Coastal sea level trends in Southern Europe

Transcription

1 Geophys. J. Int. (2008) 175, doi: /j X x Coastal sea level trends in Southern Europe Marta Marcos and Michael N. Tsimplis National Oceanography Centre, Southampton SO14 3ZH, UK. Accepted 2008 June 23. Received 2008 June 23; in original form 2007 October 25 GJI Marine geoscience SUMMARY Low frequency sea level variability in the Mediterranean Sea and in the Atlantic Iberian coast is investigated by use of tide gauge records. The five tide gauge records that span most of the 20th century show positive trends between 1.2 and 1.5 ± 0.1 mm yr 1 and negative accelerations between 0.3 ± 0.3 and 1.5 ± 0.4 mm yr 1 century 1. Sea level trends obtained from the 21 longest records (>35 yr) are smaller in the Mediterranean (0.3 ± 0.4 to 0.7 ± 0.3 mm yr 1 ) than in the neighbouring Atlantic sites (1.6 ± 0.5 to 1.9 ± 0.5 mm yr 1 )for the period Decadal sea level trends in the Mediterranean are not always consistent with global values, in particular for the 1990s, during which the Mediterranean has shown enhanced sea level rise of up to 5 mm yr 1 compared to the global average (mostly attributed to higher warming). The atmospheric and steric contributions to the observed sea level trends for are also examined. The atmospherically induced sea level is obtained from a barotropic model forced by wind and atmospheric pressure. The atmospheric contribution accounts for per cent of the observed yearly sea level variability and introduces negative trends of 0.2 to 0.9 mm yr 1. The steric sea level, obtained from T and S climatologies, has negative trends ranging from 2.1 ± 0.6 to 0.1 ± 0.3 mm yr 1. Other shorter tide gauge records (>7 yr) are used to quality check longer series and to explore their consistency with the long-term records and identify short but apparently consistent tide gauge records. Key words: Time series analysis; Sea level change; Europe. 1 INTRODUCTION Sea level is an important environmental parameter because of its impacts on the coastal zone as well as an indicator of climate change. The Fourth Assessment Report of the Intergovernmental Panel for Climate Change (IPCC AR4) (2007) recognises the variability of sea level at regional scales (Cazenave & Nerem 2004). Coastal sea level is measured by tide gauges referenced to a point on land. Thus the measurement includes land movements that can significantly contaminate the sea level signal (Emery & Aubrey 1991; Zerbini et al. 1996). Land movements can be due to different processes: long-term changes like glacial isostatic adjustment (GIA) due to the ongoing viscous response of the solid earth following the removal of the great ice loads following deglaciation, local subsidence either natural (e.g. sedimentary loading of delta) or anthropogenic (like coal mine collapses or extraction of gas or water) or fast changes caused by for example seismic activity (Emery & Aubrey 1991). Zerbini et al. (1996) used the tide gauge network in the Mediterranean referred to a global reference system to determine the rates of vertical crustal motion. They found that at most stations these rates are of the order of ±1mmyr 1, a value which is small compared to decadal rates of sea level change during most of Now at: IMEDEA, Mallorca, Spain. the time periods but of the same order of magnitude with the longterm trends. In the NW Mediterranean for example the rates of crustal movements decontaminated for PGR are 0.5 mm yr 1 (with positive values meaning land uplift). They also found that the crustal motion due to tectonics in this region is smaller than the postglacial rebound (PGR) and the response to surface loads. Fenoglio-Marc et al. (2004) used differences in sea surface heights measured by tide gauges and TOPEX/POSEIDON altimeter to estimate vertical land motion in the Mediterranean region. They obtained smaller rates in the Northwestern Mediterranean (0.5 ± 0.9 mm yr 1 )and the largest in the Italian Peninsula and the eastern basin ( 3.0 ± 1.6 mm yr 1 in Antalya, for example), with an average accuracy of 2.3 ± 0.8 mm yr 1. More recently Woppelmann et al. (2007) have used the longest GPS series available globally to determine vertical land movements directly from measurements. Changes in sea level are also caused by meteorological forcing, variations in the density structure, mass addition and changes in oceanic circulation. Instrumental errors, unrecorded changes or updates in the vertical referencing system and changes in the configuration of the proximity of the instrument, like dredging, also affect the quality of the tide-gauge records and should be taken into account when sea level trends are estimated. Each tide gauge record integrates all the above forcing factors, reference point movements and instrumental problems and thus tide gauge based estimates of sea level trends are, strictly speaking, local 70 C 2008 The Authors

2 Coastal sea level trends in Southern Europe 71 in character, although they have been routinely used to estimate far field long-term changes on the assumption that such larger scale changes are reflected in most of the available tide gauge data, that is, they are a significant part of the signal. Semi-enclosed basins, like the Mediterranean Sea, pose additional problems, because of their restrictions in their communication with the open ocean. In the Mediterranean Sea coastal sea level derived from the longest tide gauges indicates a rate of sea level rise for the 20th century of mm yr 1 (Tsimplis & Baker 2000). For the period an increase in the average atmospheric pressure over the basin caused negative sea level trends (Tsimplis & Baker 2000; Tsimplis & Josey 2001). During the same period, sea level was rising in the Atlantic stations although with a lower rate than before From the 1980s onwards sea level rise appears to have increased globally (Holgate & Woodworth 2004; Holgate 2007), while in the Mediterranean fast sea level rise was observed in the late 1990s (Cazenave et al. 2001; Fenoglio-Marc 2001). Despite the above general statements which are derived, as customarily done in sea level research, on the basis of the longest tide gauges available and in spite of the well known bias in their spatial distribution (Tsimplis & Spencer 1997), there are several other