Surface Geostrophic Circulation of the Mediterranean Sea Derived from Drifter and Satellite Altimeter Data

Transcription

1 JUNE 2012 P O U L A I N E T A L. 973 Surface Geostrophic Circulation of the Mediterranean Sea Derived from Drifter and Satellite Altimeter Data PIERRE-MARIE POULAIN, MILENA MENNA, AND ELENA MAURI Istituto Nazionale di Oceanografia e di Geofisica Sperimentale, Trieste, Italy (Manuscript received 6 September 2011, in final form 26 January 2012) ABSTRACT Drifter observations and satellite-derived sea surface height data are used to quantitatively study the surface geostrophic circulation of the entire Mediterranean Sea for the period spanning After removal of the wind-driven components from the drifter velocities and low-pass filtering in bins of week, maps of surface geostrophic circulation (mean flow and kinetic energy levels) are produced using the drifter and/or satellite data. The mean currents and kinetic energy levels derived from the drifter data appear stronger/higher with respect to those obtained from satellite altimeter data. The maps of mean circulation estimated from the drifter data and from a combination of drifter and altimeter data are, however, qualitatively similar. In the western basin they show the main pathways of the surface waters flowing eastward from the Strait of Gibraltar to the Sicily Channel and the current transporting waters back westward along the Italian, French, and Spanish coasts. Intermittent and long-lived subbasin-scale eddies and gyres abound in the Tyrrhenian and Algerian Seas. In the eastern basin, the surface waters are transported eastward by several currents but recirculate in numerous eddies and gyres before reaching the northward coastal current off Israel, Lebanon, and Syria and veering westward off Turkey. In the Ionian Sea, the mean geostrophic velocity maps were also produced separately for the two extended seasons and for multiyear periods. Significant variations are confirmed, with seasonal reversals of the currents in the south and changes of the circulation from anticyclonic (prior to 1 July 2007) to cyclonic and back to anticyclonic after 31 December Introduction Observations of surface currents in the Mediterranean using drifters have been obtained since the advent of satellite tracking in the 1980s, providing unique information on the currents over a wide spectrum of scales (from tidal/inertial to interannual, from a few kilometers to thousands of kilometers). Due to their Lagrangian nature, drifters move with the currents and can cover substantial geographical areas if they are not picked up or stranded prematurely. Thus, descriptions of the circulation in specific Mediterranean subbasins have been produced and have provided novel insights on the near-surface dynamics. However, Mediterranean drifters are rather short lived (not necessarily due to power autonomy but to the high probability of stranding and pickup in the highly navigated and confined Mediterranean areas) and, as a result, these descriptions are only Corresponding author address: Pierre-Marie Poulain, Istituto Nazionale di Oceanografia e di Geofisica Sperimentale, Borgo Grotta Gigante, 42/c, Sgonico, Trieste, Italy. ppoulain@ogs.trieste.it valid for limited time periods. More generally speaking, since deployments cannot be made everywhere and frequently due to limited resources, drifter datasets are usually gappy in both space and time. In particular, the temporal distribution of the data can be very intermittent. This is even more true for the entire Mediterranean Sea, which has been sampled sporadically by drifters since the mid-1980s. The main consequence is that a statistical description of the surface circulation, due to the nonuniform sampling in space and time, can be quite biased and strongly dependent on the specific data distribution. Despite the aforementioned limitations of drifter datasets, novel quantitative descriptions of the circulation have been produced for several Mediterranean areas, such as the Adriatic Sea (Poulain 1999, 2001; Ursella et al. 2006), the Sicily Channel (Poulain and Zambianchi 2007), and the Eastern (Gerin et al. 2009) and Western (Salas et al. 2001; Poulain et al. 2011, manuscript submitted to Boll. Geofis. Teor. Appl.) Mediterranean basins. A first attempt to map the entire Mediterranean Sea surface circulation using the drifter data available in the 1980s and 1990s was made by Mauerhan (2000). DOI: /JPO-D Ó 2012 American Meteorological Society

