Elsevier

Ocean Modelling

Volume 146, February 2020, 101538
Ocean Modelling

Using multi-platform 4D-Var data assimilation to improve modeling of Adriatic Sea dynamics

https://doi.org/10.1016/j.ocemod.2019.101538Get rights and content

Highlights

  • The year-long 4D-Var data assimilation was applied in the Adriatic Sea.

  • The 4D-Var system improved modeling and revealed previously unknown ocean dynamics.

  • Corrections to atmospheric forcing, open boundary and initial states were computed.

  • 4D-Var estimate shows data mismatch to be significantly smaller than in the baseline solution.

  • Assimilation indicates that many fine scale features have possibly effected previous studies.

Abstract

A year-long Four-Dimensional Variational (4D-Var) data assimilation experiment, formulated on the physical-space statistical analysis system (PSAS) algorithm, has been applied to the Adriatic Sea for the very first time. High-resolution multi-platform observations were assimilated into the Regional Ocean Modeling System (ROMS) forced by mesoscale atmospheric model ALADIN/HR (Aire Limitée Adaptation dynamique Développement InterNational), including important freshwater river inflows and realistic open boundary conditions for the period between October 2014 and September 2015. The observations included sea surface temperature (SST) measured by satellites, in situ temperature and salinity data measured by various moving (Argo profiling floats, shipborne CTDs, sea glider, towed CTD profiler) and moored platforms, ocean current profiles measured by moored Acoustic Doppler Current Profilers (ADCPs), and 30-minute de-tided surface currents from high-frequency (HF) radars. Three model simulations were integrated: (1) a non-assimilative simulation over a year-long period — baseline simulation; (2) a non-assimilative simulation initiated by the previous assimilation 4-day cycle — background simulation, and (3) a fully assimilative simulation that used all available observations during the 4-day assimilation cycle — analysis simulation. The assimilation significantly improved the modeling system performance, especially in SST, with time average rmse equaling 0.9, 0.7 and 0.5 °C and bias equaling 0.39, 0.15, 0.01 °C for the baseline, background and analysis simulations, respectively. These reductions were mostly achieved during the wintertime outbreaks of a cold and dry bora wind, caused by an underestimation of heat and momentum fluxes by the atmospheric model. Such fine adjustments in the northern and middle Adriatic may be responsible for improved reproduction of dense water generation and the associated Adriatic thermohaline circulation.

Introduction

Recent advances in computing power, developments in ocean data assimilation (hereafter DA) techniques, improvements in satellite radiometer resolution and increased number of locally available observations (i.e. coastal HF radars, ADCPs and CTD mooring/cruises, coastal buoys), have allowed for substantial advances in objective and dynamically consistent mapping of the ocean state (Dobricic and Pinardi, 2008, Li et al., 2008). The Four-Dimensional Variational (4D-Var) DA technique (Courtier et al., 1994, Courtier, 1997) is the most advanced version of the variational schemes, with its usage becoming more prevalent in ocean research (Weaver et al., 2003, Moore et al., 2009, Moore et al., 2011a, Moore et al., 2011b, Usui et al., 2015).

There are many benefits of using 4D-Var DA over simpler methods like 3D-Var, optimal interpolation, or the simplest nudging techniques (Powell et al., 2008, Powell et al., 2012, Moore et al., 2009, Moore et al., 2011a, Moore et al., 2011b, Matthews et al., 2012, Powell, 2017). The notable feature of 4D-Var DA is its capability to ingest observations at their exact time and location, together with their time-evolving error covariance matrix (Weaver and Courtier, 2001). This feature retains temporal information that would otherwise be lost through averaging, thus enhancing our ability to simulate rapidly changing coastal ocean dynamics, as occurs during severe and abrupt winter storms that are associated with extreme cooling processes (Janeković et al., 2014).

