Benchmarking the mesoscale variability in global ocean eddy-permitting numerical systems
- 956 Downloads
The role of data assimilation procedures on representing ocean mesoscale variability is assessed by applying eddy statistics to a state-of-the-art global ocean reanalysis (C-GLORS), a free global ocean simulation (performed with the NEMO system) and an observation-based dataset (ARMOR3D) used as an independent benchmark. Numerical results are computed on a 1/4 ∘ horizontal grid (ORCA025) and share the same resolution with ARMOR3D dataset. This “eddy-permitting” resolution is sufficient to allow ocean eddies to form. Further to assessing the eddy statistics from three different datasets, a global three-dimensional eddy detection system is implemented in order to bypass the need of regional-dependent definition of thresholds, typical of commonly adopted eddy detection algorithms. It thus provides full three-dimensional eddy statistics segmenting vertical profiles from local rotational velocities. This criterion is crucial for discerning real eddies from transient surface noise that inevitably affects any two-dimensional algorithm. Data assimilation enhances and corrects mesoscale variability on a wide range of features that cannot be well reproduced otherwise. The free simulation fairly reproduces eddies emerging from western boundary currents and deep baroclinic instabilities, while underestimates shallower vortexes that populate the full basin. The ocean reanalysis recovers most of the missing turbulence, shown by satellite products , that is not generated by the model itself and consistently projects surface variability deep into the water column. The comparison with the statistically reconstructed vertical profiles from ARMOR3D show that ocean data assimilation is able to embed variability into the model dynamics, constraining eddies with in situ and altimetry observation and generating them consistently with local environment.
KeywordsMesoscale variability Eddy statistics Eddy detection system Global ocean Assimilation
Advances towards a complete understanding of the impact of oceanic eddies on the general circulation were made in the last two decades, fostered by the release of high-resolution altimetric maps together with dedicated and now fully operational satellite programs (Le Traon et al. 2015). Eddies are present almost everywhere in the world ocean, transporting heat, salt, chemical species, and organisms around the globe. They are therefore of great significance for resource distribution as well as for climate change (Zhang et al. 2014; Chi and Shang 2014; Saenko et al. 2011). The horizontal scale of an eddy ranges from dozens to hundreds of kilometers, with a lifetime spanning between few weeks and few months. The mesoscale eddies include coherent vortices, as well as a rich cascade of other structures such as filaments, squirts, and spirals. They are characterized by temperature and salinity anomalies with associated flow anomalies that are nearly in geostrophic balance. Although only the surface expression of mesoscale eddies is visible in satellite images of sea surface height or temperature, they are in fact three-dimensional structures that reach down into the pycnocline. Ocean mesoscale structures play an important role in mixing and energy exchange among different energy scales by transforming potential into kinetic energy.
Eddies act against level stratification, enhancing biological activity (Gaube et al. 2014; Mahadevan et al. 2012) or simply behaving as effective carriers. The impact on local ecosystems (McKiver et al. 1998) and their interplay with western boundary current (Straneo 2006) has a vast literature. However, important aspects of the mesoscale variability are still unknown. In numerical studies, mesoscale dynamics is still largely parametrized in eddy-permitting configurations. At these resolution, global ocean models can well reproduce large-scale circulation and currents, but do not explicitly resolve mesoscale turbulence. On the other hand, data assimilation procedures have been shown to correct the ocean state towards a more realistic description, improving global accuracy and recovering possible missing energies or transports caused by poor resolution (Masina et al. 2015; Storto et al. 2015).
In this study, in order to quantify the eddy variability generated by the assimilation scheme in an eddy-permitting model, we employ a global ocean simulation, referred as CTRL, at 1/4∘ resolution, performed with the NEMO system (Madec and Imbard 1996) and a global ocean reanalysis (C-GLORS version 4) (Storto et al. 2015). The two systems share exactly the same configuration, physical schemes and atmospheric forcing, in order to reduce disparities coming from sources different than the assimilation of ocean observations. A statistically reconstructed dataset (ARMOR3D) that provides three-dimensional fields based on sea level anomaly (SLA), sea-surface temperature (SST) and in situ observations (Guinehut et al. 2012; Mulet et al. 2012) is used as benchmark against numerical results. To our knowledge, ARMOR3D is the only observation-based three-dimensional reconstructed dataset currently available at eddy-permitting resolution.
