Segmentation and tracking of anticyclonic eddies during a submarine volcanic eruption using ocean colour imagery

Sensors 2015, 15, 8732-8748; doi:10.3390/s150408732 sensors ISSN 1424-8220 www.mdpi.com/journal/sensors Article Segmentation and Tracking of Anticyclonic Eddies during a Submarine Volcanic Eruption Using Ocean Colour Imagery Javier Marcello 1,*, Francisco Eugenio 1, Sheila Estrada-Allis 2 and Pablo Sangrà 1 1 Instituto de Oceanografía y Cambio Global, Universidad de Las Palmas de G.C., Campus Universitario de Tafira, 35017 Las Palmas de Gran Canaria, Spain; E-Mails: [email protected] (F.E.); [email protected] (P.S.) 2 Departamento de Física, Universidad de Las Palmas de Gran Canaria, 35017 Las Palmas de Gran Canaria, Spain; E-Mail: [email protected]lpgc.es * Author to whom correspondence should be addressed; E-Mail: javier[email protected]; Tel.: +34-928-457365; Fax: +34-928-451243. Academic Editor: Fabrizio Lamberti Received: 15 January 2015 / Accepted: 7 April 2015 / Published: 14 April 2015 Abstract: The eruptive phase of a submarine volcano located 2 km away from the southern coast of El Hierro Island started on October 2011. This extraordinary event provoked a dramatic perturbation of the water column. In order to understand and quantify the environmental impacts caused, a regular multidisciplinary monitoring was carried out using remote sensing sensors. In this context, we performed the systematic processing of every MODIS and MERIS and selected high resolution Worldview-2 imagery to provide information on the concentration of a number of biological, physical and chemical parameters. On the other hand, the eruption provided an exceptional source of tracer that allowed the study a variety of oceanographic structures. Specifically, the Canary Islands belong to a very active zone of long-lived eddies. Such structures are usually monitored using sea level anomaly fields. However these products have coarse spatial resolution and they are not suitable to perform submesoscale studies. Thanks to the volcanic tracer, detailed studies were undertaken with ocean colour imagery allowing, using the diffuse attenuation coefficient, to monitor the process of filamentation and axisymmetrization predicted by theoretical studies and numerical modelling. In our work, a novel 2-step segmentation methodology has been developed. The approach incorporates different segmentation algorithms and region growing techniques. In particular, the first step obtains an initial eddy segmentation using thresholding or clustering methods and, next, the fine OPEN ACCESS Sensors 2015, 15 8733 detail is achieved by the iterative identification of the points to grow and the subsequent application of watershed or thresholding strategies. The methodology has demonstrated an excellent performance and robustness and it has proven to properly capture the eddy and its filaments. Keywords: volcano; eddies; segmentation; clustering; thresholding; watershed; MODIS; MERIS; Worldview-2 1. Introduction Mesoscalar oceanic eddies are ubiquitous structures that can be found anywhere in the world’s oceans, with a predominance of anticyclonic over cyclonic vortexes [1]. Several studies [2–4] point out that these mesoscalar structures can exert an important role on the global climate and local oceanic dynamic due to its influence in the transport of ocean heat and momentum, as well as, in the enhancement of vertical shear mixing and both biotic and abiotic tracers. In the North East Atlantic Ocean, the presence of the Canary Archipelago acts as an obstacle to the oceanic and atmospheric flows, originating a rich mesoscale variability mainly represented by anticyclonic and cyclonic eddies spun off from the islands’ flanks [5,6]. The mechanisms of such variability are invoked through the perturbation of the Canary Current and Trades Winds by the tall topography islands such as El Hierro [7]. Eddy radius varies between the origin island radius (~12–25 km) at their initial stage and c.a. 75 km at their dissipative stage [5]. They are between 300 and 700 m depth depending on their intensity [8]. Anticyclonic eddies are initially more intense with a rotation rate of about 3 days whereas the initial rotation rate for cyclones is ca. 5 days [6]. Canary Island eddies are long-lived eddies (>4 months) being at the origin of the Canary Eddy Corridor which is the main pathway for long-lived eddies in the subtropical North Atlantic Ocean [9], therefore they have not only a local impact but also a remote impact. In this context, active and passive satellite sensors have been operating to investigate the variability of