Elsevier

Ocean Modelling

Volume 155, November 2020, 101704
Ocean Modelling

Impact of assimilating altimeter data on eddy characteristics in the South China Sea

https://doi.org/10.1016/j.ocemod.2020.101704Get rights and content

Highlights

  • A simple data assimilation method raises the fraction of simulated eddies from 71% to 82.5%.

  • The seasonality of the cyclone/anticyclone asymmetry is improved by assimilation.

  • Residual errors after assimilation vary strongly with location and season.

  • The eddies near Mindoro Island benefit the least from data assimilation.

Abstract

From satellite altimetry, it is known the benefit of assimilating sea level anomalies (SLA) has been shown in the context of operational ocean forecast systems. However, how much data assimilation (DA) of altimetry data improves the representation of mesoscale eddies has still not been investigated in previous studies. Especially in the South China Sea (SCS), no estimation for that has been done in a long time. In this study, a nested SCS model system uses the Ensemble Optimal Interpolation method to assimilate along-track SLA data from 1993 to 2011. We assess the representation of eddy characteristics in two hindcast simulations – one with DA and the other without – and compare them to satellite-derived eddy characteristics. In the whole SCS, the assimilation improves the number of simulated cyclonic (anticyclonic) eddies by 10.3% (13.6%). The corresponding improvement in the eddy-rich northern SCS is 17.9% (19.6%). Assimilation improves the seasonality of eddy occurrence, with cyclones and anticyclones showing an obviously asymmetric seasonality. However, diagnosed assimilation effects and associated residual errors show large spatial and temporal dependencies. The radii of anticyclonic eddies smaller than 70 km are not changed with DA. The results show that deficiencies of cyclonic eddies in winter and anticyclonic eddies in summer around 13N in the southeastern SCS are not well corrected by DA, where one of the shortcomings resulted from the used wind forcing. Although this study does not conduct one-to-one forecasting experiments for each eddy track, improved eddy reproduction is the first step towards detailed validation metrics for eddy forecasting systems.

Introduction

Ocean mesoscale eddies are stable closed-circulation features found over almost the entire global oceans and acting over time scales ranging from a few days to more than a hundred days. They are characterized by their strong kinetic energy; more than ten times the background mean kinetic energy (Richardson, 1983, Chelton et al., 2007). Eddies generally propagate westward at a phase speed close to that of baroclinic Rossby waves (Chelton et al., 2007, Early et al., 2011). Typical mesoscale eddies are horizontally symmetrical vortices with radii ranging from 50 km to 500 km. They are regarded as the “ocean weather” that stirs the stratified ocean (Le Traon and Dibarboure, 1999, Li and Pohlmann, 2002, Halo et al., 2014) and facilitates the exchange of properties between different water masses (Wunsch, 2008, Dufois et al., 2016).

The South China Sea (SCS) is a semi-enclosed marginal sea, having only one main connection to the western Pacific through Luzon Strait. Its circulation, relatively independent from the tropical ocean circulation, is mainly driven by the alternating northeast and southwest monsoon winds in winter and summer respectively (Qu, 2000, Metzger, 2003, Gan et al., 2006). However, the Kuroshio intrusion combined with complex local topographic steering directly affects the ocean circulation through Luzon Strait, resulting in mesoscale eddies easily spreading into the northern SCS (NSCS) (Wang et al., 2003, Chow et al., 2008, Chen et al., 2012). Eddy genesis mechanisms have been related to the wind stress curl, ring shedding from the Kuroshio, frontal instability related to the Kuroshio intrusion and orographic wind jets (Wu and Chiang, 2007, Wang et al., 2008, Zhang et al., 2017). Numerical simulations of eddies rely strongly on high-quality atmospheric forcing, boundary transports, and ocean initial state. Numerous cases of improvements in simulations or forecasts of specific eddies due to data assimilation have been reported (e.g., Wu and Chiang, 2007, Wang et al., 2012, Xu et al., 2019). However, no studies quantitatively evaluate the extent to which the predicted eddies and their trajectories over a long time are attributed to data assimilation (DA) except for a few cases under ocean reanalysis frameworks. This may be a result of the intrinsic chaotic nonlinear dynamics involved and practical limits in accurately characterizing real eddies and their tracks. In a two-years-assimilation experiment, De Vos et al. (2018) noted a generally positive impact of assimilating along-track sea-level anomaly (SLA) on simulated eddy characteristics in the Agulhas Current System. Based on an eddy-resolving ocean reanalysis, Pilo et al. (2018) found the model’s response to DA distorts the eddies and were cautious about interpreting their results because the adjusted vertical advection terms are about 1.5 times larger after assimilation. This issue is hopefully irrelevant in the Arbitrary Lagrangian–Eulerian coordinates model used for this study, which represents vertical velocities by vertical displacements of material coordinates smoothed by the model.

