Mesoscale Eddy Detection Research Articles

A Metadata-Enhanced Deep Learning Method for Sea Surface Height and Mesoscale Eddy Prediction

Predicting the mesoscale eddies in the ocean is crucial for advancing our understanding of the ocean and climate systems. Establishing spatio-temporal correlation among input data is a significant challenge in mesoscale eddy prediction tasks, especially for deep learning techniques. In this paper, we first present a deep learning solution based on a video prediction model to capture the spatio-temporal correlation and predict future sea surface height data accurately. To enhance the performance of the model, we introduced a novel metadata embedding module that utilizes neural networks to fuse remote sensing metadata with input data, resulting in increased accuracy. To the best of our knowledge, our model outperforms the state-of-the-art method for predicting sea level anomalies. Consequently, a mesoscale eddy detection algorithm will be applied to the predicted sea surface height data to generate mesoscale eddies in future. The proposed solution achieves competitive results, indicating that the prediction error for the eddy center position is 5.6 km for a 3-day prediction and 13.6 km for a 7-day prediction.

Mesoscale Eddy Detection and Classification From Sea Surface Temperature Maps With Deep Neural Networks

Mesoscale Eddy Detection and Classification From Sea Surface Temperature Maps With Deep Neural Networks

Deep Learning for Mesoscale Eddy Detection With Feature Fusion of Multisatellite Observations

Deep Learning for Mesoscale Eddy Detection With Feature Fusion of Multisatellite Observations

Evaluating the Detection of Oceanic Mesoscale Eddies in an Operational Eddy-Resolving Global Forecasting System

In this study, a global analysis and forecasting system at 1/12° is built for operational oceanography at the National Marine Environmental Forecasting Center (NMEFC) by using the NEMO ocean model (NMEFC-NEMO). First, statistical analysis methods are designed to evaluate the performance of sea level anomaly (SLA) forecasting. The results indicate that the NMEFC-NEMO performs well in SLA forecasting when compared with the Mercator-PSY4, Mercator-PSY3, UK-FOAM, CONCEPTS-GIOPS and Bluelink-OceanMAPS forecasting systems. The respective root-mean-squared errors (RMSEs) of NMEFC-NEMO (Mercator PSY4) are 0.0654 m (0.0663 m) and 0.0797 m (0.0767 m) for the lead times of 1 and 7 days. The anomaly correlation coefficients between forecasting and observations exceed 0.8 for the NMEFC-NEMO and Mercator-PSY4 systems, suggesting that the accuracy of SLA predicted using NMEFC-NEMO is comparable to Mercator PSY4 and superior to other forecasting systems. Moreover, the global spatial distribution of oceanic eddies are effectively represented in NMEFC-NEMO when compared to that in the HYCOM reanalysis, and the detection rate reaches more than 90% relative to HYCOM reanalysis. Regarding the strong eddies in the Kuroshio region, the NMEFC-NEMO reproduces the characteristic for anticyclonic and cyclonic eddies merging and splitting alternatively. As for the detective eddies in the Gulf Stream, NMEFC-NEMO effectively represents the spatial distribution of mesoscale eddies from different seasons. The amplitude of oceanic eddies, including both cyclones and anticyclones, were much stronger on 1 July 2019 than 1 January 2019. Overall, NMEFC-NEMO has a superior performance in SLA forecasting and effectively represents the oceanic mesoscale eddies for operational oceanography.

SymmetricNet: end-to-end mesoscale eddy detection with multi-modal data fusion

Mesoscale eddies play a significant role in marine energy and matter transportation. Due to their huge impact on the ocean, mesoscale eddy detection has been studied for many years. However, existing methods mainly use single-modal data, such as the sea surface height (SSH), to detect mesoscale eddies, resulting in inaccurate detection results. In this paper, we propose an end-to-end mesoscale eddy detection method based upon multi-modal data fusion. Particularly, we don’t only use SSH, but also add data of other two modals, i.e., the sea surface temperature (SST) and the velocity of flow, which are closely related to mesoscale eddy detection. Moreover, we design a novel network named SymmetricNet, which is able to achieve multi-modal data fusion in mesoscale eddy detection. The proposed SymmetricNet mainly contains a downsampling pathway and an upsampling pathway, where the low-level feature maps from the downsampling pathway and the high-level feature maps from the upsampling pathway are merged through lateral connections. In addition, we apply dilated convolutions to the network structure to increase the receptive field without sacrificing resolution. Experiments on multi-modal mesoscale eddy dataset demonstrate the advantages of the proposed method over previous approaches for mesoscale eddy detection.

