The anatomy of recent large sea level fluctuations in the Mediterranean Sea

Authors


Corresponding author: F. W. Landerer, Jet Propulsion Laboratory/California Institute of Technology, 4800 Oak Grove Dr., MS 238-600, Pasadena, CA 91109-8099, USA.

(Felix.W.Landerer@jpl.nasa.gov)

Abstract

[1] During the boreal winter months of 2009/2010 and 2010/2011, Mediterranean mean sea level rose 10 cm above the average monthly climatological values. The non-seasonal anomalies were observed in sea surface height (from altimetry), as well as ocean mass (from gravimetry), indicating they were mostly of barotropic nature. These relatively rapid basin-wide fluctuations occurred over time scales of 1–5 months. Here we use observations and re-analysis data to attribute the non-seasonal sea level and ocean mass fluctuations in the Mediterranean Sea to concurrent wind stress anomalies over the adjacent subtropical Northeast Atlantic Ocean, just west of the Strait of Gibraltar, and extending into the strait itself. The observed Mediterranean sea level fluctuations are strongly anti-correlated with the monthly North-Atlantic-Oscillation (NAO) index.

1 Introduction

[2] Sea level change at the regional and local levels can deviate significantly from the global mean change [e.g., Meyssignac et al., 2012]. The latter has risen fairly linearly at a rate of 1.7 ± 0.2 mm/yr over the last century [Church and White, 2011], but the impacts of sea level change occur mostly locally. The Mediterranean coasts are particularly vulnerable due to large population centers and important infrastructure in low-lying delta regions [Rowley et al., 2007]. During the winter months of 2009/2010 and 2010/2011, mean sea level in the Mediterranean Sea reached amplitudes of about 10 cm above the average climatological values. These sub-seasonal variations are the highest recorded values since altimeter measurements started in 1993. As GRACE observations became available in 2002, the amplitudes and phases of seasonal sea level and ocean mass variations in the Mediterranean Sea have been directly observed and analyzed [e.g., Fenoglio-Marc et al., 2006; García-García et al., 2010]. Seasonal sea level variations in the Mediterranean Sea consist of steric and non-steric (i.e., mass) contributions, whereas de-seasonalized sub-annual sea level variations are mostly of non-steric origin [Fukumori et al., 2007]. Model simulations suggest that intra-seasonal sea level changes are mainly mass related, driven by zonal wind stress over the Strait of Gibraltar and over the adjacent Atlantic Ocean [Fukumori et al., 2007; Calafat et al., 2012] and hence water exchange through the Strait of Gibraltar. Furthermore, inter-annual to decadal non-steric sea level changes are closely related to the North Atlantic Oscillation, but the particular mechanisms may differ from the intra-annual fluctuations [e.g., Gomis, 2008; Tsimplis et al., 2008; Calafat et al., 2010; Calafat et al., 2012].

[3] Here we use satellite observations of sea surface height (SSH, from altimetry) and ocean mass (OcM, from GRACE), as well as re-analysis data of atmospheric surface wind stress from 1993 to 2012 to investigate the nature of the recent, large non-seasonal, and sub-annual sea level fluctuations and to assess their connection to large-scale North Atlantic climate variations during this time-period.

2 Data and Methods

2.1 Sea Surface Height From Altimetry

[4] We use the Mediterranean Sea gridded SSH anomalies from October 1992 to May 2012 at 1/8-degree resolution from AVISO [Ducet et al., 2000]. The data are produced by merging observations of Topex/Poseidon, ERS1/2, Jason1, Jason2/OSTM, Envisat, and Geosat Follow-On (GFO) altimetry missions. Several corrections for instrumental errors, geophysical effects, tides, and atmospheric wind and pressure effects are applied [e.g., Le Traon et al., 2003; Carrère and Lyard, 2003; Volkov et al., 2007]. Aliasing that results from the barotropic ocean response to the high-frequency atmospheric forcing is corrected for by applying a dynamic atmospheric correction (DAC) derived from a barotropic ocean model. The DAC combines the high frequencies (<20 days) of the MOG2D-G model with the low frequencies (>20 days) of the inverted barometer (IB) correction [Carrère and Lyard, 2003], making the SSH maps directly comparable to GRACE's OcM observations.

