The eastern Mediterranean surface circulation is highly energetic and composed of structures interacting stochastically. However, some main features are still debated, and the behavior of some fine-scale dynamics and their role in shaping the general circulation is yet unknown. In the following paper, we use an unsupervised neural network clustering method to analyze the long-term variability of the different mesoscale structures. We decompose 26 years of altimetric data into clusters reflecting different circulation patterns of weak and strong flows with either strain or vortex-dominated velocities. The vortex-dominated cluster is more persistent in the western part of the basin, which is more active than the eastern part due to the strong flow along the coast, interacting with the extended bathymetry and engendering continuous instabilities. The cluster that reflects a weak flow dominated the middle of the basin, including the Mid-Mediterranean Jet (MMJ) pathway. However, the temporal analysis shows a frequent and intermittent occurrence of a strong flow in the middle of the basin, which could explain the previous contradictory assessment of MMJ existence using in-situ observations. Moreover, we prove that the Levantine Sea is becoming more and more energetic as the activity of the main mesoscale features is showing a positive trend.
The Levantine Sea surface circulation is controlled by a complex mesoscale system composed of eddies, jets, and filaments interacting stochastically with each other
Several studies have aimed to characterize the surface dynamics of this basin, such as
Additionally, there is a previous contradictory assessment of the presence of the Mid-Mediterranean Jet (MMJ, see Fig.
Currently, satellite altimetry is a widely used observational tool for analyzing sea surface physical dynamics and mesoscale activity and provides a continuous coverage for more than 25 years
In light of the major gaps in characterizing the surface currents of the Levantine Sea and previous contradictory assessments, the present paper aims to improve the understanding of its surface dynamics and mesoscale structures using machine learning techniques. To reach this aim, we adapt the neural network clustering method, the SOM method, and the hierarchical ascendant classification (HAC) method as used in
The paper is structured as follows. We start by presenting the data and the clustering method used to obtain the different clusters in Sect.
This section presents the data and methodology used for the restitution of the surface circulation regimes based on the concept of SOM and HAC methods. We then present the approach used to divide the basin into six geographical sub-regions (or boxes) that include the different mesoscale features.
The altimeter satellite gridded sea level anomaly (SLA) is estimated by optimal interpolation, merging the measurement from the different altimeter missions available. This product is processed by the DUACS (Data Unification and Altimeter Combination System) multi-mission altimeter data processing system. It processes data from all altimeter missions: Jason-3, Sentinel-3A, HY-2A, Saral/AltiKa, Cryosat-2, Jason-2, Jason-1, T/P, ENVISAT, GFO, and ERS1/2. To produce reprocessed maps in delayed time, the system uses the along-track altimeter missions from products called
The daily geostrophic surface velocity fields between 1993 and 2018, from the Herodotus Abyssal Plain and until the easternmost part of the Levantine sea, form the input layer of the SOM. In addition to the zonal and meridional components of the geostrophic velocities, the fluid parameter of Okubo–Weiss (OW) is included in the input layer (see Fig.
SOM is an unsupervised neural network method used for data visualization. It projects higher-dimensional data into lower-dimensional space by leveraging topological similarity properties. By this method, multidimensional data are clustered into neurons automatically associated into an orderly organization, where similar neurons are adjacent, and the less similar neurons are situated far from each other in the grid. This way allows for obtaining an insight into the topographic relationships of the initial data set
Panels
The SOM is structured in two layers: the input layer (in our case, a 3-D input layer composed of the zonal and meridional components and the Okubo–Weiss parameter) and the resulting neuron grid. Each neuron, representing a cluster with data presenting common characteristics, is associated with a referent vector obtained from a learning data set. Each vector of the input layer will be attributed to the neuron with the closest Euclidean distance to the referent vector. This referent vector is called the best matching unit (BMU), and its associated neuron is the “called” winning neuron. The determination of the referent vectors and the topological order of the SOM maps is done by minimizing the cost function
After the training phase, the SOM is well organized where there is a gradient of zonal and meridional velocities. This distribution of
The SOM allowed for classifying the velocity field into neurons that represent the different circulation patterns of the targeted grid based on
To analyze the activity of the most dominant features in the Levantine basin, we targeted the area from the Herodotus Abyssal Plain until the easternmost part of the Mediterranean Sea. The Mersa Matruh Eddy (MME) (see Fig.