tide gauges in the Mediterranean Sea providing of information regarding local sea level variability. In this work, we estimate the sea level trends from the available tide-gauge records in the Mediterranean Sea and in the Atlantic Iberian coasts. Sea level trends are estimated on the basis of records longer than 35 yr. These estimated sea level trends include contributions by long-term changes in the meteorological forcing (Tsimplis & Josey 2001) as well as changes in the steric contribution (Tsimplis & Rixen 2002). For the period and for each tide gauge we separate the direct atmospheric forcing and steric contributions from the other forcing factors and present the residual trends. The main objective of the paper is to explore the low frequency behaviour (interannual and decadal variations and long-term trends) of all the tide gauge records available in the region spanning at least 7 yr in recent decades, using their most updated version. The ultimate aim of this work is to provide the potential user of tide gauge data in this region with an assessment of the quality of the time-series and with estimates of sea level trends in the different areas of study. This is done through a detailed analysis of the longest records (>35 yr) and their intercomparison with shorter ones. Additionally, the causes of this variability are also investigated for the period , when atmospheric and steric corrections are available. The improvement of this research in respect to previous works is the large number of tide gauge records involved and the application of the corrections in order to explain the physical mechanisms. 2 DATA SETS 2.1 Tide gauge data Monthly sea level values with benchmark datum history (Revised Local Reference, RLR) from the Permanent Service for Mean Sea Level (PSMSL) database (Woodworth & Player 2003) which cover the last decades of the 20th century were considered. Because of the lack of long-term reliable measurements in the Eastern Mediterranean (Fig. 1) the measurements in Alexandria (Egypt) were also included in the analysis. However, this station is not an RLR station and this must be kept in mind during the analysis. The PSMSL performs quality checks and provides comments on the quality of the records. A well known problem with the station of Marseille is the high values found during the beginning of the 1950s, which are not reproduced in any of the nearby records and that are claimed to be due to bad operation of the instrument ( Therefore this part ( ) of the record was removed. For Cascais (Portugal) 20 yr of quality controlled data ( ) have been kindly provided by the local authorities (Instituto Geográfico Português) and added to the PSMSL long record after they were quality checked. Among the 68 selected sea level records 21 span more than 35 yr (marked with circles on Fig. 1 and listed in Table 1). The remaining 47 records are longer than 7 yr (Table 2) and span the last decades of the 20th century. The shorter tide gauges used in this study together with their period of operation and their percentage of data gaps are listed in Table 2. The shorter records are useful in three capacities. First, they can be used for quality checks on the longer records. Second, if they are consistent with the long records they can be used for filling in gaps, where needed, of the longer record. Third, their consistency or inconsistency with the longer records behaviour can Figure 1. Location of the tide gauges. Black dots correspond to the longest time-series (>35 yr) which are labelled, while red squares are the shorter records Journal compilation C 2008 RAS

3 72 M. Marcos and M. N. Tsimplis Table 1. Tide gauges records in Southern Europe longer than 35 yr. Station Name Area Latitude (, Minutes) Longitude (, Minutes) Period Number years Per cent gaps Variance (cm 2 ) Trend (mm yr 1 ) S.Jean de Luz Atlantic N W ± 0.3 Santander N W ± 0.2 Coruña N W ± 0.2 Vigo N W ± 0.2 Cascais N W ± 0.1 Lagos N W ± 0.1 Cadiz Strait N W ± 0.3 Tarifa N W ± 0.2 Ceuta N W ± 0.1 Malaga N W ± 0.2 Alicante W. Med N W ± 0.2 Marseille N E ± 0.1 Genova N E ± 0.1 Venice Adriatic N E ± 0.1 Trieste N E ± 0.1 Rovinj N E ± 0.2 Bakar N E ± 0.1 Split I N E ± 0.2 Split II N E ± 0.2 Dubrovnik N E ± 0.2 Alexandria E. Med N E ± 0.2 Note: The percentages of data gaps, the variances and the trends are computed on the basis of monthly values. Errors in sea level trends correspond to standard errors. give early indications on whether they can be suitable for long-term monitoring of sea level. Shorter records can be contaminated by decadal signals arising from a variety of forcing factors (Tsimplis & Spencer 1997). Thus an effort was made to identify shorter records which can be considered as reliable. This is done by comparing the short records with a nearby long-term reference record using a two-step process. The first step consists of calculating the correlation between the shorter records and the chosen reference station for each area, over the period where both tide gauges were operating and after the trends have been removed (Table 2). The common period between each short series and the corresponding reference station is used to compute the covariance matrix from which the correlation coefficient is obtained. Statistically significant correlation indicates that these records confirm the signals observed on the reference stations and, in addition, can be used for filling in gaps in the reference stations. Thus they provide some redundancy in the monitoring network. The second step, which is only performed where the correlation is statistically significant, involves the comparison of the trends of the differences between the short station and the reference station over the common period. Where the trend of their difference is not statistically significant then we can conclude that the shorter stations can be expected to provide, in the future, additional good long-term monitoring points for the Mediterranean Sea. In addition the confidence to the longer stations improves. The comparison between reference stations and shorter records is a very reliable quality check. Thus the discussion of the suitability of short records is included in the Appendix although the identification of the consistent short records and the