mesoscale features in the ocean. In particular, sea level anomaly fields from altimeters [10] have systematically been used to detect and monitor eddies. Some recent studies can be found in [11–15]. Unfortunately, these altimeter products have spatial resolution of tens of kilometres and, thus, they are suitable for global studies but are not adequate to perform detailed analysis at submesoscale level, to analyse their filaments or to study eddies with horizontal scales below 50 km, which is close to twofold the climatological first baroclinic Rossby radius of deformation for the Canary Islands region. This scale is the natural scale for mesoscale eddies at their initial stage and decreases with increasing latitude. However, these fine studies can be overcome from space by optical/infrared sensors [16–19]. In this way, satellite derived full resolution data of chlorophyll-a concentration or sea surface temperature can be used to thoroughly characterize eddies. Unfortunately, not all these structures can be clearly seen in ocean colour or thermal imagery, as this depends on which way they rotate, apart from the climatological inconvenience. Anti-cyclonic eddies in the northern hemisphere are particularly difficult to detect and analyse with these data. Sensors 2015, 15 8734 The eruption of a submarine volcano at El Hierro Island provided a unique and outstanding source of tracer that enabled us to study a variety of structures. El Hierro, an island off the Atlantic Ocean coast of North Africa was rocked by thousands of tremors and earthquakes since July 2011. On October 10 of the same year an underwater volcanic eruption started 300 meters below sea level. Figure 1 displays the location of the volcano and images during the first days after the eruption. It has to be pointed out that this natural tracer release has provided an exceptional opportunity to perform a detailed monitoring of an anticyclonic eddy. Low and high-resolution satellite images obtained from MODIS, MERIS and WorldView sensors were systematically processed [20] to provide information on the concentration of a number of marine parameters: chlorophyll-a concentration, suspended matter, PIC, POC, etc. These oceanographic remote sensing data also played a fundamental role during field campaigns guiding the Spanish Government oceanographic vessel to the appropriate sampling areas [21]. (a) (b) (c) Figure 1. Submarine volcano eruption at El Hierro Island during October 2011: (a) location of the underwater volcano; (b) photographs during the eruption; and (c) satellite image by NASA EO-1 ALI© sensor. Traditional eddy detection methods using altimeter data can be categorized into three classes [22]: (i) those based on a physical parameter, usually the Okubo–Weiss approach; (ii) those based on flow geometry, with the winding-angle approach being a representative method that identifies eddies by clustering closed or spiral streamlines; and (iii) those based on sea surface height or level anomaly in which a threshold is always required to delimit eddy dimensions. Several thresholds have been applied, typically ranging from 6 cm to 10 cm, depending on the region of study and size of the structures to consider. In our work, as we deal with higher resolution optical ocean colour data, the detection approaches are different, mainly addressing segmentation techniques. Segmentation is performed based on different attributes of an image such as size, colour, texture etc. Several image segmentation techniques have been developed for a variety of applications in various fields of image analysis [23–27], from feature-based approaches, such as clustering and histogram thresholding, as well as spatial domain-based methods, such as region growing-splitting-merging, energy-driven active contours, Sensors 2015, 15 8735 graph cuts, and watersheds. Our methodology for the automatic and precise detection of oceanographic features is based on an initial structure detection followed by a fine detail growing process. This paper illustrates the capabilities of multispectral satellite remote sensing systems to improve the understanding and monitoring of submarine volcanic processes and, mainly, it presents a novel image processing methodology for the segmentation of oceanographic structures. To that respect, automated detection methods are essential for understanding the complex movements and the dynamic characteristic of mesoscale eddies. The organization of contents is structured as follows: Section 2 includes de satellite imagery used and describes the two-steps segmentation algorithm. Section 3 presents the main results. Finally, Section 4 summarizes the main conclusions. 