2 974 J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y VOLUME 42 Satellite-tracked drifters can slip with respect to the near-surface water because of the direct effect of the local wind and waves acting on the drifter parts protruding above the sea surface. This slip varies substantially among the different drifter designs. Wind-induced slips and wind-driven Ekman surface currents can be estimated from drifter data using surface wind products and simple complex regression models (Ralph and Niiler 1999; Rio and Hernandez 2003; Poulain et al. 2009; Centurioni et al. 2009). Assuming negligible Stokes and wind-independent ageostrophic currents (e.g., in the open sea and with low sea states), the wind-driven component (direct slip and Ekman component) can be removed from the drifter velocities to obtain estimates of the surface geostrophic currents. In contrast, satellite altimetry data can be used to estimate statistics of the surface geostrophic circulation with less biases since sampling is uniform in both space and time. A drawback is that altimetry data are generally low-pass filtered due to the satellite subtrack patterns and repeat times, not to mention inaccuracies in coastal areas. In particular, the Archiving, Validation, and Interpretation of Satellite Oceanographic data (AVISO) products are interpolated on a regular grid with mesh size of 1 /88 every week. Interpolation and sampling errors are not negligible but are believed not to bias the derived geostrophic velocity statistics. Most importantly, the absolute geostrophic currents can only be obtained if the geoid (equipotential surface) is known with sufficient accuracy and spatial resolution. Given the aforementioned considerations, it becomes obvious that the combination of drifter and altimetry data is ideal to provide unbiased representations of the surface circulation. Drifter and satellite altimetry data have already been combined to estimate the surface currents of the Tyrrhenian Sea (Rinaldi et al. 2010), the entire Mediterranean (Rio et al. 2007), and the World Ocean (Rio and Hernandez 2004). In this paper, the combination is applied to the Mediterranean Sea following the method developed by Niiler et al. (2003) and Centurioni et al. (2008). The goal of this paper is to produce the best quantitative representation of the Mediterranean Sea surface geostrophic circulation during the period After a brief description of the drifter, wind, and altimeter data used and some explanations of the methodologies employed to separate wind-driven and geostrophic velocity components and to combine the drifter and altimetry data (section 2), pseudo-eulerian maps of Mediterranean geostrophic velocity statistics (means and energy levels) are presented and discussed in section 3. In the central Mediterranean (Ionian Sea), statistics are also described separately for the different seasons and multiyear periods to assess long-term variations of the basinwide circulation. Section 4 includes a discussion of the results and final conclusions. 2. Data and methods a. Datasets 1) DRIFTERS The majority of the Mediterranean drifters considered in this work are of three types: Surface Velocity Program (SVP) drifters, which are the standard design of the Global Drifter Program (Lumpkin and Pazos 2007); Coastal Ocean Dynamics Experiment (CODE) drifters, which were developed by Davis (1985) in the early 1980s to measure coastal surface currents; and Compact Meteorological and Oceanographic Drifters (CMOD) or XAN-1 drifters (Selsor 1993), which were mainly operated by the U.S. Navy. In addition, a limited number of A106/111 drifters were considered. These systems are fitted with a holey sock drogue centered at a nominal depth of 10 m (Font et al. 1998; Salas et al. 2001) but no sensor indicates the drogue presence. The SVP drifters used in the Mediterranean starting in 2005 are the mini World Ocean Circulation Experiment (WOCE) SVP drifters. They consist of a surface buoy tethered to a holey-sock drogue, centered at a nominal depth of 15 m, that holds the drifter almost motionless with respect to the horizontal layer studied [for details on the SVP design, see Sybrandy and Niiler (1991)]. They haveadragarearatioofthedroguetothetetherand surface buoy in excess of 40. A tension sensor, located below the surface buoy where the drogue tether is attached, indicates the presence or absence of the drogue. Measurements of the water-following capabilities of the SVP have shown that, when the drogue is attached, they follow the water to within 1 cm s 21 in 10 m s 21 winds (Niiler et al. 1995). CODE drifters consist of a slender, vertical, 1-m-long negatively buoyant tube with four drag-producing vanes extending radially from the tube over its entire length and four small spherical surface floats attached to the upper extremities of the vanes to provide buoyancy (Poulain 1999). Comparisons with current meter measurements (Davis 1985) and studies using dye to measure relative water movements (D. Olson 1991, personal communication) showed that the CODE drifters follow the surface currents to within 3 cm s 21, even during strong wind conditions. More recent slippage measurements (Poulain et al. 2002) with acoustic current meters positioned at the top and bottom of the drifter showed that the CODE drifters follow the surface currents within 2cms 21 and that they move in a manner consistent with

3 JUNE 2012 P O U L A I N E T A L. 975 TABLE 1. Quantity of drifter data in the Mediterranean for the period corresponding to the different types of drifters considered (CMOD, CODE, SVP, and A106/A111). The time span covered by the different drifter categories is indicated with the first and last days of available data. Drifter type Number of drifter tracks Drifter days First day Last day Max life (days) Mean half-life (days) CMOD Oct Mar CODE Nov Dec SVP drogued Sep Dec SVP undrogued Sep Dec SVP unknown Aug Dec A106/A Oct Jan Tot the near-surface Ekman dynamics with a velocity component to the right of the prevailing wind. CMOD drifters are sonobuoys that consist of a 60-cmlong aluminum cylindrical hull with a floatation collar (35-cm overall diameter). They are drogued with the sonobuoy case on a 100-m-long (4 m for a few of them) tether, resulting in a wet to dry area ratio of about 5 (Matteoda and Glenn 1996). Water-following capabilities of the CMOD have not been assessed and there is no sensor of drogue presence. Given its small wet to dry area ratio and that the case is believed to detach rapidly from the surface buoy after deployment, we expect that direct wind effects can be important, so we can consider that all the CMOD drifters measure currents at the surface with significant errors due to wind slippage. All drifters were localized by, and transmitted data (sea surface temperature, voltage, drogue presence indicator, etc.) to, the Argos Data Collection and Location System (DCLS) onboard polar-orbiting satellites. Some units (mostly CODE drifters) were also equipped with GPS receivers to obtain more accurate (;10 m) and more frequent (hourly) positions. The drifter positions were edited for outliers and spikes using statistical and manual techniques with criteria based on maximum distance, maximum speed, and maximum angle between two consecutive points, as described in Poulain et al. (2004). Edited positions were interpolated at regular intervals with a kriging optimal interpolation technique (Hansen and Poulain 1996). Velocities were estimated by central finite differencing the interpolated positions. Positions and velocities were subsequently low-pass filtered using a Hamming filter with 36 h to remove high frequency motions and subsampled every 6 h. The data were finally archived in several databases (organized by subbasins or projects) accessible through the Mediterranean SVP (MedSVP) website ( In total, for the time period October 1992 December 2010, more than 221 drifter years worth of data were obtained in most areas of the Mediterranean Sea, corresponding to 1218 individual tracks. The SVP units were sorted into three categories: drogued, undrogued, and unknown (no information on drogue presence). The quantities of drifter tracks and drifter days are listed for all the drifter types/categories considered in Table 1, along with the specific time spans covered. Information on the drifter life times is also included (maximal operational periods and mean half-lives of the drifters). The drifters were deployed throughout the Mediterranean (see Mediterranean geography map with deployment locations in Fig. 1) as part of several projects, especially in the Adriatic Sea and the Sicily Channel. The drifter tracks are depicted in Fig. 2. It can be seen that CODE drifters mostly cover the Adriatic, Ionian, Aegean, Tyrrhenian and the Liguro Provencxal subbasins and that drogued/undrogued SVP units dominate in the eastern Mediterranean basin. SVP drifters with no information on the drogue presence and CMOD drifters are found in most areas. Drifters A106/111 mainly provided data in the southern Algerian subbasin. The temporal distribution of the drifter data (Fig. 3) is intermittent owing to the relative short lifetime of the drifters, with maxima (exceeding 40 drifters operating simultaneously) occurring in November 1995, March 1998, May 2003 [mostly CODE units in the Adriatic Sea (Poulain 2001; Ursella et al. 2006) and in the Sicily Channel and Ionian Sea (Poulain and Zambianchi 2007)], and May 2006 (SVP units in the eastern basin, Gerin et al. 2009). The longest drifter track (575 days) corresponds to a SVP drifter with unknown drogue presence. The mean halflives vary between 38 (CMOD) and 94 days (SVP). The drifter velocities were not corrected explicitly for wind slippage because correction algorithms are not available for the A106/A111, CMOD, CODE, and SVP with unknown drogue presence, that is, for the majority of the drifters. To obtain good coverage of the entire Mediterranean Sea, the velocities of all the drifter designs were considered together, despite their different windinduced errors and drogue depths. Hence, we should bear in mind that the drifter dataset used here represents measurements of the near-surface currents (between 0 and 15 m) that can be affected by wind slippage errors.

