Comparison of the Atlantic meridional overturning circulation between 1960 and 2007 in six ocean reanalysis products

Abstract

The mean and variability of the Atlantic meridional overturning circulation (AMOC), as represented in six ocean reanalysis products, are analyzed over the period 1960–2007. Particular focus is on multi-decadal trends and interannual variability at 26.5°N and 45°N. For four of the six reanalysis products, corresponding reference simulations obtained from the same models and forcing datasets but without the imposition of subsurface data constraints are included for comparison. An emphasis is placed on identifying general characteristics of the reanalysis representation of AMOC relative to their reference simulations without subsurface data constraints. The AMOC as simulated in these two sets are presented in the context of results from the Coordinated Ocean-ice Reference Experiments phase II (CORE-II) effort, wherein a common interannually varying atmospheric forcing data set was used to force a large and diverse set of global ocean-ice models. Relative to the reference simulations and CORE-II forced model simulations it is shown that (1) the reanalysis products tend to have greater AMOC mean strength and enhanced variance and (2) the reanalysis products are less consistent in their year-to-year AMOC changes. We also find that relative to the reference simulations (but not the CORE-II forced model simulations) the reanalysis products tend to have enhanced multi-decadal trends (from 1975–1995 to 1995–2007) in the mid to high latitudes of the northern hemisphere.

Appendices

Appendix 1: Details of the ocean reanalysis and No Assimilation products

1.1 ECDA

The GFDL Ensemble Coupled Data Assimilation version 3.1 system (ECDA; Zhang et al. 2007) employs an ensemble-based filtering algorithm with the GFDL fully coupled model version 2.1 (CM2.1; see Delworth et al. 2006 for details). The CM2.1 uses the MOM version 4 (Griffies 2005) at nominal 1° horizontal resolution telescoping to $\frac{1}{3}^\circ$ meridional spacing near the equator and 50 vertical levels (22 levels of 10 m thickness in the top 220 m) and includes a dynamical ice model.

The filter used in both the ocean and atmosphere is based on the Ensemble Adjustment Kalman Filter (EAKF) described in Anderson (2003). The atmosphere is updated at 6-hourly intervals based on temperature and winds from NCEP-RA gridded reanalysis dataset (version 1 before 1979 and version 2 afterward). Ocean subsurface observations of temperature and salinity from the WOD09 (Boyer et al. 2010) and Reynolds (e.g., Reynolds et al. 2002; before 1990) and AVHRR (Casey et al. 2010; after 1989) SST are assimilated in 5-day windows using non-stationary covariance structures diagnosed sequentially from the coupled model. Because subsurface salinity observations were extremely sparse prior to the mid 2000s, statistically derived ‘pseudo-salinity’ profiles are constructed based on the observed temperature and altimetry (Chang et al. 2011). These pseudo-salinity profiles are assimilated in the ocean for 1993–2002. Altimetry is not used to constrain sea surface height directly. In the ocean, the EAKF uses an adaptive inflation algorithm that is designed to enhance the consistency of upper and deep ocean adjustments (Zhang and Rosati 2010). A weak restoring to climatological temperature and salinity is applied below 2000 m to control the deep ocean drift.

1.2 GECCO2 and GECCO2-REF

The GECCO2 ocean synthesis (Köhl 2015) uses the 4DVar framework from the German contribution of the Estimating the Circulation and Climate of the Ocean project but builds on a higher resolution (1° zonally and between 30 and 100 km meridionally with 50 levels in the vertical) global configuration of the MITgcm (Adcroft et al. 2004). This configuration of the MITgcm includes a prognostic ice model. After a 1 year integration from a resting circulation with temperature and salinity from the WOA09 (Locarnini et al. 2010), the model is forced with the NCEP-RA1 atmospheric state using the Large and Yeager (2009) bulk formula from January 1948 onward. From this first guess, the GECCO2 employs the adjoint method to adjust the initial (1 January 1948) temperature and salinity and the atmospheric surface forcing every 10 days to bring the model into consistency with the assimilated data. The GECCO2-REF No Assimilation control run is initialized on 1 January 1948 from the unadjusted first guess, also forced with the NCEP-RA1. The GECCO2-REF uses a 30-day relaxation to WOA09 sea surface salinity and HadISST to control for drift.