2. Methodology 2.1. Ocean Colour Imagery Satellite data from AQUA/TERRA-MODIS and ENVISAT-MERIS sensors have been the main source of remote sensing image information for monitoring El Hierro submarine volcanic processes. The NASA Moderate-resolution Imaging Spectroradiometer (MODIS) has a viewing swath width of 2330 km, it images the Earth in 36 spectral bands, ranging from 0.405 µm to 14.385 µm, and it collects data at three spatial resolutions: 250, 500, and 1000 meters. The ESA Medium Resolution Imaging Spectrometer (MERIS) has a swath width of 1,150 km, measuring the solar radiation reflected by the Earth in 15 spectral bands from 0.4125 µm to 0.9 µm and with a pixel field of view of 0.019°, generating full resolution data at 300 m resolution. We have implemented a methodology for providing multitemporal and multisensor imagery information, at 1000 m (MODIS) and 300 m (MERIS) resolution, on the concentration of specific oceanographic parameters, mainly chlorophyll-a (Chl-a) concentration and diffuse attenuation coefficient (Kd). The diffuse attenuation coefficient is an important water property related to light penetration in the blue-green wavelengths of the spectrum. It can provide valuable information about the transparency or the amount of dissolved and suspended materials in the water column. Satellite observations of this coefficient, at the wavelength 490 nm, are an effective method to provide turbidity maps for ocean and coastal waters at high spatial and temporal resolutions. Several empirical and semi-analytical models have been developed to estimate Kd; however, they are generally applicable to clear open ocean waters. Recently, improved models have appeared [28] mostly applicable to turbid coastal waters. In our case, satellite estimations of the diffuse attenuation coefficient and chlorophyll-a concentration under these abnormal conditions were assessed [20]. After comparing in-situ measurement with Chl-a products obtained using the MODIS and MERIS algorithms, we concluded that chlorophyll estimations were erroneous in such areas with extreme turbidity. On the other hand, estimations of Kd using common operational algorithms were not fully accurate due to the physical-chemical alterations provoked by the underwater volcano. For this reason, for MODIS and MERIS, we obtained and validated improved Kd models for coastal turbid waters [20]. As a consequence, only Kd products have been considered appropriate for the monitoring of the oceanographic structures during the event. Sensors 2015, 15 8736 Using the mentioned algorithms, we produced periodic K d maps during the timeframe of the volcanic eruption to allow the monitoring of the plume evolution and other generated structures. As an example, Figure 2 presents multitemporal and multisensor K d maps of the plume and the eddy progression during specific days of October and November 2011. Apart from the regular monitoring of the volcano activity with the provision of all cloud free oceanographic products to the governmental and research users, a database composed by K d MODIS and MERIS images of November 2011, as shown in Figure 2, displaying the evolution of the eddy is used to describe the segmentation methodology developed. Figure 2. Multitemporal K d product generated from MODIS (Terra and Aqua) and MERIS (ENVISAT) data during October and November 2011. On the other hand, the submarine volcano was also monitored using medium and high resolution data. In particular, the very high resolution Worldview-2 satellite was selected due to its excellent characteristics having eight multispectral bands with 2 m resolution. Thus, WorldView-2 spectral bands were atmospherically corrected and deglinted prior to obtain a very high-resolution K d model. This novel model was developed and validated using matchup data sets with MERIS, MODIS data, in-situ transmittances measurements and a seawater radiative transfer model [20]. This extraordinary resolution allows the analysis of the marine structures at the highest level of detail. Sensors 2015, 15 8737 Figure 3 displays a colour composite WorldView-2 image acquired on 26 October. The image was pre-processed to remove the atmospheric and sun-glint effects in order to reliably obtain the reflectance emitted from the sea surface. We can notice the brown patch in the volcano cone location and greener waters indicating the presence of lower concentrations of volcanic material. Thus, we can see extremely turbid waters with values above 1.5 m−1 in the centre of the eruption and gradually decreasing in the surroundings. The diffuse attenuation coefficient was estimated using our improved algorithm [20]. Unfortunately, due to the large horizontal extension of eddies, their continuous monitoring and tracking using this source of information is not suitable and extremely expensive. WorldView-2 was programmed to collect data during these months but, unfortunately, only the displayed image was appropriate (October 2011), while for the rest, the volcano activity had temporarily stopped or was very low in the sensed area. (a) (b) Figure 3. Worldview-2 image of October 26, 2011: (a) colour composite (bands 5-3-2) and (b) very high resolution diffuse attenuation coefficient map (clouds masked in black). 