4 976 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 42 FIG. 1. Geography and bathymetry of the Mediterranean Sea. Drifter deployment positions are shown with colored dots. 2) SATELLITE ALTIMETRY The sea level anomaly (SLA) data used for this study are the gridded (1/ 88 Mercator projection grid) SSALTO/ DUACS weekly, multimission [Jason-1, Ocean Topography Experiment (TOPEX)/Poseidon, Envisat, and Geosat Follow-On (GFO)], delayed time (quality controlled) altimetry product from AVISO (AVISO 2011). The time span for the AVISO data is from 14 October 1992 to 31 December SLA data are defined with respect to a 7-yr mean ( ). The corresponding anomalies of surface geostrophic velocities were also downloaded. In addition, data of the sea level height above the geoid, or the absolute dynamic topography, and the absolute surface geostrophic currents were obtained from AVISO. The absolute dynamic topography is defined by adding the SLA and the synthetic mean dynamic topography (SMDT) of Rio et al. (2007). 3) WIND PRODUCTS Cross-calibrated, multiplatform (CCMP) ocean surface wind velocity products were downloaded from the NASA Physical Oceanography Distributed Active Archive Center (DAAC) for the period July 1987 December 2010 (Atlas et al. 2009). These products were created using a variational analysis method to combine wind FIG. 2. Composite diagram of the interpolated and low-pass-filtered drifter tracks in the Mediterranean Sea. Different colors correspond to different drifter types and drogue depths.

5 JUNE 2012 P O U L A I N E T A L. 977 TABLE 2. Results of the regression model, Eq. (1), applied to extract the wind-driven currents from the velocities of the different types of drifters: R 2 is the coefficient of determination and N is the number of observations considered. Drifter type b R 2 (%) RMSE (cm s 21 ) N CMOD 0.02 exp(2188i) CODE 0.01 exp(2338i) SVP drogued exp(2348i) SVP undrogued 0.01 exp(2268i) SVP unknown exp(2328i) A106/ exp(2218i) FIG. 3. Temporal distribution of the Mediterranean Sea drifter dataset (number of drifter days per day), spanning 14 Oct Dec Different colors correspond to different drifter types and drogue depths. measurements derived from several satellite scatterometers and microwave radiometers. Six-hourly gridded analyses with 25-km resolution were used (level 3.0, firstlook version 1.1). b. Methods 1) PSEUDO-EULERIAN VELOCITY STATISTICS First, the velocities obtained from the different sources (winds, geostrophic currents from altimetry and drifter velocities) were all low-pass filtered with a 36-h Hamming filter in order to use data filtered in the same way. Second, the filtered velocities obtained from altimetry and from the drifters were averaged in overlapping (50%) geographical bins of organized on a 0.58 grid and in time intervals of one week. The averages corresponding to bins and weeks with less than three drifter observations were not considered. Then, maps of mean circulation and kinetic energy of the mean flow [mean kinetic energy (MKE)] and of the fluctuations [eddy kinetic energy (EKE)] were constructed. See Poulain (2001) for definitions of these statistics. These statistical results, known as pseudo-eulerian statistics, are only considered in bins where the number of observations is above a minimum threshold (chosen practically as 20 weeks in this work). The bin size was chosen as a compromise to resolve the basin and subbasin circulation of the Mediterranean Sea. Mesoscale and submesoscale velocities are not resolved and are included in the velocity variance and the EKE, as well as temporal variations at scales ranging between a few weeks and several years. Two kinds of pseudo-eulerian statistics were calculated. The first one considers only observations at the locations and times of the drifters. Given the nonuniform spatial and temporal distribution and the scarcity of the drifter data, these statistics are called biased and written as h i b. An example is the map of geostrophic surface currents derived from the drifter data. The other kind is based on regularly sampled observations, such as satellite altimetry data. We denote these results as unbiased and use the symbol h i u. 2) ESTIMATION OF WIND-DRIVEN CURRENTS The filtered CCMP winds (W) and the absolute geostrophic velocities derived from altimetric measurements and the SMDT of Rio et al. (2007) (U AGRIO ) were interpolated at the drifter locations with a bilinear scheme. Then, the absolute geostrophic velocities were removed from the drifter velocities U D and the following regression model was applied to estimate the winddriven component (U wind-driven ) (slip and Ekman) of the drifter velocities following Ralph and Niiler (1999) and Poulain et al. (2009), U D 2 U AGRIO 5 be iu W 1 error 5 U 1 error, wind-driven (1) where b is a real constant, u is an angle (positive anticlockwise), and velocities are expressed as complex numbers (e.g., W 5 w 1 1 iw 2, indices 1 and 2 corresponding to the zonal and meridional directions). Introducing the variation of the Coriolis parameter with latitude in the regression model did not yield better or significantly different results. Likewise, adding a constant term in the regression did not change significantly the results. Equation (1) was applied to the different drifter systems separately, over the whole Mediterranean Sea, and the results are presented in Table 2. The skill (or coefficient of determination) and the rms error (RMSE) range from 1% to 27% and 2.4 to 4.8 cm s 21, respectively. The SVP drifters with drogue attached measured the smallest wind-driven currents, amounting to 0.5% of