1.3 ORAS4 and ORAS4-CNTRL

ORAS4 (Ocean Reanalysis System 4; Balmaseda et al. 2012) has been produced by combining, every 10 days, the output of an ocean model forced by atmospheric reanalysis fluxes and quality controlled ocean observations. ORAS4 uses the NEMO ocean model (Madec 2001) and the NEMOVAR (Daget et al. 2009; Mogensen et al. 2012) data assimilation system in its 3DVar configuration. A model bias correction scheme (Balmaseda et al. 2007b) is used to reduce potential spurious variability resulting from changes in the observing system. The bias correction first guess, a seasonal cycle of 3-dimensional model error, is estimated from the data-rich Argo era. The ocean model horizontal resolution is approximately 1°—refined meridionally down to $\frac{1}{3}^\circ$ at the equator. There are 42 vertical levels with separations varying smoothly from 10 m at the surface to 300 m at the bottom. The ocean model is forced by daily atmospheric-derived surface fluxes from the ERA-40 reanalysis (Uppala et al. 2005) until December 1989, and from ERA-I (Dee et al. 2011) thereafter. The heat fluxes are adjusted using a strong relaxation to gridded SST products.

Ocean observations consist of temperature and salinity profiles from the Hadley Centres EN3 data collection (Ingleby and Huddleston 2007), who acquired much of the data from the WOD09. Altimeter-derived along track sea level anomalies from Aviso are also assimilated. Gridded maps of SST (combined from ERA-40 and NCEPOI-v2; Reynolds et al. 2002) are used to adjust the heat fluxes via strong relaxation, altimeter-based global mean sea-levels are used to constrain the global average of the fresh-water flux, and surface salinity is weakly restored to the monthly climatology of surface salinity from the WOA05 (Antonov et al. 2006).

ORAS4 consists of five ensemble members, which differ in their spin up procedure, but here we only present the central ensemble. The initial conditions in 1958 for this central ensemble came from a 18-year climatological spin-up, where the NEMO model starting from rest and the WOA05 temperature and salinity fields from January climatological conditions, is forced with fluxes from the ERA-Interim climatology. A strong relaxation (1-year time-scale) to WOA05 is employed to accelerate the spin-up process.

ORAS4-CNTRL is the equivalent model-only simulation to ORAS4, spanning the period 1958–2009. It uses the same model and forcing fluxes but no assimilation, bias correction or SST relaxation. An exception is the freshwater flux, which is constrained in the same way as in ORAS4. To reduce drift, the ORAS4-CNTRL utilizes a different spin-up strategy. Initial conditions for ORAS4-CNTRL were produced by a 18-year climatological spin up followed by a 50-year ocean simulation forced by the ERA-40/ERA-I time-varying daily surface fluxes from 1958 to 2008.

1.4 DEPRESYS

The Met Office Decadal Prediction System (Smith et al. 2007, 2013) is based on the HadCM3 (Gordon et al. 2000), with a resolution of 2.5° × 3.75° in the atmosphere and 1.25° in the ocean. HadCM3 has an active ice model. The assimilation is performed by integrating HadCM3 while relaxing to analysis products of oceanic temperature and salinity, and atmospheric horizontal winds, temperature and surface pressure. Six-hourly atmospheric analyses are taken from ERA-40 and operational analysis from 2002 onwards. Monthly ocean analyses are created by four-dimensional, multivariate optimal interpolation (Smith and Murphy 2007) of salinity and sub-surface temperature observations and analyzed SST from HadSST2 (Rayner et al. 2006) and HadISST.

The original Smith and Murphy (2007) analysis updated with improved covariances obtained using an iterative approach. For the first iteration, analyses for the period 1950-2006 were made using covariances computed from a nine member ensemble of perturbed physics variants of HadCM3 (Collins et al. 2010). Two further iterations were then performed using covariances computed directly from the analyses of the previous iteration. Covariances in the final iteration are potentially more accurate than the model covariances used in the first iteration because they are progressively influenced by real observations.

The relaxation timescales are three hours for the atmosphere and six hours for the ocean, and the 6-hourly atmosphere and monthly ocean analyses are linearly interpolated to the required model time. The ‘full-field’ version of DEPRESYS (Smith et al. 2010a) is used in this study such that the model is relaxed to the full observed values rather than anomalies.