Comparison of two deep learning model YOLOF and Detectron2 for mesoscale eddies identification in the South China Sea

ABSTRACT Ocean mesoscale eddy is a special mesoscale phenomenon in the ocean, which widely exists in all oceans and marginal seas around the world. Compared with ordinary ocean circulation, mesoscale eddy has high rotating speed, strong current velocity, and average vertical depths of several kilometres, which play an important role in ocean circulation and material transport. The traditional mesoscale eddy identification and recognition have great subjectivity and usually depends on the parameters predefined or adjusted by experts, which cannot guarantee the accuracy. With the development of artificial intelligence, deep learning method has also been widely used in the detection and recognition of ocean eddies. Based on the sea-level anomaly (SLA) data provided by Copernicus Marine Environment Detection Service, this study compares the results of the ‘you only look once level feature’ (YOLOF) model and Detectron2 model in deep learning on marine mesoscale eddy detection. The mesoscale eddies in the South China Sea from 2011 to 2013 are detected and identified. The accuracy of Detectron2 model is about 57%, while the accuracy of YOLOF model can reach 90%. Comparing the performance of the two models on the same data set, it can be found that YOLOF model improves the accuracy and speed of recognition to a certain extent. The deep learning network model provides an effective technical method for the study of mesoscale eddy detection by sea surface height.

Detection and Identification of Mesoscale Eddies in the South China Sea Based on an Artificial Neural Network Model—YOLOF and Remotely Sensed Data

Mesoscale eddies are typical mesoscale ocean phenomena that exist widely in all oceans and marginal seas around the world, playing important roles in ocean circulation and material transport. They also have important impacts on the safe navigation of ships and underwater acoustic communications. Traditional mesoscale eddy identification methods are subjective and usually depend on parameters that must be pre-defined or adjusted by experts, meaning that their accuracy cannot be guaranteed. With the rise of deep learning, the “you only look once” (YOLO) series target recognition models have been shown to present certain advantages in eddy detection and recognition. Based on sea level anomaly (SLA) data provided over the past 30 years by the Copernicus Marine Environment Monitoring Service (CMEMS), as well as deep transfer learning, we propose a method for oceanic mesoscale eddy detection and identification based on the “you only look once level feature” (YOLOF) model. Using the proposed model, the mesoscale eddies in the South China Sea from 1993 to 2021 were detected and identified. Compared with traditional recognition methods, the proposed model had a better recognition effect (with an accuracy of 91%) and avoided the bias associated with subjectively set thresholds; to a certain extent, the model also improved the detection of and the identification speed for mesoscale eddies. The method proposed in this paper not only promotes the development of deep learning in the field of oceanic mesoscale eddy detection and identification, but also provides an effective technical method for the study of mesoscale eddy detection using sea surface height data.

Oceanic mesoscale eddy detection and convolutional neural network complexity

Deep learning has drawn the attention of oceanographic researchers over the past few years, making the research community adopt computer vision techniques for oceanic mesoscale eddy detection on satellite altimetry gridded products. In this paper, we describe a convolutional neural network designed to detect eddies in satellite altimetry maps after being trained using segmentation masks provided by the OpenEddy detection algorithm. Against the current trend, in which increasingly complex neural networks are being proposed to address this problem, our design is relatively simple and yet provides competitive performance when compared to any of the previous deep learning methods reported in the literature. Furthermore, we show that our model is less sensitive to timely variations than the traditional models based on physical and geometric features defined by human experts, making it possible for our model to use the general data context to identify eddies that those traditional models would have missed. These results prove that overly complex neural network architectural designs are not required to solve the eddy detection problem on altimetry maps and generate a sufficiently good model for most practical applications in the field of marine sciences. • Increasingly complex CNN architectures are being proposed for eddy detection. • We prove that excessively complex CNN designs are not needed to solve this problem. • Our proposal improves state-of-the-art designs and requires to train much less parameters. • Our data augmentation method improves performance without artificial transformations. • Our model is less sensitive to timely variations and detects eddies that traditional models would miss.