The capability of data assimilation procedures to enhance variability, has been assessed by comparing kinetic energy statistics under different hypothesis, i.e., with or without assimilative corrections (Masina et al. 2015; Storto et al. 2015). This approach, although quite straightforward, cannot discern between realistic physical entities and transient noise occasionally generated by the systems. Inferring sub-surface features from a 2D method can lead to an overestimation of the variability that is difficult to inspect and that can only emerge when vertical analysis is added. In order to overcome this empirical approach, we implement a three-dimensional eddy detection system that seeks for realistic mesoscale eddies at global scale and is able to reduce the number of falsely detected eddies.
Detailed eddy characteristics are available from satellite altimetry (Chelton et al. 2011) but provide no information about depth. The classical algorithms for eddy detection are based on 2D approaches. A vast number of automatised 2D detection and tracking systems have been proposed in literature. They usually rely on some necessary conditions that are combined and parametrised to evolve in sufficient constraints for eddy identification. These algorithms exploit either physical properties or geometric characteristics of eddy. The most widely adopted detection method is based on the Okubo-Weiss (OW) criterion (Okubo 1970; Weiss 1991) that aims at separating the velocity field into regions of high vorticity and strain. This method identifies and segments eddies through the local balance between deformation and vortical flow that is expressed in terms of a physical parameter (W) computed from the horizontal velocity field. Eddies are ocean domains where W-parameter is less than a certain regional-dependent threshold, tuned ad-hoc in order to limit false-positive detection. Despite the implementation of this cutoff, OW has been proved to be rather affected by noise generated in W-fields that increases falsely detected eddies (Souza et al. 2011a; D’Ovidio et al. 2009). To overcome this issue, some novel approaches based on flow geometry characteristics or additional physical constraints were developed (e.g., Ari Sadarjoen and Post 2000; Turiel et al. 2007; Chaigneau et al. 2008; Nencioli et al. 2010; Ubelmann and Fu 2011; Beron-Vera et al. 2013; Halo et al. 2014). These methods include additional empirical parameters to identify the correct structures, complicating the eddy identification procedure. A new scheme has been proposed in Ashkezari et al. (2016) where a phase-angle system for eddy detection is combined with a machine learning algorithm for forecasting eddy lifetime, fed by 23 years of satellite data. A third approach for eddy identification is based on the analysis of sea level anomaly fields, in which a threshold is nevertheless always required to delimit eddy dimensions (e.g., Chaigneau and Pizarro 2005; Chelton et al. 2011). Eddies are identified as depression/elevation areas. Depression in SLA field leads a cyclonic structure, anti-cyclonic being an elevation. At regional scale, these algorithms are generally implemented with additional hypotheses (Mason et al. 2014), combined with filter on relative vorticity detection schemes (Morrow et al. 2004b; Chaigneau and Pizarro 2005), or employed together with OW parameter (Kang and Curchitser 2013; Yi et al. 2014; Viikmäe and Torsvik 2013). At global scale, a first milestone analysis of mesoscale structures through SLA fields came from Chelton et al. (2011), and more recently from Faghmous et al. (2015) and Xu et al. (2011). These methods try to identify eddies by exploiting 2D surface variables without any filter or check on the vertical. They impose constraints on eddy life-time to lessen positive biases in the number of structures detected, although an actual assessment of their accuracy is not simple to achieve without a constraint of vertical profiles.
Differently from observations, numerical simulations of the ocean easily provide full three-dimensional velocity and tracer fields that lend themselves to automated eddy census and tracking algorithms. A number of studies used ocean simulations to investigate eddy characteristics in selected areas, generally limited to regional domains on continental shelves (e.g., Doglioli et al. 2007; Colas et al. 2012; Liu et al. 2012). In a recent study, Petersen et al. (2013) developed a new algorithm to characterise eddies in particular properties involving depth that are somewhat sparse in observational studies, in a global ocean simulation with a 1/10 ∘ resolution and a duration of 7 years (Petersen et al. 2013). Their algorithm is based on the analysis of W-field, and eliminates the dependency on regional-threshold choice, introducing a global mathematical cutoff called “confidence threshold.” This parameter judges the likelihood of an eddy based on the similarity of certain functional fits with an idealised Gaussian vortex. In order to remove the positive bias that affects the W-field, an high value of “confidence” level is employed, although the main eddy features (e.g., the horizontal size and vertical thickness) seem tightly depending on this parameter at a global level. No hints about the improvements brought by 3D with respect to a realistic 2D method is shown, except for a comparison with the number of eddy detected by OW field and a global threshold value. Other works as Dong et al. (2014) apply 2D surface analysis scheme and consider the full water column down to the seabed, neglecting realistic vertical extension.