1.5 SODA and SODA-NOASSIM

The SODA (Simple Ocean Data Assimilation) data used in this study is from two reanalyses: SODA 2.2.4 (SODA) and SODA 2.2.0 (SODA-NOASSIM) as described by Giese and Ray (2011). These SODA runs use the POP2 (Smith et al. 2010) ocean model in conjunction with the SODA assimilation optimal interpolation data assimilation system (Carton and Giese 2008). The two experiments are identical except that SODA includes data assimilation, and SODA-NOASSIM does not. The ocean model has a horizontal resolution that is on average 0.4° × 0.25° and with 40 levels in the vertical. Rivers are included with climatological seasonal discharge. There is no explicit sea ice model, although surface heat flux is modified when the surface temperature reaches the freezing point of seawater.

The assimilation is carried out sequentially using a 10-day update cycle with model error covariances determined from a simulation that does not include assimilation. The error covariances evolve in time as a function of the local velocity field and mixed layer depth. Updating is done incrementally following Bloom et al. (1996) to suppress excitation of spurious variability. The ocean model surface boundary conditions are provided from the 20CRv2 atmospheric dataset (Whitaker et al. 2004; Compo et al. 2011). The surface wind stress from 20CRv2 is used in the ocean model for the surface momentum fluxes. Solar radiation, specific humidity, cloud cover, 2-m air temperature, precipitation and 10-m wind speed from 20CRv2 are used for computing heat and fresh-water fluxes.

The temperature and salinity profile data have been obtained from the WOD09. The XBT and MBT observations used in SODA are from the WOD09 ‘standard level data’, which have been corrected for the fall-rate error as described by Levitus et al. (2009). Sea surface temperature observations are from the ICOADS Release 2.5 (Woodruff et al. 2011). In addition to assimilating temperature and salinity profile data we extract and assimilate mixed layer properties such as temperature, depth and barrier layer distribution.

1.6 MOVE-CORE and MRI-CORE

MOVE-CORE (Fujii et al. 2015) is a global ocean data assimilation system based on the Multivariate Ocean Variational Estimation/ Meteorological Research Institute Community Ocean Model (MOVE/MRI.COM; Usui et al. 2006; Fuet12). In the configuration used here, the MRI.COM3 model (Tsujino et al. 2001, 2011) is run at 1° zonal and 0.5° meridional resolution with 50 vertical levels and a prognostic sea ice model. The ocean model is forced with atmospheric data sets from the CORE-II interannual forcing (Griffies et al. 2012; Large and Yeager 2009). The system adopts a 3DVar analysis scheme based on Fujii and Kamachi (2003) and the incremental analysis update scheme of (Bloom et al. 1996) to assimilate in situ temperature and salinity observations from the WOD09 and GTSPP database (Hamilton 1994). The vertical and horizontal covariances of the first-guess temperature and salinity (T/S) fields are represented by multivariate empirical orthogonal functions based on historical T/S profiles in WOD09 and GTSPP and analysis increments are applied above 1750 m depth only. This yields anisotropic and spatially inhomogeneous horizontal decorrelation length scales ranging from O (100–1000 km). Other ocean variables including horizontal currents and sea surface height are not directly incremented, but are allowed to adjust through the model integration. No satellite data are assimilated. To suppress drift, the system also relaxes the first-guess forecasts to a monthly climatology of temperature and salinity above 1750 m based on the WOA09 with a restoring-time of 100 months.

MRI-CORE is a forced simulation of the same MRI.COM3 ocean model used in MOVE-CORE, but without assimilation and climatological restoring. As in MOVE-CORE, the simulation was forced with the CORE-II interannually varying atmospheric fields, and follows the experimental protocol of Griffies et al. (2012). Spin up for the MRI-CORE run consists of 4 cycles of a 60-year simulation over the 1948–2007 period using the CORE-II forcing. MRI-CORE is then the ocean state from the fifth cycle from 1948 to 2007.

The initial fields for the MOVE-CORE was generated by initializing from MRI-CORE at the beginning of 1998 in the fourth cycle and assimilating T/S from 1998 to 2007 while relaxing to climatology with a strong restoring time-scale of 20 months. The MOVE-CORE assimilation then progressed over the fifth forcing cycle. Both the MOVE-CORE and MRI-CORE were used in the CORE-II ocean model inter comparison project described in Danabasoglu et al. (2014), but are referred to as MRI-A and MRI-F (respectively) in that work.