2.2. Methodology for the Precise Segmentation of Eddies Methods for detecting eddies are very important and can help to better understand the processes involved in eddy generation, evolution and decaying, as well as their impacts on marine environment and climate change. In addition, such eddy segmentation allows for the evaluation of the numerical modelling performance in terms of eddy centre, positions, sizes and motion. The precise detection approach we have developed is presented in Figure 4. The methodology is composed by two steps: (i) Initial eddy detection: First, a pre-processing is applied to the MODIS/MERIS Kd images. Next, the initial segmentation stage is applied with the possibility to select different clustering and thresholding algorithms. Finally, a post-processing is required to remove isolated noisy pixels or Sensors 2015, 15 8738 to fill holes in the eddy structure due to clouds. After the initial detection, the structure is well identified; however the fine detail of the emerging filaments cannot be preserved without incorporating unwanted areas. (ii) Eddy growth: The precise delimitation is of fundamental importance when analysing the structure and its evolution, so controlled growing techniques are applied only where filaments appear. Points to grow and their directions are found using the skeleton of the initial structure detected. At each point, a window with the corresponding orientation and size masks the image and new pixels are added to the structure after the application of thresholding or watershed region growing algorithms. This process is iteratively applied to achieve the desired fine segmentation. Figure 4. Methodology for the automatic segmentation of eddies using optical imagery. The different modules that integrate the methodology are next explained in detail. 2.2.1. Initial Eddy Detection Pre-Processing The proper segmentation of eddies is interfered by the presence of clouds, land, etc. So, a cloud mask (CM), as well as the land mask (LM) are applied to the geometrically corrected (GC) Kd image. Despite the good performance of the cloud detection algorithm [29], in terms of detection and false alarm probabilities, not all the existing cloud pixels are masked. Therefore they may appear as noise in the segmented image. In order to minimize its effect, an optional 3 × 3 median filtering (median) can be applied to the corrected and masked image. This filtering method has also been selected because of its smoothing effect without edge deterioration. Thus, the pre-processed image ( ) is obtained as follows:   (,)=󰇟󰇝(,)󰇞∩(,)∩(,)󰇠 (1) Initial Segmentation Several segmentation algorithms have been developed over the years, however oceanographic structures are difficult to be properly detected with conventional algorithms. After testing a variety of techniques and analysing their results with expert oceanographers, the following methods for the Sensors 2015, 15 8739 preliminary eddy segmentation have been selected: K-means Clustering, Fuzzy K-means Clustering, Mean Shift Clustering and Thresholding. In [30,31] a detailed evaluation of 36 automatic global and local thresholding techniques was performed and the best methods belonging to different categories (histogram shape-based, clustering-based, entropy-based, object attribute-based, spatial methods, etc.) were incorporated to the eddy segmentation tool. In our study, satisfactory results to segment marine structures are achieved by techniques like Riddler, Otsu, Gaussian Mixture Modelling, Pun or Li. In particular, Otsu algorithm provides an intermediate, yet accurate, solution and it can be selected as a compromise technique, providing robustness and excellent results for the majority of images. After the application of the segmentation algorithm, we get a binary image corresponding to the Initial Segmented Eddy (ISE): (,) =  (,) (2) Some examples of possible automatic segmentations are included in Figure 5. As it can be