In the following, we develop a new complementary analysis based on two physical quantities: the SSH anomaly field for horizontal identification and the rotational velocity for vertical detection. One of the main benefits of such a system consists on the possibility to switch off the vertical filter thus quantifying the improvements coming from the additional vertical analysis, with respect to a standard SSH surface analysis. The horizontal eddy extension is firstly outlined from an depression/elevation area on SLA field. In order to retain the eddy, this area must embed an effective rotation in the surface currents, whose signature persists through the water column into the subsurface currents. The eddy bottom is found at the depth where its mean rotational velocity vanishes. This is a crucial feature that behaves as an actual discriminator among eddies and spurious patterns, providing at the same time the vertical thickness. We apply the eddy detection algorithm to quantify the impact of data assimilation in recovering the 3D ocean mesoscale variability not resolved by eddy-permitting simulations.
The paper is organized as follows. Section 2 provides a description of the eddy detection method together with the global numerical simulation and observation-based dataset. Section 3 compares the performances of 2D and 3D detection systems that are independently applied over one dataset (i.e., reanalysis). In particular, Section 3.1 describes the differences arising in the eddy layout for one single snapshot, stressing the effect of the vertical filter. The corresponding total eddy population and statistics are then discussed in Section 3.2 . Section 4 covers the dataset inter-comparison, where several eddy features are addressed. Section 5 is devoted to final remarks. We refer to Appendix for technical descriptions and further validations of the method.
2 Methods and datasets
2.1 The 3D eddy detection and tracking method
A picture of trajectories longer than 14 ∘ of longitude, detected with this method, are shown in Fig. 1a for the reanalysis dataset. Starting point of each trajectory is also highlighted. About 527 anti-cyclonic and 620 cyclonic eddies propagating along more than 14 ∘ of longitude are detected within 10-year time-window, consistently with the analysis performed by Chelton et al. (2011).
The control simulation (CTRL) of the global ocean based on version 3.4 of NEMO ocean model (Madec et al. 1998). A tripolar ORCA grid (Madec and Imbard 1996) is employed with an 1/4 ∘ horizontal resolution at the Equator, corresponding to 27 km, that increases with latitude to be 13.8 km at 60 ∘’N or 60 ∘’S. The model configuration has 50 vertical levels, with level thickness decreasing from 1m near the surface (22 levels in the first 100 m) to about 450 m at the ocean bottom. Bottom topography is represented with partial steps (Barnier et al. 2006). Physical and dynamical schemes follow the ones implemented in the reanalysis product used here and presented in Storto et al. (2015).
the reanalysis (C-GLORS version 4), which uses the same NEMO model set-up and grid configuration as CTRL. Data assimilation is carried out through a 3DVAR scheme (Storto et al. 2014; Storto et al. 2015) that ingests in situ temperature and salinity observation from EN4 data set of the Met Office Hadley Centre (Good et al. 2013) and along-track satellite altimetric data (AVISO 2012), together with surface nudging to sea surface temperature (SST) from the NOAA daily analyses (Reynolds et al. 2007) and sea ice concentration from Defense Meteorological Satellite Program (DMSP) SSM/I (Cavalieri et al. 1999). It is worth mentioning that SLA data are fully integrated in the 3DVAR scheme. They contribute to correct sub-surface temperature and salinity without directly changing the SSH. The overall effect of such a scheme on density profile is not limited to the mixed layer, but extents down to the thermocline or even below (Storto et al. 2010) where the vertical redistribution is driven by the data assimilation background-error covariances. The reanalysis product is available on-line at C-GLORS website (www.cmcc.it/c-glors) for the 1982–2013 period.