rejection of the others is also a result of this work. The inconsistency amongst the shorter records, and by comparison with the longer records, prohibits their further use. Thus the shorter records will not be used in the following discussion, except where needed to confirm or question parts of the longer records. Our assessment of which of the short-terms tide gauges can be considered reliable can be found in Table Methodology Sea level trends are estimated by a linear robust fit and their uncertainties are defined as the standard errors. This alternative to a simple linear regression minimizes a weighted sum of squares, but the weight given to each data point depends on how far the point is from the fitted line by using a bisquare weighting function, resulting in a smaller sensitivity to outliers. The variance-covariance matrix of the coefficient estimates is computed as V = inv(x X) σ 2 where X is the vector of times and σ is the root mean square error between the prediction and the data. The standard errors are derived from the matrix V. Prior to the trend computation the monthly time-series have been deseasonalized. To do this, the mean monthly values for each calendar month have been computed to obtain the mean seasonal cycle, and this cycle has been subtracted from the monthly time-series. Only complete years were used to compute the mean seasonal cycle in order to avoid biases. Acceleration is estimated by a second order least-square fit regression on time as the coefficient of the quadratic term. Sea level accelerations during the 20th century have been computed for the five longest tide gauges. Again uncertainties correspond to the standard errors. For the longest tide gauge records yearly time-series have been produced for the period (Table 2). Lagos and Alexandria have been discarded for this period because of lack of data during the 1990s decade, that is, they cover less than 75 per cent of the period A yearly value was assigned only if at least 11 months of the year were available; otherwise the value is taken as a gap. In order to compute variances and correlations the time-series have been previously detrended. Correlations presented are always significant to the 99 per cent confidence level. 2.3 Atmospherically induced sea level variations The meteorological contribution to sea level has been quantified using the output of a barotropic oceanographic model. In the

4 Coastal sea level trends in Southern Europe 73 Table 2. From left- to right-hand columns: tide gauge stations with shorter records, the area where they are located, their position (in latitude and longitude), period of operation, percentage of data gaps, reference station for computation of correlations, common period with the reference station, correlations and trends of the difference between the short series and the reference station. Station name Area Latitude Longitude Period Per cent gaps Reference st. Common period Correlation Relative trend Quality (, Minutes) (, Minutes) Bilbao Atlantic N W Cascais ± 1.2 G Santander N W ± 1.1 G Gijon N W ± 1.9 Q Coruña N W ± 1.3 Q Villagarcia N 8 46 W ± 2.9 Q Vigo N W ± 2.7 G Setroia N W ± 0.5 G Huelva Strait 37 8 N 6 50 W Ceuta ± 3.7 G Bonanza N W ± 1.6 G Algeciras N W ± 1.2 G Ceuta N W ± 1.2 G Malaga N W ± 1.3 G Almería W. Med N W Marseille ± 0.6 G Cartagena N W ± 1.4 G Valencia N W ± 1.6 Q Barcelona N E ± 1.5 G L Estartit N E ± 1.7 G Sete N E ± 2.7 G Nice N E ± 0.4 G Valleta N E ± 0.9 G Koper Adriatic N E Trieste ± 0.4 G Luka Koper N E ± 1.8 Q Zadar N E ± 1.8 G Sucuraj N E ± 0.9 U Bar N E ± 0.5 U Preveza E. Med N E Antalya ± 1.1 Q Levkas N E ± 0.8 Q Posidhonia N E ± 2.2 Q Patrai N E ± 0.8 Q Katakolon N E ± 0.8 Q Kalamai N E ± 1.0 G North Salaminos N E ± 0.9 Q Piraievs N E NS Q Khalkis South N E ± 1.0 Q Khalkis North N E ± 0.8 Q Skopelo N E ± 0.9 Q Thessaloniki N E ± 0.9 G Kavalla N E ± 2.0 Q Alexandropoulis N E ± 0.9 Q Khios N E ± 0.9 G Siros N E ± 4.9 U Leros N E ± 0.7 Q Soudhas N E ± 0.8 Q Rodhos N E ± 1.1 Q Izmir N E ± 3.0 Q Hadera N E ± 1.4 G Antalya N E Notes: Uncertainties correspond to standard errors. Trends significantly different from zero are highlighted in boldface. NS means not significant. Last column is a quality flag of the record: G, good; Q, questionable; U, unknown. framework of the Hindcast of Dynamic Processes of the Ocean and Coastal Areas of Europe (HIPOCAS) project (Guedes Soares et al. 2002), atmospheric pressure and wind fields were produced by a dynamical downscaling of the reanalysis of NCEP/NCAR for the period (García-Sotillo et al. 2005). These fields were used to force a barotropic version of the Hamburg Shelf Circulation Model (HAMSOM) model covering the Mediterranean Sea and the Eastern Atlantic coast, with a spatial resolution of 1/4 1/6 in latitude and longitude, respectively. The comparison between the HIPOCAS sea level hourly output and the tidal residuals Journal compilation C 2008 RAS at coastal sites is very good with correlations between 0.8 and 0.9 (Marcos et al. 2005). Atmospherically induced sea level trends derived from HIPOCAS data set have already been computed in previous works (Tsimplis et al. 2005; Gomis et al. 2008). The comparison of HIPOCAS with another 2-D barotropic model in the region did not reveal any artificial drift of the model (Pascual et al. 2008). HIPOCAS monthly values have been used to correct sea level observations for the atmospheric contribution. For each tide gauge station, data from the closest point of the HIPOCAS grid has been