6 978 J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y VOLUME 42 FIG. 4. Spatial distribution of the number of weeks worth of drifter data in bins of for the period 14 Oct Dec the wind speed and directed about 348 to the right of the wind. These currents represent only 1% of the total velocity variance. The SVP without drogue and CODE drifters had wind-driven currents of 1% of the wind speed at an angle of 268 and 338 to the right of the wind, respectively. The corresponding explained variance is about 22% and 11%. The difference between the results obtained for the SVP with and without drogue is compatible with the findings of Pazan and Niiler (2001) and Poulain et al. (2009); that is, the undrogued SVP has a downwind slip of 0.5% 1% of the wind speed with respect to the drogued SVP. The regression coefficients for the CODE and drogued/undrogued SVP drifters are similar to those estimated by Poulain et al. (2009) in the Eastern Mediterranean. As expected given their worst water-following capabilities, the CMOD drifters move more downwind (about 188 to the right of the wind) with wind-driven currents amounting to 2% of the wind speed. A significant portion of the velocity variance is due to these wind-driven currents (about 27%). The SVP drifters with unknown drogue presence and the A106/111 units have similar wind effects: that is, a wind-driven current of about 0.7% of the wind speed at 328 and 218 to the right of the wind, respectively, and a low explained variance (5% 7%). The regression of Eq. (1) was also applied without removing the AVISO absolute surface geostrophic currents. The results are essentially identical to those listed in Table 1, except that the explained velocity variance is smaller, varying between 1% and 19% (Menna et al. 2010). 3) COMBINATION OF DRIFTER AND SATELLITE ALTIMETRY DATA TO CREATE UNBIASED GEOSTROPHIC VELOCITY STATISTICS Drifter and satellite altimetry data were combined following the method described in Niiler et al. (2003) and Centurioni et al. (2008). The drifter velocities and the anomalies of surface geostrophic velocities were first preaveraged in spatial bins of and over 7-day intervals. The spatial bins and time intervals with less than three drifter observations were not considered. The spatial distribution of the number of weeks with observations is shown in Fig. 4. This number reaches values above 100 weeks in the northwestern Mediterranean, Sicily Channel, Adriatic Sea, and the southern Levantine subbasin. Spatial bins with less than 20 weeks are not considered in the following statistical analyses. With this threshold the following areas are excluded due to their scarcity of drifter data: the Alboran Sea, Tunisian shelf, Aegean Sea, and northwestern Levantine. It is important to note that 100 weeks corresponds to a temporal window of only ;10% of the entire time period considered (14 October December 2010). Consequently, the pseudo-eulerian velocities calculated from the drifter data are based on data that cover only 2% 10% of the time period considered and, as a result, can be significantly biased and are not representative of the period considered. In contrast, the satellite data has continuous and uniform coverage over the entire Mediterranean Sea. The complex correlation (Kundu 1976) between U DG and U SLA was first calculated, where U DG and U SLA are the bin-averaged ( week) drifter velocities from which the wind-driven components, based on Eq. (1) and the coefficients listed in Table 1, that is, U DG 5 U D 2 U wind-driven, were removed, and the anomalies of surface geostrophic velocities, obtained from AVISO, respectively. The magnitude of the correlation is depicted in Fig. 5. Large values (above 0.5) are found in the Algerian subbasin, in the central Tyrrhenian and Ionian Seas, and in the central Levantine subbasin. In contrast, the correlation between drifter- and altimetry-derived velocities is low in the Adriatic and Aegean Seas and

7 JUNE 2012 P O U L A I N E T A L. 979 FIG. 5. Spatial distribution of the magnitude of the complex correlation, between drifter geostrophic velocities U DG and the anomalies of AVISO geostrophic velocity U SLA. on the Tunisian shelf. These regions were discarded in the following computations. Within each spatial bin the following regression model was applied: U DG 5 AU SLA 1 B 1 error, (2) where the unknowns A and B are complex numbers or 2D vectors. The slope A was subsequently low-pass filtered with a running spatial mean to remove insignificant noise. In contrast, the offset B was kept at the 18 resolution because we believe that it contains useful information related to the geoid and the mean dynamic topography at this smaller scale. It is actually the time-mean circulation for the period used to define the SLA. The magnitude of the low-pass filtered A,(jAj), is shown in Fig. 6a, whereas B is depicted in Fig. 6b with vectors and color-coded magnitudes. The magnitude of A varies mostly between 1 and 2. It exceeds 1 in areas such as the southwestern Balearic Sea, the Algerian subbasin, the northern Tyrrhenian Sea, the central Ionian Sea, the southern Cretan Passage, and the central Levantine subbasin. In contrast, it is around or slightly less than one in the Liguro Provencxal subbasin, the Sicily Channel, the eastern Tyrrhenian Sea, the northern Ionian, and in some areas of the Levantine. The veering angle induced by A (no shown) is not significantly from zero in most places. Deviations of jaj from unity are mainly due to oversmoothing of the satellite altimeter data and to the existence of residual wind-driven components, nonlinear boundary currents, and ageostrophic acceleration (Niiler et al. 2003) in the drifter velocities. The offset B can be as large as 20 cm s 21 in the fast currents of the Northern Current (Liguro Provencxal subbasin), the Algerian Current, and the Sicily Channel and strong circulation features in the Levantine subbasin (e.g., the Ierapetra gyre, the Mersa-Matruh eddy, and the Asia Minor Current). The goodness of fit of the regression model expressed by Eq. (2) can be represented by the RMSE (see Fig. 6c). The discrepancy between the drifter- and satellite-derived surface geostrophic velocities can be as large as 6 cm s 21 in some areas of the Mediterranean Sea where currents are strong and highly variable such as in the Algerian and Levantine (vicinity of the Ierapetra gyre and Mersa-Matruh eddy) subbasins. The RMSE can be quite small (,2 cms 21 )in the following areas: the Liguro Provencxal subbasin, the Tyrrhenian Sea, the Sicily Channel, and the southeastern Levantine. Once the unknowns in Eq. (2), A and B, have been estimated, unbiased maps of pseudo-eulerian statistics can be estimated from the satellite altimetry data; for instance, the mean surface geostrophic circulation can be written as hu G i u 5 AhU SLA i u 1 B, (3) where h i u indicates unbiased temporal averages in each spatial bin and U G is the surface geostrophic current. Equation (3) is simply the unbiased mean of Eq. (2). 4) COMPARISON OF BIASED AND UNBIASED ESTIMATES OF MEAN SURFACE GEOSTROPHIC CIRCULATION Before exploiting the unbiased velocity statistics to describe the kinematics of the surface waters in the Mediterranean Sea, it is interesting to assess the difference between the biased maps obtained from drifters and the unbiased results coming from the combination of satellite altimetry and drifter data. Averaging Eq. (2) in the bins where drifter data exist (biased average denoted as h i b ), we obtain