appreciated, oceanographic structures are not simple to automatically detect due to the fact that gradients are weak, generating several unconnected border lines and making of great difficulty the structure identification based on edge techniques; the presence of noise, mainly due to atmospheric phenomena; the low signal received at sensor level; and the strong morphological variation that impedes an accurate geometric representation and the absence of a valid analytical model for the structures. (a) (b) (c) (d) Figure 5. Structures detected by different segmentation algorithms: (a) K d pre-processed imagery; (b) K-means; (c) Fuzzy K-Means; and (d) Otsu. Trying a segmentation using manual thresholding techniques provides the results included in Figure 6a–c. It can be appreciated that even using a manual procedure and adjusting the K d threshold to different values (0.04 to 0.06) it is not possible to achieve at the same time the fine detail of the complete filaments without adding undesired lower levels of K d . Post-Processing The segmented image includes the eddy plus additional artefacts due to the presence of noise, isolated turbid water zones and the remaining cloud pixels not properly masked previously. Thus, a post-processing step is required consisting mainly on the application of morphological operators [32]. Sensors 2015, 15 8740 A closing operator (∅), using a structuring element (typically of size 3 pixels) to smooth and connect pixels, mainly belonging to broken filaments; followed by a removal of holes (RH) operation usually to fill cloudy pixels belonging to the eddy. The resulting initial segmented eddy ( ) is obtained as follows:   (,)=󰇟∅󰇝(,)󰇞󰇠 (3) An additional capability included in the methodology is the elimination of isolated objects whose size is smaller than a threshold defined by the user. (a) (b) (c) (d) Figure 6. Structures detected by manual thresholding algorithms with global K d threshold values of: (a–c) 0.06, 0.05 and 0.04, respectively; (d) Segmentation desired by the oceanographers. 2.2.2. Eddy Growing After the preceding tasks, the structure is identified and, selecting the corresponding parameters, it is possible to achieve different levels of segmentations to get the desired eddy detection. Unfortunately, none of the tested methods is able to produce the results required by our oceanographers, as presented in Figure 6d, who want to preserve the core of the eddy but also to perfectly delimitate the emerging filaments but without including noise associated to water with less turbidity surrounding its core. The precise delimitation of the filaments provides fundamental information about the submesoscale variability such as the filamentation of the eddy not observed before with such. In addition, such precise delimitation is essential to perform a comparison of a model outcome with observational data from remote sensing imagery. To achieve this fine segmentation, a two-step procedure has been developed, as shown in Figure 4. It allows the application of controlled growing techniques but only in regions where filaments appear. The procedure is as follows: Identification of Points to Grow and Direction First, the skeleton (SK) of the segmented eddy is used to identify the points to grow and it provides, as well, the growing direction (see Figure 7a in red colour). In this work, the skeleton is obtained by Sensors 2015, 15 8747 6. Sangrà, P.; Auladell, M.; Marrero-Díaz, A.; Pelegrí, J. L.; Fraile-Nuez, E.; Rodríguez-Santana, A.; Martín, J.M.; Mason, E.; Hernández-Guerra, A. On the nature of oceanic eddies shed by the Island of Gran Canaria. Deep-Sea Res. I 2007, 54, 687–709. 7. Jiménez, B.; Sangrà, P.; Mason, E. A numerical study of the relative importance of wind and topographic forcing on oceanic eddies shedding by deep water islands. Ocean Model. 2008, 22, 146–157. 8. Lasternas, S.; Piedeleu, M.; Sangrà, P.; Duarte, C.M.; Agustí, S. Forcing of dissolved organic carbon release by phytoplankton by anticyclonic mesoscale eddies in the subtropical NE Atlantic Ocean. Biogeosciences 2013, 10, 2129–2143. 9. Sangrà, P.; Pascual, A.; Rodríguez-Santana, A.; Machín, F.; Mason, E.; McWilliams, J.C.; Pelegrí, J.L.; Dong, C.; Rubio, A.; Arístegui, J.; et al. The Canary Eddy Corridor: A major pathway for long-lived eddies in the subtropical North Atlantic. Deep-Sea Res. I 2009, 56, 2100–2114. 10. Quartly, G.; Chen, G. Introduction to the Special Issue on “Satellite Altimetry: New Sensors and New Applications”. Sensors 2006, 6, 616–619. 11. Liu, Y.; Dong, C.M.; Guan, Y.; Chen, D.; McWilliams, J.; Nencioli, F. Eddy analysis in the subtropical zonal band of the North Pacific Ocean. Deep-Sea Res. I 2012, 68, 54–67. 