Evaluation of HY-2 Series Satellites Mapping Capability on Mesoscale Eddies

With the launches of HY-2B, HY-2C and HY-2D, the National Satellite Ocean Application Service (NSOAS) successfully established the first batch of HY-2 constellation on 19 May 2021, dedicated to detecting the marine dynamic environment. While the validation and calibration of HY-2 series satellites have been well studied, using both in situ and cross-track measurements, further evaluation is required of HY-2 series satellites mapping capability on mesoscale eddies. In this study, we conducted two groups of pseudo-observations obtained by ocean models output at positions and times of the satellite along ground tracks. The two altimetry satellite combinations are the Jason-3+Sentinel-3A/B configuration (reference group) and the Jason-3+Sentinel-3A/B+HY-2B/C/D configuration (evaluation group). Three regions at different latitudes were selected as the study areas. Then, we gridded the satellite along ground tracks into synthetic SSH products for each satellite configuration group. Finally, the attributes of mesoscale eddies extracted from gridded products were analyzed to evaluate the mapping capability of HY-2 series satellites. Our results suggest that the gridded products of evaluation groups are more similar to model data compared to reference groups, with a maximum difference reduced by 23%. In terms of eddy identification, the evaluation groups using HY-2 series satellites can alleviate eddy incorrect estimation and add more details to the spatial distribution of eddy radius and amplitude. Additionally, the evaluation groups displayed better performance in mesoscale eddy detection, with a maximum enhancement of 14% and 16% in eddy recognition rate and accuracy.

The Attenuation Effect of Jet Filament on the Eastward Mesoscale Eddy Lifetime in the Southern Ocean

Abstract The interaction between the Antarctic Circumpolar Current (ACC) jets and mesoscale eddies plays a prominent role in the dynamics of the ACC. However, few studies have investigated the influence of the jets on the statistics of the mesoscale turbulence field. This study combines a mesoscale eddy detection algorithm with the jet-detection technique to extract their interaction in the Southern Ocean from 1993 to 2016. It is found that stronger jet filaments can effectively shorten eddy lifetime, thus introducing a negative correlation between eddy lifetime and matched jet filament velocities. It can also be seen along the pathway of the ACC whether the eddy is located upstream or downstream of topography. A series of numerical experiments suggest that the stronger zonal jet plays the role of a waveguide to facilitate the stronger and faster eastward linear Rossby wave field induced by the mesoscale eddy. A stronger eastward Rossby wave phase speed, enabled by the jet-induced Doppler shift, results in a more rapid loss of eddy energy and shortens the eddy lifetime.

The Identification and Prediction of Mesoscale Eddy Variation via Memory in Memory With Scheduled Sampling for Sea Level Anomaly

Ocean mesoscale eddies are ubiquitous in world ocean and account for 90% oceanic kinetic energy, which dominate the upper ocean flow field. Accurately predicting the variation of ocean mesoscale eddies is the key to understand the oceanic flow field and circulation system. In this article, we propose to make an initial attempt to explore spatio-temporal predictability of mesoscale eddies, employing deep learning architecture, which primarily establishes Memory In Memory (MIM) for sea level anomaly (SLA) prediction, combined with the existing mesoscale eddy detection. Oriented to the western Pacific ocean (125°−137.5°E and 15°−27.5°N), we quantitatively investigate the historic daily SLA variability at a 0.25° spatial resolution from 2000 to 2018, derived by satellite altimetry. We develop the enhanced MIM prediction strategies, equipped with Gated Recurrent Unit (GRU) and spatial attention module, in a scheduled sampling manner, which overcomes the gradient vanishing and complements to strengthen spatio-temporal features for long-term dependencies. At the early stage, the real value SLA input guides the model training process for initialization, while the scheduled sampling intentionally feeds the newly predicted value, to resolve the distribution inconsistency of inference. It has been demonstrated in our experiment results that our proposed prediction scheme outperformed the state-of-art approaches for SLA time series, with MAPE, RMSE of the 14-day prediction duration, respectively, 5.1%, 0.023 m on average, even up to 4.6%, 0.018 m for the effective sub-regions, compared to 19.8%, 0.086 m in ConvLSTM and 8.3%, 0.040 m in original MIM, which greatly facilitated the mesoscale eddy prediction. This proposed scheme will be beneficial to understand of the underlying dynamical mechanism behind the predictability of mesoscale eddies in the future, and help the deployment of ARGO, glider, AUV and other observational platforms.