8 980 JOURNAL OF PHYSICAL OCEANOGRAPHY VOLUME 42 FIG. 6. Spatial distribution of (a) the slope jaj and (b) the offset B of the regression model [Eq. (2)] in bins of for the period 14 Oct Dec The slope jaj has been low-pass filtered with a running mean. (c) Spatial distribution of the RMSE corresponding to Eq. (2). hug ib 5 AhUSLA ib 1 B, (4) and by subtracting Eqs. (3) and (4) the difference between the biased and unbiased mean maps of surface geostrophic circulation becomes hug ib 2 hug iu 5 A(hUSLA ib 2 husla iu ). (5) This difference (Fig. 7) is large in strong currents and eddies of the Algerian subbasin, the northern Ionian Sea, the southern Cretan Passage, and the northeastern

9 JUNE 2012 P O U L A I N E T A L. 981 FIG. 7. Difference between the biased and unbiased means [hu G i b hu G i u 5 A (hu SLA i b hu SLA i u )] of the surface geostrophic circulation in the Mediterranean for the period 14 Oct Dec Levantine (Asia Minor Current) where it can reach amplitudes of 5 cm s 21. As a result the Algerian Current, the southward Mid-Ionian Jet, and the Asia Minor Current are stronger in the biased map obtained from drifter data only. 3. Results a. Surface geostrophic circulation in the entire Mediterranean Sea The biased and unbiased mean surface geostrophic circulation maps in the entire Mediterranean are presented in Fig. 8. The mean circulation derived from the drifter data only (Fig. 8a) shows the following major features in the Western Mediterranean: 1) a strong anticyclonic western Alboran gyre, with mean speeds exceeding 30 cm s 21 ; 2) a westward Algerian Current, with mean speeds over 30 cm s 21 off central Algerian and Tunisia and related instability features southeast of Sardinia; 3) a branching of the Algerian Current into a strong current in the southern Tyrrhenian and a flow entering the Sicily Channel; 4) a major cyclonic eddy east of the Strait of Bonifacio in the northern Tyrrhenian; and 5) a southwestward mean flow, the Northern Current, in the northern Ligurian Sea and in the vicinity of the Gulf of Lion. In the Eastern Mediterranean, the Atlantic Ionian Stream is evident in the Sicily Channel, extending into the Ionian in two branches, one crossing the basin in its central part and headed toward the southern Levantine and the other forming a basinwide anticyclonic circulation in the Ionian, with northeastward currents south of Italy, the southward Mid-Ionian Jet in the eastern side (at E), and a return flow to the northwest and north in the southern Ionian to close the basinwide anticyclone. A strong anticyclone, the Pelops eddy, is also evident south of the Peloponese, as well as the west Cretan cyclone southwest of Crete. In the southern Levantine, a coastal/slope current is not as evident as the Algerian Current in the western basin. Eastward flow is significant in the central Crete Passage. More to the east, the Ierapetra gyre and Mersa-Matruh eddies are coarsely represented with disorganized strong mean currents. East of the Nile River delta, there are strong coastal currents flowing off Egypt, Israel, Lebanon, Syria, and Turkey (Cilician and Asia Minor Currents). In the Adriatic and Aegean Seas, the surface circulation is essentially cyclonic with enhanced southeastward currents on the western sides. As discussed above, the unbiased mean circulation map calculated with Eq. (3) (Fig. 8b) shows essentially the same general circulation patterns of the Mediterranean. Differences are substantial in areas of strong currents and strong variability, such as the Algerian subbasin, the northern Tyrrhenian Sea, the northern Ionian Sea, and the Levantine subbasin. The mean circulation estimated from absolute geostrophic velocities derived from altimetric measurements and the SMDT of Rio et al. (2007) hu AGRIO i u is substantially different from the two other mean maps (Fig. 8c). First, strong currents such as the Algerian Current, the North Current, the Atlantic Ionian Stream, the Mid-Ionian Jet, and the Cilician and Asia Minor Currents are weaker. Second, the cyclonic features in the northern Tyrrhenian and southwest of Crete are less pronounced. Third, the Ierapetra gyre and Mersa- Matruh eddies are smoothed out in the eastern Levantine subbasin. Fourth and mostly importantly, the basinwide anticyclones in the northern and southern Ionian are almost nonexistent, and most of the eastward transport of surface waters appears to take place as a

10 982 J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y VOLUME 42 FIG. 8. Pseudo-Eulerian mean surface geostrophic circulation in the Mediterranean Sea in spatial bins of and for the period 14 Oct Dec 2010; bins with less than 20 weekly observations are not considered. (a) Biased estimates directly derived from the drifter data, (b) unbiased estimates using Eq. (3) and satellite altimeter data (excluding the Adriatic Sea, Aegean Sea, and Tunisian shelf), and (c) unbiased estimates using the SMDT of Rio et al. (2007) and satellite altimeter data.