5 74 M. Marcos and M. N. Tsimplis subtracted from the observations for the period covered by HIPOCAS data ( ). the shorter records would make them consistent with the long records. 2.4 Steric sea level Steric sea level variability has been estimated from temperature and salinity climatologies by integration from the surface to 300 m of the specific volume anomaly caused by density changes. The steric component of each tide gauge has been computed at the closest gridpoint. In the Atlantic stations the Ishii climatology was used (Ishii et al. 2003). This consists of 1 1 gridded temperature anomalies given as yearly means for the upper 700 m from 1945 to 2005 (Ishii et al. 2003). This database was preferred instead of the alternative of Levitus et al. (2000) simply because it spans a longer time period. Differences in the steric sea level from the various databases in the Mediterranean are in general small at basin scales both at annual and interannual scales (Fenoglio-Marc, personal communication, 2007). In the Mediterranean the MEDAR database was preferred (Rixen et al. 2005), consisting of 1-yr temperature and salinity fields from 1945 to 2002 with a spatial resolution of 0.2. It is still a subject of discussion how the steric signals propagate between the open ocean and coastal sites. Previous works assume that tide gauge records are representative of the steric changes in a basin wide area (Miller & Douglas 2004) while others have used close gridpoints to account for local steric sea level variations at the coast (Plag 2006; Marcos & Tsimplis 2007). Here we have corrected for the steric effects only those tide gauges whose atmospherically corrected record is significantly correlated with the steric signal corresponding to the closest gridpoint deeper than 300 m. We have not used the simply closest gridpoint to avoid very local effects (e.g. river runoff) affecting our estimation of steric signal. It was hoped at the beginning of the study that the removal of the direct atmospheric forcing and the steric contribution from 2.5 Postglacial rebound The PGR effects have been corrected by means of the Glacial Isostatic Adjustment model ICE-5G (VM2) ( ac.uk/psmsl/peltier/index.html, Peltier 2004). The model does not provide uncertainties for vertical land movements associated to this effect, so the error bars for this correction have been neglected. In the Mediterranean the effect of the isostatic compensation due to PGR is of the order of ±0.3 mm yr 1 (Zerbini et al. 1996). 3 RESULTS In the first part of this section we discuss sea level trends as directly derived from the observations and for the longest period for each tide gauge. Then we discuss trends for the common period and the contribution of the atmospheric and steric forcing to sea level trends. The next section discusses accelerations of sea level for the longest records. Finally, the last part of this section discusses the change in the decadal trends and helps us link the long-term trends with those dominating the last four decades of the last century. 3.1 Sea level trends The observed sea level trends and the variance of the 21 longest records are shown in Table 1, together with their location, period of operation and percentage of data gaps. Yearly time-series for these records are plotted in Figs 2(a) (c), where nearby stations have been grouped into three different areas: Atlantic, Western Mediterranean Figure 2. Yearly observations and linear sea level trends for the longest sea level records in Southern Europe. (a) Atlantic area and Gibraltar, (b) remaining stations in Gibraltar and Western Mediterranean and (c) Adriatic and Eastern Mediterranean.

6 Coastal sea level trends in Southern Europe 75 Table 3. From left- to right-hand columns: Name of the stations, values of variance of yearly observations, percentage of variance explained by the atmospheric contribution, observed trends, atmospherically induced trends, trends of atmospherically corrected series (observation minus atmospheric contribution) and steric sea level trends for the period ; residual (observations minus atmospheric and steric contributions) trends. Station name Variance (cm 2 ) Per cent variance atm. Observed trend Atm. trend Atm. residual Steric trend Residual Residual + PGR S.J. de Luz ± ± ± ± ± ± 0.6 Santander ± ± ± ± ± ± 0.6 Coruña ± ± ± ± ± ± 0.5 Vigo ± ± ± ± ± ± 0.5 Cascais ± ± ± ± ± ± 0.7 Cadiz ± ± ± ± ± ± 0.8 Tarifa ± ± ± ± ± ± 0.7 Ceuta ± ± ± ± ± ± 0.8 Malaga ± ± ± ± ± ± 0.8 Alicante ± ± ± ± ± ± 0.4 Marseille ± ± ± ± ± ± 0.4 Genova ± ± ± ± ± ± 0.5 Venice ± ± ± ± ± ± 0.4 Trieste ± ± ± ± ± ± 0.3 Rovinj ± ± ± ± ± ± 0.3 Bakar ± ± ± ± ± ± 0.4 Split I ± ± ± ± ± ± 0.4 Split II ± ± ± ± ± ± 0.4 Dubrovnik ± ± ± ± ± ± 0.4 Notes: Those stations where the steric and the atmospherically corrected series are significantly correlated at 99 per cent confidence level are highlighted in boldface; residual-pgr trends are the previous value corrected for PGR. Steric trends are calculated using MEDAR database for the Mediterranean stations and Ishii for the Atlantic ones. Trends are in mm yr 1 and uncertainties correspond to standard errors. and Adriatic Eastern Mediterranean. The corresponding linear trends are also plotted for each time-series (Fig. 2) The Atlantic Iberian coasts On the Atlantic coasts the two longest stations, Cascais and Lagos (Fig. 1), cover almost one century of data and display sea level trends of 1.3 ± 0.1 and 1.5 ± 0.1 mm yr 1. The rest of the records, which have up to 60 yr of data covering the more recent decades, have linear trends ranging from 2.0 ± 0.2 mm yr 1 in Santander to 2.5 ± 0.2 mm yr 1 in Vigo, with the exception of Coruña which shows a trend of 1.4 ± 0.2 mm yr 1. Marcos et al. (2005) detected a reference jump in the Coruña tide gauge in 1963 which artificially decreased the observed trend by up to 0.9 mm yr 1 for the period By correcting the value at this station the trend becomes 2.4 mm yr 1 for the period , which is consistent with nearby tide gauges. For the period , the observed, atmospheric and steric sea level trends are compared in Table 3. Note that Lagos and Alexandria are not listed because they have significant gaps over this period. Yearly variances and the percentage explained by the atmospheric contribution are also listed. For (Table 3) sea level trends in the Atlantic vary between 1.6 ± 0.5 and 1.9 ± 0.5 mm yr 1 in the Bay of Biscay and the Northern Spanish coast, and 0.0 ± 0.4 mm yr 1 in Cascais, mainly caused by a continuous sea level drop since 1980 at this station. The variance of the observed yearly series range between 7 and 19 cm 2,being larger at the Northern Spanish coast. The meteorological contribution to the sea level variability explains between 20 and 30 per cent of the yearly variance, with atmospherically induced trends varying between 0.2 ± 0.2 mm yr 1 in Vigo and 0.7 ± 0.2 mm yr 1 in Sant Jean de Luz. The atmospherically corrected series, that is, the observations minus the meteorological contribution, have very consistent trends in the Northern