examining, so no information can be extracted from these sites.
 We have estimated spatial gradients for the corrected sea level trends in the Atlantic tide gauges. The NW Atlantic sector displays a gradient not significantly different from zero, using both steric corrections. Higher uncertainty is expected in this side of the Atlantic due to the larger differences between the corrections for PGR using both VM2 and VM4 in the GIA model. Wake et al.  using hydrographic data, suggest differential steric corrections in the NW Atlantic sites varying up to 0.5 mm/yr. These facts, together with the small length in the latitudinal dimension, prevent us from making an estimate of melting rates using this area. In the NE Atlantic area we have found a spatial gradient of −0.07 ± 0.04 mm/yr/°lat (Figure 4). Differences in the gradient using Levitus and Ishii data bases are 0.005 mm/yr/°lat and have been included in the uncertainty range. Ignoring the outlier of Dublin the gradient becomes −0.05 ± 0.03 mm/yr/°lat (Figure 4, dashed line).
 The M01 model predicts a spatial gradient of −0.04 mm/yr/°lat in the same NE Atlantic stations. The higher values derived here (−0.07 ± 0.04 mm/yr/°lat) are thus translated into a higher rate of ice melting in Greenland of 1.0 ± 0.6 mm/year, which corresponds to a mass loss of 360 ± 216 Gt/yr. The most recent direct observations however provide lower values for recent years: Chen et al.  estimated a mass loss of 239 ± 23 Gt/yr in Greenland for the period 2002–2005, while Velicogna and Wahr  provided a value of 248 ± 36 Gt/yr for 2002–2006 both from GRACE measurements. Rignot and Kanagaratnam  stated that the ice discharge was accelerated during the last decade from 90 to 220 Gt/yr. The work by Shepherd and Wingham  reviews the last results, establishing an interval of 101 ± 16 and 219 ± 21 Gt/yr for Greenland mass loss for recent years. For the period 1960–2000 Greenland mass loss is estimated to be much less (E. Rignot, personal communication, 2007). It is worth mentioning that part of the discrepancies between these estimations and our result could be due to signals not linked to Greenland like the melting of mountain glaciers which has been accelerated during the last decade [Dyurgerov and Meier, 2000].
 Note that the model presented by M01 is based on a different deglaciation history (ICE-3G) than the one used here. Thus further work is needed in evaluating the consistency of the spatial gradients from models with the spatial patterns presented here.
 M. Marcos is under a post-doc fellowship funded by the Spanish Ministry of Education and Science. The work has been carried out in the framework of the VANIMEDAT project (CTM2005-05694-C03/MAR) funded by the Spanish Marine Science and Technology Program. We thank Natacha Barnier and Roger Flather, respectively for the provision of model data, Alix Lombard for providing information on the steric sea level trends, M. Tamisiea for the earth model and to J. Gregory, P. Woodworth, G. Milne, R. Gehrels and E. Rignot for their helpful comments.