The growth of the satellite altimeter constellation to a mature state has enabled statistics of eddy trajectories by various automatic eddy detection algorithms to be determined (e.g. Penven et al., 2005, Chelton et al., 2007, Souza et al., 2011, Faghmous et al., 2015). Previous efforts to document SCS eddies by manual detection from altimetry have been made (Wang et al., 2003, Chen et al., 2011, Du et al., 2016, Zhu et al., 2016). However, there are only a few attempts to evaluate automatically simulations of eddy trajectories against satellite altimeter products in the SCS. This is asserted to be valuable in the context of the SCS serving as a useful example for other semi-enclosed marginal seas. Xiu et al. (2010) give a census of long-lived eddies (>30 days) in the SCS during 1993–2007. They first validate a Regional Ocean Model System (ROMS) simulation of eddies against merged weekly SLA maps from Archiving, Validation, and Interpolation of Satellite Oceanographic Data (AVISO; now available through CMEMS, Copernicus Marine Environmental Monitoring Services). Neglecting eddies in waters shallower than 1000 m, they counted 32.8 ± 3.4 eddies in satellite maps each year, with 53% of them being cyclonic and 70% of these having radii smaller than 100 km. Zhu et al. (2016) contrast eddies originating from a global and a regional forecast system to a daily 1/4°deg resolution satellite SLA data product in 2012. For eddies with lifetimes longer than 28 days, they found 24, 47, and 24 eddies in altimeter data, global and regional forecast systems respectively, but the spatial distributions of eddies’ origins (their Figure 12) uncover large differences between the different forecasting systems. Further, all the above studies disregard eddies with lifetimes between 10 days and 28 days. Since these account for over 50% of all eddies, referred to as “short-lived” in Chen and Han (2019), they are an important component of the ocean’s dynamic variability.

Due to various technical reasons, in situ profiles of temperature and salinity are notably poor and intermittent in the SCS. For instance, yearly in situ profiles in 2004–2007 are fewer than 500 as shown in Fig. 4b of Zeng et al. (2016). This further highlights the important role of assimilating altimeter data in representing mesoscale eddy activity in the SCS and the importance of multivariate corrections by DA. An eddy-resolving numerical model such as the one used in this study should, in principle, reproduce the statistics of eddy activity, but not the timing of the chaotic events during its lifetime (genesis, motions, decay, merging). Although data assimilation experiments in the SCS do show the benefits of altimeter data for reducing errors in SLA (e.g., Xie et al., 2011, Lyu et al., 2014), the extent to which reproduced eddies are attributable to DA has not yet been addressed. Two initial challenges arise from sensitivities to the selection of parameters used in the eddy tracking process, as well as to differing spatial resolutions between the models and altimetry data. This has a bearing on the fair comparison of the performance of the different products. To understand the role of DA on simulated ocean mesoscale variability, Cipollone et al. (2017) compared eddy statistics of a global ocean reanalysis, a free ocean simulation, and an observation-based dataset on a common 1/4°horizontal grid. They concluded that DA enhances the mesoscale variability of some features which cannot be well-reproduced in the free-running model.

Another challenge is to select appropriate eddy characteristics for evaluation, which is both useful and available from long time series of observations. In general, data assimilation techniques also assume the model to be bias-free. If this is the case, they should effectively constrain eddy lifecycles, but residual errors can be higher with model biases, which may be stronger in certain locations or during certain seasons. Biases may arise from various reasons; for example, due to numerical schemes (Backeberg et al., 2009), biases in the forcing fields or errors in the bottom topography.