Lagrangian eddy tracking reveals the Eratosthenes anticyclonic attractor in the eastern Levantine Basin

Abstract. Statistics of anticyclonic eddy activity and eddy trajectories in the Levantine Basin over the 2000–2018 period are analyzed using the DYNED-Atlas database, which links automated mesoscale eddy detection by the Angular Momentum Eddy Detection and Tracking Algorithm (AMEDA) algorithm to in situ oceanographic observations. This easternmost region of the Mediterranean Sea, delimited by the Levantine coast and Cyprus, has a complex eddying activity, which has not yet been fully characterized. In this paper, we use Lagrangian tracking to investigate the eddy fluxes and interactions between different subregions in this area. The anticyclonic structure above the Eratosthenes Seamount is identified as hosting an anticyclone attractor, constituted by a succession of long-lived anticyclones. It has a larger radius and is more persistent (staying in the same position for up to 4 years with successive merging events) than other eddies in this region. Quantification of anticyclone flux shows that anticyclones that drift towards the Eratosthenes Seamount are mainly formed along the Israeli coast or in a neighboring area west of the seamount. The southeastern Levantine area is isolated, with no anticyclone transfers to or from the western part of the basin, defining the effective attraction basin for the Eratosthenes anticyclone attractor. Co-localized in situ profiles inside eddies provide quantitative information on their subsurface physical anomaly signature, whose intensity can vary greatly with respect to the dynamical surface signature intensity. Despite interannual variability, the so-called Eratosthenes anticyclone attractor stores a larger amount of heat and salt than neighboring anticyclones, in a deeper subsurface anomaly that usually extends down to 500 m. This suggests that this attractor could concentrate heat and salt from this subbasin, which will impact the properties of intermediate water masses created there.

Three Mesoscale Eddy Detection and Tracking Methods: Assessment for the South China Sea

AbstractComplex topography and the Kuroshio eddy-shedding process produce active mesoscale eddy activity in the South China Sea (SCS). Three eddy detection and tracking methods, the Okubo–Weiss (O-W), vector-geometry (V-G), and winding-angle (W-A) algorithms, have been widely applied for eddy identification. This study provides a comprehensive assessment of the O-W, V-G, and W-A methods in the SCS, including their detection, statistical analysis, and tracking capabilities. The mean successful detection rates of the O-W, V-G, and W-A methods are 51.9%, 56.8%, and 61.4%, respectively. The O-W and V-G methods preferentially detect eddies with medium radii (½°–1°), whereas the W-A method tends to detect eddies with larger radii (>1°). The V-G method identifies an excessive number of weak (radius < ⅓°) eddylike structures in the SCS, accounting for 48.2% of the total eddy number. The highest mean excessive detection rate of the V-G method biases the data on eddy number, probability, and propagation direction. With the lowest mean successful tracking rate (STR), the O-W method might not be suitable for tracking long-lived eddies in the SCS. The V-G method performs well with regard to the overtracking issue and has the lowest mean questionable tracking rate of 1.1%. Among the three methods, the W-A method tracks eddies most accurately, with the highest mean STR of 80.6%. Overall, the W-A method produces reasonable statistical eddy characteristics and eddy tracking results. Each method has advantages and disadvantages, and researchers should choose wisely according to their needs.

Neural Network Training for the Detection and Classification of Oceanic Mesoscale Eddies

Recent advances in deep learning have made it possible to use neural networks for the detection and classification of oceanic mesoscale eddies from satellite altimetry data. Various neural network models have been proposed in recent years to address this challenge, but they have been trained using different types of input data and evaluated using different performance metrics, making a comparison between them impossible. In this article, we examine the most common dataset and metric choices, by analyzing the reasons for the divergences between them and pointing out the most appropriate choice to obtain a fair evaluation in this scenario. Based on this comparative study, we have developed several neural network models to detect and classify oceanic eddies from satellite images, showing that our most advanced models perform better than the models previously proposed in the literature.