the ARMOR3D dataset, which provides full ocean state from surface to 5500 m of depth: 3D temperature, salinity, geopotential height, and geostrophic currents are defined on a regular 1/4 ∘ grid. These variables are generated by the statistical analysis of several different products: satellite data, in situ profiles and three-dimensional temperature, and salinity fields from ARIVO climatology (Gaillard and Charraudeau 2008) . At first, SLA and SST satellite observations are projected onto the vertical via a multiple linear regression method and using covariance deduced from historical observations (EN3 datasets from Ingleby and Huddleston 2007). The synthetic profiles produced in this way are then combined, through an optimal interpolation, with T,S in situ profiles (Guinehut et al. 2012). The latter gather Argo floats together with XBT, CTD and moorings measurements. Deep currents in ARMOR3D follow the scheme of thermal-wind approximation (Surcouf3D). They are evaluated starting from surface geostrophic velocities and adding baroclinic contributions coming from horizontal density gradient for each vertical layer (Guinehut et al. 2012). Gridded maps for satellite SST combine AVHRR, AMSR, and in situ observations distributed by the National Climatic Data Center at NOAA (Reynolds et al. 2007). Gridded SLA products comes from an optimal combination of all available satellite altimeters (AVISO 2012).
A 10-year time-window is spanned from 1st January 2003 to 31st December 2012. Frequency output of reanalysis and ARMOR3D datasets is 7-day means, while 5-day means for CTRL. Five meters correspond to five vertical levels for CTRL and reanalysis, while we consider the two upper levels for ARMOR3D fields corresponding to 10 m depth.
3 2D and 3D analysis
In this section, we investigate the benefits coming from the implementation of a 3D analysis with respect to a surface detection where the vertical filter is not considered. 3D and 2D procedures are independently applied to the same dataset, the ocean reanalysis. Performances for the two methods are compared for one single time-step in the first subsection. The overall statistics and decay rate are then evaluated over the entire dataset.
3.1 Single record analysis
In literature, the life-time filter is widely employed to screen out false-positive patterns that appear and disappear frame after frame. However, several noisy patterns still persist. Green contours in Fig. 2a represent those patterns that are able to pass the life-time filter thus affecting the final population. This is particularly true when a second tracking system is added between two non-consecutive timestep t and t + 2 to recover eddies that are eventually not detected at t + 1. This mechanism is also employed in order to increase the tracking efficiency (Chelton et al. 2011). On the other hand, it can lead to mismatches, for example, a “real” deep eddy can be erroneously matched with a transient surface anomaly arising in the same area two frame later (or two frame earlier). This inevitably extends the lifetime of the eddy at the cost of having an heterogeneous composition in the end.
Figure 2b–c show the number of eddies found in 1 ∘ latitudinal bands as function of their thickness, for a specific region in the North Atlantic highlighted in Fig. 2a. Figure 2b refers to eddies that are segmented by the 2D method with no check on their vertical extension. For the sake of clarity each eddy is assigned to the location of its mean center. Superficial eddies, less than 1 m deep, are almost evenly spread at any latitude. Those eddies also affect the abundance at depth; the distribution is slightly different with respect to results from 3D method plotted in Fig. 2c. Some irregular-shape eddies, extending about 5 m, are gathered by the 2D method at high latitudes; those eddies are not present in the 3D analysis.
3.2 Full dataset analysis
4 Inter-comparison of datasets
Coherent structures of mesoscale patterns progressively emerge at global scale once the eddy detection algorithm is applied. Eddies are hereby schematized as three-dimensional cylinder-like structures whose vertical thickness is defined by the persistence of the rotation inside the core. The 3D analysis is always applied hereafter.
4.1 Eddy population
Although the number of eddies cannot be considered a trustful measure of the mesoscale variability itself, it represents a meaningful proxy when comparing different datasets, giving at the same time indications on the efficiency of the tracking system. Roughly 2200 eddies are detected for each record in CTRL simulation (Fig. 4a). The number of detected eddies increases in the reanalysis and ARMOR3D datasets to approximately 3200 and 3500 eddies per record, respectively. CTRL underestimates the eddy numbers by 35%, showing a rather noisier time-series. Both reanalysis and ARMOR3D do not show any trend over time with variations of about 300 eddies. These stable behaviours can be seen as an hint about the robustness of our algorithm. It is worth to note that only the mean number of eddies detected per frame is usually provided in literature at global scale. Reanalysis single-record statistics are in agreement with Chelton et al. (2011), provided that they only implemented a surface detection. The percentage of undissipated eddies as function of time, i. e., the decay rate, is plotted in Fig. 4b. For CTRL, the population decreases down to 16% , from 105700 eddies to 17500 eddies, passing from eddies lasting more than 4 and 16 weeks, respectively. The percentage increases up to 20% for ARMOR3D and reanalysis where the overall numbers changes from 136000/30000 and 130000/29000, respectively. For reanalysis datasets, the number of eddies living more than 4, 16, or 28 weeks are respectively, 2700, 1400, or 700 per frame. ARMOR3D and reanalysis show similar behaviours, in agreement with Chelton et al. (2011) for eddies lasting less than 20 weeks.