Spanish coast ranging between 2.0 ± 0.5 and 2.2 ± 0.4 mm yr 1, again with the exception of Cascais that presents a lower value of 0.6 ± 0.3 mm yr 1. The steric contribution to sea level trends for the period is also negative in the Atlantic sector (Table 3). The steric trends decrease (increase in magnitude) southwards, varying from 0.1 ± 0.3 mm yr 1 in Sant Jean de Luz to 0.7 ± 0.5 mm yr 1 in Cascais. Only in Sant Jean de Luz the steric signal is significantly correlated with the atmospherically corrected time-series with a correlation larger than 0.3. The residual trend at this station is 2.6 ± 0.6 mm yr 1. At the other stations the residual trends after the steric sea level corrections are around 2.6 ± 0.6 mm yr 1 with a smaller value for Cascais of 1.2 ± 0.7 mm yr 1 for 1960 onwards. Thus for this area the atmospherically corrected trends are 2.1 ± 0.5 mm yr 1 with the steric correction slightly increasing this value to around 2.6 ± 0.6 mm yr The Strait of Gibraltar Close to the Strait of Gibraltar the tide gauges at Cádiz, Tarifa, Málaga and Ceuta, span between 40 and 60 yr (Table 1). Each tide gauge shows significantly different trend (Fig. 2) respectively 4.0 ± 0.3 mm yr 1 in Cádiz, 0.4 ± 0.2 mm yr 1 in Tarifa, 2.4 ± 0.2 mm yr 1 in Málaga and 0.6 ± 0.1 mm yr 1 in Ceuta. The lack of coherence in the trends cannot be explained by the different periods of operation as Fig. 2 indicates. In Cádiz, the trend is continuously present during the whole period, while in Málaga the trend appears after Data gaps at both stations are only 20 per cent thus the discrepancy cannot be attributed to missing observations either. Problems with the Cádiz tide gauge were reported in Ross et al. (2000), who corrected for what they considered a spurious downward trend of 14 mm yr 1 in The discrepancies do not disappear or reduce when the common period is considered (Table 3), with trends ranging from 1.3 ± 0.5 mm yr 1 in Tarifa to 4.5 ± 0.7 mm yr 1 in Cádiz. The atmospherically induced trends for this period in the area are 0.6 ± 0.1 mm yr 1 for all the tide gauges and the steric trends are between 2.1 ± 0.6 and 1.3 ± 0.6 mm yr. When the atmospheric correction Journal compilation C 2008 RAS

7 76 M. Marcos and M. N. Tsimplis is applied the sea level trends show the same discrepancies ranging between 0.8 ± 0.5 and 5.0 ± 0.6 mm yr 1. For Tarifa tide gauge the steric signal is correlated with the atmospherically corrected sea level (with a correlation larger than 0.3 again). When the steric correction is applied together with PGR the residual trend is 1.0 ± 0.7 mm yr 1. In summary, no regional trend can be defined for this region due to the large scatter of the results from the observations. In the quality control section we identified the contrast between Ceuta and the short tide-gauges located at the European side of the Strait of Gibraltar. We note that the same differences occur in respect of the Malaga and Cadiz and only Tarifa appears to have a trend smaller than Ceuta. The anomalously large trend in Cádiz has been checked against the two closest stations Huelva and Bonanza, also located in the Atlantic side of the Strait and thus expected to be subjected to the same forcing. The linear trend of the differences between each of these series and the Cadiz sea level, are 20 and 12 mm yr 1 for their common periods and , respectively. This indicates that the observed trend in Cádiz is due to either a large local subsidence or a malfunctioning of the instrument, but in any case cannot be considered as representative of this area. The tide gauge in Málaga has been compared with the shorter series in Málaga2, Algeciras and Almería for their common periods. The correlation of Málaga with Malaga2 and Algeciras is 0.9 and lower (0.6) with Almeria. The trends of the difference of these stations with Malaga are 6.5 ± 1.6 mm yr 1 for Málaga2, 7.7 ± 1.2 mm yr 1 for Algeciras and 6.7 ± 0.6 mm yr 1 for Almería, indicating that the long tide gauge trend in Málaga is spurious. Thus, one must utilise the two remaining long tide gauges of Tarifa and Ceuta which are not, as noted, consistent in their trends. Thus we cannot provide a sea level trend estimate for this area. This is problematic because the Strait of Gibraltar is a crucial area for sea level observation for the whole of the Mediterranean Sea. On one hand the Atlantic inflow has been found to be significantly correlated with the across Strait component (Tsimplis & Bryden 2000) and, on the other hand the along and across Strait slopes are considered to be relevant in determining the maximal or sub-maximal character of the exchange. Knowledge of the slopes can be an efficient way of monitoring the Mediterranean Sea. However, accurate sea level measurements are essential for the correct estimation of these slopes The Western Mediterranean The longest series in the Western Mediterranean are Alicante, Marseille and Genova, the two last ones starting by the end of the 19th century (Table 1). The variance in Marseille and Genova is very similar (between 50 and 55 cm 2 ) and they show the same trend for their entire period of 1.2 ± 0.1 mm yr 1. Alicante shows less variability with a variance of 24 cm 2 and a sea level trend of 0.3 ± 0.2 mm yr 1. For the period the observed trends become coherent among the three series with values between 0.0 ± 0.4 and 0.7 ± 0.3 mm yr 1 (Table 3). The atmospheric contribution with trends of 0.7 ± 0.2 and 0.9 ± 0.2 mm yr 1 accounts for per cent of the total yearly variance of the series. In general, the variability explained by the atmospheric forcing in the Mediterranean is larger than in the Atlantic (Table 3). By applying the atmospheric correction the sea level trends become positive and more coherent in the area with values between 0.0 ± 0.2 mm yr 1 in Alicante and 0.6 ± 0.2 and 0.7 ± 0.2 mm yr 1 in Genova and Marseille respectively. Thus we consider the atmospherically corrected trends of this area to be slightly positive with zero values in the western part of the basin. Steric trends range between 1.3 ± 0.3 mm yr 1 in Genova, 0.1 ± 0.4 mm yr 1 in Marseille and 0.7 ± 0.3 mm yr 1 in Alicante. Only the steric signal corresponding to Marseille tide gauge is correlated with the meteorological residual time-series. The timeseries in