2.2 Ocean Mass from Gravimetry

[5] To infer the non-steric sea level change over the Mediterranean, we use monthly GRACE Release-04 gravity coefficients provided by the Center of Space Research (U of Texas, Austin). The data processing follows standard procedures: we restore background geoid rates removed during processing, replace several low-degree geoid coefficients—which are not or only poorly observed by GRACE, with ancillary data: the C20 coefficient (related to Earth's oblateness) is replaced with one obtained from Satellite-Laser-Ranging [Cheng and Tapley, 2004]—and geocenter changes are based on Swenson et al. [2008]. To analyze ocean mass changes, we add back the monthly atmospheric and oceanic mass de-aliasing fields (GAD), but we subtract the mean atmospheric pressure over the oceans as this pressure term does not change SSH as seen by altimeters. Additionally, we filter the data to reduce correlated errors [Swenson and Wahr, 2006; Duan et al., 2009] and apply a Gaussian smoothing filter of 300 km radius to reduce any remaining noise. Two further processing steps are necessary to retrieve the mean Mediterranean Sea mass changes from GRACE: (1) signal leakage from the surrounding land areas is minimized by applying an iterative technique [Wahr et al., 1998; Chambers et al., 2007] and (2) signal attenuation, which occurs due to the spectral filters and limited bandwidth [e.g., Landerer and Swenson, 2012], is corrected for by applying a gain factor of 1.42 to the basin averaged OcM anomaly. We estimate this gain factor from synthetic surface mass changes which are obtained from a combination of the ECCO ocean model [Fukumori, 2002] and the GLDAS-NOAH land surface model [Rodell et al., 2004]. Our gain factor is somewhat smaller than previously published values [e.g., Fenoglio-Marc et al., 2006; García-García et al., 2010], likely because of smaller averaging radius and differences in the spectral error filters. Finally, the domain average is computed by convolving the basin mask with the GRACE gravity coefficients [Wahr et al., 1998; Swenson and Wahr, 2003]. The error of the mean monthly basin-mean OcM anomalies is ±14 mm of water-equivalent-height (mmH2O), accounting also for the gain factor [Wahr et al., 2006; Landerer and Swenson, 2012].

2.3 Surface Wind Stress

[6] Monthly mean surface wind stress is from the monthly means of the daily atmospheric state data from ECMWF's ERA-Interim re-analysis project [Dee et al., 2011]. The data were obtained at a spatial resolution of 0.75° for the period commensurate with GRACE and altimeter measurements. In the following analysis, we use wind stress over the Mediterranean Sea and over the Eastern Atlantic Ocean (East of 20°W, and between 28 and 40°N). This choice of the region of influence is motivated by findings in previous studies [e.g., Calafat et al., 2010, 2012; Fukumori et al., 2007].

2.4 Methods

[7] To evaluate the temporarily co-varying but not necessarily spatially co-located signals in the sea level (SSH as well as OcM) and wind stress fields, we use a coupled EOF analysis [e.g., Wallace et al., 1992; von Storch and Zwiers, 1999]. The coupled EOF decomposition of two variables extracts those variable-specific spatial patterns that explain most of the common non-seasonal signal variances. We present these patterns in the form of heterogeneous correlation maps (e.g., the EOF pattern of SSH correlated with the PC time series of wind stress and vice versa, etc.); all correlations shown are significant at the 95% confidence level based on a t-test. To focus our analysis on the non-seasonal variations, we remove the multi-year (2005–2009) monthly mean climatology from all data sets.

3 Results

[8] Mean monthly SSH and OcM observations from altimetry and GRACE show considerable differences if the seasonal cycle is not removed because altimetric SSH includes steric signals related to seawater density changes, whereas GRACE only measures the total non-steric mass variations (Figure 1, top). Once we subtract the mean seasonal cycle (Figure 1, bottom), SSH and OcM agree very well, with a correlation of 0.85 and a RMS difference of 15.9 mm. Therefore, the non-seasonal Mediterranean SSH variations are mostly of barotropic nature, while the seasonal SSH variations are mostly steric, consistent with model simulations [Fukumori et al., 2007; García-García et al., 2010; Calafat et al., 2012]. Non-seasonal SSH variations account for only 35% of the overall monthly SSH variance, whereas non-seasonal OcM variations account for 75% of the total OcM variance, indicating that seasonal net surface heat flux exchange with the atmosphere explains most of the total variance of steric SSH.

Figure 1.