By definition, variational methods aim to minimize model to data discrepancy (JO) and changes to the model background (associated model part of the cost function - JM) at the same time during each DA cycle, usually referred to as total cost function J=JM+JO (Eq. (1)). This is achieved by adjusting the control parameters for initial and boundary conditions, and atmospheric forcing fields: J=JM+JO=12δxTD1δx+12GδxdTR1Gδxdwhere x is state vector (i.e. free surface, currents or tracer), δx represents deviation of x from its background, D is the normalized background error covariance matrix, R is the normalized observational error matrix and Gδxd represents the mismatch between the Tangent Linear model mapped to the observations (Gδx) and the innovation vector d (defined as difference between observations and the model equivalent of the observations computed from the background circulation). The goal of minimization is to find optimal perturbation (δxa), such as to minimize J with respect to the deviation of the state vector from its background δx (Eq. (2)): δxa=[D1+GTR1G]1GTR1dOne of the challenges in this approach is to very closely proximate the D matrix, which in our case was done by using correlation modeling via a generalized diffusion equation (Weaver and Courtier, 2001). These methods rely on the linear variational approach (Tangent Linear model and the Adjoint model) that limits spatial and temporal scales specific to the modeled region. In high resolution coastal applications, nonlinearity of the system increases with smaller spatial and temporal scales which can be of order of 100 m in space and seconds in time. Consequently, this can pose significant computational burdens. Despite the demanding computational requirements, 4D-Var has also been successfully applied in highly dynamical, regional coastal seas (Janeković et al., 2013, Iermano et al., 2015, Souza et al., 2015, Kerry et al., 2016, Sperrevik et al., 2017).

The scope of the study is the Adriatic Sea, the 800 × 200 km northernmost extension of the Mediterranean Sea, connected to the adjacent Ionian Sea through the Otranto Strait (Fig. 1). The Adriatic bathymetry exhibits a shallow (<80 m) and wide northern Adriatic shelf, gently increasing in depth towards the middle Adriatic depressions of approx. 250 m (Jabuka Pit, JP in Fig. 1), being separated from the 1200 m deep and circular Southern Adriatic Pit (SAP) by the 170 m deep Palagruža Sill (PS in Fig. 1). General circulation of the Adriatic is cyclonic, with Eastern Adriatic Current (EAC) flowing along the eastern perimeter of the sea and transporting saline Levantine Intermediate Water (LIW, Zore-Armanda, 1963), while the Western Adriatic Current (WAC) transports fresher surface water having its origin in the northern Adriatic (Orlić et al., 1992, Artegiani et al., 1997).

Although the northern Adriatic is shallow and its bathymetry mostly flat, its circulation is particularly challenging to reproduce. This is because of: (1) the influence of large rivers, of which the largest is the Po River with an annual mean flux of approx. 1500 m3/s (Raicich, 1994, Vilibić et al., 2016); and (2) extreme forcing during winters when across-basin, cold, and dry bora winds blow downslope along the coastal mountains and extensively cool the shallow sea (Grisogono and Belušić, 2009). Heat losses during such periods may reach 2000 W/m2 (Janeković et al., 2014). The consequence is the generation of extremely dense and cold waters (1030.0 kg/m3 and 5–7 °C during extreme episodes, e.g. 2012) (Mihanović et al., 2013, Janeković et al., 2014) which sink to form a bottom density current (Nof, 1983) and fill the middle Adriatic depressions (Artegiani and Salusti, 1987, Vilibić and Supić, 2005) and the Southern Adriatic Pit (Querin et al., 2016). Conversely, during the summer season, sea surface temperatures can reach as high as 30°C with sharp thermocline located at 20 m depth. Additionally, the effect of tides can be important in the northern Adriatic, particularly in narrow channels (Janeković and Kuzmić, 2005).

In this study, we present a year-long reanalysis of the Adriatic dynamics carried out by using 4D-Var DA, formulated on the physical-space statistical analysis system algorithm (PSAS) (Moore et al., 2011b). In this system, applied minimization procedures that estimate optimal analysis increments are defined in the observation space, in contrast to the primal definition and full model space (I4D-Var).