12. Xiu, P.; Chai, F.; Xue, H.; Shi, L.; Chao, Y. Modeling the mesoscale eddy field in the Gulf of Alaska. Deep-Sea Res. I 2012, 63, 102–117. 13. Ruiz, S.; Pelegri, J.L.; Emelianov, M.; Pascual A.; Mason, E. Geostrophic and ageostrophic circulation of a shallow anticyclonic eddy off Cape Bojador. J. Geophys. Res.-Oceans 2014, 119, 1257–1270. 14. Mason, E.; Pascual, A.; McWilliams, J.C. A New Sea Surface Height-Based Code for Oceanic Mesoscale Eddy Tracking. J. Atmos. Ocean. Technol. 2014, 31, 1181–1188. 15. Lohrenz, S.E.; Cai, W.-J.; Chen, X.; Tuel, M. Satellite Assessment of Bio-Optical Properties of Northern Gulf of Mexico Coastal Waters Following Hurricanes Katrina and Rita. Sensors 2008, 8, 4135–4150. 16. Shapiro, G.I.; Stanichny, S.V.; Stanychna, R.R. Anatomy of shelf-deep sea exchanges by a mesoscale eddy in the North West Black Sea as derived from remotely sensed data. Remote Sens. Environ. 2010, 114, 867–875. 17. Watanabe, E. Beaufort shelf break eddies and shelf-basin exchange of Pacific summer water in the western Arctic Ocean detected by satellite and modeling analyses. J. Geophys. Res.-Oceans 2011, 116, C08034. 18. Alpers, W.; Brandt, P.; Lazar, A.; Dagorne, D.; Sow, B.; Faye, S.; Hansen, M.W.; Rubino, A.; Poulain, P.M.; Brehmer, P. A small-scale oceanic eddy off the coast of West Africa studied by multi-sensor satellite and surface drifter data. Remote Sens. Environ. 2013, 129, 132–143. 19. Dong, C.; Nencioli, F.; Liu, Y.; McWilliams, J.C. An Automated Approach to Detect Oceanic Eddies from Satellite Remote Sensed Sea Surface Temperature Data. IEEE Geosci. Remote Sens. Lett. 2011, 8, 1055–1059. 20. Eugenio, F.; Martin, J.; Marcello, J.; Fraile-Nuez, E. Environmental monitoring of El Hierro Island submarine volcano, by combining low and high resolution satellite imagery. Int. J. Appl. Earth Obs. Geoinf. 2014, 29, 53–66. Sensors 2015, 15 8748 21. Fraile-Nuez, E.; González-Dávila, M.; Santana-Casiano, J.M.; Arístegui, J.; Alonso-González, I.J.; Hernández-León, S.; Blanco, M.J.; Rodríguez-Santana, A.; Hernández-Guerra, A.; Gelado-Caballero, M.D.; et al. The Submarine Volcano Eruption at the Island of El Hierro: Physical-chemical Perturbation and Biological Response. Sci. Rep. 2012, doi:10.1038/srep00486. 22. Yi, J.; Du, Y.; He, Z.; Zhou, C. Enhancing the accuracy of automatic eddy detection and the capability of recognizing the multi-core structures from maps of sea level anomaly. Ocean Sci. 2014, 10, 39–48. 23. Vantaram, S.R.; Saber, E. Survey of contemporary trends in color image segmentation. J. Electron. Imaging 2012, 21, 9011–90128. 24. Thakur, P.; Madaan, N. A survey of Image Segmentation Techniques. Int. J. Res. Comput. Appl. Robot. 2014, 2, 158–165. 25. Agarwal, M.; Singh, V. A Methodological Survey and Proposed Algorithm on Image Segmentation using Genetic Algorithm. Int. J. Comput. Appl. 2013, 67, 7–17. 26. Dey, V.; Zhang, Y.; Zhong, M. A review of image segmentation techniques with remote sensing perspective. In Proceedings of the ISPRS TC VII Symposium–100 Years ISPRS, Vienna, Austria, 5–7 July 2010. 27. Narkhede, H.P. Review of Image Segmentation Techniques. Int. J. Sci. Mod. Eng. 2013, 1, 54–61. 28. Zhang, Y.; Liu, X.; Yin, Y.; Wang, M.; Qin, B. A simple optical model to estimate diffuse attenuation coefficient of photosynthetically active radiation in an extremely turbid lake from surface reflectance. Opt. Express 2012, 20, 18. 29. Ackermann, S.; Frey, R.; Strabala, K.; Liu, Y.; ; Gumley, L.; Baum, B.; Menzel, P.H. Discriminating clear-sky form cloud with MODIS. In NASA_MODIS Algorithm Theoretical Basis Document (MOD35); University of Wisconsin: Madison, WI, USA, 2010; Version 6.1. 30. Marcello, J.; Marqués, F.; Eugenio, F. Automatic tool for the precise detection of upwelling and filaments in remote sensing imagery. IEEE Trans. Geosci. Remote Sens. 2005, 43, 1605–1616. 31. Marcello, J.; Eugenio, F.; Marqués, F. Evaluation of thresholding techniques applied to oceanographic remote sensing imagery. In Proceedings of the SPIE-Remote Sensing Europe, Maspalomas, Spain, 10 November 2004. 32. Soille, P. Morphological Image Analysis: Principles and Applications; Springer-Verlag: Berlin, Germany, 2003. 33. AVISO. Available online: http://www.aviso.altimetry.fr/en/home.html (accessed on 3 March 2014). 34. Dong, C.; McWilliams, J.C.; Shchepetkin, A. Island wakes in deep water. J. Phys. Oceanogr. 2007, 37, 962–998. 35. Koumoutsakos, P. Inviscid Axisymmetrization of an Elliptical Vortex. J. Comput. Phys. 1997, 138, 821–857. © 2015 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/4.0/).