Mean Mediterranean SSH (AVISO) versus OcM (GRACE); (top) monthly mean with seasonal variations (not scaled). The annual amplitudes and phases of the time series in Figure 1 agree with previously published values; (bottom) non-seasonal variations (2005–2009 climatology removed and scaled with a gain factor of 1.4). The correlation between non-seasonal SSH and OcM is 0.85, with an RMS difference of 1.59 cm.

[9] Focusing on the non-seasonal variations, two events stand out from the recent observational SSH and OcM records: in January 2010 and December 2010, non-steric sea level anomalies reached amplitudes of 12 and 10 cm, respectively, among the highest values since altimetric SSH measurements became available in 1992. We also note that SSH anomalies similar to the ones observed in January and December 2010 do not occur often as can be seen in the de-seasonalized tide gauge average from Marseille and Trieste [PSMSL, 2012; Woodworth and Player , 2003] (see Supporting Information for details). The December 2010 SSH anomaly is the highest in the 1960–2012 period, while January 2010 is matched in amplitude only by 3–4 other events (Figure S1), and after averaging with a 3-month running mean to remove some of the local variability near the tide gauges, the sea level extremes in January and December 2010 and 2011 are clearly unique over the last 50 years.

[10] Using an adjoint model simulation, Fukumori et al. [2007] found that along-strait concurrent zonal wind stress forcing leads to a stationary balance between wind stress and a pressure gradient between the Atlantic and the Mediterranean, effectively generating a downwind transport into the Mediterranean until a balanced pressure head develops across the connecting region. The region over which the zonal wind stress builds up consists of the Strait of Gibraltar but also the Alboran Sea and the Atlantic Ocean immediately to the west of the strait [Fukumori et al., 2007, Figure 11]. To evaluate if the relationship between wind stress and sea level is also found in the observations, we calculate the coupled EOFs between SSH and wind stress vectors, and the coupled EOFs between OcM and wind stress vectors. The EOFs for SSH and wind stress reveal that a fairly uniform basin-wide sea level anomaly is highly correlated with zonal wind stress immediately to the west of the Strait of Gibraltar, consistent with Fukumori et al. [2007]. The first combined SSH and taux EOF mode explains 93.2% of the total variance, and the corresponding PC time series are correlated at 0.73 (Figure 2a). Meridional wind stress anomalies have a similar pattern but explain slightly less of the variance (88% for the first EOF), and the PC time series have a lower correlation of 0.66 (Figure 2b); some similarity between taux and tauy EOF modes is expected since the two wind components are spatially and temporally correlated. Inside the Mediterranean, SSH in the Adriatic region has the highest correlation with wind stress variations near the Strait of Gibraltar. The coupled EOF patterns and correlations for the two wind stress components and OcM from GRACE are quite similar to the SSH-wind stress patterns (Figure 3). The main difference is the spatial resolution: while the altimetric SSH pattern displays meso-scale variability on top of the basin mode, GRACE cannot resolve scales smaller than 300 km, and hence, mostly a large-scale basin mode is detected. Nonetheless, the highest correlations are obtained toward the Central and Eastern Mediterranean, in particular for the coupled OcM-taux EOF mode. Higher sea level is indeed also seen in the SSH-taux EOF mode, suggesting that the observed GRACE pattern is real and the wind-related basin-mode is actually not quite as uniform as previous coarser-resolution model results may have indicated.

Figure 2.

Coupled first EOF heterogeneous correlation patterns and corresponding principle component time series (normalized by their standard deviation) for non-seasonal SSH and zonal surface wind stress (left panels) and for SSH and meridional surface wind stress (right panels).

Figure 3.

Coupled first EOF heterogeneous correlation patterns and corresponding principle component time series (normalized by their standard deviation) for non-seasonal OcM and zonal surface wind stress (left panels) and for OcM and meridional surface wind stress (right panels).

[11] The combined results so far show that the observed non-seasonal SSH fluctuations are of non-steric nature and mostly originate outside the Mediterranean. This, in turn, necessities a mass transport through the Strait of Gibraltar. The large SSH fluctuations in January and December 2010 occurred over a relatively short period of only 2–4 months. A basin-wide sea level anomaly of 10 cm building up over 2 months implies an influx of approximately 0.05 Sv, which would come on top of the long-term average time-mean net inflow into the Mediterranean through Gibraltar that is of the same order of magnitude [Candela, 2001; García Lafuente et al., 2002]. Turning to the large-scale atmospheric conditions that cause the wind stress anomalies identified in the EOF analysis, we observe that non-seasonal Mediterranean sea level and the monthly PC-based NAO index [Hurrell, 2012] are highly inversely correlated at −0.62 (from 1993 through 2012; Figure S2). As we are using a PC-based NAO index, the peaks in January and December 2010 are associated with large-scale North Atlantic variations, rather than with isolated local pressure variations near the Azores west of Gibraltar Strait. Therefore, the intra-annual near-Gibraltar wind bursts that force the SSH fluctuations appear to be associated with short-term but basin-wide North Atlantic variability.