The Ensemble Optimal Interpolation method (EnOI; Oke et al., 2002) has been applied to assimilate SLA data in different ocean operational forecast centers (e.g., Oke et al., 2008, Lyu et al., 2014, Tanajura et al., 2014) and has proven able to improve forecasts of mesoscale eddies and fronts (Backeberg et al., 2014, Counillon and Bertino, 2009, Oke et al., 2008). Other DA methods relying on static ensembles are used in several operational centers (Tonani et al., 2015 and references therein). In this study, applying the EnOI method, we only assimilate along-track altimetry data weekly into a nested eddy-resolving SCS ocean model. The DA system is used in hindcast mode with the model forced by reanalyzed winds rather than in real-time prediction. As such, higher performance than might be achieved by operational forecast modes can be expected. Based on the diagnostic evaluation of two model runs without and with DA, its effects are quantified in terms of the basic eddy characteristics such as eddy centroids and radii as a pilot study. This evaluation initially highlights their geographic and seasonal variations, disregarding the irregular shapes of the eddies, and neglecting the eddy distance error. Mechanism analysis related to the eddies cyclogenesis and propagation speeds has not been pursued in this study.

Section snippets

Altimetry data for assimilation and evaluation

The delayed-time (DT) along-track sea level anomaly (TSLA) data were produced Data Unification and Altimeter Combination System (DUACS) and released by AVISO (http://www.aviso.oceanobs.com). To assimilate a fine resolution of TSLA equivalent to the model resolution, the unfiltered TSLA product of DUACS DT2014 is used with a 7 km horizontal resolution along-track (Pujol et al., 2016). Nowadays, another similar TSLA product but with a lower resolution (14 km) is available through the CMEMS

Automatic eddy detection and tracking procedure

Using closed contours of SLA from satellite maps, Wang et al. (2003) identify SCS eddies with a lifetime longer than 30 days and in depths deeper than 1000 m. They find 58 anticyclonic eddies and 28 cyclonic eddies during the years of 1993–2000. In automatic detection methods, the Okubo–Weiss​ parameter (Okubo, 1970, Weiss, 1991) is one of the most widespread tools to treat both satellite data and numerical model results. It is defined as W=sn2+ss2ζ2,where sn=uxvy is the normal component

Results

Having verified the expected effect of DA on the location and vertical structure of eddies, their detailed characteristics are investigated here. We focus on the eddies whose centers once appeared in the deep basin (> 100 m) of the SCS domain (105–120.5°E, 2–22.5°N). The main expected eddy hotspots follow the continental slopes from the northwest of Luzon Strait to the coast of Vietnam until 6°N. Another eddy hotspot (>70 eddy days) resides northwest of Luzon island and is known as the “west

Discussion and conclusions

Since the early 2000s, altimetry data have been routinely assimilated into regional and global ocean forecast models to improve the eddy forecasts. However, little documentation of the accuracy of eddy characteristics has been produced, even less so in a regional sea like the SCS. Even though our experiments have been conducted in reanalysis mode, the eddy characteristics evaluated here are useful as an upper limit of the eddy prediction skill one would expect in a general ocean operational

CRediT authorship contribution statement

J. Xie: Initiated the collaboration from assimilation experiments to writing -original draft. M. De Vos: Improved the original draft by the methodology polishing, Writing - review & editing. L. Bertino: Interpretations of result, Methodology, Writing - review & editing. J. Zhu: Interpretations of result, Methodology. F. Counillon: Interpretations of result, Methodology, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

We are grateful to Dr. P. Penven for the software package to detect eddy and feature scales and thank Dr. R.P.Raj for fruitful discussions. Grants of computing time (nn2993k and nn9481k) and storage (ns2993k) from the Norwegian Sigma2 infrastructures are gratefully acknowledged. This work also has received partial support from the Hong Kong Research Grants Council under the Theme-based Research Scheme (TRS) through Project No. T21-602/16-R and the National Science Foundation of China (Grant

References (65)

  • BackebergB.C. et al.