1.7 NCAR-CORE and the CORE-II set of forced simulations

NCAR-CORE is the NCAR contribution to the CORE-II set of experiments (Danabasoglu et al. 2014, 2015). It uses the POP2 (Smith et al. 2010) and CICE4 (Hunke and Lipscomb 2008) ocean and sea-ice components of the CESM1 (Gent et al. 2011). POP2 is run in a global configuration with nominal 1° horizontal resolution. It has 60 vertical levels, with thickness monotonically increasing from 10 m in the upper ocean to 250 m in the deep ocean. Integration and forcing details follow the CORE-II protocol described in Griffies et al. (2012) and Danabasoglu et al. (2014)—see also below. As allowed by the protocol, a weak restoring of surface salinities (at a 4 year time-scale over the upper 50 m) to monthly mean salinity climatology is used. Further details of the ocean model configuration and parameterizations can be found in Danabasoglu et al. (2012), and a detailed description of the sea-ice model configuration in Holland et al. (2012).

The CORE-II set of simulations are described in detail in Danabasoglu et al. (2014) – here we briefly summarize. Twenty modeling group are participants in the intercomparison, wherein the CORE-II interannually varying atmospheric forcing data set (Large and Yeager 2009) are used to force a diverse set of global ocean–sea-ice models for the period 1948–2007. The simulations follow a common spin-up and forcing protocol. The spin-up consists of five repeat cycles of the 60-year forcing. Results only from the last forcing cycle, corresponding to years 1960–2007, are used in the analysis to avoid the impact of the spin-up and unphysical jump in forcing from 2007 back to 1948. Many of the participating models employ surface salinity relaxation to observed monthly mean climatology to avoid model salinity drift that can result in a weak or collapsed AMOC. The details of each model simulation in this set are presented in Danabasoglu et al. (2014). The MOVE-CORE and MRI-CORE simulations described earlier in this appendix are excluded from the CORE-II set here because they are included as reanalysis and No Assimilation sets, respectively.

Appendix 2: Acronyms

AVHRR:

Advanced Very High Resolution Radiometer

CESM1:

Community Earth System Model, version 1

CICE4:

Sea-Ice Model version 4

CORE-II:

Coordinated Ocean-Ice Reference Experiment, phase II

DEPRESYS:

U.K. Met Office Decadal Prediction System

ECDA:

Ensemble Coupled Data Assimilation

ERA-40 (ERA-I):

ECMWF atmosphere reanalysis and ECMWF atmosphere reanalysis Interim

ECMWF:

European Centre for Medium-Range Weather Forecasts

GECCO2:

The German Estimating the Circulation and Climate of the Ocean, version 2

GFDL:

Geophysical Fluid Dynamics Laboratory

GTSPP:

Global Temperature and Salinity Profile Programme

HADCM3:

Hadley Centre Climate Model version 3

HADISST:

Hadley Centre Global Sea Ice and Sea Surface Temperature dataset

HADSST2:

Hadley Centre Sea Surface Temperature dataset, version 2

ICOADS:

International Comprehensive Ocean–Atmosphere Data Set

MOM:

Modular Ocean Model

MOVE-CORE:

Multivariate Ocean Variational Estimation forced with CORE-II forcing

MRI.COM:

Meteorological Research Institute Community Ocean Model

MITgcm:

Massachusetts Institute of Technology [ocean] general circulation model

NCAR:

National Center for Atmospheric Research

NCEPOI-v2:

National Centers for Environmental Prediction optimal interpolation SST (version 2)

NCEP-RA1 (RA2):

National Centers for Environmental Prediction, atmospheric reanalysis version 1 (version 2)

NEMO:

Nucleus for European Modelling of the Ocean

NEMOVAR:

NEMO-based variational data assimilation system

ORAS4:

Ocean Reanalysis System version 4

POP2:

Parallel Ocean Program, version 2

RAPID-MOCHA or RAPID:

United Kingdom–United States Rapid Climate Change, Meridional Overturning Circulation and Heatflux Array

SODA:

Simple Ocean Data Assimilation

WOA09 (05):

World Ocean Atlas 2009 (2005)

WOD09:

World Ocean Database 2009

20CRv2:

20th Century [atmospheric] Reanalysis, version 2

XBT/MBT/CTD:

Expendable bathymetric thermistors, mechanical bathymetric thermistors, conductivity-temperature-depth sensors