4 Concluding Discussion

[12] Our results, based on satellite measurements of sea level from altimetry and ocean mass from GRACE, lend observational support for the model-derived hypothesis [Fukumori et al., 2007] that wind setup over a relatively small region can lead to relatively large and basin-wide sea level changes over the Mediterranean Sea. At higher frequencies (i.e., sub-annual), the along-strait wind-setup forces water into the Mediterranean Sea until the zonal SSH pressure gradient balances the wind stress [Fukumori et al., 2007; Menemenlis et al., 2007]; this is essentially a barotropic response. The wind-setup region extends into the Atlantic Ocean, just to the West of the Strait of Gibraltar. Compared to the adjoint model sensitivity patterns of basin-mean SSH to surface wind stress in Fukumori et al. [2007, Figure 11], our coupled EOF correlation patterns for the wind stress (Figures 2 and 3) are somewhat broader and extend further out into the Atlantic Ocean. Whether this reflects actual differences between the model dynamic response and the real ocean response to surface wind stress, forcing cannot be fully resolved here because the coupled EOFs may establish a correlation, but this does not necessarily imply that the wind stress over the entire area is causing the Mediterranean SSH fluctuations.

[13] The influence of the NAO on Mediterranean sea level over longer time scales has been described in terms of the atmospheric pressure loading (IB effect) and also through NAO-related variations in evaporation and precipitation over the area [e.g., Tsimplis and Josey, 2001]. Our preliminary analysis of ERA-Interim precipitation fields indicates that the two sea level events in January and December 2010 cannot be explained by anomalous land or atmosphere freshwater fluxes (not shown here). More recently, tetcalafat:12 used tide gauges distributed along European coasts to demonstrate that on decadal time scales, NAO-related Mediterranean sea level variations can be traced to long-shore wind forcing northward of 25° and subsequent northward boundary wave propagation of coastally trapped waves. Upon reaching the Strait of Gibraltar, any SSH difference between the Atlantic and Mediterranean is rapidly equilibrated through net mass flow through the Strait. Furthermore, Calafat et al. [2012] demonstrate that at decadal time scales, most of the NAO-related Mediterranean SSH variance is induced by baroclinic processes over the Eastern North Atlantic. In contrast, our results here indicate that monthly wind stress anomalies near the Strait of Gibraltar and the resulting Mediterranean sea level change appear also to be related to large-scale North Atlantic variability.

[14] The high-resolution AVISO SSH data set also reveals some deviations from a uniform, basin-wide sea level response in the Mediterranean to non-seasonal wind stress forcing. SSH in the Adriatic region has the highest correlation with wind stress variations near the Strait of Gibraltar (Figure 2). The correlation pattern between Mediterranean SSH fluctuations and the near-strait zonal wind forcing suggests a seiche-like response within the Mediterranean, likely aided by the wind stress over areas with higher correlation immediately to the East of the Strait of Gibraltar. The GRACE observations, although much coarser and hence smoother, also indicate deviations from a simple uniform basin-wide sea level response (Figure 3). Since the wind-stress-driven Mediterranean sea surface height fluctuations described here occur rapidly at time scales of a month or less, it appears warranted to carefully assess possible aliasing effects in the GRACE data fields over the Mediterranean, in particular when different ocean de-aliasing models are used.

5 Acknowledgments

[15] This work represents one phase of research carried out at the Jet Propulsion Laboratory/California Institute of Technology. FWL and DV were supported by NASA's Physical Oceanography Program. SSH observations are processed by SSALTO/DUACS and distributed by AVISO with support from CNES; ERA-Interim Re-analysis data are provided by the European Center for Medium Range Weather Forecast (ECMWF). We thank the German Space Operations Center (GSOC) of the German Aerospace Center (DLR) for providing continuously and nearly 100% of the raw telemetry data of the twin GRACE satellites.

Ancillary