Marseille corrected for the atmospheric effects and for both the atmospheric and steric effects are plotted in Fig. 3 (upper plot) for the period The sea level trend corrected for the atmospheric, steric and PGR effects in this station is 1.1 ± 0.4 mm yr 1. The discrepancy between the steric signal in Marseille and Genova is problematic. These stations are highly correlated and the atmospherically corrected trends are consistent. The addition of the steric signal would make them inconsistent by about 0.8 mm yr 1, with Genova showing higher trends of about 2.0 ± 0.5 mm yr 1. This value appears consistent with the values at the Atlantic coasts. However, it contradicts the values in Marseille and Alicante which are significantly lower. In addition, only Marseille appears to be correlated with the steric sea level signal. Thus we consider that the lower trends values in Marseille are representative of the area in respect of the steric correction The Adriatic Sea The Adriatic Sea is a region where seven stations have been operating continuously during the last 50 yr, including two, Venice and Trieste, covering more than one century (Table 1). All the available long-term tide gauge records are of good quality with low percentages of data gaps, between 1 and 6 per cent, except for Bakar which, although of good quality, has about 15 per cent. Interannual oscillations appear to be very coherent among all the time-series (Fig. 2c) and the same holds for observed trends, which vary between 0.5 ± 0.2 and 1.2 ± 0.1 mm yr 1. The only exception is Venice with a trend of 2.5 ± 0.1 mm yr 1 during the 20th century due to local subsidence caused by water extraction which was reduced after As a result this tide gauge is unsuitable for regional or global trend studies (Woodworth 2003). Monthly variability is also very similar amongst stations (40 70 cm 2 ) with the exception again of Venice, which presents larger variance (Table 1). For the period sea level trends are consistent between the stations with values between 0.4 ± 0.4 and 0.3 ± 0.4 mm yr 1 (Table 3). Both the atmospheric and steric contributions to sea level trends are negative in the area. The corrections applied to Trieste time-series are plotted in Fig. 3 (lower plot). The atmospheric trends are 0.8 ± 0.2 mm yr 1, while the steric trends vary between 1.6 ± 0.3 and 1.7 ± 0.3 mm yr 1 (Table 3). The atmospheric contribution accounts for almost 50 per cent of the total yearly sea level variability in the Adriatic, except in Venice where only 30 per cent is explained by meteorological effects. However, this could be partly explained by the influence of local wind conditions on sea level which are not reproduced in most atmospheric reanalyses accurately (Wakelin et al. 2000) as well as the location of Venice tide gauge inside a lagoon. The atmospherically corrected series have trends ranging between 0.4 ± 0.3 and 0.9 ± 0.3 mm yr 1, with the exception of Venice. In three of the tide gauges located in the Northern Adriatic, Trieste, Rovinj and Bakar, the atmospherically corrected series are correlated with the steric signal. It is worth mentioning that due to the shallowness of this basin, the closest point to these stations deeper than 300 m is located in the southern part of the Adriatic Sea. However, if the closest gridpoint to each tide gauge is used for

8 Coastal sea level trends in Southern Europe 77 Figure 3. Tide gauge observations (black lines), atmospherically corrected records (dark grey lines) and atmospherically corrected records minus the steric correction (light grey lines) in Marseille (upper plot) and Trieste (lower plot). the steric correction the trends remain negative, although of smaller magnitude in the northern area. The residual trends, further corrected by PGR, are 2.6 ± 0.3, 2.3 ± 0.3 and 2.6 ± 0.4 mm yr 1 in Trieste, Rovinj and Bakar, respectively. These are considered as representative of the area. However, a significant part of these trends of about 1.6 ± 0.3 mm yr 1 is due to steric variations. The residual atmospheric trends are consistently around 0.8 ± 0.2 mm yr 1.Notably, in this area the direct atmospheric forcing contributes almost 1 mm yr 1 of sea level reduction between 1960 and The Eastern Mediterranean There is only one long tide gauge station in the Eastern Mediterranean located in Alexandria. As mentioned above, this tide gauge is not RLR and the data stops at 1989, but has been considered because of the lack of long-term good quality records in this region (Table 1). During the period of operation Alexandria has few data gaps (4 per cent) and its monthly variability (40 cm 2 ) is consistent with other observations in the Adriatic. The observed sea level trend of 1.9 ± 0.2 mm yr 1 is higher than those in the rest of the basin. The value would be expected to be even higher if the 1990s were covered, since the enhanced sea level rising was higher in the Eastern Mediterranean than in the rest of the basin. For the period the fraction of variance explained by the atmospheric contribution is only 15 per cent, much smaller then than in the rest of the basin and comparable to the Atlantic stations. The conclusion is that despite the large number of tide gauges in operation in the Eastern Mediterranean there are, yet, no reliable stations that can be used for the estimation of sea level trends. 