11 JUNE 2012 P O U L A I N E T A L. 983 southeastward current connecting the Sicily Channel to the southern Cretan Passage in the central Ionian, eventually extending toward the east as the Libyan Egyptian Current and splitting into two branches near 288E one coastal and the other flowing toward Cyprus in the central Levantine (sometimes referred to as the Mid-Mediterranean Jet). The mean kinetic energy of the surface geostrophic circulation derived from the drifter data (not shown) is maximal in the Alboran gyre, Algerian Current, Northern Current, and Atlantic Ionian Stream, in the vicinity of the Ierapetra gyre and Mersa-Matruh eddies, and in the Cilician and Asia Minor Currents, reaching values above 900 cm 2 s 22, that is, 30 cm s 21 rms. Elsewhere, the MKE is essentially less than 100 cm 2 s 22. The unbiased map of MKE (not shown) is similar to the biased estimate except that energy levels can be slightly lower in most areas of the Mediterranean Sea. It is even lower when computed with the Rio et al. (2007) absolute geostrophic surface currents (now shown). The biased map of the kinetic energy of fluctuating geostrophic currents (EKE, Fig. 9a) derived directly from the drifter data shows large values (.400 cm 2 s 22 )inthe Alboran Sea, the Algerian subbasin with extension toward Sardinia, the northwestern Ionian Sea, the southern Cretan Passage, in the vicinity of the Ierapetra gyre and Mersa- Matruh eddies, and the northeastern Levantine. In contrast, the unbiased map of EKE (Fig. 9b), shows values 5 20 times smaller than those of the biased estimates, reaching only values of ;200 cm 2 s 22 in the Algerian Current. It is interesting to compare these results to the values of EKE derived from U AGRIO, which are even smaller (maximum near 100 cm 2 s 22 in the Algerian Current, Fig. 9c). The ratio between our new unbiased estimates and those of Rio et al. varies between 1 and about 4 (not shown), consistent with the square of jaj depicted in Fig. 6a. Thus, the strength of the velocity fluctuations at temporal scales longer than one week and averaged in bins is much more important when calculated directly from the drifters with respect to those obtained from the satellite altimeter data corrected with Eq. (2) or those using the SMDT of Rio et al. (2007). The main causes for this discrepancy are 1) the possible residual wind-driven and nongeostrophic currents measured by the drifters; 2) the spatial smoothing applied to the satellite altimeter data [100-km scale, see Pujol and Larnicol(2005)]; and 3) the spatially varying and sometime low number of data used to compute EKE with the drifters, that is, the varying and large error bars). b. Surface geostrophic circulation in the Ionian Sea Several experimental and numerical studies have documented that the basinwide circulation in the central Mediterranean Sea, in particular the Ionian Sea, varies significantly at seasonal and decadal scales [see the numerical simulations of Pinardi and Navarra (1993), Pinardi et al. (1997), and Demirov and Pinardi (2002); the altimeter data of Larnicol et al. (2002) and Pujol and Larnicol (2005); and the drifter measurements of Poulain (1998), Mauerhan (2000), Poulain and Zambianchi (2007), and Gerin et al. (2009)]. This variability is related to variations in the atmospheric forcing and to the vorticity transfer due to the redistribution of water masses produced by the Eastern Mediterranean transient (EMT) (Borzelli et al. 2009; Gačić et al. 2011). Hereafter, we present statistical results on the temporal variability of the surface geostrophic circulation for the Ionian Sea, with partial extensions to the Sicily Channel and Cretan Passage. These statistics are only estimated for the Ionian subbasin because 1) data are more abundant in that region and 2) the long-term variations and reversals of mean circulation are mostly well documented in this Mediterranean subbasin. The seasonal variability of the surface circulation in other Mediterranean subbasins, for example, in the Tyrrhenian (Rinaldi et al. 2010) and Adriatic (Poulain 2001) Seas, are not discussed in this paper. 1) SEASONAL STATISTICS The satellite altimeter products U SLA were used in concert with Eq. (3) to estimate the mean surface geostrophic circulation in the Ionian Sea using the values of A and B illustrated in Fig. 6. The mean was defined over two perpetual extended seasons spanning from October 1992 to December 2010: the extended winter (November April) and the extended summer (May October). The results are depicted in Fig. 10. Some circulation features appear to be independent of the seasons, such as the Pelops and west Cretan eddies centered near N, E and N, E, respectively. The basinwide clockwise circulation in the northern Ionian remains essentially invariant with the seasons. In contrast, significant seasonality is evident in the Atlantic Ionian Stream south of Sicily (with maximum in summer), in the outflow toward the Levantine (stronger eastward zonal flow off Libya near longitude E in winter). In the southern Ionian the westward surface flow is stronger near N during the extended summer. The eastward coastal current off Libya between 138 and 198E appears more robust in winter. 2) MULTIYEAR STATISTICS Unbiased mean surface geostrophic circulation maps of the Ionian Sea were also constructed for three consecutive periods: from October 1992 to June 1997, July 1997 to December 2005, and January 2006 to December

12 984 JOURNAL OF PHYSICAL OCEANOGRAPHY FIG. 9. As in Fig. 8, but for the EKE: (a) biased estimates directly derived from the drifter data, (b) unbiased estimates using Eq. (2) and satellite altimeter data (excluding the Adriatic Sea, Aegean Sea, and Tunisian Shelf), and (c) unbiased estimates using the SMDT of Rio et al. (2007) and satellite altimeter data. VOLUME 42