Detection and tracking of mesoscale eddies in the Mediterranean Sea: A comparison between the Sea Level Anomaly and the Absolute Dynamic Topography fields

Mediterranean Sea eddies are often generated in fixed geographical locations linked to either bathymetric fluctuations, coastline changes, instabilities around islands, or orographic wind forcing. Because of that, the mean circulation exhibits features with the same special scales as the mesoscale field. We provide here a first comparison of eddies detected and tracked using two altimetry products, the Sea Level Anomaly (SLA) and the Absolute Dynamic Topography (ADT), from 2000 to 2015. We showed that more individual eddies and trajectories are detected in the SLA data than in the ADT, having larger radius but lower Eddy Kinetic Energy. The spatial distribution of the mesoscale activity is different between the two data sets: the ADT-detected eddies closely follow the Mean Dynamic Topography (MDT) patterns which include the mean eddy generation sites whereas the SLA-detected eddies show more spatial homogeneity. The larger eddies are generally well detected in both the ADT or the SLA fields, but the numerous smaller and coastal eddies detected in the SLA fields have a tendency to mismatch the ADT detected eddies. Only 20–30% of eddies are detected exclusively in the SLA or the ADT product (i.e. have no collocation in the other field), whereas to 65–80% of the eddies are detected both in the SLA and the ADT fields. The Ierapetra Eddy is a typical example of the limitation of the SLA-detected eddies, as this orographically controlled feature is present in the MDT and thus leads to an artificial cyclonic eddy in the SLA field that shifts around its most frequent position. This study recommends the use of ADT fields for the detection of mesoscale structures in the Mediterranean Sea where the mean circulation and these mesoscale features have the same spatial scale and intensity, and similarly for any other ocean regions where the MDT contains mesoscale patterns. Yet the ADT fields depend on the accuracy of the MDT, and the altimeter-based SLA fields may be preferable in specific locations where the MDT is not reliable.

A new automatic oceanic mesoscale eddy detection method using satellite altimeter data based on density clustering

The mesoscale eddy is a typical mesoscale oceanic phenomenon that transfers ocean energy. The detection and extraction of mesoscale eddies is an important aspect of physical oceanography, and automatic mesoscale eddy detection algorithms are the most fundamental tools for detecting and analyzing mesoscale eddies. The main data used in mesoscale eddy detection are sea level anomaly (SLA) data merged by multi-satellite altimeters’ data. These data objectively describe the state of the sea surface height. The mesoscale eddy can be represented by a local equivalent region surrounded by an SLA closed contour, and the detection process requires the extraction of a stable closed contour structure from SLA maps. In consideration of the characteristics of mesoscale eddy detection based on SLA data, this paper proposes a new automatic mesoscale eddy detection algorithm based on clustering. The mesoscale eddy structure can be extracted by separating and filtering SLA data sets to separate a mesoscale eddy region and non-eddy region and then establishing relationships among eddy regions and mapping them on SLA maps. This paper overcomes the problem of the sensitivity of parameter setting that affects the traditional detection algorithm and does not require a sensitivity test. The proposed algorithm is thus more adaptable. An eddy discrimination mechanism is added to the algorithm to ensure the stability of the detected eddy structure and to improve the detection accuracy. On this basis, the paper selects the Northwest Pacific Ocean and the South China Sea to carry out a mesoscale eddy detection experiment. Experimental results show that the proposed algorithm is more efficient than the traditional algorithm and the results of the algorithm remain stable. The proposed algorithm detects not only stable single-core eddies but also stable multi-core eddy structures.

Helmholtz–Hodge decomposition-based 2D and 3D ocean surface current visualization for mesoscale eddy detection