The motivation for the study was to (a) quantify the usefulness of 4D-Var DA in such a dynamic environment due to linear constraints inherent inside Tangent Linear and Adjoint models, (b) quantify how long the improvements achieved by DA persist in the model, and (c) quantify the improvements in reproducing the ocean state, i.e. to reach the best ocean state description in relation to different observing platforms used in the Adriatic Sea.

The structure of the paper is as follows: the modeling system setup is described in Section 2; multi-platform observations assimilated in our numerical experiments in Section 3; Section 4 is an assessment of the results and a verification using the independent dataset; and Section 5 contains the discussion and summarizes the major conclusions.

Section snippets

Numerical model setup and experiments

ROMS was used as a base ocean model in the DA experiment. ROMS is three-dimensional, free surface, bathymetry following s-coordinate model solving set of the Reynolds-Averaged Navier–Stokes equations with finite difference approximation and a time splitting approach (Shchepetkin and McWilliams, 2005, Shchepetkin and McWilliams, 2009).

The standard set of daily averaged lateral boundary conditions (free surface, temperature, salinity and velocity) were interpolated from the wider Adriatic

Observations

The observational period ranged from 1 October 2014 to 30 September 2015, with more than 15 million, high-resolution observations used in the DA, most of which were sea surface temperatures (SST) measured by satellite radiometers and surface ocean currents measured with high-frequency (HF) radars. In addition, the following in situ data has been used for the DA: (1) vertical temperature and salinity profiles from various ocean platforms (Argo profiling floats, shipborne yo-yo profiler,

Cost function

During the year-long DA experiment, applied model minimization procedure reduced observation-to-model discrepancy scaled with appropriate observation error covariance matrix (JO inside the cost function J in Eq. (1), Fig. 3a) and at the same time penalizing mismatch with the model background (JM in the Eq. (1) and Fig. 3b; for details see Broquet et al., 2011, Moore et al., 2009, Moore et al., 2011a, Moore et al., 2011b). Overall, the mismatch between the model and the observations (JO) was

Discussion and conclusions

Until now, reproducing the dynamics of the Adriatic at high resolution has been predominantly achieved by ocean models forced by mesoscale atmospheric models (e.g. Bergamasco et al., 1999, Beg Paklar et al., 2001) or coupled atmosphere-ocean models (e.g., Loglisci et al., 2004, Pullen et al., 2006, Pullen et al., 2007, Ličer et al., 2016), some of them using a simple bias correction for improving the model performance (e.g. Benetazzo et al., 2014).

To improve the model dynamics, we implemented a

CRediT authorship contribution statement

I. Janeković: Conceptualization, Methodology, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing. H. Mihanović: Data curation, Formal analysis, Investigation, Writing - review & editing. I. Vilibić: Data curation, Formal analysis, Investigation, Writing - review & editing, Funding acquisition. B. Grčić: Software, Investigation, Visualization.

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

This project has been supported by Croatian Science Foundation under the Grants ADAM-ADRIA (IP-2013-11-5928), ADIOS (IP-2016-06-1955), and SCOOL (IP-2014-09-5747). Some observational data were acquired within FP7 PERSEUS-ADREX and EUROFLEETS2 ESAW (Grant No. 312762) projects. Numerical simulations were made possible using HPC infrastructure through ECMWF project “PSAS Data Assimilation for the Adriatic Sea using Regional Ocean Modeling System (ROMS)” and Pawsey Supercomputer Centre, Australia.

References (92)

  • MooreA.M. et al.

    The Regional Ocean Modeling System (ROMS) 4-dimensional variational data assimilation systems: Part I – System overview and formulation

    Prog. Oceanogr.

    (2011)
  • NofD.

    The translation of isolated cold eddies along a sloping bottom

    Deep-Sea Res.

    (1983)
  • PowellB.S. et al.