4.2 Eddy spatial distribution
Our results show that, although model resolution puts severe constraints in the capability of producing such a mesoscale variability, the latter can be improved by applying a data assimilation procedure. Reanalysis map in Fig. 5b provides a much more spread distribution, vortexes are generated almost everywhere, mainly through SSH assimilation.
It should be noted that the sea level anomaly is not directly relaxed to observed SLA data. Altimetric data are here ingested in the 3D-var scheme and go to consistently modify subsurface temperature and salinity profiles. This is crucial for a correct projection of variability over sub-surface layers.
Looking only at the surface layer, the similar behaviour of reanalysis and ARMOR3D reflects a similar surface observational constraint, i.e., both methods process satellite altimetry and SST data although in very different ways. The main differences come from the analysis of ocean interior, where only sparse data are provided for temperature and salinity and no data of currents are assimilated in either dataset.
4.3 Eddy thickness
Only sparse observations are available about the effective distribution of eddy thickness. Drifters and floats typically can extract these values regionally, from the number of loops or their vorticity sign (e.g., Trani et al. 2011; Griffa et al. 2008; Shoosmith et al. 2005). However, it is difficult to quantify the horizontal and vertical extents of eddies with these sparse Lagrangian data. Argo profiles have been recently combined with altimetric data to improve vertical and horizontal analysis (Chaigneau et al. 2011) or heat transport (Souza et al. 2011b). In Castelao (2014), gridded SLA data are firstly analysed through an SSH surface eddy-detection system (Chelton et al. 2011). Then eddies are combined with Argo floats to investigate their vertical structure in the South Atlantic Bight . To the authors’ knowledge, the only former attempt to estimate eddy depth distribution with 3D detection systems at global level was performed by Petersen et al. (2013).
In ARMOR3D, the difference between Northern and Southern hemisphere is much reduced. Eddies become deeper than 1000 m soon after 20 ∘’N and 20 ∘’S of latitude. This large discrepancy with numerical results is likely to be ascribed to a different evaluation of deep currents. In ARMOR3D they are calculated starting from surface geostrophic velocities and adding baroclinic contributions coming from horizontal density gradients, i.e., Surcouf3D method in Mulet et al. (2012). However, as already noticed by the authors, the synthetic density field does not have enough vertical density gradient structure and therefore seems not able to compensate the altimetric signal (Mulet et al. 2012). On the other hand, in the reanalysis dataset, deep currents come from the integration of the model primitive equations, combining contributions from the density field produced by a data assimilation scheme as well as contributions from other processes (e.g., barotropic circulation, advection, viscosity). It is worth to take in consideration that data assimilation provides the ocean state closest to observations only in a variational sense. Where deep observations are not present, the SLA signal affects the deep density structure driven by the spatially varying temperature and salinity error covariance that is estimated by the model anomalies with respect to climatological fields. Thus, the eddy thickness emerging from this analysis can be considered as a lower bound, being constrained by the limitation of current eddy-permitting model.
4.4 Eddy amplitude distribution
Any discrepancy in eddy mean amplitude among datasets yields to a corresponding difference in the mean surface geostrophic motion. Strong ocean currents are fairly reproduced by the CTRL simulation. Eddies usually higher than 30 cm are associated to the Gulf Stream and the Kuroshio currents, the Agulhas current, and the Brazil confluence. Nevertheless, the zonal extensions are only correctly reproduced in the reanalysis. At surface, reanalysis and ARMOR3D are much more comparable in term of number of eddies, relative distribution, and amplitude over the entire ocean. A finer measure of variability is shown in the next sub-section, analyzing the amount of turbulent kinetic energy conveyed by eddies through their trajectories.
4.5 Relative eddy kinetic energy (REKE)
Differences are enhanced for full 3D results, shown in Fig. 9b,d,f, where surface-to-bottom vertical integration is considered (see Eq. 2), and eddies are schematized as three-dimensional structures in cylinder-like approximation. The physical origin of eddies can now be inferred by comparing surface and three-dimensional results: shallower eddies, living in the mixed layer, can be distinguished by shear-driven or baroclinic ones . This cannot be achieved from a surface analysis only. The CTRL simulation is capable of catching turbulence generated by instabilities in the western boundary currents and the ACC, that embed up to 60% of the total EKE in those regions. This percentage sharply decreases only a few hundreds of kilometer away. In the reanalysis dataset at mid latitudes, eddies can still give a significant contribution to EKE, ranging from about 40% of surface REKE to 20% of 3D REKE.