3.2 Sea level accelerations For the five longest tide gauges, Cascais, Marseille, Genova, Venice and Trieste, the sea level accelerations during the 20th century are found to be negative: 0.3 ± 0.2, 0.9 ± 0.2, 0.3 ± 0.3, 1.5 ± 0.4 and 0.8 ± 0.3 mm yr 1 century 1, respectively (Fig. 4). However, we noted in the Section 3.1 that Cascais presents a downward trend from the 1980s, the period of data added to the PSMSL original data. If the acceleration in Cascais is computed until 1980, then a value 0.50 ± 0.36 mm yr 1 century 1 is found. The established view is that sea level has risen globally during the 1990s at rates higher than before (IPCC 2007). As we will see below this is also true for the Mediterranean Sea and the Atlantic stations. Thus, we consider that the Cascais values after 1980 indicate either a problem with the tide-gauge or a real local problem of land uplift. Negative values in the Mediterranean are consistent with the wellknown behaviour of the sea level rise pattern in this area (Tsimplis & Baker 2000) during the last century: from the beginning of the 20th century and up to 1960 sea level was rising between 1.2 and 1.5 mm yr 1 (Tsimplis & Baker 2000). At this time it started decreasing mainly due to an increase in atmospheric pressure, up to 1990, at rates of 1.3 mm yr 1 (Tsimplis & Baker 2000). This period is highlighted in Fig. 4 at the Marseille time-series. During the 1990s a fast sea level rise has been observed in the Mediterranean, as can also be seen in these five tide gauges, which was stopped after 1999 (Fenoglio-Marc 2001). It is worth noting the significantly different values of accelerations in Genova and Marseille in spite of their proximity, and the similarity between Marseille and Trieste. The largest deceleration value corresponds to Venice which is subjected to local subsidence and therefore not representative of the region. Journal compilation C 2008 RAS

9 78 M. Marcos and M. N. Tsimplis Figure 4. Monthly sea level values for the five longest tide gauges and the fitted quadratic curves. The dashed square in Marseille indicates the period of sea level drop identified from 1960 to We do not have sufficient atmospheric data to run hindcast models for the whole of the last century. A good proxy to the atmospherically driven part of the mean sea level variability is the North Atlantic Oscillation index (Tsimplis & Josey 2001; Gomis et al. 2006). Thus, we check whether the negative accelerations are due to changes in the meteorological forcing caused by the NAO. When the effect of NAO is removed from these records by subtracting the linear regression between winter NAO and the time-series, the negative accelerations decrease (in absolute values) in the Mediterranean and become 0.4 ± 0.3, 0.1 ± 0.3, 1.1 ± 0.6 and 0.1 ± 0.5 mm yr 1 century 1 in Marseille, Genova, Venice and Trieste, respectively, while in Cascais it remains unaltered. As a conclusion we note that for the period for which data are available we do not detect acceleration of sea level in the Mediterranean Sea or at the Atlantic coasts when the NAO contribution is removed. 3.3 Decadal variability of the time-series We compute the decadal rates of sea level change for the 21 long tide gauges as linear trends for 10-yr segments of the records overlapping year to year and compare the results with the global rates (Fig. 5). There are several studies estimating global sea level values (Holgate & Woodworth 2004; Church & White 2006; Jevrejeva et al. 2006; Holgate 2007). Here we compare the global decadal rates derived from the nine longest tide gauges in the world corrected for the inverted barometer effect (Holgate 2007) with those in Southern Europe (Figs 5a and b). Holgate (2007) used the HadSLP2 air pressure data set (Allan & Ansel 2006) for correcting the tide gauge data for the inverted barometer effect. Although this assumption is not totally valid for semi-enclosed seas such as the Mediterranean, here we have used the same dataset to correct our tide gauge records for consistency. Both in the Atlantic sites and in the Mediterranean interannual signals with periods of yr are evident. The average decadal rate in the Atlantic Iberian stations (Fig. 5a) does not always match the global values and it is in general larger, except for the period from mid-1970s to mid-1980s. During the 1920s and 1950s the regional rate presents the largest differences with the global values, although the former is computed with only two stations. From 1970s onwards the agreement is better. The Mediterranean mean decadal rate (Fig. 5b) shows good agreement with the global values except during the 1940s and beginning of 1970s, when much smaller values are found in the region than globally. Also during the 1990s the largest decadal rate of sea level change is observed,

10 Coastal sea level trends in Southern Europe 79 a Trends (mm/yr) b Year Trends (mm/yr) Year Figure 5. Decadal rates of observed sea level change, overlapped year to year, for the longest tide gauges separated in the Atlantic sites (a) and Mediterranean stations (b) (grey shadowed area). Red solid line represents the mean values and the global average derived by Holgate (2007) is also plotted as the blue line. much larger (by more than 5 mm yr 1 ) than the global values and only comparable to the observations of the beginning of the 20th century (although again these are computed with fewer stations). The differences found in the last decade clearly indicate that global estimates are not always applicable to the Mediterranean and that these can be much larger than global values of sea level rise. The observed decadal rates of sea level change have been compared with the decadal rates of NAO (not shown). The correlation found with the average decadal rate in the Atlantic stations was 0.46 while in the Mediterranean is 0.60 (both significant at the 99 per cent level), indicating that the decadal fluctuations are partly atmospherically driven. Journal compilation C 2008 RAS

11 80 M. Marcos and M. N. Tsimplis 4 SUMMARY AND CONCLUSIONS Long-term sea level variability and trends have been analysed for 21 coastal records longer than 35 yr in Southern Europe, covering the Iberian Atlantic region and the Mediterranean Sea. We have presented updated sea level trends for the area both in respect of observed sea levels and after correcting them for steric and atmospheric effects. The six longest records spanning more than 90 yr have sea level trends between 1.2 and 1.5 ± 0.1 mm yr 1 during the 20th century. The only exception corresponds to the tide gauge in Venice, which is affected by local subsidence (Woodworth 2003). These values are of the same order of magnitude as those found by Zerbini et al. (1996) for crustal movement rates. The secular sea level change in this region is therefore smaller than the global average of 1.8 ± 0.5 mm yr 1 (IPCC 2007). Sea level accelerations in the 20th century computed with the longest records are negative, as a consequence of the particular behaviour of sea level in the basin, that is, a decrease from 1960 to 1980s. This contrasts with the suggestions about global sea level accelerations found to be positive (IPCC 2007) albeit for longer