    4DVAR data assimilation in the Intra-Americas Sea with the Regional Ocean Modeling System (ROMS)

    Ocean Model.

    (2008)
  • ShchepetkinA.F. et al.

    The regional oceanic modeling system: a split-explicit, free-surface, topography-following coordinate ocean mode

    Ocean Model.

    (2005)
  • SmolarkiewiczP.K. et al.

    MPDATA: A finite-difference solver for geophysical flows

    J. Comput. Phys.

    (1998)
  • TurchettoM. et al.

    Physical and biogeochemical processes controlling particle fluxes variability and carbon export in the Southern Adriatic

    Cont. Shelf Res.

    (2012)
  • VilibićI. et al.

    Modelling the formation of dense water in the northern Adriatic: sensitivity studies

    Ocean Model.

    (2016)
  • ZhangW.F.G. et al.

    Towards an integrated observation and modeling system in the New York Bight using variational methods. Part I: 4DVAR data assimilation

    Ocean Model.

    (2010)
  • ArtegianiA. et al.

    The Adriatic Sea general circulation, part I: air– sea interactions and water mass structure

    J. Phys. Oceanogr.

    (1997)
  • ArtegianiA. et al.

    Field observation of the flow of dense water on the bottom of the Adriatic Sea during the winter of 1981

    Oceanol. Acta

    (1987)
  • BatistićM. et al.

    Interannual variations in Adriatic Sea zoo- plankton mirror shifts in circulation regimes in the Ionian Sea

    Clim. Res.

    (2014)
  • BuljanM. et al.
  • CardinV. et al.

    Thermohaline properties in the Eastern Mediterranean in the last three decades: is the basin returning to the pre-EMT situation?

    Ocean Sci.

    (2015)
  • ChapmanD.C.

    Numerical treatment of cross-shelf open boundaries in a barotropic coastal ocean model

    J. Phys. Oceanogr.

    (1985)
  • CourtierP.

    Dual formulation of four-dimensional variational assimilation

    Q. J. R. Meteorol. Soc.

    (1997)
  • CourtierP. et al.

    A strategy for operational implementation of 4D-Var, using an incremental approach

    Q. J. R. Meteorol. Soc.

    (1994)
  • DormanC.E. et al.

    Winter 2003 marine atmospheric conditions and the Bora over the northern Adriatic

    J. Geophys. Res. Oceans

    (2007)
  • Dutour SikirićM. et al.

    A new approach to bathymetry smoothing in sigma-coordinate ocean models

    Ocean Model.

    (2009)
  • FairallC.W. et al.

    Bulk parameterization of air-sea fluxes for tropical ocean-global atmosphere coupled-ocean atmosphere response experiment

    J. Geophys. Res.

    (1996)
  • FlatherR.A.

    A tidal model of the northwest European continental shelf

    Mem. Soc. R. Sci. Liege

    (1976)
  • FragosoM.D. et al.

    A 4D-variational ocean data assimilation application for Santos Basin, Brazil

    Ocean Dyn.

    (2016)
  • GačićM. et al.

    Can internal processes sustain reversals of the ocean upper circulation? The Ionian Sea example

    Geophys. Res. Lett.

    (2010)
  • GačićM. et al.

    Thermohaline properties and circulation in the Otranto Strait

    Bull. Inst. Oceanogr. (Monaco)

    (1996)
  • GomezR. et al.

    Real-time quality control of current velocity data on individual grid cells in WERA HF radar

  • GrisogonoB. et al.

    A review of recent advances in understanding the meso- and microscale properties of the severe Bora wind

    Tellus A

    (2009)
  • IermanoI. et al.

    Impacts of a 4-dimensional variational data assimilation in a coastal ocean model of southern Tyrrhenian Sea

    J. Mar. Syst.

    (2015)
  • Ivatek-ŠahdanS. et al.

    Use of high-resolution dynamical adaptation in operational suite and research impact studies

    Meteorol. Z.