The daily variation of each cluster frequency in each of the selected boxes between the beginning of 1993 and late 2018. This frequency variation reflects the percentage of pixels that were affected by each of the five clusters in a designated box.
The eddies usually reveal elevations (anticyclones) or depressions of the sea surface. Accordingly, and after decomposing the Levantine surface circulation (e.g. Fig.
In this section, we present the results of decomposing the surface circulation of the Levantine basin into a daily time series of five clusters obtained by the HAC and SOM methods.
The daily dominant cluster in each box from 1993 and 2018.
The frequency variation of the five clusters in each of the selected boxes, i.e., Bei, Shik, MME, AMC, Nile, and CE, is seen in Fig.
The seasonal variation of the C1 (high kinetic energy), C5 (high vorticity), and C3 (low kinetic energy and low vorticity) average in each box and their resulting linear regression.
Although clusters of strain-dominated flow (C1 and C2) were not frequent everywhere, C1 and C2 were frequently observed in MME and AMC, respectively, during the entire period. Such a high frequency occurred at the expense of other clusters, especially the cluster of weak flow C3, which was less observed in these two boxes. Regarding the vortex-dominated clusters (C4 and C5), C5 was the most frequent. C4 was quasi-absent in AMC and scarcely existed in the other boxes. When comparing between boxes, C5 was most frequent in the MME. Overall, all the cluster occurrences fluctuated greatly over time. To better analyze such a frequency variations, we presented the daily dominant cluster in each box in Fig.
These results showed that MME and AMC are two zones of a special regime of flow. This latter zone is represented by clusters of intense current, the so-called C1 and C2. The other boxes are zones of relatively weaker currents. In all the boxes, there were sporadic events of intense eddy activity, exhibited by the intermittent periods of C5 dominance.
The daily mean kinetic energy of the mean flow per unit of mass (MKE) computed in each box (see Fig.
The tendency of the dominant cluster could change with time, where previous results show, for example, that C5 was rare in Bei before being more frequently observed as a dominant cluster between 1993 and 1997.
Figure
These results show that the activity of the dominant mesoscale is increasing with time. Previous altimetric data observations from 1993–2003 revealed increasing variability of the Mediterranean Sea activity that is maximal in the Levantine Sea, especially in the Mersa Matruh area, where increasing energetic structures were observed
Percentage of C1, C2, C3, C4, and C5 occurrence between 1993 and 2018, superimposed onto the main bathymetric iso-lines of
It should be mentioned that C2 and C4 did not show a clear tendency. However, C2 was only significantly present in AMC with values around 40 %, while in all other boxes the frequency was less than 20 % (see Fig.
The spatial variation of clusters' frequencies is shown in Fig.
To further investigate the time evolution of the potentially existing MMJ, we present a Hovmöller diagram in Fig.
Both C4 and C5 are vortex-dominated clusters. However, the previous results showed that C4 is a peripheral cluster that is scarcely observed, dominating only very few pixels close to the coast. C5 was the main cluster that mainly reflected the eddies' presence. Here we present a more detailed analysis of the C5 evolution that reveals eddy activity in the Levantine Sea.
Figure
Because of the coarse resolution in both space and time of the altimeters and the fact that mesoscale structures move continuously, eddies can be missed or artificially created, smoothed, misplaced, or aliased into larger features compared to the true eddies
The average velocity field shows differences in CE after assimilation (see Fig.
In this study, we analyzed the surface circulation of the Levantine basin using the SOM
The variation of the daily average kinetic energy (MKE cm
The seasonal variation of the C2 and C4 average in each box and their resulting linear regression.
The SOM-HAC has been deposited into a public domain repository accessible at
The study was conceptualized by GB, REH, and LM. The methodology was developed by GB, REH, and LM. Any software used was developed by GB, REH, and GF. Validation was done by GB, REH, MF, JB, LI, GF, and LM. Formal analysis was conducted by GB, REH, and LM. Investigation was made by GB, REH, LI, JB, and LM. Resources were obtained by MF and LM. Data curation was done by GB, REH, and GF.
The contact author has declared that none of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank Milena Menna for providing the drifter data.
This research has been supported by Council for Scientific Research of Lebanon (CNRS-L). It was partially funded by the ALTILEV (in the framework of the PHC-CEDRE project) and O'LIFE programmes (Observatoire Libano-Français de l'Environnement).
This paper was edited by Aida Alvera-Azcárate and reviewed by two anonymous referees.