    Evaluating two numerical advection schemes in HYCOM for eddy-resolving modelling of the Agulhas Current

    Ocean Sci.

    (2009)
  • BackebergB. et al.

    Assimilating along-track SLA data using the EnOI in an eddy resolving model of the Agulhas system

    Ocean Dyn.

    (2014)
  • ChassignetE.P. et al.

    US GODAE: Global ocean prediction with the hybrid coordinate ocean model (HYCOM)

    Oceanography

    (2009)
  • ChassignetE.P. et al.

    Impact of horizontal resolution (1/12° to 1/50°) on Gulf Stream separation, penetration, and variability

    J. Phys. Oceanogr.

    (2017)
  • CheltonD.B. et al.

    Global observations of large oceanic eddies

    J. Geophys. Res.

    (2007)
  • ChenG. et al.

    Eddy heat and salt transports in the South China Sea and their seasonal modulations

    J. Geophys. Res.

    (2012)
  • ChenG. et al.

    Contrasting short-lived with long-lived mesoscale eddies in the global ocean

    J. Geophys. Res.

    (2019)
  • ChenG. et al.

    Mesoscale eddies in the South China Sea: Mean properties, spatiotemporal variability, and impact on thermohaline structure

    J. Geophys. Res.

    (2011)
  • ChoiJ.-W. et al.

    Possible relation of the western North Pacific monsoon to the tropical cyclone activity over western North Pacific

    Int. J. Climatol.

    (2016)
  • ChowC.-H. et al.

    Mesoscale Dongsha cyclonic eddy in the northern South China Sea by drifter and satellite observations

    J. Geophys. Res.

    (2008)
  • CipolloneA. et al.

    Benchmarking the mesoscale variability in global ocean eddy-permitting numerical systems

    Ocean Dyn.

    (2017)
  • CohenJ.

    Statistical Power Analysis for the Behavioral Sciences

    (1988)
  • CounillonF. et al.

    High-resolution ensemble forecasting for the Gulf of Mexico eddies and fronts

    Ocean Dyn.

    (2009)
  • De VosM. et al.

    Using an eddy-tracking algorithm to understand the impact of assimilating altimetry data on the eddy characteristics of the Agulhas system

    Ocean Dyn.

    (2018)
  • DibarboureG. et al.

    A demonstration of the potential of Cryosat-2 to contribute to mesoscale observation

    J. Adv. Space Res.

    (2011)
  • DufoisF. et al.

    Anticyclonic eddies are more productive than cyclonic eddies in subtropical gyres because of winter mixing

    Sci. Adv.

    (2016)
  • EarlyJ.J. et al.

    B: The evolution and propagation of quasigeostrophic ocean eddies

    J. Phys. Oceanogr.

    (2011)
  • EvensenG.

    The ensemble Kalman filter: theoretical formulation and practical implementation

    Ocean Dyn.

    (2003)
  • FaghmousJ.H. et al.

    A daily global mesoscale ocean eddy dataset from satellite altimetry

    Sci. Data

    (2015)
  • FaghmousJ.H. et al.

    Data from: A Daily Global Mesoscale Ocean Eddy Dataset from Satellite Altimetry

    (2016)
  • GanJ. et al.

    Modeling South China Sea circulation: response to seasonal forcing regimes

    J. Geophys. Res.

    (2006)
  • GoodS.A. et al.

    EN4: quality controlled ocean temperature and salinity profiles and monthly objective analyses with uncertainty estimates

    J. Geophys. Res.

    (2013)
  • Cited by (5)

    • Verification of eddy properties in operational oceanographic analysis systems

      2022, Ocean Modelling
      Citation Excerpt :

      Cipollone et al. (2017) use both a surface and three-dimensional eddy detection algorithm to demonstrate the positive impact that data assimilation has on the representation of eddies. Xie et al. (2020) also demonstrate the positive impact of data assimilation in terms of the number and mean radius of eddies in the South China Sea. Similarly, de Vos et al. (2018) show the impact of data assimilation on eddy population, density and lifetime in the Agulhas Region.

    View full text