13 JUNE 2012 P O U L A I N E T A L. 985 FIG. 10. Unbiased estimates of the surface geostrophic circulation in the Ionian Sea during two extended seasons, (a) winter (Nov Apr) and (b) summer (May Oct), in spatial bins of and for the period 14 Oct Dec These periods were selected by several authors to describe long-term changes in the basinwide circulation of the Ionian Sea, partially related to the EMT (Pujol and Larnicol 2005; Borzelli et al. 2009; Gačić et al. 2011). Before 1 July 1997 (Fig. 11a) the surface geostrophic circulation in the northern Ionian was essentially anticyclonic, with most of the water coming from the Sicily Channel in the AIS turning northward southeast of Sicily, proceeding in the north and northeast direction to reach the very northern Ionian (almost reaching the Otranto Channel). From there, the surface flow continues toward the south, in a meridional jet centered on E, as far south as 348N and then it splits into two branches, one eastward to exit in the Cretan Passage as a weak mean current and the other heading southward and westward in the southern Ionian (near 328N). A northwestward and northward flow completes the basinwide anticyclonic loop between Libya and Malta in the Sicily Channel. Between 1 July 1997 and 31 December 2005 (Fig. 11b) most of the surface geostrophic transport between the Sicily Channel and Cretan Passage is in the form of an eastward current crossing the central Ionian. Near 348N, E this current partially feeds into an anticyclonic circulation feature dominating the southern Ionian. The anticyclonic circulation to the north is much reduced, and there is even a sign of southwestward current along the Italian (Calabria) coast. The currents entering the Cretan Passage and the nearby west Cyprus cyclone and Pelops anticyclone appear to be stronger during this period. After 1 January 2006 (Fig. 11c) the Ionian surface geostrophic circulation is restored to patterns similar to the situation before July 1997, although the northern anticyclone is not as strong and the direct southwestward pathway is still strong in the central Ionian. The west Cyprus cyclone and Pelops anticyclone are still strong. 4. Discussion and conclusions Drifter and satellite altimeter data have been used to describe the surface geostrophic circulation of the entire Mediterranean Sea for the period Regression models have been used to remove the currents directly induced by the winds from the drifter velocities [Eq. (1)] and to combine the drifter and satellite altimeter data [Eq. (2)]. Pseudo-Eulerian velocity statistics (mean flow and kinetic energy levels) have been estimated from drifter data only, altimeter data only, and the combination of drifter and altimeter data. The latter statistics are less affected by the nonuniform drifter sampling and hence correspond to a better unbiased representation of the structure and variability of the Mediterranean surface currents. The difference between the mean circulation produced from drifter data only and drifter/altimeter data can be important in some areas (Fig. 7), such as in strong current areas, but the major circulation patterns are quite similar, at least qualitatively, throughout the Mediterranean. Once regression coefficients have been estimated, A and B in Eq. (2), mean circulation maps can be produced for any time period when altimeter data are available, independently of the availability of drifter data. Actually, Eq. (2) can be viewed as a correction of the altimeter data using the drifter measurements. In particular, the mean coastal currents, which usually are not accurately measured by satellite altimeters, are now well

14 986 J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y VOLUME 42 FIG. 11. Unbiased estimates of the surface geostrophic circulation in the Ionian Sea in spatial bins of and for the following periods: (a) October 1992 June 1997, (b) July 1997 December 2005, and (c) January 2006 December represented owing to the drifter correction (via the offset B). Values of jaj range mainly between 1 and 2, indicating that 1) drifter currents are generally stronger and contain contributions, such as residual wind-driven components, nonlinear boundary currents, and ageostrophic acceleration, not included in the geostrophic currents estimated from satellite altimetry, and 2) the currents calculated from satellite altimeter data are oversmoothed. Similar values of jaj have been found in the North Pacific Ocean (Niiler et al. 2003; Centurioni et al. 2008). The kinetic energies of the mean flow, derived from the drifters only and calculated from a combination of drifter/satellite data, are quite similar since 1) the mean error in Eq. (2) is zero by definition and 2) the biased and unbiased means of the satellite-derived velocities anomalies are comparable. In contrast, the fluctuating currents derived from the drifters, even though they are nonuniformly and scarcely sampled, are more energetic compared to the results obtained using the model of Eq. (2) (by a factor of 5 20). Since both datasets have been postprocessed in the same way (averaged in bins of week), this difference is due to 1) the fact that there might be residual wind-driven and nongeostrophic currents measured by the drifters; 2) the objective interpolation of the satellite data using a correlation radius of 100 km; and 3) the small number of drifter observations in some bins producing large error bars on the velocity statistics. It is interesting to note that the EKE levels of the geostrophic currents estimated from satellite altimeter data and the SMDT of Rio et al. (2007) are generally lower than our unbiased estimates. This is expected since the method of Rio et al. assumes a unity value for jaj. Following the above remarks on the methodologies used to study the surface geostrophic circulation, we now

15 JUNE 2012 P O U L A I N E T A L. 987 FIG. 12. Schematized representation of the mean surface geostrophic circulation in the Mediterranean Sea in based on the circulation maps depicted in Figs. 8a,b. discuss the salient results obtained with the methodology combining the drifter and satellite altimeter data and describe some aspects of the spatial structure and temporal variability of the Mediterranean surface geostrophic currents. For the areas where drifter and satellite data cannot be combined because of the poor correlation between them, the discussion is based on the drifterderived biased statistics. Note that, in this paper, notable vortices created by wind and topographic effects at given fixed locations are referred to as gyres, whereas the structures that appear to vary in location and maybe more variable in time, resulting mostly from the instability of strong coastal currents, are called eddies. These definitions might differ from those found in the literature (e.g., Robinson et al. 1991; Malanotte-Rizzoli et al. 1997; Millot and Taupier-Letage 2005; Pinardi et al. 2006). A schematic diagram of the Mediterranean surface geostrophic circulation based on the mean circulation maps depicted in Figs. 8a,b is presented as Fig. 12. In the western basin, our results agree well with the schema of Millot (1999) and Pinardi et al. (2006). It includes the two Alboran gyres, the Algerian Current and the anticyclonic eddies resulting from its instability southwest of Sardinia, the northwestward current along the Italian Peninsula in the Tyrrhenian Sea, and the northern Tyrrhenian gyre forced by the vorticity input of mistral winds funneling into the Strait of Bonifacio. To the north, the Northern Current dominates along the coasts of Italy, France, and Spain, and a basinwide cyclonic circulation pattern prevails in the combined Liguro Provencxal and Catalan subbasins. In the Sicily Channel, part of the Algerian Current enters the channel as a meandering Atlantic Ionian Stream from which a portion separates to feed a vein heading toward the southern Tunisian coast. Upon exiting the channel (east of Malta) the flow splits into 1) a branch moving northward and forming a basinwide anticyclonic circulation in the northern Ionian and 2) a southeastward current in the central Ionian. These branches eventually join north of Libya near 208E. The central current is actually the northern rim of another anticyclonic circulation located in the southern Ionian. To the south off the Libyan coast, there is a hint of coastal waters flowing eastward (especially in winter, see Fig. 11a). The northern Ionian anticyclone is in good agreement with the maps of Robinson et al. (1991) and Malanotte-Rizzoli et al. (1997), but contradicts the results of Pinardi et al. (2006) where a cyclonic circulation prevails. The southern Ionian anticyclone observed with the drifter and satellite altimetry are not compatible with the recent map of Hamad et al. (2005) based on satellite thermal imagery. Actually the North African coastal current that Hamad et al. advocate as a continuous current from the Sicily Channel to the southeastern Levantine subbasin is only confirmed by our results off the Nile Delta east of 278E. In the Adriatic and Aegean Seas the circulation is essentially characterized by a basinwide cyclonic pattern with an enhanced southeastward coastal current on the western side of the basin (Poulain 2001; Olson et al. 2007). In the Adriatic, three recirculation cyclonic cells are observed in agreement with the results of Poulain. In good agreement with most circulation maps available in the literature, the anticyclonic Pelops and Ierapetra gyres and the cyclonic west Cretan and Rhodes gyres are well delineated in the Cretan Passage and