This sensible difference between surface and 3D-integrated REKE maps cannot be sketched in ARMOR3D, where the two maps have almost the same intensity, especially in the Southern ocean. While surface variables are correctly constrained by SLA and SST maps, deep currents are responsible for the eddy vertical extension.
4.6 REKE vertical profiles
4.7 Lagrangian trajectories
Figure 11b,d,f compares cyclonic, anti-cyclonic behavior of eddies for ARMOR3D, reanalysis, and CTRL simulation, respectively. No big discrepancy can be seen in the fraction of cyclonic vs. anti-cyclonic eddies. In the reanalyis, about 13900/12900 cyclonic/anti-cyclonic eddies lasting longer than 16 weeks are detected, while the number rises to 71300/ 69800 for 4 weeks eddies. About 8000 eddies propagate eastward (not shown), corresponding to 30% of all long-living eddies. Similar results are found for ARMOR3D. Numbers are smaller for CTRL, with 4700 eastward over 16000 total eddies. There is no clear preference for the eddy type: cyclonic and anticyclonic are equally occurring.
A new detection system is formulated for use in global configuration, bypassing the need of regional-dependent definition of threshold. This method provides full three-dimensional statistics of eddies by segmenting vertical profiles from eddy internal rotational velocity. This feature is crucial for selecting actual 3D eddies discerning it from transient surface noise. The latter can indeed affect any 2D detection method that infers real 3D eddies from surface analysis only. Our analysis shows that the corresponding discrepancy is estimated in about 10% of eddy population.
The new three-dimensional eddy detection system has been used to evaluate the role of data assimilation in representing ocean mesoscale variability. Data assimilation schemes have been broadly implemented in literature to guide simulations towards a more realistic ocean state. While their effects on energetic level have been extensively demonstrated in literature, a comprehensive study of their impacts on eddy statistics was not yet properly carried out to the authors’ knowledge. This is of primary importance to see the effect of assimilation procedures at different spatial scales.
We present a comparison of the eddy content among different datasets aimed to stress the fraction of mesoscale variability specifically coming from the assimilation procedure. We analyzed a state-of-the-art global reanalysis (C-GLORS), together with a twin-simulation (CTRL) with exactly the same physical parametrizations but where the assimilation was switched off. A third, independent comparison with an observation-derived dataset (ARMOR3D), is proposed. To our knowledge, this is the only 3D reconstructed dataset currently available at eddy-permitting resolution. ARMOR3D provides 3D ocean state from statistical analysis of satellite altimetric and SST maps together with in-situ data. Datasets are at 1/4 ∘ resolution and the time-range analyzed spans 10 years from 1st January 2003 to 31st December 2012.
At surface, reanalysis and ARMOR3D datasets show a wider spread horizontal distribution of eddies with respect to CTRL simulation. The CTRL simulation fairly reproduces eddies coming from shear stress of boundary currents and the ones generated from baroclinic instabilities. Eddies, living in the mixed layer, seldom occur. Data-assimilation is able to recover the most of surface variability missed by CTRL at the same eddy-permitting resolution, in particular, the eddy population increases by about 35% in the reanalysis. Lagrangian trajectories of eddies now pervade the full basin and surface EKE carried by eddies is quantitatively comparable with observation-derived results (ARMOR3D).
The analysis of 3D eddy vertical profiles shows very different behaviors between ARMOR3D and reanalysis. As already noticed by the authors of ARMOR3D, a residual correlation between the spatial variability at surface and at depth in the density field induces an overestimation of the mesoscale vertical penetration. Recently, (2017), a new release of ARMOR3D has been disseminated with an improved synthetic density field. In the reanalysis dataset, deep currents come from the integration of the model primitive equations, which include contributions from the density field produced by a full data assimilation scheme as well as other processes (e.g., advection, diffusion). These factors impact on the variability at depth, reducing the vertical extension of eddies. From our 3D analysis, eddies are responsible for more than 60% of total EKE in regions where deep baroclinic instabilities occur and where eddy cores are shown to be located well-below the ocean surface.
This work has received funding from the Italian Ministry of Education, University and Research and the Italian Ministry for the Environment, Land and Sea under the GEMINA project.