periods. When the NAO contribution is removed from the sea level records the accelerations become indistinguishable from zero in the Western Mediterranean and very close to zero, but negative in the Atlantic coast. For the period atmospheric (due to wind and pressure) correction is available in this region. The atmospherically induced sea level derived from the HIPOCAS data set is very well correlated ( ) with observations and therefore can be used reliably to correct sea level for the meteorological effects. Observed trends in the Atlantic sites for the period are in the range of 1.6 ± 0.5 and 1.9 ± 0.5 mm yr 1, except for Cascais with 0.0 ± 0.4 mm yr 1, and in the Mediterranean they are between 0.3 ± 0.4 and 0.7 ± 0.3 mm yr 1. That is, sea level trends are smaller in the Mediterranean than in the nearby Atlantic area. This behaviour has been partly attributed to the effect of the atmospheric pressure, which increased in average for this period in Southern Europe (Tsimplis & Baker 2000; Tsimplis et al. 2005). Its contribution to the sea level trends is negative in the entire region, with larger values (in absolute terms) in the Mediterranean side than in the Atlantic. Thus, when the atmospheric correction is applied to the observed records these differences become smaller. Sea level trends corrected for meteorological effects vary between 2.0 and 2.2 ± 0.4 mm yr 1 in the Northern Spanish coast, between 0.0 ± 0.2 and 0.7 ± 0.2 mm yr 1 in the western basin and between 0.4 ± 0.3 and 0.9 ± 0.3 mm yr 1 in the Adriatic Sea. In terms of interannual variability the atmospheric contribution is responsible of per cent of the total sea level yearly variance, with higher values in the Mediterranean that in the Atlantic sites. The steric contribution, obtained from T and S climatologies, varies amongst regions: maximum values are found at the Strait of Gibraltar with 2.1 ± 0.6 mm yr 1 in Ceuta and southern Adriatic with 1.7 ± 0.3 mm yr 1 for Dubrovnik. The overall steric effect appears to be negative over this period, again linked with cooler winters and negative temperature trends in the upper waters of particular basins in the winter (Tsimplis & Rixen 2002; Painter & Tsimplis 2003). However, the steric correction is based on climatological data which in turn are produced by extensive interpolation of non-systematic observations. Thus significant discrepancies may be introduced by this correction. It is worth noting that most of the Atlantic and Adriatic sites as well as Genova agree on a residual trend of around mm yr 1,after the atmospheric, steric and PGR corrections have been applied. The way the steric signal propagates between the coast and the open ocean is a subject under investigation and further research is needed. The atmospheric and steric corrections lead to better agreement between the sea level trends within the Mediterranean Sea on one hand and those at the coasts of the Iberian Peninsula on the other hand. It was also expected that the corrections we apply would have made trends from shorter records consistent with those estimated for the longer ones and thus provide some confidence in their use. However, the trends of many of the shorter than 35 yr records, even after they were corrected, remained inconsistent with the longer stations. Only land movements due to ongoing GIA have been considered in this work. Other sources of land motion have been neglected and could partly explain the differences observed between some sites. The approach by Woppelmann et al. (2007) is applied only to those tide gauges with collocated GPS with time-series long enough. In the Mediterranean only Marseille and Genova comply with the requirements and the obtained vertical motion is 0.3 ± 0.2 mm yr 1, that is, land subsidence in the NW Mediterranean. This value would decrease the atmospherically corrected sea level trend in these two stations by 50 per cent. However, it is worth mentioning that they used GPS time-series shorter than 7 yr in their computation and this could partly explain the discrepancies with previous works. Decadal changes in sea level trends in the Atlantic and the Mediterranean Sea were not found to be consistent with published global values, at least for parts of the records. Thus, the enhanced sea level rise observed globally in the decade of the 1990s (Cazenave et al. 2001; Holgate & Woodworth 2004) is larger in the MediterraneanSeabyaround5mmyr 1, indicating that global values are not always appropriate to describe long-term changes in sea level in the Mediterranean and that is can be significantly underestimated. On the contrary, during the 1940s and 1970s the mean decadal rate in the Mediterranean is more than 5 mm yr 1 smaller than the global rate. Thus, we conclude that although the Mediterranean Sea decadal sea level changes are partly due to global signals, the local atmospheric and steric variability have dominated both the trends and the decadal variability. We also explored the consistency of short sea level records with the long-term records and identified presently short but apparently consistent tide gauges from others that are not consistent. The consistent records provide redundancy in cases where the long-term record is interrupted. The application of atmospheric and steric corrections does not in general improve the agreement between the short-term records and longer ones. This indicates that other factors not covered by this study contribute to the observed trends. However, the use of shorter records is advisable for checking spurious trends and the reliability of measurements of the longer records over the common period of operation. It is well known that the spatial distribution of tide gauge records in the Mediterranean is biased towards the northern coasts (Tsimplis & Spencer 1997). In addition we have found that most of the tide gauges in the Eastern Mediterranean Basin (except the Adriatic) contain spurious and inconsistent signals which make them useless for future monitoring of sea level changes. Significant efforts by MedGLOSS ( ESEAS ( and the PSMSL promise to improve in the long-term the recovery of data from the north coasts of Africa. However not much progress appears to have been made in this respect since Tsimplis and Spencer (1997) was published.