Ocean surface current (or ocean flow) visualization plays a significant role in the understanding of dynamical processes of ocean. It has been a hot research topic in both computer science and oceanography. Ocean surface current is a turbulent flow field mixing of multi-scale ocean dynamics such as large-scale ocean circulations ( $$100\,\hbox {km}\sim$$ ), mesoscale eddies (10– $$100\,\hbox {km}$$ ), submesoscale processes (1– $$10\,\hbox {km}$$ ). Mesoscale eddies, which are strong but short-life movement relative to the large-scale ocean circulations, have great importance on the transportation of ocean water masses, momentum and energy. However, their detection and recognition, which are treated as the foundation of exploring their dynamical mechanisms, is still a challenging issue. For one thing, in the mixed ocean flow field, different ocean flows depended and influence with each other making existing methods difficult to identify among them. For another, mesoscale eddies are active signals on the ocean. They may change their forms and velocities at any time. This challenges existing works to deal with their boundary ambiguity and unremitting transitions. To solve these problems, this paper proposes a novel 2D and 3D ocean surface current visualization approach based on an amended Helmholtz–Hodge decomposition (HHD), which can be widely used for mesoscale eddy detection. In our method, HHD decomposes each mixed ocean flow field to two components: curl component and divergence component. Different ocean flows can be represented by these two components independently. In addition, to improve the performance of eddy identification and to reveal the 3D structure of ocean flows simultaneously, HHD transforms the 2D ocean flow field to 3D potential surfaces. Finally, comprehensive experiments are performed on both global and local ocean flow field (Black Sea and Mediterranean Sea) calculated from satellite maps of sea level anomaly to verify our method. Experimental results demonstrate the good effectiveness of our method.

Basin- to mesoscale surface circulation of the Western Mediterranean manifested by satellite-derived data products

In this article, there is an overview of the potential of different (mostly altimetry-derived) products for manifesting features of basin- and mesoscale circulation of the Western Mediterranean Basin. In terms of detection of mesoscale eddies and larger gyres, the comparison was performed between eddy statistics extracted directly from fields of sea surface temperature (SST), on the one hand, and reconstructed from different scalar and vector altimetry-derived fields, on the other hand. Among the products being analysed, there are fields of (i) sea level anomaly (SLA), (ii) geostrophic currents, and (iii) finite size Lyapunov exponents (FSLE). Comparison of the eddy statistics provided by such different data sources revealed that altimetry-based techniques tended to overestimate the number of big cyclones and thus failed to reproduce cyclonic/anticyclonic eddy asymmetry which was shown by fields of SST. Defining the spatial scale of eddies via altimetry-based techniques seemed particularly problematic. FSLE product analysed in the study showed significant disagreements with vortical structures observed in fields of SST both in terms of the spatial scale and the sign of rotation. In order to scrutinize basin-scale features of surface circulation in the region of interest, fields of (i) geostrophic currents, (ii) total currents reconstructed by the GlobCurrent project, and (iii) surface currents provided by the reanalysis technique were analysed. As a result of the comparison performed, a conclusion was made on the complementarity of the datasets analysed.

An evaluation of sea surface height assimilation using along-track and gridded products based on the Regional Ocean Modeling System (ROMS) and the four-dimensional variational data assimilation

Remote sensing products are significant in the data assimilation of an ocean model. Considering the resolution and space coverage of different remote sensing data, two types of sea surface height (SSH) product are employed in the assimilation, including the gridded products from AVISO and the original along-track observations used in the generation. To explore their impact on the assimilation results, an experiment focus on the South China Sea (SCS) is conducted based on the Regional Ocean Modeling System (ROMS) and the four-dimensional variational data assimilation (4DVAR) technology. The comparison with EN4 data set and Argo profile indicates that, the along-track SSH assimilation result presents to be more accurate than the gridded SSH assimilation, because some noises may have been introduced in the merging process. Moreover, the mesoscale eddy detection capability of the assimilation results is analyzed by a vector geometry–based algorithm. It is verified that, the assimilation of the gridded SSH shows superiority in describing the eddy’s characteristics, since the complete structure of the ocean surface has been reconstructed by the original data merging.

Automatic detection of oceanic mesoscale eddies in the South China Sea

This study focuses on the spatial and temporal distribution characteristics of mesoscale eddies in the South China Sea (SCS). An automatic eddy detection method, based on the geometry of velocity vectors, was adopted to obtain an eddy dataset from 21 years of satellite altimeter data. Analysis revealed that the number of anticyclonic eddies was nearly equal to cyclonic eddies; in the SCS, cyclonic eddies are generally stronger than anticyclonic eddies and anticyclonic eddies are larger and longer-lived than cyclonic eddies. Anticyclonic eddies tend to survive longer in the spring and summer, while cyclonic eddies have longer lifetimes in the autumn and winter. The characteristics and seasonal variations of eddies in the SCS are strongly related to variations in general ocean circulation, in the homogeneity of surface wind stress, and in the unevenness of bottom topography in the SCS. The spatial and temporal variation of mesoscale eddies in the SCS could, therefore, be an important index for understanding local hydrodynamics and regional climate change.