16 988 J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y VOLUME 42 northern Levantine subbasins. Anticyclones are also observed in the southern and central Levantine, and some of them can be referred to as Mersa-Matruh eddies. They have been shown to be highly variable and to be generated by the instability of the African coastal current (Hamad et al. 2005, 2006; Millot and Taupier-Letage 2005; Gerin et al. 2009). The formation mechanism and evolution of these anticyclonic features is not studied in this paper, but their possible connection to the African coastal current is not confirmed as this current is not demonstrated by our combined drifter/altimeter data off the coasts of eastern Libya and western Egypt. Farther to the east, the mean circulation map shows clearly the basinwide cyclonic circulation in the Levantine subbasin, including a strong meridional coastal current off Israel, Lebanon, and Syria and the Asia Minor and Cilician Currents. Two more anticyclonic eddy features are apparent south and southeast of Cyprus, that is, the Cyprus and Shikmona eddies. More details on these circulation structures and their possible interaction and generation by the coastal current are discussed in Menna et al. (2012). An eastward zonal flow is also observed in the central Cretan Passage and Levantine subbasin, meandering between the cyclonic and anticyclonic gyres. It is often referred to as the Mid-Mediterranean Jet (Robinson et al. 1991; Robinson and Golnaraghi 1993). The combined drifter/altimeter proves the existence of this current and contradicts the recent findings of Hamad et al. (2005, 2006) and Millot and Taupier-Letage (2005). It is definitely not a data analysis artifact, as recently suggested by Millot and Gerin (2010). The mean kinetic energy of the surface geostrophic velocities dominates in the strong currents along most of the perimeter of the Mediterranean Sea: that is, the Algerian Current; the coastal current off eastern Egypt, Israel, Lebanon, and Syria; the Asia Minor and Cilician Currents, the western Adriatic and Aegean Currents; and the Northern Current. For the fluctuating velocities (Fig. 9) kinetic energy levels estimated using Eq. (2) (unbiased estimates) reach maximal values of cm 2 s 22 in the Algerian subbasin, off the eastern coast of Libya, and in the vicinity of the Ierapetra gyres and the Mersa-Matruh eddies. The geographical distribution of this EKE is similar to the mean kinetic energy levels estimated by Pujol and Larnicol (2005) with only satellite altimeter data. Given the values of jaj. 1in Eq. (2), it was expected that our unbiased estimates be larger than those directly computed from altimeter data. However, the spatial smoothing in bins lower the energy levels to values below or equal to the ones presented by Pujol and Larnicol. The EKE estimated directly from the filtered drifter velocities (biased map in Fig. 9a) reaches values near 1000 cm 2 s 22, especially in the western basin, comparable with the estimates of Mauerhan (2000) from a drifter dataset spanning the years using bins of size 80 km and 6-h temporal resolution. In the eastern basin the unbiased EKE levels are comparable to those of Gerin et al. (2009). Seasonal variations in the central Mediterranean surface geostrophic currents have been confirmed in the Atlantic Ionian Stream, the southern Ionian, and the western Cretan Passage. However, they are less pronounced than the estimates obtained directly from drifter data (Poulain 1998; Mauerhan 2000; Poulain and Zambianchi 2007; Gerin et al. 2009). In the northern Ionian, the seasonal variability advocated by Pinardi and Navarra (1993) as a wind-forced signal is not confirmed by our results. In contrast, the multiannual or decadal variations already divulged by Pujol and Larnicol (2005), Borzelli et al. (2009), Gačić et al. (2011), and others is well confirmed by our new results. As explained by these authors, this long-term variability is most likely related to the variations in formation of dense waters in the Adriatic and Aegean Seas, which appear typical on decadal scales and is known as the Eastern Mediterranean transient. This paper has demonstrated the occurrence of significant temporal changes in the Mediterranean Sea surface currents and the existence of important permanent, quasipermanent, and seasonal structures of the circulation. A quantitative statistical and kinematical description of these structures has been provided using drifter and satellite altimeter data. It is strongly recommended to concentrate future international efforts to continue the sampling of currents throughout the Mediterranean Sea using a combination of surface drifters, subsurface floats, and other instruments in order to be able to detect and study future circulation changes that can be important for the functioning of the Mediterranean system and to the communities living along its coasts. Acknowledgments. We thank all the people who deployed drifters in the Mediterranean and shared their data with us and the two anonymous reviewers for the constructive comments on the original manuscript. Discussions with Luca Centurioni and Marie-Hélène Rio helped to improve and clarify the paper. This work was partially supported by the U.S. Office of Naval Research under Contract N and by the Italian Space Agency as part of the ESA Endorsement GOCE Italy project. This paper is dedicated to the memory of Peter Niiler. REFERENCES Atlas, R., J. V. Ardizzone, R. Hoffman, J. C. Jusem, and S. M. Leidner, 2009: Cross-calibrated, multi-platform ocean surface