- AVISO (2012) SSALTO/DUACS User Handbook: (M)SLA and (M)ADT near-real time and delayed time products. SALP-MU-P-EA- 21065-CLS Edn. 2.9, 73 ppGoogle Scholar
- Barnier B, Madec G, Penduff T, Molines J-M, Treguier A-M, Sommer JL, Beckmann A, Biastoch A, Boning C, Dengg J, Derval C, Durand E, Gulev S, Remy E, Talandier C, Theetten S, Maltrud M, McClean J, Cuevas BD (2006) Impact of partial steps and momentum advection schemes in a global ocean circulation model at eddy-permitting resolution. Ocean Dyn. doi: 10.1007/s10236-006-0082-1
- Faghmous J H, Frenger I, Yao Y, Warmka R, Lindell A, Kumar V (2015) A daily global mesoscale ocean eddy dataset from satellite altimetry. Sci Data 2. doi: 10.1038/sdata.2015.28
- Forget G, Ponte RM (2015) The partition of regional sea level variability. Progress Oceanogr 137, Part A 173–195. doi: 10.1016/j.pocean.2015.06.002
- Gaillard F, Charraudeau R (2008) New climatology and statistics over the global Ocean, MERSEA-WP05-CNRS-STR-001-1AGoogle Scholar
- Halo I, Backeberg B, Penven P, Ansorge I, Reason C, Ullgren JE (2014) Eddy properties in the mozambique channel: a comparison between observations and two numerical ocean circulation models. Deep Sea Res Part II: Topical Stud Oceanogr 100:38–53. doi: 10.1016/j.dsr2.2013.10.015 CrossRefGoogle Scholar
- Le Traon P-YL, Antoine D, Bentamy A, Bonekamp H, Breivik LA, Chapron B, Corlett G, Dibarboure G, DiGiacomo P, Donlon C, Faugre Y, Font J, Girard-Ardhuin F, Gohin F, Johannessen JA, Kamachi M, Lagerloef G, Lambin J, Larnicol G, Borgne PL, Leuliette E, Lindstrom E, Martin MJ, Maturi E, Miller L, Mingsen L, Morrow R, Reul N, Rio MH, Roquet H, Santoleri R, Wilkin J (2015) Use of satellite observations for operational oceanography: recent achievements and future prospects. J Oper Oceanogr 8(sup1):s12–s27. doi: 10.1080/1755876X.2015.1022050 Google Scholar
- Madec G, Delecluse P, Imbard M, Levy (1998) OPA 8.1 ocean general circulation model reference manual, vol. Note du Ple de modlisation Inst. Pierre-Simon Laplace (IPSL) France. 91ppGoogle Scholar
- Masina S, Storto A, Ferry N, Valdivieso M, Haines K, Balmaseda M, Zuo H, Drevillon M, Parent L (2015) An ensemble of eddy-permitting global ocean reanalyses from the MyOcean project. Clim Dyn 1–29. doi: 10.1007/s00382-015-2728-5
- Nencioli F, Dong C, Dickey T, Washburn L, McWilliams JC (2010) A vector geometry–based eddy detection algorithm and its application to a high-resolution numerical model product and high-frequency radar surface velocities in the southern california bight. J Atmos Ocean Technol 27(3):564–579. doi: 10.1175/2009JTECHO725.1 CrossRefGoogle Scholar
- Nychka D, Furrer R, Sain S (2014) Fields: tools for spatial data, R package. http://CRAN.R-project.org/package=fields
- Pierce D (2014) ncdf: interface to Unidata netCDF data files. R package. http://CRAN.R-project.org/package=ncdf
- R Core Team (2014) R: a language and environment for statistical computing. http://www.R-project.org/
- Storto A, Masina S, Navarra A (2015) Evaluation of the CMCC eddy-permitting global ocean physical Reanalysis system (C-GLORS, 1982-2012) and its assimilation components. Quart J R Meteorol Soc, pp. n/a–n/a. doi: 10.1002/qj.2673
- Straneo F (2006) On the connection between dense water formation, overturning, and poleward heat transport in a convective basin. J Phys Oceanogr 1822–1840. doi: 10.1175/JPO2932.1
- VanDerWal J, Falconi L, Januchowski S, Shoo L, Storlie C (2014) SDMTools: species distribution modelling tools: tools for processing data associated with species distribution modelling exercises. R package. http://CRAN.R-project.org/package=SDMTools
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.