    (2004)
  • JanekovićI. et al.

    Numerical solution of the Adriatic Sea principal tidal constituents

    Ann. Geophys.

    (2005)
  • JanekovićI. et al.

    Extreme cooling and dense water formation estimates in open and coastal regions of the Adriatic Sea during the winter of 2012

    J. Geophys. Res.

    (2014)
  • JanekovićI. et al.

    4D-Var data assimilation in a nested, coastal ocean model: A Hawaiian case study

    J. Geophys. Res. Oceans

    (2013)
  • JPL/OBPG/RSMAS

    GHRSST level 2P global skin sea surface temperature from the moderate resolution imaging spectroradiometer (+MODIS) on the NASA Aqua satellite. Ver. 10 PO.DAAC, CA, USA

    (2006)
  • KerryC. et al.

    Development and evaluation of a high-resolution reanalysis of the East Australian current region using the Regional Ocean Modelling System (ROMS 3.4) and Incremental Strong-Constraint 4-Dimensional Variational (IS4D-Var) data assimilation

    Geosci. Model Dev.

    (2016)
  • KokkiniZ. et al.

    A multiplatform investigation of Istrian Front dynamics (north Adriatic Sea) in winter 2015

    Mediterranean Mar. Sci.

    (2017)
  • KourafalouV.H.

    Process studies on the Po River plume, North Adriatic Sea

    J. Geophys. Res.

    (1999)
  • KuzmićM. et al.

    Examining deep and shallow Adriatic bora events

    Q. J. R. Meteorol. Soc.

    (2015)
  • KuzmićM. et al.

    Modeling the northern Adriatic double-gyre response to intense bora wind: A revisit

    J. Geophys. Res.

    (2006)
  • Cited by (17)

    • A data-assimilative model reanalysis of the U.S. Mid Atlantic Bight and Gulf of Maine: Configuration and comparison to observations and global ocean models

      2022, Progress in Oceanography
      Citation Excerpt :

      The system is described in detail by Moore et al. (2011a, 2011b, 2011c). In addition to our own work in the MAB (Levin et al. 2020, 2021a, 2021b), ROMS 4D-Var has been applied to studies of circulation in the California Current system (Neveu et al., 2016), Intra-Americas Sea (Powell et al., 2008), East Australian Current (Kerry et al., 2016), Yellow Sea (Lee et al., 2018), Adriatic Sea (Janeković et al., 2020) and elsewhere. The assimilation algorithm used in Doppio is the incremental, strong constraint, dual formulation of 4D-Var using the Restricted B-preconditioned Lanczos (RBL4DVAR) formulation of the conjugate gradient method in which the cost function (comprising contributions from the model-data misfit and the perturbation from the background, or prior, estimate) is minimized directly in the space spanned by the observations (Gürol et al., 2014).

    • Impact of 4D-Var data assimilation on modelling of the East China Sea dynamics

      2022, Ocean Modelling
      Citation Excerpt :

      The free-run model and 4D-Var configurations are listed below, followed by the introduction of the observation impact method. As an advanced DA system, ROMS 4D-Var has prevalent usage in ocean research around the world (Powell et al., 2008; Li et al., 2015; da Rocha Fragoso et al., 2016; Janeković et al., 2020). Here, only a brief introduction to 4D-Var will be given.

    • Dissolved organic carbon accumulation during a bloom of invasive gelatinous zooplankton Mnemiopsis leidyi in the northern Adriatic Sea; case of the anomalous summer in 2017

      2021, Journal of Marine Systems
      Citation Excerpt :

      The grid used has a 4 km resolution which is adequate here. The surface forcing is obtained from the ECMWF forecasts, while the boundary conditions at the Otranto strait (temperature, salinity, currents and water height), as well as initial state, were taken from the climatology obtained from the model runs of the ADAM-ADRIA project (Janeković et al., 2020). The river inflow was obtained from Vilibić et al. (2016).

    View all citing articles on Scopus
    View full text